This dissertation is dedicated to my mother, Song Guizhi, and father, Liu Qinghe, to my wife, Yi Yue, and my daughter, Liu

Chang

Acknowledgments
First of all, I would like to express my sincere gratitude and appreciation to my promoter, Prof. F. De Smedt, for providing me the opportunity to fulfil this PhD, and for his patience, encouragement and financial support throughout my PhD research in the Frije Universiteit Brussel. I really thank him for guiding me to conduct my PhD research in an area that is challenging and enjoyable in the field of water resources engineering: Development and application of a GIS-based hydrological model for flood prediction and watershed management. It was a fruitful and enriching experience for me in my life. Thank you very much. I would like to thank Mr. O. Batelaan, lecture of IUPWARE (Interuniversity Program in Water Resources Engineering), for his constructive suggestions and fruitful discussions on my study. Special thanks are given to Prof. J.W. Delleur, (Purdue University, Indiana, USA), for whom I served as a teaching assistant for the course Advanced Hydraulics, IUPWARE, for 4 academic years, and I really gained a lot from this experience. My thanks also go to former and current fellow colleagues: Z. Wang, T. Van Dale, L. Feyen, A. Deng, J. Cools, J. Corluy, A. Aish, A. Bahremand, I. Houcyne, L.Q. Hung, M. Charles, Y. Meyus, J. Rwetabula, J. Severyns, B. Verbeiren, S. Tuccu, and the technical staff members of the department: H. De Coninck, E. Van Den Stoeme, E. Verbeken and A. Cosemans, for their help during my PhD, and for making my stay at VUB enjoyable. During the study of my PhD, I had the opportunities to participate in several Belgian and European projects, e.g. the project of modelling floods in three Belgian watersheds, Barebeek, Ijse and Veuren-Ambacht; the FRHYMP (Flood Risk and HYdrological MAPping towards sustainable flood risk management in the Rhine and Meuse river basins) project; and the Tisza River project. These projects also form the case studies of this PhD research. I thank Dr. L. Pfister of the Centre de Recherche Public Gabriel Lippmann, Grand-Duchy of Luxembourg for his kind cooperation and for providing all the required data of the Alzette River basin. Thanks are also given to Mrs. J. Poorova, L. Velcicka and M. Dobiasova of the Slovak

Acknowledgments Hydrometeorological Institute, Slovakia, for providing all necessary data of the Hornad River Basin and the fruitful discussions in Bratislava and Brussels. I am really grateful to my jury members: Prof. J. Wastiels (Vrije Universiteit Brussel), Prof. J. Vereecken (Vrije Universiteit Brussel), Prof. W. Bauwns (Frije Universiteit Brussel), Prof. L. Hoffmann (Centre de Recherche Public-Gabriel Lippmann, GrandDuchy Luxembourg), Prof. Jacques W. Delleur (School of Engineering, Purdue University, USA), and Prof. J. Bogaert (Universit Libre de Bruxelles),for their willingness to review and evaluate this thesis. Their critical comments and suggestions make this PhD thesis a reality. I am grateful to my parents who are living in China. I owe much of my success in this endeavour to their loving efforts and never-ending support. This thesis would not have been possible without the complete support of my wife, Yue Yi. Her support, patience and understanding during the research and writing process, along with the love of my beautiful daughter, Chang Liu, made this task more worthwhile and fulfilling. I am deeply indebted for not having spent much time with them throughout my study at Katholieke Universiteit Leuven and Vrije University Brussel in the Belgium. Their love reminded me of what is truly important and helped me maintain perspective on life while completing this research. I love you all!

ii

Abstract
A GIS-based distributed hydrological model, WetSpa Extension, that operates on catchment scale is developed for flood prediction and watershed management. It is a continuous simulation model that may operate at a different time scale, e.g. hourly or daily, and at a different spatial resolution, which well represent the river basin characteristics. The required inputs to the model include digital maps of elevation, soil type and land use, and time series of precipitation, temperature, potential evapotranspiration and flow discharge, where temperature data is optionally used for modelling snowmelt, and discharge data is required for model calibration. The model enables one to simulate the complex hydrological regimes of a river basin within a GIS framework, estimate runoff for each grid cell and route the flow along its flow path to the basin outlet or any converging points in the stream network, and eventually to simulate the spatial distribution of hydrological variables in a river basin, such as surface runoff, interflow, soil moisture and groundwater recharge, etc. The hydrological processes considered in the model include precipitation, interception, snow accumulation and melting, depression storage, infiltration, evapotranspiration, percolation, surface runoff, interflow and groundwater flow. Surface runoff in each cell is computed by a modified rational method, which is controlled by the rainfall intensity and soil moisture content. Potential runoff coefficient is obtained from literature and a lookup table is created linking potential runoff coefficient with different categories of slope, soil type, land use and the proportions of bare soil and impervious areas in a grid cell. The excess rainfall is routed along the flow paths by using the diffusive wave approximation. The interflow and percolation are controlled by soil characteristics and modelled by Darcys law and kinematic approximation. The groundwater flow is modelled by a linear reservoir method on small GIS derived subcatchment scale, while a non-linear reservoir method is optional in the model. Actual evapotranspiration is composed of evaporation from interception and depression storage, and transpiration from root zone and groundwater storage. Snowmelt is estimated using a simple degree-day model. Algorithms derived as much as possible from physical processes, together with more conceptual or empirical algorithms have been selected. The structure of the model with regard to interception,

Abstract depression, root-zone and groundwater storage compartments is variable, allowing much flexibility to simulated different systems. The application of the WetSpa model is demonstrated in three case studies, i.e. Barebeek (Belgium), Steinsel (Luxembourg) and Margecany (Slovakia), which are described in this dissertation. The Barebeek catchment is a typical suburban watershed with drainage area of 67.8 km2 situated northeast of Brussels, Belgium. The resulting calculated hydrographs with WetSpa model compare favourably with measurements at the gauging sites. The usefulness and utility of the model are subsequently demonstrated by forecasting peak discharges resulting from an observed 102 years precipitation series. The resulting hourly discharges were analyzed statistically to determine the characteristics of extreme flood events and compared with the results computed from design storms. Comparison of the two methods shows that the model is capable to predict both normal and extreme floods. The Steinsel catchment is a highly urbanized watershed with a drainage area of 407 km2 located in the upstream part of the Alzette River basin, Grand Duchy of Luxembourg. Results of WetSpa model simulation show that the models level of representativeness to be quite satisfactory. Next, the runoff contribution from different land use areas and the impacts of land use change and natural river restoration for the headwater areas on the flood behaviours are assessed using WetSpa model on an hourly time scale and 5050 m resolution results. The Margecany catchment is a typical mountainous watershed with a drainage area of 1133 km2 situated in the upstream part of the Hornad River basin, Slovakia. Simulation results of WetSpa model show that the flow hydrographs of both snow melting floods and storm floods are well reproduced on a daily time scale and 100100 m resolution. Moreover, it is demonstrated that the spatial distribution of input data has a large influence on the modelling results, particularly for a mountainous catchment. The WetSpa model makes full use of the remote sensed data and calculations are for the most part performed by standard GIS tools, such that the model is especially useful for flood prediction on complex terrain and analyzing the effects of topography, soil type, and land use or soil cover on the flood. Additionally, the model can be easily coupled with other water quality and soil erosion models, and used for simulating spatial hydrological behaviour of a river basin. iv

Table of contents
Acknowledgments.......................................................................................................... i Abstract........................................................................................................................ iii Table of contents........................................................................................................... v List of figures............................................................................................................. xiii List of tables ............................................................................................................. xvii List of publications..................................................................................................... xix General introduction........................................................................... 1

Chapter I: 1. 2. 3.

Background and significance............................................................................ 1 Objectives of the research............................................................................. 3 Outline of the dissertation............................................................................. 4 References..................................................................................................... 7

Chapter II: 1. 2. 2.1. 2.2. 2.3. 3. 3.1. 3.2. 3.3. 4. 4.1. 4.2. 5. 5.1. 5.2. 6.

GIS-based hydrological modelling and watershed analysis............ 9

Abstract ............................................................................................................ 9 Introduction ...................................................................................................... 9 Effects of watershed characteristics on runoff............................................ 11 Effects of topography...................................................................................... 11 Effects of soil type........................................................................................... 13 Effects of land use........................................................................................... 14 GIS applications in watershed modelling........................................................ 16 Watershed description..................................................................................... 16 Hydrological parameter determination............................................................ 17 Integration with hydrological models.............................................................. 18 Modelling of watershed hydrology................................................................. 20 Popular GIS-based hydrological models......................................................... 21 Assessment of future scenarios....................................................................... 23 WetSpa model overview................................................................................. 24 Model history.................................................................................................. 24 WetSpa Extension........................................................................................... 25 Summary......................................................................................................... 28

Table of contents References ...................................................................................................... 29

Chapter III: Development of a diffusive transport approach for flow routing in GIS-based watershed modelling.............................................................. 35 Abstract .......................................................................................................... 35 1. 2. 3. 4. 4.1. 4.2. 4.3. 4.4. 4.5. 5. Introduction .................................................................................................... 35 Methodology................................................................................................... 38 Model application ........................................................................................... 43 Sensitivity analysis...................................................................................... 50 Effect of hydraulic radius................................................................................ 50 Effect of channel roughness............................................................................ 51 Effect of minimum slope................................................................................. 52 Effect of area threshold in delineating channel networks............................... 53 Other effects.................................................................................................... 55 Conclusions..................................................................................................... 56 References ...................................................................................................... 58

Chapter IV: Flood modelling for complex terrain using GIS and remote sensed information..................................................................................................... 61 Abstract .......................................................................................................... 61 1. 2. 2.1. 2.2. 2.3. 3. 3.1. 3.2. 3.3. 4. 4.1. 4.2. 4.3. 5. vi Introduction .................................................................................................... 61 The modelling approach.................................................................................. 63 Runoff production............................................................................... 64 Water balance.................................................................................................. 68 Flow routing............................................................................ 70 GIS implementation ....................................................................................... 72 Drainage system.............................................................................. 72 Soil and land use.......................................................................... 74 Spatial hydrological input and output............................................................. 75 Model application........................ 76 Watershed description and data availability.................................................... 76 Model calibration............................................................................................ 78 Model application using the historical and IDF data...................................... 82 Discussion and conclusions............................................................................. 85

Development and application of a GIS-based hydrological model References ...................................................................................................... 87

Chapter V:

Assessing land use impacts on flood processes using a GIS

modelling approach....................................................................................... 91 Abstract .......................................................................................................... 91 1. 2. 2.1. 2.2. 2.3. 2.4. 2.5. 3. 3.1. 3.2. 4. Introduction .................................................................................................... 91 Methodology................................................................................... 94 Description of the WetSpa model................................................................... 94 Description of the study area........................................................................... 96 Data collection......................................................................... 98 Model calibration and verification................................................................ 100 Model evaluation........................................................................................... 103 Results and discussion.......................................................... 105 Evaluating runoff partitions from different land use classes......................... 105 Assessing the impact of land use changes on flood...................................... 112 Conclusions....................... 116 References .................................................................................................... 117

Chapter VI: Assessing the effects of river restoration on the reduction of floods in a river basin.................................................................................. 120 Abstract ........................................................................................................ 120 1. 2. 2.1. 2.2. 2.3. 2.4. 3. 3.1. 3.2. 3.3. 4. Introduction .................................................................................................. 120 Methodology................................................................................. 124 Model description.......................................................................................... 124 Description of the study area......................................................................... 124 Stream classification ........................................................................ 124 Modelling approach............................................................... 126 Results and discussion................................................................................... 129 Model calibration and evaluation.................................................................. 129 Model prediction........................................................................................... 131 Evaluation of a future flood scenario............................................................ 134 Conclusions......................... 136 References .................................................................................................... 137

vii

Table of contents

Chapter VII: Flow simulation in a Carpathian catchment accounting for topographic controls.................................................................................... 139 Abstract ........................................................................................................ 139 1. 2. 2.1. 2.2. 2.3. Introduction .................................................................................................. 139 Methodology................................................................................. 141 Description of the study area and data available........................................... 141 Modelling snowmelt...................................................................................... 145 Topographic adjustment for the input variable............................................. 146

2.3.1. Adjustment for temperature........................................................................... 147 2.3.2. Adjustment for precipitation......................................................................... 148 2.3.3. Adjustment for PET....................................................................................... 150 2.3.4. Topographic correction................................................................................. 151 3. 3.1. 3.2. 3.3. 5. Model simulation........................................................................................... 151 Parameter identification................................................................................ 152 Automated calibration................................................................................... 154 Modelling results........................................................................................... 156 Discussion and conclusions.......................................................................... 160 References .................................................................................................... 162

Chapter VIII: Integrating GIS and hydrological process modelling in medium and large watersheds................................................................................... 165 Abstract ........................................................................................................ 165 1. 2. 3. 4. 5. 6. Introduction .................................................................................................. 165 Study area and data availability..................................................................... 168 Effects of grid size on runoff and flow responses......................................... 170 Transforming WetSpa into a semi-distributed model................................... 173 Results and discussion................................................................................... 178 Conclusions................................................................................................... 183 References .................................................................................................... 185 Chapter IX: Summary and conclusions.............................................................. 189 1. 1.1. viii General summary.......................................................................................... 189 Model development....................................................................................... 189

Development and application of a GIS-based hydrological model 1.2. 1.3. 2. Model applications........................................................................................ 192 Model limitations.............................................................................. 195 Future perspectives........................................................................................ 197

Appendix A: WetSpa Extension: A GIS-based hydrological model for flood prediction and watershed management Documentation and User Manual.............................................................. 201 1. 1.1. Model description.......................................................................................... 201 Model construction........................................................................................ 202

1.1.1. Model objectives........................................................................................... 202 1.1.2. Model structure.............................................................................................. 203 1.1.3. Model assumptions........................................................................................ 205 1.2. Data preparation............................................................................................ 205 1.2.1. Digital data.................................................................................................... 206 1.2.2. Hydro-meteorological data............................................................................ 208 2. 2.1. 2.2. 2.3. 2.4. 2.5. Model formulation......................................................................................... 210 Precipitation................................................................................................... 211 Interception.................................................................................................... 212 Snowmelt....................................................................................................... 214 Rainfall excess and infiltration...................................................................... 215 Depression and overland flow....................................................................... 217

2.5.1. Formulation of depression storage................................................................ 218 2.5.2. Mass balance of depression storage.............................................................. 219 2.5.3. Formulation of overland flow........................................................................ 220 2.6. 2.7. Water balance in the root zone...................................................................... 221 Evapotranspiration from soil......................................................................... 222

2.7.1. Potential evapotranspiration.......................................................................... 222 2.7.2. Actual evapotranspiration.............................................................................. 225 2.8. 2.9. 2.10. Percolation and interflow.............................................................................. 226 Groundwater storage and baseflow............................................................... 229 Overland flow and channel flow routing....................................................... 231

2.10.1. Flow response at a cell level.......................................................................... 231 2.10.2. Flow response at a flow path level................................................................ 233 2.10.3. Flow response of the catchment.................................................................... 235 ix

Table of contents 2.11. 2.12. 3. 3.1. Subcatchment integration.............................................................................. 235 Catchment water balance............................................................................... 237 Parameter identification and model evaluation............................................. 239 Default model parameters............................................................................. 239

3.1.1. Parameters characterizing soil texture classes............................................... 239 3.1.2. Parameters characterizing land use classes................................................... 240 3.1.3. Potential runoff coefficient............................................................................ 243 3.1.4. Depression storage capacity.......................................................................... 247 3.2. 3.3. 4. 4.1. 4.2. Global parameters.......................................................................................... 250 Model evaluation........................................................................................... 255 Model operation............................................................................................ 258 Program installation...................................................................................... 258 Program description...................................................................................... 260

4.2.1. Avenue scripts and their tasks....................................................................... 260 4.2.2. Lookup tables................................................................................................ 261 4.2.3. FORTRAN programs and their tasks............................................................ 261 4.2.4. PEST files and their tasks.............................................................................. 262 4.3. GIS pre-processing........................................................................................ 262 4.3.1. Surface grid preparation................................................................................ 262 4.3.2. Soil based grid preparation............................................................................ 266 4.3.3. Land use based grid preparation.................................................................... 267 4.3.4. Potential runoff coefficient and depression storage capacity........................ 268 4.3.5. Flow routing parameters................................................................................ 269 4.3.6. Thiessen polygon........................................................................................... 270 4.3.7. Drainage systems for a complex terrain........................................................ 271 4.4. Creation of input files.................................................................................... 272 4.4.1. Input file of time series.................................................................................. 272 4.4.2. Global parameters and spatial output specifications..................................... 275 4.5. Model calibration and verification................................................................ 278 4.5.1. Calibration and verification processes.......................................................... 278 4.5.2. Manual parameter adjustment....................................................................... 282 4.5.3. Parameter sensitivity..................................................................................... 283 4.6. Model output................................................................................................. 285 4.6.1. Intermediate output........................................................................................ 285 x

Development and application of a GIS-based hydrological model 4.6.2. Final output.................................................................................................... 287 4.6.3. Post processing of model outputs.................................................................. 291 5. 5.1. 5.2. 5.3. 5.4. 5.5. 6. Case study...................................................................................................... 291 Description of the study area......................................................................... 291 Data availability............................................................................................ 293 Basin delineation and parameter determination............................................ 297 Model calibration and validation................................................................... 299 Discussion..................................................................................................... 304 Concluding remarks...................................................................................... 306 References..................................................................................................... 307

xi

List of figures
II-1: GISHydrological modelling integration methods............................................. 19 II-2: Hydrological processes considered in the original WetSpa model..................... 24 III-1: (a) Unit response functions for an expected travel time of 3600 s and different standard deviations, and (b) Unit response functions for an expected standard deviation of 3600 s and different travel times................... 41 III-2: Location plan showing the study area, the Attert and Alzette river basin... 44 III-3: DEM of the study area..................................................................................... 44 III-4: Land use map of the study area....................................................................... 44 III-5: Distribution of potential runoff coefficient..................................................... 45 III-6: Distribution of hydraulic radius for a flood with a 2-year return period......... 45 III-7: (a) Average flow time to the basin outlet and (b) its standard deviation........ 47 III-8: Observed and predicted stream flow and baseflow at Ell station.................... 49 III-9: Measured vs. simulated peak direct discharges............................................... 49 III-10: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of hydraulic radius with expected flood frequency, p.. 51 III-11: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of channel Mannings roughness coefficient, n................ 52 III-12: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of the threshold of minimum slope, Smin.......................... 53 III-13: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of cell number threshold, Cn, in delineating channel networks.......................................................................................................... 54 IV-1: Location of the Barebeek catchment................................................................. 77 IV-2: Drainage system of the Barebeek catchment..................................................... 77 IV-3: Land use map of the Barebeek catchment......................................................... 77 IV-4: Distribution of potential runoff coefficient...................................................... 79 IV-5: Distribution of average flow time to the catchment outlet................................ 79 IV-6: Observed Vs calculated discharges at MO6...................................................... 80 IV-7: Observed Vs calculated discharges at MO3...................................................... 80 IV-8: Observed Vs calculated flow hydrographs at MO6 for the period of Sept. 1998 to Dec. 1999............................................................................................ 80

List of Figures IV-9: Simulated distribution of relative soil wetness on 8/10/1999, 4:00................... 82 IV-10: Simulated distribution of surface runoff on 8/10/1999, 4:00-5:00.................. 82 IV-11: Design summer and winter storms with 100-year return period..................... 83 IV-12: Simulated floods for the 100-year design storms............................................ 83 IV-13: Comparison of the design peak flow discharges............................................. 84 IV-14: Simulated maximum flood at the watershed outlet......................................... 84 V-1: V-2: V-3: V-4: V-5: V-6: V-7: V-8: V-9: Location of the Alzette basin and Steinsel sub-basin...................................... 97 Topography and gauging network of the Steinsel sub-basin.......................... 97 Soil type map of the Steinsel sub-basin........................................................... 99 Land use map of the Steinsel sub-basin.......................................................... 99 Observed Vs calculated flow at Steinsel for the floods in Feb. 1997........... 102 Observed and calculated daily flow at Steinsel for the year 2000................ 102 Storm runoff partitions at Steinsel for the flood events in Feb. 1997........... 105 Storm runoff contributions at Steinsel for the flood events in Apr. 1999..... 106 Plot of event to event variations of the different runoff contributions (a), normalized relative runoff contribution from urban (b), cropland (c), grassland (d), woodland (e), and other areas (f), and error in flood volume (g) and error in peak discharge with respect to the flow coefficient.109 V-10: Contributions of monthly flow at Steinsel from different land use classes... 111 V-11: Land use change scenarios for the Steinsel sub-basin................................... 113 V-12: (a) Simulated surface runoff distribution under present land use condition for the storm on Feb. 24-26, 1997, and (b) Simulated surface runoff distribution after urbanization for the same storm event............................... 114 V-13: Simulated hydrographs for each scenario for a storm in Dec. 1999............. 115 V-14: Peak discharges for each scenario over the simulation period...................... 115 VI-1: (a) Stream orders and their drained area, and (b) percentage of stream length, percentage of drained area and average slope for different order streams............................................................................. 125 VI-2: Observed and simulated flow hydrographs for the flood events in Oct. and Nov. 1998............................................................................................... 130 VI-3: Observed versus simulated peak flows for the simulation period................. 130 VI-4: (a) Average flow travel time to the sub-basin outlet for the present condition, and (b) Increases in flow travel time after river restoration.......................... 132 VI-5: Flood events showing the effect of natural river restoration......................... 133 xiv

Development and application of a GIS-based hydrological model VI-6: Present versus restored simulated peak discharges for the simulation period indicating a 14% reduction in average after river restoration....................... 133 VI-7: Simulated hydrograph under present condition and after river restoration for a future storm scenario.................................................................................. 135 VII-1: Location of the Margecany catchment.......................................................... 142 VII-2: Monthly temperature, precipitation and PET at Spisske Vlachy.................. 142 VII-3: Land use map of the Margecany catchment.................................................. 142 VII-4: Soil textural map of the Margecany catchment............................................. 144 VII-5: Topographical map of the Margecany catchment......................................... 144 VII-6: Gauging sites and Thiessen polygons for the Margecany catchment........... 144 VII-7: Lapse rates for mean monthly temperature................... 147 VII-8: Distribution of yearly precipitation over the Hornad River basin................. 149 VII-9: Vertical gradient of yearly precipitation with elevation................................ 149 VII-10: Vertical gradient of monthly PET with elevation........................................ 151 VII-11: Potential runoff coefficient for the Margecany catchment........................... 153 VII-12: Mean travel time to the basin outlet at Margecany...................................... 154 VII-13: Observed and calculated daily flow at Margecany for the year 1997.......... 157 VII-14: Variation of precipitation, temperature, evapotranspiration, and relative soil saturation for the Margecany catchment during the year 1997.............. 159 VII-15: Distribution of surface runoff for the storm event 1/8-6/8, 1997................. 160 VII-16: Observed and simulated mean daily PET at Spisske Vlachy....................... 161 VIII-1: Study area and observation network............................................................. 168 VIII-2: Land use of the Alzette River basin.............................................................. 168 VIII-3: Mean parameters obtained from DEMs with grid sizes: (a) flow length, flow time and its standard deviation, (b) slope, curvature and depression capacity, (c) runoff coefficient, velocity and hydraulic radius, (d) IUHs for the entire catchment................................................................................. 171 VIII-4: Comparison of flow hydrographs at Ettelbruck calculated from DEMs with different grid size for a flood event in Dec. 1999......................................... 172 VIII-5: Comparison of surface flow hydrographs at Ettelbruck calculated from DEMs with different grid size for a flood event in Dec. 1999................................. 173 VIII-6: Mean travel time to the basin outlet.............................................................. 179 VIII-7: River reaches and divided subwatersheds..................................................... 179 VIII-8: Calculated Vs observed flows at Ettelbruck for the floods in Feb. 1997...... 181 xv

List of Figures VIII-9: Simulated Vs observed daily flows at Ettelbruck for the year 1999............. 182 A-1.1: Model structure of WetSpa Extension at a pixel cell level........................... 204 A-2.1: Annual variation of grass interception storage capacity............................... 214 A-2.2: Relationship between rainfall excess coefficient and soil moisture.............. 217 A-2.3: Sketch of depression storage as a function of excess rainfall....................... 219 A-2.4: Graphical presentation of excess rainfall and overland flow........................ 220 A-2.5: Graphical presentation of soil water balance................................................ 221 A-2.6: Observed and simulated daily EP at Ukkel for the year 1997....................... 224 A-2.7: Simulated hourly EP at Ukkel with EPd = 3mm........................................... 224 A-2.8: Graphical presentation of soil evapotranspiration......................................... 226 A-2.9: Effective hydraulic conductivity as a function of moisture content.............. 227 A-2.10: Flow path response functions with different ti and i2................................. 234 A-3.1: Potential runoff coefficient vs. slope for forest and different soil types....... 245 A-3.2: Depression storage capacities vs. slope for grass and different soil types.... 249 A-4.1: Schematic view of the models project folders............................................. 259 A-4.2: Screenshort of surface menu......................................................................... 263 A-4.3: Screenshort of parameter menu..................................................................... 267 A-5.1: Location of the Bissen catchment................................................................. 292 A-5.2: Topography map of Bissen............................................................................ 293 A-5.3: Land use map of Bissen................................................................................. 293 A-5.4: Soil type map of Bissen................................................................................. 293 A-5.5: River network and Thiessen polygons of Bissen......... 293 A-5.6: Hydraulic radius of Bissen............................................................................ 298 A-5.7: Runoff coefficient of Bissen.......................................................................... 298 A-5.8: Mean travel time to the basin outlet of Bissen.............................................. 299 A-5.9: Standard deviation of flow time to the basin outlet ...................... 299 A-5.10: Observed and calculated flow at Bissen for the floods in Dec. 1999.......... 301 A-5.11: Observed and calculated hourly flow at Bissen for the year 1999............... 303 A-5.12: Peak Qm Vs Peak Qc selected from the whole simulation period................ 303 A-5.13: Observed and calculated hourly flow frequency curves at Bissen............... 304

xvi

List of tables
II-1: II-2: V-1: V-2: V-3: V-4: V-5: Primary topographic attributes that can be computed by terrain analysis from DEM Data............................................................................................... 12 Samples of popular distributed and semi-distributed hydrological models.... 22 Description of the area, slope and main soil types for each land use class..... 98 Evaluation criteria for the assessment of model performance...................... 104 Watershed characteristics and model performance....................................... 104 Simulated runoff contributions from different land use classes.................... 107 Land use change scenarios compared with the present situation.................. 113

VII-1: Information of weather stations in the Margecany catchment...................... 143 VII-2: Regression analysis between monthly temperature and elevation................ 148 VII-3: PET stations used for regression analysis..................................................... 150 VII-4: Regression analysis between PET and elevation.......................................... 151 VII-5: Parameters and their ranges in the PEST control file..................................... 155 VII-6: List of input parameter values, water balance, and evaluation results.......... 157 VIII-1: Mean parameter values calculated from maps with different grid size....... 170 VIII-2: Flood characteristics estimated from maps with different grid size..... 173 VIII-3: Subwatershed characteristics and model performance.. 183 A-3.1: Default parameters characterizing soil textural classes................................. 240 A-3.2: Default parameters characterizing land use classes...................................... 241 A-3.3: Potential runoff coefficient for different land use, soil type and slope......... 243 A-3.4: Slope constant S0 for determining potential runoff coefficient.................... 244 A-3.5: Impervious percentages associated with selected land use classes............... 246 A-3.6: Depression storage capacity for different land use, soil type and slope........ 248 A-4.1: Sample file of precipitation series p.txt......................................................... 273 A-4.2: Sample file of potential evapotranspiration series pet.txt............................. 274 A-4.3: Sample file of temperature series t.txt........................................................... 274 A-4.4: Sample file of discharge series q.txt.............................................................. 275 A-4.5: Template of global model parameters........................................................... 275 A-4.6: Template of spatial output specifications...................................................... 276 A-4.7: Parameter sensitivity for model calibration................................................... 284 A-4.8: Sample output file of mean.txt...................................................................... 285

List of Tables A-4.9: Parts of output file uh_cell_h.txt.................................................................. 286 A-4.10: Sample output file of q_tot.txt..................................................................... 287 A-4.11: Sample output file of q_sub.txt.................................................................... 288 A-4.12: Sample output file of balance.txt................................................................. 289 A-4.13: Parts of the output file runoff.asc................................................................. 289 A-4.14: Model evaluation result evaluation .txt........................................................ 290 A-5.1: Default parameter values in the PET formula for different land uses........... 296 A-5.2: Data available and characteristics of the Bissen catchment........................... 297 A-5.3: Statistics and model performance for the calibration/validation period........ 302 A-5.4: Water balance estimation at Bissen for the whole simulation period............ 305

xviii

List of publications
Liu, Y.B., Gebremeskel. S., De Smedt, F., Hoffmann, L. and Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modeling, Journal of Hydrology, 283, 91-106, 2003. Liu, Y.B., De Smedt, F., Hoffmann, L. and Pfister, L., Assessing land use impact on flood processes in complex terrain by using GIS and Modeling approach, Environmental Modeling and Assessment, 2004 (in press). Liu, Y.B. and De Smedt, F., Flood modeling for complex terrain using GIS and remote sensed information, Water Resources Management, 2004 (accepted). Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. and Pfister, L., Predicting storm runoff from different land use classes using a GIS-based distributed model, Hydrological Processes, 2004 (accepted). Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. and Pfister, L., Simulation of flood reduction by natural river rehabilitation using a distributed hydrological model, Hydrology and Earth System Sciences, 2004 (accepted). Liu Y.B., De Smedt F, Hoffmann L. and Pfister L., Parameterization using ArcView GIS in medium and large watershed modeling, in Chen, Y.B., Takara, K., Cluckie, I. and De Smedt, F. (eds.), GIS and Remote Sensing in Hydrology, Water Resources and Environment, IAHS Publ. 289, 50-58, 2004. Liu, Y.B., Batelaan, O., Huong, N.T., Tam, V.T. and De Smedt, F., Flood prediction in the karstic Suoimuoi catchment, Vietnam, in Batelaan, O., Dusar, M., Masschelein, J., Tam, V.T., Van, T.T. and Khien, N.X. (eds.), Trans-KARST, Proceedings of the International Transdisciplinary Conference on Development and Conservation of Karst Regions, Hanoi, Vietnam, 139-144, 2004. Liu, Y.B., Gebremeskel. S., De Smedt, F. and Pfister, L., Flood prediction with the WetSpa model on catchment scale, in Wu, B.S., Wang, Z.Y., Wang, Q., Huang, G.H., Fang, W. and Huang, J.C. (eds.), Flood Defence 2002, 499-507, Science Press, New York Ltd, 2002. De Smedt, F., Liu, Y.B., Gebremeskel, S., Hoffmann, L. and Pfister, L., Application of GIS and remote sensing in flood modeling for complex terrain, in Chen, Y.B., Takara, K., Cluckie, I. and De Smedt, F. (eds.), GIS and Remote Sensing in Hydrology, Water Resources and Environment, IAHS Publ. 289, 23-32, 2004

Listof publications De Smedt, F., Liu, Y.B., and Qiao, Y., Prediction of floods with the WetSpa model, Annals of Warsaw Agricultural University SGGW, Land Reclamation, 33: 7180, 2002. De Smedt, F., Liu, Y.B. and Gebremeskel, S., Hydrologic modeling on a catchment scale using GIS and remote sensed land use information, in Brebbia, C.A. (ed.), Risk Analysis II, 295-304, WTI press, Southampton, Boston, 2000. De Smedt, F., Liu, Y.B. and Gebremeskel, S., Integrated Hydrologic Modeling on a Catchment Scale for Prediction of Floods, in Hoeben, R., Herpe, Y.V. and De Troch, F.P. (eds.), ERB 2000 Proceedings on CDROM, 11pp, Ghent, Belgium, September 27-29, 2000. Gebremeskel S., Liu, Y.B., De Smedt, F. and Pfister, L., 2002. GIS based distributed modelling for flood simulation. J. A. Ramirez (ed), Proceedings of the Twenty Second Annual American Geoplhysical Union Hydrology Days, Fort Collins, Colorado: 98-109. Liu, Y.B., De Smedt, F., Hoffmann, L. and Pfister, L., Integrating GIS and process modeling for runoff prediction in medium and large watersheds, 2004 (submitted to Journal of Spatial Hydrology). Liu, Y.B., Yi, Y., Batelaan, O. and De Smedt, F., Assessing grid size effects on runoff and flow response using a GIS-based hydrologic model, 2004 (submitted to Environmental Modeling and Assessment). Liu, Y.B., Batelaan, O., De Smedt, F., Poorova, J. and Velcicka, L., Automated calibration applied to a GIS-based flood simulation model using PEST, 2005 (to be submitted to the Third International Symposium on Flood Defence, 25-27, May, 2005, Nijmegen, the Netherlands). Liu, Y.B., Corluy, J., Bahremand, A., De Smedt, F., Poorova, J. and Velcicka, L., Estimation of runoff and phosphorus loading for the Margecany catchment, Hornad, Slovakia, 2005 (to be submitted to the International Symposium on Wetland Pollutant Dynamics and Control, 4-8 September 2005, Ghent, Belgium).

xx

Chapter I General introduction

1. Background and significance

Due to the increased water resources demand, there is a growing requirement to predict their natural processes in order to address the environmental problems today and in the future. A hydrological model is a simplified representation of the natural hydrological system, and represents different physical processes at a wide range of time and space scales. This has generally been associated with an increase in model complexity, a lack of appropriate observational data to constrain model states, and an increasing number of model outputs (Wagener et al., 2001). In particular, distributed hydrological models allow for detailed description of the hydrological and energy cycle and provide opportunities for dealing with forcing variables that fluctuate in space and time. Hydrologists are therefore trying to implement these models increasingly as a means to capture the state of knowledge on basins of interest, and provide valuable information regarding hydrological state variables and potentially important distributed information on existing and future streamflow conditions. Flood prediction and control is one of the greatest challenges facing the world today, which have become more frequent and severe due to the effects of global climate change and human alterations of the natural environment. As the watershed becomes more developed, it also becomes more hydrologically active, changing the streams flow components as well as the origin of flow. In turn, flood flows that once occurred infrequently during pre-development periods have now become more frequent and more severe due to the transformation of the watershed from rural to urban land uses (Boyle et al., 1997; Weng, 2001). Consequently, the extent of the floodplain also is altered and the risk of flooding imposed on surrounding areas is increased. The forecast of flooding would benefit greatly from the use of hydrological models, which are designed to simulate the processes of surface or subsurface water flow. Because the flow processes are spatially distributed, a great amount of spatially related physical data needs to be prepared and analyzed in order to construct an adequate

Chapter I simulation model, including variables related to the water, the air, the landscape, the soil, the plant, and all of the geophysical and environmental constituents. Therefore, in order to successfully manage natural hydrological systems, one must have a thorough understanding of these variables with regard to their characteristic temporal and spatial scale. Modern geographic information systems (GIS) offer new opportunities for the collection, storage, analysis, and display of spatially distributed meteorological and geophysical data (Goodchild et al., 1992). The use of GIS enables one to implement geographic data more efficiently for hydrological monitoring, analyzing, planning and management. More specifically, the use of a GIS in combination with hydrological models allows one to perform multi-objective analyses incorporating a wide range of geographical information and data in an accurate and efficient manner (Babd & Moore, 1995; McDonnell, 1996). Despite the advent of recent integrated approaches to watershed planning and development, as well as the large research effort to enhance the interface between GIS technology and hydrological models, a need remains to make the GIS-based hydrological models more reliable, effective and easy to implement for research and engineering purposes. In addition, a GIS methodology for predicting runoff and flood risks spatially within the corresponding watershed needs to be developed. These applications of GIS for flood prediction and watershed management would greatly improve the capabilities of the hydrological modelling and risk assessment, offering users a powerful capability to analyze and visually express the spatially distributed hydrological variables, and therefore, assess the impacts of land use and climate change on hydrological responses of the river basin. The purpose of this research is to develop and test a GIS-based flow and water balance simulation model, WetSpa (Water and Energy Transfer between Soil, Plant and Atmosphere) Extension, based on the pre-proposed WetSpa model (Wang et al., 1997). The model uses the spatial information of a digital elevation model (DEM), land use and soil to derive all necessary spatially distributed model parameters within a GIS framework. Hydrological computations are performed for each grid cell or small subwatershed realized by incorporating with a tightly-coupled hydrological model. The maps and databases are integrated using GIS data management tools, and the data sets and programs are integrated by applying the concepts of object-oriented 2

General Introduction programming (OOP). ArcView is selected as the host environment because it provides both spatial database management and OOP capabilities, while a procedural programming language, FORTRAN, is used to construct the hydrological model. Three study catchments with distinct basin characteristics are selected to investigate the applicability and adaptability of the model. They are: (1) Barebeek, a 67.8 km2 suburban plain watershed of the Dijle River basin situated northeast of Brussels, Belgium; (2) Steinsel, a 407 km highly urbanized hilly watershed located in the upstream part of the Alzette River basin, Grand Duchy of Luxembourg; and (3) Margecany, a 1133 km2 upland mountainous watershed situated in the upstream part of the Hornad River basin, Slovakia. Model applications are illustrated in the above study areas with an interest in flood forecasting, water balance simulation, the impact of land use change and natural river restoration on flooding behaviours, and so on.

2.

Objectives of the research

The overall goal of this research is to develop and apply a WetSpa Extension, which is a long-term, continuous simulation, physically-based, distributed parameter watershed model for flood prediction and watershed management. The specific objectives of this research regarding model development are to: Develop a comprehensive GIS-based modelling approach being compatible with GIS technology and remote sensing information, and by using the spatial information associated with the DEM, land use and soil type of the river basin for flood prediction and watershed management; Create a software package by integrating GIS with hydrological modelling, which contains a user-friendly interface allowing the model to operate in a GIS ArcView framework, a modelling database containing all lookup tables for determining model parameters, model programs, and other necessary components; Develop a practical method for flow routing in GIS-based modelling that enables to calculate response functions between any start and end point, depending upon slope, flow velocity and dissipation characteristics along the flow lines; Enable the use of the model for simulation of flow hydrographs at any location in the stream network and the spatial distribution of hydrological processes, such as runoff, soil moisture, evapotranspiration, groundwater recharge, etc.;

Chapter I Investigate the sensitivity of model predictions to identify those model parameters requiring the most careful estimation, and establish effective model evaluation criteria for assessing the model performance; Provide for a distributed model that can operate on a different spatial and temporal scale, and enable the use of the model for assessing the impacts of land use change, climate change, natural river restoration, etc., on the basin hydrological processes; Provide a platform on which the future water quality and soil erosion models can be developed at multiple scales. To demonstrate the use of the model for flood prediction and watershed management, the following specific objectives are addressed in different chapters: Simulate flood hydrographs and water balance on complex terrain; Estimate storm runoff contributions from different land use areas; Assess the impacts of land use change on flood behaviours; Assess the impacts of river restoration on the reduction of downstream flooding; Apply the model in a mountainous catchment accounting for snow and topographic adjustment for the input weather data.

3.

Outline of the dissertation

In the following chapters, the targets listed above are achieved by incorporating case studies to each of the specific subjects. The main model equations are presented in Chapter III and Chapter IV, while these equations are omitted in the following chapters to avoid unnecessary repetition. The detailed descriptions of the model, including model structure, formulation, data preparation, parameter identification, model calibration and evaluation, etc. are presented in Appendix A: Documentation and user manual of WetSpa Extension. A CD-ROM is prepared to install the full package of the WetSpa Extension operating under ArcView environment, which contains the sample project, Avenue scripts, lookup tables, help files, FORTRAN programs, and essential testing data. In Chapter II, a general literature review is conducted on the subject of GIS-based hydrological modelling. It begins with a brief review of the lumped and distributed

General Introduction hydrological modelling and the advantages of one type of modelling over the other are given. Next, the influences of watershed characteristics, i.e. topography, soil type and land use on runoff generation and flow routing are discussed, in order to stress the importance and necessity of distributed hydrological modelling. The application of GIS in hydrological modelling is highlighted including the application in watershed description, watershed interpretation, and the integration with hydrological models. A short description about some popular distributed and semi-distributed hydrological models linked with GIS is also given in this chapter. At the end, an overview of the WetSpa model is presented including the model history, assumptions and the improvement made for the WetSpa Extension. In Chapter III, a GIS-based diffusive transport approach for the determination of rainfall runoff response and flood routing through a catchment is developed. In the mean time, the sensitivity analysis for the effects of hydraulic radius, channel roughness, minimum slope, area-threshold in delineating channel networks, etc., on the flow response is discussed in this chapter. In Chapter IV, the application of WetSpa for the flood simulation in the Barebeek catchment, Belgium, is presented. This chapter gives a brief introduction at first about the concepts of the WetSpa model and the procedures of GIS implementation to derive model parameters for a complex terrain. Calibration and simulation results using the measured and the intensity-duration-frequency (IDF) data based on the analysis of historical records are presented afterwards. In Chapter V, an assessment of land use impacts on flood processes in a complex terrain is conducted for the Steinsel sub-basin of Alzette River basin, Grand-Duchy of Luxembourg. The assessment focuses on the runoff contributions from different land use classes and the potential impact of land use changes on runoff generation. Three types of possible land-use scenarios are developed and their effect on flood processes is investigated using the WetSpa distributed hydrology model. In Chapter VI, a flood reduction method by using conceptual river restoration is proposed. The scenarios are constructed considering the effects of increasing the flow resistance and re-meandering of the first and second order streams of the Alzette 5

Chapter I River basin, Grand-Duchy of Luxembourg, upstream of the Steinsel station. WetSpa model is applied to estimate in a scientific way the possible beneficial effect of river restoration on flood reduction in the main channels. In Chapter VII, the WetSpa model is applied to an 1133 km2 Carpathian watershed, Margecany, situated in the upstream part of the Hornad River basin, Slovakia. A simple snowmelt model is developed and embedded with the WetSpa model running on a daily scale. In addition, an automated calibration procedure by incorporating a model independent parameter estimator PEST is developed serving as an optimization algorithm to estimate the model parameters. Moreover, the topographic adjustment of input data, i.e. temperature, precipitation and potential evapotranspiration (PET), in a mountainous catchment is discussed. In Chapter VIII, a method of integrating GIS and process modelling for runoff prediction in medium and large river basins is presented based on a case study for the Alzette River basin, Grand-Duchy of Luxembourg. The chapter starts with a discussion of the grid size effects on runoff and flow responses. Next, the approach of transforming WetSpa into a semi-distributed model is proposed, and the procedures to derive model parameters together with the simulation results are presented. In Chapter IX, the summary and conclusions of this research are provided, in which the technique developed and knowledge acquired from this research are described and evaluated together with some comments regarding model limitations and some possible future researches of the model.

General Introduction References Babd, L.E. & Moore, I.D., Landscape attribute and Geographical information Systems. In: Scale Issues in Hydrological Modelling, eds., J.D. Kalma and M. Sivapalan, 159-180, John Willey & Sons., 1995. Boyle, C.A., Lavkulich, L., Schreier, H., & Kiss, E., Changing in land cover and subsequent effects on lower Fraser basin ecosystems from 1827 to 1990, Environ. Manage., 21(2), 185-196, 1997. Goodchild, M.F., Haining, R.P., Wise, S. & 12 others, Integrating GIS and spatial data analysis: problems and possibilities, Int. J. Geogr. Inf. Syst. 6(5), 407423, 1992. McDonnell, R.A., Including the spatial dimension: Using geographical information systems in hydrology, Prog. Physical Geography, 20(2), 159-177, 1996. Wagener, T., Boyle, D.P., Lees, M.J, Wheater, H.S., Gupta, H.V., & Sorooshian, S., A framework for development and application of hydrological models, Hydrol. Earth Syst. Sc., 5(1), 13-26, 2001. Wang, Z., Batelaan, O. & De Smedt, F., A distributed model for water and energy transfer between soil, plants and atmosphere (WetSpa), Phys. Chem. Earth, 21(3), 189-193, 1997. Weng, Q., Modelling urban growth effects on surface runoff with integration of remote sensing and GIS, Environ. Manage., 28(6), 737-748, 2001.

Chapter II GIS-based hydrological modelling and watershed analysis

Abstract GIS with its upcoming advanced technology has been a great asset to the hydrological modelling and watershed analysis. In particular, digital elevation models together with soil and land use mapping are used in a number of sub-domains in hydrology. Many hydrological models developed in the past which were useful individually and can be combined in various applications. The goal of this chapter is to provide a brief review of the extensive literature that exists in the area of GIS-based hydrological modelling; to address the potential impacts of watershed characteristics, such as topography, soil type and land use, on runoff generation and flow response that can be assessed by using GIS; to outline the rational basis for the linkage between GIS and hydrological modelling; and to indicate the type of model that could be incorporated within GIS and which are best left as independent analytical tools linked to GIS for data input and display of results. Finally, an overview of the WetSpa model and its extension is presented at the end of this chapter.

1.

Introduction

Over the past decades, extensive studies have been carried out with the aim of analyzing and modelling the natural systems in respect to the processes of runoff generation as well as the related transport of water, solutes and sediments. However, all these studies have different temporal and spatial scales from single events in micro-scaled sub-catchments. Chow et al. (1988) offered a taxonomy of hydrological models based on the randomness (deterministic/stochastic), spatial variation (lumped/distributed; space-independent/space-dependent) and time variation (steady flow/unsteady flow; time independent/time correlated) thereby drawing the attention to the pivotal position of the spatial dimension explored by the several possible applications of linking GIS with the hydrological models. The aim of these studies is mostly the same, i.e. to represent the natural system in a more or less sophisticated

Chapter II mathematical description in order to match the observed system outputs. The problem hereby is to find the right model, which is appropriate for the particular scale and aim of the study. There exists a distinct disagreement in the scientific community on which kind of model is more appropriate for the simulation of natural processes. It is widely recognized that the natural systems are extremely complex and the inherent processes are non-linearly connected and sometimes even characterized as being a chaotic system. Simple lumped models treat the hydrological system as spatially averaged and homogeneous in space with a small amount of calibration parameters for adapting the model. These models have only in part the possibility or very little opportunity for parameterization with respect to basin characteristics. Therefore measurements of the system output for an adequate period are a prerequisite for their application. However, lumped models have the disadvantage that the possibility of forecast simulations for un-gauged sites is limited and no learning effect of the inherent processes of the catchment is achieved. On the other hand, physically-based distributed models are used to offset the mentioned disadvantage of the simple models. The process description is derived from process studies at the scale of elementary spatial unit, such that the whole basin is partitioned into smaller cell elements. For each model cell the hydrological parameters and processes are assumed to be homogenous. One of the drawbacks of spatially distributed hydrological models is their requirement of detailed spatial data of the basin and of the climate. However, the availability of spatially distributed data sets (DEM, land use, soil, etc.) in the present time coupled with recent advances in computer hardwares and GIS software allows the spatial variation of model parameters and processes to be considered at a detailed resolution. Additionally, the spatio-temporal scales are also key considerations in selecting the model and in modelling the hydrological processes. The spatial scale for which a model is designed can play a significant role in how specific processes are treated. Therefore, it is an important criterion in the selection of a model because the storage characteristics may vary at different watershed scales, that is, large watersheds have well developed channel networks and channel phase, and thus, channel storage is dominant. Such watersheds are less sensitive to short duration, high intensity rainfalls. On the other hand, small watersheds are dominated by the land phase and overland 10

GIS-based hydrological modelling and watershed analysis flow, have relatively less conspicuous channel phase, and are highly sensitive to high intensity, short duration rainfalls. The temporal scale is important for modelling that operates from event to daily or even longer time scales. At the event time scale, models typically do not compute inter-storm soil moisture conditions and therefore this information must be provided as an initial condition to initiate the model run. On the other hand, continuous-time hydrological models can simulate the processes of precipitation, surface storage, snowmelt, evapotranspiration, soil moisture, and infiltration in a seasonal framework. These models typically operate on a time interval ranging from a fraction of an hour to a day. One of the advantages of continuous modelling is that it can provide a long-term simulation of the hydrological processes variability.

2.

Effects of watershed characteristics on runoff

Runoff is generated by precipitation during storm events and by groundwater entering surface channels. During dry periods, streamflows are sustained by groundwater discharges. Relations between precipitation and runoff are very complex, being influenced by many factors, such as interception, depression storage, infiltration, and evapotranspiration. Moreover, the quality and quantity of streamflow are strongly affected by its basins physical, vegetative, and land use features. Thus, basin and runoff are inter-related. These basin features are spatially distributed, and as a result there is a need for physically-based distributed hydrological models that are applicable to these environments. However, the robustness of such models would be enhanced if they were developed from a conceptual framework based on the physics of the dominant runoff processes.

2.1.

Effects of topography

Topographic features such as slope, curvature, and degree of convergence have an important impact on runoff production and the nature and strength of the flow connectivity. For instance, as the mean slope length becomes shorter, the time required to reach an effective channel decreases, leading to a steeper rising hydrograph limb and a higher peak discharge. Primary terrain attributes include slope, aspect, plan and profile curvature, flow-path length, and upslope contributing area. 11

Chapter II Table II-1: Primary topographic attributes that can be computed by terrain analysis from DEM data (Moore et al., 1991)
Attribute Altitude Upslope height Aspect Slope Definition Elevation Mean height of upslope area Slope azimuth Gradient Significance Climate, vegetation, potential energy Potential energy Solar isolation, evapotranspiration, flora and fauna distribution and abundance Overland and subsurface flow, velocity and runoff rate, precipitation, soil water content, vegetation, geomorphology, land capability Upslope slope Dispersal slope Catchment slope Upslope area Dispersal area Catchment area Specific drainage area Flow path length Mean slope of upslope area Mean slope of dispersal area Mean slope of the watershed Catchment area above a short length of contour Area downslope from a short length of contour Area draining to the catchment outlet Upslope area per unit width of contour Maximum distance of water flow to a point in the catchment Upslope length Dispersal length Catchment length Profile curvature Plan curvature Tangential curvature Elevation percentile Mean length of flow paths to a point in the catchment Distance from a point in the catchment to the outlet Distance from highest point to the basin outlet Slope profile curvature Contour curvature Plan curvature multiplied by slope Proportion of cells in a userdefined circle lower than the centre cell Flow acceleration, erosion and deposition rate, geomorphology Converging/diverging flow, soil water content, soil characteristics Provides alternative measure of local flow convergence and divergence Relative landscape position, flora and fauna distribution and abundance Overland flow attenuation Impedance of soil drainage Flow acceleration, erosion rates Runoff volume, steady-state runoff rate, soil-water content, geomorphology Erosion rates, sediment yield, time of concentration Runoff volume Soil drainage rate Runoff velocity Rate of soil drainage Time of concentration Runoff volume, steady-state runoff rate

12

GIS-based hydrological modelling and watershed analysis Most of these topographic attributes can be calculated from the directional derivatives of a topographic surface. The primary topographic attributes that can be computed by terrain analysis from DEM data and their significance are listed in Table II-1 (Moore et al., 1991). The secondary attributes that are computed from two or more primary attributes, e.g. the topographic wetness indices (TWI), the stream power indices and the radiation indices, are important because they offer an opportunity to describe pattern as a function of process. Those attributes that quantify the role played by topography in redistributing water in the landscape and in modifying the amount of solar radiation received at the surface have important hydrological, geomorphologic, and ecological consequences in many landscapes. These attributes may affect soil characteristics, distribution and abundance of soil water, susceptibility of landscapes to erosion by water, and the distribution and abundance of flora and fauna. These topographic indices are used frequently in many topography-based hydrological models, as for example the TOPMODEL (Beven & Kirkby, 1979), where TWI is applied to characterize the spatial distribution and extent of zones of saturation and variable source areas for runoff generation.

2.2.

Effects of soil type

The ability of the soil to transmit and retain water influences the rate of runoff during a storm and the rate at which the soil dries out between storms. For example, sandy soils allow for more infiltration of rain water than do heavier clay soils. Therefore, the soil constituents are important to the accurate modelling of runoff, as well as soil erosion and pollutant transport. Meanwhile, the states of surface and root zone soil moisture reservoirs are key variables controlling surface water and energy balances. Soil moisture plays an important role in various hydrological processes acting over a range of spatio-temporal scales, like partitioning of rainfall into infiltration and runoff, partitioning of net radiation into sensible and latent heat. Exchange of moisture flux between land surface and atmosphere in the form of evapotranspiration is the major link of interaction between hydrological and atmospheric processes. Hence there is a need for accurate spatio-temporal representation soil moisture in the modelling of atmospheric and hydrological processes.

13

Chapter II The evolution and variability of soil moisture are affected by various factors, such as soil properties, vegetation, solar radiation, atmospheric condition, prevailing topography and general geomorphologic conditions. Numerical modelling of soil moisture is usually based on highly nonlinear Richards equation. For stability consideration most of the existing approaches solve the equation with a fully implicit approach and use pressure head that is continuous in both saturated and unsaturated zones as primary variable. However, soil water balance models are usually used to keep track of water content changes in the soil zone, which treats the soil zone as a control volume. This approach provides a consistent means for applying physical laws to hydrological systems (Chow et al., 1988). Using this approach, direct runoff, evaporation, and percolation are treated as losses from the hydrological system. In hydrological modelling, soil texture is often used as a descriptor of soil physical properties such as porosity, saturated hydraulic conductivity, soil matric potential, and pore size distribution index (Cosby et al., 1984). Conventionally, the textural classification of a soil is determined as a function of the mass ratios of the three textural separates, namely sand, silt, and clay. Within the USDA soil textural classification, soils with different percentages of sand, silt, and clay are assigned to 12 different classes. Although other descriptors such as horizon and structural size certainly influence the hydraulic parameters of soils, Cosby et al. (1984) perform a two-way analysis of variance of nine descriptors to conclude that soil texture alone can account for most of the discernible patterns in porosity, saturated hydraulic conductivity, soil matric potential, and pore size distribution index. Under given climatic and vegetation conditions the above soil-texture-dependent physical properties, through their influence on soil water movement and the energy state of the water in the soil column, determine the soil wetness values which in turn establish the water condition of the plant (Fernandez-Illescas, 2001).

2.3.

Effects of land use

The patterns of vegetation on land surface give areas with different runoff generating characteristics. The vegetation cover density and the spatial configuration will both affect the discharge from the hill slope. As the vegetation density increases, the average infiltration rate will increase, thus leading to a reduction in discharge. 14

GIS-based hydrological modelling and watershed analysis However, as the cover becomes more fragmented, there are a greater number of pathways from the runoff source areas to the channel base. This increases in connectivity and consequently increases hill slope discharges. Moreover, different land use types have different evapotranspiration rates, because different plants have different vegetation cover, leaf area indices, root depths and albedo. During storms, interception rates are different for different land use types. Although it is recognized, that interception losses represent a significant net addition to catchment evaporative losses (Ward & Robinson, 1990), the influence of interception is noticeable only during small storms and influences only surface runoff rates. For largest storm and flood events, the interception losses are of minor importance (Calder, 1993). Land use also influences the infiltration and soil water redistribution process, because especially saturated hydraulic conductivity is influenced by plant roots and pores resulting from soil fauna (Ragab & Cooper, 1993). An extreme example is the influence of build up areas and roads on overland flow. Finally, land use and land management influences surface roughness, either by the land use type itself or by its management, which affects the overland flow velocity and floodplain flow rate. The hydrological effects of land use changes have been thoroughly described by Ward and Robinson (1990) and Calder (1993). The major changes in land use that affect hydrology are afforestation and deforestation, the intensification of agriculture, the drainage of wetlands, road construction, and urbanization. Of all the land use modifications, urbanization is by far the most forceful, by which land is transformed from its natural state or from agricultural use to an economically developed or populating region. This process can take many forms including irrigation, drainage, deforestation and logging, and urban development, all which result in numerous adverse effects on the water quality and quantity of surrounding terrestrial and aquatic ecosystems. The most significant of these effects is the alteration of the hydrological cycle and rainfall-runoff transformation of the watershed, including (1) changes in peak flow characteristics, (2) changes in total runoff, (3) changes in quality of water, and (4) changes in the hydrological amenities (Leopold, 1968). As the watershed becomes more developed, it also becomes more hydrologically active, changing the streams flow components as well as the origin of flow. Additionally, urbanization tends to increase both the flood volume and the flood peak.

15

Chapter II 3. GIS applications in watershed modelling

GIS is a software and hardware tool applied to geographical data for integration of collection, storing, retrieving, transforming and displaying spatial data for solving complex planning and management problems. It integrates common database operations including (1) database and data management operations, (2) time and spatial analysis functions, (3) image elaboration and filtering possibilities, (4) data merging and informative layers management, (5) cartographic and display functions, and (6) map realization and data presentation. GIS has the advantage of handling attribute data in conjunction with spatial features, which was totally impossible with manual cartographic analysis. The availability of GIS technology greatly enhances the capabilities for land description and interpretation by means of powerful distributes indicators, and therefore has allowed, and sometimes imposed, significant changes in the general approach to hydrological investigation and to operative hydrology since its development.

3.1.

Watershed description

Watershed characteristics relevant to hydrological investigations can be easily stored and handled by GIS, adding layers such as soil type and land use to the topographic database. The GIS produces a digital model of the basin, as detailed as required, which is easily readable by means of the appropriate software. Thus, the hydrologist can account for not only the land characteristics but also their distribution and individual localization in space. The input digital data are normally retrieved from external sources such as numerical cartography, aerial photography, satellite images, and digitalization of maps. The hydrologist usually deals with distributed objects as land use, soil type, hydrographical network, and hydraulic infrastructures. Once the databases of geo-referenced objects are setup the work begins by analyzing data, extracting information, producing synthetic maps by logical and algebraic operations, and running distributed models. The terrain complexity can be better represented by means of vectorial information, using points, polylines and polygons. Nevertheless, dealing with such geometry is still often too difficult when overlying or elaborating maps. So it is usually preferable to 16

GIS-based hydrological modelling and watershed analysis use raster maps in the analysis operations, which is rougher but easy to manage. However, an excellent degree of precision can be achieved by simply reducing the pixel dimension within reasonable levels. The morphological maps of elevation, slope and aspect are the basic informative layers, especially when dealing with mountain basins where the morphology plays a crucial role in hydrological response. Many algorithms exist to transform elevation contours into a raster DEM. The capability of DEM in term of land description is not limited to the elementary topographic attribute. A good DEM makes it possible to recognize the channel network by means of software, providing a suitable framework for routing modelling approach. The distinction between slope and channel paths can even be achieved simply by fixing a threshold drainage area for which the flow concentration is sufficient to initiate a channel (Tarboton et al., 1991; Da Ros & Borga, 1997). The use of soil and land use maps is generalized as both soil and land use strongly affects the hydrological behaviour of a single land unit, and specifically it is an effective indicator of potential direct flow generation. The land use map can be derived from remotely sensed imagery, or generated by an aerial view, particularly in terms of number of classes. The soil map, when not available, should be realized by expensive field techniques. However, in most cases the local land use, sometimes combined with a geologic map, can suggest some indications on soil characteristics.

3.2.

Hydrological parameter determination

A very active area of research is the use of GIS for model parameter estimation, in which local interpretations of the phenomena that exhibit over a watershed are provided by means of simple lookup tables or complex algorithms in combination with different kind of digital information. This role of GIS can be very beneficial for distributed parameter models which require large amounts of data. Special tools added to any commercial GIS software are able to compute powerful distributed indicators. The well-known Curve Number of the U.S. Soil Conservation Service, whose values are based on land use and soil permeability maps, is actually an indicator of potential direct flow generation. Specifically, primary terrain attributes like elevation, slope, aspect, shape of profile, upslope drained area, can be combined to give compound attributes, giving the dominant role of topography in hydrological processes control 17

Chapter II and landscape evolution. A well-known compound indicator is the topographic steady state wetness index (WI), defined, for each pixel, as the logarithm of the rate between the upslope drained area and the local slope (Western et al., 1999). Studies have shown the capabilities of slope and contributing drainage area to account topographic control for potential erosion/deposition in complex terrain (Mitasova et. al., 1996; De Roo, 1998). The drainage contributing area reflects the occurrence of flow concentration. The local slope gives account for the action of gravity force and so directly affects the flow velocity and the shear stress exerted by water on soil surface. Energy for surface processes as evapotranspiration and snow melt is provided by solar radiation, which is affected by local terrain morphology. With the model of the terrain surface provided by the GIS, Cazorzi and Dalla (1996) proposed a method to produce a potential energy distribution map, representing an indicator of the maximum solar radiation that can reach each pixel of the surface in a given time interval, without regard to weather conditions. Moreover, studies have shown the success of estimating SCS curve numbers using GIS by incorporating the land use and soil type data layers with the DEM (Hjelmfelt, 1991). Likewise, a potential runoff coefficient map can be generated based on the digital information of slope, soil type and land use (De Smedt et al., 2000; Liu et al., 2002). Maidment (1993) suggested a unit hydrograph technique that could combine the advantages of distributed modelling with the power of GIS. In his study, the time-area curve method was used to develop the synthetic unit hydrograph, while the travel time from each cell to the watershed outlet was calculated by dividing each flow length by a constant velocity. Liu et al. (2003) proposed a diffusive transport approach for flow routing in GIS-based watershed modelling, in which the unit response function of the grid cell is calculated based on the spatially distributed travel time and its standard deviation. Differing from the method of isochronal lines based only on the distance from the watershed outlet, these GIS models can route the runoff over the elevation surface and account for differences in runoff velocity due to changing slope, land use, and surface conditions.

3.3.

Integration with hydrological models

The ongoing development of GIS and associated databases offers the opportunity to simulate hydrological processes on a watershed scale using more physically based 18

GIS-based hydrological modelling and watershed analysis approaches than in the past. Three approaches exist for integrating hydrological modelling with GIS environment (Figure II-1): un-coupled integration, loosely coupled integration and tightly coupled integration (Kopp, 1996). In un-coupled integration, the GIS are used to process the spatial data into the desired model inputfile format and post-processing the model output. In loosely coupled integration, model input and output can be addressed directly by the GIS. Finally, in tightly coupled integration the model is written in an integrated programming language such as ArcView avenue language and Arc/Info AML language.

Figure II-1: GISHydrological modelling integration methods (Matson, et al., 1995) Parameters calculation for traditional lumped or semi-distributed models, when linked to a GIS, becomes much quicker and easy to change and test for a better solution. This kind of deterministic conceptual models is still very used in operative hydrology, as their behaviour is almost well known and, in most cases, they provide sufficient results. In this case the model and the GIS are essentially autonomous, and although the GIS potentiality is far to be completely exploited, it offers undoubtedly a good value added. The distributed models normally deal with square elementary land units, corresponding to raster pixels, and parameter computation, for each land unit, can be practically done through coupling with a GIS (Maidment, 1993, Wimgosta et al., 1994). Some of these models have already shown to be very important in the

19

Chapter II evolution of hydrological sciences, either for their conceptual structure, like for instance the physically based SHE (Abbott et al., 1986a, b), or for their convenience and ease of use, as for instance the TOPMODEL (Beven & Kirkby, 1979). Recent research developments brought to fully distributed hydrological models embedded into a GIS. In these cases a single database is used rather than two separate ones and the user interacts solely with the GIS, as the model is seen as a GIS function and is driven by a more or less customizable GIS menu. The leading idea of embedded models is to keep the model simple while exploiting the GIS power. An example is the LISFLOOD model, which is programmed in a dynamic GIS language called PC Raster used to simulate floods in large European drainage basins (De Roo et al., 2000). The model is defined by a regular horizontal grid of user-defined size typically varying between a few hundred meters for smaller catchment or subcatchment simulations and 5 km for simulations on European scale. Embedded models in a GIS environment have many advantages over traditional loosely GIScoupled models. However, current GIS is lacking of tools for developing physicallybased models. Especially simulating transport of water and pollutants through landscapes is a problem in a GIS environment. The development of GIS systems to handle all calculations will be an important step in hydrology, for which all aspects related to hydrological modelling can be integrated into a comprehensive GIS system, so that better simulations can be obtained, and therefore better management can be achieved. Unfortunately this approach currently remains out of reach for most practical applications due to data limitations and the lack of proven software.

4.

Modelling of watershed hydrology

Hydrological models are integral components of water management and monitoring strategies. In particular, distributed hydrological models allow for detailed description of the hydrological and energy cycle and provide opportunities for dealing with forcing variables that fluctuate strongly in space and time, such as precipitation. Hydrologists increasingly implement these models as a means to apply the state of knowledge on basins of interest, and provide valuable information regarding hydrological state variables and potentially important distributed information on existing and future streamflow conditions. Also, there is increasing interest in using 20

GIS-based hydrological modelling and watershed analysis spatially distributed meteorological data from diverse sources such as environmental satellite and weather radar. GIS provides representations of these spatial features of the Earth, while hydrological modelling is concerned with the flow of water and its constituents over the land surface and in the subsurface environment. Singh and Woolhiser (2002) provide a general review of mathematical modelling of watershed hydrology. Vieux (1991) presents a review of water quantity and quality modelling with GIS and, as an application example, employed the kinematic wave method to an overland flow problem.

4.1.

Popular GIS-based hydrological models

The employment of GIS and remotely sensed data in watershed modelling is one of the most important recent advances in hydrology. Over the last decade, there has been tremendous development in hydrological modelling using GIS. Most of the hydrological models are mathematically based, where they integrate existing knowledge into a logical framework of rules and relationships. GIS technology has been integrated with these surface or subsurface hydrological models emphasizing the utility and significance of topographic attributes of the terrain for various hydrological applications. The reason of adopting GIS technology is because it allows the spatial information to be displaced in integrative ways that are readily comprehensible and visual. The spatial information collected is further subjected to continuous GIS analysis, providing the opportunity for realistic representation of the natural landscapes under the constraints of maintaining physical consistency. Hydrological models with a spatial structure are being increasingly based on DEM or DTM (Moore et al., 1991). Many of the existing models, such as SHE, TOPMODEL, etc., have been adapted to the new type of data that can be processed by GIS software. Integration of hydrological models with remotely sensed, GIS, and DEM-based data has started to occur. Examples of newly developed or adapted models are those by Fortin et al. (2001a,b), Wigmosta et al. (1994), Julien et al. (1995), Desconnets et al. (1996), Olivera and Maidment (1999) and De Roo et al. (2000). Table II-2 lists some samples of popular distributed and semi-distributed hydrological models that support GIS and remote sensing applications.

21

Chapter II Table II-2: Samples of popular distributed and semi-distributed hydrological models
Model name gical Modelling System (HEC-HMS) Semi-distributed Land Use-based Runoff Processes (SLURP) Systeme Hydrologique Europeen/Systeme Hydrologique Europeen Sediment (SHE) Institute of Hydrology Distributed Model (IHDM) (TOPMODEL) Kinematic Runoff and Erosion Model (KINEROS) teme Hydroloque Europeen (MIKE-SHE) Waterloo Flood System (WATFLOOD) Dynamic Watershed Simulation Model (DWSM) Hydrological Model System (HMS) Hydrological Modelling System (ARC/EGMO) Hydrological Simulation (HBV) Model Distributed Hydrology Soil Vegetation Model (DHSVM) (SHETRA) Author(s) (1981) Kite (1998) Abbott et al. (1986a,b) Calver & Remarks Physically-based, semidistributed, eventbased, runoff simulation model Process-oriented, semidistributed, continuous stormflow simulation model Physically based, distributed, continuous flow and sediment simulation model Physically based, distributed, continuous Physically based, distributed, continuous Physically based, semidistributed, eventbased, runoff and water quality model Physically based, distributed, continuous, Process-oriented, semidistributed continuous flow simulation model Process-oriented, event-based, runoff and water quality simulation model Physically based, distributed, continuous hydrological simulation system Process-oriented, distributed, continuous Process-oriented, distributed, continuous streamflow simulation model Distributed, physically based, continuous hydrological simulation model Physically based, distributed, water quantity and quality simulation model Physically based, distributed, event-based runoff simulation model Physically based, distributed, continuous hydrological simulation model Physically based, distributed, continuous runoff simulation model on large scale Physically based, distributed, continuous hydrological simulation model Distributed, conceptual, continuous simulation model Hydrological Engineering Centre, Hydrolo- Feldman

Wood (1995) rainfall-runoff modelling system Kirkby (1979) hydrological simulation model Smith et al. (1995)

Physically Based Runoff Production Model Beven &

Generalized River Modelling Package, Sys- Refsgaard & Kouwen (1988) Borah et al. (2002) Yu et al. (1999) Pftzner & Lindstrm et al. (1997) Wigmosta et al. (1994) (2000) (1995) Geomorphology-Based Hydrology Simulation Model (GBHM) Physically-Based River Basin Modelling System (LISFLOOD) Distributed Hydrological Model (HYDROTEL) Soil Water Assessment Tool (SWAT) Yang et al. (1998) De Roo et al. (2000) Fortin et al. (2001a, b) Arnold et al. (1998)

Storm (1995) hydrological & hydraulic model

Becker (1995) simulation system

Systeme Hydrologique Europeen Transport Ewen et al. Cascade two dimensional Model (CASC2D) Julien et al.

22

GIS-based hydrological modelling and watershed analysis 4.2. Assessment of future scenarios

The understanding of global climate systems has considerably increased in recent years, as well as human concern about future global climatic changes. Along with these changes important consequences are expected in regional hydrological cycles and subsequent effects on regional water resources. Since hydrological processes directly depend on climate conditions, influences of possible climatic changes on these processes will differ from region to region. The magnitude and spatial distribution of the climatic changes in combination with hydrological characteristics of the study region determine which effects will be most relevant on the regional scale. Though the direction or magnitude of many important changes are not yet fully clear, studies in the last years have shown important regional vulnerabilities against changes of both temperature and precipitation patterns. They suggest that climatic changes will alter basic components of the hydrological cycle like soil moisture, groundwater availability, magnitude and timing of runoff, and water quality, which would induce dramatic environmental dislocations and widespread implications for future water resources planning and management (Lahmer et al., 2001). Land use change has been identified as a major driving force for global change, which may induce comparable effects on water quantity and quality. These changes are the result of natural processes as well as anthropogenic influences, which include such processes as vegetation dynamics, erosion, acidification, salinization, overgrazing, desertification, mining, urban and industrial development, conversion of lands into agriculture, and deforestation, etc. it is desirable that the hydrological modelling describes the spatio-temporal variability of land use effects so that the assessment could reflect the variability of the hydrological parameter at the required scales (Parkin et al., 1996). Though the problems and consequences resulting from land use change are clearly defined, clear solutions and practical applications are still challenging to the hydrologist. One of the most important reasons is the difficulty of parameterization of the physiographic properties and state variables of a basin into a model (Naef et al., 2002). In recent years, river restoration, rehabilitation and other environmentally preferable methods have become important issues in river engineering practice. The goal of 23

Chapter II stream restoration is to restore the stream to a more natural form to create environmentally favourable conditions, which do not necessarily imply that the stream will be restored to its pre-settlement condition (Morris, 1995). Besides the benefits on the ecology and morphology of a river basin, river restoration activities in the headwater streams may have a positive effect on flood reduction for the main river channels in the river basin. In this respect, distributed hydrological models based on GIS techniques have advantages in modelling and assessing the effects of river restoration on flooding dynamics based on reliable restoration scenarios.

5.

WetSpa model overview

5.1.

Model history

WetSpa is a physically based distributed hydrological model for predicting the Water and Energy Transfer between Soil, Plants and Atmosphere on regional or basin scale proposed by Wang et al. (1997). The model conceptualizes a basin hydrological system being composed of atmosphere, canopy, root zone, transmission zone and saturation zone. The basin is divided into a number of grid cells in order to deal with the heterogeneity of catchment characteristics. Each cell is further divided into a bare soil and vegetated part, for which the water and energy balance are maintained. Figure II-2 shows schematically the considered hydrological processes for a grid cell.

long w ave radiation sensible heat latent heat bare soil

short and long w ave radiation

precipitation evapotranspiration surface runoff

e on

heat to ground

infiltration percolation recharge

Figure II-2: Hydrological processes considered in the original WetSpa model

24

ca tra pil lar ns im y f ri iss ng ion e z

z ot ro

on e

groundw ater flow

r tu sa

ion at

ne zo

GIS-based hydrological modelling and watershed analysis Water movement in the soil is simplified as one-dimensional vertical flow, including surface infiltration, percolation and capillary rise in the unsaturated zone and recharge to groundwater. The model was designed to simulate the Hortonian overland flow and the variable source area concept of runoff generation. In order to have a more realistic representation of the interaction between surface runoff and groundwater storage, a groundwater flow model is integrated, for which the groundwater balance in the saturated zone is described by the two-dimensional Dupuit-Forchheimer horizontal flow equation. Under appropriate boundary conditions the water table position is determined with a finite difference scheme for each grid cell, and explicitly for each time step. The model was designed for scientific research with time resolution of minutes. Due to the complex model structure and the limit of available data, the model is difficult to be implemented for an engineering purpose. For the estimation of long-term spatial patterns of the groundwater recharge, that could be used as input in regional groundwater flow models and for the analysis of regional groundwater flow systems, a simplified model WetSpass was developed by Batelaan & De Smedt (2001) based on WetSpa. WetSpass stands for Water and Energy Transfer between Soil, Plants and Atmosphere under quasi-Steady State conditions, which is GIS based, spatially distributed hydrological model for calculating the spatially distributed yearly and seasonal evapotranspiration, surface runoff, and groundwater recharge. The model accounts for the spatial variation in the groundwater recharge, which is the result of distributed land use, soil type, and slope, etc. The total water balance for a cell in a spatially distributed grid is split up in independent water balances for vegetated, bare-soil, open-water and impervious parts of the grid cell. This allows accounting for the non-uniformity of the land use depending on the resolution of the grid cell. WetSpass model operates on a seasonal basis, and the flow processes are not included.

5.2.

WetSpa Extension

The WetSpa Extension, developed during my PhD research, is a physically-based distributed, continuous hydrological model compatible with remote sensing and GIS data, which is capable of predicting outflow hydrographs at basin outlet or any converging point in a watershed with a variable time steps (De Smedt et al., 2000; Liu 25

Chapter II et al., 1999, 2002, 2003). The model aims not only at predicting flood, but also investigating the reasons behind it, especially the spatial distribution of topography, land use and soil type. Compared with the originally proposed WetSpa model, major changes involved in this extension are: 1) The time resolution of simulated hydrological processes is changed to a variable time scale (minutely, hourly, daily, etc.). The spatial resolution is also variable to allow adequate interpretation of the landscape. The spatio-temporal scales of the model are determined according to the project purpose, the size and complexity of the catchment, data available, and computer speed and memory. 2) The Thiessen polygon method is applied to address the spatial distribution of input time series, i.e. precipitation, temperature and PET, for the current version. Topographical corrections for the input time series are considered in modelling a mountainous catchment if the necessary data are available. 3) A moisture-related runoff coefficient method is developed for calculating surface runoff in each grid cell, which allows the actual runoff coefficient to vary in time, and in function of rainfall intensity, rainfall duration and cell characteristics (slope, soil type and land use), giving an approximation to the surface runoff volume at each time step. 4) The modelling of snow accumulation and snowmelt is added in the WetSpa model capable of predicting snowmelt flood by a simple degree-day approach, in which the temperature lapse rate for each grid cell is determined based on the DEM and the elevation of reference station. 5) The hydrological processes of interception and depression storage are considered in the WetSpa model allowing the estimation of initial abstraction for a storm event. The parameter of interception storage capacity is determined based on the land use map and varies with season, while the depression storage capacity is determined based on the cells slope, soil type and land use. 6) The process of shallow subsurface lateral flow is considered in the WetSpa model allowing the simulation of interflow and return flow from adjacent upslope cells by the method of Darcy's law and kinematic approximation. The subsurface lateral flow is assumed to occur when soil moisture is higher than field capacity and join the overland flow at each grid cell. However, the routing of subsurface lateral flow to the downhill neighbours is not performed for model simplicity.

26

GIS-based hydrological modelling and watershed analysis 7) An effective hydraulic conductivity of the soil is specified based on the soil texture class, which limits the rate at which water can percolate out of the root zone. The percolated water is assumed to recharge the groundwater reservoir within the same time step. 8) Evapotranspiration from root zone is calculated for each cell as a function of PET and moisture content in the cell. A part of water is extracted for transpiration from groundwater storage when PET exceeds the evapotranspiration from soil depending upon the amount of groundwater storage. Actual evapotranspiration is then the sum of the evaporation from interception and depression storage and the evapotranspiration from soil and groundwater storage. 9) The simulation of groundwater balance and baseflow is performed on small subcatchment scale by the linear or non-linear reservoir method for the simplification of model parameterization. Baseflow is added to the streamflow producing a total-runoff hydrograph at the subcatchment outlet. 10) A flow routing model for both overland flow and channel flow is incorporated in the WetSpa model using the method of linear diffusive wave approximation. The unit response function for each grid cell is calculated based on the mean and variance of the flow time distribution, which is derived from the advectiondispersion transport equation. The flow velocity is location dependent and calculated for each cell by the Manning equation based on the local slope, roughness coefficient and hydraulic radius. The hydraulic radius is determined according to the geophysical properties of the river basin and the flood frequency. 11) An automated calibration procedure is applied to the WetSpa model by incorporating a model-independent parameter estimator PEST (Doherty & Johnston, 2003) to estimate the most sensitive parameters of the model with observed flow hydrographs as the calibration target. This scheme serves as an optimization algorithm to estimate the model parameters. 12) 5 criteria are set up for statistically evaluating the model performance, namely model bias, model confidential coefficient, Nash-Sutcliffe efficiency (Nash & Sutcliffe, 1970), logarithmic transformed Nash-Sutcliffe efficiency for low flow evaluation (Smakhtin et al., 1998), and an adapted version of Nash-Sutcliffe efficiency for high flow evaluation (Guex, 2001). 13) Some model formulas are modified in order to make the model more physically based and capable of using readily available data, for instance the equation to 27

Chapter II calculate water percolation out of the root zone and the equation to calculate evapotranspiration from soil layer. Besides, all default parameter values in the model lookup tables are recalibrated based on the literature review and practical case studies, and some new lookup tables are established, such as hydraulic radius, interception and depression storage capacity, etc. 14) Model programs are developed, which make use of spatial inputs and give spatial outputs as well. The GIS part of the model is processed using ArcView Avenue script together with functions in the ArcView Spatial Analyst extension. The programming of hydrological simulation is developed using FORTRAN language. A user friendly ArcView interface which integrates GIS with hydrological modelling is also developed.

6.

Summary

The rainfall-runoff relationship is one of the most complex hydrological phenomena due to the tremendous spatio-temporal variability of the watershed characteristics and unpredictable rainfall pattern. However, to address the effects of spatial variability of model parameters such as land-use changes, distributed hydrological models have the advantage as they can explicitly consider spatial variability of parameters. With the development of GIS and remote sensing techniques, the hydrological catchment models have been more physically based and distributed to enumerate various interactive hydrological processes considering spatial heterogeneity. A GIS-based hydrological model, WetSpa Extension, has been developed for use in flood prediction and watershed analysis. It accounts for the effects of topography, soil and land use on runoff in a spatial way operates within a GIS framework. The model automates the process of converting commonly available GIS data to input parameter files for the hydrological modelling extension. Spatial results from the model, such as runoff, groundwater recharge, soil moisture, etc., can be imported into GIS and investigated using ArcView visualization tools. Since the models operate at different spatial and temporal scales, it provides the capability to assess the spatial distribution of the impacts of catchment characteristics, such as land use and land cover change, on watershed hydrological responses.

28

GIS-based hydrological modelling and watershed analysis References Abbott, M.B., Bathurst, J.C., Cunge, J.A., OConnell, P.E. & Rasmussen, J., An introduction to the European Hydrological System Systeme Hydrologique Europeen, SHE, 1: History and philosophy of a physically based, distributed modelling system, J. Hydrol., 87, 45-59, 1986a. Abbott, M.B., Bathurst, J.C., Cunge, J.A., OConnell, P.E. & Rasmussen, J., An introduction to the European Hydrological System Systeme Hydrologique Europeen, SHE, 2: Structure of a physically based, distributed modelling system, J. Hydrol., 87, 61-77, 1986b. Arnold, J.G., Srinivasan, R., Muttiah, R.S. & Williams, J.R., Large area hydrological modelling and assessment, Part I: Model development, J. Am. Water Resour. Ass., 34(1), 73-89, 1998. Batelaan, O. & De Smedt, F., WetSpass: a flexible, GIS based, distributed recharge methodology for regional groundwater modelling, eds., H. Gehrels et al., Impact of human activity on groundwater dynamics, 11-17, IAHS Publ. No. 269, 2001. Beven, K.J. & Kirkby, M.J., A physically-based variable contributing area model of basin hydrology, Hydrol. Sci. Bull., 24(1), 43-69, 1979. Borah, D.K., Xia, R. and Bera, M., DWSM A Dynamic Watershed Simulation Model, In: Mathematical models of small watershed hydrology and applications, 113-166, eds., V.P. Singh & D.K. Frevert, Water Resou. Publ., Littleton, Colo., 2002. Calder, I.R., Hydrologic effects of land-use change, ed., Maidment, D.R., Handbook of Hydrology, 50, McGrawHill, New York, 1993. Calver, A. & Wood, W.L., Chapter 17: The Institute of Hydrology distributed model, Computer models of watershed hydrology, V.P. Singh, ed., Water Resources Publications, Littleton, Colo., 1995. Cazorzi F. & Dalla, F.G., Snowmelt modelling by combining air temperature and a distributed radiation index. J. Hydrol., 181, 169-187, 1996. Chow, V.T., Maidment, D.R. & Mays, L.W., Applied Hydrology, McGrawHill, 1988. Cosby, B.J., Hornberger, G.M., Clapp, R.B. & Ginn, T.R., A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils, Water Resour. Res., 20(6), 682690, 1984.

29

Chapter II Da Ros D. & Borga M., On the use of digital elevation model data for the derivation of the geomorphologic instantaneous unit hydrograph, Hydrol. Process., 11, 1333, 1997. De Roo, A.P.J., Wesseling, C.G. & Van Deursen, W.P.A., Physically Based River Basin Modelling within a GIS: The LISFLOOD Model, Hydrol. Process., 14, 1981-1992, 2000. De Roo, A.P.J., Modelling runoff and sediment transport in catchments using GIS, Hydrol. Process., 12, 905-922, 1998. De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrological modelling on a catchment scale using GIS and remote sensed land use information, In: Risk Analysis II, ed., C.A. Brebbia, 295-304, WTI press, Southampton, Boston, 2000. Doherty, J., & Johnston, J.M., Methodologies for calibration and predictive analysis of a watershed model, J. Am. Water Resour. Ass., 39(2), 251-265, 2003. Ewen, J., Parkin, G. & OConnell, P.E., SHETRAN: Distributed river basin flow and transport modelling system, J. Hydrological Eng., 5(3), 250-258, 2000. Feldman, A. D., HEC models for water resources system simulation: Theory and experience, Adv. Hydrosci., 12, 297-423, 1981. Fernandez-Illescas C., Porporato A., Laio F. & Rodriguez-Iturbe I, The role of soil texture in water controlled ecosystems, Water Resour. Res., 37(12), 2663-2872, 2001. Fortin, J.P., Turcotte, R., Massicotte, S., Moussa, R., Fitzback, J. & Villeneuve, J.P., A distributed watershed model compatible with remote sensing and GIS data, I: Description of model, J. Hydrological Eng., 6(2), 91-99, 2001a. Fortin, J.P., Turcotte, R., Massicotte, S., Moussa, R., Fitzback, J. & Villeneuve, J.P., A distributed watershed model compatible with remote sensing and GIS data, II: Application to Chaudiere watershed, J. Hydrological Eng., 6(2), 100-108, 2001b. Guex, F., Modlisation hydrologique dans le bassin versant de lAlzette (Luxembourg), rgionalisation des paramtres dun modle global, Travail pratique de Diplme, Ecole Polytechnique Fdrale de Lausanne, 2001. Hjelmfelt, A.T., Investigation of curve number procedure, J. Hydraulic Eng., ASCE, 117(6), 725-737, 1991. Julien, P.Y., Saghafian, B. & Ogden, F.L., Raster based hydrological modelling of spatially varied surface runoff, Water Resour. Bull., 31(3), 523-536, 1995.

30

GIS-based hydrological modelling and watershed analysis Kite, G.W., Danard, M. & Li, B., Simulating long series of streamflow using data from an atmospheric model, Hydrol. Sci. J., 43(3), 391-407, 1998. Kopp, S.M., Linking GIS and Hydrological models: where we have been, where we are going? In: Application of geographic information systems in hydrology and water resources, eds., K. Kovar & H. P. Nachtnebel, 133-139, IAHS Publ No. 235, 1996. Kouwen, N., WATFLOOD: a micro-computer based flood forecasting system based on real-time weather radar, Can. Water Resour. J., 13(1), 62-77, 1988. Lahmer, W., Pftzner, B. & Becker, A., Assessment of land use and climate change impacts on the mesoscale, Phys. Chem. Earth, 26, 565-575, 2001. Leopold, L.B., Hydrology for urban land planning: A guidebook on the hydrologic effects of urban land use, U.S. Geol. Surv. Circ. 554, USGS, Washington, DC, 1968. Lindstrm, G., Johansson, B., Persson, M., Gardelin, M. & Bergstrm, S., Development and test of the distributed HBV-96 hydrological model, J. Hydrol., 201, 272-288, 1997. Liu, Y.B., A GIS-based spatially distributed hydrological modelling of the Barebeek catchment, MSc. Thesis, Inter-University Programme in Water Resources Engineering, Katholieke Universiteit Leuven and Vrije Universiteit Brussel, Belgium, 1999. Liu, Y.B., Gebremeskel, S, De Smedt, F, Hoffmann, L., and Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modelling, J. Hydrol., 283, 91-106, 2003. Liu, Y.B., Gebremeskel, S., De Smedt, F., and Pfister, L., Flood prediction with the WetSpa model on catchment scale, In; Flood Defence 2002, ed., Wu et al, pp. 499-507, Science Press, New York Ltd., 2002. Maidment, D.R., Developing a spatially distributed unit hydrograph by using GIS, In HydroGIS 93: Application of Geographic Information Systems in Hydrology and Water Resources, Proceedings of the Vienna Conference, eds., K. Dovar and H. P. Natchnebel, 181-192, Vienna: Int. Assoc. of Hydrol. Sci., 1993. Matson, K.C., Stallings, C., Jennings, G.D., McLaughlin, R.A., and Coffey, S.W., Watershed scale hydrological modelling in a GIS environment, GIS/LIS '95 Annual Conference Exposition, 14-16, Nashville, TN., 1995.

31

Chapter II Mitasova, H., Hofierka, J., Zlocha, M. & Iverson, L.R., Modelling topographic potential for erosion and deposition using GIS. Int. J. Geogr. Inf. Syst., 10(5), 629-641, 1996. Moore, I.D., Grayson, R.B. and Ladson, A.R., Digital terrain modelling: A review of hydrological, geomorphologic and biological applications, Hydrol. Process. 5, 330, 1991. Morris, S.E., Geomorphic aspects of stream-restoration, Phys. Geogr., 16(5), 444-459, 1995. Naef, F., Scherrer, S. & Weiler, M., A process based assessment of the potential to reduce flood runoff by land use change, J. Hydrol., 267, 74-79, 2002. Nash, J.E. & Sutcliffe, J.V., River flow forecasting through conceptual model, J. Hydrol., 10, 282290, 1970. Olivera, F. & Maidment, D., Geographical information system (GIS)-based spatially distributed model for runoff routing, Water Resour. Res., 35(4), 11551164, 1999. Parkin, G., ODonnel, G., Ewen, J., Bathurst, J.C., OConnell, P.E. & Lavabtr, J., Validation of catchment models for predicting land-use and climate change impacts, 2 Case study for a Mediterranean catchment, J. Hydrol., 175, 595-613, 1996. Pftzner, B. & Becker, A., ARC/EGMO Kurzdokumentation, Bro fr Angewandte Hydrologie, 70, Seiten, Berlin., 1995. Ragab, R. & Cooper, J.D., Variability of unsaturated zone water transport parameters: implications for hydrological modelling, 1. In situ measurements, J. Hydrol., 148, 109131, 1993. Refsgaard, J.C. & Storm, B., Chapter 23, MIKE SHE, Computer models of watershed hydrology, ed., V.P. Singh, Water. Resour. Publ., Littleton, Colo., 1995. Singh V.P. & Woolhiser D.A., Mathematical modelling of watershed hydrology, J. Hydrological Eng., 7 (4), 270-292, 2002. Smakhtin, V.Y., Sami. K. & Hughes, D.A., Evaluating the performance of a deterministic daily rainfall-runoff model in a low flow context. Hydrol. Process., 12, 797-811, 1998. Smith, R.E., Goodrich, D.C., Woolhiser, D.A. & Unkrich, C.L., Chapter 20: KINEROS - A kinematic runoff and erosion model, Computer models of

32

GIS-based hydrological modelling and watershed analysis watershed hydrology, V.P. Singh, ed., Water. Resour. Publ., Littleton, Colo., 1995. Tarboton, D.G., Bras, R.L. & Rodriguez-Iturbe, I., On the Extraction of Channel Networks from Digital Elevation Data, Hydrol. Process., 5, 81-100, 1991. Vieux, B.E., Geographic information systems and non-point source water quality and quantity modelling, Hydrol. Process., 5, 101-113, 1991. Wang, Z.M., Batelaan, O. & De Smedt, F., A distributed model for water and energy transfer between soil, plants and atmosphere (WetSpa), Phys. Chem. Earth, 21(3), 189-193, 1997. Ward, R.C. & Robinson, M., Principles of Hydrology, third edition, 365, McGrawHill, London, 1990. Western, A.W., Graysen, R., Bloschl, G., Willgoose, G. & McMahon, T.A., Observed spatial organization of soil moisture and its relation to terrain indices, Water Resour. Res., 35, 797-810, 1999. Wigmosta, M.S., Vail, L.W. & Lettenmaier, D.P., A distributed hydrology-vegetation model for complex terrain, Water Resour. Res., 30(6), 1665-1679, 1994. Yang, D., Herath, S. & Musiake, K., Development of a geomorphology-based hydrological model for large catchments, Ann. J. Hydraulic Eng., 42, 169-174, 1998. Yu, Z., Yarnal, B, Barron, E.J., Duffy, C. & Schwartz, F.W., Simulating the riverbasin response to atmospheric forcing by linking a mesoscale meteorological model and a hydrological model system, J. Hydrol., 218, 7291, 1999.

33

Chapter III Development of a diffusive transport approach for flow routing in GIS-based watershed modelling
Abstract A GIS-based diffusive transport approach for the determination of rainfall runoff response and flood routing through a catchment is developed. The watershed is represented as a grid cell mesh, and routing of runoff from each cell to the basin outlet is accomplished using the first passage time response function based on the mean and variance of the flow time distribution, which is derived from the advection-dispersion transport equation. The flow velocity is location dependent and calculated for each cell by the Manning equation based on the local slope, roughness coefficient and hydraulic radius. The hydraulic radius is determined according to the geophysical properties of the river basin and the flood frequency. The total direct runoff at the basin outlet is obtained by superimposing all contributions from every grid cell. The model is tested on the Attert catchment in Luxembourg with 30 months of observed hourly rainfall and discharge data, and the results are in good agreement with the measured hydrograph at the basin outlet. Sensitivity analysis shows that parameters of flood frequency and the channel roughness coefficient have large influences on the outflow hydrograph and the calculated watershed unit hydrograph, while the threshold of minimum slope and the threshold of drainage area in delineating channel networks have a marginal effect.

1.

Introduction

Flood prediction and catchment modelling are main topics facing the hydrologist dealing with processes of transforming rainfall into a flood hydrograph and the translation of hydrographs throughout a watershed. The theory of the unit hydrograph for the prediction of stream flow in a basin has played a prominent role in hydrology for several decades since its development. This system response theory assumes that the basin response to a rainfall input is linear and time invariant. The discharge at the

Chapter III outlet of the basin is given by the convolution of the rainfall input and the instantaneous unit hydrograph (IUH) (Dooge, 1959). In engineering practice, the unit hydrograph is often determined by numerical deconvolution techniques (Chow et al., 1988) using observed stream flow and rainfall data. Since the characteristics of hydrological systems, as for instance precipitation and the generation of runoff, are extremely variable in space and time, the response of the system, i.e. the flow of water over the land surface and the river channels, is a distributed process in which the characteristics of the flow change both in time and space. This limits the use of the unit hydrograph model. Consequently, in trying to relax the unit hydrograph assumptions of uniform and constant rainfall, and to account for spatial variability of the catchment, considerable research has been conducted in recent years, and many articles dealing with these topics can be found in the literature. In an attempt to find a physical basis for the IUH, Rodriguez-Iturbe and Valdes (1979) introduced the concept of a geomorphologic instantaneous unit hydrograph (GIUH), which relates the geomorphologic structure of a basin to the IUH using probabilistic arguments. This theory was later generalized by Gupta et al. (1980) and Gupta and Waymire (1983). In their paper, Horton's empirical laws, i.e. law of stream numbers, lengths and areas, are used to describe the geomorphology of the system. The IUH is defined as the probability density function (PDF) of the droplet travel time from the source to the basin outlet, in which the time spent in each state (order of the stream in which the drop is located) is taken as a random variable with an exponential PDF. The model is relatively parsimonious in data requirements and most parameters can be obtained from DEM data. Consequently, this theory has undergone several noteworthy developments over the last two decades. Mesa and Mifflin (1986) obtained their GIUH by means of the width function and the inverse Gaussian PDF. The width function is the frequency distribution of channels with respect to flow distance from the outlet. It is an approximate representation of the area function under the assumption of a uniform constant of channel maintenance throughout the drainage basin. Similar methodologies were presented by Naden (1992) and Troch et al. (1994). Sivapalan et al. (1990) incorporated the effect of partial contributing areas, which recognizes that during a rainfall event, droplets contributing to the runoff are 36

Development of diffusive transport approach for flow routing not uniformly distributed throughout the basin but are more likely to come from areas that are saturated close to stream channels. The saturated areas can be identified through topographic indices (Beven & Kirkby, 1979), which can be easily obtained from DEM data. Van Der Tak and Bras (1990) incorporated hillslope effects in the basic formulation of GIUH by using a gamma distribution for the travel time distributions through the flow pathways and introducing a hillslope velocity term. Using the method of moments, they found that hillslope velocities are two orders of magnitude smaller than channel velocities, which has a significant impact on the GIUH. To describe the flow through individual streams, Rinaldo et al. (1991) used an advection-dispersion equation, which is obtained by introducing a diffusion term in the kinematic wave equation. They showed that not only is there a dispersion effect in the individual channels, but that the stream network structure itself causes dispersion, which is described as geomorphologic dispersion. Snell and Sivapalan (1994) showed that the geomorphologic dispersion coefficient depends on the first two moments of the flow path lengths, with the assumption of a constant flow velocity and longitudinal dispersion throughout the catchment. Lee and Yen (1997) introduced the kinematic wave theory to determine the travel times of overland and channel flows, thus relaxing the linearity restriction of the unit hydrograph theory. Maidment (1993) proposed the promising concept of using GIS to derive a spatially distributed unit hydrograph (SDUH) that reflects the spatially distributed flow characteristics of the watershed. The SDUH is similar to GIUH, except that it uses a GIS to describe the connectivity of the links and the watershed flow network instead of probability arguments. The travel time from each cell to the watershed outlet is calculated by dividing each flow length by a constant velocity. Subsequently, a timearea diagram based on the travel time from each grid cell is developed. A more elaborate flow model, which accounts for both translation and storage effects in the watershed, is presented by Maidment et al. (1996). In his paper, the watershed response is calculated as the sum of the responses of each individual grid cell, which is determined as a combined process of channel flow followed by a linear reservoir routing. Olivera and Maidment (1999) proposed a method for routing spatially distributed excess precipitation over a watershed using response functions derived from a digital terrain model. The routing of water from one cell to the next is accomplished by using the first-passage-time response function, which is derived 37

Chapter III from the advection-dispersion equation of flow routing. The parameters of the flow path response function are related to the flow velocity and the dispersion coefficient. The watershed response is obtained as the sum of the flow path response to spatially distributed precipitation excess. De Smedt et al. (2000) proposed a flow routing method, in which the runoff is routed through the basin along flow paths determined by the topography using a diffusive wave transfer model, that enables to calculate response functions between any start and end point, depending upon slope, flow velocity and dissipation characteristics along the flow lines, and all the calculations performed with standard GIS tools. In this study, a diffusive transport approach for flow routing in GIS-based flood modelling is presented. A response function is determined for each grid cell depending upon two parameters, the average flow time and the variance of the flow time. The flow time and its variance are further determined by the local slope, surface roughness and the hydraulic radius. The flow path response function at the outlet of the catchment or any other downstream convergence point is calculated by convoluting the responses of all cells located within the drainage area in the form of the PDF of the first passage time distribution. This routing response serves as an instantaneous unit hydrograph and the total discharge is obtained by convolution of the flow response from all spatially distributed precipitation excess. The model is applied to the Attert basin in the Grand-duchy of Luxembourg, for which topography and soil data are available in GIS form, and land use data is obtained from remote sensed images. River discharges are estimated on hourly basis from October 1998 to March 2001. Consequently, a sensitivity analysis is conducted to study the effect on the IUH and the predicted hydrograph at the basin outlet such as the hydraulic radius, the channel roughness coefficient, the threshold of minimum slope, and the area threshold of delineating permanent channel networks. The parameters, which significantly affect the IUH and the general applicability of the model, are also discussed.

2.

Methodology

Starting from the continuity equation and the St. Venant momentum equation, assuming one-dimensional unsteady flow, and neglecting the inertial terms and the 38

Development of diffusive transport approach for flow routing lateral inflow to the flow element, the flow process can be modelled by the diffusive wave equation (Cunge et al., 1980):
2Q Q Q D 2 =0 +c x x t

(3.1)

where Q [LT-1] is the discharge at time t and location x, t [T] is the time, x [L] is the distance along the flow direction, c [LT-1] is the kinematic wave celerity and is interpreted as the velocity by which a disturbance travels along the flow path, and D [LT-1] is the dispersion coefficient, which measures the tendency of the disturbance to disperse longitudinally as it travels downstream. Such dispersion is induced by turbulence initiated from the shearing effects of channel boundaries (Mesa & Mifflin, 1986; Rinaldo et al., 1991). Assuming that the bottom slope remains constant and the hydraulic radius approaches the average flow depth for overland flow and watercourses, c and D can be estimated using the relation of Manning, by c = (5/3)v, and D=(vR)/(2S) (Henderson, 1966), where v is the flow velocity, R the hydraulic radius and S the bed slope. Parameters c and D are assumed to be independent of the discharge, Q. Hence, the partial differential equation (3.1) becomes parabolic, having only one dependent variable, Q(x, t). Considering a system bounded by a transmitting barrier upstream and an adsorbing barrier downstream, the solution of (1) at the cell outlet with cell size of l [L], can be obtained using Laplace transforms for a unit impulse input (Eagleson, 1970), which results in a PDF of the first passage time distribution as:
(ct l )2 u (t ) = exp 4 Dt 2 Dt 3 l

(3.2)

where u(t) [T-1] is the cell response function, and is equal to the PDF of the travel time spent in a flow element, X [T], which is considered to be a random variable independent of those in the other flow elements. From a physical point of view, the independence of flow elements implies that the travel time a water particle spends in a grid cell is not related to the time spent in any other cells, and the transport dynamics depend solely on local variables and parameters and not on the conditions in the 39

Chapter III surrounding cells (Maidment et al., 1996). Consequently, the first three moments can be derived from the moment generating function of the first passage time distribution (DeGroot, 1986, p. 201) as E(X)=l/c, Var(X)=2Dl/c3, Skw(X)=12D2l/c5, where E(X), Var(X) and Skw(X) are the mean, variance and skewness of the random variable X. Since the total time spent in the flow path, Y [T], is equal to the sum of the times spent in each of its components along the flow path, Y is also a random variable independent of those in the other flow paths. In probability theory, the PDF of the sum of a finite number of random variables is defined as the sequential convolution of their probability density functions. Therefore, the flow path redistribution function, which is equal to the PDF of the random variable Y, can be obtained through the sequential convolution of the PDFs of the random variable X within the flow path. Mathematically, this convolution can be performed only by numerical integration and therefore has no analytical representation (Olivera & Maidment, 1999). For a flow path consisting of N elements, N-1 convolutions have to be performed in order to get the flow path redistribution function. Furthermore, this process has to be worked out for each flow path in the watershed. Due to the enormous amount of calculations that have to be performed, the method of numerical integration is not feasible and difficult to realize in the hydrological models. Hence, an approximate numerical solution is preferable in finding the PDF of Y, given that the probability density functions of all X in the flow path are known. Although it is not possible to obtain an exact solution to the sequential convolution, the moments of the sequential convolution can be determined using the probability theory. DeGroot (1986, p. 188, p. 197) proves that the expected value and the variance of the sum of the random variables are equal to the sum of their expected values and variances. For a first passage time distribution, the equations can be expressed as:
1 E (Y ) = t0 = dx c

(3.3)

Var (Y ) = 2 = 2

D dx c3

(3.4)

40

Development of diffusive transport approach for flow routing where t0 [T] is the travel time from the cell to the basin outlet along the flow path, and 2 [T] is the variance of the flow time. Likewise, it can be proven that the skewness of the sum of the independent variables is equal to the sum of their skewnesses.
D2 dx c5

Skw(Y ) = 12

(3.5)

An approximate solution of the flow path response function is then obtained in the form of a first passage time distribution, which satisfies the statistical requirement of the first three moments as described above. The equation is written as:
(t t 0 ) 2 exp 2 3 2t 3 t 0 2 t / t 0 1

U (t ) =

(3.6)

where U(t) [T-1] is the flow path unit response function, and [T] is the standard deviation of the flow time. The parameters t0 and in equation (3.6) are spatially distributed, so that each flow path has different parameters depending on the length of the flow path and the physical characteristics of the flow path elements. From a hydraulic point of view, Equation (3.6) describes an elementary wave serving as an IUH of the flow path. Examples of such IUH at the end of the flow path are presented in Figure III-1a and III-1b as a function of time. It is seen that the IUH is asymmetric with respect to time caused by the wave attenuation.

(a)

0.0008

(b) 0.0008
= 7200 7200 s s = d
d 3600 s s = 3600 d t0 = = 7200 5400s s 0.0006 d t0 = = 3600 3600s s

0.0006

d 1800 s s = 1800 d 600s s = 600

t0 = = 1800 2700s s d
= 600 2300 t0 = d ss

U (s )

0.0004

U (s )

-1

-1

0.0004

0.0002

0.0002

0.0000 0 1800 3600 5400 7200

0.0000 0 1800 3600 5400 7200

t (s)

t (s)

Figure III-1: (a) Unit response functions for an expected travel time of 3600 s and different standard deviations, and (b) Unit response functions for an expected standard deviation of 3600 s and different travel times 41

Chapter III

Figure III-1a and III-1b show that the approximate solution of the diffusive wave equation satisfies the general characteristics of longitudinal wave dispersion along a flow path, i.e. for a given variance of the flow time, more travel time results in less wave attenuation, and for a given average travel time, more variance of the flow time results in more wave attenuation. The IUH tends to a normal distribution when 2 is small and the wave propagates as a pure translation at the limit 0. Olivera and Maidment (1999) compare the goodness of the approximation of three probability distributions: normal, gamma and first-passage-time, with the exact numerical integral solution of the sequential convolution. They conclude that no statistical reasons make one function better than the others. The first passage time distribution is chosen in this study, because the two parameters t0 and 2 are physically based and can be estimated conveniently by using standard GIS functions, e.g. Equations (3.3) and (3.4) can be calculated with the weighted flow length function, included in all commercially available GIS software that operates on raster data. Moreover, the first passage time distribution has been used in other studies (Mesa & Mifflin, 1986; Naden, 1992; Troch et al., 1994; Olivera & Maidment, 1999) for modelling the time spent by water in hydrological systems. The total flow hydrograph at the basin outlet can be obtained by a convolution integral of the flow response from all grid cells.

Q (t ) =

I ( )U (t )ddA
t A 0

(3.7)

where Q(t) [LT-1] is the outlet flow hydrograph, I(t) [LT-1] is the excess precipitation in a grid cell, [T] is the time delay and A [L] is the drainage area of the watershed. For the purpose of model parameter optimization and sensitivity analysis, a watershed unit response function is proposed in this study based on the flow path redistribution function described above. The watershed IUH differs from the traditional GIUH, which uses the drainage basin hillslope function weighted by the channel network width function (Troch et al., 1994), because it integrates the flow path response functions in the basin weighted by the spatially distributed runoff coefficient

42

Development of diffusive transport approach for flow routing

UH (t ) =

CU (t )dA

(3.8)

CdA

where UH(t) [T-1] is the IUH of the catchment or subcatchment, and C [-] is the default runoff coefficient of the grid cell, which is assumed to depend upon slope, soil type and land use. Values of the default runoff coefficient can be collected from the literature (Kirkby, 1978; Chow et al., 1988; Browne, 1990; Mallants & Feyen, 1990; Pilgrim & Cordery, 1993) The numerator on the right hand side of Equation (3.8) serves as the direct runoff hydrograph at the outlet resulting from a unit volume of rainfall but spatially distributed surface runoff, while the denominator is the total volume of the runoff. The watershed IUH described in Equation (3.8) can also be used in lumped or semi-lumped rainfall runoff models to predict outlet hydrographs with an average excess precipitation input on subcatchment or catchment scale.

3.

Application

The diffusive flow routing model was tested on a subcatchment with outlet at Ell in the Attert basin, which is a main tributary of the Alzette river in the Grand-Duchy of Luxembourg (Figure III-2). The topography and soil data of the catchment are available in GIS form, and land use data was obtained from remote sensed images. The elevation in the 96.8 km watershed ranges from 273 to 530 m above mean sea level, with an average basin slope of 9.6%. Figure III-3 shows the topographic elevation map of the Attert subcatchment upstream of Ell gauging station, and Figure III-4 shows the land use map of the study area. This subcatchment is partly located in Belgium and partly in the Grand-Duchy of Luxembourg. Deciduous shrub and forest are the dominant land use types of the watershed (41.1%); other land use types are agriculture (21.4%), grassland (34.1%) and urban areas (3.4%). Left-bank tributaries of the Attert are located on schistous substratum, characteristic of the Ardennes massif, whereas right-bank tributaries are located on marls and sandstone, belonging to the Paris Basin Mesozoic deposits. A very small area is covered by marshes. The dominant soil textures are loam (67.6%) and sandy loam (29.8%), while the rest is sand, loamy sand and sandy clay loam, which are scattered near the basin outlet.

43

Chapter III

#
#

Towns Rivers

Clervaux
GERMANY

Alzette basin
Grand-Duchy of Luxembourg
Study area

Wiltz

BELGIUM

Ettelbruck

rt Atte
#

Echternach
W

N E S

Ell

Mersch

Alzette
#

Luxembourg-city
#

Remich

10

20 km

Esch/Alzette
FRANCE

Figure III-2: Location plan showing the study area, the Attert and Alzette river basin
Elevation (m) ( ) 270 - 300 300 - 330 330 - 360 360 - 390 390 - 420 420 - 450 450 - 480 480 - 510 510 - 540
N W S E

Land use

crop grass marsh shrub forest urban

N W S E

2 km

2 km

Figure III-3: DEM of the study area

Figure III-4: Land use map of the study area

The climate in the region has a northern humid oceanic regime. Rainfall is the main source of runoff and is relatively uniformly distributed over the year. High runoff occurs in winter and low runoff in summer due to the higher evapotranspiration. Winter storms are strongly influenced by the westerly atmospheric fluxes that bring humid air masses from the Atlantic Ocean (Pfister et al., 2000) and floods happen frequently because of saturated soils and low evapotranspiration. The average annual precipitation in the region varies between 800 mm and 1000 mm, and the annual potential evapotranspiration is around 570 mm. Precipitation generally exceeds potential evapotranspiration except for four months in summer. A total of 30 months of hourly precipitation, discharge and potential evapotranspiration data are available

44

Development of diffusive transport approach for flow routing at Ell station. The average flow during the monitoring period is 2.41 m/s with flows ranging from 0.4 to 29.8 m/s.

Model parameters are identified using GIS tools and lookup tables, which relate default model parameters to the base maps, or a combination of the base maps. Starting from the 50 by 50 m pixel resolution digital elevation map, hydrological features including surface slope, flow direction, flow accumulation, flow length, stream network, drainage area and sub-basins are delineated. The threshold for delineating the stream network is set to 10, i.e. the cell is considered being drained by ditches or streams when the drained area becomes greater than 25,000 m. A map of Mannings roughness coefficients is derived from the land use map, and a map of potential runoff coefficients (Figure III-5) is calculated from the slope, soil type and land use class combinations (Liu et al., 2000). Due to the 50 m grid size, urban cells may not be 100% impervious in reality. In this study, the percentage of impervious area in a grid cell is computed based on land use classes, with 30% for residential area, 70% for commercial and industrial area and 100% for open water areas (lakes and ponds). Default potential runoff coefficients for these areas are calculated by adding the impervious percentage with a grass runoff coefficient multiplied by the remaining area. This results in runoff coefficients of 40 to 100% in urban areas, while other areas have much smaller values, down to 5% for forests in valleys with practically zero slopes.
(a)
Runoff coefficient
0.0 - 0.1 0.1 - 0.2 0.2 - 0.3 0.3 - 0.4 0.4 - 0.5 0.5 - 0.6 0.6 - 1.0

(b)
Hydraulic radius (m) y ( ) 0.005 - 0.01 0.1 - 0.05 0.05 - 0.1 0.1 - 0.2 0.2 - 0.3 0.3 - 0.5 0.5 - 1.0

N W S E
W

N E S

2 km

2 km

Figure III-5: Distribution of potential runoff coefficient

Figure III-6: Distribution of hydraulic radius for a flood with a 2-year return period 45

Chapter III

For calculation of the spatially distributed flow velocity and dispersion coefficient, both parameters are assumed to depend on local slope, hydraulic radius and vegetation type. This differs from previous work, where the flow velocity and dispersion coefficient are considered to be uniform distributed over the hillslope and the channel networks and estimated by model calibration (Van Der Tak & Bras, 1990; Troch et al., 1994; Gyasi-Agyei et al., 1996; Olivera and Maidment, 1999). In this study, the roughness coefficients for river courses and different land uses are obtained from literature (Chow, 1964; Yen, 1991; Ferguson, 1998), while the hydraulic radius is determined by a power law relationship with an exceeding probability (Molnar & Ramirez, 1998), which relates hydraulic radius to the drained area and is seen as a representation of the average behaviour of the cell and the channel geometry, i.e.

Rp = a ( Ad ) b

(3.9)

Where Rp [L] is the hydraulic radius with exceeding probability p, Ad [L] is the drained area upstream of the cell, which can be easily determined by the flow accumulation routine in standard GIS, a [-] is a network constant and b [-] a geometry scaling exponent both depending on the discharge frequency. In determining the parameters a and b for a fixed flood frequency, the minimum and maximum hydraulic radius, corresponding to a drained area of a single cell and the whole catchment, are determined firstly based on basin characteristics or estimated when catchment geohydrological data is available. By substituting these values into Equation (3.9), a and b can be determined. Consequently, the hydraulic radius for each grid cell in the basin is calculated with Equation (3.9). In this study, the exceeding probability p is set to a 2-year return period for normal floods with corresponding a and b values of 0.10 and 0.50. This causes the minimum hydraulic radius for overland flow to be 0.005 m and the maximum hydraulic radius for channel flow 1 m at the basin outlet. The values of a and b can be increased for more extreme floods. Figure III-6 shows the spatial distribution of the hydraulic radius for a flood with a 2-year return period.

46

Development of diffusive transport approach for flow routing Because the local slope in some cells derived from the DEM can be very small and even can reach zero particularly in the river valleys in the flood plain area, the calculated flow time and its variance become very large and the computed flow path IUH is unrealistic. Therefore, a threshold for the minimum slope should be fixed, in order to make the flow path IUH more reasonable. In this study, the threshold of the minimum slope is set to 0.05%, i.e. the local slope is considered to be at least 0.05%. Thereafter, by combining the maps of the hydraulic radius, Mannings roughness coefficient, and surface slope, the average flow velocity in each grid cell can be calculated using Mannings equation, which results in velocities in the order of 0.005 m/s for overland flow on upland areas in the watershed, and up to 2 m/s for some parts of the main river. The contributing area is then determined from topographic data for a particular downstream convergence point, normally the cells corresponding to the main river or the basin outlet. Figure III-7a shows the spatial distribution of the average flow time to the basin outlet from each grid cell, and Figure III-7b shows the spatial distribution of the standard deviation of the flow time. The average flow time is less than 4 h for the main river and up to 15 h for the most remote areas, and the standard deviation increases with flow length up to 5 h for the most remote cells. With the above information, the flow path unit response functions are calculated for each grid cell to the basin outlet using Equation (3.6), and the watershed unit response function can be calculated using Equation (3.8), weighted by the spatially distributed runoff coefficient. The calculated watershed IUH is shown in Figure III-10b.

(a)
Flow time ( ) (h)
0-1 1-2 2-3 3-4 4-5 5-8 8 - 12 > 12

b) (b)
Standard deviation (h)
0.0 - 1.0 0 1.0 - 1.5

3 1.5 - 2.0
2.0 - 2.5 2.5 - 3.0 3.0 - 3.5 2 3.5 - 4.0 > 4.0

N W S E

N W S E

2 km

2 km

Figure III-7: (a) Average flow time to the basin outlet and (b) its standard deviation

47

Chapter III The generation of surface runoff is performed using the WetSpa (Water and Energy Transfer between Soil, Plants and Atmosphere) model developed by Wang et al., (1997), De Smedt et al., (2000), and Liu et al., (2002), in which the runoff production in the cell is calculated by the method of default runoff coefficients and controlled by the rainfall intensity and the soil moisture content. A linear relationship is assumed between the actual surface runoff and the soil moisture content in the root zone, where wet soils tend to generate more runoff and dry soils tend to generate less or even no runoff. The soil moisture content for each cell is simulated on the basis of a soil water balance on hourly time scale, which relies on the rate of the infiltration, percolation, interflow and evapotranspiration in and out of the root zone. Finally, the hydrograph at the basin outlet is obtained by the convolution integral of the excess precipitation and the flow path IUH from all cells in the watershed with Equation (3.7). In order to evaluate the performance of the diffusive wave approximation method for the routing of surface runoff, 30 months observed hourly discharge data at the Ell station in the Attert catchment are selected for the model verification. The baseflow is separated from the total hydrograph by the nonlinear reservoir algorithm (Wittenberg & Sivapalan, 1999), in which the baseflow is assumed to be proportional to the square of the groundwater storage as:

Qg = kS 2

(3.10)

where Qg [LT-1] is the baseflow, S [L] is the groundwater storage, and k [LT-1] is a reservoir recession coefficient, which is related to the area, shape, pore volume and transmissivity of the watershed, and can be derived from the analysis of the recession curves. Combined with the soil water balance equation, the groundwater storage can be determined and used for baseflow separation with Equation (3.10). It turns out that the computed surface runoff hydrographs compared very well with the observations. As a typical example, we show the results for a flood event that occurred from October 23 to November 13, 1998, shown in Figure III-8, where the baseflow volume takes about 69% of the total flood volume, and the direct flow about 31%. The diffusive flow routing model is then applied with spatially distributed excess rainfall as input and the hydrograph at the basin outlet as output. The predicted direct flow

48

Development of diffusive transport approach for flow routing plus baseflow versus observed hydrograph is shown in Figure III-8 for the same period. The maximum recorded rainfall intensity during this period is 12 mm/h, yielding an observed peak discharge of 29.8 m3/s, while the simulated peak flow is 31.4 m3/s. As can be seen in the figure, the predicted hydrograph is in good agreement with the observations.

Rainfall

Simulated direct discharge (m /s)

40

25

30

Observed
Predicted Baseflow

20

10

20

20

P (mm/h)

Q (m /s)

15

10

10

30

0 23/10

40 26/10 29/10 1/11 4/11 7/11 10/11

0 0 5 10 15 20
3

25

Time (d/m)

Measured direct discharge (m /s)

Figure III-8: Observed and predicted stream flow and baseflow at Ell station

Figure III-9: Measured vs. simulated peak direct discharges

The results for other periods of the 30 months observation series are similar. A scatter plot of observed versus simulated peak direct discharges of the 24 largest storm events that occurred during the 30 months simulation period are presented in Figure III-9, in which the measured peak direct discharge is given as the observed peak discharge minus the baseflow. As can be seen in the figure, peak floods are reproduced fairly well, while the low floods tend to be somewhat overestimated by the model. This is because the frequency used to estimate the hydraulic radius in the model is a 2-year return period, which may not be correct for simulating more frequent flood events. For assessing the model performance, 3 evaluation criteria were applied to the simulation results for the whole simulation period: (1) the model reproduces the volume of surface runoff with 8% under estimation, (2) the model Nash-Sutcliffe efficiency (Nash & Sutcliffe, 1970) for reproducing the direct discharges is 83%, and (3) the average correlation coefficient between the measured and predicted hydrograph is 76%. Also, the prediction errors of the time to the peak of the 24 flood events are within 3 hours, which proves that the diffusive transport model is very well suited for flood prediction in the Attert basin. 49

Chapter III

4.

Sensitivity analysis

The basic purpose of the sensitivity analysis is to determine differences in the model responses as a result of changes in the values of specific parameters. In the present study, a sensitivity analysis was conducted for the hydraulic radius, the channel roughness coefficient, the threshold for minimum slope, and the area threshold in delineating channel networks. The sensitivity results are, however, site specific and may vary with locations of different catchment size, soils, land use, and slope configurations. The effect of each parameter is studied by varying its value while keeping other parameters constant. In all cases, the predicted hydrograph for a flood event in October 1998 is considered as references. The calculated watershed IUH by Equation (3.8) is also presented to give a graphical view of the effect on the mean, variance and skewness of the average travel time, even though it is not used to calculate the outlet hydrograph.

4.1.

Effect of hydraulic radius

Instead of using a constant hillslope velocity and channel flow velocity to calculate the flow path response and watershed response as in many of the previous works, the concept of minimum energy expenditure is applied here to derive the hydraulic radius. The flow velocity is considered to be location dependent relying on the roughness coefficient, the local slope, and the hydraulic radius. The average hydraulic radius is obtained by the power law relationship given by Equation (3.9) (Molnar & Ramirez, 1998), which is assumed to be constant for a flood event, but may vary from event to event according to the flood frequency. Three flood frequencies, namely 0.1, 0.5 and 2.0, were considered to study their influence on the runoff hydrograph at the outlet and the watershed IUH, while keeping other parameters constant. The frequencies, 0.1, 0.5 and 2.0, correspond to return periods of 10, 2 and 0.5 years respectively. The corresponding values of calculated hydraulic radius at the basin outlet are about 1.5, 1.0 and 0.5 m respectively, while the minimum value of the hydraulic radius remain constant at 5 mm for surface runoff in the upstream part of the catchment. It is found from Figure 50

Development of diffusive transport approach for flow routing III-10a that a change in the flood frequency causes a considerable alteration in the peak value of the simulated direct hydrographs and the catchment IUH. The peak discharge increases from 16.7 m/s to 17.8 /s and shifts one hour ahead as the flood frequency decreases from 0.5 to 0.1, and decreases to 14.7 m/s with one hour time delay as the flood frequency increases to 2.0. This is logical because big storms lead to higher peak discharges and shorter travel times. Figure III-10b shows the effect of the hydraulic radius on the calculated watershed IUH. The mean, variance and the skewness of the travel time are decreasing with increased flood frequency, because these parameters are inversely depending on the celerity, as can be seen from Equations (3.3), (3.4) and (3.5), and any increase in hydraulic radius will result in less damping and faster response of the flood wave.

(a)

25

(b)
p= 0.1
-1

0.25

D irect discharge (m /s)

p= 0.5
15

Watershed IUH (h )

20

0.20

p= 0.1

p= 0.5
0.15

p= 2.0

p= 2.0

10

0.10

0.05

0 24/10 19

0.00
25/10 3 25/10 12 25/10 20 26/10 4 26/10 13

10

15

20

Time (d/m h)

Time (h)

Figure III-10: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of hydraulic radius with expected flood frequency, P

4.2.

Effect of channel roughness

Since surface runoff from each grid cell will contribute to the stream flow, the roughness coefficient has a direct impact on the travel time and amount of dissipation that will occur when routing a flood hydrograph through a river basin. Roughness coefficients for hydrological routing models are typically in the form of Mannings n values, and estimated based on the channel geometry. Generally, the roughness coefficient is higher for upstream channels, and decreases with stream order when the channel slope becomes small. For the convenience of model computation and result comparison, the channel roughness coefficient is considered to be constant in this example regardless of the effect of stream order. Figure III-11a shows the simulated 51

Chapter III direct runoff hydrographs and the calculated watershed IUH with three different values of Mannings roughness coefficient. The value 0.03 corresponds to clean and straight streams without riffles or deep pools, 0.04 to clean and winding streams with some pools and shoals, and 0.05 to clean and winding streams with stones (Chow, 1964). It is found that the peak discharge decreases from 16.7 to 14.3 m/s and is somewhat delayed as the roughness coefficient increases from 0.04 to 0.05, and increases to 20.2 m/s with one hour shifting ahead as the roughness coefficient decreases to 0.03. Since the total runoff volume remains constant, reduction in peak discharge and delay in peak time are compensated by prolonged flow recession, and vice versa. This is also reflected in the calculated watershed IUH as shown in Figure III-11b. The mean, variance and the skewness of the travel time are increasing with increasing roughness, due to the fact that any increase in roughness coefficient results in higher shear stresses, causing more damping and slowing down of the flood wave.

(a) 25 Direct discharge (m /s)

n = 0.03

(b) 0.25
n = 0.03

Watershed IUH (h )

20

0.20

-1

n = 0.04
15

n = 0.04
n = 0.05

n = 0.05

0.15

10

0.10

0.05

0 24/10 19

0.00

25/10 3

25/10 12

25/10 20

26/10 4

26/10 13

10

15

20

Time (d/m h)

Time (h)

Figure III-11: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of channel Mannings roughness coefficient, n

4.3.

Effect of minimum slope

The present approach considers the changes in velocity with respect to distance, but ignores the changes in velocity with respect to time. Therefore, it can be used to route slow rising floodwaves through very flat slopes, but errors in the amount of damping will occur when routing rapidly rising flood waves through extremely flat channel slopes, because the inertia terms are not included in the diffusion wave method. In GIS, the slope of the cell is derived from the DEM and calculated from the 3x3 neighbourhood using the average maximum technique. Inevitably, nearly zero slopes 52

Development of diffusive transport approach for flow routing may occur in some areas, especially in the river valleys in the flood plain area, resulting in nearly infinity travel time and damping. To mitigate the impact of the extremely flat slopes on the flow path function, it is necessary to import a threshold for minimum slope, i.e. the cell slope is put equal to the threshold value when the calculated slope is smaller than the threshold. Keeping all other parameters constant, three values of minimum slope, namely, 0.01%, 0.05% and 0.1% are considered to study the effect of the threshold value on the outflow hydrograph and the calculated watershed IUH. Results are shown in Figure III-12a and 12b. It is found that the peak discharge and the time to the peak of the watershed IUH decrease slightly with a smaller threshold for minimum slope. This is because a decrease in slope will reduce the flood wave celerity, and therefore increase the travel time and the amount of hydrograph attenuation. Since the number of cells with a slope lower than the thresholds is small in this catchment, the influence of the minimum slope is not very significant. However, the minimum slope may have a large influence on the outflow hydrograph for catchments with flatter slopes.

(a) 20

(b)
Smin = 0.1% 0.1% Smin =
-1

0.25
Smin =0.1% 0.1% Smin =
Smin =0.05% 0.05% Smin =
Smin Smin = =0.01% 0.01% 0.15

Direct discharge (m /s)

15

Smin = 0.05% 0.05% Smin =

Smin Smin = = 0.01% 0.01%

10

Watershed IUH (h )

0.20

0.10

0.05

0 24/10 19

0.00

25/10 3

25/10 12

25/10 20

26/10 4

26/10 13

10

15

20

Time (d/m h)

Time (h)

Figure III-12: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of the threshold of minimum slope, Smin

4.4.

Effect of area threshold in delineating channel networks

In standard GIS applications, such as ArcInfo and ArcView, watershed channels are delineated based on the upstream area of each cell. It is assumed that any upstream area smaller than the threshold value does not produce enough runoff to support a

53

Chapter III channel. The area required to develop a channel depends on regional and watershed characteristics such as climatic conditions, soil properties, surface cover, and slope characteristics (Martz & Garbrecht, 1992). In cells that are not part of the stream network, overland flow occurs. Therefore, with a small area threshold value, GIS derived stream networks are more meticulous and may represent ephemeral and intermittent streams that are too small to be represented on topographical maps. The effect of the area threshold in delineating channel networks on the outflow hydrograph and the calculated watershed IUH is investigated by varying the cell number threshold, namely 5, 10 and 50, which corresponds to draining areas of respectively 12500, 25000 and 125000 m, while keeping other model parameters constant. It can be seen from Figure III-13a and III-13b, that there is no significant effect on the peak discharge and the calculated watershed IUH in this catchment. This is due to the fact that changes in the threshold area will result in expansion or shrinking of the stream network with lengths that are however relatively short compared to the whole flow paths. Hence, the impact will only become significant when using large threshold values, because in this case hillslope effects become important due to their high overland flow roughness, which will result in a longer flow time, and a prolonged flow response at the end of the flow path.

(a)
Direct discharge (m /s)

25
Cn =5 5 C n =

(b)

0.25
Cn Cn = 5

Watershed IUH (h )

20

0.20

Cn = 10 10 C n =
Cn C = 50 50 n =

-1

Cn Cn = 10
Cn Cn = 50

15

0.15

10

0.10

0.05

0 24/10 19

0.00

25/10 3

25/10 12

25/10 20

26/10 4

26/10 13

10

15

20

Time (d/m h)

Time (h)

Figure III-13: (a) Simulated direct hydrographs and (b) calculated watershed IUH showing the effect of cell number threshold, Cn, in delineating channel networks

54

Development of diffusive transport approach for flow routing 4.5. Other effects

In addition to the effects discussed above, the variation of channel geometry and the temporal and spatial resolutions of the model will also have considerable influence on the outflow hydrograph and the watershed IUH. In this study, flow is routed using a velocity calculated for each land use category both for overland flow and channel flow. The velocity is determined from Mannings equation by assuming that the hydraulic radius equals the average flow depth without considering the effect of channel width and type. This assumption is warranted if the width of the river is much larger than its depth for a flood event. However, as the width of a channel decreases, the hydraulic radius does not tend towards the average flow depth. Also, the effect of flood plains on the propagation of a floodwave can be very significant, when water overflows the riverbanks. It is expected that an expanded channel width will slow down the flow velocity and therefore reduce the peak discharge and delay the resulting runoff hydrograph. Hence, more reliable results can be obtained when calculating the hydraulic radius combined with measured or estimated channel width. The time and space scale of the model not only influence on rainfall intensity and the surface runoff distribution, but they also have impacts on the watershed IUH derived from the diffusive transport method. Errors may arise when modelling flash flood for a small catchment with a long time scale. This is because floodwater can flow out of the catchment within the first time step, which the IUH cannot calculate accordingly. Therefore, a higher time resolution is necessary in this case. However, when modelling floods in a large catchment with relatively long concentration times, the effect of time scale is not important. On the other hand, changes in spatial resolution of the model will lead to variations of the GIS derived slope, flow direction, and spatial distribution of the flow paths. In general, higher spatial resolution tends to generate longer flow paths, and hence increases the hydrodynamic and the geomorphologic attenuation of the flood wave. The first is due to increased flow time, and the second to increased variability of the flow paths. Both impacts will play an important role in the prediction of transport phenomena, especially in large basins (Rinaldo et al., 1991). It is expected that reduction in spatial resolution will result in a decrease of peak discharge and prolonged time to flood peaks, and vice versa.

55

Chapter III As pointed out by Horritt and Bates (2001), a high-resolution model is advantageous when small scale processes have a significant effect on model predictions, but have to be balanced against the increased computation onus. Predictions with a low-resolution may also give an essentially correct result in many cases. In practice, determination of temporal and spatial resolution of the model should rely on the data available, the catchment characteristics and the model accuracy requirement. However, quantitative analysis of these effects on the outflow hydrograph and the watershed IUH in GIS flood modelling is beyond the scope of this study.

5.

Conclusions

A physically based distributed unit hydrograph method derived from the diffusive transport approach is presented in this study for GIS based modelling on catchment scale. The method differs from the previous work in that it is based on a location dependent velocity field. The basic modelling approach is to use raster GIS functions to calculate the travel time from each point in the watershed to the outlet by determining the flow path and the travel time through each cell along the path. The flow velocity in each grid cell is calculated by the Manning equation, which depends upon the local slope, roughness coefficient and hydraulic radius. The travel time through each individual cell along the flow path is integrated to obtain the cumulative travel time to the outlet. Based on the mean and the variance of the flow time, the first passage time distribution density function is applied as a flow response function. Runoff is routed over the surface flow path, and accounts for the differences in runoff amount and velocity, due to changing slope, land use, soil type and other surface conditions. Finally, the total direct discharge at the downstream convergence point is obtained by superimposing all contributions from every grid cell. The watershed IUH is calculated based on the flow path functions and the spatially distributed runoff coefficient, and can be used for model parameter sensitivity analysis or as the IUH for lumped prediction models. Model parameters based on surface slope, land use, soil type and their combinations are collected from literature, and can be prepared easily using standard GIS techniques. The model was tested on the Attert catchment in Luxembourg with 30 months of observed hourly rainfall and discharge data, where the spatial distributed surface 56

Development of diffusive transport approach for flow routing runoff was generated by the WetSpa model. The results show an excellent agreement with the measured hydrograph at the basin outlet. Consequently, a sensitivity analysis was conducted to study the effect of the hydraulic radius, the channel roughness coefficient, the threshold for minimum slope, and the area threshold in delineating channel networks on the outflow hydrograph and the calculated watershed IUH. It was found that the hydraulic radius and channel roughness coefficient are the most sensitive parameters. The hydraulic radius corresponding to a 2-year return period can meet the requirements of flood prediction for normal floods, but should be increased for more extreme flood. Also, the channel roughness coefficient shows a strong impact on the model output. More reliable results are expected when the channel roughness is determined according to the stream order. The thresholds of minimum slope and the area in delineating channel networks have only marginal effects on the outflow hydrograph and the calculated watershed IUH. However, all these parameters should be chosen properly when applying the model in practice. The diffusive wave transport approach assumes a unique relationship between flow and stage at each point for both overland flow and channel flow, and so does not require the specification of a downstream stage. It also generally operates satisfactorily with less detailed ditch and channel geometry information than required by dynamic wave models and is much more stable and easy to use in GIS based flood modelling. Moreover, this approach allows the spatially distributed excess precipitation and hydrological parameters of the terrain to be used as inputs to the model, and is especially useful to analyze the effects of topography, and land use or soil cover on the hydrological behaviour of a river basin. The method is worth to be applied in flood modelling for a wide range of slopes from flood plains to the hilly areas. However, accuracy of the diffusive wave approach increases with increasing slope, and it cannot be used in situations where flow reversals occur. Application of the methodology suggests that simulations of the hydrological response based on diffusive wave approximation and GIS specification of the topographical network are validated in the study area. This is sustained by a proper adjustment of the parameter values characterizing the flow travel time and its variance, which is deemed to cover most cases of engineering interest.

57

Chapter III References

Beven, K.J. & Kirkby, M.J., A physically based variable contributing area model of basin hydrology, Hydrol. Sci. Bull., 24(1), 43-59, 1979. Browne, F.X., Stormwater Management, Standard Handbook of Environmental Engineering, ed., R.A. Corbitt, McGraw-Hill, New York, 7.1-7.135, 1990. Chow, V.T., Handbook of Applied Hydrology, 7-25, McGraw-Hill Book Company, New York, 1964. Chow, V.T., Maidment, D.R. & Mays, L.W., Applied Hydrology, McGraw-Hill, New York, 1988. Cunge, J. A., Holly, F.M. & Verwey, A., Practical Aspects of Computational River Hydraulics, p. 45, Pitman Publ. Ltd, London, GB, 1980. DeGroot M.H., 1986, Probability and Statistics, Addison-Wesley, Reading, MA, USA. De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrological modelling on a catchment scale using GIS and remote sensed land use information, In: Risk Analysis II, ed., C.A. Brebbia, 295-304, WTI press, Southampton, Boston, 2000. Dooge, J.C.I., A general theory of the unit hydrograph, J. Geophys. Res., 64, 241-256, 1959. Eagleson, P.S., Dynamic Hydrology, 364, McGraw-Hill Pub., 1970. Ferguson, B.K., Introduction to Stormwater, Concept, Purpose and Design, 111, John Wiley & Sons, Inc., 1998. Gupta, V.K. & Waymire, E., On the formulation of an analytical approach to hydrological response and similarity at the basin scale, J. Hydrol., 65, 95-123, 1983. Gupta, V.K., Waymire, E., & Wang, C.T., A representation of an instantaneous unit hydrograph from geomorphology, Water Resour. Res., 16(5), 855-862, 1980. Gyasi-Agyei, Y., De Troch, F.P. & Troch, P.A., A dynamic hillslope response model in a geomorphology based rainfall-runoff model, J. Hydrol., 178, 1-18, 1996. Henderson, F.M., Open Channel Flow, 522, McMillan, New York, 1966.

58

Development of diffusive transport approach for flow routing Horritt, M.S. & Bates, P.D., Effect of spatial resolution on a raster based model of flood flow, J. Hydrol., 253, 239-2498, 2001. Kirkby, M.J., Hill-slope Hydrology, 235, John Wiley & Sons, Ltd., 1978. Lee, K.T. & Yen, B.C., A geomorphology and kinematic-wave-based hydrograph derivation, J. Hydraulic Eng., ASCE, 123(1), 73-80, 1997. Liu, Y.B., Gebremeskel, S., De Smedt, F., and Pfister, L., Flood prediction with the WetSpa model on catchment scale, In; Flood Defence 2002, ed., Wu et al., 499507, Science Press, New York Ltd, 2002. Maidment, D.R., Developing a spatially distributed unit hydrograph by using GIS, In HydroGIS 93: Application of Geographic Information Systems in Hydrology and Water Resources, Proceedings of the Vienna Conference, eds., K. Dovar and H. P. Natchnebel, 181-192, Vienna: Int. Assoc. of Hydrol. Sci., 1993. Maidment, D.R., Olivera, J.F., Calver, A., Eatherral, A. & Fraczek, W., A unit hydrograph derived from a spatially distributed velocity field, Hydrol. Process, 10(6), 831-844, 1996. Mallants, D. & Feyen, J., Kwantitatieve en kwalitatieve aspecten van oppervlakte en grondwaterstroming (in Dutch), 96, Volume 2, KUL, 1990. Martz, L.W. & Garbrecht, J., Numerical definition of drainage network and subcatchment areas from digital elevation models, Computers and Geosciences, 18(6), 747-761, 1992. Mesa, O.J. & Mifflin, E.R., On the relative role of hillslope and network geometry in hydrological response, in Scale Problems in Hydrology, eds., V.K. Gupta, I. Rodriguez-Iturbe, and E.F. Wood, 1-17, D. Reidel, Norwell Mass, 1986. Molnar, P. & Ramirez, J.A., Energy dissipation theories and optimal channel characteristics of river networks, Water Resour. Res., 34(7), 1809-1818, 1998. Naden, P.S., Spatial variability in flood estimation for large catchments: The exploitation of channel network structure, J. of Hydrol. Sci., 37, 53-71, 1992. Nash, J.E. & Sutcliffe, J.V., River flow forecasting through conceptual models, J. Hydrol. 10, 282-290, 1970. Olivera, F. & Maidment, D.R., Geographic information system (GIS)-based spatially distributed model for runoff routing, Water Resour. Res., 35(4), 1155-1164, 1999.

59

Chapter III Pfister, L., Humbert, J. & Hoffmann, L., Recent trends in rainfall-runoff characteristics in the Alzette river basin, Luxembourg, Climate Change, 45(2), 323-337, 2000. Pilgrim, D.H. & Cordery, I., Flood runoff, in Handbook of Hydrology, ed., D. R. Maidment, 9.1-9.42, McGraw-Hill, New York, 1993. Rinaldo, A., Marani, A. & Rigon, R., Geomorphological dispersion, Water Resour. Res., 27(4), 513-525, 1991. Rodriguez-Iturbe, I. & Valdes, J.B., The geomorphologic structure of hydrological response, Water Resour. Res., 15(6), 1409-1420, 1979. Sivapalan, M., Wood, E.F. & Beven, K., On hydrological similarity: 3. A dimensionless flood frequency model using a generalized geomorphologic unit hydrograph and partial area runoff generation, Water Resour. Res., 26(1), 43-58, 1990. Snell, J.D. & Sivapalan, M., On geomorphologic dispersion in natural catchments and the geomorphologic unit hydrograph, Water Resour. Res., 30(7), 2311-2323, 1994. Troch, P.A., Smith, J.A., Wood, E.F. & de Troch, F.P., Hydrological controls of large floods in a small basin, J. Hydrol., 156, 285-309, 1994. Van Der Tak, L.D. & Bras, R.L., Incorporating hillslope effects into the geomorphologic Instantaneous Unite Hydrograph, Water Resour. Res., 26(1), 2393-2400, 1990. Wang, Z., Batelaan, O. & De Smedt, F., A distributed model for Water and Energy Transfer between Soil, Plants and Atmosphere (WetSpa), Phys. Chem. Earth, 21(3), 189-193, 1997. Wittenberg, H. & Sivapalan, M., Watershed groundwater balance estimation using streamflow recession analysis and baseflow separation, J. Hydrol., 219, 20-33, 1999. Yen, B.C., Channel Flow Resistance: Centennial of Manning's Formula, p. 43, Water Resources Publications, Littleton, CO., 1991.

60

Chapter IV Flood modelling for complex terrain using GIS and remote sensed information
Abstract A spatially distributed hydrological model, WetSpa (Water and Energy Transfer between Soil, Plants and Atmosphere) Extension, working on an hourly time scale is presented in this chapter. The model combines elevation, soil, and land use data, and predicts flood hydrograph and the spatial distribution of hydrological characteristics in a watershed. The model is tested on a small catchment in Belgium for which topography and soil data are available in GIS form, while the land use and soil cover is obtained from remote sensed images. The resulting calculated discharges compare favourably with the field measurements. Next a 102-year series of measured hourly precipitation data is processed with the model and the resulting hydrographs are analyzed statistically to determine the characteristics of extreme floods. Finally, the simulated extreme peak discharges are compared to the results calculated with design storms. Comparison of the two methods shows that the model is capable to predict both normal and extreme floods in a complex terrain.

1.

Introduction

Flood is a natural phenomenon that poses serious challenges to many countries and regions in the world. It is among the most severe risks on human lives and properties, as well as economical development. The forecast and simulation of floods is therefore essential for planning and operation of civil protection measures and for early flood warning. In applied hydrology, the prediction of peak flow and the simulation of flood hydrographs in a stream or river is a very complex process, because the hydrological variables vary both in space and time as a function of the meteorological inputs, and the spatial variability of topography, land use and soil types.

Chapter IV In contrast to lumped models, distributed models attempt to account for the spatial variability of basin parameters, as well as their physical significance. In flood prediction and rainfall-runoff computation, physically based distributed hydrological models have become a more feasible approach in recent years. In addition to the development of improved computational capabilities, DEMs, digital data of soil type and land use, as well as the tools of GIS, give new possibilities for hydrological research in understanding the fundamental physical processes underlying the hydrological cycle and the solution of mathematical equations representing those processes. In a distributed model, the watershed is subdivided into a grid of cells, and model parameters are assigned to each grid cell based on the physical land, soil and vegetation characteristics that exist in that cell. Precipitation and other meteorological data are then applied over each cell in the watershed, and the runoff is computed and routed along its flow direction to the collecting channel. In such a way, distributed models are able to account for the spatial variability of hydrological processes within a watershed. In addition, the model parameters of this approach are largely physically based and the spatial information of the land, soil, vegetation and precipitation can be captured with much greater detail than the lumped watershed modelling. This gives the advantage for the model to be used in complex terrains and ungauged river basins without model optimization. Recently, many hydrological models with a flood prediction component using information on topography available from DEM have been developed (Moore, 1991; Palacios-Vlez & Cuevas-Renaud, 1992; Robson et al., 1993; Garrote & Bras, 1995), whereas models like SHE and TOPMODEL were adapted to a new type of data which can benefit from the GIS techniques (Quinn et al., 1991; Ewen et al., 2000). At the same time, hydrological models compatible with remotely sensed data and GIS have been developed or updated from their previous version, such as the model DHSVM (Wigmosta et al., 1994), CASC2D (Julien et al., 1995), DWSM (Borah et al., 2002), HYDROTEL (Fortin et al., 2001), and so on. These models are either loosely or tightly coupled with the GIS and remote sensed data. Along with the rapid development of GIS technology and remote sensing techniques, especially the concomitant availability of high resolution DEM and the advances in integrating GIS with hydrological modelling, flood prediction with distributed models tends to be more advantageous and competent by linking GIS with hydrological modelling. 62

Flood modelling for complex terrain using GIS and Remote sensed information

In this chapter, a physically based distributed hydrological model, WetSpa Extension, is presented, which is tightly coupled with GIS technology and remote sensed information. The model takes into account the detailed basin characteristics to predict flood hydrographs and other spatially distributed hydrological variables on catchment scale. The parameters of the model are derived from DEM, land use and soil maps in raster format. The model is validated by comparing calculated and observed hourly discharges for a 16 months period in a small watershed, Barebeek, located northeast of Brussels, Belgium, where the topography and soil data are available in GIS form, and land use data was obtained from remote sensed images. The utility of the model is demonstrated by forecasting peak discharges resulting from an observed 102-year precipitation series. The simulation results are then compared with the results computed from design storms.

2.

The WetSpa model

The WetSpa model is a grid-based distributed hydrological model for Water and Energy Transfer between Soil, Plants and Atmosphere, which was originally proposed by Wang et al. (1997) and adopted for flood prediction on hourly and daily time step by De Smedt et al. (2000) and Liu et al. (2002, 2003). For each grid cell, four layers are considered in the vertical direction as vegetation zone, root zone, transmission zone and saturated zone. The hydrological processes considered in the model are precipitation, interception, depression, surface runoff, infiltration, evapotranspiration, percolation, interflow, ground water flow, and water balance in the root zone and the saturated zone. The total water balance for a raster cell is composed of the water balance for the vegetated, bare-soil, open water and impervious parts of each cell. This allows accounting for the non-uniformity of the land use per cell, which is dependent on the resolution of the grid. The processes in each grid cell are set in a cascading way, which means that an order of occurrence of the processes is assumed after a precipitation event. A mixture of physical and empirical relationships is used to describe the hydrological processes in the model. The model predicts peak discharges and hydrographs, which can be defined for any numbers and locations in the channel network, and can simulate the spatial distribution of catchment hydrological characteristics. 63

Chapter IV

2.1.

Runoff production

Three runoff components, surface runoff, interflow and groundwater flow, are considered in the model. The model takes Hortonian flow as the main overland flow process, which occurs when rainfall intensity exceeds the infiltrability of soil. However, for a complex terrain, particularly suburban and urban areas, precise estimation of infiltration parameters is rather difficult due to the high heterogeneity of the land and soil characteristics. Hence, simplified methods are still widely used by the hydrologists for surface runoff estimation in water resources planning, design and practices, for instance the rational method and the soil conservation service (SCS) method. The rational formula is the most commonly used method of determining peak discharge from small drainage areas since it was developed in the late nineteenth century (Kuichling, 1889). This method is traditionally applied to design storm sewers, channels, and other drainage structures by making use of intensity-durationfrequency (IDF) curves, which are statistical summaries of the historical precipitation records. Since the method does not produce a hydrograph, and does not account for the change of time dependent conditions such as soil moisture or rainfall intensity, it is in general incapable to predict floods that result from individual predefined runoff events (Singh, 1992). The SCS method is an empirical approach to estimate infiltration within a watershed, in which the runoff depth is calculated as a function of the rainfall and the SCS retention factor, which is estimated from the land surface properties using an empirical relationship depending upon a curve number (Maidment, 1993). The excess rainfall is then available for runoff routing over the watershed surface and through the stream network. The SCS method is widely used for estimating floods in small to medium-sized ungauged catchments in the US. However, intensive calibrations need to be performed for the model parameters against local conditions when applying this method to other areas. In this study, a moisture-related runoff coefficient method is proposed for calculating surface runoff in each grid cell, which allows the actual runoff coefficient to vary in time, and in function of rainfall intensity, rainfall duration and cell characteristics, giving an approximation to the surface runoff volume at each time step. The initial losses due to interception and depression are considered separately in the formula: 64

Flood modelling for complex terrain using GIS and Remote sensed information

V = c s c r (P I a ) D a

(4.1)

where V [LT-1] is the surface runoff in depth over the time, P [LT-1] is the rainfall intensity, Ia [LT-1] is the interception loss, Da [LT-1] is the depression loss, cs [-] is a moisture related coefficient relying on the relative soil moisture content of the root zone, and cr [-] is the potential runoff coefficient, which is assumed to depend upon slope, soil type, land use and the proportions of bare soil and impervious areas in a grid cell. The values of default runoff coefficients are taken from literature references (Kirkby, 1978; Chow et al., 1988; Browne, 1990; Mallants and Feyen, 1990; Pilgrim and Cordery, 1993) and a table was generated, linking values of the runoff coefficient to slope, soil type and land use classes. The potential runoff coefficient is then the area-weighted average of the land use classes within the grid cell. The moisture related coefficient cs is time dependent and calculated as a function of the soil moisture and rainfall intensity as:

cs =( t s )

(4.2)

where t [L3L-3] is the soil moisture content at time t [T], s [L3L-3] is the soil porosity, and [-] is an exponent variable reflecting the effect of rainfall intensity and the modelling time step on the volume of surface runoff, which can be expressed as:
10 P Pm

= Min 1, 0 +

(4.3)

in which 0 [-] is the exponent at a near zero rainfall intensity, 0 1, and Pm [LT-1] is the rainfall intensity, over which equals 1 and the volume of surface runoff will not be affected by the rainfall intensity. Normally, 0 approaches to 1 for a short time interval, and increases when the time scale is enlarged. These two parameters mainly affect the amount of surface runoff for small storms so that more water will infiltrate into the soil. Pm has no sense if 0 is set to 1 for which a linear relationship is assumed between the surface runoff and the soil moisture content. Equation (4.1) reveals that the surface runoff achieves its potential rate when the soil is saturated, and approaches

65

Chapter IV to zero as the soil dries out. This is logical from a hydrological point of view that wet soil tends to produce more surface runoff, and dry soil tends to give more infiltration. The first term on the right side of Equation (1) is the excess rainfall calculated with the net precipitation (P - Ia) multiplied by a potential runoff coefficient and a moisture related coefficient. The difference between net precipitation and excess rainfall is the amount of infiltration into the soil. Surface runoff that is available for runoff routing is then computed with the excess rainfall subtracted by the depression losses. The coefficient cs is determined based on the intermediate soil moisture of the time step. A first trial is performed using cs of the last step to estimate excess rainfall and infiltration, and soil moisture at the end of this time step is calculated by means of water balance of the cell. The intermediate soil moisture is then estimated as the arithmetic mean of the moisture content at the end of last step and the result of the first trial. Thereafter, excess rainfall and soil moisture content at the end of the time step are recalculated using cs computed from the intermediate soil moisture. In such a way, cs is controlled not only by the antecedent soil moisture, but also by the rainfall intensity and rainfall duration. Hence, high rainfall intensity or rainfall with long duration tends to give higher percentage of runoff. The product of cs and cf forms the actual runoff coefficient, which varies both with time and rainfall intensity depending upon the soil moisture content, and allows computing excess rainfall for each time step during the model simulation. The actual runoff coefficient is set to one under the condition where saturation happens from below and groundwater resurgence occurs. The sum of interception and depression losses forms the initial abstraction, which does not contribute to runoff. In studies of major storm events, the interception loss is generally neglected. However, it may be a very significant factor for small or medium storms, and water balance computations would be significantly in error if evaporative losses of intercepted moisture were not included (Singh & Szeicz, 1979). Interception is a complicated process, which is mainly a function of the storm characteristics, the season, and the species, age and density of plants. In this study, interception loss is evaluated using a simple reservoir model, in which the rainfall rate is reduced until its storage capacity is achieved. If the total rainfall during the first time increment is greater than the interception storage capacity, the rainfall rate is reduced by the capacity. Otherwise, all rainfall is intercepted in the canopy, and the remainder of 66

Flood modelling for complex terrain using GIS and Remote sensed information interception is removed from the rainfall in the following time increments. Typical interception capacity values can be found in the literature (Lull, 1964; Zinke, 1967; Rowe, 1983). A lookup table of maximum and minimum interception storage capacity corresponding to the extreme points during summer and winter periods is established linking values of interception storage capacity to different land use classes. Thereafter, a simple sine-shaped variation curve was proposed allowing the interception storage capacity to vary continuously with time. The intercepted water in the canopy is lost by evaporation and returns to the hydrological cycle with a potential evaporation rate. Depression storage may have a considerable magnitude and plays an important role in flood modelling for small or medium storms. The rainfall excess begins to fill depressions once rainfall intensity exceeds the local infiltration capacity. Due to the extreme variability of these characteristics, it is very difficult to specify a general relationship for the losses due to depression storage. In this study, a simple empirical equation suggested by Linsley (1982) is used to calculate depression loss, in which the depression storage is assumed to be a function of depression storage capacity and increases exponentially with rainfall intensity up to the point where depression storage capacity is reached. This allows overland flow and depression storage to occur simultaneously. The depression storage capacity is a function of landform, soil type and vegetation. Based upon the typical values found in the literature (ASCE, 1969; SINCE, 1972; Sheaffer et al., 1982), a lookup table of default depression storage capacity is set up according to the category classes of slope, land use and soil type. Water held in depressions at the end of a rainfall event depletes either by evaporation with a potential evaporation rate or contributes to the soil moisture. With the component of depression storage, the model can handle surface runoff production more properly. For instance, the depression storage capacity in forest areas is much higher than that of other land use areas. Hence, there will be no or little surface runoff generated in those areas during the initial phase of a storm or for a small rainfall, because most of the produced excess rainfall will contribute to the depression storage. In this case, dominant surface runoff comes from impervious area or areas with high runoff coefficient and low depression storage capacity.

67

Chapter IV 2.2. Water balance

WetSpa has four water stores, i.e. the plant canopy, land surface, root zone and saturated zone, for each of which the water balance is maintained. The water balance in the root zone is the most important one, as it controls the amount of surface runoff, interflow, evapotranspiration and groundwater recharge. Assuming that the groundwater table is below the root zone, the water balance in the root zone can be modelled continuously for each grid cell by equating inputs and outputs:
= P I V E R F t

(4.4)

where D [L] is the root depth, [L3L-3] is the change in soil moisture, t [T] is the time interval, I = Ia+Da, [LT-1], is the initial abstracts including interception and depression storage within time t, E [LT-1] is the actual evapotranspiration from soil, R [LT-1] is the percolation out of the root zone, and F [LT-1] is the amount of interflow in depth over the time. Evapotranspiration from soil and vegetation is calculated based on the relationship developed by Thornthwaite and Mather (1955) as a function of potential evapotranspiration, vegetation type, stage of growth and soil moisture content

E=0 w E = (c v E p E i E d ) t w f E = cv E p Ei E d

for t < w for w t < f for t f

(4.5)

where cv [-] is a vegetation coefficient, which varies throughout the year depending on growing stage and vegetation type, Ep [LT-1] is the potential evapotranspiration, Ei [LT-1] and Ed [LT-1] are evaporations from interception storage and depression storage, w [L3L-3] is the moisture content at permanent wilting point, and f [L3L-3] is the moisture content at field capacity. The soil evapotranspiration varies linearly with Ep, when the soil moisture content is above the permanent wilting point, and zero, when the soil moisture is below the permanent wilting point. Potential evapotranspiration 68

Flood modelling for complex terrain using GIS and Remote sensed information data can be obtained from field measurements, estimated from the historical records through statistical analysis, or calculated with Penman-Monteith equation when hourly meteorological data of net radiation, air temperature, relative humidity and wind speed are available. The process of evapotranspiration is assumed to occur in an order from interception storage, depression storage and soil subsequently. If the water content of interception storage and depression storage is equal to or greater than the potential evapotranspiration, all the evapotranspiration comes from those storages. Otherwise, the difference is calculated with Equation (4.5) from the soil water. For the surface layer, actual evapotranspiration is computed as the area-weighted mean of the land use composition, for which transpiration happens from the vegetated parts, evaporation happens from the bare soil, and there is no evaporation from impervious areas. Finally, the total evapotranspiration is calculated as the sum of evaporation from interception storage, depression storage and the actual evapotranspiration from the soil and a part from the groundwater storage. Percolation and interflow are important components in the root zone water balance. Both processes are assumed to be gravity driven. The rate of percolation or groundwater recharge is determined by Darcys law (Hillel, 1980) in function of the hydraulic conductivity and the gradient of hydraulic potential. When an assumption is made that the pressure potential only varies slightly in the soil, its gradient can be approximated to zero, such that the percolation is controlled by gravity alone (Famiglietti & Wood, 1994). Therefore, the percolation out of root zone is simply the hydraulic conductivity corresponding to the moisture content in the soil layer. The Brooks and Corey relationship between hydraulic conductivity and moisture content is used to define percolation, which is simply (Eagleson, 1978):
( 2+3B) B

r R = K( ) = Ks s r

(4.6)

where K() [LT-1] is the unsaturated hydraulic conductivity, Ks [LT-1] is the saturated hydraulic conductivity, s [L3L-3] is the soil porosity, r [L3L-3] is the residual moisture content, and B [-] is the soil pore size distribution index. The vertical transport of water through the unsaturated soil matrix is slow. It generally takes days or months

69

Chapter IV before the percolating water reaches the saturated zone. Nevertheless, precipitation is followed by an almost immediate rise of the groundwater table owing to a rapid transfer of increased soil-water pressure through the unsaturated zone (Myrabo, 1997). In addition, macropores in the subsurface layers resulting from root and fauna activity may allow rapid bypassing of the unsaturated zone when the rate of precipitation is high (Beven, 1982). The model assumes that percolation input affects the groundwater table within one time step. Interflow is assumed to occur in the root zone after percolation and becomes significant only when the soil moisture is higher than field capacity. Darcys law and kinematic approximation are used to estimate the amount of interflow generated from each cell, in function of hydraulic conductivity, the moisture content, the slope angle, and the root depth

F = c f DS0 K( ) W

(4.7)

where S0 [LL-1] is the surface slope, W [L] is the cell width, and cf [-] is a scaling parameter depending on land use, used to consider river density and the effects of organic matter on the horizontal hydraulic conductivity in the top soil layer. Apparently, with Equation (4.7), rapid interflow will be generated in areas with high moisture, steep slope and well vegetation. For other areas, little interflow will be produced. Since little is known about the hydro-geological conditions in the saturated layer at the cell level, the groundwater storage is simulated with a semi-distributed model for simplicity, in which the groundwater balance is maintained in a small subcatchment scale with input of average groundwater recharge and output of average evapotranspiration and groundwater discharge at the subcatchment outlet.

2.3.

Flow routing

The routing of overland flow and channel flow is implemented by the method of the diffusive wave approximation, which has been described in detail in chapter III. A two-parameter response function, based on the average flow time and its standard deviation of the flow time, is applied for routing both overland flow and channel flow.

70

Flood modelling for complex terrain using GIS and Remote sensed information The flow time and its standard deviation are determined by the local slope, surface roughness and the hydraulic radius for each grid cell. The flow path response function at the outlet of the catchment or any other downstream convergence point is calculated by convoluting the responses of all cells located within the drainage area in the form of the probability density function (PDF) of the first passage time distribution. This routing response serves as an instantaneous unit hydrograph and the total discharge is obtained by a convolution integral of the flow response from all generated spatially distributed runoff. Interflow is assumed to contribute to the surface runoff at the outlet of each cell, and routed to the catchment outlet together with surface runoff without redistribution among downslope cells for simplicity. Groundwater flow is modelled with a linear reservoir method on small subcatchment scale, while a non-linear reservoir method is optional in the model with storage exponent of 2 (Wittenberg & Sivapalan, 1999). The groundwater outflow is added to any runoff generated at the subcatchment outlet to produce the total streamflow. The general groundwater flow equation can be expressed as:

G(t ) = kSm

(4.8)

where G [L3T-1] is the groundwater flow of the subcatchment at time t [T], S [L3] is the groundwater storage, m [-] is an exponent, m=1 for linear reservoir and m = 2 for non-linear reservoir which is the same as Equation 3.10, k is a baseflow recession constant, has a dimension of [T-1] for a linear reservoir and [L-3T-1] for a non-linear reservoir, which is related primarily to area, shape, pore volume and transmissivity of the subcatchment. The groundwater storage in each subcatchment is obtained by equating the water balance of the saturated zone as:
S = RdAG Eg As t As

(4.9)

where As [L2] is the subcatchment area, and Eg [LT-1] is the average transpiration from groundwater storage giving the effect of a steeper recession during dry period.

71

Chapter IV Eg is assumed to occur only when the PET is higher than soil evapotranspiration, and equated as a function of PET and groundwater storage:
E g = (cv E p Ei Ed E )

G Gm

(4.10)

where Gm is the maximum active groundwater storage of the subwatershed [L]. Hence, the flow routing consists of tracking runoff along its topographic determined flow path, and evaluating groundwater flow for each small subcatchment. In order to consider the damping effect of the river, overland flow and interflow are routed firstly from each grid cell to the main channel, and joined with groundwater flow at the subcatchment outlet. Then the total hydrograph is routed to the outlet of the catchment by the channel response function derived from Equation (3.6). The total discharge is the sum of the overland flow, interflow and groundwater flow.

3.

GIS implementation

WetSpa operates within a GIS framework. Input maps to the model include DEM, soil type and land use. Besides, the digital information of gauging sites, watershed boundary and river network, sewer systems, main hydraulic and civil infrastructures are necessary geo-referenced data for a complex terrain. Once the project database is setup, the work begins by analyzing data, extracting information, producing parameter maps, and running the distributed model. GIS functions greatly enhance the capabilities for watershed description and interpretation by means of powerful distributed indicators, which account not only for the watershed characteristics but also for their distribution and individual localization in space.

3.1.

Drainage system

Elevation data in the form of a DEM are the principle digital data source for acquiring watershed properties in the GIS-based WetSpa model. The raster-type DEM has to be compatible with remotely sensed data layers such as land use and soil type. Based on the digital elevation model, hydrological GIS tools are used to extract information on the watershed boundaries, such as slope, flow length, flow direction and accumulation, 72

Flood modelling for complex terrain using GIS and Remote sensed information configuration of stream network, subwatershed, etc., providing a suitable framework for the modelling approach. The flow accumulation map is used for synthetic network extraction, where each pixel is associated to an upstream drainage area. The distinction between hill slopes and channel paths can be achieved simply by fixing a threshold drainage area for which the flow concentration is sufficient to initiate a channel. For a small and steep watershed, processing a DEM is relatively straightforward. After filtering the initial data to detect and remove erroneous extreme values, the slope and aspect of each cell are determined and the flow direction can be obtained allowing water flowing to the basin outlet from each grid cell. However, in flat areas that are often present in large watersheds or floodplains, a raster cell within a DEM may not have neighbouring cells with a lower elevation, and thus has no down-slope exit. Such features could occur due to insufficient data, data noise, or interpolation errors during DEM production. As a result, a channel network cannot be captured from the raster DEM. A practical way to solve this problem is by the method of filling sinks (Band, 1986), in which the elevation of sink cells is increased until a downslope path to a neighbouring cell becomes available, under the constraint that the flow is not directed to another depression cell. Caution should be paid when using this method in the presence of a large water surfaces, such as lakes, reservoirs, ponds, etc., and large plains with little or no elevation variation. Additionally, difficulties may arise in delineating meandering rivers or actual watershed boundaries in flat areas. To account for these cases, data from the hydrographical layer of a digitized map can be used in combination with the DEM to identify output cells and revise flow directions, in order to produce realistic and topographically consistent drainage patterns. In a natural drainage basin with very little or no human interactions, aspect information obtained from a DEM alone is a good indication of flow direction, and the derived internal drainage structure of a watershed can be a perfect reflection of the natural reality. However, in complex terrain, such as urban or suburban watersheds, sewer systems, roads, artificial channels, etc., are important elements in drainage structure configuration, and often govern the flow direction more strongly than the derived slopes at local scale. Since most of these features are not sufficiently represented in a DEM, additional procedures for deriving more realistic flow direction 73

Chapter IV map have to be performed using GIS overlaying techniques. A general flow direction map can be generated using geographic data alone. Thereafter, direction maps of sewer areas, main water routes and a fine river network can be created separately based on the DEM and available coverage maps. Finally, the general flow direction map can be overlaid by the flow direction map of sewers, drain ditches, and streams subsequently. In such a way, normal flow directions are altered fundamentally by the presence of artificial drainage systems. The derived flow direction map is then used for further drainage structure delineation. Local slopes are normally derived from a DEM and calculated from the 3x3 neighbourhood using the average maximum technique with GIS. These are used in several fundamental equations that form the foundation of the model. Among them, the channel slope is most important, because all water from individual cells contributes to the river flow. The slope of a channel cell is determined by the elevation difference and distance between the upward and downward cells along a streamline. Therefore, the channel slope should be calculated separately from the general slope map in the model using the available DEM data and stream network information.

3.2.

Soil and land use

In addition to the topography, soil and land use properties are utilized in the WetSpa model to specify the land surface characteristics that determine the partitioning of incident rainfall into infiltration and runoff, as well as the simulation of subsurface flow and the vertical water and energy budget. Parameters, which depend upon soil type and land use, are incorporated in the model as attribute tables of the land use and soil type maps. The soil texture is classified based on the US Department of Agriculture (Soil Survey Staff, 1951) classification in the model for the identification of soil dependent parameters, such as porosity, field capacity, permanent wilting point, residual moisture content, pore size distribution index, saturated hydraulic conductivity, etc. An assumption made in the model is that soil hydraulic properties remain constant throughout the root zone for each grid cell. Default values are available in the model, but can be adjusted by the user to more appropriate values in case more specific information is available. 74

Flood modelling for complex terrain using GIS and Remote sensed information

The digital land use map is normally obtained from a high-resolution remotely sensed image. Fourteen land use classes are identified in the WetSpa model, which are significantly different from each other on the basis of their effects on hydrological processes. Each of these classes is characterized by quantitative attributes, as for instance, canopy resistance, albedo, root depth, interception capacity, Mannings roughness coefficient, etc. The proportion of impervious area and bare soil are considered separately for setting up the overall land use dependent parameters for each grid cell. Using the roughness map derived from the land use, combined with the slope map and hydraulic radius map derived from the DEM, the average flow velocity, celerity and dispersion coefficient can be calculated for each cell. Consequently, the average travel time and standard deviation are obtained using the weighted GIS flow length routine, and used for determining flow response function for each grid cell. Afterwards, the land use is regrouped into six classes, namely forest, pasture, crop, bare soil, urban and open water surface, for deriving potential runoff coefficient and depression storage capacity in combination with slope and soil texture classes.

3.3.

Spatial hydrological input and output

Rainfall and potential evapotranspiration are the two meteorological variables needed for the WetSpa model. If the Penman-Monteith equation is used for estimating potential evapotranspiration, the data of net radiation, air temperature, relative humidity, and wind speed are required. The method of Thiessen polygons is applied to interpolate precipitation and other meteorological variables observed at different meteorological stations. This implies that the rainfall or other meteorological variables of each grid cell is set to the value recorded at the nearest gauging site. The average rainfall and potential evapotranspiration for each subwatershed can be estimated by integration of the values on the cells belonging to that subwatershed. For long-term flood simulation in a watershed, the initial soil moisture content distribution is less important, as it affects runoff production only in the initial part of the simulation. However, for short-term flood prediction or event based flood simulation, the antecedent soil moisture condition is one of the most important factors in storm-runoff generation. The concept of topographic wetness index (TWI) (Beven 75

Chapter IV & Kirkby, 1979) is introduced to the model to evaluate antecedent moisture condition in a watershed with TWI = ln(A/S), where A [L2] is the upslope drained area, and S [-] the local slope. The TWI distribution can be easily obtained from a high resolution DEM. Those cells with high TWI values have larger upslope contributing areas or smaller element slopes or a combination of the two properties that lead to accumulation of soil moisture. The antecedent moisture distribution can be obtained by simply relating moisture content to the TWI values. Cells with very high TWI values are assumed saturated and are generally distributed along the main river or in the depression areas of the watershed. WetSpa model computes and generates time series of flow hydrographs at selected stations in a watershed and maps of spatial outputs. These maps include interception, surface runoff, infiltration, soil moisture, actual evapotranspiration, percolation, interflow, as well as the gross and net rainfall distribution. For each of these, maps in GIS format can be saved at a specified time increment, which can be used for graphical presentation to see the complete temporal and spatial variation of each of the above state variables during a model simulation.

4.

Model application

4.1.

Watershed description and data availability

The model is tested on a 67.8 km2 catchment, Barebeek, which is located in the downstream part of the Dijle River basin, Belgium. It is a typical suburban area situated northeast of Brussels (Figure IV-1). The Brussels international airport is located in the upper area of the catchment. Four main traffic lines cross the watershed in different directions, and many country roads crisscross the area from one village to another. The Leuven canal passes through the area in the north. A small lake covering about 0.55 km is located near the basin outlet. Several residential areas with sewer systems exist in the watershed occupying about 28% of the total area. The watershed drainage system, together with the main civil infrastructures and measuring stations are presented in Figure IV-2.

76

Flood modelling for complex terrain using GIS and Remote sensed information
Netherlands North Sea Brugge Gent
Mechelen
W S
0 20 40 km

Netherlands
Antwerp

City City River River

Barebeek Barebeek Flanders Flanders

N E

Brussel
France Wallonia (South-Belgium)

Leuven

Figure IV-1: Location of the Barebeek catchment

MO6

# Rain gauge
Stream gauge River Canal High way Sewer area Boundary
N W S E

$ P1 $ # $ $ MO1
MO2

P1

Land use Crop Grass Forest Urban Water

$
MO3

P2

$ $ #MO4 MO5

N W S
4 km

4 km

Figure IV-2: Drainage system of the Barebeek catchment

Figure IV-3: Land use map of the Barebeek catchment

The study area is discretized into grid cells of 50 m by 50 m. The topography is digitalized from 1/10,000 maps and the soil types are obtained from the physical system map of Flanders. The catchment is rather low and flat with average basin slope of 0.63%. Elevation differences are small with extreme values ranging from 5 to 68 m. The dominant soil types are sandy loam (66.2%) and loamy sand (29.8%), while the rests are sand, loam, silt loam and clay scatted around the catchment. The soil cover is obtained from the digital land use map of Flanders, which is based on remote sensed data of 1995. The resampled land use map for use in the model is presented in Figure IV-3. The study area is well vegetated. Forest (16.8%) is predominant in the river valleys, while the higher terrains consist of agricultural areas, with pasture (24.7%) and crops (36.9%), strongly intermixed with urban areas (16.2%), as villages, roads and Brussels international airport in the south. 77

Chapter IV

The study area has a maritime temperate climate with no proper distinction between rainy and dry seasons. Rainfall is relatively uniformly distributed through out the year, but storms have low intensities and long durations in winter, and are intense and of short duration in summer. High runoff occurs in winter and low runoff in summer due to dry soils and high evapotranspiration. Heavy storms usually last 2 to 3 days with peak rainfalls concentrated into 3 to 6 hours. The average annual precipitation in the region is about 800 mm, and the annual potential evaporation from free water surface is around 650 mm. During the period of Dec. 1998 to Feb. 1999, an intensive hydrological research was carried out to study the water quantity and quality in the Barebeek watershed. Two temporary rainfall gauges and five stream gauges were set up, MO1 to MO5, as shown in Figure IV-2. MO6 is a regular flow monitoring station located in the downstream part of the river with a relatively long discharge record. The main meteorological station is situated at Ukkel, located south of Brussels about 12 km away from the catchment. At this station, a very long meteorological record starting from the year 1898 is available, which can be used for model simulation.

4.2.

Model calibration

Calibration of the model was performed using the measured precipitation and discharge data during the period of Dec. 1998 to Feb. 1999. The PET data was obtained from Ukkel station. Predicted hydrographs were compared with the measured streamflows at each stream gauge. Model calibration was performed at the most down gauging site MO6, while simulation results at other stream gauging sites can be seen as model verifications. Further verification was carried out at MO6 for the period of Sept. 1998 to Dec. 1999, considering the seasonal parameter verification and using the available precipitation and PET data of Ukkel. Figure IV-4 shows the distribution of potential runoff coefficients that result from the different slope, soil type and land use class combinations. The watershed is covered by either dense plants or impervious areas. Impervious areas have a significant influence on runoff production, because they can generate direct runoff even during small storms. Due to the model grid size, cells may not be 100% impervious in reality. In this study, the percentage of impervious area in a grid cell is computed based on 78

Flood modelling for complex terrain using GIS and Remote sensed information land use classes, with 30% for residential areas, 70% for commercial and industrial areas and 100% for streams, lakes and roads. Default runoff coefficients for these areas are calculated by adding the impervious percentage with a grass runoff coefficient multiplied by the remaining percentage. This results in potential runoff coefficients of 40 to 100% in urban areas, while other areas have much smaller values, down to 3% for forests in valleys with practically zero slopes.

Runoff coefficient
0.0 0.1 0.1 0.2 0.2 0.3 0.3 0.4 0.4 0.5 0.5 1.0
& $ % &

Flow time (h)

05 5 10 10 15 15 20 20 25 25 30 30 40
N

N W S
0 2 4 km 0

W S
2

4 km

Figure IV-4: Distribution of potential runoff coefficient

Figure IV-5: Distribution of average flow time to the catchment outlet

A flow direction map is generated considering the effect of sewer systems. The average hydraulic radius is computed for normal floods, i.e. with a return period of 2 years. The parameters of network constant and the geometry scaling exponent are set to 0.07 and 0.43 respectively resulting in a minimum hydraulic radius for overland flow of 0.005 m and a maximum hydraulic radius for channel flow of 0.5 m at the watershed outlet. These two parameters can be increased for more extreme floods. Next, the overland flow velocity is calculated using Mannings equation. The urban areas have a remarkable influence, due to the artificial drainage facilities as sewer systems that result in flow velocities of order 0.5 m/s or more. On the other hand, the overland flow velocity in valley areas is very small, due to the high resistance of the soil cover, being mostly forest, and the very faint slopes. With this information the flow path response functions are calculated for each cell using the diffusive wave approximation. The exponent in Equation 4.2 is set to 1 after model calibration.

79

Chapter IV This is possible because that there are no extreme rainfalls during the simulation period. Figure IV-5 shows the resulting average flow time from each grid cell to the outlet of the watershed, in which the average flow time is less than 10 hours for the main river and up to 40 hours for the most remote areas. A comparison of calculated and observed discharges at the most downstream measuring station MO6 is presented in Figure IV-6. A comparison at an internal site MO3 for the same period is shown in Figure IV-7. The simulation results for other measuring stations are similar. Six storms occurred during this calibration period, with quite small rainfall intensities lower than 3 mm/h, but lasting for relatively long periods, i.e. 2 to 5 days for each. Flow volumes are mainly determined by the baseflow, about 82% for MO6 and 71% for MO3. Peak discharges result from surface runoff during each flood, and are mainly generated in urban areas.

1.0

Precipitation (mm/h)

Precipitation Measured Calculated

0.6

Precipitation Measured Calculated

0.4

0.2

0 12/20/98 12/30/98

1/9/99

1/19/99

1/29/99

10 2/8/99

0.0 12/20/98 12/30/98 1/9/99

1/19/99 1/29/99

10 2/8/99

Figure IV-6: Observed Vs calculated discharges at MO6

Figure IV-7: Observed Vs calculated discharges at MO3

Discharge (m/s)

Precipitation 3
Measured
Calculated

10

15

20

0 9/1/98

10/19/98

12/6/98

1/23/99

3/12/99

4/29/99

6/16/99

8/3/99

9/20/99

11/7/99

25 12/25/99

Figure IV-8: Observed Vs calculated flow hydrographs at MO6 for the period of Sept. 1998 to Dec. 1999 80

Precipitation (mm/h)

Precipitation (mm/h)

0.8

Discharge (m/s)

Discharge (m/s)

Flood modelling for complex terrain using GIS and Remote sensed information

A comparison at MO6 for the period of Sept. 1998 to Dec. 1999 is presented in Figure IV-8. One can notice a reasonable agreement between the simulation results and the observed hydrograph. Peaks in the hydrograph are rather well simulated, as well as for their shape and time of occurrence. From the 16 months simulation results, about 11% of the total rainfall is lost by interception, 53% of the total rainfall returns to the atmosphere as evapotranspiration, and 32% is recharged to the groundwater, which mainly happens during the winter season. The simulated flow volume is composed of surface runoff (28%), groundwater flow (70%), and Interflow (2%). Interflow is not an important flow component in this study area, due to the fact that the slope is too small to generate lateral flow in the unsaturated zone. Infiltrated water either stays in the soil and is lost as evapotranspiration, or is recharged to the saturation zone for generation of baseflow. Floods occur frequently in the winter season, because of saturated soils and high baseflow, even though the storms were not very intensive. The largest storm occurred on August 10, 1999, with maximum rainfall intensity of 15 mm/h, and the accumulative rainfall was 39 mm within 15 hours. This storm did not result in a severe flood, since the antecedent soil moisture of the watershed was much lower than the field capacity, thus leading to very high infiltration and storage in the soil. Additionally, a large amount of rainfall was lost by interception and depression storage during the initial phase. The measured peak discharge was 1.43 m/s, and the calculated peak discharge was 1.82 m/s. Figure IV-9 shows the simulated spatial distribution of relative soil wetness (/s) on Oct. 8, 1999, before the main storm. It is found that higher soil wetness was present in river valleys and lower wetness in the upper areas. In general, the soil moisture content is quite small in the watershed due to little precipitation and high evapotranspiration in the previous month. Additionally, it is seen from the figure that the relative soil wetness in the downstream areas close to the river outlet was much lower than that in other areas. This is because the area is covered by sandy soils with rather small water holding capacity, and most soil water contributed to the evapotranspiration and percolation. Figure IV-10 gives the simulated surface runoff produced in the following hour. High surface runoff was mainly generated from the impervious areas and open water surfaces, while surface runoff for other areas was rather small, especially for the areas covered by forest and sandy soils.

81

Chapter IV

Relative saturation 0.1 0.2 0.2 0.3 0.3 0.4 0.4 0.5 0.5 0.6 0.6 0.7 0.7 0.8 0.8 0.9 0.9 1.0
N W S
0 2 4 km

Surface runoff (mm) 0.0 1.0 1.0 2.0 2.0 3.0 3.0 4.0 4.0 5.0 5.0 6.0 6.0 7.0 7.0 8.0 8.0 15
N

W S
0 2

4 km

Figure IV-9: Simulated distribution of relative soil wetness on 8/10/1999, 4:00

Figure IV-10: Simulated distribution of surface runoff on 8/10/1999, 4:00-5:00

The model performed well with the Nash-Sutcliffe efficiency criteria for the calibration period, with water volume under estimated by 2%, the efficiency for reproducing river discharges by 72.4%, and the ability to reproduce low flow and high flow by 83.8% and 76.8% respectively. This indicates that the model is suitable for both peak flow prediction and hydrograph simulation in this watershed.

4.3

Model application using the historical and IDF data

To demonstrate the usefulness and performance of the model, a historical 102-year series of precipitation data from Ukkel was processed under the present land use condition. The daily and hourly potential evapotranspiration was estimated from the historical records of Ukkel through statistical analysis. The resulting hydrographs for the whole catchment were then analyzed statistically to determine the characteristics of peak discharges. Also, hourly design storms developed by Willems (2000) were introduced as rainfall input to the model for calculating corresponding design floods to compare with the modelling results of the 102-year rainfall series. The IDF relationships were established based on the long rain gauge record of 10-min precipitation depths for the period 1967-1993 at Ukkel, which is the same rainfall station for the model simulation

82

Flood modelling for complex terrain using GIS and Remote sensed information of the Barebeek. Storms of two different types, air mass thunderstorms and cyclonic/frontal storms, are separated based on their distribution of peak-overthreshold intensity. This is done for each of the durations in the range of 10 min to 15 days, using a two-component exponential distribution. Different mixtures of the two type storms are estimated for summer and winter conditions.

100 Summer Winter 10

10
Summer
Winter

Discharge (m/s)
336

P (mm/h)

0 0 48 96 144 192 240 288

0 0 48 96 144 192 240 288 336

T (h)

Time (h)

Figure IV-11: Design summer and winter storms with 100-year return period

Figure IV-12: Simulated floods for the 100-year design storms

The hourly design summer and winter storms with 15 days duration and 100 year return period are shown in Figure IV-11, in which the peak storm intensities are 36.3 mm/h and 13.1 mm/h respectively for summer and winter design storms, while the storm volumes are the same, i.e. 174 mm. Summer design storms are more intensive corresponding mostly to air mass thunderstorms, and winter design storms are less intensive corresponding mostly to cyclonic/frontal storms. The resulting flood hydrographs from the 100-year design storms were simulated with the WetSpa model, as shown in Figure IV-12, resulting in peak discharges of 8.55 m/s and 5.96 m/s respectively for the design summer and winter storms. The winter floods were simulated with the assumption that the initial soil moisture content was equal to the field capacity, while for summer design floods, the initial soil moisture content was assumed to be half of the field capacity. Figure 12 shows clearly that the design summer storm produces higher peak discharge and lower flood volume compared with that induced from the design winter storm due to the high rainfall intensity, high evapotranspiration rate, and low antecedent soil moisture. These results reflect the typical pattern of summer and winter floods of the catchment on one hand and the importance of soil moisture in controlling the runoff production on the other hand.

83

Chapter IV

The comparison of design peak flows for the two methods is presented in Figure IV13. For each return period, the design winter peak flow is smaller than the design summer peak flow, but is getting closer with each other for short return periods. The statistical result of the 102-year series model simulation is very close to the result calculated from the summer design storms. This is because the annual maximum flood occurs mostly in the summer season.

10

10

0 4 8 12 16 20 9/6/96

Discharge (m/s)

Discharge (m/s)

8 6 4 2 0 8/23/96 8/25/96

0 1 10 100

8/28/96 8/31/96

9/3/96

Return period (y)

Time (m/d/y)

Figure IV-13: Comparison of the design peak flow discharges

Figure IV-14: Simulated maximum flood at the watershed outlet

The simulated largest flood at the watershed outlet on Aug. 30, 1996 is presented in Figure IV-14. It shows a typical pattern that is present in most of the precipitation events in the study area that lead to flood discharges. The actual storm was preceded by another storm 5 days before the main event. The simulated antecedent soil moisture was very low at that moment with average moisture content around 43% of the saturation capacity. This resulted in a very small actual runoff coefficient for pervious soils, and the surface runoff was mainly produced from the impervious areas. This storm did not lead to flooding, but increased the root zone average soil moisture to field capacity, or about 54% of the saturation capacity. On Aug. 29, 1996, a storm event with 108 mm fell on the entire catchment, causing a severe flood in the watershed. The calculated surface runoff was generated from every grid cell of the catchment. Complete saturation of the root zone occurred in the river valleys and the areas with sandy soils, forest cover and very flat slopes, but not on the entire watershed. The average moisture content of the watershed was 70% of the saturation capacity after the storm event. This information, together with the simulated stream

84

Precipitation (mm/h)

Summer Winter Model

Precipitation Discharge

Flood modelling for complex terrain using GIS and Remote sensed information flow hydrographs, gives a more complete view of the hydrological behaviour and allows a better understanding of the hydrological processes.

5.

Discussion and Conclusion

A physically based distributed hydrological model compatible with remote sensing and GIS is presented for simulating the hydrological behaviour of a watershed. The generation of runoff depends upon rainfall intensity and soil moisture status and is calculated in function of slope, land use and soil type. The runoff is subsequently routed through the watershed along flow paths determined by the topography using a diffusive wave transfer model that leads to response functions between any start and end point, depending upon slope, flow velocity and dissipation characteristics along the flow paths. The model can predict not only the flood hydrograph at any controlling point of the river, but also the spatially distributed hydrological processes, such as surface runoff, soil moisture, interflow, groundwater recharge, actual evapotranspiration and so on, at each time step during a simulation. The modified runoff coefficient for surface runoff production is no longer the conventional runoff coefficient used in the rational method, but a measure of rainfall partitioning capacity, which allows varying with time, rainfall intensity, rainfall duration and the cell geophysical characteristics. With such improvements, the model can be used for the computation of storm hydrographs for any size watershed. Processes in water and energy transfer between soil, plant and atmosphere are simulated using simplified equations for each grid cell. Among these, water balance in the root zone is important, as the moisture in the root zone is a key factor to control the amount of surface runoff, interflow, actual evapotranspiration and groundwater recharge. All model parameters can be obtained from DEM, land use and soil type data of the watershed or combinations of these three fundamental maps. The spatial distribution of rainfall intensity, potential runoff coefficient and the antecedent soil moisture content are governing factors of the flood volume, while the hydraulic radius and the channel roughness coefficient are sensitive for flow routing simulations. GIS provides a powerful platform for developing the model, calibrating parameters and displaying model results in a spatial way, so that it is possible to capture local 85

Chapter IV complexities of a watershed and compare model results to the field measurements. The model was validated on a small watershed in Belgium for which topography and soil data are available in GIS form, while the land use and soil cover was obtained from remote sensed images. The resulting calculated hydrographs compare favourably with measurements. The usefulness and utility of the model are subsequently demonstrated by forecasting peak discharges resulting from an observed 102 years precipitation series. The resulting discharges were analyzed statistically to determine the characteristics of extreme flood events and compared with the results computed from design storms. Comparison of the two methods shows that the model is capable to predict both normal and extreme floods. For flood modelling on a large catchment scale, computing with small grid size leads to huge memory cost and is time consuming. To solve this problem, the model can be easily converted to a semi-distributed model, where water and energy budget are maintained for each very small subwatershed derived from the high resolution DEM, hydrographs at each subwatershed outlet are firstly calculated using GIS derived subwatershed response function and then routed to the basin outlet using the channel response function. Model parameters and meteorological data input for each subwatershed can be obtained by integration of the values from all cells of that subwatershed. This has the advantage of maintaining the correct internal drainage structure within each subwatershed, which could not be the case with a distributed model with larger grid cells. However, division of the catchment should be done with caution in order to take care of the spatial variability of the hydrological processes. The model makes full use of the remote sensed data and calculations are for the most part performed by standard GIS tools, such that the model is especially useful for flood prediction on complex terrain and analyzing the effects of topography, soil type, and land use or soil cover on the flood. Additionally, the model can be easily coupled with other water quality and soil erosion models, and used for simulating spatial hydrological behaviour of a river basin.

86

Flood modelling for complex terrain using GIS and Remote sensed information References ASCE, Design and Construction of Sanitary and Storm Sewers, Manuals and Reports of Engineering Practice, No. 37, New York, 1969. Band, L.E., Topographic partition of watersheds with digital elevation models, Water Resour. Res., 23, 15-24, 1986. Beven, K.J. & Germann, P., Macropores and water flow in soils, Water Resour. Res., 18, 1311-1325, 1982. Beven, K.J., Kirkby, M.J., Schoffield, N. & Tagg, A., Testing a physically based flood forecasting model (TOPMODEL) for three UK catchments, J. Hydrol., 69, 119-143, 1984. Borah, D.K., Xia, R. & Bera M., DWSM - A dynamic watershed simulation model, eds., Singh V.P. and Freyert D.K., Mathematical Models of Small Watershed Hydrology and Applications, 113-166, Water Resour. Publ., Highlands Ranch, Colorado, 2002. Browne, F.X., Stormwater management, ed., Corbitt, R.A., Standard Handbook of Environmental Engineering, 7.1-7.135, McGraw-Hill, New York, 1990. Chow, V.T., Maidment, D.R. & Mays, L.W., Applied Hydrology, McGraw-Hill, New York, 1988. De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrological modelling on a catchment scale using GIS and remote sensed land use information, 295-304, ed., Brebbia, C.A., WTI press, Boston, 2000. Eagleson, P.S., Climate, soil, and vegetation, a simplified model of soil moisture movement in liquid phase, Water Resour. Res., 14(5), 722-730, 1978. Ewen, J., Parkin, G. & O'Connell, P.E., SHETRAN: distributed river basin flow and transport modelling system, J. Hydrological Eng., 5, 250-258, 2000. Famiglietti, J.S. & Wood, E.F., Multiscale modelling of spatially variable water and energy balance processes, Water Resour. Res., 30(11), 3061-3078, 1994. Fortin, J.P., Turcotte, R., Massicotte, S., Moussa, R., Fitzback, J. & Villeneuve, J.P., A distributed watershed model compatible with remote sensing and GIS data, I: Description of the model, J. Hydrological Eng., 6, 91-99, 2001. Garrote, L. & Bras, R.L., A distributed model for real-time flood forecasting using digital elevation models, J. Hydrol., 167, 279-306, 1995. Hillel, D., Application of Soil Physics, Academic Press, New York, 1980. 87

Chapter IV Horritt, M.S. & Bates, P.D., Effect of spatial resolution on a raster based model of flood flow, J. Hydrol., 253, 239-249, 2001. Julien, P.Y., Saghafian, B. & Ogden, F.L., Raster-based hydrological modelling of spatially-varied surface runoff, Water Resour. Bull., 31, 523-536, 1995. Kirkby, M.J., Hill-slope Hydrology, 235, John Wiley & Sons, Ltd., 1978. Kuichling, E., The relation between rainfall and the discharge in sewers in populous district, Trans., AECE, No 20, 1889. Linsley, Ray K., Jr., Kohler, M.A. & Joseph Paulhus, L.H., Hydrology for Engineers, 237, McGraw-Hill, New York, 1982. Liu, Y.B., Gebremeskel, S., De Smedt, F. & Pfister, L., Flood prediction with the WetSpa model on catchment scale, eds., Wu et al., Flood Defence 2002, 499507, Science Press, New York Ltd., 2002. Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. & Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modelling, J. Hydrol., 283, 91-106, 2003. Lull, H.W., Ecological and silvicultural aspects, ed., Chow, V.T., Handbook of Applied Hydrology, 6.1-6.30, McGraw-Hill, New York, 1964. Maidment, D.R., Handbook of Hydrology, McGraw-Hill, Inc., New York, 1993. Mallants, D. & Feyen, J., Kwantitatieve en kwalitatieve aspecten van oppervlakte en grondwaterstroming (in Dutch), 96, Vol. 2, KUL, 1990. Moore, I.D., Grayson, R.B. & Ladson, A.R., Digital terrain modelling: a review of hydrological, geomorphologic, and biological applications, Hydrol. Process., 5, 3-30, 1991. Myrobo, S., Tempral and spatial scale of response area and groundwater variation in till, Hydrol. Process., 11, 1861-1880, 1997. Palacios-Velez, O.L. & Cuevas-Renaud, B., SHIFT: A distributed runoff model using irregular triangular facets, J. Hydrol., 134, 35-55, 1992. Pilgrim, D.H. & Cordery, I., Flood runoff, ed., Maidment, D.R., Handbook of Hydrology, 9.1-9.42, McGraw-Hill, New York, 1993. Quinn, P., Beven, K., Chevallier, P. & Planchon, O., The prediction of hillslope flow paths for distributed hydrological modelling using digital terrain models, Hydrol. Process., 5, 59-79, 1991. Rinaldo, A., Marani, A. & Rigon, R., Geomorphologic dispersion, Water Resour. Res., 27, 513-525, 1991. 88

Flood modelling for complex terrain using GIS and Remote sensed information Robson, A.J., Whitehead, P.G. & Johnson, R.C., An application of a physically based semi-distributed model to the Balquhidder catchments, J. Hydrol., 145, 357-370, 1993. Rowe, L.K., Rainfall interception by an evergreen beech forest, Nelson, New Zealand, J. Hydrol., 66, 143-158, 1983. Sheaffer, J.R., Wright, K.R., Taggart, W.C. & Wright, R.M., Urban Storm Drainage Management, Marcel Kekker, New York, 1982. SINCE, WatfloodSPL8, Flood forecasting system, developed for surveys and information branch ecosystem science and evaluation directorate, Environment Canada, 1972. Singh, B. & Szeicz, G., The effect of intercepted rainfall on the water balance of a hardwood forest, Water Resour. Res., 15, 131-138, 1979. Singh, V.P., Elementary Hydrology, Prentice Hall, Englewood Cliffs, New Jersey, 1992. Soil Survey Staff, Soil Survey Manual, Handbook, No. 18, US Department of Agriculture, 1951. Thornthwaite, C.W. & Mather, J.R., The Water Balance, Laboratory of Climatology, Publ. No. 8, Centerton NJ., 1955. Wang, Z., Batelaan, O. & De Smedt, F., A distributed model for Water and Energy Transfer between Soil, Plants and Atmosphere (WetSpa), Phys. Chem. Earth, 21, 189-193, 1997. Wigmosta, M.S., Vail, L.W. & Lettenmaier, D.P., A distributed hydrology vegetation model for complex terrain, Water Resour. Res., 30(6), 1665-1679, 1994. Willems, P., Compound intensity/duration/frequency-relationships of extreme precipitation for two seasons and two storm types, J. Hydrol., 233, 189-205, 2000. Wittenberg, H. & Sivapalan, M., Watershed groundwater balance estimation using streamflow recession analysis and baseflow separation, J. Hydrol., 219, 20-33, 1999. Zinke, P.J., Forest interception studies in the United States, eds., Sopper, W.E. & Hull, H.W., International Symposium on Forest Hydrology, 137-161, Pergamon Press, Oxford, 1967.

89

Chapter V Assessing land use impacts on flood processes using a GIS modelling approach
Abstract A distributed hydrological modelling and GIS approach is applied for the assessment of land use impact in the Steinsel sub-basin, Alzette, Grand-Duchy of Luxembourg. The assessment focuses on the runoff contributions from different land use classes and the potential impact of land use changes on runoff generation and flow responses. The results show that the direct runoff from urban areas is dominant for a flood event compared with runoff from other land use areas in this catchment, and tends to increase for small floods and for the dry season floods, whereas the interflow from forested, pasture and agricultural field areas contributes to the recession flow. For assessing the hydrological effects of land use changes, three scenarios, urbanization, deforestation and afforestation, are established and investigated. Significant variations in flood volume, peak discharge, time to the peak, etc., are found from the model simulation based on the three hypothetical land use change scenarios.

1.

Introduction

Flood risk is among the most severe risks on human lives and properties, and has become more frequent and severe along with local economical development. As the earths population has been growing rapidly and more stress is put on the land to support the increased population, hydrological resources are affected both on local and global scale. One of the recent thrusts in hydrological modelling is the assessment of the effects of land use and land cover changes on water resources, and the influence on storm runoff generation is one of the main research topics in the last decade. The influence of land use on storm runoff generation is very complex as land use and soil cover have an effect on interception, surface retention, evapotranspiration, and resistance to overland flow. For instance, cropland and urban land yield more flood

Chapter V volumes, higher peak discharges and shorter flow travel times than grassland or woodland. Increased runoff from cropland is mainly due to the removal of native vegetation and soil compaction, which decrease soil infiltration capacity. Increased runoff from urban areas results from impervious surfaces that prevent infiltration of water into soils. Urban land uses also reduce the surface roughness and therefore shorten the overland flow retention time. In contrast, less runoff is produced from undisturbed grassland and woodland areas. This is due to factors such as interception of precipitation by the vegetation canopy, the dense network of roots that increase infiltration capacity and soil porosity, as well as the accumulated organic debris on the surface that increases depression storage capacity and overland flow retention time. Moreover, dense vegetation causes higher evapotranspiration and affects the longterm water and energy balance. Evidently, areas with a high percentage of cropland or urban land use yield more storm runoff than the areas of similar soils and topography with grassland or woodland. Land uses generally do not affect the slope and soil composition. Although certain practices such as terraced agriculture, mining, and residential landscaping can potentially have drastic effects on the slope and the physical properties of the soil. Land use can have a big effect on the water quality of rivers, lakes and estuaries, since land reclamation, cultivation, deforestation and urbanization influence the water environment by non-point source pollution, soil erosion, sedimentation, etc. Bultot et al. (1990) employed a conceptual hydrological model on a small catchment in Belgium to assess the impacts of possible land use changes on the water balance. The study was carried out by converting the present diversified vegetation cover for the un-built areas into a single vegetation species, and simulation results showed that minimum streamflow occurs for a coniferous cover and maximum streamflow for pasture. Nandakumar and Mein (1997) combined a Monte-Carlo simulation method with a conceptual rainfall-runoff model to examine the effect of random errors in land use model parameters on flood predictions. One of their conclusions is that small changes in catchment land use may cause changes in runoff, which cannot be detected statistically from the output errors with a lumped model. Lorup et al. (1998) adopted a methodology combining common statistical methods with conceptual hydrological modelling to distinguish between the effects of climate variability and the effects of land use for six semi-arid catchments in Africa. Their analysis indicated a decrease in 92

Assessing land use impacts on flood processes using a GIS modelling approach the low flow for most of the testing catchments, with the largest changes occurring for catchments located within communal land, where large increases in population and agricultural intensity have taken place. Karvonen et al. (1999) developed a semilumped model to study the influence of land use on catchment runoff. The modelling is based on the subdivision of the catchment into smaller units, which aggregate areas of hydrologically similar behaviour, e.g., land use, soil, slope, and vegetation, and allows calculating runoff from different land use areas separately. In their model, a characteristic profile was used as a basic component to represent various sizes of homogenous land use areas and the hydraulic properties were defined at the profile scale and were applied as such for areas of different sizes. In general, conceptual rainfall-runoff models usually consider the entire catchment or sub-catchment as one unit, and describe the transformation of rainfall to runoff with simple concepts. Due to limitations in the model conceptualizations of the hydrological processes involved, lumped catchment models must be used carefully in predicting the impacts of land use change on catchment runoff (Kuczera et al., 1993). In this type of analysis, it is desirable that the hydrological modelling describes the spatio-temporal variability of anthropogenic effects so that the assessment could reflect the variability of the hydrological parameter at the required scales. These kinds of hydrological models have the advantage of reflecting the effects of spatially distributed model parameters such as land use on stream flows. As for instance, Niehoff et al (2002) combined spatially distributed land use scenarios and a processoriented hydrological model to study land use impacts on storm runoff generation in a meso-scale catchment in SW-Germany. The simulation results showed that the influence of land use conditions on storm runoff production depends largely on the rainfall event characteristics and on the related spatial scale. However, distributed physically-based rainfall-runoff models use a high number of parameters which are difficult to determine, and are usually accompanied by a high degree of data and parameter uncertainty, which impose limitations to the prediction of land use impacts on the hydrological behaviour. Along with the rapid development of GIS technology and remote sensing techniques, especially the concomitant availability of high resolution DEM and the advances in integrating GIS with hydrological modelling, hydrological assessment with distributed models tends to be more advantageous and competent by linking GIS with hydrological modelling. 93

Chapter V

In this chapter, a GIS modelling approach for predicting flood hydrograph and runoff contribution from different land use areas and assessing the land use change on flood processes is presented. The model takes into account the spatial heterogeneity of the basin parameters to predict flood hydrographs and spatially distributed hydrological characteristics in a watershed, and therefore making it suitable for analysis of the effect of land use change on stream flows. The input of the model includes observed data of precipitation and evaporation together with parameters derived from a combination of a DEM, land-use and soil map in raster format. The model is validated by comparing calculated and observed hourly discharges for a 52 months period at four stream flow stations in the Steinsel watershed, located in the upstream part of the Alzette River basin, Grand-Duchy of Luxembourg. Three land use scenarios, urbanization, deforestation and afforestation, are considered for assessing the effects of land use change on runoff and flow processes in the Steinsel catchment.

2.

Methodology

2.1.

Description of the WetSpa model

WetSpa is a grid-based distributed hydrological model for water and energy transfer between soil, plants and atmosphere, which was originally proposed by Wang et al. (1997) and adopted for flood prediction on a variable time step by De Smedt et al. (2000), and Liu et al. (2002, 2003). The detailed theory and formulation of the WetSpa model have been described in chapter III and IV. For each grid cell, four layers are considered in the vertical direction as vegetation zone, root zone, transmission zone and saturated zone. The hydrological processes considered in the model are precipitation, interception, depression, surface runoff, infiltration, evapotranspiration, percolation, interflow, ground water flow, and water balance in the root zone and the saturated zone. The total water balance for a raster cell is composed of the water balance for the vegetated, bare-soil, open water and impervious parts of each cell. This allows accounting for the non-uniformity of the land use per cell, which is dependent on the resolution of the grid. The processes in each grid cell are set in a cascading way, which means that an order of occurrence of the processes is assumed after a precipitation event. A mixture of physical and 94

Assessing land use impacts on flood processes using a GIS modelling approach empirical relationships is used to describe the hydrological processes in the model. The model predicts peak discharges and hydrographs, which can be defined for any numbers and locations in the channel network, and can simulate the spatial distribution of catchment hydrological characteristics. The simulated hydrological system consists of four control stores: the plant canopy, the soil surface, the root zone, and the saturated groundwater aquifer. Among the process variables, soil moisture content is a crucial factor in the model as it affects the hydrological processes of surface runoff, actual evapotranspiration, interflow and percolation out of the root zone. The precipitation that falls from the atmosphere before it reaches the ground surface is abstracted by canopy interception storage. The remaining rainfall reaching to the ground is separated into rainfall excess and infiltration. Rainfall excess is calculated using a moisture-related modified rational method with potential runoff coefficient depending on the land cover, soil type, slope, the magnitude of rainfall, and the antecedent moisture content of the soil. The calculated rainfall excess fills the depression storage at the initial stage and runs off the land surface simultaneously as overland flow (Linsley, 1982). The infiltrated part of the rainfall may stay as soil moisture in the root zone, move laterally as interflow or percolate as groundwater recharge depending on the moisture content of the soil. Both percolation and interflow are assumed to be gravity driven (Famiglietti & Wood, 1994) in the model. Percolation out of the root zone is equated as the hydraulic conductivity corresponding to the moisture content as a function of the soil pore size distribution index (Eagleson, 1978). Interflow is assumed to occur in the root zone after percolation and becomes significant only when the soil moisture is higher than field capacity. Darcys law and a kinematic approximation are used to estimate the amount of interflow generated from each cell, in function of hydraulic conductivity, the moisture content, slope angle, and the root depth. The actual evapotranspiration from soil and plant is calculated for each grid cell using the relationship developed by Thornthwaite and Mather (1955) as a function of potential evapotranspiration, vegetation and stage of growth, and moisture content in the cell. A percentage of the remaining potential evapotranspiration is taken out from the water content in the groundwater reservoir as a function of the maximum reservoir storage, giving the effect of a steeper baseflow recession during dry period. The total evapotranspiration is the sum of evaporation from intercepted water, depressed water and the bare soil 95

Chapter V surface, and the transpiration from plants through the root system and a small part from the groundwater storage. A simple structure is used in the model because the emphasis here is on developing and testing parameterizations for the root zone. Excess runoff, infiltration, evapotranspiration, interflow and percolation estimates are point calculations. Different slope, land use and soil properties in different grid cells of a watershed result in different amounts of excess runoff when subjected to the same amount of rainfall. Runoff from different cells in the watershed is routed to the watershed outlet depending upon flow velocity and wave damping coefficient by using the diffusive wave approximation method. A two parameter, mean travel time and its variance, approximate solution proposed by De Smedt et al. (2000) in the form of an instantaneous unit hydrograph (IUH) was used in the model relating the discharge at the end of a flow path to the available runoff at the start of the flow path. The mean travel time and its variance for each grid cell are spatially distributed, and can be obtained by integration along the topographic determined flow paths as a function of flow celerity and dispersion coefficient. Although the spatial variability of land use, soil and topographic properties within a watershed are considered in the model, the groundwater response is modelled on small GIS-derived subcatchment scale due to the fact that groundwater flow is much slower than surface flow and little is known about the bedrock. The simple concept of a linear reservoir is used to estimate groundwater discharge on a small subwatershed scale, while a non-linear reservoir method is optional in the model with storage exponent of 2 (Wittenberg & Sivapalan, 1999). The groundwater outflow is added to any runoff generated to produce the total streamflow at the subwatershed outlet. All model equations are specifically chosen to maintain a physical basis and well supported by previous studies.

2.2.

Description of the study area

The Steinsel watershed covers approximately 407 km and is located in the upstream part of the Alzette river basin. The study area is situated in the southern part of the Grand Duchy of Luxembourg, with a small part in the south located in France, as shown in Figure V-1. The elevations in the watershed range from 450.0 m to a low elevation of 225.5 m at the watershed outlet, with an average basin slope of 7%. 96

Assessing land use impacts on flood processes using a GIS modelling approach Figure V-2 shows the topographic elevation map and measuring stations in the watershed. The local topography is characterized by a natural sandstone bottleneck, located near Luxembourg-city. The valley is up to 2.5 km wide upstream of the bottleneck, and only 75 m in the Luxembourg sandstone, which extends approximately 80 m into the ground (Pfister et al., 2000).

Stei nsel Loren tzweiler# Waldhaff Mull endo rf # # M amer # B ea l ir # Reckan ge/Mess # Pfaffe nthal
#Fin del

River Rive r Boundary Bou nd ary Ele Elevation vatio n (m) (m ) 225 -2 225 - 70 270 m Hesperange Contem # 270 -3 270 - 15 315 315 -3 315 - 60 360 360 4 05 360 - 405 405 - 4 50 405 - 450
N W S
0 5 10 k

Raingauge # Raingauge #

Streamg au

Streamgau

Belvaux # Schifflan ge #

Livange Ro eser #

Figure V-1: Location of the Alzette basin and Steinsel sub-basin

Figure V-2: Topography and gauging network of the Steinsel sub-basin

The dominant soil types are loamy sand (29.1%) and silt (37.7%) distributed in the higher terrains, while the rests are silt clay loam (13.3%), sandy clay loam (10.2%) and clay loam (9.5%) mainly in the river valleys, and other soil types covering very small areas scattered around the catchment as shown in Figure V-3. The watershed has undergone rapid urbanization and extensive cultivation since the 1950s. Urban areas cover about 20.5% of the watershed with Luxembourg-city in the downstream and Esch-Alzette city in the upstream part. Cultivated lands occupy about 22.1% of the total area distributed beside the river valleys with main crop types of maize and wheat. Forest (28.9%) and grass (24.4%) are predominant in the river valleys and the high terrain, intermixed with urban areas and cultivated lands. There are also some former mining areas located in the high terrain of the upstream watershed covering about 2.5% of the total area, where surface runoff is seldom generated. The watershed is well drained with a dense stream network, open water occupies about 1.6% of the total area. Table V-1 gives a general description of the area, average slope and main soil types for each land use class.

97

Chapter V

Table V-1: Description of the area, slope and main soil types for each land use class
Land use Cropland Grassland Woodland Mining area Urban Water surface Total Area (km) 94.0 97.3 115.6 10.2 83.4 6.5 407.0 Relative area (%) 23.1 23.9 28.4 2.5 20.5 1.6 100.0 Average slope (%) 5.77 4.73 10.10 11.50 5.93 1.31 7.01 Main soil types Silt, loamy sand Silt, clay loam, sandy clay loam Silt, loamy sand, silt clay loam Loamy sand Silt, silt clay loam Clay loam, silt clay loam Silt, loamy sand, silt clay loam

The climate in the region has a northern temperate humid oceanic regime without extremes. The mean annual temperature is around 10C, with average temperature of 0.7C in January and 17.3C in July (Pfister et al., 2002). Rainfall has a relatively uniform distribution throughout the year. High runoff occurs in winter and low runoff in summer due to the higher evapotranspiration. Winter storms are strongly influenced by the westerly atmospheric fluxes that bring humid air masses from the Atlantic Ocean (Pfister et al., 2000), and floods happen frequently because of the saturated soils and low evapotranspiration. The average annual precipitation in the region varies between 800 mm to 1,000 mm. Precipitation generally exceeds potential evapotranspiration except for four months in the growing season.

2.3.

Data collection

Three digital base maps are prerequisite in the model to define the watershed drainage work and derive spatial model parameters, i.e. DEM, soil type and land use. A DEM with 5050 m grid size for the watershed was constructed using 2-meter resolution elevation contours and the official river network. Information of soil types was obtained from the digital 1:100,000 Soil Map of the European Communities, and converted to 12 USDA soil texture classes based on textural properties. The land use information was obtained from the digital land use map of Luxembourg and France derived from remote sensed image with respect to the watershed condition in the year 1995. The original land use map was classified to fourteen classes for use in the WetSpa model, and reclassified to five hydrological land use classes for simulation of 98

Assessing land use impacts on flood processes using a GIS modelling approach storm runoff partitions from different land use areas, i.e., crops, grassland, forest, urban areas, surface water, and mining areas, as shown in Figure V-4.

Sand Sand Loamy san Loamy sand Silt Silt Sandy Sandy clay clay Silt Silt clay clay loa Clay Clay loam coam

Crop Grass Forest Mining Urban Water

N W S E

N W S
10

10 km

Figure V-3: Soil type map of the Steinsel sub-basin

Figure V-4: Land use map of the Steinsel sub-basin

A dense hydrological observation network has been set up in the Alzette river basin, as shown in Figure V-2, where 4 stream gauges, namely Steinsel, Pfaffenthal, Hesperange and Livange, are located in the study area and recording water levels at a 15-minute time step, and 10 rain gauges are located in and around the watershed recording at an hourly or daily time step. Daily rainfall was disaggregated into hourly rainfall series according to the nearest hourly reference rain gauges for being used in the model. Potential evapotranspiration was estimated using the Penman-Monteith formula (Monteith & Unsworth, 1990) with daily meteorological data measured at the Luxembourg airport, and extended to each rainfall Thiessen polygon based on the proportions of different land use type over the polygon (Drogue et al., 2002). A total of 52 months of hourly rainfall, discharge and potential evapotranspiration data from December 1996 to March 2001 are available for model application. The average flow at Steinsel during the monitoring period is 5.6 m/s, with flows ranging from 0.07 to 40.7 m/s, and the measured maximum hourly rainfall intensity is 23 mm/h, which occurred on July 2, 2000.

99

Chapter V 2.4. Model calibration and verification

The application procedures for WetSpa include database development, watershed segmentation, model calibration and validation. Model parameters are identified firstly using GIS tools and lookup tables, which relate default model parameters to the base maps, or the combination of these base maps. Starting from a 50 by 50 m pixel resolution digital elevation map of the Steinsel catchment, hydrological features including surface slope, flow direction, flow accumulation, flow length, stream network and drainage area are delineated. Maps of porosity, field capacity, wilting point, residual moisture content, saturated hydraulic conductivity and pore size distribution index are obtained from the soil type map. Maps of root depth, Mannings roughness coefficient and interception storage capacity are derived from the land use map. Maps of potential runoff coefficient and depression storage capacity are obtained from the slope, soil type and land use combinations. Impervious areas have significant influence on runoff production in a watershed, because they can generate direct runoff even during small storms. Due to the model grid size, cells may not be 100% impervious in urban areas. In practice, the percentage of impervious area in a grid cell is computed based on land use classes, with 30% for residential area, 70% for commercial and industrial area and 100% for streams, lakes and bare exposed rock. Default potential runoff coefficients for these areas are calculated by adding the impervious percentage with a grass runoff coefficient multiplied by the remaining percentage. This leads to runoff coefficients of 40 to 100% in urban areas, while other areas have much smaller values, down to 5% for forests in valleys with practically zero slopes. The model was calibrated against hourly streamflow measurements at the four stations for the time period of December 1996 to December 1999, while the period of January 2000 to March 2001 was used for model validation. The calibration was not carried out for all model parameters, but for some global parameters only, including the evapotranspiration correction factor in controlling water balance, the interflow scaling factor in controlling the amount of interflow, and the groundwater flow recession coefficient governing the routing process of baseflow. Other parameters, such as potential runoff coefficient, soil properties, overland flow roughness coefficient, interception and depression storage capacity, etc., were set to values 100

Assessing land use impacts on flood processes using a GIS modelling approach obtained from the literature, which have shown to yield reliable results in previous model applications on different basin (De Smedt et al. 2000; Liu et al. 2003). Calibration of the evapotranspiration factor can be performed independently by comparing the calculated and observed flow volume for a long time series. The interflow scaling factor is calibrated by matching the computed discharge with the observed discharge for the recession part of the flood hydrograph. Groundwater flow recession coefficient can be obtained by the analysis of recession curves at discharge gauging stations. Refinement of this coefficient is necessary to get a better fit for the low flows. Inputs to the model are spatially distributed precipitation interpolated by the method of Thiessen polygon and potential evapotranspiration, while the outputs are hydrographs at each gauging site and the simulated spatial distribution of hydrological characteristics. Model performance for calibration and validation were evaluated through qualitative and quantitative measures, involving both graphical comparisons and statistical analysis for hourly, daily and monthly values. In addition to the above comparisons, the water balance components for individual land uses were reviewed by displaying model results including precipitation, surface runoff, interflow, baseflow, actual evapotranspiration and groundwater recharge. Although observed values were not available for each of the water balance components listed above, the average annual values must be consistent with expected values for the region, depending upon the individual land use categories. This is a separate consistency check with data independent of the modelling to ensure that land use categories and overall water balance reflect local conditions. Finally, the spatial outputs of simulated hydrological variables were used to assess the reasonability of hydrological processes distribution, where the processes of surface runoff, soil moisture, interflow, and percolation etc. would be spatially distributed depending upon the cells physical characteristics. Figure V-5 gives a graphical comparison between simulated and observed hourly streamflows at Steinsel for a sequence of floods that occurred in February and March 1997. The total rainfall was 184.3 mm with measured runoff of 107.8 mm and simulated runoff of 111.0 mm. A small flood happened in early February, followed by three large floods successively. The simulated hydrographs of surface runoff, 101

Chapter V interflow and baseflow correspond to respectively 38%, 27% and 35% of the predicted total flood volume, which were obtained by summation of flow responses from all contributing cells. The figure shows a very good agreement between the predicted and measured hydrograph, in which the rising and high water limb are dominated by surface runoff, while interflow is a few hours delayed and mainly contributes to recession flow. Groundwater discharge forms the baseflow of the total hydrograph. Due to the high antecedent soil moisture content and groundwater storage, the amount of interflow and groundwater flow is abundant in these flood events, being 62% in total of the whole flood volume.

Figure V-5: Observed Vs calculated flow at Steinsel for the floods in Feb. 1997

40

30

30

Precipitation Measured Calculated

20 60

10

90

0 1/00 2/00 3/00 4/00 5/00 6/00 7/00 8/00 9/00 10/00 11/00 12/00

120

Figure V-6: Observed and calculated daily flow at Steinsel for the year 2000

102

Daily precipitation (mm/d)

Daily discharge (m/s)

Assessing land use impacts on flood processes using a GIS modelling approach Figure V-6 presents a graphical comparison of calculated and observed daily flows at Steinsel for the validation year 2000. The year 2000 was a very high flow year in the region with an annual precipitation of 1004 mm and an annual mean discharge of 6.61 m/s at Steinsel, which is 1.4 times the average flow for the previous 3 years. Floods happened both in winters due to the saturated soils and high groundwater storage, and in summer due to intensive rainfall intensity. With the initial hydrological condition at the end of the simulation period, the validation results for the year 2000 are in a good agreement with the measured daily discharges. This indicates that a fairly high degree of model precision is obtained, and the general hydrological trends are well captured by the model.

2.5.

Model evaluation

For the assessment of the model performance and model efficiency for the simulation period, 5 evaluation criteria were selected as listed in Table V-2. In all equations, Qs and Qo are the simulated and observed streamflows at time step i, Qo is the mean
observed streamflow over the simulation period, and N is the number of time steps. CR1 is the model bias, for which the value 0 represents a perfect simulation of the flow volume. CR2 is a model determination coefficient representing the proportion of the variance in the observed discharges that are explained by the simulated discharges with the best value of 1. The Model efficiency is measured by the Nash-Sutcliffe coefficient (Nash & Sutcliffe, 1970) expressed as CR3. A CR3 of 1 indicates a perfect fit, while a negative CR3 means that the prediction is worse than simply using the observed mean. CR4 is a logarithmic transformed Nash-Sutcliffe criterion, giving emphasize for evaluating the quality of low-flow simulations (Smakhtin et al., 1998). An arbitrarily small value may be introduced to the discharges in case of zero flows for which the logarithm does not exist. Moreover, an adapted version of the NashSutcliffe criterion CR5 is proposed, which is in fact a combination between the calibration criteria used by Guex (2001) for the hydrological study on the Alzette river basin and the HEC-1 objective function (USACE, 1998). As seen in the formula, more weight is given to high discharges, and therefore, the criterion CR5 can be used for evaluating model efficiency for high flows.

103

Chapter V Table V-2: Evaluation criteria for the assessment of model performance
Code Criteria CR1 Description
N i =1 i N

Qs Qo
i =1 i

1
2 N

Model bias for evaluating the ability of reproducing water balance

i =1 i

CR2

(Qs
N i =1
N i =1

Qo

) (Qo
N i =1

Qo

Determination coefficient representing the simulation variance Model efficiency for evaluating the ability of reproducing streamflows
2

2 CR3 1 (Qs i Qoi )

(Qo

Qo
N

2 CR4 1 [ln (Qs i ) ln (Qo i )] i =1 N

[ln (Qo ) ln (Qo )]

i =1 i

Model efficiency for evaluating the ability of reproducing low flows

2 CR5 1 (Qoi + Qo )(Qsi Qoi ) i =1

(Qo
N i =1

+ Qo Qoi Qo

)(

Model efficiency for evaluating the ability of reproducing of high flows

The model performance was evaluated both qualitatively by visual comparison of the simulated and observed hydrographs and quantitatively using the above statistical indexes at Steinsel and other three stations inside the catchment. The model performance is found to be satisfactory as illustrated in Table V-3, which shows the subcatchment area, the percentage of main land use classes, and the results of the five assessment criteria for both calibration and validation periods on hourly scale. Model biases are within the range of -0.04 to 0.02, and the mean value is 0.81, 0.80 and 0.85 respectively for the three efficiency criteria, which indicate that the model has a high confidence and can give a fair representation of both low-flow and high-flow hydrographs for the study catchment. Table V-3: Watershed characteristics and model performance
Station Livange Area Urban Crop Grass Forest (km) (%) 233 18.6 17.8 19.2 20.5 (%) 28.9 27.4 25.4 23.2 (%) 22.9 25.3 26.8 24.3 (%) 24.7 25.4 25.2 29.0 Period CR1 CR2 CR3 CR4 CR5 0.82 0.80 0.87 0.84 0.92 0.87 0.85 0.86

Calibration -0.02 0.76 0.78 0.83 Validation -0.04 0.79 0.75 0.78 Calibration -0.03 0.81 0.83 0.78 Validation -0.02 0.75 0.79 0.81 Calibration 0.02 0.80 0.81 0.82 Validation -0.03 0.82 0.80 0.76 Calibration 0.01 0.87 0.85 0.83 Validation -0.03 0.83 0.84 0.82

Hesperange 291 Pfaffenthal Steinsel 350 407

104

Assessing land use impacts on flood processes using a GIS modelling approach 3. Results and discussion

3.1.

Evaluating runoff partitions from different land use classes

Since the WetSpa model calculates runoff and flow path response in a spatial way for each grid cell, it is capable of evaluating storm runoff partitions from different land use areas. By convolution of the flow responses from the cells belonging to a certain land use category, runoff partition to the flood hydrograph from this land use category can be estimated. The total flow hydrograph at the watershed outlet is obtained by the sum of runoff partitions from different land use areas in the watershed. For assessing runoff partitions from different land use areas of the catchment, the calibrated WetSpa model is run for the whole simulation period, and the flow components for different land use classes are calculated at each time step. Figure V-7 gives a graphical presentation for the same flood event used in the model calibration but shows storm runoff contributions from different land use classes. Clearly, surface runoffs from urban areas, cropland and grassland form the high water peak of the hydrograph, representing 39.1%, 11.6% and 9.0% respectively of the storm runoff (excluding baseflow). Interflow from woodland, grassland and cropland yields 16.7%, 8.8% and 7.5% of the storm runoff. Other storm runoff components in the figure are mainly surface runoff from water surfaces and forested areas, accounting to about 7.2% of the storm runoff, while surface runoff and interflow from mining areas and interflow from urban areas are negligible for this flood event.

Figure V-7: Storm runoff partitions at Steinsel for the flood events in Feb. 1997

105

Chapter V

Another simulated partitioning of hourly outflow at Steinsel for a flood event, that occurred in April 1999, is given in Figure V-8, for which the runoff contribution from urban, forested, pasture, agricultural, water surface and mining areas are identified separately. The total rainfall is 49.7 mm with a measured runoff of 21.3 mm and a calculated runoff of 19.9 mm. Runoff from urban areas, being 35.4% of the total flood volume (including baseflow), dominates the high water flow in this flood event. A part of rainfall infiltrates in urban areas in gardens and parks, and is used mainly for evapotranspiration. Therefore, interflow and groundwater recharge in urban areas are very small and contribute very little to the total runoff. Runoff from woodland (16.4%), grassland (23.2%) and cropland (22.3%) are basic components of the flood hydrograph, contributing to the flow in the period of peak flow and recession, as well as baseflow. The interflow and baseflow contribution from forested areas is the highest compared to other land use areas, because it is one of the most common land types in the catchment, and moreover, most rainfall falling on woodland infiltrates into the soil and contributes to interflow in areas with steep slopes or otherwise to groundwater recharge. Note that a small flood occurred 2 days before the main flood, for which runoff from urban areas was dominant, being 58% of the flood volume, while other runoff contributions were relatively small compared to the main flood. Other runoff contributions for this flood event are direct flows from water surfaces (3.5%), and groundwater drainage from mining areas (1.3%). Surface runoff and interflow from mining areas are negligible in the catchment.

Figure V-8: Storm runoff contributions at Steinsel for the flood events in Apr. 1999

106

Assessing land use impacts on flood processes using a GIS modelling approach

In order to analyze the controlling factors on runoff at the basin outlet from different land use classes, 18 flood hydrographs with peak discharges higher than 20 m/s within the simulation period are selected, and the partitions as well as the flow coefficient for the flood event and relative errors in flood volume and peak discharges are calculated individually. Table V-4 contains the simulated runoff partitions for the selected storm events occurring at Steinsel and the statistics of the simulation errors for each flood event. Table V-4: Simulated runoff contributions from different land use classes
Other contributions (%) Urban contribution (%) Cropland contribution (%) Grassland contribution (%) Woodland contribution (%) Flow coefficient (-) Observed peak discharge (m/s) 40.2 27.5 33.9 31.5 40.5 32.9 38.9 39.9 32.6 24.8 21.4 23.8 24.3 25.8 26.1 28.6 40.7 32.6 31.4 Observed flood volume (mm)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

10/02-06/03 97 17/12-24/12 97 01/01-12/01 98 15/01-24/01 98 23/10-07/11 98 12/04-18/04 99 10/12-17/12 99 23/12-02/01 00 03/03-06/03 00 09/07-12/07 00 13/07-19/07 00 24/07-31/07 00 18/08-23/08 00 08/10-14/10 00 06/11-11/11 00 21/11-30/11 00 01/01-10/01 01 21/01-31/01 01 Average

156.0 43.2 56.7 35.0 174.7 48.4 79.4 67.0 19.4 25.3 41.6 63.3 43.4 42.0 37.5 49.4 70.3 63.3 62.0

34.1 32.0 22.6 22.2 29.3 39.9 35.1 22.9 23.8 42.0 38.3 35.2 53.7 35.7 26.9 22.7 24.0 22.7 31.3

20.9 22.0 21.7 23.1 22.8 23.7 24.1 22.3 23.0 24.1 22.9 23.7 22.4 23.5 22.6 22.6 22.5 23.4 22.8

23.8 22.1 25.1 25.2 22.1 18.7 22.1 24.2 25.5 22.7 19.3 22.4 11.8 22.1 24.3 24.8 23.9 24.6 22.5

16.1 19.0 26.2 25.1 20.8 12.7 13.5 25.8 23.1 6.0 14.3 13.7 5.9 14.0 21.6 25.4 25.1 24.8 18.5

5.1 4.9 4.5 4.4 4.9 5.1 5.1 4.8 4.6 5.1 5.2 4.9 6.3 4.6 4.6 4.4 4.6 4.5 4.9

82.2 25.6 43.5 28.7 73.7 19.2 30.4 47.4 15.4 9.8 12.3 19.3 8.3 17.6 19.6 29.0 38.3 39.7 31.1

1.6 -9.4 -10.0 -17.8 6.0 -10.6 1.0 -18.1 -13.2 -10.8 18.1 13.5 18.6 -9.3 -5.9 2.7 5.0 -1.5 -2.3

0.55 0.61 0.80 0.85 0.44 0.41 0.40 0.73 0.82 0.40 0.31 0.32 0.20 0.43 0.54 0.61 0.56 0.65 0.53

12.7 -7.9 -8.7 -12.4 15.7 -15.2 16.7 -6.0 -10.2 -14.8 16.4 16.6 -1.8 -5.1 -17.9 -7.5 14.1 -11.8 -2.0

The flow coefficient for a storm event defined in Table V-4 is the ratio of the outflow water volume at the catchment outlet to the volume of water precipitated over the catchment during this event. The simulated flow coefficients can be computed in a similar manner by incorporating the simulated flow volume at the basin outlet and 107

Error in peak discharge (%)

Error in flood volume (%)

Flood period (d/m-d/m y)

Precipitation (mm)

No.

Chapter V should be close to those of the observed flow coefficient. As can be seen in the table, both flow coefficient and runoff partitions from different land use areas vary from one storm event to another, because they depend upon antecedent soil moisture, groundwater storage and storm behaviour. The WetSpa model takes the soil moisture content and rainfall intensity into account, and makes it possible to better explain these variations. Variations in runoff contribution and runoff partition are directly tied to soil moisture and groundwater storage. Among the 18 storm events listed in the table, 13 occurred during the winter season, and 5 events between April and November. Winter storms are usually characterized by high flow coefficients, due to the high soil moisture content and high groundwater storage, causing high baseflow, interflow and saturation overland flow. Under such conditions, an amount of the river discharges is generated from natural areas. This can be illustrated by the two flood events which occurred in January 1998 and one flood in March 2000, in which flow coefficients were higher than 0.80, and the runoff partitions from urban, agricultural, pasture and forested areas were of the same magnitude. This indicates that groundwater drainage plays an important role in the winter season, which is mostly produced by previous storms. For instance, the simulated baseflow accounts to 30% of the total volume for the flood event in March 2000, and 38% for the flood in January 1998. However, the urban contribution increases greatly if we consider only the storm runoff excluding baseflow. For instance, the simulated urban contribution increases from 22.6% of the total runoff to 31.2% of the storm runoff for the first flood in January 1998, and from 22.2% to 36.5% for the second flood in the same month. On the contrary, summer storms usually have low flow coefficients, due to the low soil moisture content and low groundwater storage. Runoff from urban areas is dominant in all flood hydrographs of this catchment, while other contributions are relatively small, especially the runoff from forested areas. An extreme example is the storm event, which occurred in August 2000, for which the total rainfall was 43.3 mm, causing a peak discharge of 24.3 m/s at Steinsel and a flood volume of 8.3 mm with a flow coefficient of 20%. The calculated runoffs from urban, agricultural, pasture and forested areas were 53.7%, 22.4%, 11.8% and 5.9% respectively. The soil was very dry before the storm, and most rainfall infiltrated in the soil resulting in rather small runoff contributions to the flood event. 108

Assessing land use impacts on flood processes using a GIS modelling approach

60

(a) Land-use classes contributions

Urban Grassland Others Cropland Woodland

2.5 2.0 1.5 1.0 0.5

(b) Urban relative contribution

40

20

0 1 5 9 13 17

0.0 0.0 0.2 0.4 0.6 0.8 1.0

Event
2.5 2.0 1.5 1.0 0.5 0.0 0.0 0.2 0.4 0.6 0.8 1.0

Flow coefficient
2.5 2.0 1.5 1.0 0.5 0.0 0.0 0.2 0.4 0.6 0.8 1.0

(c) Cropland relative contribution

(d) Grassland relative contribution

Flow coefficient
2.5 2.0 1.5 1.0 0.5 0.0 0.0 0.2 0.4 0.6 0.8 1.0

Flow coefficient
2.5 2.0 1.5 1.0 0.5 0.0 0.0 0.2 0.4 0.6 0.8 1.0

(e) Woodland relative contribution

(f) Other relative contribitions

Flow coefficient
20 10 0 -10 -20 0.0 0.2 0.4 0.6 0.8 1.0

Flow coefficient
20 10 0 -10 -20 0.0 0.2 0.4 0.6 0.8 1.0

(g) Error on flood volume

(h) Error on peak discharge

Flow coefficient

Flow coefficient

Figure V-8: Plot of event to event variations of the different runoff contributions (a), normalized relative runoff contribution from urban (b), cropland (c), grassland (d), woodland (e), and other areas (f), and error in flood volume (g) and error in peak discharge with respect to the flow coefficient

109

Chapter V

Storm characteristics, such as volume, duration, intensity and shape, have a big influence on the flow coefficient and the runoff contributions from different land use areas. Large storms with long duration and high intensity produce more runoff under similar antecedent soil moisture conditions, but the flow coefficient may not response positively due to the lower baseflow. For small rainfall events, most runoff is generated from impermeable areas and open water surfaces, while runoff from natural areas can be ignored. For the three storm events in July 2000 for example, the rainfall volume for each storm event was 25.3 mm, 41.6 mm and 63.3 mm, resulting in urban runoff of 42.0%, 38.3% and 35.2 respectively, while the contributions from other natural land use classes increased with storm volume accordingly. In addition, runoff varies also with time and storm shapes. The flow at the start of the flood stems exclusively from direct runoff from urban areas and water surfaces because of the quick response and short travel times. Other runoff contributions join the flow afterwards, but the magnitude of which is strongly depending upon soil moisture conditions. For instance, for the flood in April 1999, shown in Figure V-8, a considerable rainfall occurred 2 days before the main storm, causing a small flood in which urban runoff was dominant. However, this was no longer the case for the main flood, because the first storm also increased the soil wetness, such that runoff from agricultural, pasture and forested areas was strongly increased for the main flood. A series of plots based on the resules in Table V-4 is presented in Figure V-8 showing the event to event variations of the different runoff contributions, the evolution of each contribution normalized by the percentage of its land-use class area, the error on flood volume and the error on peak discharge with respect to the flow coefficient. As can be seen V-8a, both flow coefficient and runoff partitioning from different landuse areas vary from one storm event to another, which demonstrates the effects of antecedent soil moisture, storm characteristic and groundwater storage on the flood behaviour of the catchment. Urban contribution to storm runoff decreases with the flow coefficient with mean of 1.53 and standard deviation of 0.43 (Figure V-8b). The relative contribution value is larger than one indicating its contribution being higher than its percentage of land-use class area of the catchment. Cropland relative contribution is almost constant for different storms with mean of 0.99 and standard deviation of 0.04 (Figure V-8c), which indicates its contribution being more or less 110

Assessing land use impacts on flood processes using a GIS modelling approach equal to the cropland percentage of the catchment. The relative contribution from grassland increases slightly with the flow coefficient with mean of 0.94 and standard deviation of 0.14 (Figure V-8d). The relative contribution from woodland areas increases significantly with the flow coefficient with mean of 0.64 and standard deviation of 0.24 (Figure V-8e). The relative contribution values from grass and forest land-use class areas are less than one indicating their contributions being less than the percentage of their land-use class areas. Other contributions (mining plus water surface) decrease with the flow coefficient with mean of 1.19 and standard deviation of 0.11 (Figure V-8f), which is mainly due to the change of runoff contribution from water surface. The relative error on the flood volume strongly decreases with the flow coefficient (Figure V-8g). This indicates that the Thiessen polygon method may not give a precise estimate of the rainfall distribution for small and large storms, or may because of the model deficiency itself. The relative error on the peak discharge is presented in Figure V-8h with mean of -1.5 m3/s and standard deviation of 12.8 m3/s, which indicates that the model gives a fairly good prediction of the flood peaks.

Figure V-10: Contributions of monthly flow at Steinsel from different land use classes Figure V-10 shows the simulated contributions to the monthly flow at Steinsel from different land use classes from December 1996 to March 2001. The runoff contributions from urban (29.3%), agricultural (22.8%), pasture (22.2%), forested (21.5%), mining (1.5%) and water surfaces (2.7%) are indicated separately. The runoff contribution from urban areas is the highest, contributing mainly to direct flow.

111

Chapter V Contributions from agricultural, pasture and forested areas are more or less equal, contributing to surface runoff, interflow and baseflow, while the contribution from mining areas is the smallest contributing only to baseflow.

3.2.

Assessing the impact of land use changes on flood

Changes in land use may have significant effects on infiltration rates, on the water retention capacity of soils, on sub-surface transmissivity and thus on the runoff production. It is evident that the Alzette watershed has undergone rapid urbanization since 1950. In general, the flood potential of a catchment significantly increases by urbanization. The introduction of impervious surfaces and good drainage systems increases the volume of runoff and results in flood hydrographs which are faster to peak, faster to recede, and of increased peak discharge (Crooks, 2000). As the WetSpa model accounts for spatially distributed hydrological and geophysical characteristics of the watershed, it is suitable for assessing the impact of land use changes on hydrological behaviours in a complex terrain with reliable land use change scenarios. A realistic set up of land use scenarios requires scenarios of future regional development. Land use decision-making is strongly influenced by socio-economic factors. As these particular future land use policies are complicated, three distinct scenarios are considered in the Steinsel sub-basin of the Alzette (Figure V-11), where urban areas are increased at the expense of crops and grassland for urbanization, all forests are converted into crops or grassland for deforestation, and forests are increased at the expense of crops and grassland for afforestation. The urbanisation scenario was elaborated on the basis of information regarding the changes in land use planned by the government. The afforestation scenario was meant to recreate in a simple manner the conditions that might have prevailed some 200 years ago when the Grand-Duchy of Luxembourg was largely covered by forests. The deforestation scenario was more or less chosen to evaluate what the behaviour of the basin would be without this important forest cover. Starting from the land use map with respect to the 1995 watershed situation, the first scenario increases urban areas by 31.8%, and decreases agricultural and grasslands by 11.7% and 15.4% respectively. The second scenario reduces forested areas by almost 100%, and expands agricultural and grasslands by 108% and 16% accordingly. The third scenario increases forested areas 112

Assessing land use impacts on flood processes using a GIS modelling approach by 52.6% and decreases agricultural and grasslands by 48.8% and 16% with respect to the present land use situation (Table V-5). Table V-5: Land use change scenarios compared with the present situation
Scenarios Present Urbanization Deforestation Afforestation Urban areas 20.5 27.0 20.5 20.5 Crops (%) 23.2 20.5 48.3 11.9 Grass (%) 24.3 20.5 28.2 20.4 Forest (%) 29.0 29.0 0.0 44.2 Others (%) 3.0 3.0 3.0 3.0

Urbanization

Deforestation

Afforestation

Crops

Grassland

Forest

Urban areas

Surface water

Mining

Figure V-11: Land use change scenarios for the Steinsel sub-basin Based on these land use change scenarios, model parameters were recalculated and the model was run to deliver the modified runoff and streamflows. Figure V-12(a) shows the simulated surface runoff distribution for the storm of Feb. 24-26, 1997 under the present land use condition, corresponding to the fourth flood in Figure V-5. The total storm rainfall was 63 mm with a calculated average surface runoff volume of 12.8 mm. Due to the very high antecedent soil moisture content of this storm event, almost all areas contribute to the storm runoff but with different volumes according to their land use types. As can be seen from the map, high surface runoff was produced on surface water and urban areas, while low surface runoff occurred in the areas with forest cover and sandy soils. There was very little surface runoff generated from the former mining areas because of its specific surface characteristics, and most water in these areas recharged to the groundwater reservoir. Figure V-12(b) presents the simulated surface runoff distribution for the same storm event after urbanization. The 113

Chapter V calculated average surface runoff volume becomes 17.0 mm, being increased 32.8% compared with the result under present land use condition.
(a) (b)
Surface runoff (mm)

0- 5 5 - 10 10 - 20 20 - 30 30 - 40 > 40

N W S E

10 km

Figure V-12: (a) Simulated surface runoff distribution under present land use condition for the storm on Feb. 24-26, 1997, and (b) Simulated surface runoff distribution after urbanization for the same storm event Figure V-13 gives the simulated flood hydrographs for the present and three land use scenarios at Steinsel for a flood event that occurred on December 12, 1999. The results indicate that the urbanization scenario produces the highest peak flow, followed by the deforestation and afforestation scenario. The rainfall originated from slow moving westerly atmospheric fluxes with long duration and low intensity. The 2day rainfall was 65 mm, but the highest rainfall intensity was only 4.6 mm/h. The simulated peak discharge for the present land use is 45.3 m3/s, for the urbanization scenario 61.7 m3/s, the deforestation 50.6 m3/s and the afforestation 41.7 m3/s. Accordingly, the urbanization scenario increases the peak discharge for this storm by 36%, the deforestation scenario increases the peak discharge by 12%, while the afforestation scenario decreases the peak discharge by 8%. In addition to the difference in the magnitude of the simulated peak discharges, differences in time to peak of the modelled discharges are also observed. The peak discharge occurred around 3 hours after the main rainfall for the present condition, after 2 hours for the urbanization and deforestation scenarios, and 3 hours for afforestation scenario.

114

Assessing land use impacts on flood processes using a GIS modelling approach
80 0
70

60

Qscenario (m /s)

Q (m /s)

40

15

P (mm/h)

Rainfall Urbanisation Deforestation Present Afforestation

60

Urbanisation Deforestation Afforestaion Present

10

50

40

20 20 25

30

30

20 20 30 40 50
3

11/12/99 12/12/99 13/12/99 14/12/99 15/12/99

60

70

Time (d/m/y)

Qpresent (m /s)

Figure V-13: Simulated hydrographs for each scenario for a storm in Dec. 1999

Figure V-14: Peak discharges for each scenario over the simulation period

Figure V-14 gives the present versus the scenario peak discharges selected from the whole simulation period. It shows that afforestation has a mild positive effect in reducing the peak discharge in comparison to the present situation. On the contrary, urbanization and deforestation lead to an increase of the simulated peak discharges. In addition to the effects of land use change on flood volume, runoff composition, evapotranspiration and soil moisture were also evaluated quantitatively from the model results. It was found that urbanization and deforestation result in increasing the flood volume and the amount of surface runoff, but decreasing the amount of interflow and baseflow, as well as soil moisture and the amount of evapotranspiration from a long term simulation, while this is the contrary for afforestation. The magnitudes of changes, however, differed from one storm to another depending upon the antecedent soil moisture content. This can be explained by the fact that a change in land cover will alter the leaf area index, the interception storage capacity, the soil infiltration capacity and thus the evolution of soil moisture. High soil moisture leads to more evapotranspiration, groundwater recharge and interflow, and vice versa. Investigation of low flows indicates that the effect on baseflow is not pronounced in summer, due to the fact that most soil water is used for evapotranspiration, and the baseflow is very small for all the three scenarios. However, considerable differences in baseflow are found in winter with the afforestation scenario producing the highest baseflow, while the urbanization scenario producing the lowest baseflow.

115

Chapter V

4. Conclusions Distributed models have proven to be useful in such kind of analyses because of their ability to predict the effect of spatially changing variables, like land-use change. In this chapter, a spatially distributed continuous simulation model, WetSpa, running on hourly time scale and compatible with GIS and remote sensed information, is applied to assess the land use impacts on flood processes in the Steinsel sub-basin, Alzette, Grand-Duchy of Luxembourg. Model calibration and validation have shown the models level of representativeness to be quite satisfactory. The outflow at the catchment outlet has been especially well reproduced. 5 criteria are used to evaluate the model performance. The model bias is between the range of -0.04 to -0.02 for the four gauging stations over the simulation period, and model determination coefficient from 0.75 to 0.87, Nash- Sutcliffe efficiency from 0.75 to 0.85, and the adapted NashSutcliffe efficiency for low flow and high flow from 0.76 to 0.83 and from 0.80 to 0.92 respectively. It is demonstrated from the model simulation that the land use composition and soil moisture condition play an important role in generating flood hydrographs at basin outlet. Simulation results show that the important runoff processes, which contribute to storm runoff, are mainly surface runoff from urban areas and partly from cropland, grassland for big storms. Interflow from woodland, grassland and cropland forms the recession of the flood hydrograph, but also contributes considerably to the peak discharges for the floods in the wet season. Other areas with high infiltration and storage capacity contribute very little to the storm runoff. Simulations show that the flow coefficient and the runoff partitions from different land use classes vary from one storm event to the other due to differences in soil moisture and storm behaviour. The relative runoff contribution from urban areas decreases with the flow coefficient. Cropland relative runoff contribution tends to be a constant being more or less equal to the cropland area percentage of the catchment. The relative runoff contributions from grassland and woodland increase with flow coefficient, and toward their percentage of land-use class areas of the catchment for large storms. It can be concluded that the runoff from urban areas is dominant for a flood event compared to other land use classes in this catchment, and tends to increase for small floods and for 116

Assessing land use impacts on flood processes using a GIS modelling approach flood events with low antecedent soil moisture. Other runoff contributions tend to increase for large storms and for storm events with high antecedent soil moisture. Interflow and baseflow from natural areas are important during the wet season but not for small floods during the dry season. For assessing the hydrological effects of land use changes on floods, three hypothetical scenarios, namely urbanization, deforestation and afforestation scenario, were considered based on the present land use configuration and possible land use trends in the study area. It is found from the model simulation that the urbanization scenario has a large impact on increasing peak discharge and flood volume, as well as time to the peak. Likewise, deforestation has a fair negative impact, while afforestation has a moderate positive impact on the floods. Investigation of the peak flow shows that land-use changes can have remarkable effects on peak discharges in comparison to the present land use condition. The urbanization and deforestation scenarios increase the peak discharges by 26% and 9.1% in average, respectively, while afforestation has a positive impact, decreasing the peak flow by -5.3% in average. The model can also be used for the assessment of the land use change impacts on other hydrological processes, such as interception, depression, soil moisture, evapotranspiration, low flow, etc. in a river basin. However, since this research focuses on the estimation of flow contributions from different land use classes and the assessment of land use change effects on the floods processes, detailed discussions of these effects are not given in this report.

References

Bultot, F., Dupriez, G.L. & Gellens, D., Simulation of land use changes and impacts on the water balance - A case study for Belgium, J. Hydrol., 114, 327-348, 1990. Crooks, S., Davies, H. & Goodsell, G., Rainfall runoff modelling and the impact of land use change in the Thams catchment, in: European Conference on Advances in Flood Research, 115-130, eds., A. Bronstert, Ch. Bismuth & L. Menzel, Potsdam, Germany, 2000.

117

Chapter V De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrological modelling on a catchment scale using GIS and remote sensed land use information, In: Risk Analysis II, ed., Brebbia, C.A., 295-304, WTI press, Southampton, Boston, 2000. Drogue, G., Leviandier, T., Pfister, L., El Idrissi, A., Iffly, J.F., Hoffmann, L., Guex, F., Hingray, B. & Humbert, J., The applicability of a parsimonious model for local and regional prediction of runoff, Hydrol. Sci. J., 47, 905-920, 2002. Eagleson, P.S., Climate, Soil, and Vegetation, a simplified model of soil moisture movement in liquid phase, Water Resour. Res., 14(5), 722-730, 1978. Famiglietti, J.S. & Wood, E.F., Multiscale modelling of spatially variable water and energy balance processes, Water Resour. Res., 30(11), 3061-3078, 1994. Guex, F., Modlisation hydrologique dans le bassin versant de lAlzette (Luxembourg), Rgionalisation des paramtres dun modle global. Travail pratique de Diplme, EPFL/CRP-GL, Luxembourg, 2001. Karvonen, T., Koivusalo, H., Jauhiainen, M., Palko, J. & Weppling, K., A hydrological model for predicting runoff from different land use areas. J. Hydrol., 217, 253-265, 1999. Kuczera, G., Raper, G.P., Brah, N.S. & Jayasuriya, M.D.A., Modelling yield changes following strip thinning in a mountain ash catchment: An exercise in catchment model validation, J. Hydrol., 150, 433-457, 1993. Linsley, Ray K., Jr., Kohler, M.A. & Joseph Paulhus, L.H., Hydrology for Engineers, 237, McGraw-Hill, New York, 1982. Liu, Y.B., Gebremeskel, S., De Smedt, F. & Pfister, L., Flood prediction with the WetSpa model on catchment scale, In; Flood Defence 2002, eds., Wu et al, pp. 499-507, Science Press, New York Ltd., 2002. Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. & Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modelling, J. Hydrol., 283, 91-106, 2003. Lorup, J.P., Refsgaard, J.C. & Mazvimavi, D., Assessing the effect of land use change on catchment runoff by combined use of statistical tests and hydrological modelling: Case studies from Zimbabwe, J. Hydrol., 205,147-163, 1998. Monteith, J.L. & Unsworth, M., Principles of Environmental Physics, 291, London, Edward Arnold, 1990.

118

Assessing land use impacts on flood processes using a GIS modelling approach Nandakumar, N. & Mein, R.G., Uncertainty In rainfall runoff model simulations and the implications for predicting the hydrological effects of land use change, J. Hydrol., 192, 211-232, 1997. Nash, J.E. & Sutcliffe, J.V., River flow forecasting through conceptual model, J. Hydrol., 10, 282290, 1970. Niehoff, D., Fritsch, U. & Bronstert, A., Land use impacts on storm-runoff generation: Scenarios of land use change and simulation of hydrological response in a mesoscale catchment in SW-Germany, J. Hydrol., 267, 80-93, 2002. Pfister, L. & Hoffmann, L., Experimental hydro-climatological atlas of the Alzette river basin, Grand-Duchy of Luxembourg, Centre de Recherche Public, Gabriel Lippmann, 2002. Pfister, L., Humbert, J. & Hoffmann, L., Recent trends in rainfall-runoff characteristics in the Alzette river basin, Luxembourg, Climate Change, 45(2), 323-337, 2000. Smakhtin, V.Y., Sami, K. & Hughes, D.A., Evaluating the performance of a deterministic daily rainfall-runoff model in a low flow context, Hydrol. Process. 12, 797-811, 1998. Thornthwaite, C.W. & Mather, J.R., The water balance, Laboratory of Climatology, Publ. No. 8, Centerton NJ., 1955. USACE, HEC-1 Flood Hydrograph Package: Users Manual, U.S. Army Corps of Engineers, Davis, CA., 1998. Wang, Z.M., Batelaan, O. & De Smedt, F., A distributed model for water and energy transfer between soil, plants and atmosphere (WetSpa), Phys. Chem. Earth, 21(3), 189-193, 1997. Wittenberg, H. & Sivapalan, M., Watershed groundwater balance estimation using streamflow recession analysis and baseflow separation, J. Hydrol., 219, 20-33, 1999.

119

Chapter VI Assessing the effects of river restoration on the reduction of floods in a river basin
Abstract This chapter discusses the effects of river restoration on flood reduction in the Steinsel sub-basin of the Alzette River basin, the Grand-Duchy of Luxembourg. The basic approach of restoration is carried out firstly by classifying the streams into different orders and by assessing the response of stream channels to the resistance or obstruction of flows, and the river re-meandering in the headwater streams. Based on this assessment, the roughness to the flow in the first and second order streams is adjusted for the river restoration while the roughness of downstream higher order channels is kept unchanged. An increase of 10% of channel sinuosity is proposed in this study to account for possible future activities in river re-meandering. The hydrological analysis is performed using the WetSpa distributed hydrological model based on the spatial information of topography, soil type, and land use. Flow delay as a result of increased channel roughness offered by obstruction of the flow from the headwater channels results in a reduced peak discharge at the outlet of the basin. The model simulation indicates that peak flow can be reduced by as much as 14% and delay the time of concentration by as much as 2 hours after river restoration.

1.

Introduction

The Alzette River basin has encountered a series of serious floods since the early 1990s. For instance, floods in 1993, 1995 and 1998, caused a lot of damage to properties within municipal areas in the river basin. In addition to the positive trend noticed in winter rainfall totals due to an increase in westerly atmospheric fluxes since 1970s (Pfister et al., 2000), the change of land-use pattern in the Alzette River basin is a major reason causing rapid runoff into channelized streams, which may increase flood frequency and enhance downstream peaks in flood hydrographs. The proportion of urban areas in the Alzette River basin had increased by 30% between 1954 and

Chapter VI 1979, and by 15% between 1979 and 1995. Major changes in the future land-use are also anticipated. The changes in land-use pattern within a watershed and especially adjacent to rivers and streams have a great influence on the river hydrological processes. For instance, cropland increases runoff due to the removal of native vegetation and soil compaction, which decreases the soil infiltration capacity. Urbanization increases runoff even more due to impervious areas, reduced vegetation cover and depression storage, and by concentrating and accumulating runoff by sewer systems. Such impacts result in increasing the volume of surface runoff, increasing the velocity and concentration time of storm runoff, reducing infiltration into the soil and ultimately reducing baseflow (Schueler, 1994). Urbanization can also change the balance of forces in the stream towards channel incision. In addition to the aforementioned effects, urbanization can damage the zone surrounding the channel that influences the hydrology and ecology. Trees and vegetation along the bed of the riverbank play an important role in the hydrodynamic behaviour and the ecological equilibrium of a river. Therefore, urbanization reduces the amount of wood and vegetation that enter into the river and can deprive the stream of stabilizing elements that help to dissipate the flow energy (Booth, 1991). The valleys of the upper Alzette River have been densely populated for centuries. Modification of rivers and their riparian areas has been undertaken considerably. Physical degradation has been particularly great since 1950s as a result of artificial drainage, and flood defence structures. Such changes also occurred in the headwater areas of the river basin. Many stream reaches were straightened and channelized, and the landscape formerly covered by forest is converted into farmland. As pointed out by Nienhuis and Leuven (2001), such practices may lead to irreversible changes, not only in riverine habitats caused by the disruption of the natural evolution of backwaters, but also in the disappearance of aquatic-terrestrial transition zones. Because of the change in the response to rainfall of a basin following urbanization, peak flows become higher causing more flood damage compared to the predevelopment stage. Previous studies have shown the local negative effect of heavily urbanized areas on peak runoff in small streams in the vicinity of the city of Luxembourg (Pfister et al., 2002). 122

Assessing the effects of river restoration on the reduction of floods in a river basin

Channelization is one of the common approaches to reduce floods caused by urbanization of a basin. However, flood conveyance benefits of channelization are often offset by ecological losses resulting in an increased stream velocity and reduced habitat diversity. Channelization by creating smoother and faster flow also increases the risk of flooding in downstream areas. Because of these negative effects, there is a trend to restore streams to their pristine conditions to minimize similar flood repercussions in the future. Riparian plants, which influence strongly the headwater streams, play a key role to dam up the stream water and have major impacts on lowland stream ecosystems. Increasing the capacity of high water retention in noncritical areas may reduce the extent of flooding in the downstream inhabited areas, and improve the ecological stabilization of the watercourse with reinforcement of natural metabolisms. Currently, stream restoration programs are seeking to improve aquatic habitats and regulate stream flows by natural means. The goal of stream restoration is to restore the stream to a more natural form to create environmentally favourable conditions, which do not necessarily imply that the stream will be restored to its pre-settlement condition (Morris, 1995). The effects will be reduction of damage to property, risk of accident, making water flow rates favourable to the development of aquatic life and avoiding damage due to destructive erosion. The purpose of this study is to investigate the effect of a conceptual headwater restoration on flooding in the Steinsel sub-basin of the Alzette River basin mainly located in Luxembourg. The restoration is considered for first and second-order streams, characterized by moderate steep slopes, flowing through forest and grassland intermixed with agricultural and urban areas, and therefore potentially suitable for river restoration. The simulation results illustrate the importance of headwater restoration on the reduction of flood peak discharge for the downstream main channels. The analysis is performed with the WetSpa model, which is a GIS-based distributed runoff and flow routing model calculating hourly runoff occurring at any point in a watershed and providing spatially distributed hydrological characteristics in the river basin. The study is not intended to address the biological and ecological value of river restoration, nor does it focus on the planning and design of future river restoration activities.

123

Chapter VI 2. Methodology

2.1.

Model description

The WetSpa model is used to assess the possible effects of river restoration on flood reduction in a river basin. The model is a grid-based distributed hydrological model for water and energy transfer between soil, plants, and atmosphere. The model was originally developed by Wang et al. (1997) and adapted for flood prediction by De Smedt et al. (2000) and Liu et al. (2003). The theory and formulas of the model have been described in detail in chapter III and chapter IV, and will not be discussed repeatedly in this chapter.

2.2.

Description of the study area

A case study is performed in the 407 km2 Steinsel sub-basin, Alzette, Grand Duchy of Luxembourg, as shown in Figure V-1. Detailed descriptions about the topography, soil type, land use, climate, geology, hydrology, gauging network, and data available of the study area have been provided in chapter V, and will not be addressed repeatedly in this chapter. According to Pfister et al. (2002), the maximum stormflow coefficient, which is the maximum slope of the double-mass curve of rainfall and stormflow, is stable for each subcatchment of the Steinsel sub-basin from one winter to the next in recent years. The maximum stormflow coefficient for the upper part of the basin (sandy soils, deciduous shrub mixed with agricultural land) is 0.29, whereas the lower part has a maximum stormflow coefficient of up to 0.64, due to the extensive sewage system of Luxembourg City. The average stormflow coefficient is 0.37 for the entire Steinsel sub-basin. The maximum stormflow coefficient is strongly related to the basin characteristics as shape of drainage area, topography, soil type, land-use, etc.

2.3.

Stream classification

The classification of stream channels can be helpful in the interpretation and assessment of the response of stream channels to the resistance or obstruction of flow. 124

Assessing the effects of river restoration on the reduction of floods in a river basin Based on this assessment and interpretation a decision can be made on which section of the basin to focus in order to restore streams. There are different methods of classifying streams depending on the intended use. Morphologically described stream types (Rosgen, 1994) are delineated by slope, channel materials, width/depth ratio, sinuosity, and entrenchment ratio. This classification system has been widely used in stream restoration and mitigation, because it allows for predicting the behaviour of the system, extrapolating knowledge of one system to another, and provides a consistent frame of reference for communication among those interested in these systems. Usually streams are classified according to their number of tributaries and confluences. Smaller tributaries are assigned the lowest order and main rivers are assigned the highest order. One of the most popular methods for assigning stream orders was proposed by Strahler (1957). The uppermost channels in a catchment with no upstream tributaries are first order. The confluence of two first-ordered streams gives a second-order stream, and so on, but the confluence of a channel with another channel of lower order does not raise the order of the stream after the confluence. The main purpose in classifying the streams of the Steinsel basin is to make a sound judgement about the areas of the basin that can be restored to mitigate floods. Figure VI-1 shows the different stream orders and their specific characteristics of the Steinsel basin according to Strahlers method of ordering that is extracted from a 50m by 50m resolution DEM of the basin.

(a)
Drained area 1st order 2nd order 3rd order 4th order 5th order

(b) 80
60

Stream length Drained area Average slope

Percentage (%)

40

N W S
0 5 10 km

20

0 1 2 3 4 5

Stream order

Figure VI-1: (a) Stream orders and their drained area, and (b) percentage of stream length, percentage of drained area and average slope for different order streams 125

Slope (%)

Chapter VI

The streams of the study area are ordered from first to fifth order corresponding to the threshold value of 100 cells when delineating the stream network based on the flow accumulation theme. This implies that the flow concentration is sufficient to initiate a channel if the drainage area is greater than 0.25 km2. Figure VI-1a represents the area of the sub-basin that is drained by different order streams. Among them, the first and second order streams drain about 322 km2 area representing 79% of the total study area (Figure VI-1b). In addition the length of the first and second order stream constitutes 425 km and around 76% of the total length of the GIS extracted stream network of the Steinsel sub-basin (Figure VI-1b). Headwater streams are predominantly accumulators, processors, and transporters of materials from the terrestrial system. Therefore, restoring these stream channels can have significant implications in the reduction of downstream floods in the river basin. Overlapping of the stream order and the land-use of the Steinsel sub-basin shows that the first and second order streams are channels that usually pass through forest and grassland. These stream channels have a moderate gradient with calculated average slopes of 0.013 and 0.016 for the first and second streams respectively (Figure VI-1b). In addition, the characteristics of these streams are much more influenced by riparian vegetation and geomorphology than the downstream higher order channels.

2.4.

Modelling approach

Environmentally acceptable river restoration prefers features such as non-uniform cross-sectional profiles, vegetation, meanders, islands, riffles and pools, in order to have diverse habitats (Hansen, 1996). In addition, allowing nature restoration by wood and debris to fall in streams for regulating river flows is becoming an important component of current stream and river restoration (Larson et al., 2001). However, one of the difficult tasks is to predict the amount and distribution of flow obstacles in the streams, which is not uniform throughout the whole stream network and depends on the morphology of the river system and the dynamics of the flow. Generally, small channels tend to contain abundant wood and debris that is distributed randomly, and affected by the density and species composition of the riparian area. Due to variable roughness and extra turbulence, estimation of flow behaviour after restoration is rather difficult. 126

Assessing the effects of river restoration on the reduction of floods in a river basin

Because of the lack of prior detailed studies in the Alzette headwater streams about the channel morphology and the distribution of woods and vegetation in the streams and their movement during and after floods, a conceptual river restoration is proposed based on some facts of the basin. In the study area, the small streams originate upstream on the plateaus with channel riparian areas mainly characterised by forest land-use, while much of the downstream area is characterised by urban settlement. The first and second order streams have steeper slopes than that of higher order streams (Figure VI-1b), which have a potential of eroding and entraining the riparian vegetation. The mean bank-full width of these streams is generally less than 2 m, and the mean bank-full depth smaller than 0.5 m, which are favourable conditions to collect woody debris that can form obstacles and increase the resistance to the flow. Moreover, first and second order streams flow through forest areas enabling woods being entrained into the stream. Based on these facts, natural restoration of the first and second order streams is proposed in order to mitigate floods in the downstream main channels of the sub-basin. Riparian and stream strategies may include increasing desired vegetation, decreasing invasive species, and increasing stream sinuosity. Following factors before and after river restoration are therefore considered in the modelling approach. (1) Change of flow resistance: Change in stream flows is linked directly to the change of flow resistance of the streams. Such resistance to flow reflects the rate of energy dissipation and incorporates resistance offered by in-stream vegetation and natural obstruction as wood and debris accumulation. The effect of vegetation and natural obstructions in the stream channels is estimated in terms of the Manning roughness coefficient. Literature indicates that the Manning roughness coefficient can increase by as much as three times compared to without in-stream woods vegetation (Shields & Gippel, 1995). Therefore, an increase of the Manning roughness coefficient from the value of 0.04 to 0.1 m-1/3s is proposed for the first and second stream orders, while the Manning coefficient of the higher order streams remains unchanged. (2) Stream re-meandering: Stream re-meandering is one of the major measures for river rehabilitation and restoration. Re-meandering of a watercourse can remedy 127

Chapter VI some of the consequences of former channelization and improve the interplay between watercourses and their river valley. Moreover, it can have a positive influence on water quality, not just in the watercourse itself, but also in other aquatic areas. Typical examples were given by Hansen (1996), who described the river restoration projects in Denmark involving re-meandering and increased inundation of the floodplain during the recent two decades. More specifically, a positive response was monitored in their study for an upper reach, which was remeandered from a 2.7 km straight and channelized channel into a new 3.2 km meandering course. Considering the non-uniformity of the stream meandering for the Alzette headwater streams, an increase of 10% of channel sinuosity is proposed, which will increase 42.5 km in total length for the first and second order streams. (3) Change of stream slope: The longitudinal stream slope after restoration is reduced due to the effect of stream re-meandering. 10% increase of channel sinuosity results in a same ratio of decrease, i.e. 0.0012, of the average channel slope of the first and second order streams. Changes in slope may also be associated with changes in channel material and in-stream vegetation, which slow down the flow velocity and induce more sediment deposit along the river channel. However, only the slope change caused by stream re-meandering is considered in this study. (4) Hydraulic radius: Hydraulic radius is a measure of the relative channel shape, which is governed by the channel cross section and a particular flow level. It is often approximated by flow depth for broad and shallow channels. Therefore, a consistent way would be to model the dynamic interaction between water level and channel flows from the characteristic profiles. This interaction cannot be modelled unless detailed information is available about the hydraulic properties of the entire channel network. As such information does not exist for the Steinsel sub-basin, a diffusive wave approximation routing procedure was adopted in the model. This method assumes time-independent flow velocities parameterized as a function of the topographic gradient, which is commonly used in the GIS-based flow routing schemes (Lee & Yen, 1997; Olivera & Maidment, 1999). The model determines the hydraulic radius by a power law relationship with an exceedance probability (Molnar & Ramirez, 1998), which relates hydraulic radius to the 128

Assessing the effects of river restoration on the reduction of floods in a river basin drained area and is seen as a representation of the average behaviour of the cell and the channel geometry. Due to the increased flow resistance, more water would be retained in the headwater streams during the rising stage of a flood event. However, this extra flood volume may partly be compensated by the extended river reaches after stream re-meandering. Accordingly, the estimation scheme for hydraulic radius remains unchanged in this study.

3.

Results and discussion

3.1.

Model calibration and evaluation

The model was verified for the present situation before being applied for analysis of the effects of river restoration to mitigate flooding of the study area. The procedures of model parameterization and calibration results have been provided in chapter V. Normally, the model improvement and calibration proceed in three phases, focusing on different flow conditions. Initially, the volume of precipitation, evapotranspiration and outflow is examined at each flow station, for which the correction factor of potential evapotranspiration can be determined. Next, the interflow and groundwater flow recession coefficients are adjusted by comparing the low flow and the recession part of flood hydrographs. The third phase of model calibration focused on improving the timing, magnitude and hydrograph shape of various flood pulses. In conjunction with this, major parameters, including channel roughness coefficient and hydraulic radius, are adjusted in order to achieve the best agreement with the measurements. A value of 0.04 m-1/3s is obtained as a proper streams Manning roughness coefficient for the present situation by model calibration, which is typical for clear streams without vegetation, or any other major obstructions. The calibrated minimum hydraulic radius for overland flow is 0.005 m, and maximum 1.2 m for channel flow at the sub-basin outlet for a normal flood corresponding to a 2-year return period. These values can be increased for extreme floods. Observed hourly time series of streamflow are used to verify the models performance and to adjust model parameters. Based on graphical and numerical evaluation of the results, model parameters are adjusted, and the model is re-run until a good match between the observed and simulated hydrographs is obtained.

129

Chapter VI
50 0

40

30

10

20

15

10

20

0 23/10

25 25/10 27/10 29/10 31/10 2/11 4/11 6/11 8/11 10/11 12/11 14/11

Time (d/m)

Figure VI-2: Observed and simulated flow hydrographs for the flood events in Oct. and Nov. 1998

45 40

Simulated (m /s)

1:1

35 30 25 20 15 15 20 25 30 35
3

40

Observed (m /s)

Figure VI-3: Observed versus simulated peak flows for the simulation period

Specifically, Figure VI-2 presents an observed and simulated hydrograph for a compound flood at Steinsel, which is the largest flood during the simulation period that occurred in October and November 1998. As can be seen from the figure, the general evolution of the observed hydrograph is reproduced rather well. The total rainfall was 193.2 mm with a measured runoff of 87.6 mm and a simulated runoff of 84.3 mm. A small flood occurred on October 24, followed by three large successive floods with observed peak discharges of 33.3, 38.8 and 40.5 m3/s. The predicted peak discharges are 34.3, 42.7 and 38.4 m3/s with relative errors of 3.0%, 10.0% and -5.2%

130

Precipitation (mm/h)
45

Discharge (m /s)

Observed Simulated

Assessing the effects of river restoration on the reduction of floods in a river basin respectively. Figure VI-3 gives the observed versus simulated peak flows selected from 60 independent storm events that occurred throughout the simulation period. The high peak discharges are reproduced reasonably well, while the estimations are slightly poorer for small floods. Some points are far from the 1:1 line, which might be caused by small scale thunderstorms for which the spatial distribution of rainfall was not well captured by the rainfall stations (Drogue et al., 2002). Five evaluation criteria are applied for the assessment of the model performance for both calibration and validation period at the four flow stations. They are model bias, model determination coefficient, Nash-Sutcliffe efficiency, and adapted NashSutcliffe efficiency for evaluating high flow and low flow, as have been described in chapter V. A good model performance, especially for the high flow, has been obtained after model calibration, which indicates that the model is suitable to predict runoff and flow responses for the study area.

3.2.

Model prediction

Once the model is verified for the present conditions, it can be applied for the simulation of the effect of river restoration in first and second-order streams in the study area. The modelling approach includes: (1) increasing the roughness coefficient by 250% to account for the effect of in-stream vegetation and natural obstructions as wood and debris accumulation after river restoration, (2) extending the channel length for the first and second order streams by 10% as a consequence of increasing channel sinuosity, and (3) reducing the average stream slope by 10% as a result of river remeandering. Since the detailed information of the hydraulic properties is not available for the entire channel network, the changes in flow resistance, flow length and stream slope are assumed to be uniformly distributed over the headwater streams. By applying these changes, a new roughness coefficient grid and a new channel slope grid of the first and second order streams are constructed, for which the Mannings coefficient is increased from 0.04 to 0.1 m-1/3s, and the slope is reduced by 0.0012 for each channel cell. The new grids are then merged by the previous roughness coefficient and slope map of the sub-basin. The new grids of average flow time and its standard deviation are obtained using the weighted GIS FLOWLENGTH routine, giving an extra weight of 1.1 for the first and second order streams to account for the 131

Chapter VI increase of flow length for each channel cell, while the grid of flow direction and flow accumulation is kept unchanged. In this way, the flow response function for each grid cell after river restoration can be obtained. The immediate effect of these changes in the first and second order streams is to decrease the flow velocity in these channels. However, there is no change in velocity once the flow enters in the main channel. The lower velocities in the first and second order streams result in a prolonged travel time from the headwater areas to the subbasin outlet. Figure VI-4a gives the average flow travel time to the sub-basin outlet in hours for the present condition, in which the flow time is less than 10 hours for the main river and up to 35 hours for the most remote areas. Figure VI-4b shows the increase in average flow travel time after river restoration, in which the travel time of areas of the sub-basin that are drained by first and second order streams is delayed by several hours, while the flow time of areas drained by high order streams remains unchanged.
(a)
Time (h) (h) Time 0- 5 1 0 1 - 10 2 5 2 - -315 10 3 - -420 15 4 - -525 20 > >5 25
N W

(b)
Time (h) 0-1 1-2 2-3 3-4 4-5 >5
N
E
E E S
10 km 10 km

NN
S

NN
S

E
E E

WW

WW

S
0

S
0

S
10 km 10km
10km

5 5
5

5 5
5

10km

Figure VI-4: (a) Average flow travel time to the sub-basin outlet for the present condition, and (b) Increases in flow travel time after river restoration The changes of stream features after river restoration result not only in a time delay from the headwater streams to the downstream main river, but also in a reduction of peak flows at the outlet of the sub-basin. Figure VI-5 shows the impact of river restoration on peak discharges for the compound flood event in October and

132

Assessing the effects of river restoration on the reduction of floods in a river basin November 1998, which was described in the model calibration. Under the proposed conceptual mitigation plan, the peak discharges are reduced to 29.8, 38.2 and 35.1 m3/s, indicating a 13%, 10% and 9% reduction respectively in calculated peak discharge after river restoration. The reduction in the discharge occurs in the rising stage of the flood hydrograph and continues until reaching the peak value. During the recession limb, the runoff becomes larger in the river restoration case compared to the present situation. This shows that the flow is retarded and stored in the headwater streams, and later released during the recession stage of the flood. Figure VI-5 also indicates that the travel time of the peak discharges are shifted by 1, to 2 hours by the water flow delaying effect.

50

Discharge (m /s)

40

10

30

15 20

20

10 23/10

25 25/10 27/10 29/10 31/10 2/11 4/11 6 /11 8/11

Time (d/m)

Figure VI-5: Flood events showing the effect of natural river restoration

45

1:1

After restoration (m /s)

40 35 30 25 20 15 15 20 25 30

Present (m /s)

35

40

Figure VI-6: Present versus restored simulated peak discharges for the simulation period indicating a 14% reduction in average after river restoration 133

Precipitation (mm/h)
45

Present condition After restoration Depletion Accumulation

Chapter VI

The effect of river restoration is also evident in Figure VI-6, which gives a general overview of reduction in peak discharges. This figure shows the calculated peak flows before and after river restoration for the 60 independent storm events that occurred during the entire simulation period. This result shows that peak flows are reduced for all storms by an average of about 14%. Hence, the reduction is substantial and can contribute in a significant way to the mitigation of floods in the downstream rivers.

3.3.

Evaluation of a future flood scenario

Studies have shown that there has been a marked increase in the contribution of the westerly component of atmospheric circulation to rainfall in the Alzette River basin since the 1970s (Pfister et al., 2000). These changes in atmospheric circulation are usually accompanied with an increase in rainfall intensity and duration, which result in a significant increase in the winter maximum daily storm and river flow. Therefore, the model approach can be used to investigate the consequences of future climate change and the potential effect of river restoration for a given storm scenario under present land-use conditions. Future regional climate scenarios are firstly constructed by using the outputs from HadCM3, which is the third generation coupled atmosphere-ocean general circulation model developed by the U.K. Meteorological office, Hadley centre (Gordon et al., 2000). The atmospheric component of HadCM3 has 19 levels with a horizontal resolution of 2.5o latitude by 3.75o longitude, which produces a global grid of 96 by 73 cells that is equivalent of about 417 km by 278 km at the Equator. Future time series of precipitation and temperature are synthesised, the latter to estimate the potential evaporation, and together with precipitation to be used as input to the WetSpa model. The downscaling of precipitation and temperature for the study area is performed with the Statistical Downscaling Model (SDSM) (Wilby et al., 2002). First, a daily statistical relationship is established between surface and upper-atmospheric circulation variables with locally observed precipitation and temperature data for the period from 1961 to 1990. Next, the SDSM model is calibrated using observed precipitation and temperature data of the baseline period and the selected predictor variables, namely, the geopotential height and relative humidity at 500 hPa, the 134

Assessing the effects of river restoration on the reduction of floods in a river basin geopotential height and relative humidity at 850 hPa, the near surface specific humidity, the westerly wind component at 10 m elevation, and the maximum temperature at 2 m height (Gebremeskel, 2003). Hourly precipitation and temperature series are simply extracted from the daily SDSM predictions by amplification of the hourly baseline series. A worst simulated storm scenario in February, 2050 is selected from the SDSM predictions to study the potential effect of river restoration on flood shape at the subbasin outlet as shown in Figure VI-7. The total rainfall is 72.5 mm within 18 hours with a maximum rainfall intensity of 11.3 mm/h, corresponding to a winter storm with a frequency of 2% (Gebremeskel, 2003). A new grid of hydraulic radius is generated by the model with an exceedance probability of 50-year return period, which results in a maximum value of 2 m at the basin outlet, which is nearly the double of values obtained for a normal flood with a 2-year return period. Thereafter, the grids of flow velocity, average travel time and its standard deviation are created. By keeping other parameter maps as used in model calibration and prediction, the outflow hydrograph at Steinsel is estimated by the WetSpa model using the input data from the selected scenario. The calculated peak discharge for this flood under the present condition is 70.1 m3/s. The estimated peak discharge after river restoration is 59.5 m3/s and the peak time is delayed by 2 hours. The peak discharge at Steinsel is reduced by 10.6 m3/s indicating a 15% reduction after river restoration.

80

Discharge (m /s)

60

After restoration

40

12

20

18

Depletion
0 1 25 49

Accumulation
24 73 97 121

Time (h)

Figure VI-7: Simulated hydrograph under present condition and after river restoration for a future storm scenario

135

Precipitation (mm/h)

Reduced peak
3

Present condition

Chapter VI

Considering the water volume for this scenario flood, about 1 million m3 of water is retained during the first two days and released thereafter during the next four days. This water volume is stored in the first and second order riverbeds, which may induce a considerable increase of water depth in these watercourses. As a result, river restoration may cause local flooding, which however will be less damaging and dangerous than flooding of the downstream urban areas.

3.4.

Discussion

In this study, an increase of roughness coefficient by 150% is assumed for the first and second order streams after river rehabilitation. This increase is associated to the rehabilitation measures on the headwater streams and their riparian areas. Typical Mannings roughness coefficient values are with the range 0.08-0.12 m-1/3s for minor streams with dense vegetation and irregular alignment and cross section (Chow, 1959). Much higher values have been reported in some recent river rehabilitation studies in Europe, e.g. Helmi & Jrvel (1998) and Seara & Newsonb (2004). The Mannings coefficient value of 1.0 m-1/3s is therefore feasible reflecting the stream condition after rehabilitation. However, Mannings roughness coefficient are site-specific, depending on the channel surface roughness, irregularity, shape variation, obstructions, type and density of vegetation, degree of meandering, flow depth, seasonal changes in vegetation, and so on. These factors are unique to each stream reach and change in space and time. Using a unique roughness coefficient value and applying it to the entire headwater streams in this study may misrepresent site conditions and add uncertainty to the simulation results. The flow routing in this study is modelled by a linear approximation of the diffusive wave equation. This method assumes that flood waves propagate at a constant velocity on a river reach. WetSpa model calculates flow velocity at each grid cell using Mannings equation as a function of roughness coefficient, slope and hydraulic radius. The hydraulic radius is estimated as a power function of the upstream drainage area of the cell and varies with flood frequencies. Parameters controlling the distribution of hydraulic radius for different flood magnitudes are adjusted at each gauging station using historical records during model calibration. Consequently, the 136

Assessing the effects of river restoration on the reduction of floods in a river basin average travel time and its standard deviation to the basin outlet can be obtained, which are location dependent and vary with floods frequencies. Since there is no detailed channel geometric data available for this study, the effects of overbank flow and flood plain storage are accounted for in the model by cautiously setting the hydraulic radius parameters for extreme flood. The resulting values are compared with the measurement at different stations and adjusted through model calibration. However, such simplification may not represent precisely the real situations under heavy flooding, and may reduce the value of model validation. In addition to the rehabilitation measure of increasing natural storage capacities in lowland areas downstream of the river segment, flood mitigation strategies in upstream areas are also important (WWF, 2002). The main purposes of these measures as described in this study are to reduce rapid runoff on upland and riparian areas, to retain more floodwater in the upstream tributaries, and consequently to mitigate the flood risk in the downstream areas. Compared with other flood control measures, river rehabilitation, as means of increasing channel resistance in headwaters, also provide a wide range of additional benefits in the form of reducing soil erosion, increasing water quality, maintaining biodiversity and areas for recreation, and so on. These strategies are therefore essential for the integrated river basin management. The rehabilitation measures are, however, feasible for a long time period, which can be realized by stopping deforestation, using buffer zoning and strips, planting of tree species alongside the river channel, providing tree barriers and other engineering measures, etc. This study focuses on the possible river rehabilitation effects on the Steinsel sub-basin. It covers an area of 34.7% of the Alzette River basin (1175 km2), and is only 0.22% of the Rhine River basin (185000 km2). Simulation results of this study show that the natural river rehabilitation in the headwater areas can produce a prolonged flow time, and result in a remarkable reduction in the flood peaks in the downstream main channels. Considering the vast headwater areas of the Rhine River basin, these measures may significantly mitigate the flood risk on the main river channels. This is especially important for the downstream delta, which strongly depends on the flood defence strategies for the entire river basin and the time of peaks of all tributaries. However, the impact of river rehabilitation of headwater areas can not be assessed by 137

Chapter VI simply amplifying the results obtained from a small subcatchment. Collection of detailed data and studies on the complex flow systems are required in this respect.

4.

Concluding remarks

In this chapter, the effects of river restoration on flood reduction in the Steinsel subbasin of the Alzette River basin Grand-Duchy of Luxembourg is presented. A conceptual method is proposed to account for the effect of in-stream vegetation and channel re-meandering after river restoration. The restoration is focused on the first and second order streams ordered according to Strahlers method. The implementation is performed using the WetSpa model, applied in a GIS environment. The simulation results indicate a significant decrease in peak discharge and a delay of flood peak occurrence after river restoration. The reduction of peak discharge is as much as 14% on average compared to the present situation, and the time delay of flood peak can be as much as 2 hours. A reduction in the discharge occurs during the rising limb of the flood hydrograph, while the discharge increases during the falling limb of the hydrograph, which results in longer sustained flows than in the present conditions. However, while large floods in the main stream channels are reduced or avoided, a local flooding may occur in the headwater stream areas due to river restoration. It can be concluded that river restoration in such a way as to increase flow resistance, restore more natural conditions, and enable watercourse locally to flow their meadows, has a positive influence on reducing the risk of flooding further downstream, where the consequences can be more severe due to the size of river basin and the magnitude of the discharge. This work focused on the beneficiary effects of river restoration for flood reduction in the Steinsel sub-basin. However, it is clear that river restoration can have additional beneficiary as well as adverse effects on the ecology and morphology of river basins that are not covered in this study.

References: Booth, D.B., Urbanization and the natural drainage system-impacts, solutions and prognosis, Northwest Environ. J., 7, 93-118, 1999. Chow, V.T., Open-Channel Hydraulics, New York, McGraw-Hill, 680 pp., 1959.

138

Assessing the effects of river restoration on the reduction of floods in a river basin De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrologic modelling on a catchment scale using GIS and remote sensed land-use information, In: Risk Analysis II, ed., C. A. Brebbia, 295-304, WTI press, Southampton, Boston, 2000. Drogue, G., Leviandier, T., Pfister, L., El Idrissi, A., Iffly, J.F., Hoffmann, L., Guex, F., Hingray, B. & Humbert, J., The applicability of a parsimonious model for local and regional prediction of runoff, Hydrol. Sci. J., 47, 905-920, 2002. Gebremeskel, S., Modelling the effect of climate and land-use changes on hydrological processes: An integrated GIS and distributed modelling approach. Doctoral Thesis, Vrije Universiteit Brussel, Belgium, 2003. Gordon, C., Cooper, C., Senior, C.A., Banks, H., Gregory, J.M., Johns, T.C., Mitchell, J.F.B. & Wood, R.A., The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments, Clim. Dynamics 16, 147-168, 2000. Hansen, H.O., River Restoration, Danish experience and examples, 99, National Environmental Research Institute, Denmark, 1996. Helmi, T. & Jrvel, J., Assessing the hydraulic performance in river rehabilitation projects - Myllypuro Brook case study, In: Kajander, J. (ed.), XX Nordic Hydrological Conference, Helsinki, Finland, 357364, 1998. Larson, M.G., Booth, D.B. & Morley, S.A., Effectiveness of large woody debris in stream rehabilitation projects in urban basins, Ecol. Eng., 18, 211226, 2001. Lee, K.T. & Yen, B.C., A geomorphology and kinematic-wavebased hydrograph derivation, J. Hydraulic Eng., ASCE 123(1), 7380, 1997. Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. & Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modelling, J. Hydrol., 283, 91-106, 2003. Molnar, P. & Ramirez, J., Energy dissipation theories and optimal channel characteristics of river networks, Water Resour. Res., 34(7), 1809-1818, 1998. Morris, S.E., Geomorphic aspects of stream-restoration, Phys. Geogr. 16(5), 444-459, 1995. Nienhuis, P.H. & Leuven, R.S.E.W., River restoration and flood protection: controversy and synergism, Hydrobiologia, 444, 85-99, 2001. Olivera, F. & Maidment, D.R., Geographic information system (GIS)-based spatially distributed model for runoff routing, Water Resour. Res. 35(4), 11551164, 1999. 139

Chapter VI Pfister, L., Humbert, J, Iffly, J.F. & Hoffmann, L., Use of regionalised stormflow coefficients in view of hydro-climatological hazard mapping, Hydrol. Sci. J., 47(3), 479-491, 2002. Pfister, L., Humbert, J. & Hoffmann, L., Recent trends in rainfall-runoff characteristics in the Alzette river basin, Luxembourg, Climate Change 45, 323-337, 2000. Rosgen, D.L., A classification of natural rivers, Catena, 22, 169-199, 1994. Schueler, T., The importance of imperviousness, Watershed Protection Techniques, 1(3), 100-111, 1994. Seara, D.A. & Newsonb, M.D., The hydraulic impact and performance of a lowland rehabilitation scheme based on pool-riffle installation: The river Waveney, Scole, Suffolk, UK., River Res. Applic., 20, 847863, 2004. Shields, F.D. & Gippel, C.J., Prediction of effects of woody debris removal on flow resistance, J. Hydraul. Div. ASCE 121(4), 341354, 1995. Strahler, A.N., Quantitative analysis of watershed geomorphology, Trans. Am. Geophys. U. 38, 913920, 1957. Wang, Z., Batelaan, O. & De Smedt, F., A distributed model for Water and Energy Transfer between Soil, Plants and Atmosphere (WetSpa), Phys. Chem. Earth 21(3), 189-193, 1997. Wilby, R.L. & Dettinger, M.D., Steamflow changes in the Sierra Nevada, California, simulated using a statistically downscaled General Circulation Model Scenario of climate change. In: Linking Climate Change to Land Surface Change, eds., McLaren, S., and Kniveton, D., Kluwer Academic Publishers, Netherlands. 6.1- 6.20, 2000. WWF, Managing floods in Europe: The answers already exist, WWF Background Briefing, Paper 26, Brussels, 2002. Yen, B.C., Hydraulic resistance in open channel, in: Channel Flow Resistance, Centennial of Mannings Formula, ed., B.C. Yen, 1-135. Wat. Resour. Publ. Littleton, Colorado USA, 1991.

140

Chapter VII Flow simulation in a Carpathian catchment accounting for topographic controls
Abstract An application of the WetSpa model based on GIS and remote sensing is presented for hydrological modelling of a Carpathian catchment on daily scale. This is the Hornad River catchment upstream of the Ruzin reservoir in the eastern Slovak Republic, which is characterised by mountainous terrain and altitudinal variation is an important determinant of the local climate and soil characteristics in all the state. It is required to develop further the WetSpa model to account for these terrain features, which includes the snow accumulation and snowmelt processes in order to predict spring flood. Particular attention is devoted to the analysis of the temporal and spatial distribution of temperature, PET and precipitation within the catchment. The derivation of distributed models parameters is based on an extensive database of catchment characteristics available for the region, including a 100 m resolution DEM, and digital maps of soil type and land use. An automated calibration scheme is employed to the WetSpa model in this study serving as an optimization algorithm to estimate the model parameters. The encouraging results in spite of the highly complex catchment morphology underline the importance of the availability of spatially distributed data to be used for model identification and parameterisation. However, as expected, the model is strongly sensitive to the parameters describing the runoff generation processes and the routing of water in surface, subsurface and groundwater reservoirs. The study concludes about the evident need for enlarging data availability or, alternatively, in developing a robust parameter calibration method, which rely on data that are generally available.

1.

Introduction

In March 1999 and April 2000, major flood events occurred in Slovakia affecting most regions of the Hornad River basin. These events were caused by incessant heavy

Chapter VII rainfall, preceded by a considerable snowmelt over a large area of the region, particularly in the Margecany catchment upstream of the Ruzin reservoir, due to particularly mild weather conditions at the end of an extremely snow-rich winter. Because of the complexity of the regulation of the lakes in relation to the downstream flood conveyance capacity, it is of interest to understand the dynamics of the events and investigating the long-term vulnerability of the basin to similar circumstances. Furthermore, the basin is highly complex in terms of topography, geology, climatology, and soil properties. Floods occur both in springs due to snowmelt, and in summers due to heavy rainfall. Snow accumulation in winter and snowmelt in spring and early summer represent a significant component of the hydrological cycle and play an important role in generating floods. Reliable estimates of meteorological inputs like precipitation, temperature and PET are key elements to hydrological models, particularly when modelling a mountainous catchment. The analyses of spatial variability of hydrological processes in general and precipitation processes in particular have been of interest to water resources planners and managers for quite some time (Tabios and Salas, 1985). Such analyses have applications in such studies as determination of water budget at different spatial and temporal scales, validation of different hydrological models and recently global climate change impact studies (Nijssen et al., 2001; Arnell, 1999). A number of techniques for the spatial interpolation of long-term mean precipitation for the whole territory of Slovakia have been tested with acceptable results (Parajka, 1999). The importance of spatial variability in temperature and PET to the hydrological modelling has been addressed in many publications (Shevenell, 1999; Xu and Li, 2003). Distributed models have the advantage to estimate the evapotranspiration in a spatial way in combination with the measured meteorological and land-use data (Flerchinger et al., 1996). In this study, a simple linear regression method is used for the spatial interpretation of precipitation, temperature and PET inputs based on measured meteorological data. Further physically-based studies on the spatial variability of meteorological variables are necessary once additional data are available. The investigation of all of these aspects requires the availability of a tool that allows a robust simulation of the hydrological processes. Therefore, the WetSpa model, working on a daily basis, is applied to simulate both spring and summer floods for the 142

Flow simulation in a Carpathian catchment accounting for topographic controls Margecany catchment accounting for topographic controls on the input

meteorological variables. Some developments of the original formulation of the model are illustrated in order to account for peculiarities of the mountainous environment. A preliminary set of results is illustrated, and the adequacy of the WetSpa model to represent the response of mountainous catchments is also discussed.

2.

Methodology

2.1.

Description of the study area and data available

The Margecany catchment covers the upper Hornad River basin upstream of the Ruzin reservoir in the eastern Slovak Republic (Figure VII-1), with elevation ranging from 333 m a.s.l. near Margecany to 1556 m a.s.l. The Ruzin reservoir has several other tributary streams, like the Hnilec River, entering the reservoir, which were not evaluated in this study. The total watershed size of Ruzn Reservoir is 1929 km2, where the Margecany catchment accounts for 1130 km2 (58.6%) of that area. The basin has a northern temperate mountainous climate with four distinct seasons. January is the coldest month and July is the warmest month of the year. The highest amount of precipitation occurs in the period from May to August and the least is in January and February. The mean annual precipitation ranges from about 640 mm in the valley to more than 1000 mm in the vicinity of the water divide from the analysis of the Slovak hydrometeorological institute, Bratislava, Slovakia. Based on the statistical analysis of the 10-year (1991-2000) weather data observed at Spisske Vlachy, located in the downstream reach of the Margecany catchment at an elevation of 382 m a.s.l., the annual precipitation ranges from 453 mm to 745 mm with mean of 610 mm, in which about 10% of the total in annual precipitation is snowfall mainly concentrated in December, January and February. The average annual temperature at Spisske Vlachy is 7.2 C with a monthly minimum temperature of 8.0 C in December and an average monthly maximum temperature of 19.9 C in August. The average annual PET measured at Spisske Vlachy is 518 mm with a monthly minimum PET of zero mm in December and January, and an average monthly maximum PET of 96 mm in July and August. Figure VII-2 shows the average monthly distribution of

143

Chapter VII temperature, precipitation and PET at Spisske Vlachy based on the 10-year measured weather data.

Figure VII-1: Location of the Margecany catchment

100

T ( c) and P, PET, (mm)

80 60 40 20 0 -20 1

Temperature
Precipitation
Potential evaporation

10

11

12

Month

Figure VII-2: Monthly temperature, precipitation and PET at Spisske Vlachy The daily precipitation, temperature and discharge data are obtained from SHMU, the Slovak Hydrometeorological Institute, whereas the PET data are obtained from VUVH, the Water Research Institute of Slovakia. The sets include daily precipitation for 9 stations, temperature for two stations, PET at Spisske Vlachy, and discharge data at Margecany from 1991 to 2000 (Table VII-1). Flow stations containing daily discharge data at 8 locations are available within the catchment, but only the outlet

144

Flow simulation in a Carpathian catchment accounting for topographic controls station, Margecany, is used for model calibration in this case study. The precipitation in a given raster cell is obtained from the precipitation of the representative weather station and is corrected for the altitude of that cell within its Thiessen polygon (Figure VII-3) with the use of elevation data from the DEM. The same procedures are applied for the temperature and PET for each raster cell. Table VII-1: Information of weather stations in the Margecany catchment
Station Spisske Vlachy Ganovce Vikartovce Hranovnica Levoca Rudnany Spisske Podhradie Krompachy Mlynky Margecany Number 5 6 15 16 17 18 19 20 22 Type P, T, E P, T P P P P P P P Q Elevation (m) 382 691 752 613 577 546 427 366 1007 339 X_coord (m) 485522 450468 437883 449580 470566 477042 482033 490591 458204 498007 Y_coord (m) 5421194 5431519 5427066 5426359 5430167 5413764 5427029 5418827 5416373 5417377

# # #

Vikartovce

Ganovce
$
#

Levoca # Spisske Podhradie

#

U $
#

Hranovnica $
$ $

Spisske $ Vlachy
# U $

Mlynky #
# # U

Krompachy $
#

Evaporation station Flow station Precipitation station Stream network Thiessen polygon Catchment boundary

$
#

Rudnany

$
#

Figure VII-3: Gauging sites and Thiessen polygons for the Margecany catchment Land cover data were obtained from the third hierarchy CORINE geographic information system coverage developed from the European PHARE Project. This land cover information was reclassified into 14 categories and converted to a 100 m cell size grid for use in the WetSpa model, and re-grouped into 5 classes for deriving model parameters of potential runoff coefficient and depression storage capacity (Figure VII-4). The soil information was provided by VUPU, the Soil Science and Conservation Research Institute, Slovakia. A simple nearest neighbour interpolation method was applied based on the point information. The map is further reclassified 145

Chapter VII into 12 USDA soil texture classes based on their textural properties and converted to a 100 m cell size grid (Figure VII-5). The elevation data for the river basin was obtained from SHMU, the Slovak Hydrometeorological Institute, digitized from an elevation map. This elevation data was interpolated to construct a 100 m grid size DEM, from which the drainage system and area were determined (Figure VII-6).

Landuse Crop Grass Forest Urban area Open water

Figure VII-4: Land use map of the Margecany catchment

Figure VII-5: Soil textural map of the Margecany catchment

Figure VII-6: Topographical map of the Margecany catchment 146

Flow simulation in a Carpathian catchment accounting for topographic controls

2.2.

Modelling snowmelt

Physical processes within the snowpack and involved in snowmelt are very complex. They involve mass and energy balances as well as heat and mass transport. Formation of ice layers further complicates evolution of snowpack resulting in processes known from soil physics like fingering or lateral flow (Parajka, 2001). Snowmelt is basically an energy driven process. Incoming solar radiation, absorption and emission of long wave radiation, turbulent transfers by sensible and latent heat fluxes, and energy exchanges at snow-ground base are the main driving components. The volume of water released by melting snow is a major state variable that characterises the snowmelt processes. Numerous snowmelt models have been developed to describe the evolution of this variable (Leavesley, 1989). Generally, they can be divided into three groups: index models, energy based models and detailed models using full solutions of the energy and mass flow equations. Energy based models use more correct physical description of basic processes affecting snow accumulation and melt. Detailed models based on energy and mass flow equations are physically correct, but demand a lot of data that are not easily available. The conceptual temperature index or degree-day method used in this study is a simple method but has a strong physical foundation. The method replaces the full energy balance with a term linked to air temperature. It is physically sound in the absence of shortwave radiation when much of the energy supplied to the snowpack is atmospheric long wave radiation. The equation can be expressed as: M = Max[0, C snow (T T0 ) + C rain P(T T0 )] (7.1)

where M [LT-1] is the daily snowmelt, T [C] is the cell daily mean temperature, T0 [C] is a threshold melt temperature, Csnow [LC-1T-1] is a melt-rate factor, and Crain [LL-1C-1T-1] is a degree-day coefficient regarding to the heat contribution from rainfall. The critical melt temperature, T0, is often intuitively set to 0C. The melt-rate factor, Csnow, is an effective parameter and may vary with location and snow

147

Chapter VII characteristics. However, this parameter is assumed as a constant in this study for model simplicity. Melt water reaching permeable surfaces can either infiltrate or flow overland depending on soil structure, moisture and thermal status. Outside permafrost regions, overland flow of melt water is rare in rural areas, especially in forests where the soil cover is thick and permeable (Espeby, 1990). On the other hand, melt water from snow overlying a saturated, frozen and compacted urban soil can flow overland, particularly towards the end of the melt period (Bengtsson & Westerstrm, 1992). In hydrological modelling, the infiltration of melt water into soil is usually modelled by a coefficient and the ratio of water stored in the soil to the maximum storage (Mocko & Sud, 2001; DHI, 1994). The HBV model, as another example, treats soil with a bucket approach, for which rain and melt water infiltrate freely until the soil layer is saturated (Lindstrm, 1997). In this study, the same scheme as for surface runoff from rainfall is used to calculate the snowmelt runoff and its infiltration:

V = c r (M + P ) s

(7.2)

where V [LT-1] is the surface runoff resulting from snowmelt and rainfall, cr is a potential runoff coefficient depending on slope, soil type and land use, P [LT-1] is the rainfall intensity, [L3L-3] is the soil moisture content, and s [L3L-3] is the soil porosity. The difference between snowmelt, rainfall and surface runoff (M + P - V) is the snowmelt infiltration, which contributes further to the lateral subsurface flow, evapotranspiration and percolation to the groundwater storage. The snowmelt surface runoff, together with lateral subsurface flow and the groundwater flow at the subcatchment out is routed along its flow path to the basin outlet using the diffusive water approximation method as describe in chapter III.

2.3.

Topographic adjustment for the input variables

Reliable estimates of meteorological data such as temperature, precipitation and PET are key elements to the hydrological modelling. The traditional approach is to use

148

Flow simulation in a Carpathian catchment accounting for topographic controls observed values from nearby stations, which are usually located in the river valleys and can not be representative for the high terrain areas. Since the study area is located in a temperate, mountainous region, vertical variation of theses elements are of major importance in hydrological modelling, such as snow accumulation and snowmelt, runoff, evapotranspiration, soil moisture content, etc.

2.3.1. Adjustment for temperature Air temperature is strongly related to altitude. In average the temperature decreases with about 0.65-1.0 C/100m vertically in the free atmosphere. Close to the surface, this vertical temperature gradient is different due to local and regional characteristics, influenced by the local terrain, distance from sea, etc. 28 temperature stations with 10 years daily average temperature data (1991-2000) in and surrounding the Hornad River basin are selected to study the vertical temperature gradient. Figure VII-7 shows the lapse rates for the three mean monthly temperatures, as well as the yearly temperature lapse rate.

14

Year
10

February April

Marc h

y = -0.0049x + 9.9263 R2 = 0.8561 y = -0.0064x + 11.045 R2 = 0.9058 y = -0.0055x + 4.7451 R2 = 0.8704 y = -0.0032x - 0.2008 R2 = 0.7497

Temperature ( C)

-2

-6 0 500 1000 1500 2000

Elevation (m)

Figure VII-7: Lapse rates for mean monthly and yearly temperature Table VII-2 gives a summary of monthly temperature characteristic values and the regression relationships between monthly temperature and elevation, which shows a very close correlation between mean monthly temperature and elevation with average correlation coefficient of 0.83. The vertical temperature gradient varies considerably with season with low slope in winter and steep slope in summer. The yearly

149

Chapter VII regression slope is -0.50% (0.5oC/100 m) with correlation coefficient of 0.86, which conforms to the normal lapse rate of the region. The air temperature from October to March is important in controlling the snow accumulation in this catchment, whereas the temperature in March and April is critical in controlling the snowmelt. Table VII-2: Regression analysis between monthly temperature and elevation
Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Year Mean (oC) -2.50 -1.39 2.71 8.72 13.73 17.26 18.66 18.38 13.21 8.22 2.83 -2.13 8.10 Max. (oC) 2.45 4.17 9.75 14.48 19.51 20.84 24.16 24.82 19.00 13.75 9.77 3.10 9.20 Min. (oC) -7.47 -8.39 -5.91 -1.54 4.33 10.54 10.88 9.12 4.00 -1.60 -7.05 -8.02 7.29 Slope (%) -0.21 -0.32 -0.55 -0.64 -0.61 -0.62 -0.61 -0.56 -0.52 -0.44 -0.42 -0.28 -0.50 Intercept (oC) -1.72 -0.20 4.75 11.05 15.97 19.50 20.91 20.44 15.11 9.84 4.35 -1.11 9.93 R2 0.63 0.75 0.87 0.91 0.89 0.90 0.88 0.79 0.82 0.82 0.89 0.77 0.86

2.3.2. Adjustment for precipitation The increase of precipitation with altitude is well known, chiefly because landforms obstruct the movement of air and cause it to rise. The effects of altitude on precipitation are very complex, including (1) forced ascent, (2) blocking or retardation of storms, (3) lifting by landforms, (4) local convection, and (5) condensation processes. Besides, the local variability of precipitation in a mountainous catchment may be poorly cast with a function of topographic height along and it strongly depends on other physiographical factors (slope, aspect, broad-scale topographic environment), especially for a small-scale storm or precipitation within a short time period. However, due to the lack of dense observation, linear precipitation-height regression analysis over a larger region can be used to compensate for the local variation of precipitation as a function of altitude. In this study, 44 precipitation stations in and surrounding the Hornad River basin with elevation ranging from 105 to 1240 m and available daily precipitation data over the period of 1991 to 2000 were

150

Flow simulation in a Carpathian catchment accounting for topographic controls selected to study the precipitation-height correlation. As illustrated in Figure VII-9 (hollow points), the annual precipitation is poorly correlated with its altitude.

Precipitation (mm/year) 590 - 630 630 - 660 660 - 700 700 - 740 740 - 780 780 - 820 820 - 870 880 - 930 930 - 1020 1020 - 1220 River network

N W S E

10

20 km

Figure VII-8: Distribution of yearly precipitation over the Hornad River basin

1250 Obtained from the distribution map

Precipitation (mm/year)

Observed 1000

750 y = 0.3921x + 484.79 R2 = 0.9838 500 0 200 400 600 800 1000 1200 1400 1600 1800 2000

Ele vation (m)

Figure VII-9: Vertical gradient of yearly precipitation with elevation A spatial distribution of yearly precipitation with grid size of 10001000 m over the Hornad River basin is given in Figure VII-8 provided by SHMU, in which the mean annual precipitation was created based on the climatological and geophysical information, taking account for the effects of altitude, slope and aspect. It gives a highly satisfactory qualitative representation of the gross features over the river basin, showing a continuous wet zone in the south-western mountain chain, high mean 151

Chapter VII values along the mountain ridges, and low values in the broad river valleys. By ordering the elevation for each cell and calculating the mean precipitation for each 100 m interval, the mean annual precipitation was found to have a close correlation with elevation as shown in Figure VII-9 (solid points) with a slope of 0.39 mm/year/100m and a correlation coefficient of 0.98.

2.3.3. Adjustment for PET To study the topographic effect on PET, 8 weather stations were selected in and surrounding the Hornad River basin with elevation ranging from 239 to 904 m and observation period from 10 to 40 years (Table VII-3). A linear regression analysis was performed between average monthly measured PET and its altitude. All months show a decrease in PET values with increasing elevation. Although these data are not available over the entire range of elevations in the region, the data are sufficient to demonstrate that PET decreases with increasing elevation, except the 5 winter months in which the regression slope is close to zero, the intercept is more or less equal to the monthly mean PET, and no correlations are found due to the low temperature (Table VII-4). 4 typical regression lines between monthly PET and its elevation are shown in Figure VII-10. However, the latitude, slope and aspect may cause a remarkable variation in the angle at which solar radiation intersects a hill slope, and therefore have considerable effects on the PET distribution. These adjustments on PET are not performed in this study due to data limit. Table VII-3: PET stations used for regression analysis
Station ID 11938 11934 11945 11963 11949 11955 11968 Station name Telgart Poprad Svedlar Jakubovany Spisske Vlachy Presov - vojsko Kosice - letisko Period (year) 1961-2000 1961-2000 1961-2000 1963-2000 1965-2000 1991-2000 1961-2000 Altitude (m) 904 695 475 398 382 295 239 X coordinate (m) 440382.01 444753.06 477080.25 509890.95 485521.91 521340.24 516238.02 Y coordinate (m) 5410897.25 5435331.20 5406705.20 5439402.77 5421193.81 5430299.31 5391001.42

152

Flow simulation in a Carpathian catchment accounting for topographic controls Table VII-4: Regression analysis between PET and elevation
Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Year Mean (mm/d) 0.03 0.17 0.87 1.90 2.74 3.19 3.35 2.87 1.85 0.88 0.23 0.02 1.52 Max. (mm/d) 0.07 0.30 1.19 2.29 3.22 3.68 3.92 3.43 2.21 0.98 0.36 0.06 1.82 Min. (mm/d) 0.00 0.03 0.48 1.55 2.34 2.70 2.87 2.49 1.61 0.78 0.08 0.00 1.27 Slope (%) 0.00 0.00 -0.09 -0.09 -0.11 -0.13 -0.13 -0.11 -0.06 0.00 0.00 0.00 -0.06 Intercept (mm/d) 0.04 0.22 1.30 2.35 3.27 3.83 3.98 3.40 2.14 0.90 0.22 0.04 1.82 R2 0.05 0.08 0.89 0.79 0.76 0.77 0.78 0.59 0.50 0.03 0.05 0.13 0.71

4.5 March May July September y = -0.0013x + 3.9764 3.0 R = 0.7804 y = -0.0011x + 3.273 R = 0.7633 1.5 y = -0.0006x + 2.1444 R = 0.4972 y = -0.0009x + 1.3041 R = 0.8912 0.0 200 400 600 800 1000 1200
2 2 2 2

PET(mm/d)

Elevation (m)

Figure VII-10: Vertical gradient of monthly PET with elevation 2.3.4. Topographic correction Using GIS ArcView, knowing the coordinates of each weather station, the Thiessen polygons for precipitation, temperature and PET can be created. Next, the daily mean precipitation, temperature and PET are computed for each grid cell within its Thiessen polygon by applying the regression expressions. The data used in the calculation are measured daily series at the reference station, elevations of the grid cell and the reference station, and the regression slope. For instance, the reference precipitation

153

Chapter VII series can be adjusted for each grid cell or subcatchment in which the precipitation is assumed to vary linearly with the elevation (Dingman et al., 1988):
Pi = Pref + Pref (H i H ref )

(7.3)

where Pi [LT-1] is the precipitation at cell i, Pref [LT-1] is the precipitation at the reference precipitation station, Hi and Href are the height at cell i and at the reference station, and is the precipitation regression slope. Likewise, the temperature and PET at each cell can be estimated based on the data series at reference station, lapse rate and the elevation difference. These topographical adjusted spatially distributed precipitation, temperature, and PET form the input to the WetSpa model. Using mean monthly temperature, yearly precipitation and monthly PET to describe their spatial distributions in a daily scale is under the assumption of linear relationship between elevation and climate features. However it is not sufficient to describe conditions strongly controlled by local terrain and climate features. Nevertheless, these adjustments provide an estimate of temperature, precipitation and PET that account for topographic effects obtained on a grid, and hence the model can account for the spatial variations of these input datasets more properly than using a unique value within each Thiessen polygon. 3. 3.1. Model Simulation Parameter identification

Once the required data are collected and processed for use in the WetSpa model, identification of spatial model parameters is undertaken. Terrain features at each grid cell including elevation, flow direction, flow accumulation, stream network, stream link, stream order, slope, and hydraulic radius are firstly extracted from the 100100 DEM. The grid of stream network is delineated by applying a threshold value to subset cells with an accumulated flow, for which cells that have more than 10 cells flowing into them corresponding to a drainage area of 0.1 km2 are assigned as stream network. The threshold value for determining subcatchments is set to 100, on which 585 subcatchments are divided with average subcatchment area of 1.93 km2. When

154

Flow simulation in a Carpathian catchment accounting for topographic controls creating the grid of surface slope, a minimum slope threshold of 0.01% is given in order to keep water moving on those cells without extremely low velocity. The grid of hydraulic radius is generated with an exceeding frequency of 0.5 (2-year return period), for which the network constant ap and the geometry scaling exponent bp are set to 0.05 and 0.48 resulting in average hydraulic radius of 0.005m for the upland cells and up to 1.5 m at the outlet of the watershed. Next, the grids of soil hydraulic conductivity, porosity, field capacity, residual moisture, pore size distribution index, and plant wilting point are reclassified based on the soil texture grid by means of its attribute lookup table. Similarly, the grids of root depth, interception storage capacity, and Mannings n are reclassified from the land use grid. Specifically, the Mannings n for stream channels is linearly interpolated based on the stream order grid with 0.055 m-1/3s for the lowest order and 0.025 m-1/3s for the highest order. The grids of potential runoff coefficient and depression storage capacity are obtained by means of attribute table combining the grids of slope, soil and land use, for which the percentage of impervious area in an urban cell is set to 30%. As can be seen in Figure VII-11, the non-afforested and steeper grid generates a very high runoff coefficient, whereas the afforested and gentle grid generates less surface runoff. The calculated average potential runoff coefficient is 0.43 for the entire catchment.

Runoff Coefficient 0.071 - 0.174 0.174 - 0.278 0.278 - 0.381 0.381 - 0.484 0.484 - 0.587 0.587 - 0.69 0.69 - 0.794 0.794 - 0.897 0.897 - 1

Figure VII-11: Potential runoff coefficient for the Margecany catchment The grids of Thiessen polygons for precipitation, temperature and PET are created based on the geographical coordinates of each measuring station and the catchment boundary using the Thiessen polygon extension of the ArcView Spatial Analyst.

155

Chapter VII Finally, the grids of flow velocity, travel time to the basin outlet and to the main river, as well as their standard deviation are generated, on which the IUH from each grid cell to the basin outlet and the main river is calculated. Figure VII-12 shows the estimated average flow time from each grid cell to the basin outlet. Flow time for the main rivers is generally less than 20 hours, whereas the flow time for the most remote area is around 2 days. The mean travel time for the entire catchment is 24 hours.

Flow time (h) 0-3 3-6 6-9 9 - 12 12 - 15 15 - 18 18 - 21 21 - 24 24 - 27 27 - 30 30 - 35 35 - 40 40 - 45 45 - 50

Figure VII-12: Mean travel time to the basin outlet for the Margecany catchment 3.2. Automated calibration

Distributed models are generally parameterized by deriving estimates of parameters from the topography and physical properties of the soils and land use of the basin. The reliability of model predictions depends on how well the model structure is defined and how well the model is parameterized. However, estimation of model parameters is difficult due to the large uncertainties involved in determining the parameter values, which can not be directly measured in the field. Therefore model calibration is necessary to improve the model performance. Because of the large number of model parameters and the complexity in simulating the hydrological response of a catchment, automated calibration techniques are becoming popular methods to account for spatial parameter variability, while reducing the model calibration effort. An automated calibration procedure is applied to the WetSpa model by incorporating a model-independent parameter estimator PEST (Doherty, 1994; Doherty & Johnston, 2003). The automated calibration approach is applied in this study to focusing only on

156

Flow simulation in a Carpathian catchment accounting for topographic controls the global parameters of the WetSpa model with observed flow hydrographs at selected stations as the calibration targets, including interflow scaling factor, baseflow recession constant, evapotranspiration correction factor, initial soil moisture, initial groundwater storage and maximum groundwater storage, three snowmelt parameters, and two surface runoff parameters. The spatial model parameters calculated with GIS tools are not calibrated in this study and remain as they are. Automated calibration of these spatially distributed parameters with PEST is recommended for the following research, for which more organization and computation efforts are needed. To accomplish the automated calibration process, three input files must be prepared at first, including an input template file, an output instruction file and a PEST control file. The template file is one for each model input file on which parameters are identified, and is needed only for those input files which contain parameters requiring optimization. The instruction file is one for each model output file containing the directions which PEST must follow in order to read or write that file. The PEST control file supplies PEST with the names of all template and instruction files together with the model input/output files to which they pertain. It also provides PEST with the model name, parameter initial estimates, measurements to which model outcomes must be matched, prior parameter information, and a number of PEST variables which control the implementation of the optimization method. The above three input files must be prepared before the model is run, and can be constructed using any text editor. Table VII-5: Parameters and their ranges in the PEST control file
Parameter Ci Cg K_ss K_ep G0 G_max T0 K_snow K_rain K_run P_max Description Interflow scaling factor (-) Baseflow recession constant (d ) Initial soil moisture Vs field capacity (-) PET correction factor (-) Initial groundwater storage (mm) Maximum groundwater storage (mm) Threshold melt temperature ( C) Melt-rate factor (mm C d ) Rainfall melt-rate factor ( C d ) Surface runoff exponent (-) Maximum rainfall intensity (mm)
o -1 -1 o -1 -1 o -1

Initial 1.5 0.01 1.0 1.0 10 20 0.0 2.0 0.0 3.0 40

Minimum 1.0 0.001 0.8 0.8 5.0 10 -0.5 0.0 0.0 1.0 10

Maximum 5 0.1 1.2 1.2 150 250 1.0 5.0 0.02 6.0 100

157

Chapter VII Table VII-5 lists the parameters, their initial estimates and possible range during automated model calibration. The best set of parameters is selected from within reasonable ranges by adjusting values until the discrepancies between observed and simulated hydrographs is reduced to a minimum in the weighted least squares sense. This scheme serves as an optimization algorithm to estimate the model parameters. A further manual calibration approach is implemented to avoid the ill-posed problems, which is commonly associated with direct inverse procedures.

3.3.

Modelling results

10 years (1991-2000) measured daily precipitation, temperature, PET, and discharge data as described in section 2.1 are used for both automated and manual model calibration. The calibration processes are performed mainly for the global model parameters, whereas the spatial model parameters are kept as they are. The initial global model parameters are specifically chosen according to the basin characteristics as discussed in chapter IV. The automated calibration procedures are firstly performed to initially estimate global model parameters. Next, a manual calibration approach is implemented for avoiding ill-posed problems and parameter optimization. The simulation results are compared to the observed hydrograph at Margecany both graphically and statistically. As for the first step, the parameters of base temperature and degree-day coefficients are adjusted independently in order to get a proper fit of snowmelt flood that occurred normally in late February and early March. The initial groundwater flow recession coefficient is estimated by analyzing the baseflow, which is separated from the observed hydrograph at Margecany. Adjustment of this parameter is necessary in accordance with the fitting of baseflow and the total flow volume. The interflow scaling factor is adjusted for the peak and recession part of the flood hydrograph, which is sensitive for both high and low flow evaluation. The additional two parameters controlling the amount of surface runoff, i.e. the surface runoff exponent for a near zero rainfall intensity and the rainfall intensity corresponding to a surface runoff exponent of 1, are adjusted mainly for small storms, in which the actual runoff coefficients are small due to the low rainfall intensity. The initial soil moisture and active groundwater storage are adjusted by comparison of the hydrographs and water balance for the initial phase. The maximum active 158

Flow simulation in a Carpathian catchment accounting for topographic controls groundwater storage controls the amount of vapour transpirated from the groundwater, and therefore can be adjusted by comparison of the flow volume and low flow during dry period. Figure VII-13 gives a graphical comparison between observed and calculated daily flow at Margecany for the year 1997. The input global model parameters after automated and manual calibration, the calculated water balance, and the model evaluation results are listed in Table VII-6.

100

0
Precipitation
Observed
Calculated

Discharge (m 3/s)

80 60 40 20 0 1/1

20 40 60 80 100

1/2

1/3

1/4

1/5

1/6

1/7

1/8

1/9

1/10 1/11 1/12

Date (d/m)

Figure VII-13: Observed and calculated daily flow at Margecany in 1997 Table VII-6: List of input parameter values, water balance, and evaluation results
1) Global model parameters Ci 2.0 Cg K_ss K_ep 1.10 I 551 DS 26.5 G0 30.0 F 85.3 1.66 C5 0.825 G_max 40.0 E 69.6 1.35 T0 0.0 Perc 28.6 0.56 K_snow K_rain K_run P_max 2.20 Rs 446 6.3 0.12 Ri 346 4.9 0.10 0.00 Rg 1310 18.4 0.36 3.0 R 2102 29.6 0.58 60.0 DG -2.25 -0.03 37.7 0.0085 0.95 P Sum (mm) 7108 % of P Mean C1 0.05 C2 0.919 1.95

2) Calculated water balance 6067 4947 2036

7.75 0.37 0.16 287 C3 0.738 C4 0.703

3) Model evaluation

Figure VII-13 shows that both the spring and summer flood hydrographs are well reproduced by the model. The simulation of snowmelt flood is important in this study as it not only contributes to the results of model evaluation, but also provides reliable soil moisture estimation at the end of snow melting period, which affects following

Precipitation (m m/d)

159

Chapter VII rainfall runoff processes. The calibrated base temperature and degree-day coefficient are 0oC and 2.2 mm/day/ oC, whereas the heat contribution from rainfall to the snowmelt is not important for this case study. The calibrated groundwater flow recession coefficient at Margecany is 0.0085 day-1, which coincides with the calculated value (0.01 day-1), and gives a good estimation for the whole simulation period. The peak discharges, concentration time, and flow volumes are especially well predicted for the three summer floods in 1997, and similar simulation results are obtained for other hydrological years. The model performance is satisfactory from the statistical evaluation results, in which the flow volume is 5.0% over estimated, model determination coefficient is 0.919, the Nash-Sutcliffe efficiency is 0.738 for the time evolution of stream flows, and the modified Nash-Sutcliffe efficiency is 0.703 and 0.825 respectively for low and high flows. These indicate that the model is able to consider the precipitation, antecedent moisture and runoff-generating processes in a spatially realistic manner based on topography, land use and soil type, resulting in a fairly high accuracy for both high and low flows, and the general hydrological trends being well captured by the model. The estimated annual precipitation after topographic correction is 710.8 mm/year over the 10 years simulation period, which coincides with the result obtained from Figure VII-8 for the Margecany catchment (704.8 mm/year). The estimated annual precipitation without considering the topographic correction is only 662.1 mm/year, which is much lower than the expected value, and may not reflect properly the spatial precipitation distribution over the catchment. From the results of model simulation, as shown in Table VII-6, 7.75% of the precipitation is intercepted by the plant canopy, 85.3% infiltrate to the soil, 69.6% evapotranspirates to the atmosphere, 28.6% recharges to the groundwater reservoir, and 29.6% becomes runoff, of which direct flow, interflow, and groundwater flow possesses 21.3%, 16.5%, and 62.2% respectively. These values are reasonable in view of the catchment hydrological characteristics. Figure VII-14 presents the simulated variation of precipitation, temperature, evapotranspiration, and relative soil saturation for the Margecany catchment during the year 1997. The precipitation series is decreased by 20 mm/d in the figure in order to give a clear view from other three time series. Obviously, large storms are concentrated in July and early August of the year. Mean temperature is below freezing in December, January and February, and fluctuates around zero in 160

Flow simulation in a Carpathian catchment accounting for topographic controls March, April, October and November. Accordingly, high evapotranspiration occurs from May to August, and the value is zero during winter season. The simulated average soil saturation varies accordingly, in which it remains stable during winter season as there is no water movement into or out of the root zone, increases temporarily in spring due to the infiltration taking place during snowmelt, and decreases dramatically afterwards due to the high evapotranspiration. High soil moisture is present in July and August as a result of intensive rainfall, while low moisture is present in June and September duo to the intensive evapotranspiration and the insufficient rainfall to recharge the soil moisture.

80

15
Relative saturation (%)

P, T, relative saturation

60 40 20 0
Precipitation-20 (mm/d)

10
Evapotranspiration (mm/d)

5 0 -5 -10

Temperature ( C)

-20 1/1 1/2 1/3 1/4 1/5 1/6 1/7 1/8 1/9 1/10 1/11 1/12

Date (d/m)

Figure VII-14: Variation of precipitation, temperature, evapotranspiration, and relative soil saturation for the Margecany catchment in 1997 A spatial distribution of simulated surface runoff for the storm event occurred in early August, 1997, is presented in Figure VII-15. One can see that the rainfall distribution is highly variable over space and time. The average rainfall over the catchment is 70 mm with measured maximum storm volume over the 5 days of 101 mm at Spisske Vlachy located in the downstream part, and the minimum storm volume of 50.2 mm at Ganovce located near the north-western boundary of the catchment. The calculated average surface runoff for this storm event over the catchment is 8.8 mm with surface runoff coefficient of 0.126. Apparently, high runoff occurs in the areas with high rainfall intensity, steep slope, clay soil and afforested land uses, while low runoff occurs in the areas with low rainfall intensity, gentle slope, sandy soil and forest land uses. The model can also produce other spatial output maps, such as soil moisture

Evapotranspiration

161

Chapter VII content, interflow, groundwater recharge, actual evapotranspiration, etc., in different time scales, which are not presented here.

50.2 Ganovce
#

51.4 Vikartovce Hranovnica # # 55

Levoca # 66

81.1 Spisske Podhradie #

63.6 Mlynky #
# #

Spisske Vlachy # 101 Krompachy

#

# 1/8/1997 - 5/8/1997 Surface runoff (mm) 0-5 5 - 10 Ch 10 - 15 15 - 20 20 - 30 30 - 40 > 40

62.3 Rudnany #
#

96.6

Mean = 8.8 mm

Figure VII-15: Distribution of surface runoff for the storm event 1/8-5/8, 1997 4. Discussion and conclusion 1) Spatial and temporal variability of input time series The spatial and temporal variability of rainfall, temperature, PET, and snow accumulation in the western mountains of the Hornad River basin is an important but poorly documented phenomenon. It governs the hydrological processes such as runoff generation, soil moisture condition, etc., and is one of the major uncertainties to the modelling results. A linear topographical correction is applied in the WetSpa model to compensate the elevation effects on the spatial distribution of input time series. However, the regression is performed based on the monthly temperature, PET, and yearly precipitation data, which is not sufficient to account for the nonlinearity of daily variables. Moreover, local variability in input time series results from variations in latitude, aspect, wind speed and direction, humidity, as well as sun radiation, is not taken into account in this case study. In particular, the snow pack is assumed to be accumulated from the solid precipitation without considering the wind-induced snow drift. These local factors have considerable effects on the spatial distribution of input time series and the modelling results as well when conducting water studies for a large mountainous catchment. Further researches may be undertaken on this subject once the relevant data are available.

162

Flow simulation in a Carpathian catchment accounting for topographic controls In this case study, the WetSpa model is conducted using measured precipitation, temperature and PET data on daily scale, which has shown success in the simulation of flood hydrographs at Margecany. Due to the large time resolution of precipitation time series, the effects introduced by the internal structure of rainfall are ignored. The complex structure of rainfall has been found to exhibit a large degree of temporal variability under natural conditions in the region (Kostka & Holko, 2002), which substantially affects the processes of runoff production and the downward water movement in the root zone. Studies have shown that accuracy of hydrological models can be improved by using fine time step resolution of rainfall data in the determination of infiltration excess runoff (Finnerty et. al., 1997). Therefore, hyetographs with a fine time resolution are required to capture rainfall variability for further research. The daily PET data observed at Spisske Vlachy is used as a reference series in the model, on which the topographical corrections are conducted over the catchment. As an alternative, an empirical sine curve as a function of date solely is developed based on the 10 years daily observed PET data. This curve can be used as a PET input for simulating future flood scenarios. The equations can be expressed as:
1.41

d 90 EPd = 0.12 + 1.311 + sin 2 365

(7.4)

where EPd (mm/day) is the daily PET, d is the day of the year, starting from 1 for the first of Jan. and ending with 365 for the 31st of Dec. Figure VII-16 gives a graphical presentation between observed and simulated mean daily PET at Spisske Vlachy.

4.5
Observed mean
Simulated

PET (mm/d)

3.0

1.5

0.0
1 31 61 91 121 151 181 211 241 271 301 331 361

Date

Figure VII-16: Observed and simulated mean daily PET at Spisske Vlachy 163

Chapter VII

2) Operation of the model for real time forecasts WetSpa has relatively modest requirements for input variables (temperature, precipitation, and PET) and therefore it is possible to be used for real time forecasts on hourly or daily basis. For short term forecasts, temperature and precipitation must be forecasted or predetermined for the coming days and substituted into the model, which are becoming increasingly available from meteorological services. For long term forecasts, such as monthly or seasonal runoff volumes, the PET series is becoming essential, which can be extracted from the mean PET curve or estimated based on the predicted temperature series. Differing from the conventional lumped model predictions, the distributed WetSpa model can predict not only the flood hydrograph at the basin outlet, but also the hydrographs at any selected stations inside the catchment, and the spatial distribution of hydrological processes for a certain period as well. The advantage of this modelling approach is to have the knowledge of how much, and where runoff will occur for a real time prediction by combining terrestrial GIS/RS data with meteorological information. In this respect, GIS techniques can be efficiently used to store, manage and display the spatial distribution of real time variables, such as rainfall, snow cover, runoff, soil moisture, as well as remotely sensed images, etc. The model performance in the forecasting mode is naturally affected by the reduced accuracy and reliability of temperature and precipitation forecasts. The propagation of errors can be avoided by periodical updating. Therefore, further possibilities of updating can be made available to users when more experience in real time situations is accumulated, for instance, to adjust some parameters (e.g., the potential runoff coefficient) in the progress of the forecast within hydrologically and physically acceptable limits. In any case, false forecasts of temperature and precipitation should be updated whenever a correction by new data is indicated.

References Arnell, N.W., The effect of climate change on hydrological regimes in Europe: a continental perspective, Glob. Environ. Change, 9, 5-23, 1999. 164

Flow simulation in a Carpathian catchment accounting for topographic controls Bengtsson, L. & Westerstrm, G., Urban snowmelt and runoff in northern Sweden, Hydrol. Sci. J., 37, 263-275, 1992. Dingman, S.L., Barry, R.G. & Reynolds, R.C., Application of kriging to estimating mean annual precipitation in a region of orographic influence, Water Resour. Bull., 24, 329-339, 1988. DHI, Danish Hydraulic Institute. MouseNAM Reference Manual 1.0, 1994. Doherty, J. & Johnston, J.M., Methodologies for calibration and predictive analysis of a watershed model, J. Am. Water Resour. Ass., 39(2), 251-265, 2003. Doherty, J., PEST: a unique computer program for model-independent parameter optimization, Watermark Computing, Washington, DC, 1994. Espeby, B., Tracing the origin of natural waters in a glacial till slope during snowmelt, J. Hydrol., 118, 107-127, 1990. Finnerty, B., Smith, M., Seo, D.J., Koren, V. & Moglen, G., Space time scale sensitivity of the Sacramento model to radar-gauge precipitation inputs, J. Hydrol., 102, 69-92, 1997. Flerchinger, G.N., Hanson, C.L. & Wight, J.R., Modelling evapotranspiration and surface energy budgets across a catchment, Water Resour. Res., 32(8), 25392548, 1996. Kostka, Z. & Holko, L., Analysis of rainfall-runoff events in the mountain catchment, eds., Holko L., Miklnek P., Parajka J. & Kostka Z., ERB and NE FRIEND Proj. 5 Conf., Interdisciplinary approaches in small catchment hydrology: monitoring and research, Slovak IHP UNESCO/IH SAS, 10-13, 2002. Leavesley, G.H., Problems of snowmelt runoff modelling for a variety of physiographic and climatic conditions, Hydrol. Sci. J., 34(6), 617-634, 1989. Lindstrm, G., Johansson, B., Persson, M., Gardelin, M. & Bergstrm, S., Development and test of the distributed HBV-96 hydrological model, J. Hydrol., 201, 272-288, 1997. Mocko, D.M. & Sud, Y.C., Refinements to SSiB with an emphasis on snow physics: Evaluation and validation using GSWP and Valdai data, Earth Interactions, 5, 131, 2001. Nijssen, B., ODonnell, G.M., Hamlet, A.F. & Lettenmaier, D.P., Hydrologic sensitivity of global rivers to climate change, Climatic Change, 50, 143-175, 2001.

165

Chapter VII Parajka, J., Mapping long-term mean annual precipitation in Slovakia using geostatistical procedures. In: Problems in Fluid Mechanics and Hydrology, Institute of Hydrodynamics, ASCR, Prague, 1999. Parajka, J., Holko, L., Kostka, Z., Snowmelt modelling and GIS, GIS at Development Magazine, 5(10), 23-27, 2001. Shevenell, L., Regional potential evapotranspiration in arid climates based on temperature, topography and calculated solar radiation, Hydrol. Process., 13, 577-596, 1999. Tabios, G.Q. & Salas, J.D., A comparative analysis of techniques for spatial interpolation of precipitation, Water Resour. Bull., 21(3), 365-380, 1985. Xu, Z.X. & Li, J.Y., A distributed approach for estimating catchment evapotranspiration: comparison of the combination equation and the complementary relationship approaches, Hydrol. Process., 17, 1509-1523, 2003.

166

Chapter VIII Integrating GIS and hydrological process modelling in medium and large watersheds
Abstract This chapter examines the effects of the DEM grid size on hydrological modelling by comparing the resulting runoff and flow responses from different resolution DEMs (50, 100, 200, 400, and 800 m) of the Alzette River basin, an 1176 km medium-sized watershed in the Grand-duchy of Luxembourg. Next, a comprehensive subwatershed parameterization method is developed for modelling medium and large-scale watersheds based on the distributed WetSpa model. A simplified approach is used in order to take advantage of the existing spatial analysis function within ArcView GIS and its Spatial Analyst and hydrological modelling extensions. The method of diffusive wave approximation is applied in tracing water for both overland flow and channel flow, while the water and energy budgets are maintained for each very small subwatershed derived from the high resolution DEM. Hydrographs at each subwatershed outlet are firstly calculated using the GIS derived subwatershed response function and then routed to the basin outlet using the channel response function. Model parameters and meteorological data input for each subwatershed are obtained by integration of the values from all cells of that subwatershed, allowing for the internal drainage structure within each subwatershed. Calibration and validation are performed on hourly basis using observed rainfall and discharge data from Dec. 1996 to Dec. 2000. Good agreement between the predicted and measured hydrographs has been achieved according to the graphical comparison and statistical assessment.

1.

Introduction

Watershed models have become useful tools in water resources planning and management, allowing prediction of stream flow from measured meteorological data. Evidently the spatial distribution of soil moisture and the production of runoff are dependent on basin topography, soil and land use patterns, especially for medium and

Chapter VIII large watersheds. Consequently, accounting for this spatial heterogeneity within watershed, models has long been considered as a prerequisite for improving water and energy flux predictions. Recent advances in computer hardware and software including increased speed and storage, debugging tools, and GIS technology have made possible the hydrological simulations of large catchments. The challenge then is to develop a basin-scale model that is computationally efficient, capable of simulating land management scenarios, allowing considerable spatial detail, requiring readily available inputs, and giving reasonable results (Arnold et al., 1998). Along with the rapid development of GIS technology and remote sensing techniques, especially the concomitant availability of high resolution DEM and the advances in integrating GIS with hydrological modelling, flood prediction with distributed models tends to be more advantageous and competent. However, model simulation for long time series on medium and large catchment scale with small grid size and short time interval is tedious, costly, and time consuming. This is because the computation time and the use of computer memory are affected by the number of cells involved in the catchment. If the watershed is large or the cell dimension too small, the number of cells increases so that the computation time increases and the free memory of the computer is often insufficient, and therefore the simulation is sometimes difficult or impossible to realize with a personal computer. One approach to cope with this problem is to increase the cell size, which may introduce errors by aggregation of spatial input data and misrepresentation of the true basin characteristics. For instance, increasing grid size causes the information content of the slope gradient to decrease slightly and the curvature of the landscape to decrease greatly. Studies have shown that grid size has remarkable effects on the simulation results of both runoff generation and flow routing. The coarser resolution results in a lower peak discharge and earlier time-to-peak, especially for the storm events of short duration and lower antecedent soil moisture condition (Braun et al., 1997; Molnar and Julien, 2000). This is in agreement with the argument used by Beven (1995) that the aggregation approach towards macro-scale hydrological modelling, using averaged parameter values, is inadequate for representing hydrological processes at a large scale. In addition, terrain information is lost as a result of data aggregation at a larger grid-cell size, and the high-resolution data is not fully utilized in the watershed modelling.

168

Integrating GIS and hydrological process modelling in medium and large watersheds It has been widely recognized that spatial scale, e.g. the size of grid cell or sub-basin, generally lead to predictive uncertainty in distributed hydrological modelling (Blschl & Sivapalan, 1995). Although such difficult and serious scale problems are yet far from any form of solution, many valuable ideas have been proposed to attempt to solve it and hence improve model reliability. For example, the simple scaling and multi-scaling frame (Gupta et al., 1994), the REA (Representing Element Area) concept (Wood et al., 1988), the GLUE (Generalized Likelihood Uncertainty Estimation) framework (Beven & Binley, 1992), the HRU (Hydrological Response Units) concept (Flgel, 1995), as well as the basin-scale model equations (Kavvas et al., 1998). In addition, many efforts have been focused on the effects of grid size on model parameters and performances. For instance, Quinn et al. (1991), Zhang and Montgomery (1994), Bruneau et al. (1995), and Wolock and Price (1994) looked at how grid size affected the computed topographic characteristics, wetness index, and outflow using TOPMODEL. In general, they found that the finer grid size gave more accurate results. Similar results have been obtained in recent studies (Franchini et al., 1996; Saulnier et al., 1997; Horritt & Bates, 2001; Moglen & Hartman, 2001). In this study, an operational method for the automated physiographic parameterization of a hydrological model is developed. The purpose is to provide an enabling technology for GIS-based medium and large watershed modelling, utilizing highresolution information as much as possible. The automated model parameterization is implemented using ArcView, which is a friendly desktop mapping and GIS tool that enables users to quickly select and display different combinations of data for creatively visualizing information. Moreover, the Avenue programming language and the spatial extension of ArcView enable the modelling of water budgets on a grid cell basis and the tightly integrating of GIS outputs with hydrological models. Using this technique, it is possible to derive all necessary model parameters rapidly and accurately, and at different discretizing levels by varying the number of GIS derived subwatersheds. A case study in the Alzette river basin, located in the Grand-Duchy of Luxembourg, is presented. The potential application, limitation and the major influencing factors of this method with respect to watershed modelling for flood prediction and water balance simulation on medium and large catchment scale are also discussed.

169

Chapter VIII 2. Study area and data availability

The analysis of grid size effects and the test of the modified WetSpa model are performed on the Alzette river basin using available hydro-meteorological data from Dec. 1996 to Dec. 2000. The topography and soil data are available in GIS form, while the land use data is obtained from remote sensed images. The study area covers an area of 1176 km2 and is located mostly in the Grand-duchy of Luxembourg. Highmagnitude floods occurred frequently and have caused important damages since the early 1990s. Figure VIII-1 shows the location of the study area, as well as the river system and the monitoring network. The bedrock of the basin is of sedimentary origin, consisting of flat fractured layers of marls, limestone, sandstone and schist (El Idrissi et al., 2002).

# Et telbru ck

Ell
#

Mer s ch
#

Mining Wate r Crop Grass Forest Urban

Luxembou rg-cit y
#

Esc h/Alzette
#

W S

10 km

Figure VIII-1: Study area and observation network

Figure VIII-2: Land use of the Alzette River basin

The study area has generally a rolling topography, with elevation ranging from 195 to 545 m and an average basin slope of 8.7%. Most tributaries of the Alzette are leftbanked and located in the western part of the basin. The soil is mostly a sandstone/limestone mixture with spots of clay and marl, and can be classified as

170

Integrating GIS and hydrological process modelling in medium and large watersheds texture categories of loamy sand (16.7%), silt loam (8.1%), silt (21.4%), loam (22.7%), sandy clay loam (10.9%), silt clay loam (15.4%) and clay loam (4.8%). The soil geological formations partially condition the land use patterns. Thus, in general, agricultural areas coincide with marls and forest areas with sandstones. The dominant land use types in the catchment are deciduous shrub and forest (33.7%), grassland (30.7%) and cropland (23.3). Because the study area is highly developed, urban area (11.2%) is one of the major land use types in the river basin, including several towns such as Ettelbruck, Mersch, Luxembourg-city, and Esch/Alzette (Figure VIII-2). Other land use types are free water surface (0.2%), and former mining areas (0.9%), which are located in the right bank tributaries of the upstream catchment and disturb the local hydrological behaviour during storm events. The climate in the region has a northern temperate humid oceanic regime without extremes. Rainfall has a relatively uniform distribution over the year. High runoff occurs in winter and low runoff in summer due to the higher evapotranspiration. Winter storms are strongly influenced by the westerly atmospheric fluxes that bring humid air masses from the Atlantic Ocean (Pfister et al., 2000), and floods happen frequently because of the saturated soils and the low evapotranspiration. The average annual precipitation in the region varies between 800 mm to 1000 mm. Precipitation generally exceeds potential evapotranspiration except for the four months in the growing season. A dense rainfall and discharge monitoring network was set up to study the hydrological behaviours of the river basin in 1995 as shown in Figure VIII-1. Water levels are recorded at 16 streamgauges at a 15-minute time step, while rainfall is collected via 19 daily raingauges and 4 hourly raingauges, covering the study area and modelling period. To define the hourly rainfall distribution over the catchment, the daily areal rainfall interpolated via Thiessen polygons is time disaggregated according to the temporal structure of the 4 hourly reference raingauges (El Idrissi et al., 2002). Potential evapotranspiration is estimated using the Penman-Monteith formula with measured meteorological data at Luxembourg-city (Findel Airport), and extended to each rainfall Thiessen polygon based on the proportions of different land use type over the polygon (Drogue et al., 2002).

171

Chapter VIII 3. Effects of grid size on runoff and flow responses

The 50 x 50 m maps of elevation, soil type, land use, and watershed boundaries for the study area are aggregated to produce 5 sets of data with grid sizes of 50, 100, 200, 400, and 800 m. Due to the gentle relief for some areas along the catchment boundary, DEM aggregation using a simple GIS resample command may cause divergence of the stream lines, and hence can not represent properly the drainage system of the catchment. To serve this problem, the ArcInfos TOPOGRID function is used to create DEMs with different grid size based on the information of 2 m contour map, the catchment boundary and the digitalized stream network. The digital land use and soil type maps with the same cell size are resampled from the 5050 m land use and soil type grid by means of the nearest-neighbour assignment. The spatial analysis of the physical elements of the catchment and the preparation of model parameters are implemented using GIS ArcView. Table VIII-1: Mean parameter values calculated from maps with different grid size
Parameters Elevation Flow length Slope RMS profile curvature Hydraulic radius Flow velocity Flow time Standard deviation Urban percentage Loam percentage Potential runoff coefficient Depression storage capacity unit (m) (km) (%) (%m-1) (m) (m/s) (h) (h) (%) (%) (-) (mm) 50 m 332 37.9 8.18 12.4 0.017 0.08 22.4 6.71 11.3 22.6 0.426 2.01 100 m 332 36.2 6.41 6.26 0.022 0.11 20.5 6.26 11.3 22.6 0.410 2.24 200 m 332 33.7 4.41 3.12 0.038 0.14 18.9 6.20 11.1 22.6 0.389 2.59 400 m 333 31.8 2.68 1.28 0.066 0.18 16.8 5.59 11.3 22.6 0.369 2.97 800 m 333 30.1 1.64 0.45 0.112 0.27 11.2 4.40 12.0 22.6 0.359 3.27

The effects of grid size on parameters including elevation, flow length, slope, curvature, hydraulic radius, flow velocity, flow time and its standard deviation, urban and loam percentage, potential runoff coefficient and depression storage capacity, are investigated as listed in Table VIII-1. The value of the model parameter derived from the 5050 m DEM is used as the reference value, with estimates of model parameters for DEMs of coarser resolutions compared against it. It is found from the table that the mean elevation, urban percentage and loam percentage remain more or less the 172

Integrating GIS and hydrological process modelling in medium and large watersheds same for different grid size DEMs. As the resolution grows coarser, estimates of mean flow length, slope, root mean square (RMS) profile curvature, potential runoff coefficient, flow time and its standard deviation decrease consistently. On the other hand, the mean hydraulic radius, flow velocity and depression storage capacity increase consistently as the resolution grows coarser (Figure VIII-3). The decrease of potential runoff coefficient and increase of depression storage capacity are mainly caused by the decrease of the derived grid slope, while the decrease of flow time is due to the decrease of flow length and the increase of flow velocity.

(a) 50
40
Flow length (km) Flow time (h) Standard deviation (h)

(b) 15
12
Para meter v alues .

Parameter values

Slope (%) RMS curvature (%m-1 ) Depression capacity (mm)

30 20 10 0 0 200 400 600 800

9 6 3 0
0 200

Grid size (m)

400 600 Grid size (m)

800

(c) 0.6
Potential runo ff coefficien t Flow velocity (m/s) Hydraulic radius (m) 0.4

(d) 0.08
50 m 100 m 200 m 400 m 800 m

Para meter val ues .

0.06
IUH (h )
-1

0.04

0.2

0.02

0 0 200 400 600 Grid size (m) 800

0 0 10 20 30 Time (h) 40 50

Figure VIII-3: Mean parameters obtained from DEMs with grid sizes: (a) flow length, flow time and its standard deviation, (b) slope, curvature and depression capacity, (c) runoff coefficient, velocity and hydraulic radius, (d) IUHs for the entire catchment.

173

Chapter VIII Figure VIII-3(d) presents the calculated catchment IUHs obtained from DEMs with different grid sizes. Obviously, the IUHs peak increases consistently with considerably shorter times to its peak value as the resolution grows coarser. These biases lead to a tendency to overestimate peak discharge for a certain amount of runoff when coarser data are used. In addition, the calculated lateral interflow decreases dramatically as the grid size increases due to the decreasing slope gradient. Figure VIII-4 presents the hydrographs at Ettelbruck for a flood event in Dec. 1999 calculated from base maps with grid size of 50, 200 and 800 m. One can see that the lag time decreases consistently with the increasing map resolution, but the flood volume and peak discharge are decreasing on the contrary. These are due to the decrease of estimated interflow as the resolution grows coarser (Table VIII-2). Figure VIII-5 presents the surface runoff hydrographs independently calculated from maps with grid size of 50, 200 and 800m, in which the peak discharge increases consistently with the increasing grid size. The characteristics for the flood event in Dec. 1999 estimated from maps with different grid size are presented in Table VIII-2. Apparently, the estimated surface runoff, interflow, total runoff, peak discharge and its lag time decrease consistently with the increasing grid size, while the amount of infiltration, percolation, groundwater flow, and soil moisture are increasing on the contrary. The calculated groundwater flow is not important for this flood event. Water percolated to the groundwater reservoir is released slowly feeding low flow and further evapotranspiration from groundwater reservoir.

200 160 Discharge (m /s)

3

0 Precipitation Q observed Q 50 m Q 200 m Q 800 m 4 8 12 16 20 17/12

120 80 40 0 11/12

12/12

13/12

14/12 Time (d/m)

15/12

16/12

Figure VIII-4: Comparison of flow hydrographs at Ettelbruck calculated from DEMs with different grid size for a flood event in Dec. 1999 174

Precipitation (mm/h) .

Integrating GIS and hydrological process modelling in medium and large watersheds

160 Precipitation Q 50 m Surfa ce flow (m /s) 120 Q 200 m Q 800 m 80

10

40

15

0 11/12

12/12

13/12

14/12 Time (d/m)

15/12

16/12

20 17/12

Figure VIII-5: Comparison of surface flow hydrographs at Ettelbruck calculated from DEMs with different grid size for a flood event in Dec. 1999 Table VIII-2: Flood characteristics estimated from maps with different grid size
Characteristics Surface runoff Infiltration Interflow Percolation Groundwater flow Total runoff Soil moisture on 13/12 Peak discharge Lag-time unit (mm) (mm) (mm) (mm) (mm) (mm) (m3/m3) (m3/s) (h) 50 m 18.1 64.8 16.2 34.1 2.2 36.5 0.606 176.5 11 100 m 17.6 65.4 14.3 35.2 2.4 34.3 0.617 164.3 8 200 m 17.2 65.8 10.4 36.3 2.6 30.2 0.629 158.5 6 400 m 16.8 66.1 7.9 39.2 2.8 26.5 0.642 153.8 5 800 m 16.1 66.7 4.1 41.3 3.0 23.2 0.654 151.0 4

3.

Transforming WetSpa into a semi-distributed model

From the analysis above, it can be concluded that coarser grid size may result in significant bias for model predictions. Therefore, flood modelling using a high resolution DEM is preferred to improve model reliability. In order to deal with the computing time and memory problem when applying the distributed WetSpa model to a medium or large river basin with a fine grid size, the model is modified to a semidistributed pattern, where the water budget is computed for each very small subwatershed, built up from high resolution DEM data, rather than large grid cells with approximately the same area as the subwatershed. The GIS derived very small

175

Precipitation (mm/h) .

Chapter VIII subwatershed serves as a relatively homogeneous hydrological unit, with the same precipitation and potential evapotranspiration. The advantage of this method is to maintain the internal drainage structure of the subwatershed, for which surface runoff, interflow and groundwater flow can be estimated at the subwatershed outlet, and the respecting water and energy are balanced for each subwatershed. Simulations using large grid cells do not have this advantage, because the cell boundary may not reflect the true drainage boundary adequately. In the modified WetSpa model, the soil water balance in the root zone layer for a controlling unit is described as:
= P I V E R F t

(8.1)

in which [LL-1] represents the change of average soil moisture of the subwatershed during time t [T] in the root zone layer with depth D [L]. The flow across the controlling interface consists of infiltration, which is calculated by precipitation P [LT-1] subtracted by the initial abstraction I [LT-1], including interception and depression, and the surface runoff V [LT-1], evapotranspiration E [LT-1], percolation out of root zone R [LT-1] and lateral interflow F [LT-1]. Using GIS ArcView, the meteorological and hydrometrical stations are georeferenced to the base map and a spatial analysis of the precipitation, and PET is undertaken for the entire catchment with ArcView Thiessen Polygon extension. The mean precipitation and PET for each subwatershed as inputs to the model are then calculated by weighing the data of the neighbouring meteorological stations, for which the weight of a given station is predetermined by its area percentage in the subwatershed, using a high resolution base map and the ArcView Spatial Analyst extension. In such a way, the spatial distribution of meteorological variables is fairly taken into account in the model. Model parameters such as interception storage capacity, depression storage capacity, potential runoff coefficient, overland roughness coefficient, root depth, soil property parameters (saturation hydraulic conductivity, porosity, field capacity, wilting point, residual moisture content, pore size distribution index, etc.), average travel time to the outlet, dispersion coefficient and so on, are firstly calculated for each grid cell using ArcView lookup tables and the high resolution DEM, soil type and land use maps, or a combination of the three base maps. 176

Integrating GIS and hydrological process modelling in medium and large watersheds The mean parameters for each subwatershed are obtained by integrating the values from all cells of that subwatershed. In such a way, the minority categories of land use and soil type are taken into account in each subcatchment, and the flow time and dispersion coefficient are calculated based on the high resolution DEM. These result in a better estimation of the natural properties compared with the method using large grid cells, since a unique value is assigned to each pixel for grid type data and therefore the grid resample enumeration may cause errors in reflection of the cell characteristics. Using the predetermined parameters above for each subwatershed, the model is operated within the ArcView interface and with the input meteorological data of hourly precipitation and PET of different gauging stations. For continued simulation purpose, the storages of the four stores are estimated for each control unit and for each time step. The sum of the interception and depression storage forms the initial loss at the beginning of a storm, and does not contribute to the storm flow. Typical interception capacity values for different land use categories can be found in the literature, and the mean value of the controlling unit is calculated accordingly. Default values of depression storage capacity for different combinations of slope, land use and soil type are interpolated based on the values found in the literature, and the mean value of the controlling unit is obtained by means of weighting average for the cells within the subwatershed weighted by the land use area and the corresponding potential runoff coefficient. Water held in depressions at the end of rain depletes either by evaporation with a potential evaporation rate or contributes to the soil moisture with a linear decay function. The modified coefficient method is used to estimate surface runoff in the model, where the volume of surface runoff is a function of the potential runoff coefficient and the average soil moisture of the simulation unit, so that the actual runoff coefficient varies with time, rainfall intensity, rainfall duration and unit characteristics as described in chapter IV, which gives an approximation to the surface runoff volume at each time step. The potential runoff coefficient serves as a measure of rainfall partitioning capacity under ideal soil moisture conditions, depending upon slope, soil type, land use and the proportions of bare soil and impervious areas on the land surface. Default runoff coefficients are interpolated from the values found in the 177

Chapter VIII literature, and may be revised during calibration. The actual evapotranspiration in the model includes evaporation from interception and depression storage, the evapotranspiration from root zone limited by the PET rate and the soil moisture content, and the evapotranspiration from groundwater storage. Water stored in depressions evaporates with a potential rate, while a vegetation coefficient depending on growing stage and vegetation type is multiplied for calculating evapotranspiration from the soil. Deep evapotranspiration from groundwater storage is then calculated by the residuals of the PET multiplied by a variable coefficient depending on the groundwater storage and its capacity. The percolation out of the root zone is assumed to be controlled by gravity alone, and is defined by the Brooks and Corey relationship between hydraulic conductivity, effective saturation and soil pore size distribution index (Eagleson, 1978). The amount of interflow generated in the subwatershed is estimated based on Darcys law and the kinematic wave approximation, which is a function of the subwatershed scale effective values of hydraulic conductivity, the degree of soil saturation, the slope angle and the root depth. Interflow is assumed to occur after percolation and becomes significant only when the water content of the root zone soil is between field capacity and saturation. Finally, the groundwater flow is estimated towards the subwatershed outlet, as a function of groundwater storage and a recession coefficient, which can be computed, if measured stream flow data is available, or adjusted by model calibration. The routing of overland flow and channel flow is carried out using the method of diffusive wave approximation of the continuity equation and the St. Venant momentum equation by assuming one-dimensional unsteady flow, and neglecting the inertial terms and the lateral inflow to the flow element. A linear approximate solution proposed by De Smedt et al. (2000) is used to serve the equation in the form of a first passage time distribution, relating the discharge at the end of a flow path to the available runoff at the start of the flow path as described in chapter III. All routing parameters are estimated firstly for each grid cell with a high resolution DEM. The unit response functions for each simulation unit at their outlets are then obtained by integration of the cell response functions within that subwatershed weighted by its potential runoff coefficient, whereas the channel response functions from each subwatershed outlet to the basin outlet are obtained by integration of the cell response functions along the stream channel. If the discretization of the watershed is highly 178

Integrating GIS and hydrological process modelling in medium and large watersheds intensive, the water may flow out of the subwatershed within the first time step, and the process of flow routing within each subwatershed can be omitted. However, this process is maintained in the model in case of coarse discretization of the catchment or very small time intervals of the model simulation. With the above unit response functions defined for each simulation unit, stream flow can be routed accumulatively downstream. Three options are considered in the outflow hydrograph simulation. The first option is to compute and add interflow and groundwater flow to surface runoff at the subwatershed outlet, and route the total flow from each subwatershed accumulatively down to the watershed outlet. The second option is to add interflow to groundwater, and route surface runoff and groundwater downstream separately. The third option is to route all three components separately. The last two options are not computationally efficient compared to the first option, but can give a clear view of subsurface hydrological responses to the storm event and the distribution of surface runoff, interflow and groundwater throughout the watershed. The challenge in model calibration is that virtually no observed interflow and groundwater flow data are acquirable for medium or large watersheds. A workable solution might be to use the available baseflow separation models to separate baseflow and interflow adequately from the measured outflow hydrographs, for example the method developed by Wittenberg (1999). All model parameters can be derived within GIS ArcView framework, using lookup tables based on the three base maps. The hydrological data series needed for the input of the model are hourly precipitation and PET at each measuring station, and the outputs of the model are hydrographs at the basin outlet or any other flow measuring point inside the catchment, the variation of basin mean hydrological variables, as well as the distribution of hydrological characteristics for each subwatershed of the terrain. The division of the watershed should be performed according to the project purpose and of the terrain. Dividing the watershed into very small subwatersheds is necessary in order to take care of the spatial variability of the hydrological processes. However, for a medium sized plain watershed, if the project purpose is flood prediction only at the outlet of the watershed, dividing that watershed into a few hundreds or even tens of subwatersheds might be sufficient. Then, the modified WetSpa model could be considered as becoming more and more semidistributed as the dimensions of the 179

Chapter VIII subbasins increase. A few simulations are necessary to decide how many subwatersheds are necessary to meet various objectives going from flow forecasting at the basin outlet to detailed stream flow simulation at a number of points on the river network, together with interest in the spatial variability of hydrological characteristics, such as surface runoff, interflow, soil moisture, etc.

4.

Results and Discussion

Based on the point elevation measurements, a 50 by 50 m pixel resolution DEM is generated. The digital land use and soil type map with the same cell size are converted from the available land use and soil coverage. The cell-based parameters are then identified using GIS tools and lookup tables, which relate default parameters to the base maps, or the combination of base maps. Starting from the high resolution DEM, hydrological features including aspect, surface slope, flow direction, flow accumulation, flow length, stream network, and drainage area are delineated. Maps of porosity, field capacity, wilting point, residual moisture, saturated hydraulic conductivity, and pore size distribution index are obtained from the soil type map. Maps of root depth, Mannings roughness coefficient and interception storage capacity are derived from the land use map. Maps of default runoff coefficient and depression storage capacity are calculated from the slope, soil type and land use class combinations. For the urban areas, due to the model grid size, cells may not be 100% impervious in reality. In this study, the percentage of impervious area in the grid cell is computed based on land use classes, with 30% for residential area, 70% for commercial and industrial areas. Default runoff coefficients for these areas are calculated by adding the impervious percentage with a grass runoff coefficient multiplied by the remaining area. These result in runoff coefficients of 40 to 100% in urban areas, while other areas have much smaller values, down to 3% for forests with sandy soils and practically zero slopes. The average hydraulic radius for each grid cell is calculated according to the controlling drained area using a power law relationship. For normal floods, the estimated minimum hydraulic radius for overland flow was 0.005 m, and maximum 1.5 m for channel flow at the basin outlet. The values can be increased for extreme floods. The channel roughness coefficients are set in the model based on the stream orders with 0.025 m-1/3s for the highest order, 0.045 m-1/3s for the lowest order, and linear interpolations in between, considering the effect of river bed 180

Integrating GIS and hydrological process modelling in medium and large watersheds composition and geometry. By combining the maps of the average hydraulic radius, the Mannings roughness coefficient and the surface slope, average flow velocity in each cell is calculated using Mannings equation, which results in values of 0.005 m/s for overland flow on upland areas in the watershed, and up to 2.7 m/s for some parts of the main river. Next, the celerity and dispersion coefficients in each cell are produced by the equations described in chapter III. The contributing area is then determined from topographic data for a particular downstream convergence point, i.e. normally the cells corresponding to the stream flow gauging sites and the basin outlet. For each contributing grid cell, the flow time and its standard deviation are calculated by integration along the flow paths, using the GIS FLOWLENGTH routine. A spatial distribution of the average flow time to the basin outlet is shown in Figure VIII-6, in which the average flow time is less than 10 h for the main river and up to 55 h for the most remote areas.

5 - 10 10 - 20 20 - 30 30 - 40 40 - 50 > 50

h h h h h h
N W S E

N W S E

10 km

10 km

Figure VIII-6: Mean travel time to the basin outlet

Figure VIII-7: River reaches and divided subwatersheds

For the simulation of hydrographs both at the basin outlet and the flow monitoring points inside the catchment, and the interest in the spatial variability of hydrological

181

Chapter VIII characteristics over the catchment, the basin is divided into 2276 subwatersheds, as shown in Figure VIII-7, corresponding to the threshold value of 100 when delineating the stream network based on the flow accumulation theme. The areas of the GIS derived subwatersheds range from 0.025 to 3.8 km2 with average subwatershed area of 0.52 km2 corresponding to a grid mesh with cell size of 720 m. The flow unit response functions at each subwatershed outlet and from the subwatershed outlet to any downstream converging point can be obtained by the method described in chapter III. The flow routing within each subwatershed is not important if the GIS derived subwatersheds are very small, as is the case in this study, since most excess water may flow out of the subwatershed during the first time step with little damping effects. Routing of the flow is therefore mainly governed by the channel geophysical conditions. However, flow routing within each subwatershed becomes more and more important along with the increasing subwatershed scale. In this study, stream net delineating thresholds of 200, 300, 500, and 1000, resulting in the subwatershed numbers of 1207, 797, 455 and 233, are also tested during model calibration, for which the predicted hydrograph at the basin outlet did not change too much, but the simulation results for the small watershed inside the catchment are affected considerably. Both automated and manual calibration of the model are realized at Ettelbruck for the period of 1997-1998 and the model is validated using the available hourly precipitation, PET and stream discharge data at different stations for the period of 1999-2000. The parameters that need to be adjusted during model calibration are mainly interflow scaling factor, groundwater flow recession coefficient, plant coefficient, the initial relative soil moisture and initial groundwater storage, while the runoff exponent is set to one and other distributed model parameters are predetermined using GIS tools before the model operation and normally remain constant as they are. The predetermined distributed parameters of potential runoff coefficient, hydraulic radius, Mannings coefficient, etc., can also be adjusted during calibration. The predicted hydrographs at other flow monitoring sites inside the catchment can also be seen as model verifications, but only the simulation results at Ettelbruck, the basin outlet, are presented here. Figure VIII-8 shows the observed versus the simulated stream flow and the estimated base flow (interflow plus groundwater drainage) at Ettelbruck for a compound flood event that occurred in Feb. 182

Integrating GIS and hydrological process modelling in medium and large watersheds and Mar. 1997. Three large storms happened successively during the period, and caused serious floods over the catchment. The total recorded rainfall over the catchment during the period was 147.8 mm, with 56.5, 22.8 and 67.4 mm for the three storms respectively, corresponding to observed peak discharges of 115, 105 and 190 m3/s at Ettelbruck for each individual flood. As can be seen in the Figure, the predicted hydrograph fits the measured hydrograph very well. The volume of surface runoff takes about 53% of the total flood volume, while the interflow and groundwater flow is about 47% of the total flood volume. The first storm did not produce a relevant big flood compared with its storm volume, as a large quantity of rainfall was used for saturating the soil, while the second and the third storm generated a higher proportion of runoff. This was approved by the predicted hydrograph, where the calculated surface runoff coefficient was 0.30, 0.35 and 0.39 for the three floods respectively.

250

200 Precipitation

3 Calculated outflow

150

Calculated baseflow Observed outflow

100

50

12

0 10/2/97 13/2/97 16/2/97 19/2/97 22/2/97 25/2/97 28/2/97 3/3/97 6/3/97 9/3/97 12/3/97 15/3/97

15

Figure VIII-8: Calculated Vs observed flows at Ettelbruck for the floods in Feb. 1997 A graphical comparison between calculated and measured flows at Ettelbruck for the validation year 1999 is presented in Figure VIII-9, in which the hourly precipitation and discharges are integrated to a daily value, and the vertical axis is on a logarithmic scale in order to give a clear distinction for the low flows. With the initial hydrological condition at the end of calibration period, the validation results for the year 1999 and 2000 are in fairly good agreement with the measured discharges. The

183

Precipitation (mm /h)

Discharge (m 3/s)

Chapter VIII model can well reproduce high flows, but low flows are a little over estimated. This may be because of the simplification of the groundwater component in the model, or less accurate estimation of the evapotranspiration from groundwater storage during dry periods. Similar simulation results can be obtained for other flow gauging sites inside the catchment, but the model performance for the stations with small drained area is not as good as the stations with large areas. However, by discretizing the large catchment into very small subwatersheds, the model is able to consider the precipitation, antecedent moisture and runoff-generating processes in a spatially realistic manner based on topography, land use and soil type, resulting in a fairly high accuracy for high flows, and the general hydrological trends being well captured by the model.

10000

1000

20 Precipitation

100

Sim ulated
Observed

40

10

60

80

0.1 1/99

100
2/99 3/99 4/99 5/99 6/99 7/99 8/99 9/99 10/99 11/99 12/99

Figure VIII-9: Simulated Vs observed daily flows at Ettelbruck for the year 1999 Three model performance criteria, i.e. model bias, model confidential coefficient and the Nash-Sutcliffe coefficient (Nash & Sutcliffe, 1970) as described in chapter V, are performed to the simulation results simultaneously after the model run. The calculated values of the three criteria at Ettelbruck are 0.06, 0.84 and 0.88 for the calibration period, and 0.04, 0.81 and 0.85 respectively for the validation period, which are satisfactory in the view of model assessment, and indicate that the modified WetSpa model is able to provide good fits to the observed hydrographs at the basin outlet. The model performance at Ettelbruck and other 15 flow stations inside the catchment covered by the three assessment criteria for the whole simulation period on hourly 184

Precipitation (mm/d)

Discharge (m 3/s)

Integrating GIS and hydrological process modelling in medium and large watersheds scale, together with the subwatershed characteristics, are presented in Table VIII-3. Model biases are within the range of -0.08 to 0.05. The model confidential coefficients are with the range of 0.53 to 0.84 with mean value of 0.73, and the NashSutcliffe coefficients are within the range of 0.47 to 0.87 with mean value of 0.74. It is found that better performance exists for the watershed with large areas, while for small watersheds the performance is less satisfactory. For example, the three evaluation values at Huncherange, a 7.3 km small watershed, are 0.05, 0.53 and 0.47, and the outflow hydrograph is poorly predicted. This indicates that the modified WetSpa model can give a good representation of flow hydrographs for larger catchments. For a small watershed, a fully distributed model based on high resolution DEM and with smaller time interval is preferred. Table VIII-3: Subwatershed characteristics and model performance
River Station Area Slope Urban Crop Grass Forest Water CR1 (km) (%) (%) (%) (%) (%) (%) 1176 8.7 705 8.0 408 7.0 349 6.5 291 6.7 233 7.2 294 8.8 255 8.9 107 9.4 47.1 11.1 34.6 6.1 172 8.9 47.2 5.0 84.7 7.8 18.3 6.3 7.3 4.9 11.2 15.3 20.5 19.2 17.8 18.6 4.8 4.1 3.5 4.8 3.9 6.6 6.4 11.6 8.9 6.2 23.3 22.8 23.1 25.4 27.4 28.9 23.7 24.7 20.9 32.4 19.1 23.4 31.3 22.7 30.0 45.9 30.7 28.0 24.3 26.8 25.3 22.9 36.8 37.2 33.7 25.8 51.6 33.2 45.0 33.9 50.6 15.8 33.7 32.0 29.0 25.2 25.4 24.7 34.5 33.9 41.8 36.7 25.0 36.6 17.3 31.6 10.5 32.0 0.2 0.3 0.4 0.4 0.5 0.5 0.1 0.1 0.0 0.2 0.2 0.1 0.0 0.1 0.0 0.2 0.05 0.02 -0.01 -0.03 -0.05 -0.08 0.05 0.03 -0.01 -0.02 0.03 -0.04 -0.01 0.04 0.04 0.05 CR2 0.83 0.78 0.82 0.83 0.78 0.71 0.76 0.84 0.81 0.63 0.75 0.77 0.61 0.72 0.55 0.53 CR3 0.87 0.76 0.85 0.82 0.79 0.77 0.79 0.82 0.85 0.58 0.73 0.81 0.64 0.70 0.53 0.47

Alzette Ettelbruck Alzette Mersch Alzette Steinsel Alzette Pfaffenthal Alzette Hesperange Alzette Livange Attert Bissen Attert Useldange Attert Ell Roudbach Platen Pall Niederpallen Eisch Hunnebour Eisch Hagen Mamer Schoenfels Mamer Mamer Mierbech Huncherange

5.

Conclusions

The effects of grid size on catchment characteristics, runoff and flow responses are investigated in this chapter. It is concluded that significant bias in the modelling results can be generated when using a coarse resolution data. On the other hand, the computing time and memory problems arise when modelling a medium or large 185

Chapter VIII catchment with a fine resolution data. To overcome these problems, a practical approach by dividing a large catchment into very small subwatersheds based on highresolution digital graphic data is developed. The modified WetSpa model uses elevation, soil and land use data in a simple way to estimate runoff and soil moisture within root zone on the scale of GIS derived natural units, allowing for the internal drainage structure of WetSpa for each small subwatershed. The discretization of the catchment is implemented according to the model purpose, terrain complexity, and the interest of spatial variable distributions. In case of fine discretization, channel characteristics are more important in controlling the routing of stream flow, which have to be determined properly. On the other hand, flow routing inside the subwatershed has to be taken into account for coarse discretization, in order to predict outflow hydrographs more accurately. The model has been applied to the 1176 km Alzette river basin in Luxembourg with 4 years observed hourly rainfall and stream flow data. Results from the integrated output show that it adequately predicts stream flow at the basin outlet and the flow gauging sites with relatively large drained area inside the catchment. Working on a very small simulation unit, as illustrated in the case study, the model is also capable of predicting spatial distributions of hydrological variables on a natural unit scale. Despite the good performance in the case study, the model suffers some limitations in practice. Firstly, the model requires high resolution DEM, land use and soil type maps as input, and continuous data of precipitation, PET, and discharge for model calibration, which cannot be fully met for many watersheds. Secondly, the model employs many default parameters, which are interpolated from the literature and used over the entire catchment. Due to the large variation range, parameters such as hydraulic conductivity, roughness coefficient, etc., may change greatly when applying the model to other areas with quite different environment, and therefore model calibration is necessary. In addition, the model generates runoff using a modified coefficient method, which is empirical-based rather than on equations more closely representing physical processes. Though definitely a limitation, the use of the method has its advantages of closely relating runoff with cell characteristics such as slope, land use, soil type and moisture content, and has the potential to predict the impact of human activities on hydrological behaviours over the watershed.

186

Integrating GIS and hydrological process modelling in medium and large watersheds The use of the modified WetSpa model with its physiographic parameters provided by ArcView GIS has the attraction of reducing the cost and time required for parameterization and model simulation. This advantage is more outstanding when modelling a large watershed with long data series and short time intervals. Finally, with its display options allowing monitoring of various variables for each simulation unit during model operation, the model appears to be a good tool for understanding and managing phenomena related to hydrological processes. The implementation of the model entirely within ArcView using Avenue scripts along with the Spatial Analyst and hydrological extensions enables the integration of modelling capabilities within a GIS environment. The method makes full use of the spatially distributed hydrological and geophysical characteristics of the catchment, and therefore is suitable for stream flow forecasting and scenario simulations.

References Arnold, J.G., Srinivasan, S., Muttiah, R.S. & Williams, J.R., Large area hydrologic modeling and assessment, Part I: Model development, J. Am. Water Resour. Ass., 34, 73-87, 1998. Beven, K.J., Linking parameters across scales-subgrid parameterizations and scale dependent hydrological models, Hydrol. Process., 9, 507-525, 1995. Beven, K.J. & Binley, A.M., The future of distributed models: model calibration and predictive uncertainty, Hydrol. Process., 6, 279-298, 1992. Blschl, G. & Sivapalan, M., Scale issues in hydrological modelling: A review, Hydrol. Process., 9, 251-290, 1995. Braun, P., Molnar, T. & Kleeberg, H.B., The problem of scaling in grid-related hydrological process modelling, Hydrol. Process., 11, 1219-1230, 1997. Bruneau, P., Gascuel-Odoux, C., Robin, P., Merot, P.H. & Beven, K., Sensitivity to space and time resolution of a hydrological model using digital elevation data, Hydrol. Process., 9, 69-81, 1995. De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrologic modeling on a catchment scale using GIS and remote sensed land use information, In: Risk Analysis II, (ed), C.A. Brebbia, 295-304, WTI press, Southampton, Boston, 2000. Drogue G., El Idrissi, A., Pfister, L., Leviandier, T., Iffly, J.F. & Hoffmann, L., Calibration of a parsimonious rainfall-runoff model: a sensitivity analysis from 187

Chapter VIII local to regional scale, In: Rizzoli A.E. & Jakeman A.J. (eds.), Integrated Assessment and Decision Support, Proceedings of the First biennal meeting of the International Environmental Modeling and Software Society, Lugano (Switzerland), volume 1, 464-469, 2002. Eagleson, P.S., Climate, Soil, and Vegetation, a simplified model of soil moisture movement in liquid phase, Water Resour. Res., 14, 722-730, 1978. El Idrissi, A., Drogue, G., Pfister, L., Iffly, J.F., Hoffmann, L., Hingray, B. & Guex, F., Estimation of high floods by three rainfall-runoff models with short rainfallrunoff series (Alzette river basin, Luxembourg), In: Integrated Assessment and Decision Support, (eds), A.E. Rizzoli & A.J. Jakeman, volume 1, 470-475, Lugano, Switzerland, 2002. Flgel, W.A., Delineating hydrological response units by geographical information system analyses for regional hydrological modelling using PRMS/MMS in the drainage basin of the river BRL, Germany, In: Scale Issues in Hydrological Modelling, (eds.), J.D. Kalma & M. Sivapalan, 181-194, John Wiley & Sons Ltd, UK, 1995. Franchini, M., Wendling, J., Obled, Ch. & Todini, E., Physical interpretation and sensitivity analysis of the TOPMODEL, J. Hydol., 175, 293-338, 1996. Gupta, V. K. & Dawdy, D. R., Regional analysis of flood peaks: multi-scaling theory and its physical basis, In: Advances in Distributed Hydrology, (eds.), R. Rosso, A. Peano, I. Becchi & A. Bemporad, 147-168, Water Resour. Pub., Fort Collins, USA, 1994. Horritt, M.S. & Bates, P.D., Effects of spatial resolution on a raster based model of flood flow, J. Hydrol., 253, 239-249, 2001. Kavvas, M.L., Chen, Z.Q., Tan, L., Soong, S.T, Terakawa, A., Yoshitani, J. & Fukami, K., A regional-scale land surface parameterization based on areally-averaged hydrological conservation equations, Hydrol. Sci. J., 43(4), 611-631, 1998. Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. & Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modeling, J. Hydrol., 283, 91-106, 2003. Liu, Y.B., De Smedt, F. & Pfister, L., Flood prediction with the WetSpa model on catchment scale, In: Flood Defence 2002, (eds) Wu et al, 499-507, Science Press, New York Ltd., 2002.

188

Integrating GIS and hydrological process modelling in medium and large watersheds Moglen, G.E. & Hartman, G.L., Resolution effects on hydrologic modeling parameters and peak discharge, J. Hydrol. Eng., ASCE, 6(6), 490-497, 2001. Molnar, D.K. & Julien, P.Y., Grid-size effects on surface runoff modeling, J. Hydrol. Eng., ASCE, 5, 8-16, 2000. Nash, J.E. & Sutcliffe, J.V., River flow forecasting through conceptual models, Part 1: A discussion of principles, J. Hydrol., 10, 282-290, 1970. Pfister, L., Humbert, J. & Hoffmann, L., Recent trends in rainfall-runoff characteristics in the Alzette river basin, Luxembourg, Climate Change, 45, 323337, 2000. Quinn, P.F. & Beven, K.J., The prediction of hillslope flow parths for distributed hydrological modelling using digital terrain models, Hydrol. Process., 5, 59-79, 1991. Saulnier, G.M., Obled, C. & Beven, K.,Analytical compensation between DTM grid resolution and effective values of saturated hydraulic conductivity within the TOPMODEL framework, Hydrol. Process., 11(9), 1331-1346, 1997. Wang, Z., Batelaan, O. & De Smedt, F., A distributed model for Water and Energy Transfer between Soil, Plants and Atmosphere (WetSpa), Phys. Chem. Earth, 21, 189-193, 1997. Wittenberg, H. & Sivapalan, M., Watershed groundwater balance estimation using streamflow recession analysis and baseflow separation, J. Hydrol., 219, 20-33, 1999. Wolock, D.M. & Price, C.V., Effects of digital elevation model map scale and data resolution on topography-based watershed model, Water Resour. Res., 30, 30413052, 1994. Wood, E.F., Sivapalan, M., Beven, K. & Band, L., Effects of spatial variability and scale with implications to hydrologic modelling, J. Hydrol., 102, 29-47, 1988. Zhang, W. & Montgomery, D.R., Digital elevation model grid size, landscape representation, and hydrologic simulations, Water Resour. Res., 30, 1019-1028, 1994.

189

Chapter IX Summary and conclusions

1. General summary

The core of the research in this PhD study is to develop a GIS-based continuous hydrological model and to apply the model in the context of flood prediction and watershed management in a river basin, for instance, the assessment of land use change impacts on hydrological processes, the assessment of river restoration effects on the flood reduction, etc. The model simulates water and its movement coming from precipitation in the system of soil, plant and atmosphere within a GIS framework. The areas of major emphasis in this study are: (1) to develop a practical method of estimating runoff that relates the rainfall runoff relationship to basin characteristics, (2) to develop a diffusive transport approach for flow routing in GIS-based watershed modelling, (3) to develop efficient ArcView scripts and Fortran codes capable of identifying spatial model parameters and hydrological computation, (4) to understand the nature of hydrological interactions between soil, plant and atmosphere, (5) to develop an approach by integrating GIS and hydrological process modelling in medium and large watersheds, and (6) to demonstrate the applicability of the model by applying the model in 3 watersheds with different characteristics.

1.1.

Model development

A GIS-based distributed-parameter model, WetSpa Extension, has been developed to simulate runoff and flow responses in a river basin. The model conceptualizes a basin hydrological system being composed of atmosphere, canopy, root zone, transmission zone and saturation zone layers. The basin is divided into a number of grid cells, for which the water and energy balance is maintained. The hydrological system consists of four control stores: the interception storage of the plant canopy, the depression storage on the soil surface, the soil moisture content of the root zone, and the active groundwater storage of the saturated aquifer. The hydrological processes simulated by the model include precipitation, interception, snow accumulation and melt, depression,

Chapter IX infiltration, surface runoff, interflow, percolation, evapotranspiration, and the flow routing on hillslopes and in stream channels. A modified rational method is applied to estimate surface runoff and infiltration based on rainfall intensity, soil moisture status and cell characteristics, i.e. slope, soil type and land use. Interflow and percolation out of the root zone are simplified to be gravity driven flows and simulated by Darcys law and kinematic approximation. Actual evapotranspiration is limited by the potential evapotranspiration and calculated from the available water in the interception storage, depression storage, root-zone soil and groundwater storage. Flow routing is characterized by a two-parameter diffusive wave approximation approach, for which the travel time and its standard deviation are spatially distributed and calculated as a function of the cells roughness coefficient, hydraulic radius and slope gradient. For model simplification, groundwater flow is modelled using a reservoir method and on a small subcatchment scale. The model requires three base maps, i.e. DEM, soil type and land use, to identify spatial model parameters and process model simulations. The coverage maps of gauging site, stream network and catchment boundary are necessary in order to delineate drainage network more accurately and create Thiessen polygons. Site specific data input to the model include precipitation, temperature and PET, while discharge data is optional for model calibration. The model predicts flow hydrographs at the basin outlet or any convergent point inside the catchment, and the spatial distribution of hydrological characteristics at any time step over the catchment, such as surface runoff, interflow, groundwater recharge, soil moisture, and actual evapotranspiration, etc. All computational processes are implemented within a GIS ArcView framework. The model results on free flow travel time and its standard deviation within a river basin are very encouraging for handling overland and channel flow routing. As discussed in chapter III, one of the primary assumptions of the method employed in this research is a zero loss of water to storage during overland and channel flow. Although this is not a trivial assumption, it is important to make the model computationally efficient. Rather than calculating the inflow of water to and outflow of water from each cell in the basin and then routing the outflow from each cell based on its connectivity to its neighbouring cells, the two-parameter diffusive approach cuts the calculations down to the assessment of inflow of water into a cell from 192

Summary and conclusions precipitation and then travel time of the water to the outlet. Even if one only dealt with cells in the basin that have non-zero inflow in the calculations, the complete accounting and routing of water across a landscape is a computationally expensive undertaking. This new method is able to provide satisfactory estimates of the flood hydrographs and save computational cost greatly in a GIS-based hydrological modelling system. The raster-based approach as used in the WetSpa model is often simpler than other finite difference and finite element approaches, which aim to discretize and solve the equations governing fluid flow and energy dynamics, and thus may present a lesser computational burden and development cost. Despite its crude representation of hydrological processes, the model has shown to give good results in different applications. Moreover, the raster-based model can benefit from GIS technology greatly extracting hydrological features from the DEM and other digital maps. Once the DEM, land-use and soil maps are prepared, model parameters in the form of raster GIS maps are derived for each cell. In addition, the raster schematisation of the basin enables one to consider spatial characteristics at a small scale with reasonable accuracy. The fact that model parameters are derived from known basin characteristics makes the model suitable for analysis of the effects of changes of distributed model parameters, such as land-use changes. In order to reduce the model calibration effort, an automated calibration procedure is applied by incorporating a model-independent parameter estimator. This approach is used in this study to estimate the most sensitive parameters of the WetSpa model with observed flow hydrographs at selected stations as the calibration targets. The best set of parameters is selected from within reasonable ranges by adjusting the values until the discrepancies between observed and simulated hydrographs is reduced to a minimum in the weighted least squares sense. This scheme serves as an optimization algorithm to estimate the model parameters. A further manual calibration is necessary to avoid the ill-posed problems. In addition to the evaluation based on a visual comparison between calculated and observed hydrographs, five evaluation criteria are selected for assessing the model performance including model bias, model determination coefficient, the Nash--Sutcliffe efficiency, the logarithmic version of Nash-Sutcliffe efficiency for low flow evaluation, and the adapted version of Nash193

Chapter IX Sutcliffe efficiency for high flow evaluation. These statistical measures provide quantitative estimates for the goodness of fit between observed and predicted values, and are used as indicators of model performance. In order to identify the input parameters that had the biggest impact on the WetSpa model, a series of sensitivity analyses are performed. The potential runoff coefficient, soil hydraulic conductivity, interflow scaling factor, soil moisture status, and the runoff exponent have the biggest influence on the flood volume. The channel and overland flow roughness, hydraulic radius, channel flow threshold, and rainfall excess intensity have the biggest influence on the prediction of peak flow and the time to the peak. The size of the grid cell may also have a great influence on derived catchment characteristics, runoff and flow responses as described in chapter VIII. Generally, the estimated surface runoff, interflow, total runoff, peak discharge and its lag time decreases consistently with the increasing grid size, while the amount of infiltration, percolation, groundwater flow, and soil moisture are increasing on the contrary. Therefore, significant bias in the modelling results can be generated when using a coarse resolution data. These problems are typically encountered when modelling a large catchment. In this research, a practical approach by dividing a large catchment into very small subwatersheds based on high-resolution digital graphic data is developed, which allows for the internal drainage structure of WetSpa for each small subwatershed. Encouraging results have been achieved through the model application in the Alzette River basin located in the Grand-duchy of Luxembourg.

1.2.

Model applications

The WetSpa distributed hydrological model was initially developed and tested on a small catchment, Barebeek, in Belgium for flood prediction and hydrological design using long-term historical precipitation records and design storms. The study catchment is very complex and highly urbanized with the Brussels international airport located in the upper area of the catchment, four main traffic lines crossing the watershed in different directions, and many country roads crisscrossing the area from one village to another. Besides, a main canal passes through the area in the north, and a small lake is situated near the watershed outlet. The residential areas with sewer systems exist in the watershed occupying about 28% of the total area. Simulation 194

Summary and conclusions results show that the model is capable of handling hydrological processes in a complex terrain as for the case in the Barebeek catchment. The flood volume and its peak discharge are especially well reproduced. In addition, GIS provides a powerful tool for developing the model, calibrating parameters and displaying model results in a spatial way, and the local complexities of the watershed can be well captured. Next, the WetSpa model on hourly time scale was applied to the Alzette River basin, located in the Grand-duchy of Luxembourg, to predict floods and assess the impacts of land use change on flood behaviours. The study catchment has generally a rolling topography, and is highly developed with several towns inside the catchment such as Ettelbruck, Mersch, Luxembourg-city, and Esch/Alzette. Urban areas occupy 11.2% of the total area and have a significant effect on runoff and flow responses of the river basin. Good performance in predicting flow hydrographs and the spatial distribution of hydrological variables over the catchment has been obtained by using the fully distributed and the semi-distributed WetSpa model. The assessment of land use impacts on flooding was performed in the Steinsel sub-basin situated in the upstream area of the Alzette basin. Model simulation shows that (1) the runoff from urban areas is dominant for a flood event compared to runoff from other land use areas in this catchment, and its partition tends to increase for small floods and for the flood events with low antecedent soil moisture, (2) other runoff contributions tend to increase for large storms and for storm events with high antecedent soil moisture, (3) interflow and baseflow from natural areas are important during wet season but not for small floods during the dry season, and (4) the flow coefficient and the runoff partitions from different land use classes vary from one storm event to the other due to the difference in soil moisture and storm behaviours. For assessing the hydrological effects of land use changes on floods, three hypothetical scenarios, namely urbanization, deforestation and afforestation scenario, were considered based on the present land use configurations in the Steinsel sub-basin. The result of model simulation shows that that the urbanization scenario has a large impact on increasing peak discharge and flood volume, as well as time to the peak compared to the present land-use. The result from the simulation of deforestation scenario shows the same trend as the result from the urbanisation scenario but with reduced impact than the urbanisation scenario. Afforestation is found to increase the 195

Chapter IX infiltration water and evapotranspiration while reduces the peak discharge and the total runoff. Urbanisation with high proportion of impervious areas change the water balance of the river basin towards increased surface runoff while forest areas by allowing more water to pass through the soil surface change the water balance into increased infiltration and subsequently increasing the groundwater recharge. Investigation of the peak flow shows that land-use changes can have remarkable effects on peak discharges in comparison to the present land use condition. The urbanisation and deforestation scenarios may increase the peak discharges by 26% and 9.1% in average, while afforestation has a positive impact, decreasing the peak flow by 5.3% in average. A conceptual flood reduction methodology by natural river restoration for the headwater streams is evaluated for the Steinsel sub-basin, Alzette, the Grand-duchy of Luxembourg. The streams are firstly classified into 5 orders, and the response of stream channels to the resistance and the river re-meandering in the first and secondorder streams are assessed. These streams have deeper channel slopes, strongly influenced by morphology, soils, and vegetation of their channels than downstream higher order streams. Under natural conditions, these streams may have dense instream vegetation and low fluvial power, which eventually increase the flow resistance. Accordingly, the roughness coefficient of these channels is increased for simulating the effect on downstream flooding. The hydrological simulation indicates that the reduction of peak discharge is as much as 14% on average compared to the present situation, and the time delay of flood peak can be as much as 2 hours. A reduction in the discharge occurs during the rising limb of the flood hydrograph, while the discharge increases during the falling limb of the hydrograph, which results in longer sustained flows than in the present conditions. However, while large floods in the main stream channels are reduced or avoided, a local flooding may occur in the headwater stream areas due to river restoration. Finally, an application of the WetSpa model on daily scale based on GIS and remote sensed land use information is performed for a Carpathian catchment, Margecany, in the upstream part of the Hornad River basin, Slovak Republic. The study catchment is characterised by mountainous terrain. Altitudinal variation is an important determinant of the local climate and soil characteristics in all the state. Snowmelt is an 196

Summary and conclusions important process in the river basin and forms the main source of the spring floods. Model simulations indicate that both spring and summer floods are well reproduced, and the model is suitable to be used in a mountainous catchment situated in the temperate region. An analysis of the temporal and spatial distribution of temperature, PET and precipitation over the catchment is implemented during the case study, which illustrates the importance of capturing the spatial pattern of the input variables in hydrological modelling. It is also concluded from the calibration and validation processes that the model is strongly sensitive to the parameters describing the runoff generation and the routing of water in surface, subsurface and groundwater reservoirs.

1.3.

Model limitations

Hydrological modelling is an attempt to simulate real hydrological processes through the use of input data describing physical characteristics of the system, a set of algorithms to transform input data to output of interest, and simplifying assumptions to limit the scope of the model. Therefore, model limitations must be considered in running the model and interpreting its output. Followings are major limitations associated with the WetSpa model simulation. 1) WetSpa is a continuous simulation model. Therefore, the check of data continuity and reliability must be carried out in the phase of data preparation. If missing data exist in the precipitation or PET series, logical interpolation must be made for the time period. 2) The land use categories are grouped, for which some of the categories might be somewhat ambiguous. For instance, the category agriculture may include farmsteads, lawns, disturbed areas, and other land uses that are not identifiable as one of the other specified land use categories. Furthermore, the annual crop rotation is not taken into account in the model. In addition, lower level highways and country roads are not modelled uniquely, but are combined within the rural residential category. This may reduce the amount of runoff and alter the flow direction expected from these areas. 3) Values assigned to any raster or grid cell represents an average value over the area of each cell. The greater the variability over the cell, the greater will be the error 197

Chapter IX induced through the use of an average value. Therefore, the grid size should be well defined. A small grid size may better represent the variability of physical watershed characteristics, but leads to more memory cost and time consumption during model simulation, particularly when modelling a large watershed. Balance should be made between the model accuracy and computer efficiency. 4) The time resolution should be well defined. As for instance, it is not feasible to predict flood using hourly or daily scale for a very small watershed, since excess water may flow out within the first time step without any flood wave attenuation. In this case, a shorter time interval should be chosen if field measurements are available. 5) The snow accumulation and snowmelt are modelled in a simple way by the degree-day coefficient method with a constant degree-day factor. Moreover, the redistribution of snow pack and the effects of aspect, local slope, and land use, etc., on snowmelt are not taken into account. These may reduce the reliability of model performance in modelling snow processes. 6) WetSpa model generates runoff by an empirical-based modified coefficient method rather than from equations more closely representing physical processes. Therefore, calibration is necessary when the model is applied in a different environment. Though definitely a limitation, the use of the method has its advantages of closely linking runoff with rainfall intensity and cell characteristics such as slope, land use, soil type and moisture content, and thus has a great potential to predict the impact of land use change on hydrological behaviours in the watershed. 7) The impervious fractions for urban areas are set subjectively depending upon cell size due to the lack of detailed measurements. For instance, for a 5050 m grid, 30% is set for residential area, 70% for commercial and industrial area and 100% for major communication lines, parking lots, etc. This may not actually reflect the real world and increase uncertainties to the model result. 8) WetSpa model employs many default parameters, which are interpolated from the literature and used over the entire catchment. Due to the vast variation range, parameters such as hydraulic conductivity, roughness coefficient, etc. may change 198

Summary and conclusions greatly when applying the model to another place with quite different environment. This brings difficulties in model parameterization in an un-gauged river basin. 9) WetSpa model assumes that the groundwater table is below the root zone, which constrains the use of the model in wetland areas where groundwater table is close to the land surface. Moreover, the model simulates groundwater flow on small subcatchment scale. It estimates the groundwater flow and groundwater storage for each small subcatchment at each time step, but cannot predict the spatial distribution of groundwater table, as well as its variation during the simulation period.

2.

Future perspectives

There are many directions to improve the WetSpa model in further research. One of the most important aspects is to complete a detailed quantitative sensitivity analysis and uncertainty assessment of the model, in order to examine the relative contribution of the model parameters, initial conditions, and input meteorological variables to the models overall predictive uncertainty. In applications of this modelling system, there will be cost restraints on the collection and preparation of necessary geophysical and meteorological input data. It is essential to know which of the inputs are most important and what spatial and temporal resolution are required to the generation of accurate results. Another important area of future research is to study the spatial characteristics of global model parameters used in the model, so as to create all model parameters in a spatial way in relation with terrain features, and enable the model to be used in un-gauged river basins without model optimization. Other possible future researches on the WetSpa model can be as following: 1) Automated calibration of the most important spatial model parameters, which can be implemented by coupling WetSpa model with PEST, a model-independent nonlinear parameter estimator as described in chapter VII. The proposed scheme is to multiply the spatial parameters by a factor and run the model as many times as it needs to adjust selected parameters within their predetermined range until the discrepancies between model outputs and a complementary set of flow observations is reduced to a minimum in the weighted least squares sense.

199

Chapter IX 2) Development of a practical method to account for the joint effect of altitude, slope, aspect, general circulation of the atmosphere, etc., on the spatial distribution of precipitation, temperature and PET. This may highly increase the reliability of model inputs and decrease the uncertainty of model outputs, especially for modelling in a large mountainous catchment. The radar information may also be coupled in the WetSpa model to estimate the spatial distribution of rainfall at each time step. 3) Incorporation of variable travel time into flow routing schemes, for which the flow velocity is estimated as time variant variable depending upon the channel geometry and runoff volumes. This may overcome the shortage that flow velocity is assumed time invariant for a flood event in the current modelling approach. 4) Coupling an intelligent interflow routing scheme into the WetSpa model, for which the interflow out of one cell is routed to the downhill neighbouring cells depending on their elevation difference. The scheme of multiple flow path division may give a more realistic prediction of soil wetness distribution and be used to predict the saturated runoff source areas in regions with convergent topographies. 5) Improvement of the simple snowmelt model used in the WetSpa model taking account the variability of degree-day constant, the effect of radiation on snowmelt, snow drift and deposition in steep terrain, and so on. This will make the snowmelt model more realistic enabling the simulation of snow cover and melting runoff more accurately over extensive and heterogeneous landscapes. 6) Incorporating the influence of lake and reservoir operation in the WetSpa model by combining efficient hydraulic models for engineering purpose. This will make the modelling system more flexible for flow simulation of large river basins with lakes and reservoirs involved. 7) Updating the current WetSpa into a real fully distributed model by combining with a distributed groundwater model. Groundwater balance is then calculated on grid cell basis allowing the estimation of groundwater table fluctuation and the simulation of saturation overland flow once the water table reaches the ground surface.

200

Summary and conclusions 8) Application of the WetSpa model to study the soil erosion and deposition patterns allowing to keep a physically meaningful control on the effects of different land management scenarios on landscape-scale processes, for which the spatial parameters related to soil erosion and sedimentation will be generated. 9) Application of the WetSpa model to study the pollutant loading and transport in the surface water and ground water system for the point and non-point source contaminations, for which a range of chemical, biological and physical parameters related to contaminant generation and transport will be generated. In any of the above future perspectives, there will be a significant increase of model parameters to be estimated and consequently more complex model identifications have to be performed. This will make the model more and more complicated and difficult to be accomplished by untrained users. However, parallel extensions can be built according to the purpose of the project and focusing on specific directions. Generally, this research has laid a foundation for a GIS-based distributed hydrological modelling system for the prediction of flood and the simulation of water balance on catchment scale. Although the simulation results from the model indicate that additional works are necessary to improve model structure and model parameters, the WetSpa model at its current state of development provides a substantial framework on which further researches can be conducted.

201

Appendix A WetSpa Extension: A GIS-based hydrological model for flood prediction and watershed management Documentation and User Manual
Abstract A GIS-based distributed watershed model, WetSpa Extension, has been under development suitable for use of flood prediction and watershed management on catchment scale. The model is physically based and simulates hydrological processes of precipitation, snowmelt, interception, depression, surface runoff, infiltration, evapotranspiration, percolation, interflow, groundwater flow, etc. continuously both in time and space, for which the water and energy balance are maintained on each raster cell. Surface runoff is produced using a modified coefficient method based on the cell characteristics of slope, land use, and soil type, and allowed to vary with soil moisture, rainfall intensity and storm duration. Interflow is computed based on the Darcys law and the kinematic approximation as a function of the effective hydraulic conductivity and the hydraulic gradient, while groundwater flow is estimated with a linear reservoir method on a small subcatchment scale as a function of groundwater storage and a recession coefficient. Special emphasis is given to the overland flow and channel flow routing using the method of linear diffusive wave approximation, which is capable to predict flow discharge at any converging point downstream by a unit response function. The model accounts for spatially distributed hydrological and geophysical characteristics of the catchment and therefore is suitable for studying the impact of land use change on the hydrological behaviours of a river basin.

1.

Model description

Recent development of GIS and remote sensing technology makes it possible to capture and manage a vast amount of spatially distributed hydrological parameters and variables. Linking GIS and the distributed hydrological model is of rapidly increasing importance in studying the impact of human activity on hydrological

Appendix A behaviours in a river basin. Ideally, watershed models should capture the essence of the physical controls of topography, soil and land use on runoff production as well as the water and energy balance. Distributed parameter hydrological models are typically structured in characterizing watershed conditions such as topography, soil type, land use, drainage density, degree of soil saturation, and rainfall properties, for which it is advantageous to use the data currently available in GIS format. This report describes such a model, called WetSpa Extension.

1.1.

Model construction

1.1.1. Model objectives WetSpa is a physically based and distributed hydrological model for predicting the Water and Energy Transfer between Soil, Plants and Atmosphere on regional or basin scale proposed by Wang et al. (1997) and Batelaan et al. (1996). The model conceptualizes a basin hydrological system being composed of atmosphere, canopy, root zone, transmission zone and saturation zone. The basin is divided into a number of grid cells in order to deal with the heterogeneity. Each cell is further divided into a bare soil and vegetated part, for which the water and energy balance are maintained. Water movement in the soil is simplified as one-dimensional vertical flow, including surface infiltration, percolation and capillary rise in the unsaturated zone and recharge to groundwater. The model was designed to simulate the Hortonian overland flow and the variable source area concept of runoff generation with time resolution of minutes. Due to the complexity of the model structure and data limit, the model is difficult to be used for an engineering purpose. Therefore, a WetSpa extension is developed with a flexible data acquisition and higher computational efficiency. The main objectives of the WetSpa Extension are: 1) To provide a comprehensive GIS-based tool for flood prediction and watershed management on catchment scale, which is compatible with GIS technology and remote sensing information. 2) To enable the use of the model for simulation of the spatial distribution of hydrological processes, such as runoff, soil moisture, groundwater recharge, etc.

204

WetSpa Extension: Documentation and User Manual 3) To enable the use of the model for analysis of land use change and climate change impacts on hydrological processes. 4) To provide for a distributed model that can operate on cell scale and a variable time step, and a semi-distributed model on small subwatershed scale. 5) To provide a platform on which the future water quality and soil erosion models can be developed at multiple scales.

1.1.2. Model structure The model uses multiple layers to represent the water and energy balance for each grid cell, taking into account the processes of precipitation, interception, snowmelt, depression, infiltration, evapotranspiration, percolation, surface runoff, interflow and groundwater flow. The simulated hydrological system consists of four control volumes: the plant canopy, the soil surface, the root zone, and the saturated groundwater aquifer. The precipitation that falls from the atmosphere before it reaches the ground surface is abstracted by canopy interception storage. The remaining rainfall reached to the ground is separated into two parts depending on the land cover, soil type, slope, the magnitude of rainfall, and the antecedent moisture content of the soil. The first component fills the depression storage at the initial stage and runs off the land surface simultaneously, while the remaining part infiltrates into the soil. The infiltrated part of the rainfall may stay as soil moisture in the root zone, move laterally as interflow or percolate further as groundwater recharge depending on the moisture content of the soil. Drainage water from a given cell flows laterally depending on the amount of groundwater storage and the recession coefficient. The percolation out of the soil layer is assumed to recharge the groundwater storage. Interflow from the root zone is assumed to contribute overland flow and routed to the watershed outlet together with surface runoff. The total runoff from each pixel cell constitutes the sum of the surface runoff, the interflow and the groundwater flow. Evaporation takes place from intercepted water, depressed water and the soil surface, while transpiration takes place from the plant through root system in the soil layer, and a small part from the groundwater storage. The water balance for the interception storage includes precipitation, evaporation and through fall. The water balance for the depression storage includes through fall, infiltration, evaporation and surface runoff. The water balance for the soil column includes infiltration, evapotranspiration, percolation, and 205

Appendix A lateral subsurface runoff. The water balance for the groundwater storage includes groundwater recharge, deep evapotranspiration, and lateral groundwater flow. Figure A-1.1 shows schematically the model structure at a pixel cell level.
Evapotranspiration Interception Precipitation

CANOPY

Through fall

Depression

SOIL SURFACE

Surface runoff

Infiltration
SOIL

Interflow

Recharge
GROUNDWATER

D I S C H A R G E

Drainage

Figure A-1.1: Model structure of WetSpa Extension at a pixel cell level The simple structure in Figure A-1.1 is used in the model because the emphasis here is on developing and testing parameterizations for the root zone. Excess runoff, infiltration, evapotranspiration, interflow and percolation estimates are point calculations. Different slope, land use and soil properties in different grid cells of a watershed result in different amounts of excess runoff when subjected to the same amount of rainfall. The routing of runoff from different cells to the watershed outlet depends on flow velocity and wave damping coefficient using the method of diffusive wave approximation. Although the spatial variability of land use, soil and topographic properties in a watershed are considered in this model, the groundwater response is modelled on small subcatchment scale for the convenience of model parameterization and model simulation. Two alternatives for determining groundwater flow are used in the model, simulating groundwater flow with a simple linear reservoir method and non-linear reservoir method. All model equations are specifically chosen to maintain a physical basis and well supported by previous studies. The inputs to the model are precipitation and potential evapotranspiration (PET). Temperature data are needed if snow accumulation and melt occur during the simulation period. The digital maps of topography, land use and soil type are used to derive all necessary spatial distributed 206

WetSpa Extension: Documentation and User Manual model parameters. The main outputs of the model are river flow hydrographs and spatially distributed hydrological characteristics, such as soil moisture, infiltration rates, groundwater recharge, surface water retention or runoff, etc.

1.1.3. Model assumptions 1) Soil characteristics and landscape are isotropic and homogeneous for a single raster cell. 2) Canopy cover and ground cover are homogeneous for a single raster cell. 3) Precipitation is spatially homogeneous within a raster cell. 4) The form of Hortonian overland flow is valid for most of the areas. 5) Evapotranspiration is neglected during a rainstorm and when the soil moisture is lower than residual soil moisture. 6) Deep evapotranspiration takes place when soil is dry, and is restricted by the amount of effective groundwater storage. 7) Soil moisture content is homogeneous in a single cell, while the groundwater storage is uniformly distributed on small subcatchment scale at each time step. 8) Water flows along its pathway from one cell to another, and cannot be partitioned to more than one adjacent raster cell. 9) The method of linear diffusive wave approximation is valid for routing of both overland flow and channel flow. 10) Hydraulic radius is location dependent, varies with flood frequency, but remains constant over a flood event. 11) Interflow occurs when soil moisture content is higher than field capacity and can be estimated by Darcys law and kinematic approximation. 12) The water losses from overland and channel flow, as well as the water losses from deep percolation are not important.

1.2.

Data preparation

The preparation of the database for WetSpa Extension to a specific watershed implies the determination of the complete drainage structure of the watershed, the spatial distribution of land use classes and soil types, as well as the collection of available hydro-meteorological data related to the project. 207

Appendix A

1.2.1. Digital data The model uses geo-spatially referenced data as input for deriving model parameters, which includes most data types supported by ArcView, such as coverage, shape file, grid and ASCII file. Image can be used for reference within a view, but is not used directly by the model. The digital maps of topography, land use and soil type are 3 base maps used in the model, while other digital data are optional depending upon the data available and the purpose and accuracy requirement of the project. 1) Digital Elevation Model (DEM) The raster-type DEM, generated from point or contour topographic map, is preferred in order to be compatible with other remotely sensed data. The spatial and elevation resolutions should be fine enough to capture the essential information allowing taking care of the effects of spatial variability of the watershed characteristics on its hydrological response. In practice, the chosen resolution must allow adequate representation of the actual topography and accurate determination of the watershed area, its river network, and its subwatersheds. In the absence of large water surfaces (lakes, reservoirs, ponds, etc.) and large plains with little or no elevation variation, processing of the DEM is relatively straightforward. After filtering of the initial data to detect and remove erroneous extreme values, the slope, aspect, flow direction, flow length and flow accumulation of each grid cell are determined. Over flat areas, no slope and, hence, no direction can be computed. Also, the DEM may contain artificial pits from which no water can flow out. These specific problems have to be reserved by modifying elevations artificially to lead to flow directions as accurate as possible on any of the cells. Next, the identification of river network is performed by assuming that all cells draining more than a specified upstream area are part of that network. More or less detailed river networks can be identified, depending on the selected upstream threshold area. Finally, the stream links, stream orders and the subwatersheds corresponding to these river reaches are identified. 2) Land use and soil type Land use information is an important input to the WetSpa Extension, which is normally obtained from high-resolution remotely sensed data for the same area as the DEM, and with the same grid cell size. For hydrological simulation purpose, all land 208

WetSpa Extension: Documentation and User Manual use classes initially determined are grouped together into 14 WetSpa classes significantly distinguished from each other on the basis of their effect on hydrological processes, namely crop, short grass, evergreen needle leaf tree, deciduous needle leaf tree, deciduous broad leaf tree, evergreen broad leaf tree, tall grass, irrigated crop, bog marsh, evergreen shrub, deciduous shrub, bare soil, impervious area and open water surface. Each of these classes is characterized by quantitative attributes. The groups may vary according to the algorithms used in the model. For instance, only 5 classes are considered in defining potential runoff coefficient and depression storage capacity, i.e. crop, grass, forest, bare soil and urban areas. For simulation purpose, the percentage of bare soil and impervious areas are estimated for each grid cell based on the high-resolution land use map. Soil types of the catchment are obtained from the soil information furnished by soil maps. The soil code system used in WetSpa Extension is based on the soil texture triangle developed by the United States Department of Agriculture (USDA), which is characterized by its percentage of clay, silt and sand, ranging from the fine textures (clay), through the intermediate textures (loam), and the coarser textures (sand). Therefore, the original soil coverage map has to be converted to a raster map with WetSpa soil codes in the phase of data preparation. The grid must be adjusted to the same grid structure as the DEM and limited to the same area by using the mask grid of the catchment. The reclassification can be done within GIS framework, which makes use of a reclassification table prepared in ArcInfo GIS or ArcView Spatial Analyst. This work must be done with caution in order to make the conversion as accurate as possible. The soil properties and hydraulic characteristics of those soil types are considered constant in the present version of the model. Default values are interpolated from literature as described in section A-3.1, but users can substitute any other more appropriate values for them. 3) Optional digital data Other optional digital data that can be used in the model include point coverage or shape file of gauging station locations, line coverage or shape files of stream network and major traffic lines, polygon coverage or shape files of boundary and sewer systems, etc. These data are of great help in delineating watershed drainage path network, estimating spatial rainfall distribution, as well as properly determining 209

Appendix A model routing parameters. If two or more rain gauges exist in or around the catchment with measured data, the Thiessen polygon weighting method is then introduced to calculate the rainfall distribution, for which the weighting factors are computed for each grid cell and subwatershed. Otherwise, a uniform rainfall distribution is assumed over the catchment. The internal streamflow gauges can be used in the watershed discretization process, for which the watershed is split at those locations where gauges are present. This makes it possible to compare measured and computed flow hydrographs at a point or series points. The coverage of official river network and catchment boundary is a very important geo-referenced data, which can be combined within GIS in delineating watershed drainage network, particularly for meandering rivers in flat areas. Usually, from the topographic information present in a DEM, it is quite difficult to represent watershed boundary and meandering rivers in plain areas. To account those cases, data coming from the hydrographical layer of digital maps (boundary, rivers, lakes, ponds, etc.) are used in combination with the DEM to identify drainage areas, find input and output cells for water bodies, and make any necessary corrections to flow directions in order to have the river reaches flow where they should and to be able to estimate the flow length closer to reality. For hydrological modelling in a complex terrain, such as an urban or suburban watershed, the sewer systems, communication lines, artificial channels, etc. are important elements in drainage structure configuration, and govern flow direction more strongly than the derived aspect at a local scale. Since most of these barriers are not sufficient to be represented in a DEM, additional procedures in term of deriving more realistic flow direction map are performed using GIS overlaying technique in the model, where the general flow direction map is overlaid by the sewer flow direction map, the communication line flow direction map and the river flow direction map subsequently. This allows water draining from the sewer areas at its outlet and water crossing communication lines at the concave points to join the river. The altered flow direction map is then used for further drainage structure delineation.

1.2.2. Hydro-meteorological data The basic input requirements for the WetSpa Extension consist of model parameters, initial conditions, meteorological data and streamflow data for model calibration and 210

WetSpa Extension: Documentation and User Manual validation. The basic meteorological data requirements are rainfall and PET. Temperature data are optional used for simulation of snowmelt. In the case of calculating PET by the Penman-Monteith equation, additional meteorological data are required, including air temperature, radiation, relative humidity and wind speed. In this section, the meteorological and hydrological data are described. The model parameters and initial conditions are described in the subsequent sections. 1) Rainfall Rainfall is the fundamental driving force and pulsar input behind most hydrological processes. Rainfall-runoff models are particularly sensitive to the rainfall input and any errors in estimates are amplified in streamflow simulations. The input rainfall series must be in the same interval as the model running step. For instance, hourly rainfall data are required for each raingauge when modelling in an hourly scale. In many cases, rainfall data at certain stations are in a daily scale rather than an hourly scale. These data can be used by disaggregation according to the temporal structure of rainfall of the neighbouring hourly reference raingauges. The Thiessen polygon method is then used to estimate areal rainfall during model simulation. Depending upon the objective of the study and on the time scale of the catchment response, the time resolution of rainfall input can be enlarged to a daily scale or reduced to a finer resolution corresponding to the model time scale. The rainfall data are treated as accumulated totals so that the rainfall associated with any particular time is the rainfall volume since the end of last time step. 2) Potential evapotranspiration WetSpa Extension requires PET data as one of the inputs with the same time interval as rainfall series, which can be obtained from field measurement or estimated by physical or empirical equations. Normally, daily values of PET are sufficient, for which the value is either averaged to an hourly value or disaggregated with a simple empirical equation as a function of hour as described in section A-2.7. If only one measuring station is available, the PET data can be uniformly applied to the whole study area for a small catchment. Otherwise, the value should be revised for different virtual stations according to the local meteorological and geophysical conditions, especially when modelling in mountainous areas. The areal PET is estimated using the Thiessen polygon method. The evapotranspiration data are treated as accumulated 211

Appendix A totals so that the evapotranspiration associated with any particular time is the evapotranspiration volume since the end of last time step. 3) Discharge For the purpose of model calibration and evaluation, observed discharge data at the basin outlet with the same time interval as the precipitation series are required for visual comparison and statistical analysis. The discharge data at internal gauging sites are optional for model verification. Data conversion to another time scale is necessary according to the simulation time step. The discharge data at any particular time is the average discharge since the end of last time step. 4) Optional meteorological data Temperature data are required when snow accumulation and snowmelt occur in the catchment. Normally, daily average temperature data are sufficient in simulating snow processes. Anyhow, the temperature data should keep the same time interval as the precipitation series. If the Penman-Monteith equation is chosen to calculate the PET, when there is no measured data available in the study area, the data of air temperature, short wave radiation, relative humidity, and wind speed are required in the model, which can be obtained from the routine meteorological stations.

2.

Model formulation

WetSpa Extension is a distributed, continuous, physically based model describing the processes of precipitation, runoff and evapotranspiration for both simple and complex terrain. It is a distributed model because the watershed and channel network are represented by a grid of mesh. Each cell is described by its unique parameters, initial conditions, and precipitation inputs. It is continuous model because it has components describing evapotranspiration and soil water movement between storms, and therefore can maintain water and energy balance between storms. It is physically based because the mathematical models used to describe the components are based on such physical principles as conservation of mass and momentum. In this section, a brief description about the model formulation involved in the processes of interception, snowmelt, depression, infiltration, surface runoff, evapotranspiration, percolation, interflow and groundwater flow is presented. 212

WetSpa Extension: Documentation and User Manual

2.1.

Precipitation

Rainfall is a fundamental component of any hydrological models. To obtain information at a specific location in a catchment, either interpolation or extrapolation of the existing data is required. The spatial distribution of rainfall is often estimated by the elementary techniques from a set of fixed rainfall gauges, while the temporal distribution is ignored by averaging the rainfall over a predetermined period. The crudest method for estimating the precipitation over a region is to plot contours of equal precipitation with the assistance of a structured grid. The average precipitation is computed between successive isohyets. This method is difficult to realize for each modelling time step with sparse precipitation data, although the task of plotting isohyets is automated with the advance of GIS technology. A common interpolation approach is the Thiessen polygons, which is also the method used in the current version. In this approach, areas closest to a rainfall gauge adopt the rainfall recorded at that gauge. This results in constant rainfall regions with discontinuities between regions. In addition, there is no justification in assuming that point rainfall measurements provide reliable estimates of precipitation in the surrounding region. The inverse distance weighted method is an alternative approach, for which the rainfall at any desired location is interpolated from the given data using weights that are based on the distance from each rainfall gauge and the desired location. This approach produces a smooth rainfall distribution along with the undesirable troughs and peaks located at the rainfall gauges. However, interpolation is difficult with the inverse distance weighted method for higher dimensional data sets. A special case is the precipitation interpolation in high mountainous areas with few point measurement. In order to account for the large variations in precipitation with altitude, the reference series can be adjusted for each grid cell or subcatchment by the method of lapse rate corrections in which the precipitation is assumed to vary linearly with the elevation (Dingman et al., 1988):
Pi = Pref + Pref (H i H ref )

(2.1)

213

Appendix A where Pi is the precipitation at cell i (mm), Pref is the precipitation at the reference precipitation station (mm), Hi and Href are the height at cell i and at the reference station, and is the precipitation lapse rate. Calibration is necessary in order to get a proper precipitation lapse value. The last two methods are optional in addition to the Thiessen polygon method.

2.2

Interception

Interception is that portion of the precipitation, which is stored or collected by vegetal cover and subsequently evaporated. In studies of major storm events, the interception loss is generally neglected. However, it can be a considerable influencing factor for small or medium storms and water balance computations would be significantly in error if evaporative losses of intercepted precipitation were not included. 1) Mass balance of the interception storage Interception is a complicated process, which is affected by the storm characteristics, the species of vegetation, percentage of canopy cover, growth stage, season, and wind speed, etc. Interception loss is higher during the initial phase of a storm and approaches zero thereafter. In WetSpa Extension, the rainfall rate is reduced until the interception storage capacity is reached. If the total rainfall during the first time increment is greater than the interception storage capacity, the rainfall rate is reduced by the capacity. Otherwise, all rainfall is intercepted in the canopy, and the remainder of interception is removed from the rainfall in the following time increments. The equation can be expressed as:
I i , 0 SI i (t 1) I i (t ) = P (t ) i for Pi (t ) f I i , 0 SI i (t 1) for Pi (t ) I i , 0 SI i (t 1)

(2.2)

where Ii(t) is the interception loss at cell i over the time interval (mm), Ii,0 is the cell interception storage capacity (mm), SIi(t-1) is the cell interception storage at time step t-1 (mm), and Pi(t) is the cell precipitation amount (mm). The mass balance of interception storage at a pixel cell is computed as:

214

WetSpa Extension: Documentation and User Manual SI i (t ) = SI i (t 1) + I i (t ) EI i (t ) (2.3)

where SIi(t-1) and SIi(t) are cell interception storage at time step t-1 and t (mm), EIi(t) is the cell evaporation from interception storage (mm). EIi(t) = 0 when interception storage is zero, or during the storm event; EIi(t) = SIi(t-1) under the condition of Pi(t) = 0 and EP > SIi(t-1) > 0, in which EP is the potential evaporation (mm); and EIi(t) = EP for the rest conditions. 2) Seasonal variation of interception storage capacity Interception storage capacity is a function of leaf area index and vegetal species. Evidently, it varies with season in template regions. Typical values can be found in the literature (Horton, 1919; Clark, 1940; Lull, 1964; Simons, 1981; Rowe, 1983). Through physical analysis and interpolations, a lookup table of maximum and minimum interception storage capacity corresponding to summer and winter extremes for different vegetation types are established. Specifically, the interception storage capacity of crop is set to 0.8 mm during growing season and null for the rest. For wetting losses on impervious areas, the adsorption storage capacity is set to 0.5 mm (Bauwens et al., 1996). Since the interception storage capacity varies continuously with time, a simple sine-shaped variation curve is proposed for the convenience of model programming. The empirical equation is similar as that of estimating daily potential evaporation based on statistical analysis of long-term measurements (De Smedt, D., 1997), and is written as:

1 1 d 87 I i , 0 = I i , min + (I i , max I i , min ) + sin 2 365 2 2

(2.4)

in which Ii,min is the minimum interception storage capacity at cell i (mm), Ii,max is the maximum interception storage capacity (mm), and d is the day of the year. The exponent b controls the shape of the variation curve, and can be adjusted according to the local conditions. Hourly interception storage capacity is assumed to be constant during a day in the model. Therefore, the interception storage capacity is only a function of the date. Figure A-2.1 gives a graphical presentation for the annual variation of grass interception storage capacity, for which the minimum and

215

Appendix A maximum interception capacity is 0.5 and 2.0 mm respectively, and the exponent b is set to 1.35.

2.5

Interception storage capacity (mm)

2.0 1.5 1.0 0.5 0.0

1/1 1 1/2 1 1/3 3 1/4 4 1/5 5 1/6 6 1/7 7 1/8 8

b = 1.35

1/9 1/11 12 1/12 9 1/10 10 11

Date (d/m)

Figure 2.1: Annual variation of grass interception storage capacity By substituting Eq. (2.2) to Eq. (2.3), the interception loss and interception storage at each time step can be estimated. No interception loss exists when the interception storage capacity is achieved, and all precipitation reaches ground surface. The intercepted water in canopy loses by evaporation and returns to the hydrological cycle with potential evaporation rate modified by a correction factor. Although interception losses may be highly significant in the annual water balance, it is relatively unimportant for flood-producing storms.

2.3. Snowmelt Snow accumulation and melt are important hydrological processes in river basins, where the snow pack acts as storage in which precipitation is retained during the cold season and subsequently released as melt water during the warmer season. The snowmelt is incorporated within WetSpa Extension. This component is optional and temperature data is required additionally if the sow routine is selected. The conceptual temperature index or degree-day method (Martinec et al., 1983) is widely used in snowmelt modelling, in which the full energy balance is replaced by a term linked to air temperature. It is physically sound in the absence of short wave radiation, when much of the energy supplied to the snow pack is atmospheric long wave radiation. Its reliance on daily temperature and precipitation data make it useful for modelling snow processes in regions with a lack of regular snow observations, or historical periods

216

WetSpa Extension: Documentation and User Manual with limited data. In WetSpa Extension, an additional snowmelt caused by the advective heat transferred to the snow pack by precipitation is also considered. The total snowmelt is calculated as M i = C snow (Ti T0 ) + C rain Pi (Ti T0 ) where Mi is the daily snowmelt at cell i (mm/day), Ti (2.5)

is the cell daily mean

temperature (C), T0 is a threshold temperature (usually 0C), Csnow is the degree-day or melt factor (mm/C/day), and Crain is a degree-day coefficient regarding to the heat contribution from rainfall (mm/mm/C/day). Specifically, temperature, precipitation and snow cover often vary significantly within a mountainous catchment, and in many cases, the hydro-meteorological information from mountainous areas is quite sparse. To account for the large variations in temperature with altitude, the reference series is adjusted for each grid cell by the lapse rate correction
T i = T ref + (H i H ref )

(2.6)

where Tref is the temperature at the reference station (C), Hi and Href are the height at cell i and at the reference station, and is the temperature lapse rate. The degree-day coefficient implicitly represents all terms of the energy budget that account for the mass balance of a snow pack, and is therefore highly variable over time (Singh et al., 2000), and different between vegetation types (Kite & Kouwen, 1992). However, a constant value is used in the current version for simplicity. This factor can be determined by field experiments, or will have to be obtained by calibration otherwise. Moreover, the degree-day method by definition is only valid for daily melt simulations, whereas simulations for short time intervals require finer temporal resolutions. In this case, a fully energy balance module is suggested, and it will be incorporated in the future version.

2.4.

Excess rainfall and infiltration

Excess rainfall, or effective rainfall, is that part of rainfall in a given storm, which falls at intensities exceeding the infiltration capacity of the land surface. It may stay

217

Appendix A temporarily on the soil surface as depression, or become direct runoff or surface runoff at the watershed outlet after flowing across the watershed surface under the assumption of Hortonian overland flow. Direct runoff forms the rapidly varying portions of watershed hydrographs and is a key component for estimating the watershed response. Infiltration is the downward flow of water into the soil defined as the quantity of rainfall that does not contribute to surface runoff. Under normal conditions, the infiltration rate is mainly a function of: (1) rainfall characteristics, (2) surface conditions, (3) soil characteristics, (4) initial moisture content of the soil, etc. It is desirable to relate loss rates to physical characteristics of the watershed in a continuous simulation so that loss rates may be computed as a function of catchment characteristics and soil moisture conditions during a model simulation. In WetSpa Extension, a modified coefficient method for estimating surface runoff and infiltration processes is used relating runoff and infiltration with topography, soil type, land use, soil moisture, and rainfall intensity. The equations can be expressed as:
(t ) PE i (t ) = C i [Pi (t ) I i ((t ) )] i i ,s

(2.7)

Fi (t ) = Pi (t ) I i (t ) PE i (t )

(2.8)

in which PEi(t) is the rainfall excess on cell i over the time interval (mm), Fi(t) is the cell infiltration (mm), Ii(t) is the interception loss (mm), i(t) is the cell soil moisture content at time t (m/m), i,s is the soil porosity (m/m), is an exponent related with rainfall intensity (-), and Ci is the cell potential rainfall excess coefficient or potential runoff coefficient (-). This parameter Ci has a rather stable regularity under ideal conditions. Default rainfall excess coefficients for different slope, soil type and land cover are taken the reference from the literature (Kirkby 1978, Chow et al. 1988, Browne 1990, Mallants & Feyen 1990, and Fetter 1980). Based on the physical analysis and linear interpolations of these values, a look up table is then established, relating potential rainfall excess coefficient to the different

218

WetSpa Extension: Documentation and User Manual combinations of slope, soil type and land use. The rainfall excess is closely related with the relative soil moisture content. No rainfall excess when soil is dry, and actual rainfall excess coefficient approaches to the potential value when soil moisture content close to saturation, under which the infiltrated water is considered to be used for percolation, evapotranspiration and lateral interflow. The exponent in the formula is a variable reflecting the effect of rainfall intensity on the rainfall excess coefficient. The value is higher for low rainfall intensities resulting less surface runoff, and approaches to 1 for high rainfall intensities. The threshold value can be defined during model calibration. If = 1, a linear relationship is assumed between rainfall excess and soil moisture. The effect of rainfall duration is also accounted by the soil moisture content, in which more excess produces due to the increased soil moisture content. Figure A-2.2 shows the relationship between actual rainfall excess coefficient, relative soil moisture content and potential rainfall excess coefficient with an exponent of 2.0.

1.0

Rainfall excess co efficient

0.8 0.6 0.4 0.2 0.0 0.0

C C i==0.2 0.2 C 0.5 C i==0.5

C 0.8 C i==0.8 a = 2.0

0.2

0.4

0.6

0.8

1.0

Relative saturation

Figure A-2.2: Relationship between rainfall excess coefficient and soil moisture

2.5.

Depression and overland flow

Precipitation that reaches the ground may infiltrate, or get trapped into several small depressions, which is retained in puddles, ditches, and on the ground surface. As soon as rainfall intensity exceeds the local infiltration capacity, the rainfall excess begins to fill depression. Water held in depression at the end of rain either evaporates or contributes to soil moisture and subsurface flow by the following infiltration. Depression storage may be of considerable magnitude and may play an important role in hydrological analysis. Stock ponds, terraces, and contour farming etc. tend to moderate flood by increasing depression storage. Depression losses usually occur

219

Appendix A during the initial period of the storm and are negligible after a certain time. Factors that affect depression storage include: (1) nature of terrain; (2) slope, the more slope gradient, the less depression losses; (3) type of soil surface, the more sandy soil, the more depression losses; (4) land use, the more woody land use, the more depression losses; (5) antecedent rainfall, the more soil water content, the less depression storage; and (6) time, for which depression losses decrease with time. Depression is considered included in the potential rainfall excess coefficient in the WetSpa Extension, in order to emphasize its effects on surface runoff production, particularly for the rough surfaces and for small flood events. Therefore, default potential rainfall excess coefficient should be determined cautiously from the literature values, taking the influence of interception and depression into account.

2.5.1. Formulation of depression storage Due to the extreme variability of affecting factors, it is very difficult to specify a general relationship for depression losses. In WetSpa Extension, a simple empirical equation suggested by Linsley (1982) is used to estimate depression storage:
PC i SDi (t ) = SDi , 0 1 exp SD i ,0

(2.9)

in which SDi(t) is the cell depression storage at time t (mm), SDi,0 is the cell depression storage capacity (mm), and PCi is the accumulative excess rainfall on the soil surface (mm). The concept of Eq. (2.9) is that both overland flow and depression storage occurs simultaneously, allowing some of the water delivering as overland flow, even if excess rainfall is less than the depression storage capacity. A sketch of SDi(t) as a function of PEi is shown in Figure A-2.3. The increment of depression storage can be obtained by derivation of t for both side of Eq. (2.9) as:
PC i SDi (t ) = PE i (t ) exp SD i,0

(2.10)

220

WetSpa Extension: Documentation and User Manual where SDi(t) is the increment of depression storage at cell i over the time interval (mm), and PEi(t) is the excess rainfall for the time increment (mm). Considering that the rainfall is interrupted between storm events, the accumulative excess rainfall can be estimated based on Eq. (2.9), which is the excess rainfall at present time step plus an excess rainfall corresponding to the depression storage at last time step.
SDi (t 1) PC i = PE i (t ) SDi , 0 ln1 SDi , 0

(2.11)

Obviously, PCi equals PEi(t) when depression storage at last time step, SDt-1, is zero, and becomes a very large value when SDi(t-1) approaches to SDi,0, leading to a very small depression storage increment, SDi,t, from Eq. (2.10). The capacity of depression storage, SDi,0, is mainly affected by landform, soil type and vegetation. Based upon the analysis and linear interpolation of the typical values collected in the literature (ASCE, 1969; SINCE, 1972; Sheaffer, 1982), a lookup table for default depression storage capacity is set up according to the categories of slope, land use and soil type, which is similar as the lookup table of potential rainfall excess coefficient.

Depression storage (mm)

3.0 2.0 1.0 0.0 0.0

SDi,0 = 3.5 mm DS0

3.0

6.0

9.0

12.0

15.0

Rainfall excess (mm)

Figure A- 2.3: Sketch of depression storage as a function of excess rainfall

2.5.2

Mass balance of depression storage

As discussed above, the depressed water on soil surface will be depleted by evaporation directly or infiltrated into the soil after the rainstorm. The mass balance of depression storage can be expressed as:

221

Appendix A SDi (t ) = SDi (t 1) + SDi (t ) EDi (t ) Fi (t ) (2.12)

where EDi(t) and Fi(t) are cell evaporation and infiltration from depression storage for the time increment after the rainstorm (mm); EDi(t) = 0 when Pi(t) > 0 or DSi(t-1) = 0; EDi(t) = EP Ei(t), when Pi(t) = 0 and DSi(t-1) EP-EIi(t), in which EP and EIi(t) are the potential evaporation and the evaporation from the cell interception storage (mm); EDi(t) = DSi(t) when Pi(t) = 0 and 0 < SDi(t) < EP-Ei,(t). The infiltration from depression storage after rainstorm can be estimated using Eq. (2.7) and Eq. (2.8) by taking the depressed water as rainfall on the ground surface.

2.5.3. Formulation of overland flow Recall that the excess rainfall is a sum of overland flow and the change of depression storage, the amount of overland flow, RSt (m), can be written as:
PE i PE i (t )1 exp SD i,0

RS

(t ) =

(2.13)

Eq. (2.13) assumes that both overland flow and depression storage occur simultaneously as described in Figure A-2.4, for which the overland flow approaches to zero when the accumulative excess rainfall is very small, and approaches to PEi(t) when the depression storage closes to its capacity. This is different with the assumption that overland flow begins only after the depression storage capacity is reached as the dashed line shown in the figure.

1.0

RS i (t) / PEi (t)

0.5 SD0 SDi ,0 = 3.5 mm

0.0 0.0 3.0 6.0 9.0 12.0

PEi (mm)

Figure A-2.4: Graphical presentation of excess rainfall and overland flow 222

WetSpa Extension: Documentation and User Manual

2.6.

Water balance in the root zone

Soil moisture storage is the actual quantity of water held in the soil at any given instant, usually applied to a soil layer of root depth. Based on the different soil water content, the moisture storage can be divided into saturation capacity, field capacity, plant wilting point, residual soil moisture, etc. WetSpa Extension calculates water balance in the root zone for each grid cell. Soil water is fed by infiltration and removed from the root zone by evapotranspiration, lateral interflow and percolation to the groundwater storage, as described in Figure A-2.5.

ES

RS

RI

RG

Figure A-2.5: Graphical presentation of soil water balance The moisture storage in the root zone is expressed by a simple balance equation as: Di [ i (t ) i (t 1)]i = Fi (t ) ESi (t ) RGi (t ) RI i (t ) (2.14)

in which i(t) and i(t-1) are cell soil moisture content at time step t and t-1 (m/m), Di is the root depth (mm); Fi(t) is the infiltration through soil surface for the time increment (mm), including the infiltration during the rainstorm and the infiltration from depression storage after the rainstorm (mm), ESi(t) is the actual evapotranspiration from the soil for the time increment (mm), RGi(t) is the percolation out of root zone or groundwater recharge (mm), and RIi(t) is the interflow or lateral shallow subsurface flow out of the cell for the time increment (mm). Apparently, soil moisture content in the root zone is a crucial factor in the model, which affects the hydrological processes of surface runoff, actual evapotranspiration, interflow and percolation out of the root zone soil. 223

Appendix A

2.7.

Evapotranspiration from soil

2.7.1. Potential evapotranspiration PET is defined as the quantity of water vapour, which could be emitted by plant or soil surface per unit area and unit time under the existing conditions without water supply limit. The main influencing factors to the potential evaporation are: (1) solar radiation, providing energy or heat; (2) wind speed, transporting the moisture away from the surface, and (3) specific humidity gradient in the air above the water surface, being the driving forces for diffusion of water vapour. When taking all of these variables into account in a continuous simulation model, it would make the model much too complex. In WetSpa Extension, three options are available to estimate PET. 1) Penman-monteith equation The original Penman-Monteith method has been modified by many researchers and extended to plant surfaces by introducing resistance factors. A newly result is the FAO-56 Penman-Monteith equation derived from the original Penman-Monteith equation and the equation of aerodynamic and surface resistances, which can be used to estimate PET on an hourly or daily time basis (Allen, 2000):
37 u 2 (e s e a ) T + 273.2 + (1 + 0.34u 2 )

EP =

0.408(R n G ) +

(2.15)

where EP is the reference or PET (m), Rn is the net radiation (MJ/m), G is the soil heat flux (MJ/m), T is the air temperature (C), es is the saturation vapour pressure at air temperature (kPa), ea is the vapour pressure of air (kPa), u2 is the wind speed at 2 m (m/s), is the slope of saturation vapour pressure curve at air temperature (kPa/C), and is the psychomotor constant (kPa/C). Eq.(2.15) is an estimate of EP from a hypothetical short grass with a height of 0.12 m, a surface resistance of 70 s/m, and an albedo of 0.23. Supporting equations of Eq. (2.15) for calculating es, ea, , and G can be found from Allen (2000). Once the meteorological data of net radiation, air

224

WetSpa Extension: Documentation and User Manual temperature, relative humidity, and wind speed is available, the PET rate EP can be estimated using Eq. (2.15) in the WetSpa Extension. 2) Statistical method based on historical records Even though the Penman-Monteith method is physically based and can give good estimate to the PET, but it is always the fact that collecting all relevant meteorological data is rather difficult for many study areas. In this case, a statistical method can be applied if a long series of evaporation records are available inside the catchment or outside the catchment with similar meteorological environment. De Smedt, D. (1997) derived a simple empirical equation for estimating potential evaporation from free water surface applied to the Belgian situation. Through statistically analysis of the daily potential evaporation records over the period 1901-1993 at Ukkel meteorological station, Belgium, a mathematical equation that fits the curve of the average daily potential evaporation was obtained as:
1.35

d 87 EPd = 0.27 + 1.37 1 + sin 2 365

(2.16)

in which EPd is the daily potential evaporation (mm), d is the day of a year, starting from 1 for the first of January and ending with 365 for the 31st of December, while intercalate days are not considered. Figure A-2.6 gives a graphical comparison between observed and simulated daily potential evaporation at Ukkel for the year 1997. The hourly distribution of potential evaporation for a certain day is estimated by the empirical equation as:
h 6 1 + 0.9 sin 2 24

EP =

EP d 24

(2.17)

where EP is the hourly potential evaporation (mm), h is the hour of a day between 0 and 24. Eq. (2.17) assumes that the hourly potential evaporation is always higher or equal to 0.1EPd/24 and therefore never reaches zero. The maximum potential evaporation rate occurs at solar noon and equals 1.9EPd/24. Obviously, the integral of

225

Appendix A the above equation over a day is exactly equal to EPd of that day. A simulated curve of hourly EP is presented in Figure A-2.7 with EPd = 3 mm.

7.5

Observed

EPd (mm)

5.0

Simulated

2.5

0.0 1/2 3/3 1/3 1/1 31/1 1/4 2/4 1/5 3/5 1/6 2/6 1/7 3/7 1/8 2/8 1/9 2/10 1/10 2/11 1/11 2/12 1/12 2/9

Date (d/m)

Figure A-2.6: Observed and simulated daily EP at Ukkel for the year 1997

0.3

EPh (mm)

0.2

0.1

0.0 0 4 8 12 16 20 24

Time (h)

Figure A-2.7: Simulated hourly EP at Ukkel with EPd = 3mm For a given study area with available historical evaporation records, the constant 0.27, 1.37 and 1.35 in Eq. (2.19) can be readjusted in order to get a better fit to the measurement. Due to the lack of theoretical basement behind, errors may arise in certain hours. But in general, this method can meet the requirement of large flood prediction and other engineering purpose. The main advantage of this method is that the potential evaporation rate is only a function of the time, and can be easily coupled in the model programming. 3) Measurement using pans Evaporation pans provide a measurement of the combined effect of temperature, humidity, wind speed and sunshine on the PET. The potential evaporation can be 226

WetSpa Extension: Documentation and User Manual estimated with the pan evaporation multiplied by a pan coefficient, and used directly in the model for parameter calibration and model simulation.

2.7.2. Actual evapotranspiration Without considering the evaporation from interception storage and depression storage, actual evapotranspiration is defined as the sum of the quantities of water vapour evaporated from the soil and the plants when the ground is at its actual moisture content. Thus, if soil is fully saturated, then it is expected that the actual evapotranspiration rate equals to the PET rate. However, if the soil or vegetation is water stressed, the evapotranspiration will be less than potential evapotranspiration. Influencing factors to the actual evapotranspiration include weather, vegetation and soil condition, etc. Since the actual evapotranspiration is governed by the availability of water, soil moisture content becomes a crucial factor, which is determined by water recharge and the soil characteristics. In the WetSpa Extension, evapotranspiration consists of four parts: (1) evaporation from interception storage, (2) evaporation from depression storage, (3) evapotranspiration from soil, and (4) evapotranspiration from groundwater storage. It is assumed that water evaporates to the atmosphere in a cascade way, i.e. from interception storage, depression storage, soil matrixes, and groundwater storage consequently. The evaporation from interception storage and depression storage has been described in section A-2.2 and A-2.5, and the groundwater contribution to the evapotranspiration will be described in section A-2.9. The actual evapotranspiration from soil and plant is calculated for each grid cell using the relationship developed by Thornthwait and Mather (1955) as a function of PET, vegetation and its growing stage, and moisture content of the cell:
i (t ) i , w [c v EP EI i (t ) ED i (t )] for i , w i (t ) p i , f ES i (t ) = i, f i,w ( ) ( ) c EP EI t ED t for i (t ) i , f i i v

(2.18)

where ESi(t) is the actual soil evapotranspiration for the time increment (mm), cv is a vegetation coefficient determined by land use classes varying throughout the year, i(t) 227

Appendix A is the cell average soil moisture content at time t (m/m), i,f is the soil moisture content at field capacity (m/m), and i,w is the soil moisture content at plant permanent wilting point (m/m). It can be concluded from Eq. (2.18) that when the sum of interception and depression storage is greater than the PET, all evaporation comes from the interception and depression storage with a potential rate. When the sum of interception and depression storage is less than the amount of PET, all the remaining storage evaporates at this time step, and there is a part of evapotranspiration from the soil layer depending on the soil moisture content. For the simulation between storm events, actual evapotranspiration is mainly from the soil and plant, which varies linearly between PET when soil moisture content is at or above field capacity, and zero when soil moisture content is below the wilting point. A graphical presentation of soil evapotranspiration is given in Figure A-2.8, in which i,s is the soil porosity (m/m). For the cell in urban areas, soil evapotranspiration is reduced by the impervious areas, and is calculated by cell evapotranspiration times the pervious percentage.

ESi(t)/cvEP

0 i,w i,f Moisture content i,s

Figure A-2.8: Graphical presentation of soil evapotranspiration

2.8.

Percolation and interflow

Percolation or groundwater recharge refers to the natural process by which water is added from soil water zone to the saturation zone of the aquifer. Groundwater recharge is an important component in the root zone water balance, which connects the soil water and the saturated groundwater. The main influencing factors to the groundwater recharge are the hydraulic conductivity, root depth, and water content of the soil. In WetSpa Extension, percolation out of root zone is assumed to pass directly 228

WetSpa Extension: Documentation and User Manual to the groundwater reservoir, and estimated based on the Darcys law, being the product of hydraulic conductivity and the gradient of hydraulic potential. When an assumption is made that the pressure potential only varies slightly in the soil, its gradient can be approximated to zero, and the percolation is controlled by gravity alone (Famiglietti and Wood, 1994). Based on this assumption, the percolation amount out of root zone is simply specified as the hydraulic conductivity corresponding to the average effective saturation in the respective soil layer. The Brooks and Corey relationship between hydraulic conductivity and effective saturation is used to define percolation, which is simply (Brooks and Corey, 1966):
(t ) i , r RGi (t ) = Ki [ i (t )]t = Ki , s i t i , s i , r
A

(2.19)

where RGi(t) is the percolation out of root zone over the time interval (mm), Ki[i(t)] is the effective hydraulic conductivity corresponding to the average soil moisture content at time t (mm/h), t is the time interval (h), Ki,s is the cell saturation hydraulic conductivity (mm/h), i,s is the soil porosity (m/m), i,r is the cell residual moisture content (m/m), and A is the pore disconnectedness index, calculated by the equation A = (2+3B)/B, in which B is the cell pore size distribution index.

1.0 Sand

K[ik(/k (t)]/K s i,s

Loam 0.5 Clay

0.0 0.0 0.2 (st-sr)/(ss/sr) i,r]/(i,s-i,r) [i(t)0.4 0.6 0.8 1.0

Figure A-2.9: Effective hydraulic conductivity as a function of moisture content Figure A-2.9 gives a graphical presentation for the effective hydraulic conductivity as a function of moisture content for three different soil types: sand, loam and clay. It can be seen that the effective hydraulic conductivity varies with moisture content

229

Appendix A exponentially, reaching a maximum, the saturated conductivity, when soil is completely saturated, and zero when soil becomes dry. Interflow, or shallow subsurface lateral flow, is also a key component in the soil water balance. It is defined as the water which infiltrates the soil surface and moves laterally through the upper soil layers until it enters a channel, which includes litter flow, return flow, unsaturated through flow, saturated through flow and so on, but excludes the saturated groundwater flow. Due to the delayed flow time, interflow usually contributes to the falling limb of a flood hydrograph, but it may also be a part of peak discharge at the basin outlet, particularly for the areas with steep slope and forest cover in humid or temperate regions. Factors that influence the amount of interflow include: (1) physical properties and depth of the soil, for which coarse texture leads to more vertical flow, while fine texture or layered soil results in resistance to vertical flow and interflow may some time occur quickly; (2) vegetation cover and land use, which are directly related to the maintenance of infiltration capacity and the conditioning effect of organic material on soil structure, bulk density and porosity; (3) topography, for which the slope gradient is a major factor determining the amount and the velocity of interflow; (4) soil moisture content, for which higher moisture content tends to generate more interflow; and (5) lithology and climate of the study area. In WetSpa Extension, interflow is assumed to occur after percolation and cease when soil moisture is lower than field capacity. The quantity of interflow out of each cell is calculated from Darcy's Law and the kinematic approximation, i.e. the hydraulic gradient is equal to the land slope of the grid cell: RI i (t ) = c s Di S i K [ i (t )]t Wi (2.20)

in which RIi(t) (mm) is the amount of interflow out of the cell over the time interval t (h), Di is the root depth (m), Si is the cell slope (m/m), K[I(t)] (mm/h) is the cell effective hydraulic conductivity at moisture content i(t) (m/m), Wi is the cell width (m), and ci (-) is a scaling factor depending on land use, used to consider stream density and the effects of organic matter and root system on horizontal hydraulic conductivity in the top soil layer. Apparently, rapid interflow may generate in areas with high moisture, steep slope and well vegetation, while little is produced for other

230

WetSpa Extension: Documentation and User Manual areas with Eq. (2.20). For modelling simplification, interflow is assumed to join the surface runoff at the nearest channels or gullies serving as a supplementary discharge to the stream flow during and after storm event without further divisions among down slope neighbours. Soil hydraulic characteristics, such as porosity, field capacity, residual saturation, hydraulic conductivity, and so on, are collected from the literature, and used as default values in the WetSpa Extension.

2.9.

Groundwater storage and baseflow

Groundwater storage is defined as the quantity of water in the zone of saturation including that part of such stage when water is entering and leaving storage. Groundwater storage capacity refers to the volume of saturated groundwater that can be alternatively extracted and replaced in the deposit under natural conditions. Normally, the groundwater discharge forms a base flow to the hydrograph at basin outlet. Groundwater storage capacity is governed by the thickness and extent of the aquifer and its porosity, while the movement of groundwater is governed by the hydraulic gradient and the hydraulic conductivity of the aquifer. For the purpose of streamflow prediction, an estimate must be made of flow from the groundwater storage into the stream for each time step. Since little is known about the bedrock, the simple concept of a linear reservoir is used to estimate groundwater discharge on a small subcatchment scale, while a non-linear reservoir method is optional in the model with storage exponent of 2 (Wittenberg and Sivapalan, 1999). The groundwater outflow is added to any runoff generated to produce the total streamflow at the subcatchment outlet. The general groundwater flow equation can be expressed as:
QG s (t ) = c g [SG s (t ) 1000]

(2.21)

where QGs(t) is the average groundwater flow at the subcatchment outlet (m3/s), SGs(t) is the groundwater storage of the subcatchment at time t (mm), m (-) is an exponent, m = 1 for linear reservoir ,and m = 2 for non-linear reservoir, cg is a groundwater recession coefficient taking the subcatchment area into account, has a dimension of

231

Appendix A (m2/s) for linear reservoir and (m/s) for non-linear reservoir, which is dependent upon area, shape, pore volume and transmissivity of the subcatchment, and can be estimated from recession portions of streamflow hydrographs if measurement data at the subcatchment outlet are available. For each subcatchment, the groundwater balance can be expressed as:

SGs (t ) = SGs (t 1) +

[RG (t )A ]
i =1 i i

Ns

As

EGs (t )

QGs (t )t 1000As

(2.22)

where SGs(t) and SGs(t-1) are groundwater storage of the subcatchment at time step t and t-1 (mm), Ns is the number of cells in the subwatershed, Ai is the cell area (m2), As is the subcatchment area (m2), EGs(t) is the average evapotranspiration from groundwater storage of the subcatchment (mm), and QGi(t) is the groundwater discharge (m/s). The component of evapotranspiration from groundwater storage is considered in the WetSpa Extension, which may be produced by deep root system or by capillary drive in the areas with shallow groundwater table. It happens only when soil moisture is less than field capacity from Eq. (2.18) and has a greater impact during the summer than the winter, giving the effect of a steeper recession during dry period. A simple linear equation is used in the model relating deep evapotranspiration with PET and groundwater storage as: EGi (t ) = cd [cv EP EIi (t ) EDi (t ) ESi (t )] (2.23)

where EGi(t) is average evapotranspiration from groundwater storage (mm), EP is PET (mm), and cd (-) is a variable, calculated by SGi(t)/SGs,0, in which SGi(t) is the groundwater storage of the subwatershed at time t (mm), and SGs,0 is the groundwater storage capacity of the subwatershed (mm). Using the method of groundwater reservoir, there are only two groundwater parameters, the groundwater recession coefficient and the storage capacity, which can be determined by calibration against baseflow separated from the observed hydrograph.

232

WetSpa Extension: Documentation and User Manual 2.10. Overland flow and channel flow routing

2.10.1 Flow response at a cell level The routing of overland flow and channel flow in WetSpa Extension is implemented by the method of a linear diffusive wave approximation. This method is suitable for simulating sheet flow and channel flow at a certain degree, and one of the important advantages is that it can be solved analytically, avoiding numerical calculation and identification of the exact boundary conditions. Assuming the cell as a reach with 1-D unsteady flow and neglecting the inertial terms in the St. Venant momentum equation, the flow process in the cell can be modelled by the diffusive wave equation as (Miller and Cunge, 1975):
Q Q 2Q + ci di =0 t x x 2

(2.24)

where Q (m/s) is the flow discharge at time t (s) and location x (m), ci is the kinematic wave celerity at cell i (m/s), di is the dispersion coefficient at cell i (m/s). Considering a system bounded by a transmitting barrier upstream and an adsorbing barrier downstream, the solution to Eq. (2.24) at the cell outlet, when the flow velocity and diffusion coefficient are constant, can be obtained by the first passage time density distribution of a Brownian motion and expressed as (Eagleson, 1970):
(c t l )2 exp i i 4d i t 2 dit 3 li

ui (t ) =

(2.25)

where ui(t) is the cell impulse response function (1/s), and li is cell size (m). Two parameters ci and di are needed to define the cell response function, which can be estimated using the relation of Manning as (Henderson, 1966):
5 ci = vi 3

(2.26 )

233

Appendix A
di = vi Ri 2S i

(2.27)

where Ri is the average hydraulic radius of cell i (m), Si is the cell slope (m/m), and vi is the flow velocity of the cell i (m/s). The hydraulic radius is determined by a power law relationship with an exceeding probability (Molnar and Ramirez, 1998), which relates hydraulic radius to the controlling area and is seen as a representation of the average behaviour of the cell and the channel geometry:

Ri = a p ( Ai ) p
b

(2.28)

where Ai is the drained area upstream of the cell (km), which can be easily determined by the flow accumulation routine in ArcView GIS, ap (-) is a network constant and bp (-) a geometry scaling exponent, both depending on the discharge frequency. The flow velocity is calculated by the Mannings equation as:
2 1

vi =

1 3 2 Ri S i ni

(2.29)

where ni is the Mannings roughness coefficient (m-1/3s), which depends upon land use categories and the channel characteristics. Default Mannings roughness coefficients can be collected from literature. The velocity calculated by Eq. (2.29) may be very large or even zero due to variations in land surface slope. Therefore it is bounded between predetermined limits vmin and vmax during model calculation. Flow velocity is a time-dependent, discharge-related and location-related hydrological variable. But to be applicable of the diffusive wave approximation method for hydrological analysis, the flow must be only location-related. In reality, water depth usually increases as water goes downstream. As water deepens, the effective resistance of the streambed and banks on the flow diminishes because the hydraulic radius increases. To reflect this property, the channel roughness coefficient is set between predetermined limits nmax and nmin, depending upon the GIS derived stream orders in the WetSpa Extension. Thus, with the supporting Equations (2.26) to (2.29), the cell impulse unit response function ui(t) can be calculated for each grid cell over the entire watershed, which

234

WetSpa Extension: Documentation and User Manual reflects the redistribution tendency in the flow element serving as a flow redistribution function.

2.10.2. Flow response at a flow path level Under the assumption of linear routing system, the flow response at the end of a flow path, resulting from a unit impulse input to a single cell, can be calculated without the interference of the inputs to the other cells. Determining the flow-path response consists in routing the impulse through the corresponding sequence of cells down to the system outlet. Along the flow-path, the impulse travels through many cells, each of them having a different unit-impulse response function. In this routing process, the output of any cell becomes the input to the receiving cell, and the original input distribution is continuously modified by the flow dynamics in the cells, which are described by their impulse response functions. The flow path response is found by successively applying the convolution integral, giving:

Ui (t) = u j (t )
j =1

(2.30)

where Ui(t) is flow path response function (1/s), the subscript i refers to the cell in which the input occurs, j is the cell sequence number, and N is the total number of cells along the flow path. The diffusion equation model satisfies Eq. (2.30) within the cells, which means that it allows for longitudinal decomposability. Since the cell unit impulse response functions are time-invariant, the result of the convolutions of Eq. (2.30) is also time-invariant, and therefore, there is a linear relation between the flow path response and the impulse input. Assuming that the flow path response Ui(t) is also a first passage time distribution, De Smedt F. et al. (2000) and Liu et al. (2002, 2003) proposed an approximate numerical solution to Eq. (2.30), relating the discharge at the end of a flow path to the available runoff at the start of the flow path:
(t t i ) 2 exp 2 t i3 2 i t / t i

U i (t ) =

1 2 i2 t 3

(2.31)

235

Appendix A where ti is the mean flow time from the input cell to the flow path end (s), and i2 is the variation of the flow time (s). The parameters ti and i2 are spatially distributed, and can be obtained by convolution integral along the topographic determined flow paths as a function of flow celerity and dispersion coefficient:
N 1 ti = c j =1 j

l j

(2.32)

i2 =

2d j 3 j =1 c j
N

l j

(2.33)

The summations presented in Eq. (2.32) and (2.33) can be calculated for each grid cell as a weighted flow length to the water outlet or any downstream converging point with the routine FLOWLENGTH involved in the standard GIS tools. Examples of such flow path impulse response function are presented in Figure 2.10 for different mean flow time and its variation. It is seen that the response function is asymmetric with respect to time caused by the wave attenuation.

0.08 0.06
ti = 1800, i2 = 9002 Series1 ti = 3600, i2 = 18002 Series2 ti = 7200, i2 = 27002 Series3

Ui (t) (1/s)

0.04 0.02 0 0 1800 3600 5400 7200

9000

10800

t (s)

Figure A-2.10: Flow path response functions with different ti and i2 The flow response at the end of a flow path, to an arbitrary input at the start cell, can be calculated by convolving the input runoff volume by the flow path unit impulse response function. From a physical point of view, this is equivalent to decomposing the input into infinite impulses and adding all the responses into a single response. Thus, the outflow hydrograph to an arbitrary input can be determined as:

236

WetSpa Extension: Documentation and User Manual

Q i (t ) =

V ( )U (t )
=0 i i

(2.34)

where Qi(t) is the outflow at the end of a flow path produced by an arbitrary input in cell i (m/s); Ui(t-) is the flow path response function (1/s), being equivalent to the instantaneous unit hydrograph (IUH) used in the conventional hydrology, and is the time delay (s); Vi() is the input runoff volume at cell i and at time (m), including surface runoff and interflow, as well as groundwater runoff if cell i is located at the subcatchment outlet.

2.10.3. Flow response of the catchment Considering the areal decomposability in a linear routing system, the catchment flow response can be determined as the sum of its elements responses from all contributing cells. Thus, the catchment flow response can be calculated as:

Q (t ) =

Q (t )
i =1 i

Nw

(2.35)

where Q(t) is the total flow at the catchment outlet (m/s), Nw is the number of cells over the entire catchment. Hence, the flow routing consists of tracking runoff along its topographic determined flow path, and evaluating groundwater flow out of the subcatchment. The total discharge is the sum of the overland flow, interflow and groundwater flow, and is obtained by convolution of the flow response from all grid cells. The advantage of this approach is that it allows the spatially distributed runoff and hydrological parameters of the terrain to be used as inputs to the model, and can route runoff from a certain land use area to the catchment outlet or any downstream converging point.

2.11.

Subwatershed integration

In case of watershed modelling on medium or large scale, model parameterisation and computation on small grid size are tedious, costly and time consuming. On the other 237

Appendix A hand, working with coarse spatial resolution may introduce errors by aggregation of spatial input data and misrepresentation of the true watershed characteristics. To cope with this problem, WetSpa Extension provides a simplified semi-distributed option working on the scale of a small hydrological unit, so as to allow adequate simulation and mapping of the areal distribution of the hydrological processes. These units correspond to very small subcatchments, built up from high resolution DEM data, rather than to large grid cells with approximately the same area as the subwatersheds. This has the advantage of allowing for the internal drainage structure of the units, which would be impossible by using large grid cells. Model parameters, meteorological data input, and state variables for each simulation unit are obtained by integration of the values from all cells of that subcatchment. Meanwhile, the water and energy balance, as well as the process state variables, are computed on each unit during model simulation at each time step. The subwatershed parameters calculated by WetSpa Extension include area, slope, potential rainfall excess coefficient, interception capacity, depression capacity, soil physical properties, etc. Flow hydrographs are first calculated at the outlet of each subcatchment using the subcatchment response function, and thereafter, the flow is routed to the catchment outlet along the river channel by means of channel flow response function. Considering the effect of cell characteristics on the subwatershed IUH, the subcatchment response function is computed by integration of the flow path response functions for all cells in the subcatchment weighted by their potential rainfall excess coefficient. The equation can be written as:

U s (t ) = [CiU i (t )] Ci
i =1 i =1

Ns

Ns

(2.36)

where Us(t) is the response function or IUH of the subcatchment (1/s), Ci is the potential rainfall excess coefficient at cell i (-), Ui(t) is the flow path response function at the subcatchment outlet with runoff input at cell i (1/s), and Ns is the number of cells in the subcatchment. The flow hydrograph at subcatchment outlet is obtained by summation of its surface runoff, interflow and groundwater flow, and can be expressed as:

238

WetSpa Extension: Documentation and User Manual

Q s (t ) =

=0

V ( )U (t ) + QG (t )
s s s

(2.37)

where Qs(t) is the flow hydrograph at the subcatchment outlet (m/s), Vs() is volume of readily available runoff of the subcatchment including surface runoff and interflow (m), is the time delay (s), and QGs(t) is the groundwater flow at the subcatchment outlet (m/s). The total hydrograph at the watershed outlet is obtained by integration of the flow hydrographs produced from each subwatershed, and can be expressed as:

Q (t ) =

Q ( )U (t ) t
s =1 s r

Nr

(2.38)

where Q(t) is the flow hydrograph at the catchment outlet (m/s), Ur(t) is the channel response function from the subcatchment outlet to the catchment outlet calculated by Eq. (2.31) (1/s), t is the time interval (s), and Nr is the number of subcatchment or the number of stream links in the catchment. With the unit response functions defined for each simulation unit and the corresponding river channel, water can be routed accumulatively downstream up to the catchment outlet. However, the process of flow routing within each subcatchment can be omitted in case of highly intensive watershed discretization, since the water may flow out of the subwatershed within the first time step. In practice, division of the watershed should be performed according to the project purpose and the complexity of the terrain. A few simulations are necessary to decide the watershed discretization to meet varies objectives of the project.

2.12.

Catchment water balance

Water balance for the entire catchment is used to keep track of water changes in the hydrological system, and also a measure of model performance by comparing the simulation results with the field observations. Among the constituents in the system, soil water content is an important state variable that influence fluxes into and out of the root zone (infiltration, evapotranspiration, percolation and interflow) and the energy balance on the land surface. The stores of interception, depression, soil moisture and groundwater are treated as separate control volume, but related

239

Appendix A subsequently. Precipitation is the input to the system, while direct runoff, interflow, groundwater flow, and evapotranspiration are losses from the hydrological system. When modelling for a relatively long time period, changes in the storage of interception, depression and channel can be neglected, and the general watershed water balance can be expressed as:

P = RT + ET + SS + SG

(2.39)

where P is the total precipitation in the watershed over the simulation period (mm), RT and ET are total runoff and total evapotranspiration (mm), SS is the change in soil moisture storage for the watershed between the start and the end of the simulation period (mm), and SG is the change in groundwater storage of the watershed (mm). For a given simulation period T (s) and initial moisture and groundwater storage condition, these components can be expressed as:

P = Pi (t ) N w
t = 0 i =1

Nw

(2.40)

T Nw T Nr QG s (t ) t N r RT = [RS i (t ) + RI i (t )] N w + As t = 0 i =1 t = 0 s =1

(2.41)

ET = [EI i (t ) + EDi (t ) + ES i (t )] N w + [EGi (t )] N r

t = 0 i =1 t = 0 s =1

Nw

Nr

(2.42)

SS = Di [ i (T ) i (0)] N w
t =1

Nw

(2.43)

SG = [SG s (T ) SG s (0)] N r
s =1

Nr

(2.44)

where i(T) and i(0) are cell soil moisture content at the end and the start of the simulation period (m/m), SGs(T) and SGs(0) are subcatchment groundwater storage at the end and the start of the simulation period (mm), and the others have been 240

WetSpa Extension: Documentation and User Manual described in above sections. All of these components vary over time. A change in any one component of the watershed water balance can result in changes in the other components in the system. This is particularly useful for analysing the impact of land use changes on the watershed hydrological processes. For instance, deforestation results in more surface runoff and less infiltration, thus, decreasing the change in soil moisture storage and groundwater storage for a storm event, and the evapotranspiration is limited by the moisture content as well. When the model performs on a very long time series, the changes in soil moisture and groundwater storage will be less important, and the total precipitation is more or less equal to the sum of the runoff and the evapotranspiration.

3.

Parameter identification and model evaluation

3.1.

Default model parameters

3.1.1. Default parameters characterizing soil texture classes Soil textural classes are used to provide information concerning soil physical properties, such as porosity, hydraulic conductivity, pore size distribution index, etc. Although other descriptors such as horizon and structural size certainly influence the hydraulic parameters of soils, Cosby et al. (1984) perform a two-way analysis of variance of nine descriptors to conclude that soil texture alone can account for most of the discernible patterns. Over the last two decades, a great deal of efforts has been made to the estimation of soil hydraulic properties from the information on soil textures in the literature (McCuen et al., 1981; Rawls et al., 1982; Cosby et al., 1984; Rawls & Brakensiek, 1985; Carsel & Parrish, 1988). In WetSpa Extension, soil textures are classified into 12 USDA (U.S. Department of Agriculture) classes ranging from 1 to 12 based on the percentage of sand, silt and clay in the soil sample. Fine textured soils have a high percentage of clay and are very sticky when wet and hard when dry, while coarse textured soils have a high percentage of sand and are loose and friable. A lookup table is then established as presented in Table A-3.1 to estimate hydraulic properties as a function of soil texture classes using mean values obtained from the literature.

241

Appendix A Table A-3.1: Default parameters characterizing soil textural classes

Texture classes Sand Loamy sand Sandy loam Silt loam Silt Loam Sandy clay loam Silt clay loam Clay loam Sandy clay Silt clay Clay
1 2

Hydraulic conductiveity1 (mm/h) 208.80 61.20 25.92 13.32 6.84 5.58 4.32 2.30 1.51 1.19 0.90 0.60

Porosity1 (m/m) 0.437 0.437 0.453 0.501 0.482 0.463 0.398 0.471 0.464 0.430 0.479 0.475

Field capacity1 (m/m) 0.062 0.105 0.190 0.284 0.258 0.232 0.244 0.342 0.310 0.321 0.371 0.378

Wilting point1 (m/m) 0.024 0.047 0.085 0.135 0.126 0.116 0.136 0.210 0.187 0.221 0.251 0.251

Residual moisture1 (m/m) 0.020 0.035 0.041 0.015 0.015 0.027 0.068 0.040 0.075 0.109 0.056 0.090

Pore size distribution index2 (-) 3.39 3.86 4.50 4.98 3.71 5.77 7.20 8.32 8.32 9.59 10.38 12.13

Obtained by analysis of data presented in Rawls et al. (1982) Obtained from Cosby et al. (1984)

Soil texture is a key variable in the coupled relationship between climate, soil, and vegetation. Under given climatic and vegetation conditions the above soil-texturedependent physical properties, through their influence on soil water movement and the energy state of the water in the soil column, determine the soil wetness values which in turn establish the water condition of the vegetation (Fernandez-Illescas et al., 2001). One advantage in favour of using texture as the only distinguishing factor among components is that this approach significantly simplifies model data management. When only a single distinguishing factor is used, components with a common texture can be lumped together and the spatial soils information passed from the GIS to the hydrology model is set at 12 different specifications. Among the soil properties listed in Table A-3.1, hydraulic conductivity has by far the largest coefficient of variation based on the analysis of Carsel & Parrish (1988), and is more sensitive than other soil related parameters. These parameters allow to be revised during model calibration for refining better fit as described in Chapter A-4.

3.1.2. Default parameters characterizing land use classes Land use or land cover is an important boundary condition, which directly or indirectly influence many hydrological processes. The most obvious influence of land

242

WetSpa Extension: Documentation and User Manual use on the water balance of a catchment is on the evapotranspiration process. Different land use types have different evapotranspiration rates, due to their different vegetation cover, leaf area indices, root depths and albedo. During storms, interception and depression rates are different for different land use types. Land use also influences the infiltration and soil water redistribution process, because especially the saturated hydraulic conductivity is influenced by plant roots and pores resulting from soil fauna (Ragab & Cooper, 1993). An extreme example is the influence of build up areas and roads on overland flow. Moreover, land use influences surface roughness, which controls overland flow velocity and floodplain flow rates. Therefore, the effect of land use should be taken into account as much as possible in the simulation calculations. Table A-3.2: Default parameters characterizing land use classes
Land use classes Crop or mixed farming Short grass Evergreen needle leaf tree Deciduous needle leaf tree Deciduous broad leaf tree Evergreen broad leaf tree Tall grass Irrigated crop Bog or marsh Evergreen shrub Deciduous shrub Bare soil Impervious area Open water
1 2 3

Vegetated fraction1 (%) 85 80 80 80 80 90 80 80 80 80 80 5 0 0

Leaf area index1 (-) 0.5 6.0 0.5 2.0 5.0 6.0 1.0 6.0 1.0 6.0 5.0 6.0 0.5 6.0 0.5 6.0 0.5 6.0 0.5 6.0 1.0 6.0 0.5 2.0 0.0 0.0 0.0 0.0

Root depth1 (m) 1.0 1.0 1.5 1.5 2.0 1.5 1.0 1.0 1.0 1.0 1.0 1.0 0.0 0.0

Mannings coefficient2 (m-1/3s) 0.15 0.20 0.40 0.40 0.80 0.60 0.40 0.20 0.20 0.40 0.40 0.10 0.02 0.02-0.05

Interception capacity3 (mm) 0.05 1.00 0.05 1.00 0.10 0.80 0.05 0.80 0.05 2.00 0.15 2.00 0.10 1.50 0.05 1.00 0.05 1.00 0.10 1.50 0.05 1.50 0.05 1.00 0.00 0.00 0.00 0.00

Obtained from Dickinson et al. (1993) Obtained from Lull (1964), Zinke (1967) and Rowe (1983) Obtained from Chow (1964), Haan (1982), Yen (1992) and Ferguson (1998)

Fourteen basic land use classes are specified in the WetSpa Extension, based on the observed physical and biophysical cover of the land surface, as well as the function and the actual purpose for which the land is currently being used. Such information is obtained from ground surveys or remote sensing images. For each land use type, 243

Appendix A several vegetation parameters are defined taking the reference of previous studies as shown in Table A-3.2. In order to more correctly simulate the effect of vegetation on interception and evapotranspiration, a range of leaf area index and interception capacity is given in the table corresponding to the minimum and maximum values in a year for each vegetation class. Calculation of the temporal variation is described in Chapter A-2. Moreover, some of the parameters, such as root depth, roughness coefficient, etc., should be determined as functions of both soil type and land use. However, for the present implementation, these parameters remain a function of land use type only. Values of Mannings roughness coefficient shown in Table A-3.2 are typical values obtained from experiments reported in the literature. These values are generally representatives of very small areas when correspondence exists between reality and the mathematical model of one-dimensional flow over a plane. Therefore, if a larger grid size, e.g. larger than 100 m, is used in the model, these values should be adjusted downward to reflect the greater number of rills on long slopes (Wu et al., 1982; Hairsine & Parlange, 1986; Vieux & Farajalla, 1994). In case the model is applied to a medium or large watershed, the parameter of channel roughness coefficient, which is governed mainly by bed material and channel cross section, will have a great influence to the predicted hydrograph. In natural rivers without overbank flow, the roughness coefficient is generally small for downstream channels due to their fine bed materials, and is large for upstream channels in contrast. To account for these effects, a linear relationship is assumed in the model relating Mannings roughness coefficient to the stream order described as:
O Omin nr = nr , max O O (nr , max nr , min ) min max

(3.1)

where nr is the Mannings coefficient (m-1/3s) for stream order O, Omax and Omin are maximum and minimum stream order derived from ArcView GIS, and nr,max and nr,min are maximum and minimum Mannings coefficients corresponding to Omax and Omin (m-1/3s). Clearly, the Mannings coefficient has largest value for the channel with minimum order and smallest value for the channel with maximum order with

244

WetSpa Extension: Documentation and User Manual Equation 3.1. The value of nr,max and nr,min can be defined in the script according to the channel characteristics.

3.1.3. Potential runoff coefficient The runoff coefficient of a grid or catchment is the ratio of runoff volume to rainfall volume. A simple and practical technique is developed in WetSpa Extension to estimate the runoff coefficient under varying land use, soil type, slope, rainfall intensity and antecedent soil moisture condition as described in Chapter A-2. Undoubtedly, these variables act independently but also interact in their effects on the runoff coefficient and soil infiltration. A table of potential runoff coefficient is built for deferent land use, slope and soil type combinations and under the condition of near saturated soil moisture as shown in Table A-3.3. Water lost from the soil surface is considered to infiltrate into the soil used for further vertical percolation, evapotranspiration and lateral interflow. Table A-3.3: Potential runoff coefficient for different land use, soil type and slope
Sandy Silty Land Slope Sand Loamy Sandy Loam Silt Silt clay Clay clay Sandy Silty Clay use (%) sand loam loam loam loam loam clay clay Forest <0.5 0.5-5 5-10 >10 Grass <0.5 0.5-5 5-10 >10 Crop <0.5 0.5-5 5-10 >10 Bare <0.5 soil 0.5-5 5-10 >10 IMP 0.03 0.07 0.13 0.25 0.13 0.17 0.23 0.35 0.23 0.27 0.33 0.45 0.33 0.37 0.43 0.55 1.00 0.07 0.11 0.17 0.29 0.17 0.21 0.27 0.39 0.27 0.31 0.37 0.49 0.37 0.41 0.47 0.59 1.00 0.10 0.14 0.20 0.32 0.20 0.24 0.30 0.42 0.30 0.34 0.40 0.52 0.40 0.44 0.50 0.62 1.00 0.13 0.17 0.23 0.35 0.23 0.27 0.33 0.45 0.33 0.37 0.43 0.55 0.43 0.47 0.53 0.65 1.00 0.17 0.21 0.27 0.39 0.27 0.31 0.37 0.49 0.37 0.41 0.47 0.59 0.47 0.51 0.57 0.69 1.00 0.20 0.24 0.30 0.42 0.30 0.34 0.40 0.52 0.40 0.44 0.50 0.62 0.50 0.54 0.60 0.72 1.00 0.23 0.27 0.33 0.45 0.33 0.37 0.43 0.55 0.43 0.47 0.53 0.65 0.53 0.57 0.63 0.75 1.00 0.27 0.31 0.37 0.49 0.37 0.41 0.47 0.59 0.47 0.51 0.57 0.69 0.57 0.61 0.67 0.79 1.00 0.30 0.34 0.40 0.52 0.40 0.44 0.50 0.62 0.50 0.54 0.60 0.72 0.60 0.64 0.70 0.82 1.00 0.33 0.37 0.43 0.55 0.43 0.47 0.53 0.65 0.53 0.57 0.63 0.75 0.63 0.67 0.73 0.85 1.00 0.37 0.41 0.47 0.59 0.47 0.51 0.57 0.69 0.57 0.61 0.67 0.79 0.67 0.71 0.77 0.89 1.00 0.40 0.44 0.50 0.62 0.50 0.54 0.60 0.72 0.60 0.64 0.70 0.82 0.70 0.74 0.80 0.92 1.00

245

Appendix A To simplify the table, the original land use classes are reclassified into 5 classes as forest, grass, crop, bare soil and impervious area. The potential runoff coefficients for impervious (including open water surface) areas are set to 1. In addition, surface slope is discritized into 4 classes as shown in Table A-3.3. Values in the table are taken the reference from literature (Kirkby 1978, Chow et al. 1988, Browne 1990, & Fetter 1980) and adjusted after Mallants and Feyen (1990). In order to estimate the potential runoff coefficient on the basis of a continuous slope, a simple linear relationship between potential runoff coefficient and slope is created, which can be described as:

C = C 0 + (1 C 0 )

S S + S0

(3.2)

where C is the potential runoff coefficient for a surface slope S (%), C0 is the potential runoff coefficient for a near zero slope corresponding to the values listed on the first row of each land use class in Table A-3.4, and S0 (%) is a slope constant for different land use and soil type combinations, as listed in Table A-3.4, which is calibrated using the data in Table 3.4. Figure A-3.1 gives a graphical presentation of the grid potential runoff coefficient for a forest cover as a function of slope and different soil types. Table A-3.4: Slope constant S0 for determining potential runoff coefficient
Land use Sand Loamy Sandy Loam Silt Silt loam sand loam 0.650 0.551 0.471 0.393 0.620 0.522 0.442 0.365 0.590 0.493 0.413 0.338 0.560 0.464 0.384 0.311 0.530 0.435 0.355 0.284 Silty Sandy clay Clay clay Sandy Silty Clay loam loam loam clay clay 0.500 0.405 0.325 0.256 0.470 0.376 0.296 0.229 0.440 0.347 0.267 0.202 0.410 0.318 0.238 0.175 0.380 0.289 0.209 0.147 0.350 0.260 0.180 0.120

Forest 0.680 Grass 0.580 Crop 0.500 Bare soil 0.420

The left figure of Figure A-3.1 shows the potential runoff coefficient for a slope ranging from 0 to 20% and the supporting points, and the right one shows the potential runoff coefficient for a slope ranging from 0 to 300%. Clearly, the potential runoff coefficient approaches to C0 when slope is very small, and 1 when slope is infinite. The figure also shows that the changing magnitude of potential runoff coefficient is decreasing along with the increasing of surface slope. This conforms

246

WetSpa Extension: Documentation and User Manual that the runoff volume for a certain amount of rainfall is less or even not affected by slope beyond a critical slope (Sharma, 1986).

0.7 0.6

1.0

Potential runoff coefficient

0.5 0.4 0.3 0.2 0.1 0.0 0 5 10 15 20

Potential runoff coefficient

0.8

0.6

0.4

0.2

0.0 0 50 100 150

Sand Loamy sand Sandy loam Silt loam Silt Loam Sandy clay loa m Silt clay loam Clay loam Sandy clay Silt clay Clay

200

250

300

Slope (%)

Slope (%)

Figure A-3.1: Potential runoff coefficient vs. slope for forest and different soil types The influence of urban areas to the storm runoff is self-evident. Due to the grid size, cells may not be 100% impervious in reality. In WetSpa Extension, the remaining area is assumed to be pervious and covered by grass, and therefore, the potential runoff coefficient for urban areas is calculated as:

C u = IMP + (1 IMP ) C grass

(3.3)

where Cu and Cgrass are potential runoff coefficient for urban and grass grid, and IMP is the proportion of impervious area. Table A-3.5 is developed to associate an impervious cover percent with several of the specified land use categories. Impervious percent for residential area, commercial and industrial is estimated based on the information in Chow et al. (1988). Other estimates are considered reasonable guesses. Zero impervious percent is assumed for land use categories not listed (i.e. agriculture, grass land, and forest land).

247

Appendix A Table A-3.5: Impervious percentages associated with selected land use classes
No. 1 2 3 4 5 6 7 Land use description Residential area Commercial and industrial area Mixed urban or built-up land Transportation and communication utilities Streams, Canals, lakes and reservoirs Forest wetland Bare exposed rock Impervious percent (%) 30 70 50 100 100 100 100

In case the model is applied to a medium or large watershed, direct flow generated from the flow surface becomes an essential part of the storm runoff. Due to the effect of grid size, upstream channel cells may not be fully occupied by flow. Equation 3.4 is then used to calculate the potential runoff coefficient for these channel cells:

C r = RP + (1 RP )C

(3.4)

where Cr is the potential runoff coefficient for a channel grid, C is the potential runoff coefficient without considering the channel effect, and RP is the percentage of channel area of the grid calculated by the estimated flow width divided by the grid size. The flow width is determined by a power law relationship with an exceeding probability (Molnar & Ramirez, 1998), which relates flow width to the controlling area and is seen as a representation of the average behaviour of the cell and the channel geometry:

Wi = aW ( Ai ) W
b

(3.5)

where Ai is the drained area upstream of the cell (km), aW (-) is a network constant and bW (-) a geometry scaling exponent both depending on the flood frequency. Researches have shown that the runoff efficiency (volume of runoff per unit of area) increases with the decreasing catchment area, i.e. the larger the catchment area the smaller the runoff efficiency (Boers & Ben-Asher, 1982; Brown et al., 1999).

248

WetSpa Extension: Documentation and User Manual Analogously, the potential runoff coefficient is affected by the grid size, in which more surface runoff is produced when modelling with a small grid size, and vice versa. This can be explained by that spatial variability in climatic inputs such as rainfall and hydrometorological variables, in soil characteristics such as hydraulic conductivity and porosity, in topography, and land use, increase with spatial scale (Vijay & Woolhiser, 2002). For instance, the average saturated hydraulic conductivity and the surface retention capacity are higher when modelling in a coarser resolution, causing more infiltration and less surface runoff. These have been addressed in many of the literatures (Loague, 1988; Mazion & Yen, 1994; Saghafian et al., 1995). Therefore, the grid size should be chosen properly in order to adequately represent the spatial heterogeneity of a watershed, and the values of potential runoff coefficient are allowed to readjust during calibration.

3.1.4. Depression storage capacity Depression storage capacity is a value that is land use dependent and represents the total amount of water that can be stored in small surface depressions. Moreover, the soil type and the slope steepness also affect the depression storage capacity for ponding water and thereby the conditions for surface runoff. Generally rougher surfaces store more surface water than smoother surfaces and steeper slopes store less surface water than gentle slopes (Moore and Larson, 1979; Ullah and Dickinson, 1979a, b; Onstad, 1984). After the depression storage amount is met, runoff within a cell begins. A table of depression storage capacity, as shown in Table A-3.6, is built in WetSpa Extension for different land use, soil type and slope combinations, based on the analysis of data in ASCE (1969), SINCE (1972), Sheaffer et al., (1982), and Geiger et al. (1987). The depression storage capacity for impervious areas is considered as wetting loss, and set to 0.5 mm (Fronteau & Bauwens, 1995). In order to obtain a depression storage capacity as a function of a continuous slope used in the WetSpa Extension, a simple regression equation as in Hansen et al. (1999) is applied, in which the depression storage capacity is controlled by land use and soil type, and decreases with slope exponentially:

249

Appendix A

Sd = Sd0 exp(bS )

(3.5)

where Sd is the depression storage capacity (mm), S is the slope (%), Sd0 is the depression storage capacity for a near zero slope and different soil types (mm) corresponding to the values listed on the first row of each land use class in Table A3.6, and b = -9.5, calibrated using the data in Table A-3.6. Table A-3.6: Depression storage capacity for different land use, soil type and slope
Land Slope use (%) Forest <0.5 0.5-5 5-10 >10 Grass <0,5 0.5-5 5-10 >10 Crop <0.5 0.5-5 5-10 >10 Bare <0.5 soil 0.5-5 5-10 >10 IMP Sand Loamy Sandy Loam sand loam 8.00 7.50 7.00 6.50 6.31 5.91 5.52 5.13 3.92 3.68 3.43 3.19 1.92 1.80 1.68 1.56 5.00 4.73 4.45 4.18 3.94 3.73 3.51 3.30 2.45 2.32 2.18 2.05 1.20 1.14 1.07 1.01 3.00 2.86 2.73 2.59 2.37 2.26 2.15 2.04 1.47 1.40 1.34 1.27 0.72 0.69 0.66 0.62 1.50 1.45 1.41 1.36 1.12 1.09 1.05 1.02 0.74 0.72 0.70 0.67 0.36 0.35 0.34 0.33 0.50 0.50 0.50 0.50 Silt loam 6.00 4.73 2.94 1.44 3.91 3.08 1.92 0.94 2.45 1.94 1.20 0.59 1.32 0.99 0.65 0.32 0.50 Silt 5.50 4.34 2.70 1.32 3.64 2.87 1.78 0.87 2.32 1.83 1.14 0.56 1.27 0.95 0.63 0.31 0.50 Sandy clay loam 5.00 3.94 2.45 1.20 3.36 2.65 1.65 0.81 2.18 1.72 1.07 0.52 1.23 0.92 0.61 0.30 0.50 Clay loam 4.50 3.55 2.21 1.08 3.09 2.44 1.52 0.74 2.05 1.61 1.00 0.49 1.18 0.88 0.58 0.28 0.50 Silty clay loam 4.00 3.15 1.96 0.96 2.82 2.22 1.38 0.68 1.91 1.51 0.94 0.46 1.14 0.85 0.56 0.27 0.50 Sandy clay 3.50 2.76 1.72 0.84 2.55 2.01 1.25 0.61 1.77 1.40 0.87 0.43 1.09 0.81 0.54 0.26 0.50 Silty clay 3.00 2.37 1.47 0.72 2.27 1.79 1.11 0.55 1.64 1.29 0.80 0.39 1.05 0.78 0.52 0.25 0.50 Clay 2.50 1.97 1.23 0.60 2.00 1.58 0.98 0.48 1.50 1.18 0.74 0.36 1.00 0.75 0.49 0.24 0.50

Figure A-3.2 shows the depression storage capacity for a grass cover as a function of slope and different soil types. The left figure of Figure A-3.2 shows the depression storage capacity for a slope ranging from 0 to 20% and the supporting points, and the right one shows the depression storage capacity for a slope ranging from 0 to 100%. Clearly, the depression storage capacity approaches to Sd0 for a very small slope, and 0 for a steep slope. This conforms that the depression storage may have a significant effect for gentle slope, but is not important for a steep slope in controlling overland flow generation (Hansen et al., 1999).

250

WetSpa Extension: Documentation and User Manual

5.0 5.0

Depression storage c apacity (mm)

Depression storage capacity (mm)

4.0

4.0

3.0

3.0

2.0

2.0

1.0

1.0

Sand Loamy sand Sandy loam Silt loam Silt Loam Sandy clay loam Silt clay loam Clay loam Sandy clay Silt clay Clay

0.0 0 5 10 15 20

0.0 0 20 40 60 80 100

Slope (%)

Slope (%)

Figure A-3.2: Depression storage capacities vs. slope for grass and different soil types The computation of depression storage capacity for urban areas is the same like the process in calculating potential runoff coefficient, which is the weighted mean of the depression storage capacity for impervious area and grassland. The equation can be expressed as:

Sdu = 0.5IMP + (1 IMP) Sd grass

(3.6)

where Sdu and Sdgrass are the depression storage capacity for an urban and grass grid respectively (mm). As there is no depression loss on water surface, the depression storage capacity for a channel cell can be calculated as:

Sd r = (1 RP ) Sd

(3.7)

where Sdr (mm)is the depression storage capacity for a channel grid, and Sd (mm) is the depression storage capacity without considering the channel effect. The values of depression storage capacity are also affected by the grid size as discussed in section A-3.1.3. Therefore, cautions should be made with regards to use these values for a large grid. These parameters are allowed to modify during the GIS preprocessing in order to get a better fit.

251

Appendix A 3.2. Global parameters

For simplifying the process of parameter calibration, 12 global parameters are used in the WetSpa Extension, i.e. the correction factor of PET, interflow scaling factor, groundwater recession coefficient, initial soil moisture, initial groundwater storage, base temperature for snowmelt, temperature degree-day coefficient, rainfall degreeday coefficient, surface runoff exponent, and the rainfall intensity corresponding surface runoff exponent of 1. These parameters have physical interpretations and are important in controlling runoff production and hydrographs at basin outlet, but difficult to assign properly on a grid scale. Therefore, calibration of these global parameters against observed runoff data is preferable in addition to the adjustment of distributed model parameters. 1) Correction factor for PET The PET data used in the model are obtained from pan measurement or calculated by Pemman-Monteith or other equations using available weather data. These reference evapotranspiration rates refer to water surface or a grass cover in large fields. Actual reference or PET rates, however, may depend on local factors that are not addressed by these methods. For instance, the land use, elevation, as well as the micrometeorological conditions for the grid to be simulated may be different from those prevailing at the site of the meteorological station whose data are being used. To account for these effects, a correction factor is required in the computed PET. The correction factor is normally close to 1, and can be calibrated by the model through a long-term water balance simulation. Specifically, when modelling in a mountainous catchment, the evapotranspiration stations are usually very sparse and are located in the river valley. To account for the effect of elevation, the correction factor for PET may be much lower in this case.

2) Scaling factor for interflow computation Interflow or subsurface runoff is an essential runoff component for the humid temperate region especially for the areas with sloping landscapes and well-vegetated cover. In WetSpa Extension, interflow is assumed to occur when soil moisture exceeds the field capacity and there is sufficient hydraulic gradient to move the water. 252

WetSpa Extension: Documentation and User Manual Darcys law is then used for the simulation of interflow. Dingman (1994) pointed out that soil water preferentially flows laterally given greater lateral hydraulic conductivity than vertical due to the anisotropy of water content dependent hydraulic conductivity. Even though a uniform soil matrix is considered in the model, but in fact, the porosity and permeability of soil tend to decrease with depth given the weight of overlying soil and the translocation of material in percolating water to lateral subsurface flow. Moreover, soil water passing quickly to a stream through root canals, animal tunnels, or pipes produced by subsurface erosion may become a critical component of peak flow. To account for theses effects, a scaling factor for lateral hydraulic conductivity in computing interflow is used in the model. This scaling factor is generally greater than 1, and can be calibrated by comparing the recession part of computed flood hydrographs with the observed hydrographs.

3) Groundwater recession coefficient Groundwater flows are estimated on subcatchment scale in WetSpa Extension as described in Chapter A-2. The groundwater recession coefficient reflects the storage characteristics of the subwatershed and, therefore, is the same for all hydrographs at a given location. In accordance with Equation (2.21), the groundwater recession coefficient will remain constant if storage and discharge volumes are divided by area and expressed as depth in mm (Wittenberg, 1999). This is under the condition that groundwater flow for each subcatchment has the same recession constant, and total groundwater at the outlet of the river is only a time-shifted superposition of partial groundwater flow from each subcatchment.

In real river basins, baseflow recession coefficient for each subcatchment may not be the same, and may have a considerable deviation from the theoretical constant. A great portion of the deviation is associated with variability of subcatchment characteristics. Others may be attributed to aquifer heterogeneity and divergence from the Dupuit-Forchheimer assumption of essentially horizontal groundwater flow. For model simplification, a general value of groundwater flow recession coefficient is determined at the basin outlet in the input file. A linear correction is then performed for each subcatchment based on its drainage area and the average slope, for which

253

Appendix A higher values are assigned for the subcatchments with large drainage area and steep slope, and lower values for the subcatchments with small area and gentle slope. The shape and stream density of the subcatchment is not accounted for in the current version. The equation can be expressed as:

c g ,s = c g

Ss Ws S

(3.8)

where cg,s and cg (m2/s) are groundwater recession coefficient of the subcatchment and the entire basin, Ss and S are average slope of the subcatchment and the entire basin, and Ws is the areal weight of the subcatchment. cg can be derived by the analysis of flow records as described in Martin (1973) and Wittenberg (1999). Calibration of this parameter is necessary by comparing the computed and observed low flow hydrographs.

4) Initial soil moisture Soil moisture content is a key element in the model controlling the hydrological processes of surface runoff production, evapotranspiration, percolation and interflow. A proper initial soil moisture condition may provide a much more realistic starting point for predictions. However, for a long-term flow simulation in a watershed, the initial soil moisture condition is less important, as it affects the hydrological processes only in the initial part of the simulation. An assumption of uniform initial moisture distribution can be made in this case with modelling purpose of flood prediction under present condition. A ratio against field capacity is then defined in the input parameter file for setting up the initial soil moisture conditions. This value can be adjusted during calibration by analysis of water balance output and comparison between the computed and observed hydrographs for the initial phase.

If the model is used for short-term flow simulation or event-based flood prediction, the antecedent moisture condition becomes one of the most important factors in runoff production as well as its distribution. The concept of topographic wetness index (TWI) adapted from Moore et al. (1993) can be introduced in the model to evaluate 254

WetSpa Extension: Documentation and User Manual antecedent moisture condition of a watershed with TWI = ln(A/S), where ln(.) is the natural logarithm, A is the upslope drainage area (m2), and S is the local slope (-). The TWI distribution can be easily obtained from a high resolution DEM. Those cells with high TWI values have larger upslope contributing areas or smaller cell slopes or a combination of the two properties that lead to accumulation of soil moisture. While an assumption is made for maximum and minimum moisture content within the watershed, the antecedent moisture distribution can be obtained by simply relating moisture content to the TWI values. Cells with very high TWI values may consider to be saturated with runoff coefficient of one. These cells are normally distributed along the main river or in the depression areas in a watershed.

5) Initial groundwater storage In WetSpa Extension, groundwater balance is maintained on subcatchment scale and for the active groundwater storage, which is that part of storage in perched or shallow aquifers that contribute to the surface stream flow. Water percolating from the root zone storage may flow to active groundwater storage or may be lost by deep percolation. Active groundwater eventually reappears as baseflow, but deep percolation is considered lost from the simulated system. A value of initial groundwater storage in depth (mm) is set up in the input parameter file for all subcatchment. This value can be adjusted during calibration by comparing the computed and observed low flows for the initial phase.

6) Base temperature for snowmelt The precipitation is assumed to fall as snow if the temperature is below the base temperature. Snowmelt starts when the temperature is above the base temperature. The base temperature is typically a value near 0C, particularly for short computation period using average temperature as input. The user may specify this value during model calibration.

7) Temperature degree-day coefficient The range of the temperature degree-day coefficient is typically 1.8 3.7 mm/C/day for rain-free conditions (Anderson, 1973; Male and Gray, 1981). This value can be 255

Appendix A determined by comparison between computed and observed spring flood hydrographs during calibration. In general, the temperature degree-day coefficient is varied both in time and space. For instance, the albedo is very high for new, cold snow falling in the beginning of the accumulation season and decreases with the age of the snow, which results in an increase of the degree-day coefficient. Moreover, the temperature degreeday coefficient is also land use dependent, for which forest cover leads to a smaller value, while bare soil leads to a higher value. For simplicity purpose, these influencing factors are not accounted for in the current model, and recommended to be coupled in the future version.

8) Rainfall degree-day coefficient The rainfall degree-day coefficient determines the rate of snow melting caused by condensation of humid air on the snow surface and the advective heat transferred to the snow pack by precipitation, and is used for calculation of an additional snowmelt due to rainfall. The value of rainfall degree-day coefficient is generally very small, typically around 0.01 (mm/mm/C/day), and can be determined during model calibration. If zero value is given, the effect of rainfall on snowmelt is not considered.

9) Surface runoff exponent for a near zero rainfall intensity Rainfall intensity has a big influence in controlling the proportion of surface runoff and infiltration. As pointed by Dunne (1991), infiltration rate increases with rainfall intensity for two reasons: (1) Higher rainfall intensity tends to exceed the saturated hydraulic conductivity of larger proportions of the soil surface, and thereby to raise the spatially averaged hydraulic conductivity, and (2) Higher rainfall intensity gives more surface runoff rate and the inundated flow depth. To account for this effect on the production of surface runoff, an empirical exponent is introduced in the model as described in Eq. (2.7). The concept is that the proportion of surface runoff is very small, or even nil, under the condition of very small rainfall intensity, and the proportion increases along with the increase of rainfall intensity up to a stage for which a potential runoff coefficient is achieved. In WetSpa Extension, this exponent is assumed to be a variable starting from a higher value for a near zero rainfall intensity, and changing linearly up to 1 along with the rainfall intensity, when the

256

WetSpa Extension: Documentation and User Manual predetermined maximum rainfall intensity is reached. This value is generally less than 3 according to the previous applications. If an exponent value 1 is given, the actual runoff coefficient is then a linear function of the relative soil moisture content, and the effect of rainfall intensity on the runoff coefficient is not taken into account.

10) Rainfall intensity corresponding to a surface runoff exponent of 1 This parameter corresponds to a threshold rainfall intensity in unit of mm/h or mm/d depending upon the temporal resolution of the model simulation, over which the surface runoff exponent equals 1, and the actual runoff coefficient becomes a linear function of the relative soil moisture content. Calibration of this parameter can be performed by comparison of the observed and computed surface runoff volume and the peak discharge for high floods. This parameter is in fact spatially distributed, depending upon the cell characteristics, such as soil type, land use, and slope, etc. A constant value is assumed in the current model for simplification.

3.3. Model evaluation In order to evaluate how well WetSpa Extension reproduces an observed hydrograph, a series of statistics are used. In addition to the evaluation based on a visual comparison and an evaluation of peak flow rate and time to the peak, the bias, model confidence, and the model efficiency are also taken into account. These statistical measures provide quantitative estimates for the goodness of fit between observed and predicted values, and are used as indicators of the extent at which model predictions match observation. Based on the results of these tests, model predictive capabilities are assessed. The goodness of fit in the peak discharge and time to the peak can be evaluated by their relative and absolute errors respectively, while other evaluation criteria are described as following: 1) Model bias Model bias can be expressed as the relative mean difference between predicted and observed stream flows for a sufficiently large simulation sample, reflecting the ability

257

Appendix A of reproducing water balance, and perhaps the most important criterion for comparing whether a model is working well in practice. The criterion is given by the equation

CR1 =

(Qs
i =1 N i =1

Qoi )
i

(3.9)

Qo

where CR1 is the model bias, Qsi and Qoi are the simulated and observed stream flows at time step i (m3/s), and N is the number of time steps over the simulation period. Model bias measures the systematic under or over prediction for a set of predictions. A lower CR1 value indicates a better fit, and the value 0.0 represents the perfect simulation of observed flow volume. 2) Model confidence Model confidence is one of the important criteria in assessment of continuous model simulation, and can be expressed by its determination coefficient, which is calculated as the portion of the sum of the squares of the deviations of the simulated and observed discharges from the average observed discharge.

CR 2 =

(Qs
N i =1 N

Qo Qo

(3.10)

(Qo
i =1

where CR2 is the model determination coefficient,

Qo

is the mean observed stream

flow over the simulation period. CR2 represents the proportion of the variance in the observed discharges that are explained by the simulated discharges. It varies between 0 and 1, with a value close to 1 indicating a high level of model confidence. 3) Nash-Sutcliffe efficiency The Nash-Sutcliffe coefficient (Nash and Sutcliffe, 1970) describes how well the stream flows are simulated by the model. As pointed out by Kachroo and Natale (1992), this efficiency criterion is commonly used for model evaluation, because it involves standardization of the residual variance, and its expected value does not

258

WetSpa Extension: Documentation and User Manual change with the length of the record or the scale of runoff. The equation can be described as
2

CR 3 = 1

(Qs
i =1 N i =1

Qo i )
i

(3.11)

(Qo

Qo

where CR3 is the Nash-Sutcliffe efficiency used for evaluating the ability of reproducing the time evolution of stream flows. The CR3 value can range from a negative value to 1, with 1 indicating a perfect fit between the simulated and observed hydrographs. CR3 below zero indicates that average measured stream flow would have been as good a predictor as the modelled stream flow. A perfect model prediction has CR3 score equal to 1. 4) Logarithmic version of Nash-Sutcliffe efficiency for low flow evaluation A logarithmic transformed Nash-Sutcliffe criterion is presented in Equation 3.11, which gives emphasize for evaluating the quality of low flow simulations (Smakhtin et al., 1998).

CR 4 = 1

[ln (Qs
i =1 N i =1

+ ) ln (Qo i + )]
i

[ln (Qo

+ ) ln Qo +

)]

(3.12)

where CR4 is a logarithmic Nash-Sutcliffe efficiency for evaluating the ability of reproducing the time evolution of low flows, and is an arbitrary chosen small value introduced to avoid problems with nil observed or simulated discharges. The value of should be sufficiently low, and those observed discharges lower than value are negligible. Otherwise the CR3 criterion would present a bias. Similar as CR3, a perfect value of CR4 is 1. 5) Adapted version of Nash-Sutcliffe efficiency for high flow evaluation An adapted version of the Nash-Sutcliffe criterion is proposed as in Equation 3.12. It is in fact a combination between the calibration criteria used by Guex (2001) for the

259

Appendix A hydrological study on the Alzette river basin and the HEC-1 objective function (USACE, 1998).

CR 5 = 1

(Qo
N i =1 N i =1

+ Qo (Qs i Qo i )
i

(Qo

+ Qo Qo i Qo

)(

(3.13)

where CR5 is an adapted version of Nash-Sutcliffe criterion for evaluating the ability of reproducing the time evolution of high flows. As can be seen in the formula, more weight is given on high discharges than low ones. A perfect value of CR5 is 1.

4. Model operation

4.1. Program installation Installation of WetSpa requires a Windows 98/ME/2000/XP or Windows NT 4.0 operating system. Also required are licensed versions of ESRIs ArcView 3.2 GIS Application and Spatial Analyst v2.0 Extension. In addition, the software of Visual FORTRAN 6.1 or other FORTRAN compilers are required if the user wants to edit and modify the program source code. The minimum drive space required is 100MB. Additional space may be necessary depending on the spatial and temporal scale of the project. By simple copy and paste operation, the model can be installed and run on any computer drives and under any existing directories. Specific folders are referenced from that drive location throughout the modelling process. Figure A-4.1 gives a schematic view of the models project folders.

260

WetSpa Extension: Documentation and User Manual

Project Document DEM Soil type Land use Coverage ArcView Ascii Data Help Script Table Temp Project.apr Model Input Output PEST Program Source

Figure A-4.1: Schematic view of the models project folders Where Project is the general folder of the modelling project, and the others are: 1) Document: for storing model documents 2) ArcView: for storing ArcView GIS components 3) Ascii: for storing spatial parameter maps in ASCII format 4) Data: for storing spatial data of base maps 5) Help: for storing model help files 6) Script: for storing ArcView Avenue scripts 7) Table: for storing model lookup tables. 8) Temp: project working directory for storing intermediate and temporary files 9) Project.apr: ArcView project of the model 10) DEM: digital elevation model 11) Soil type: digital soil type map in grid format 12) Land use: digital land use map in grid format 13) Coverage: for storing coverage data including stations, streams, boundaries, etc. 14) Model: for storing model inputs, outputs and programs 15) Input: for storing model input files 16) Output: for storing model output files 17) PEST: for storing programs of automatic parameter estimation 18) Program: for storing model executive programs 19) Source: for storing program source codes

261

Appendix A

4.2. Program description

4.2.1. Avenue scripts and their tasks 1) conductivity: creates a grid of saturated hydraulic conductivity 2) delta_h: calculates standard deviation of flow time from cells to the basin outlet 3) delta_s: calculates standard deviation of flow time from cells to the main river 4) depression: calculates depression storage capacity for each cell 5) fieldcapacity: creates a moisture grid at soil field capacity 6) fillsink: fill sinks to remove small imperfections from DEM 7) flowacc: creates an accumulated flow grid at each cell 8) flowdir: creates a flow direction grid from each cell to its steepest downslope neighbour 9) flowlen: calculates a downstream distance grid along its flow path 10) interception: calculates minimum and maximum interception storage capacity 11) lai: creates a grid of leaf area index 12) manning: calculates Mannings roughness coefficient for each cell 13) mask: creates a mask grid of the watershed 14) moisture: creates an initial soil moisture grid based on the topographic index 15) poreindex: creates a grid of soil pore size distribution index 16) porosity: creates a moisture grid at soil porosity 17) radius: calculates hydraulic radius for each cell according to flood frequency 18) residual: creates a moisture grid at residual soil moisture content 19) rootdepth: creates a grid of root depth 20) runoffco: creates a grid of potential runoff coefficient 21) slope: creates a slope grid for both land surface and river channel 22) curvature: creates a curvature grid and its root mean square profile curvature 23) streamlink: assigns unique values to sections of stream network 24) streamnet: creates a grid of stream network 25) streamorder: assigns a numeric order to branches of a river network 26) streamtoline: converts stream grid to a line coverage 27) t0_h: calculates flow time from each cell to the basin outlet 28) t0_s: calculates flow time from each cell to the main river 262

WetSpa Extension: Documentation and User Manual 29) thiessen: creates a grid of Thiessen polygons 30) velocity: creates a velocity grid for both overland flow and channel flow 31) v_fraction: creates a grid of maximum fractional vegetation cover 32) watershed: determines subwatersheds based on stream links 33) wiltingpoint: creates a moisture grid at permanent wilting point

4.2.2. Lookup tables 1) depression.dbf: default values of depression storage capacity for different land use, soil texture, and near zero slopes 2) landuse_reclass.dbf: land use reclassification table for deriving potential runoff coefficient and depression storage capacity of the 5 main land use classes 3) landuse_remap.dbf: default model parameters based on land use classes, including root depth, Mannings roughness coefficient, interception capacity, vegetated fraction and leaf area index 4) radius: default parameters governing average hydraulic radius for a certain flood frequency 5) runoff_coefficient.dbf: default potential runoff coefficient for different land use, soil texture, and near zero slopes 6) soil_remap: default parameters based on soil texture categories, including hydraulic conductivity, porosity, field capacity, wilting point, residual moisture, pore size distribution index, etc.

4.2.3. FORTRAN programs and their tasks 1) mean: calculates mean parameters of each subcatchment 2) iuh: calculates the unit response function of each cell to the catchment and subcatchment outlet, the unit hydrograph of each subcatchment to the catchment and subcatchment outlet, and the unit hydrographs of main rivers. 3) model1: semi-distributed model on subcatchment scale 4) model2: fully distributed model on cell scale 5) water_balance: calculates water balance on grid scale without flow routing 6) evaluation: statistics of simulation results and model evaluation

263

Appendix A 4.2.4. PEST files and their tasks 1) input: PEST input template file 2) qtotal: PEST output instruction file 3) wetspa: PEST control file 4) pest: PEST executable file 5) pest_wetsap: PEST MS-DOS batch file

4.3. GIS pre-processing The purpose of GIS pre-processing is to create all necessary spatial parameter maps used in the WetSpa Extension. Open a new ArcView project project (or other name) under the subdirectory \project\arcview. Set the projects working directory to \project\arcview\temp, in which the intermediate and temporary GIS files are stored and all other input and output files are transferred from or to their subdirectory referencing to this path. Before performing GIS pre-processing, be sure that the ArcView Extensions: Spatial Analyst, GeoProcessing, WetSpa and Create Thiessen Polygons, are added to the ArcView project. Next, Load grid themes of elevation, landuse and soiltype from the subdirectory \project\arcview\data to the View Topography, Landuse and Soiltype separately. Set the theme names as Elevation, Landuse and Soil. Note that the extent of these three base maps must be the same in order to perform the model simulation properly.

4.3.1. Surface grid preparation Surface parameter grids based on a DEM are prepared in the view Topography of the ArcView project. The preparation of a proper DEM employs many geo-processing schemes, and can be implemented independently from the project using more powerful GIS software, such as ArcInfo etc. From the available DEM, its hydrological potential is calculated in ArcView by performing the following functions: filling sinks, determining flow direction and flow accumulation, assigning stream network, stream link and stream order, calculating slope and hydraulic radius, and delineating subcatchments, etc. Figure A-4.2 gives a screenshort of the surface grid menu. 264

WetSpa Extension: Documentation and User Manual

Figure A-4.2: Screenshort of surface menu 1) Fill Sinks A sink is a cell or set of spatially connected cells whose flow direction cannot be assigned one of the eight valid values in a flow direction Grid. This can occur when all neighbouring cells are higher than the processing cell. In ArcView GIS, sinks are considered to have undefined flow directions and are assigned a value that is the sum of their possible directions. To create an accurate representation of flow direction and therefore accumulated flow, it is required to use a data set that is free of sinks. The fill sinks request in the surface menu takes a grid theme Elevation and fills all sinks and areas of internal drainage contained within it. The process of filling sinks can create new sinks, so a looping process is used until all sinks are filled (ESRI, 1999). The output theme is named as Filled Elevation displayed in the same view, and the corresponding ASCII file elevation.asc is saved in the subdirectory /project/arcview/ascii used for estimation of altitude-distributed temperature. 2) Mask A mask grid defines the study region in the grid domain, which can be used to extract catchment boundary, determine the extent of other grids, etc. The request takes the grid theme Filled Elevation and assigns a unique value 1 for the cells within the study catchment with output theme Mask displayed in the same view.

265

Appendix A 3) Flow direction The flow direction request calculates the direction of flow out of each cell into one of its eight neighbours. The direction of flow is determined by finding the direction of steepest descent from each cell. If a cell is lower than its 8 neighbours that cell is given the value of its lowest neighbour and flow is defined towards this cell. If the descent to all adjacent cells is the same, the neighbourhood is enlarged until the steepest descent is found (ESRI, 1999). The request takes the grid theme Filled Elevation and calculates flow direction for each cell with output theme Flow Direction displayed in the same view. 4) Flow accumulation The flow accumulation request creates a grid of accumulated flow to each cell by accumulating the weight for all cells that flow into each downslope cell. The accumulated flow is based upon the number of cells flowing into each cell in the output grid. Output cells with a high flow accumulation are areas of concentrated flow, and therefore can be used to identify stream channels. Output cells with a flow accumulation of zero are local topographic highs and can be used to identify ridges. The request takes the grid theme Flow Direction and calculates flow accumulation for each cell with output Flow Accumulation displayed in the same view. 5) Stream network The results of the flow accumulation are used to create a vector stream network by applying a threshold value to subset cells with a high-accumulated flow. All cells that have more than a user-defined number of cells flowing into them are assigned a value of one; all other cells are assigned no data. The resulting stream network can be used as a predicted hydrography (ESRI, 1999). The stream network request takes the grid theme Flow Accumulation and delineates a stream network grid Stream Network displayed in the same view. 6) Stream link Links are the sections of a stream channel connecting two successive junctions, a junction and the outlet, or a junction and the drainage divide (ESRI, 1999). The stream link request takes the grid themes Flow Direction and Stream Network, and assigns unique values to sections of a stream network between intersections. The 266

WetSpa Extension: Documentation and User Manual output theme is named as Stream Link displayed in the same view, which can be used as the source grid to create drainage basins that correspond the branches of a stream network. Meanwhile, the output grid data is written to an ASCII file link.asc used to calculate IUH of stream channels. 7) Stream order The stream order request takes the grid themes Flow Direction and Stream Network, and assigns a numeric order to segments of the stream network. The Shreve method is used in the model, in which all links with no tributaries are assigned an order of 1 and the orders are additive downslope. When two links intersect, their magnitudes are added and assigned to the downslope link. The output theme is named as Stream Order displayed in the same view, and used as a source grid in assigning Mannings n for stream channels. 8) Slope The process of slope derivation calculates the rate of maximum change for locations on the elevation grid theme and creates a new grid theme Slope as output. Each cell in the output theme contains a continuous slope value represented as a percentage. Considering that the stream network is in a vector style, and its slope is determined by the elevation difference and distance between the up and down cells along the streamline, the channel slope is calculated separately from the general slope using DEM and the stream network information. This avoids the disturbance in channel slopes for a river, especially for stream channels with asymmetric side slopes of the riverbank. The final slope grid is then obtained using the general slope grid overlaid by the grid of channel slope. An ASCII file slope.asc is saved in the subdirectory /project/arcview/ascii for use in calculating interflow from each cell. 9) Hydraulic radius The hydraulic radius request takes the grid theme Flow accumulation, and calculates hydraulic radius for each grid cell. The hydraulic radius is determined by a power law relationship with an exceeding probability, which relates hydraulic radius to the controlling area and is seen as a representation of the average behaviour of the cell and the channel geometry. Generally, a flood frequency with 2-year return period is chosen for normal floods. The two controlling parameters can be adjusted in the 267

Appendix A lookup table radius.dbf to meet the specific characteristics of catchment. The output grid theme is named as Radius (m), and is used for calculation of flow velocity. 10) Watershed The watershed request takes the grid themes Flow Direction and Stream Link, and determines the subcatchment for each stream link. The output grid theme is named as Watershed displayed in the same view, and is saved as an ASCII file for semidistributed modelling and the simulation of groundwater balance. If the subcatchment does not delineate as expected, delete the grid themes Stream Network, Stream Link and Watershed by invoking the delete theme command in the edit dropdown menu, and rebuild the three grid themes by setting a new threshold value. Often it is necessary to closely zoom into the area of interest to ensure the outlet points location is positioned correctly.

4.3.2. Soil based grid preparation To calculate the soil hydraulic properties, activate the view Soiltype, select the Parameter dropdown menu, and the commands related to soil types are highlighted (Figure A-4.3), including Conductivity, Porosity, Field capacity, Residual moisture, Pore distribution index, and wilting point, etc. The commands Maximum saturation, Arithmetic mean of G and Geometric mean of G are designed for future model improvement, where G is the capillary drive (mm). By clicking each of the command, the grid files are created to redefine and display the soil units with respect to their hydrological properties, and the corresponding ASCII files are saved in the subdirectory /project/arcview/ascii.

268

WetSpa Extension: Documentation and User Manual

Figure A-4.3: Screenshort of parameter menu

Another activated function under the dropdown menu is the Initial moisture. This function creates an initial relative saturation grid of the soil using the method of the Topographical Wetness Index. A minimum ratio reflecting the moisture condition of the driest cells is asked in a pop up window, which can be selected from the provided list. The output theme is named as Initial Moisture displayed in the same view, and the ASCII file moisture.asc is saved in the subdirectory /project/arcview/ascii. Note that this operation is optional and designed for event based flood modelling, for which the initial soil moisture condition is rather important.

4.3.3. Land use based grid preparation To calculate the land use dependent model parameters, activate the view Landuse, select the Parameter dropdown menu, and the commands related to land use grid are highlighted, including Root depth, Vegetated fraction, Interception capacity, Mannings coefficient, and Leaf area index. There are two interception capacity themes created by the command Interception capacity, e.g. maximum and minimum interception capacity, corresponding to the summer and winter situation. The command Mannings coefficient creates a map of Mannings roughness coefficient for both hillslope and river channels. Therefore, the theme Stream order needs to be

269

Appendix A created firstly in the view Topography. A selection list is shown in the window asking for a Mannings n interpolation method for the stream channels. 1) Interpolation among different stream orders, for which the channel Mannings n is defined based on the stream orders with lower values downstream and higher value upstream. A maximum and a minimum Mannings n value are asked to determine corresponding to the lowest and highest stream order. 2) Remain the default constant as in the lookup table, for which a constant Mannings n is defined for the river channels using the value assigned in the lookup table. 3) Change to another constant, for which a modified constant Mannings n is defined for the river channels. The command Leaf area index is designed for future model improvement. By clicking each of the commands, the grid files are created to redefine and display the land use units with respect to their hydrological properties based on the predefined lookup table, and the corresponding ASCII parameter files are saved in the subdirectory /project/arcview/ascii.

4.3.4. Potential runoff coefficient and depression storage capacity Next, the parameter maps of potential runoff coefficient and depression storage capacity are generated in the view Runoff coefficient & depression. Since both parameter maps are functions of slope, soil type and land use, these three base maps need to be created firstly in their views. The program can load these three grid themes directly from their views, and the parameter grids are created and displayed in a separate view Runoff coefficient & depression in order to give a clear view of them. By activating the view Runoff coefficient & depression, selecting the Parameter dropdown menu, the commands Runoff coefficient and Depression capacity will be highlighted. An impervious percentage for urban cells is asked when calculating the grid of potential runoff coefficient. A default value 30% is given for a grid with cell size 100100 m. By clicking each of the commands, the resulting grid files are 270

WetSpa Extension: Documentation and User Manual created and displayed in the view, and the corresponding ASCII parameter files are saved in the subdirectory /project/arcview/ascii.

4.3.5. Flow routing parameters The flow routing parameter grids are calculated in the view Routing Parameter, including flow velocity, mean flow times to the basin outlet and to the main river from each cell, and the standard deviations of the flow times. These parameter maps are used for calculating flow response functions from each cell to the basin outlet as well as to the main river. By activating the view Routing Parameter and selecting the Parameter dropdown menu, the commands Velocity, T0_h, Delta_h, T0_s and Delta_s will be highlighted. By clicking each of the commands, the resulting grid files are created and displayed in the view, and the corresponding ASCII parameter files are saved in the subdirectory /project/arcview/ascii. 1) Run the script Velocity from the menu Parameter. This function creates a flow velocity grid based on the Mannings n, hydraulic radius and slope grid. A popup window shows and asks you if a flow velocity limit is necessary. The flow velocity is set to the upper limit when the calculated velocity is higher than the upper limit and to the lower limit vice versa. The upper and lower limits 3.0 m/s and 0.005 m/s are given by default. 2) Run the script T0_h from the menu Parameter. This function creates a flow travel time grid in hours from each cell to the catchment outlet using the weighted FLOWLENGTH routine. The ASCII file t0_h.asc is saved in the subdirectory /project/arcview/ascii. 3) Run the script Delta_h from the menu Parameter. This function creates a standard deviation grid of flow times in hours from each cell to the catchment outlet using the weighted FLOWLENGTH routine. The ASCII file delta_h.asc is saved in the subdirectory /project/arcview/ascii.

271

Appendix A 4) Run the script T0_s from the menu Parameter. This function creates a flow travel time grid in hours from each cell to its subcatchment outlet. The ASCII file t0_s.asc is saved in the subdirectory /project/arcview/ascii. 5) Run the script Delta_s from the menu Parameter. This function creates a standard deviation grid of flow times in hours from each cell to its subcatchment outlet. The ASCII file delta_s.asc is saved in the subdirectory /project/arcview/ascii.

4.3.6. Thiessen polygon Rainfall and PET data used in WetSpa Extension are tabular data gathered from point measuring stations inside or surrounding the catchment. In order to obtain a more accurate estimate of rainfall and PET values for a grid or a working unit, the Thiessen Polygon extension in ArcView is executed together with the themes of weather stations and the catchment boundary. This involves creating a Thiessen polygon theme in ArcView for all stations, then identifying each grid with the covering station identity number. The steps for creation of Thiessen polygon of rainfall data as well as its grid and ASCII file are: 1) To begin this process, three themes, rainfall stations, catchment boundary and a mask grid, need to be loaded into the View Thiessen Polygon, from which all others themes can be created. The rainfall station theme is obtained from a point shape file named as stations, which contains the fields of latitude, longitude, station name and station ID. The boundary shape file is obtained by conversion of a mask grid map to a polygon shape file. 2) Activate the theme Stations by clicking on the name in the Views theme list. Then, run the avenue script by clicking the command Thiessen polygon in the dropdown menu Surface. Select ID when prompted to "Select point field for polygon link ID", and select Boundary when prompted to Select polygon theme for boundary. Define the name of the output file as thiessen.shp in the subdirectory /project/arcview/data. The Thiessen polygon coverage theme is then displayed in the view after the execution. 272

WetSpa Extension: Documentation and User Manual

3) If it is wanted to convert the Thiessen polygon from coverage to grid, click Yes when asked Covert the Thiessen.shp to grid Thiessen?. Define the grid name as thiessen in the subdirectory /project/arcview/data. Set the output grid cell size, number of rows and number of columns the same as the mask map, and pick the field ID for cell values. A gird named Thiessen will be displayed in the view, after clicking Yes when promoted to Add grid as theme to the view. 4) Click Yes when promoted to Save the Thiessen polygon grid as Ascii file, the ASCII file Thiessen_p.asc is stored in the subdirectory /project/arcview/ascii. Following the same procedures, the Thiessen polygon grid for PET and temperature can be created using the point theme of PET and temperature stations instead of rainfall stations. The corresponding ASCII fill is named as Thiessen_e.asc and Thiessen_t.asc stored in the subdirectory /project/arcview/ascii. Note that there must be at least 2 stations in the point theme for performing the Thiessen polygon command. If only one station exists, the Thiessen polygon grid is just the same as the musk grid with cell values of station ID number.

4.3.7. Drainage systems for a complex terrain In case the WetSpa Extension is used for modelling a complex terrain, e.g. an urban or suburban watershed, on a small catchment scale, the sewer systems, communication lines, and artificial canals, lakes, reservoirs, etc., are important elements in drainage structure configuration, and govern flow direction more strongly than the derived aspect at local scale. Surface flow on these areas should thus be described with more detailed methodology, which allows a correct physical representation of the flow regime. Since most of these barriers are not sufficient to be represented in a DEM, additional procedures in term of deriving more realistic flow direction map are performed using GIS overlaying technique in the model. The procedures are:

273

Appendix A 1) Compute a general flow direction grid using the elevation grid alone without considering the effect of artificial areas, from which a stream network grid is generated. 2) Compute flow direction maps independently for sewer areas, main communication lines, artificial canals, and the stream network derived from the general flow direction grid, etc., based on the DEM and the available line and polygon themes. 3) Overlay the general flow direction map by the flow direction maps of sewer areas, communication lines, artificial canals, and the stream network subsequently, which allowing water to drain from the sewer areas at their outlets and water to cross communication lines and canals at their concave points to join the river. 4) The drainage paths delineated from the DEM are compared with existing hardcopy maps. Make any necessary corrections to the generated flow direction map in order to have the river reaches flow where they should and to be able to estimate a flow length closer to reality, particularly for the areas close to the catchment boundary, lakes, reservoirs and the meandering channel reaches. As an option, the above procedures can be integrated by modifying the elevation grid using ArcView GIS tools, in which the elevation of sewer areas, communication lines, and stream networks are lowered subsequently, e.g. 0.2, 0.4, 0.6 m. Similar flow direction grid can be obtained based on the modified elevation grid, but cautions should be made when performs this method to an even more complex terrain. The derived flow direction map is then used for further drainage structure delineation. The above procedures can be omitted, if the effects of human infrastructures are not remarkable to the flow regime in the catchment.

4.4. Creation of input files

4.4.1. Input files of time series WetSpa Extension reads input data from four input files. The names of these files are fixed during data preparation, namely p.txt, pet.txt, t.txt and q.txt. All files are in a 274

WetSpa Extension: Documentation and User Manual text format and stored in the subdirectory /project/model/input. All data are of unformatted statements, so that the exact position of each entry is not crucial. However, there must be at least one space or a comma between entries and data must be entered for each item. 1) Precipitation series The input precipitation series are in the format of year, month, day, hour, and followed by the precipitation values in mm at each gauging station. The first row of the file is year, month, day and hour, followed by the elevations of each precipitation station (m) for use in potential topographic precipitation interpolation. The precipitation series must be in an ascending order corresponding to the ID number in the precipitation Thiessen polygons. If the model runs on a daily scale, set the hour value zero. Table A-4.1 gives a sample file of precipitation series on hourly scale. Table A-4.1: Sample file of precipitation series p.txt
year 1998 1998 1998 1998 1998 1998 1998 1998 Month 10 10 10 10 10 10 10 10 day 23 23 23 23 23 23 23 23 hour 16 17 18 19 20 21 22 23 904 1.406 2.018 0.966 1.054 0.352 9.656 0.264 0.528 570 1.4 2.01 0.963 1.05 0.35 9.618 0.263 0.525 473 1.38 1.98 0.95 1.03 0.34 9.48 0.26 0.52 312 1.36 1.92 0.93 1.01 0.33 9.51 0.25 0.51

2) Potential evapotranspiration series The file pet.txt contains PET data in mm for all evaporation stations used in the model simulation. This input file is omitted if other PET calculation method is selected instead of using measured data. The format of pet.txt file is the same as the precipitation series. The first row of the file is year, month, day and hour, followed by the elevations of each evaporation station (m) for use in potential topographic evapotranspiration interpolation. The data series must be in an ascending order corresponding to the ID number in the evapotranspiration Thiessen polygons. If the model runs on a daily scale, put a zero value in the hours column. Table A-4.2 gives a sample file of PET series on hourly scale.

275

Appendix A Table A-4.2: Sample file of PET series pet.txt

Year 1998 1998 1998 1998 1998 1998 1998 1998 Month 10 10 10 10 10 10 10 10 Day 23 23 23 23 23 23 23 23 hour 16 17 18 19 20 21 22 23 901 0.05 0.048 0.048 0.05 0.05 0.05 0.04 0.04 380 0.05 0.047 0.047 0.05 0.05 0.05 0.038 0.038 270 0.048 0.043 0.043 0.048 0.048 0.048 0.036 0.036

3) Temperature series Temperature data is optional, used only when snow accumulation and snowmelt occur in the study catchment. The first row of the file is year, month, day and hour, followed by the elevations of each temperature station (m). The format of the rest of the file is the same as that in the precipitation series with temperature unit of C. If the model runs on a daily scale, set the hour value zero as shown in Table A-4.3. Note that the temperature stations should be listed in a continuously ascending order and corresponding to the station numbers in the temperature Thiessen polygons. Table A-4.3: Sample file of temperature series t.txt
Year 1991 1991 1991 1991 1991 1991 1991 1991 1991 month 1 1 1 1 1 1 1 1 1 Day 10 11 12 13 14 15 16 17 18 hour 0 0 0 0 0 0 0 0 0 295 2.5 4.7 3.5 2 -3.5 -4 -3.4 -5.3 -1.8 141 4 5.9 4.8 4.2 -1.7 -2.7 -3.6 -5.1 -3.1 702 4.4 2 2.9 -0.3 -6.2 -7.6 -7.5 -6.7 -3.8

4) Discharge series The observed discharge series are optional, used only for graphical comparison of the model outputs and statistical analysis for model evaluation. The format of the discharge file is the same as the precipitation file with values in m3/s. Set the hour value zero if the model runs on a daily scale. Table A-4.4 gives a sample file of discharge series on hourly scale.

276

WetSpa Extension: Documentation and User Manual

Table A-4.4: Sample file of discharge series q.txt

Year 1998 1998 1998 1998 1998 1998 1998 1998 month 10 10 10 10 10 10 10 10 Day 23 23 23 23 23 23 23 23 hour 16 17 18 19 20 21 22 23 q1 1.664 1.664 1.664 1.719 1.794 2.255 2.558 3.026 q2 1.784 1.829 1.93 2.031 2.069 2.713 3.092 3.905 q3 0.946 1.015 1.056 1.132 1.225 1.529 2.481 4.39

4.4.2. Global parameters and spatial output specifications 1) Global model parameters Before running WetSpa model, several global model parameters must be prepared, which are applied to each grid cell or each subcatchment. The file is named as input.txt and stored in the subdirectory /project/model/input. Table A-4.5 illustrates a template of global parameters in the input file input.txt. Table A-4.5: Template of global model parameters
dt (h) 24 Ci Cg K_ss 2.0 0.01 0.95

K_ep G0 G_max T0 K_snow K_rain K_run P_max 1.00 250.0 300.0 0.0 2.0 0.00 2.5 50.0

Where dt is the time step of the model (h), for which the value in the second row of the table can be any hours, e.g. 1 for hourly scale and 24 for daily scale. Ci is an interflow scaling factor reflecting the effect of organic material and root systems in the topsoil layer on horizontal hydraulic conductivity. Cg is a groundwater flow recession coefficient reflecting the groundwater recession regime for entire catchment. K_ss is a soil moisture ration relative to the field capacity for setting up the initial soil moisture content. This gives a uniform distribution of initial relative moisture condition and can be used for model simulation with a long time series. For performing an event based flood simulation, the initial moisture grid by the method of TWI can be applied. To do so, a negative value of K_ss should be given in the file, for instance, -1.0. K_ep is a correction factor for PET. G0 is the initial groundwater 277

Appendix A storage in depth (mm). G_max is the maximum groundwater storage in depth (mm). T0 is a base temperature (C) for estimating snowmelt, in which the precipitation shifts from rain to snow at T0. K_snow is a temperature degree-day coefficient (mm/C/day) for calculating snowmelt. K_rain is a rainfall degree-day coefficient (mm/mm/C/day) determining the rate of snowmelt caused by rainfall. Note that if there is no snow accumulation occurred in the study catchment, the parameters T0, K_snow and K_rain are set to negative values, e.g. 1.0, and the temperature input dataset t.txt is not necessary. K_run is an exponent reflecting the effect of rainfall intensity on the actual surface runoff coefficient when the rainfall intensity is very small. P_max is a threshold of rainfall intensity in mm/day or mm/hour depending on the modelling time step, over which the value of K_run is set to 1. 2) Location and time specifications for spatial output In order to obtain flow hydrographs at some specified subcatchment outlets, as well as the spatial distribution of hydrological processes, such as surface runoff, interflow, groundwater recharge, soil moisture and actual evapotranspiration, for a certain period, a station and time list must be prepared before running the model. The list is attached in the same file input.txt stored in the subdirectory /project/model/input, following the part of global model parameters. Table A-4.6 shows a template of spatial output specifications. Table A-4.6: Template of spatial output specifications
Q_sub 6 3 5 8 12 25 Surface runoff 1 1997 8 9 0 Interflow 1 1997 1 1 0 Groundwater-recharge 1 1997 1 1 0 Soil moisture 1 1997 8 9 0 Evapotranspiration 0 36

1997

10

1997

12 31

1997

12 31

1997

10

278

WetSpa Extension: Documentation and User Manual

a) Flow hydrograph at subcatchment outlet The number of the interested subcatchments is given after the mark q_sub, and the sequence number of each subcatchment is listed in the following line. This option is useful for simulating flow hydrographs simultaneously both at catchment outlet and at some gauging stations inside the catchment. The identification of specified subcatchments can be realized by modifying the stream link theme by using ArcView edit tools and making the discretization of the catchment. b) Spatial distribution of surface runoff This option gives a series of accumulative surface runoff distribution files after running the fully distributed model. The number of expected spatial outputs is given under the mark Surface runoff, and the wanted time periods are listed in the following lines. The input time period is in the format of start year, month, day, hour, and end year, month, day, hour as shown in the Table. If the model runs on a daily scale, set the hour value to be zero. If no spatial outputs are wanted, put zero value under the mark Surface runoff. c) Spatial distribution of interflow This option gives a series of accumulative interflow distribution files after running the fully distributed model. The format of the input values is the same as for the surface runoff list. d) Spatial distribution of groundwater recharge This option gives a series of accumulative groundwater recharge distribution files after running the fully distributed model. The format of the input values is the same as for the surface runoff list. e) Spatial distribution of relative soil saturation This option gives a series of soil moisture distribution files after running the fully distributed model. The format of the input values is the same as the surface runoff list. f) Spatial distribution of actual evapotranspiration

279

Appendix A This option gives a series of accumulative actual evapotranspiration distribution files after running the fully distributed model. The format of the input values is the same as for the surface runoff list.

4.5.

Model calibration and verification

4.5.1. Calibration and verification processes The purpose of calibration is to derive characteristics, equation constants, weighting factors, and other parameters that serve to define the model for a particular watershed. In distributed and continuous simulation, the calibration process is more rigorous and complex than that in model calibration for lumped model and discrete storm analysis, in that more parameters are involved in a distributed continuous model, a much greater amount of hydro-meteorological data is employed, and the fitting of the model requires a greater number of hydrological factors and more rigorous statistical procedures. To overcome these problems, calibration of WetSpa is not carried out for all model parameters, but for the most important parameters only, for instance, the channel roughness coefficient, plant coefficient, interflow scaling factor, and groundwater flow recession coefficient. Other parameters, such as hydraulic conductivity, root depth, interception and depression storage capacity, and so on, are set to values interpolated from the literature representing average conditions, and not calibrated but fixed to the selected values. Once the preparation of input data and model parameters are accomplished, the user can start to run the model for parameter calibration and model prediction. Programs can be run within the Arcview project interface, or directly executed in the subdirectory /project/model/program. Since the running of fully distributed model costs large memory space and computing time depending upon the catchment area, grid size, the length of time series and interval, it is preferable to run the semidistributed model firstly, adjust roughly the global and distributed model parameters, and then go to the fully distributed model, in order to save computing time for model calibration. The following is an outline of the steps for model calibration within ArcView interface.

280

WetSpa Extension: Documentation and User Manual 1) Calculating mean parameters for each subcatchment From the menu Model of the ArcView project or any view of the project, run the program Mean. This program computes mean model parameters of each subcatchment for use in the semi-distributed modelling and adjusting global model parameters preliminarily during model calibration. This operation can also be implemented independently by clicking the program mean in the subdirectory /project/model/program. The output file mean.txt is saved in the subdirectory /project/model/output. 2) Calculating unit hydrographs Run program IUH from the dropdown menu. This program calculates unit response function from each grid cell to the main rivers and basin outlet for use in fully distributed model, from each subcatchment to the main rivers and basin outlet for use in semi-distributed model, the unit response function for main rivers for use in both distributed and semi-distributed models, and unit response function for the entire catchment used for general parameter analysis. This operation can also be implemented independently by clicking the program IUH in the subdirectory /project/model/program. The output text files uh_cell_h.txt, uh_cell_s.txt, uh_sub_h.txt, uh_sub_s.txt, uh_river.txt and uh_watershed.txt are in the same format and saved in the subdirectory /project/model/output. 3) Modelling with a semi-distributed approach From the menu Model of the ArcView project or any view of the project, run the program Model1. Two options are available in the program: Predict outflow at catchment outlet and predict outflow both at catchment outlet and subcatchment outlets. Both options simulate flow hydrograph and water balance on a subcatchment scale, with output files q_tot.txt and balance.txt saved in the subdirectory /project/model/output. Additionally, option two routs water firstly to the subcatchment outlet, and then to the catchment outlet using channel response functions. Therefore, the produced hydrographs at the catchment outlet may be slightly different from the result of option one due to truncation errors in computing IUH. It also gives another output file q_sub.txt, which are the predicted discharges at selected subcatchment outlet saved in the subdirectory /project/model/output. Since

281

Appendix A both options give the same output file name q_tot.txt and balance.txt, the modeller needs to rename the file name if it is expected to keep the previous modelling results. 4) Model evaluation Run program Model Evaluation from the dropdown menu. This program gives a detailed description the observed data, simulation results, as well as the assessment of the current model parameters. The output file evaluation.txt is saved in the subdirectory /project/model/output. 5) Calibration of global parameters Based on the evaluation results and the visual comparison between observed and calculated hydrographs, adjust global parameters in the input file input.txt, repeat step 3 and 4, until a good match is reached. Another way of model calibration is to use the automated calibration scheme by setting properly the up and down limits of the parameter values in the PEST control file. However, manual parameter adjustment is still needed to avoid the ill-pose problems. If obvious errors exist and can not be overcome by adjusting global parameters, users may return to the GIS pre-processing phase, adjust values in the lookup table and recalculate the spatial parameter grids so as to make the input parameters more reliable. 6) Modelling with a fully distributed approach Keep the input files as in Model1, run program Model2 from the dropdown menu Model. This program simulates hydrological processes on cell scale, and predicts hydrograph at basin outlet, water balance on catchment scale, as well as spatial distribution of surface runoff, interflow, groundwater recharge, soil moisture and actual evapotranspiration at selected time periods. Output files q_tot.txt, q_sub.txt, balance.txt and other spatial distribution outputs are saved in the subdirectory /project/model/output. a) The output files q_tot.txt, q_sub.txt and balance.txt are in the same format as the outputs of Model1. If users want to keep the flow and water balance results of Model1, those files must be renamed to avoid being replaced. b) The output spatial runoff distribution files are named in the order listed in the input.txt, for instance, runoff1.asc, runoff2.asc, etc. Other spatial outputs are given similarly, such as interflow1.asc, recharge1.asc, moisture1.asc, 282

WetSpa Extension: Documentation and User Manual evaporation1.asc, and so on. All these output files are saved in the subdirectory /project/model/output. c) The computation time becomes much longer if too many spatial outputs are asked in the input file while running the fully distributed model. Therefore, it is suggested to generate less spatial outputs during model calibration. All expected spatial outputs can be given at the final run after model calibration, or using the program Water balance as described below. d) Run program Model Evaluation again to see the performance of the fully distributed model. Users are allowed to readjust global and spatial distributed input parameters in order to make a better match between calculated and observed hydrographs. 7) Simulation of water balance without flow routing This program is designed to compute water balance for each grid cell within the simulation period. Since the program does not cover the parts of flow routing, it can run more quickly and gives exactly the same water balance and spatial distribution outputs as Model2. a) Keep the input files as in Model1 or Model2, run program Water balance from the dropdown menu Model. b) The output file balance.txt and other spatial output file are saved in the c) The output file balance.txt and other spatial output file are saved in the subdirectory project/model/output. The previous output files need to be renamed if the user wants to keep them. d) The spatial input parameters can be reviewed based on the analysis of these spatial outputs, and some of the input parameter maps may need to be recalculated accordingly. 8) Model verification Model verification is being used to validate the calibrated model parameters by running the model for an independent period of record and comparing the results with observed data after calibration of the model is complete. This procedure will help to ensure that the calibration is not unique and limited to the data set employed for calibration.

283

Appendix A 4.5.2. Manual parameter adjustment In WetSpa Extension, manual calibration runs are made with trial simulation. Model output is compared with observed stream flow both at the catchment outlet and the internal discharge monitoring stations, and evaluated by the 5 assessment criteria described in section A-3.3. Based upon those comparisons and evaluations, parameter adjustments are made to improve the performance of the model. The initial choice of model parameters is not a critical concern since adjustments will be made during calibration. However, those parameters that have physical relevance should be determined to reduce the possibilities for future adjustment during model calibration. Model parameters that are typically encountered in a continuous simulation of WetSpa Extension are listed in Table A-4.6, in which the parameters that can be determined by independent analysis are indicated. For other parameters that need to be empirically determined, the initial value might be determined based upon known values in previous simulation studies, characteristic values of similar catchment, or default values collected from the literature. A desirable part of the calibration process is to make an independent estimate of the basins water balance. This calculation would yield the whole, annual or perhaps monthly estimates of basin precipitation, evapotranspiration, runoff, soil moisture and groundwater storage that can be helpful in calibrating the model parameters. Adjustments are made firstly to those parameters, which have the greatest impact on the model output, then proceeding to variables with lesser sensitivity. The process may be expressed as five basic steps with each having several trials. 1) Achieve fit of runoff volumes throughout the simulation period. This process preliminary involves adjustment of precipitation weighting factors, potential runoff coefficient, evapotranspiration factors, as well as interflow and groundwater flow production factors. Calibration fit is usually judged by comparing monthly, annual and the total runoff volumes. 2) Achieve fit of peak discharge and the time to peak. This step involves working with runoff distribution and routing factors, particularly for the components in controlling high flow hydrographs, such as hydraulic radius, channel roughness coefficient, etc.

284

WetSpa Extension: Documentation and User Manual 3) Achieve fit of hydrograph shape. This step mainly involves adjustment of model parameters in controlling low flow hydrographs, such as the interflow and groundwater flow factors, as well as evapotranspiration factors during dry period. 4) Achieve fit of snow melting floods if snow accumulation and snowmelt occurs in the study catchment. This step involves adjustment of model parameters in controlling snowmelt processes, including base temperature, temperature degreeday coefficient, rainfall degree-day coefficient, and temperature lapse rate. 5) Refine hydrograph fit. This final step involves working with different initial conditions and other distributed runoff production and flow routing parameters to refine a better hydrograph shape.

4.5.3. Parameter sensitivity Parameter sensitivity comprises the determination of changes in the individual parameters, in order to get an insight into the required precision of the model parameters relative to the precision of the model output. Table A-4.7 describes the order of parameter priority in more detail and gives a relative sensitivity of the variables, which are used in the WetSpa Extension. The Relative sensitivity in Table A-4.7 indicates the degree to which parameter affects model output. Major effects indicates which aspect of the output is primarily affected. Calibration priority suggests the order in which parameters are typically adjusted. And Independent evaluation indicates those parameters that are typically determined independent of the calibration process, because they are more physically based. All parameters in WetSpa Extension represent a physical process. It is essential that parameter values remain physically reasonable throughout the calibration process to keep the fit from being a local optimization that will not work when extrapolated to new data. Therefore, a verification step is desirable to ensure that the fit is a general solution, not one unique only to the calibration data used.

285

Appendix A Table A-4.7: Parameter sensitivity for manual calibration

Relative Major Calibration Independent Parameter sensitivity Effects priority evaluation Precipitation/Evapotranspiration Station weight High Runoff volume 1 Correction factor High Runoff volume 1 Vegetation fraction High Runoff volume 2 Vertical precipitation gradient Medium Runoff volume 2 Vertical PET gradient Medium Runoff volume 2 Maximum groundwater storage Medium Low flow shape 2 Snowmelt Base temperature High Snowmelt 1 Temperature degree-day factor High Snowmelt 1 Rainfall degree-day factor High Snowmelt 2 Temperature lapse rate High Snowmelt 2 Runoff distribution Potential runoff coefficient High Volume, high flow 1 Surface runoff exponent High Volume, peak flow 1 Threshold rainfall intensity High Volume, peak flow 1 Impervious fraction High Volume, high flow 1 Interception capacity Medium Runoff volume 2 Depression capacity Medium Runoff volume 2 Flow routing Surface roughness coefficient Medium High flow shape 2 Channel roughness coefficient High High flow shape 2 Hydraulic radius High High flow shape 2 Threshold of minimum slope Medium High flow shape 3 Threshold of stream network Medium High flow shape 3 Volume, flow Interflow scaling factor High 1 shape Baseflow recession coefficient High Low flow shape 1 Number of subcatchments Medium Low flow shape 3 Soil properties Hydraulic conductivity Medium Runoff volume 3 Porosity Low Runoff volume 3 Field capacity Low Runoff volume 3 Wilting point Low Runoff volume 3 Residual moisture content Low Runoff volume 3 Pore size distribution index Low Runoff volume 3 Root depth Medium Runoff volume 3 Initial conditions Soil moisture Low Flow shape 3 Groundwater storage Low Flow shape 3 Interception storage Low Flow shape 3 Depression storage Low Flow shape 3 Initial baseflow Low Flow shape 3

286

WetSpa Extension: Documentation and User Manual 4.6. Model output

4.6.1. Intermediate output WetSpa Extension produces the mean parameters for each subcatchment and the unit response functions for each grid cell, subcatchment and the main river channels separately, in order to avoid repeatable computations during model calibration. These intermediate outputs are further used as inputs in the distributed and semi-distributed models. Since WetSpa Extension simulates hydrological processes continuously, it uses and creates an immense amount of data, particularly if a long period of record is involved. Judging the fit of the final stream flow output along is difficult for model calibration. Reviewing these intermediate outputs therefore provides a possibility for efficiently parameter adjustments. 1) Mean parameters of each subcatchment Taking the Bissen subcatchment in the Alzette river basin, the Grand-duchy of Luxembourg, as a testing area, a sample intermediate output file mean.txt is shown in Table A-4.8. Table A-4.8: Sample output file of mean.txt
No 1 2 3 4 5 6 7 8 9 10 C 0.41 0.40 0.37 0.45 0.36 0.40 0.33 0.44 0.40 0.42 S 9.42 11.9 13.3 9.64 8.16 12.6 7.88 8.98 7.85 8.15 Kc 10.9 11.9 12.1 17.6 25.7 9.2 21.4 8.6 26.2 13.4 PS 0.49 0.48 0.47 0.49 0.48 0.48 0.47 0.49 0.48 0.49 FC 0.29 0.28 0.25 0.29 0.26 0.29 0.26 0.31 0.26 0.3 PI 11.1 11.0 11.0 10.4 9.7 11.2 9.8 11.3 9.7 10.8 WP 0.12 0.12 0.11 0.12 0.11 0.12 0.12 0.13 0.11 0.12 RM 0.05 0.05 0.07 0.04 0.05 0.05 0.07 0.04 0.05 0.04 IX 1.14 1.18 1.25 0.99 1.12 1.21 1.26 1.06 1.05 1.04 IN 0.48 0.48 0.49 0.46 0.48 0.48 0.48 0.48 0.46 0.47 DP 1.85 1.71 1.87 1.56 2.31 1.72 2.8 1.58 2.33 1.83 RD 1 1 1 1 1 1 1 1 1 1 TP 3 4 4 3 2 4 1 4 2 4 TE 3 3 3 3 2 3 1 3 2 3 TT 3 3 3 3 2 3 1 3 2 3 IMP 0 0.01 0.02 0 0 0.01 0 0 0 0 A 11.3 25.6 20.9 8.28 14.3 24.4 5.31 13 2.03 6.95

Where No is the number of the subcatchment, C is the potential runoff coefficient (-), S is the mean subcatchment slope (%), Kc is the mean hydraulic conductivity (mm/h), PS is the mean soil porosity (m/m), FC is the mean field capacity (m/m), PI is the mean pore size distribution index (-), WP is the mean wilting point (m/m), RM is the

287

Appendix A mean residual soil moisture (m/m), IX is the maximum interception capacity (mm), IN is the minimum interception capacity (mm), DP is the mean depression storage capacity (mm), RD is the mean root depth (m), TP is the Thiessen polygon number for precipitation (-), TE is the Thiessen polygon number for PET (-),TT is the Thiessen polygon number for temperature (-), IMP is the percentage of urban areas (%), and A is the subcatchment area (km). 2) Instantaneous unit hydrographs The files of instantaneous unit hydrograph (IUH) or the unit impulse response function include uh_cell_h.txt for routing water from cell to the basin outlet, uh_cell_s.txt for routing water from cell to the main river, uh_sub_h.txt for routing water from subcatchment to the basin outlet, uh_sub_s.txt for routing water from subcatchment to its outlet, uh_river.txt for routing water from subcatchment outlet to basin outlet, and uh_watershed txt which is the IUH for the entire catchment. All IUH files are in the same format. The total rows in the file uh_cell_h.txt and uh_cell_s.txt are the count of effective cells over the catchment. The total rows in uh_sub_h.txt, uh_sub_s.txt and uh_river.txt are equal to the number of subcatchments. And there is only one row in the file uh_watershed.txt. An example of the file uh_cell_h.txt is shown in Table A-4.9, where the first column is the start non-zero time step of the IUH, the second column is the end non-zero time step, and the values from the third column till the end are IUH non-zero values at each time step. Table A-4.9: Parts of output file uh_cell_h.txt
1 1 3 1 1 3 0 1 0 1 21 22 35 20 13 36 6 26 10 15 0.027 0.018 0.007 0.109 0.478 0.005 0.882 0.001 0.004 0.389 0.147 0.112 0.022 0.214 0.218 0.018 0.075 0.043 0.596 0.228 0.180 0.155 0.040 0.184 0.116 0.035 0.022 0.100 0.192 0.135 0.158 0.150 0.057 0.137 0.068 0.052 0.010 0.126 0.089 0.083 0.125 0.128 0.069 0.098 0.042 0.065 0.005 0.126 0.048 0.053 0.094 0.103 0.077 0.070 0.027 0.074 0.003 0.113 0.029 0.035 0.070 0.080 0.080 0.050 0.018 0.077 0.002 0.097 0.018 0.024 0.052 0.062 0.078 0.036 0.012 0.077 0.038 0.047 0.074 0.027 0.008 0.074 0.028 0.036 0.069 0.020 0.006 0.069 0.021 0.027 0.062 0.014 0.004 0.063 0.016 0.021 0.055 0.011 0.002 0.057

0.080 0.065 0.052 0.042 0.033 0.012 0.008 0.004 0.001 0.017 0.012 0.008 0.006 0.004

288

WetSpa Extension: Documentation and User Manual 4.6.2. Final output WetSpa Extension produces a variety of output files, depending on the selected options during the simulation run. The basic output files are the time series including predicted hydrographs at the catchment outlet or the selected subcatchment outlets, and water balance for the entire catchment over the simulation period. Other output files contain information about the spatial distributions of simulated hydrological processes at a predetermined time period. The program writes output into ASCII files, for which the file names are fixed in the program, or identified in the input file. All output files are stored in the subdirectory /project/model/output. 1) Discharge at the catchment outlet A sample output file q_tot.txt for the Bissen catchment is shown in Table A-4.10, where the first 4 columns are year, month, day and hour. If the model runs on a daily scale, the values in the Hours column are zero. P is the hourly rainfall (mm), Qs is the calculated surface runoff (m3/s), Qi is the calculated interflow (m3/s), Qg is the calculated groundwater flow (m3/s), and Q is the total runoff at the catchment outlet calculated by the summation of surface runoff, interflow and groundwater flow (m3/s). This file is the most useful output providing the simulated rainfall and runoff plot, in which the time increment for the output hydrograph is equal to the parameter dt given in the input file. Table A-4.10: Sample output file of q_tot.txt
year 1998 1998 1998 1998 1998 1998 1998 1998 1998 1998 month 10 10 10 10 10 10 10 10 10 10 day 23 23 23 23 23 23 23 23 24 24 Hour 16 17 18 19 20 21 22 23 0 1 P 1.38 1.99 0.95 1.03 0.34 9.51 0.26 0.52 0 0.43 Qs 0.274 0.569 1.535 2.262 3.033 3.497 9.084 11.122 12.723 13.545 Qi 0.539 0.542 0.544 0.549 0.554 0.560 0.567 0.585 0.606 0.628 Qg 0.947 0.948 0.949 0.950 0.951 0.952 0.952 0.953 0.953 0.953 Q 1.760 2.059 3.029 3.761 4.538 5.009 10.603 12.659 14.282 15.126

2) Discharge at the selected subcatchment outlet

289

Appendix A Table A-4.11 gives an example of output file q_sub.txt for the Bissen catchment, where the first 4 columns are year, month, day and hour, and the next 4 columns are calculated discharges at the outlet of subcatchment 1, 5, 10, and 11. This file gives simulated discharge data at a user selected location, which is useful for plotting hydrographs at an interested site, or comparing with observed hydrographs if an internal flow gauge exists at that site. Table A-4.11: Sample output file of q_sub.txt
year 1998 1998 1998 1998 1998 1998 1998 1998 1998 1998 month 10 10 10 10 10 10 10 10 10 10 day 23 23 23 23 23 23 23 23 24 24 Hour 16 17 18 19 20 21 22 23 0 1 1 0.645 0.834 1.245 1.214 1.210 1.046 3.808 2.637 1.983 1.436 5 0.809 0.927 1.204 1.227 1.247 1.156 2.853 2.324 1.937 1.579 10 0.410 0.506 0.713 0.681 0.678 0.596 2.027 1.332 1.012 0.782 11 0.516 0.684 0.960 0.785 0.769 0.633 2.975 1.093 0.772 0.615

3) Water balance for the entire catchment Both the semi-distributed and the fully distributed model produce a water balance time series. A sample output file balance.txt for the Bissen catchment is show in Table A-4.12, where T is the time step (-), P is the average hourly rainfall (mm), I is the average interception losses (mm), Sm is the average soil moisture in the root zone (mm), F is the average infiltration losses (mm), Et is the average actual evapotranspiration losses (mm), Perc is the average percolation out of root zone (mm), Rs is the average surface runoff (mm), Ri is the average interflow (mm), Rg is the average groundwater flow (mm), R is the total runoff (mm), and GT is the average active groundwater storage at this time step (mm). This file provides information on the simulated water balance for the entire catchment at each time step, which can be used for model calibration and evaluation.

290

WetSpa Extension: Documentation and User Manual Table A-4.12: Sample output file of balance.txt
T 1 2 3 4 5 6 7 8 9 10 P 1.384 1.986 0.952 1.034 0.342 9.506 0.261 0.521 0 0.431 I 0.594 0 0.049 0.049 0.049 0.049 0.049 0.039 0 0.074 Sm 282.46 283.99 284.66 285.40 285.60 292.90 293.02 293.35 293.73 293.97 F 0.616 1.546 0.703 0.766 0.228 7.324 0.163 0.371 0.42 0.275 Et 0.048 0.049 0.049 0.049 0.049 0.049 0.039 0.039 0.039 0.029 Perc 0.017 0.018 0.019 0.02 0.02 0.02 0.028 0.028 0.028 0.029 Rs 0.121 0.331 0.131 0.14 0.036 2.006 0.024 0.057 0 0.051 Ri 0.007 0.007 0.007 0.008 0.008 0.008 0.011 0.011 0.011 0.011 Rg 0.012 0.012 0.012 0.012 0.012 0.012 0.012 0.012 0.012 0.012 R 0.14 0.349 0.15 0.16 0.056 2.026 0.047 0.08 0.023 0.074 GT 150.17 150.17 150.18 150.19 150.20 150.21 150.22 150.24 150.25 150.27

4) Spatial output Table A-4.13 shows a part of the output file runoff.asc, which is the spatial distribution of surface runoff over the catchment for the time interval 14-15, Oct. 18, 1998, where ncols is the number of columns, nrows is the number of rows, xllcorner is corner coordinate in x direction (m), yllcorner is corner coordinate in y direction (m), cellsize is the cell size (m), and nodata_value is the no data value. This file contains information on simulated surface runoff on each grid cell, and can be imported to ArcView for further analysis. Other spatial distribution files, e.g. interflow, groundwater recharge, soil moisture, and actual evapotranspiration, are in the same format as for the surface runoff. The output file names are defined in the program in an ascending order, e.g. runoff1.asc, runoff2.asc, etc. Table A-4.13: Parts of the output file runoff.asc
ncols nrows xllcorner yllconer cellsize nodata_value -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 539 356 45240 84580 50 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 -1.000 2.463

-1.000 -1.000 -1.000 -1.000 -1.000 2.463 4.364

-1.000 -1.000 -1.000 -1.000 2.463 0.994 1.342

-1.000 -1.000 -1.000 2.152 2.463 1.342 1.342

-1.000 -1.000 2.152 2.463 2.863 1.342 2.261

2.463 2.463 2.463 2.863 2.863 2.261 2.261

2.463 2.463 2.863 2.863 2.863 2.261 2.261

291

Appendix A

5) Evaluation results Table A-4.14 gives a sample evaluation output evaluation.txt for the Bissen catchment after running the fully distributed model for an hourly time series in the year 1997. Table A-4.14: Model evaluation result evaluation .txt
Area of the catchment (km2) 288.8 Period of simulation 1/1/1999:0 - 31/12/1999:23 Measured precipitation, evaporation and discharge P Em Qm Qm unit (mm) (mm) (mm) (m3/s) sum 1041 630 416 33360 % of P 60.5 30.0 mean 0.12 0.07 0.05 3.81 max 9.1 0.72 0.77 61.7 Calculated water balance for the simulation period P I DS F E Perc Rs Ri Rg R DG unit (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) sum 1041 137 37.1 752 512 269 148 103 252 503 -15.2 % of P 13.2 3.56 72.2 49.2 25.8 14.2 9.90 24.2 48.3 -1.46 mean 0.119 0.016 271 0.086 0.058 0.031 0.017 0.012 0.029 0.057 132 max 9.39 1.12 352 7.23 0.713 0.289 1.69 0.120 0.066 1.78 207 Model evaluation CR1 CR2 CR3 CR4 CR5 0.048 0.825 0.792 0.807 0.832

Where P, Em and Qm are observed precipitation (mm), PET (mm) and discharge (mm) (m3/s) respectively, while the period of missing discharge data is not taken into account. I is the interception losses (mm), DS is the soil moisture difference between the start and the end time step (mm), F is the infiltration losses (mm), Et is the actual evapotranspiration (mm), Perc is the percolation out of root zone (mm), Rs is the surface runoff (mm), Ri is the interflow (mm), Rg is the groundwater flow (mm), R is the total runoff (mm), and DG is difference in groundwater storage between the start and the end time step (mm). CR1 is model bias. CR2 is model determination coefficient. CR3, CR4 and CR5 are Nash-Sutcliffe model efficiencies as described in section A-3.3. The evaluation results also contain the information on the catchment area, the period of model simulation, as well as the periods of missing discharge data

292

WetSpa Extension: Documentation and User Manual if they exist. Specifically, the change in soil moisture and groundwater storage over the simulation period is given in the evaluation output file, in order to make water balance compatible with other items, but its mean and maximum values are estimated state variables.

4.6.3. Post processing of model outputs In the current WetSpa Extension, no special effort has been paid in developing programs for the post processing of model output. However, the visual comparison between calculated and observed hydrographs can be carried out using Excel or other available software by loading the data from their text files. Moreover, the simulated hydrological processes for the entire catchment, such as precipitation, runoff, evapotranspiration, soil moisture, etc., can be viewed by plotting the data from the water balance output file. These graphs are helpful in adjusting model parameters more accurately and improving the model to have a better performance. Finally, the spatial output data including surface runoff, interflow, groundwater recharge, etc., can be imported to the ArcView project. Using the GIS tools, e.g. reclass, zoom, etc., a clear view can be obtained at the points of special interest. This information is not only a plot of spatial distribution of hydrological processes, but also a valuable feedback in refining model parameters.

5.

Case study: Bissen catchment, Luxembourg

5.1.

Description of the study area

The Bissen catchment is located in the Attert River basin covering an area of 294 km2 in the Grand-duchy of Luxembourg (Figure A-5.1). The Attert River is a main tributary of the Alzette River, where high-magnitude floods occurred frequently and have caused important damages since the early 1990s. The study catchment is homogeneous from a lithological point of view with essentially marls (El Idrissi et al., 2000). Using hourly rainfall-runoff series, the main goals are to apply the WetSpa Extension in predicting of flood hydrographs at basin outlet, estimating the spatial distribution and variability of the hydrological processes, and testing the sensitivity of model parameters with respect to catchment characteristics. 293

Appendix A

#
#

Town River

Clervaux
GERMANY

Alzette basin

Wiltz

Grand-Duchy of Luxembourg

BELGIUM

Bissen catchment
#

Ettelbruck

Ell

rt Atte#Bissen # Mersch
Alzette

Echternach
W

N E S

Luxembourg-city
#

Remich

10

20 km

Esch/Alzette
FRANCE

Figure A-5.1: Location of the Bissen catchment The climate of the region has a northern humid oceanic regime without extremes. Rainfall is the main source of runoff. The average annual precipitation varies between 800 mm to 1000 mm, which is characterized by distinctive winter and summer seasons. December is the wettest month of the year with average monthly precipitation of 84mm and April is the driest month of the year with average precipitation of 58 mm. The monthly PET values in the basin vary from 13.5 mm in winter to 81.8 mm in mid summer. High runoff occurs in winter and low runoff in summer due to the higher evapotranspiration. Winter storms are strongly influenced by the westerly atmospheric fluxes that bring humid air masses from the Atlantic Ocean (Pfister et al., 2000), and floods happen frequently because of saturated soils and low evapotranspiration. Statistical analysis of the observed data from the Luxembourg airport from 1947-1999 shows a uni-modal distribution of temperature with January being the coldest month of the year with an average temperature of 0.7C and July is the warmest month of the year with average temperature of 173C. The study area has a hilly topography, with elevation ranging from 220.6 to 545.0 m and average basin slope of 8.8% (Figure A-5.2). The land-use of the area, as shown in Figure A-5.3, is composed of agricultural land (23.7%), grassland (36.8%), forest (34.5%), urban areas (4.8%) and other land-use types (0.2%). Loam, silt loam, sandy

294

WetSpa Extension: Documentation and User Manual clay loam and loamy sand are main soil types covering 52.0%, 16.0%, 12.5% and 11.6% respectively as shown in Figure A-5.4.

N
Elevation (m) 220 - 247 247 - 274 274 - 301 301 - 328 328 - 355 355 - 382

382 - 409 409 - 436 436 - 463 463 - 490 490 - 517 517 - 545

W S

2.5

5.0 km

Landuse Crop Short grass Bog marsh Deceduous shrub Bare soil Urban area Open water

N W S E

2.5

5.0 km

Figure A-5.2: Topography map of Bissen

Figure A-5.3: Land use map of Bissen

Bissen # Reichlange # Platen # # # #Ell Niederpallen Useldange

Soil type Sand Loamy sand Silt loam Silt Loam Sandy clay loam Silt clay loam

N W S E

Hydrologic station River network

Theissen polygon
Catchment boundary

W S

2.5

5.0 km

2.5

5.0 km

Figure A-5.4: Soil type map of Bissen

Figure A-5.5: River network and Thiessen polygons of Bissen

5.2.

Data availability

1) Topographic data The topographic data is obtained from the numerical elevation data sets of the public ACT (Administration du Cadastre et de la Topographie, Luxembourg). A DEM with 5050 m grid size for the Bissen catchment is built using 2-meter resolution elevation contour map (Figure A-5.1). To check the validity of the data set, flow directions are 295

Appendix A estimated from the elevation data set and the rivers were generated. Then this is overlain with the actual river network. From this comparison as shown in Figure A5.5, it is seen that the data set has sufficient accuracy to carry out model simulation. 2) Land use data The land use information is taken from CORINE (Co-ordination of Information on the Environment) provided by the Luxembourgian Ministry of Environment, and the cadastral BD-L-TC (La Base de Donne Topo/Cartographique du Luxembourg) data. Both data sets are based on remote sensing information. These vector data sets are converted firstly to 5050 m grid according to WetSpa land use classification, as shown in Figure A-5.3, and then reclassified to 6 basic land use classes (forest, grass, crop, bare soil, urban and open water) for deriving model parameters of potential runoff coefficient and depression storage capacity. 3) Imperviousness and soil data For model simulation, the previous and impervious areas in each grid are required. For a grid size of 50 m, the impervious and pervious area ratio for different land use categories was established as described in Chapter A-3. Impervious fraction is set to 70% for commercial and industrial area, 30% for residential areas, 100% for water bodies and 0% for other land use categories. Information of soil types is obtained from the digital 1:100,000 Soil Map of the European Communities. The map is reclassified to 12 USDA soil texture classes based on their textural properties, and concerted to 50 m grid to match with the base topographic data. 4) Rainfall data 6 rainfall stations are available in the Bissen catchment as shown in Figure A-5.5. Among them, the Reichlange, located near the catchment centre, is a station recording rainfall at an hourly time step, while others are daily recording raingauges. To obtain an hourly rainfall series at each raingauge used in the WetSpa Extension, the hourly rainfall measured at Reichlange is taken as a reference, and multiplied by the ratio between the daily rainfall observed at the raingauge and the reference station:

296

WetSpa Extension: Documentation and User Manual

P day,i P P hour,i = hour,r day, r P

(5.1)

where Phour,i and Phour,r are hourly rainfall at gauging site i and the reference station (mm), and Pday,i and Pday,r are daily rainfall at gauging site i and the reference station (mm). Based on the raingauge network and the catchment boundary, the Thiessen polygon map is created as shown in Figure A-5.5 using ArcView Thiessen Polygon Extension. A unique hourly rainfall structure is then applied for each polygon, i.e. the rainfall series for each grid is set equal to the rainfall series of the nearest raingauge. 4) Potential evapotranspiration PET is estimated using the Penman-Monteith formula, as described in Chapter A-2, with daily meteorological data measured at Luxembourg airport located about 20 km south of the catchment. The same meteorological data series (net radiation, air temperature, relative humidity, and wind speed) are then uniformly applied on the whole study area. The average daily PET series for the Bissen catchment is achieved by applying weighting factor for the daily PET series obtained for the land uses as used in Drogue (2002):
EP URB. EP GRA . EP . EP d =% urb ,d + %AGR. EP agr,d + % gra,d + %FOR for,d

(5.2)

where EPd is the daily PET for the catchment, %URB, %AGR, %GRA and %FOR are weighting factors (area of land use type / area of catchment) for urban areas, cropland, grassland and forest as listed in Table A-5.2, and EPurb,d, EPagr,d, EPgra,d, and EPfor,d are daily PET series for each type of land use observed in the catchment. The PET from open water surface is neglected due do its very small percentage in the catchment. The values of canopy resistance, albedo and vegetation height considered in the PET calculation for the different land uses are given in Table A-5.1. For cropland, distinction is made between summer and winter where the land use is defined as a bare soil. The parameter values listed in Table A-5.1 are in accordance with the values used in scientific publications (Szeicz and Long, 1959; Perrier, 1982; Shuttelworth, 1989; Dickinson et al., 1993). Average values are used except for the canopy resistance, which are chosen in the range of the common values.

297

Appendix A

Table A-5.1: Default parameter values in the PET formula for different land uses
Land use Grassland Cropland (summer) Cropland (winter, = bare soil) Forest (mainly deciduous) Impervious area Canopy resistance (100 / ) 70 100 150 Albedo (-) 0.20 0.20 0.20 0.15 Vegetation height ( ) 0.12 1.00 0.12 15.0 -

The hourly PET series are finally computed from the daily data in proportion to the hourly temperature distribution (Guex, 2001).

Th,i EP h,i = EP d Td

(5.3)

where EPh,i is the hourly PET value at hour i (mm), Th,i is the hourly temperature at hour i (oC), and Td is the cumulative hourly temperature within a day (oC). In computation of hourly PET with Equation 5.3, the hourly temperature is set to zero if the actual temperature is lower than zero, and the hourly PET is considered to be zero if Td is less than or equal to zero.

5) Discharge data Six stream gauges, namely Ell, Reichlange, Useldange, Bissen, Niederpallen and Platen, as shown in Figure A-5.5, exist in the study area recording water levels at a 15-minute time step. The stream gauge Niederpallen and Platen are located at the outlet of two tributaries, while other 4 are located along the main channel with Bissen at the outlet of the catchment. Hourly discharge data are obtained through available rating curves at each gauging site. For Reichlange though, the rating curve has a low reliability, the discharge data could be used for validation purpose on peak flows. A total of 52 months of hourly rainfall, discharge and PET data from December 1996 to March 2001 are available for model calibration, except for Ell and Useldange (29 months from November 1998 to March 2001). The average hourly flow at Bissen during the monitoring period was 4.38 m/s, with flows ranging from 0.86 to 86.3

298

WetSpa Extension: Documentation and User Manual m/s, and the measured maximum hourly rainfall intensity was 21.5 mm/h occurred on July 7, 2000. Table A-5.2 presents the available data, geographical features, as well as the land use composition of each subcatchment. All hydrometeorological data sets used in this study come from the hydoclimatological database built-up and validated by the CRP-GL (Centre de Recherche Public - Gabriel Lippmann of Luxembourg). Table A-5.2: Data available and characteristics of the Bissen catchment
Station River Area (km2) Perimeter (km) Average slope (%) Raingauge type Start of data series End of data series Max. flow (m3/s) Urban (%) Crop (%) Grass (%) Forest (%) Water surface (%) Rest (%) Ell Reichlange Useldange Bissen Niederpalle Platen Attert Attert Attert Attert Pall Roubbach 107 166 255 294 34.6 47.1 49.9 64.4 75.3 82.1 32.6 33.0 9.4 9.2 8.9 8.8 6.1 11.1 Daily Hourly Daily Daily Daily Daily 22/10/98 01/12/96 02/10/98 01/12/96 01/12/96 01/12/96 01/04/01 01/04/01 01/04/01 01/04/01 01/04/01 01/04/01 25.0 13.4 51.7 86.3 22.6 11.2 3.5 4.0 4.1 4.8 3.9 4.8 20.9 23.3 24.7 23.7 19.1 32.4 33.7 37.6 37.2 36.8 51.6 25.8 41.8 34.9 33.9 34.5 25.0 36.7 0.0 0.1 0.1 0.1 0.2 0.2 0.0 0.0 0.0 0.1 0.0 0.0

5.3.

Basin delineation and parameter determination

The pre-processing starts with a creation of a depression-less DEM ensuring that positive drainage will occur. Next, flow direction and flow accumulation grids are calculated based on the flow path of steepest decent. The stream network is extracted from the master DEM using a threshold cells value of 100, which ensures that a channel is detected when the drainage area is greater than 0.25 km2. A grid of stream order used for assigning channel Mannings n is then derived from the stream network grid by the Shreve method. A slope grid is derived from the DEM and the delineated stream network, calculating slopes from each cell to its neighbours as percent rise for both land surface and stream channels. A threshold of minimum slope 0.01% is selected in order to deal with the problem of zero slopes in specific areas. The grid of hydraulic radius (Figure A-5.6) is calculated using the power law relationship

299

Appendix A described in section A-2.10.1 with a network constant a = 0.07 and a geometry scaling exponent b = 0.47, corresponding a flood frequency of 2-year return period. Finally, a map of subcatchment is extracted from the master DEM with a cells threshold value of 1000. 61 subcatchments are distinguished corresponding to an average subcatchment area of 4.73 km2 with minimum subcatchment area of 0.043 km2 and maximum subcatchment area of 14.5 km2. The resulting minimum subcatchment area is much smaller than the threshold value 0.75 km2 due to the remainder of the extraction. These subcatchments serve as working units in the semi-distributed model, and are also used for simulating groundwater balance in the full-distributed model.

Hydraulic radius (m) 0.005 - 0.01 0.01 - 0.05 0.05 - 0.1 0.1 - 0.2 0.2 - 0.3 0.3 - 0.5 0.5 - 1 1 - 1.2

N W S E

2.5

5.0 km

Runoff Coefficient 0.05 - 0.15 0.15 - 0.25 0.25 - 0.35 0.35 - 0.45 0.45 - 0.55 0.55 - 0.65 0.65 - 0.75 0.75 - 1

N W S E

2.5

5.0 km

Figure A-5.6: Hydraulic radius of Bissen Figure A-5.7: Runoff coefficient of Bissen

The physical parameters created by ArcView based on the soil type map include the saturated hydraulic conductivity, soil porosity, field capacity, plant wilting point, residual moisture content, and the soil pore size distribution index. The land use based parameters used in the model include root depth, interception capacity, and the Mannings coefficient. The Mannings coefficients for river channels are interpolated based on the GIS derived stream orders, with 0.03 m-1/3s for the highest order and 0.05 m-1/3s for the lowest order. The parameter maps of potential runoff coefficient (Figure A-5.7) and depression storage capacity are created based on the combination of the three base maps. The impervious percentage for urban cells is set to be 70%, while the rest are assumed being covered by grass. The flow routing parameters include flow velocity, average travel time and its standard deviation from cells to the catchment outlet and to the subcatchment outlet. Figure A-5.8 and A-5.9 shows the

300

WetSpa Extension: Documentation and User Manual calculated mean travel time and its standard deviation from cells to the basin outlet for the Bissen catchment.

Standard deviation (h) 0-2 2-3 3-4 4-5 5-6 6-8 8 - 10 10 - 12

N W S E

2.5

5.0 km

Travel time (h) 0-1 1-3 3-6 6 - 10 10 - 15 15 - 21 21 - 28 28 - 39

N W S E

2.5

5.0 km

Figure A-5.8: Mean travel time to the basin Outlet of Bissen

Figure A-5.9: Standard deviation of flow time to the basin outlet

Finally, the Thiessen polygons for precipitation and PET (Figure A-5.5) are created using the Thiessen polygon extension. Due to the fact that snow accumulation has a very minor effect on the runoff process in this catchment, the snowmelt flow is not accounted during the flow simulation. Therefore the preparation of temperature Thiessen polygon and temperature data series is not necessary in this case study. At this moment, all spatial parameters used in the model simulation are developed. A visual inspection is performed to ensure that the general characteristic of the parameter maps, such as the range, extreme values, etc., are logical and in the right order.

5.4.

Model calibration and validation

Model calibration for the study catchment was performed for the time period of Dec. 1996 to Dec. 1999, while the period of Jan. 2000 to Apr. 2001 was used for model validation. Both the visual and statistical comparisons for the observed and simulated flow hydrographs at Bissen station were performed for the calibration and validation periods. Comparisons at other three gauging stations inside the catchment were also implemented as a kind of model validation. The comparisons of simulated and observed values included runoff volumes, hourly time series of flow, and the time to

301

Appendix A the peak for each individual flood. In addition to the above comparisons, the water balance components (measured and simulated) were reviewed. This effort involved displaying model results for the whole simulation period for the water balance components of precipitation, infiltration, total runoff, overland flow, interflow, baseflow, PET, actual Evapotranspiration, interception, groundwater recharge, as well as the differences in soil moisture and groundwater storage between the start and end hour. Although observed values were not available for each of the water balance components listed above, the average annual values and its spatial distribution were checked for consistency with expected values for the region to ensure that overall water balance reflected local conditions, as impacted by the catchment hydrological and geographical characteristics. Calibration of the WetSpa Extension was a cyclical process of making parameter changes, running the model, producing the comparisons of simulated and observed values, and interpreting the results. The calibration process was performed mainly for the global parameters including interflow scaling factor, baseflow recession coefficient, evapotranspiration coefficient, initial soil moisture and groundwater storage, as well as the surface runoff exponent as listed in the input file. Other spatially distributed model parameters were assumed to be reasonable and remained the values as they are. Calibration of the evapotranspiration coefficient could be performed independently by comparing the calculated and observed flow volume for a long time series. The interflow scaling factor was calibrated by matching the computed discharge with the observed discharge for the recession part of the flood hydrograph. Groundwater flow recession coefficient could be obtained by the analysis of recession curves at discharge gauging stations. Refinement of this baseflow recession coefficient was necessary to get a better fit for the low flows. The initial soil moisture and initial groundwater storage were adjusted based on the comparison between the calculated and observed hydrographs for the initial period. And the runoff exponent and the rainfall intensity threshold were adjusted based on the agreement between calculated and observed flows for the small storms with lower rainfall intensity. Since these global model parameters are physically based, the interval of their variation can be predetermined based on the specific characteristics of the study catchment. For instance, the interflow scaling factor is generally within the range of 1 to 10, and the evapotranspiration coefficient should be close to 1. After the adjustment of the input 302

WetSpa Extension: Documentation and User Manual global model parameters and running the model, the post-processing capabilities of WetSpa Extension (listings, plots, statistics, etc.) were used extensively to evaluate the calibration/verification effort. Figure A-5.10 shows a typical calibration result for a flood series occurred in December 1997, corresponding to input global model parameters of Ci = 7.5, Cg = 9.0 m2/s, K_ss = 1.03, K_ep = 1.02, G0 = 280 mm, G_max = 300 mm, K_rain = 2.0 and P_max = 5.0 mm/h, where the meanings of above denotations can be found in section A-3.2.

120 Precipitation 80 Observed Q Calculated Q Baseflow Qi+Qg 40

10

20

0 3/12/97

30 7/12/97 11/12/97 15/12/97 Time (d/m/y) 19/12/97

Figure A-5.10: Observed and calculated flow at Bissen for the floods in Dec. 1999 It can be found from Figure A-5.10 that the calculated hydrograph is generally in a good agreement compared with the observed hydrograph. A big storm occurred on the fourth of December, 1997, but did not produce too much runoff due to the lower antecedent soil moisture. Most of the rainfall were therefore infiltrated and used to saturate the soil. Thereafter, another three big storms occurred successively on December 5, 8 and 12, which yielded pick discharges of 44.0, 86.3 and 66.8 m3/s respectively. The calculated pick discharges are 51.1, 73.1 and 58.1 m3/s corresponding to relative errors of 16.1%, -15.3% and -13.0% respectively. The simulated baseflow contribution was not remarkable for the first two floods, but abundant for the third and fourth flood. This can be explained that the soil moisture and the effective groundwater storage were low at beginning, and not sufficient to generate abundant interflow and groundwater flow for the first two floods. Due to the occurrence of following storms, soils were getting saturated and the surplus soil water percolated to the groundwater storage, leading to a higher baseflow for the third and fourth floods and also the following flow period. 303

Precitation (mm/h)

Discharge (m 3/s)

Appendix A

Table A-5.3: Statistics and model performance for the calibration/validation period
Station Period Total rainfall (m) 2.707 2.818 2.779 1.726 Total runoff (m) 1.511 1.455 1.202 0.798 Flow Mean coef. flow CR1 CR2 CR3 CR4 CR5 (%) (m/s) 55.8 2.25 0.035 0.765 0.772 0.653 0.786 51.6 4.68 0.012 0.815 0.798 0.715 0.824 43.3 3.66 -0.014 0.813 0.735 0.682 0.805 46.2 5.47 -0.025 0.762 0.614 0.667 0.753

Ell 22/10/98-29/01/01 Useldange 02/10/98-31/10/00 12/01/96-31/12/98 Bissen 01/01/99-12/05/00

Table A-5.3 presents the statistics of observed rainfall, runoff, the flow coefficient (ratio of the outflow water volume at the measuring station to the volume of water precipitated over the drainage area) and the mean flow discharge during the statistical period, as well as the model performance for the calibration/validation period at station Ell, Useldange and Bissen on hourly scale. The model performance is found to be satisfactory as illustrated in the table. Model bias for the simulation period is within the range of -0.025 to 0.035. Model determination coefficient is within the range of 0.765 to 0.815. The flow efficiency coefficient is within the range of 0.614 to 0.798, while the efficiency coefficient ranges from 0.653 to 0.715 for low-flow, and 0.753 to 0.824 for high-flow. These evaluation results indicate that the model has a high confidence and can give a fair representation of both low-flow and high-flow hydrographs for the study catchment. A graphical comparison between calculated and measured hourly flows at Bissen for the validation year 1999 is presented in Figure A-5.11. With the simulated initial hydrological condition at the end of the year 1998, the simulation results for the year 1999 were in fairly good agreement with the measured discharges. Similar simulation results can be obtained for other hydrological years. Figure A-5.12 shows the plots of 18 observed peak discharges at Bissen against their calculated peak discharges selected from the whole simulation period for Qpeak > 30 m3/s. The correlation coefficient is 0.96, which proves that the flow peak discharges are well reproduced. The errors of the time to the peak for the 18 floods were also examined, in which 12 of them are within the interval of -3 to 3 hours, and the rest are outside this range. The maximum error is 10 hours for the flood on April 1996, as the precipitation lasted for 3 days with lower rainfall intensity, and long peak flow duration was observed.

304

WetSpa Extension: Documentation and User Manual

Figure A-5.11: Observed and calculated hourly flow at Bissen for the year 1999

100

Calculated peak (m /s)

80

60

40

20 20 40 60 80
3

100

Observed peak (m /s)

Figure A-5.12: Peak Qm Vs Peak Qc selected from the whole simulation period Figure A-5.13 represents the observed and calculated hourly flow frequency curve for the whole simulation period. The flow frequency curve demonstrates consistent patterns between calibration and validation time periods, and in general showed good agreement. However, there are some obvious deviations for small floods, especially for the flow within the discharge interval of 2 to 6 m3/s, where the calculated flows are over estimated. These deviations may be attributed to the uncertainties inherent in modelling complex processes such as flood frequency related hydraulic radius, interflow factors, etc.

305

Appendix A

Figure A-5.13: Observed and calculated hourly flow frequency curves at Bissen

5.5.

Discussion

The hydrological modelling effort for the comprehensive study of the Bissen catchment is an attempt to apply hydrological modelling from GIS data sets. The modelling approach was developed efficiently and with consistent methodologies. The ability to define spatially distributed model parameters interactively based on topography, land use and soil maps using ArcView GIS allowed users to work quickly, and the ability to compare the intermediate results with existing maps increased the confidence in the validity of the model components. From the viewing and manipulation of the geographical data, to the development of the physical parameters, and to the post processing of the simulation results, it is clear that WetSpa Extension has its ability to calculate basin characteristics directly from terrain models allowed user to complete the comprehensive study in a timely manner. Based on the hourly hydrograph comparisons at Bissen and other internal stations, it can be concluded that the modelling results have a good to very good agreement with observed hydrographs. Table A-5.4 tabulates the measured and calculated water balance for each modelling component over the whole simulation period for the Bissen catchment. The estimated volume of interception, surface runoff and infiltration are 583.9, 688.2 and 3219 mm representing 13.0%, 15.3% and 71.5% of the total precipitation. It can also be calculated from the table that 31.5% of the infiltrated water is percolated out of the root zone, 19.4% of which becomes lateral interflow, 46.3% of which is evapotranspirated into the atmosphere from the root zone

306

WetSpa Extension: Documentation and User Manual (total evapotranspiration interception transpiration from the groundwater storage). The transpiration from groundwater storage can be estimated from the percolation amount subtracted by the groundwater volume, which is 249.1 mm in total representing 10.7% of the total evapotranspiration. The rest are remained in the soil moisture and groundwater storage. The estimated surface runoff, interflow and groundwater flow are 688.2, 623.0 and 763.9 mm representing 33.2%, 30.0% and 36.8% respectively of the total runoff. Interflow is an important flow component in this study due to the steep slope and well vegetation over the catchment. Table A-5.4: Water balance estimation at Bissen for the whole simulation period
Component Precipitation Interception Infiltration Evapotranspiration Percolation Surface runoff Interflow Groundwater flow Total runoff SM difference GWS difference Measured (mm) 4505 Calculated (mm) 4505 583.9 3219 2323 1013 688.2 623.0 763.9 2075 47.11 45.68 Percentage (%) 100 13.0 71.5 51.6 22.5 15.3 13.8 17.0 46.1 1.05 1.01 Mean (mm/h) 0.119 0.015 0.085 0.061 0.027 0.018 0.016 0.020 0.055 287.2 (mm) 176.3 (mm) Max (mm/h) 21.49 1.121 15.84 0.732 0.303 5.243 0.183 0.037 5.259 372.8 (mm) 325.7 (mm)

2467

2000

SM: soil moisture, GWS: groundwater storage. Despite the good performance of the model predictions, the model requires the user to provide the necessary elevation, soil and land use data sources that are specific to the study area. The DEM is the starting point for several processes in producing the predicted hydrographs. Moreover, a successful hydrological model requires information regarding the infiltration potential of the surface where the runoff occurs. The preferred data consists of digital maps containing area soils and land use information with associated potential runoff coefficient and depression storage capacity corresponding to each grid cell with different slope, soil and land use combinations. The functionality of WetSpa Extension was designed to accommodate both overland flow and channel flow. The routing process is accomplished by the method of linear diffusive approximation without considering the specific channel characteristics for different cross sections, for instance, the channel loss properties, 307

Appendix A channel width, compound channel roughness, etc. A linear interpolation of

Mannings n was then performed according to the stream orders by setting constant roughness values for the highest and lowest stream order. For the very flat areas (ponds, small lakes, and other zero slope cells), a minimum slope threshold was given, 0.01% for this case study, in order to keep the water moving in a right order on those areas. All these treatments will greatly facilitate the task of data collection and simplify the scheme of model calculation, but may bring errors and uncertainties to the final simulation results. 6. Concluding remarks

A GIS-based hydrological model, WetSpa Extension, in its fully and semi-distributed version compatible with remote sensing and GIS has been described in this user manual. The model runs on a microcomputer with a user-friendly interface, and can be applied to a wide range of watersheds for simulating the hydrological behaviour and especially runoff with due account for available topography, soil type, and land use data. The approach consists of the development of a spatially distributed modelling framework that accounts for spatial variability in terrain features to facilitate flood management and the physically realistic spatial integration of the complete water balance at a range of spatial and temporal scales. The model is implemented entirely within ArcView using Avenue scripts along with its Spatial Analyst and a hydrological extension integrated within a GIS environment. Encouraging results have been achieved as illustrated in the case study. The spatial characteristics of input meteorological variables, i.e. temperature, precipitation and PET, are captured by means of Thiessen polygons, on which linear topographic corrections are implemented within each polygon to account for the altitude variation of these meteorological variables if necessay. The generation of surface runoff depends upon rainfall intensity and soil moisture status and is calculated as the net precipitation times a runoff coefficient, which depends upon slope, land use and soil type. Snowmelt is estimated from typical temperature variations and a degree-day type of snowmelt model. The runoff is subsequently routed through the basin along flow paths determined from the high resolution DEM using a diffusive wave transfer model that leads to response functions between any 308

WetSpa Extension: Documentation and User Manual start and end point, depending upon slope, flow velocity and dissipation characteristics along the flow lines. Interflow and percolation is controlled by soil characteristics and modelled by Darcys law and kinematic approximation. The groundwater flow and its storage are conceptualized as a linear reservoir on small subcatchment scale with recession constant determined at reference gauging stations, and estimated for each subcatchment in relation with its drainage area and average slope. The spatial variability of model parameters used in river basin simulations is known to affect simulated results. Like other distributed models, WetSpa allows for variability of model parameters in space over a catchment by incorporating information from the spatial variability of soils, land use, and topography, which gives a more accurate representation of natural hydrological processes. However, a high degree of uncertainty exists for many model input parameters including the potential runoff coefficient, soil hydraulic conductivity, roughness coefficient, hydraulic radius, as well as the threshold values for determining stream network, minimum slope and the percentage of impervious areas within an urban cell, etc. Moreover, some global parameters, such as interflow scaling factor, plant coefficient, degree-day coefficient, etc., are used in the model due to their complexity of optimization and for the simplification of model calibration. The large number of uncertainties associated with the input meteorological variables and the model parameters may make the calibration and validation of the model a time intensive undertaking. To deal with this problem, priorities are given to the model parameters with high sensitivity during model calibration as described in chapter A-4. Further refinement of other model parameters is recommended in order to improve the model reliability. Additionally, pre-adjustment of model parameters to the channel geometry, boundary conditions, and system connectivity are necessary to achieve the quality of the final model simulation results.

Reference Allen, R.G., Using the FAO-56 dual crop coefficient method over an irrigated region as part of an evapotranspiration study, J. Hydrol., 229, 27-41, 2000.

309

Appendix A Anderson, E.A., National Weather Service River Forecast System - Snow Accumulation and Ablation Model, NOAA Technical Memorandum NWS HYDRO-17, U.S. Dept. Commerce, Silver Spring, MD., 1973. ASCE, American Society of Civil Engineers, Design and Construction of Sanitary and Storm Sewers, Manuals and Reports of Engineering Practice, No. 37, New York, 1969. Batelaan, O., Wang, Z.M. & De Smedt, F., An adaptive GIS toolbox for hydrological modelling, 3-9, eds., Kovar, K. & Nachtnebel, H.P., Application of geographic information systems in hydrology and water resources management, IAHS Publ., 1996. Bauwens, W., Vanrolleghem, P. & Smeet, M., An evaluation of the efficiency of the combined sewer-waste-water treatment system under transient conditions, Water Sci. Technol., 33, 199-208, 1996. Boers, T.M. & Ben-Asher, J., A review of rainwater harvesting, Agric. Water Manage., 5, 145-158, 1982. Brooks, R.H. & Corey, A.T., Properties of porous media affecting fluid flow, J. Irrig. Drain. Amer. Soc. Civil Eng., IR2, 61-88, 1966. Brown, V.A., McDonnell, J.J., Burns, D.A. & Kendall, C., The role of event water, a rapid shallow flow component, and catchment size in summer stormflow, J. Hydrol., 217, 171-190, 1999. Browne, F.X., Stormwater management, Standard Handbook of Environmental Engineering, ed., R.A. Corbitt, 7.1-7.135, McGraw-Hill, New York, 1990. Carsel, R.F. & Parrish, R.S., Developing joint probability distributions of soil water retention characteristics, Water Resour. Res., 24, 755-769, 1988. Chow, V.T., Handbook of Applied Hydrology, 7-25, McGraw-Hill, Book Company, New York, 1964. Chow, V.T., Maidment, D.R. & Mays, L.W., Applied Hydrology, 498, McGraw-Hill Inc., New York, 1988. Clark, O.R., Interception of rainfall by prairie grasses, weeds and certain crop plants, Ecol. Monogr., 10, 243-277, 1940. Cosby, B.J., Hornberger, G.M., Clapp, R.B. & Ginn, T.R., A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils, Water Resour. Res., 20, 682690, 1984.

310

WetSpa Extension: Documentation and User Manual De Smedt, D., Development of a continuous model for sewer system using MATLAB, MSc. Thesis, Laboratory of Hydrology, Vrije Universiteit Brussel, Belgium, 1997. De Smedt, F., Liu, Y.B. & Gebremeskel, S., Hydrological modelling on a catchment scale using GIS and remote sensed land use information, ed., Brebbia, C.A., 295-304, Risk Analyses II, WIT press, Southampton, Boston, 2000. Dickinson, R.E., Henderson-Sellers, A. & Kennedy, P.J., Biosphere Atmosphere Transfer Scheme (BATS), Version le as Coupled to the NCAR Community Climate Model, NCAR Technical Note, NCAR, Boulder, Colorado, 1993. Dingman, S.L., Physical Hydrology, Prentice-Hall, New Jersey, 1994. Dingman, S.L., Barry, R.G. & Reynolds, R.C., Application of kriging to estimating mean annual precipitation in a region of orographic influence, Water Resour. Bull., 24, 329-339, 1988. Drogue G., El Idrissi, A., Pfister, L., Leviandier, T., Iffly, J.F. & Hoffmann, L., Calibration of a parsimonious rainfall-runoff model: a sensitivity analysis from local to regional scale, eds., Rizzoli A.E. & Jakeman A.J., 464-469, Integrated Assessment and Decision Support, Proceedings of the First biennal meeting of the International Environmental Modelling and Software Society, Lugano (Switzerland), Vol. 1., 2002. Dunne, T., Stochastic aspects of the relations between climate, hydrology and landform evolution, Trans. Jpn. Geomorphol. Union, 12, 124, 1991. Eagleson, P.S., Dynamic Hydrology, 364, McGraw-Hill Pub., 1970. El Idrissi, A., Drogue, G., Pfister, L., Iffly, J.F., Hoffmann, L., Hingray, B. & Guex, F., Estimation of high floods by three rainfall-runoff models with short rainfall-runoff series (Alzette river basin, Luxembourg), eds., Rizzoli A.E. & Jakeman A.J., 470-475, Integrated assessment and decision support, Lugano (Switzerland), Vol. 1, 2002. Environmental Systems Research Institute (ESRI), ArcView GIS Version 3.2 Online Help Documentation, 1999. Famiglietti, J.S. & Wood, E.F., Multiscale modelling of spatially variable water and energy balance processes, Water Resour. Res., 30, 3061-3078, 1994. Ferguson, B.K., Introduction to Stormwater, Concept, Purpose and Design, 111, John Wiley & Sons, Inc. USA, 1998.

311

Appendix A Fernandez-Illescas, C.P., Porporato, A., Laio, F. & Rodriguez-Iturbe, I., The ecohydrologicalal role of soil texture in a water-limited ecosystem, Water Resour. Res., 37(12), 2863-2872, 2001. Fetter, C.W., Applied Hydrogeology, 6, Charles E. Merrill Pub. Co., Columbus, Ohio, 1980. Fronteau, C. & Bauwens, W., Een evaluatie van de efficientie van het rioolstelsel, RWZI, rivier systeem onder dynamische omstandigheden, Water, 84, 203-210, 1995. Geiger, W.F., Marsalek, J., Rawls, W.J. & Zuidema, F.C., Manual on drainage in urbanized areas, Vol. 1, Planning and design of drainage systems, Studies and Reports in Hydrology, UNESCO, 1987. Guex, F., Modlisation hydrologique dans le bassin versant de lAlzette (Luxembourg), rgionalisation des paramtres dun modle global, Travail pratique de Diplme, Ecole Polytechnique Fdrale de Lausanne, 67, 2001. Haan, C.T., Hydrological Modelling of Small Watersheds, 211, Published by American Society of Agricultural Engineers, USA, 1982. Hairsine, S.Y. & Parlange, J.Y., Kinematic shock waves on curved surfaces and application to the cascade approximation, J. Hydrol., 87, 187200, 1986. Hansen, B., Schjoning, P. & Sibbesen, E., Roughness indices for estimation of depression storage capacity of tilled soil surfaces, Soil Till. Res., 52, 103-111, 1999. Henderson, F.M., Open Channel Flow, 522, McMillan, New York, 1966. Horton, R.E., Rainfall interception, Monthly Weather Review , 47, 603-623, 1919. Kachroo, R.K. & Natale, L., Non-linear modelling of rainfallrunoff transformation, J. Hydrol. 135, 341369, 1992. Kirkby, M.J., Hill Slope Hydrology, 235, John Wiley & Sons, Ltd., 1978. Kite, G.W. & Kauwen, N., Watershed modelling using land classification, Water Resour. Res., 28, 3193-3200, 1992. Linsley, R.K., Kohler, J., Max A. & Paulhus, J.L.H., Hydrology for Engineers, 237, 3rd Ed., McGraw-Hill, New York, 1982. Liu, Y.B., Gebremeskel, S., De Smedt, F., Hoffmann, L. & Pfister, L., A diffusive transport approach for flow routing in GIS-based flood modelling, J. Hydrol., 283, 91-106, 2003.

312

WetSpa Extension: Documentation and User Manual Liu, Y.B., Gebremeskel, S., De Smedt, F. & Pfister, L., Flood prediction with the WetSpa model on catchment scale, eds., Wu et al., 499-507, Flood Defence 2002, Science Press, New York Ltd., 2002. Loague, K.M., Impact of rainfall and soil hydraulic property information on runoff predictions at the hillslope scale, Water Resour. Res., 24(9), 15011510, 1988. Lull, H.W., Ecological and silvicultural aspects, Handbook of Applied Hydrology, ed., Ven Te Chow, 6.1-6.30, McGraw-Hill, New York, 1964. Male, D.H. & Gray, D.M., Snowcover ablation and runoff, In Handbook of Snow: Principles, Processes, Management and Use, 360-436, eds., D.M. Gray and D.H. Male, Pergamon Press, 1981. Mallants, D. & Feyen, J., Kwantitatieve en kwalitatieve aspecten van oppervlakte en grondwaterstroming (in Dutch), 76, Vol. 2, KUL, 1990. Martin, G.N., Characterization of simple exponential baseflow recessions, J. Hydrol. 12, 57-62, 1973. Martinec, J., Rango, A. & Major, E., The Snowmelt-Runoff Model (SRM) Users Manual. NASA Reference Publ. 1100, Washington, D.C., USA, 1983. Mazion, E. & Yen, B.C., Computational discretization effect on rainfall-runoff simulation, J. Water Resour. Plan. Manage., 120(5), 715734, 1994. McCuen, R.H., Rawls, W.J. & Brakensiek, D.L., Statistical Analysis of the BrooksCorey and the Green-Ampt parameters across soil textures, Water Resour. Res., 17(4), 1005-1013, 1981. Miller, W.A. & Cunge, J.A., Simplified equations of unsteady flow, eds., K. Mahmood & V. Yevjevich, Unsteady Flow in Open Channels, Water Resources Publications, Fort Collins, CO., 1975. Molnar, P. & Ramirez, J.A., Energy dissipation theories and optimal channel characteristics of river networks, Water Resour. Res., 34, 1809-1818, 1998. Moore, I.D. & Gessler, P., Nielsen, G. & Peterson, G., Soil attribute prediction using terrain analysis, Soil Sci. Soc. Am. J., 57, 443-452, 1993. Moore, I.D., Larson, C.L., Estimating microrelief surface storage from point data, Trans. ASAE., 20, 1073-1077, 1979. Nash, J.E. & Sutcliffe, J.V., River flow forecasting through conceptual model, J. Hydrol., 10, 282290, 1970. Onstad, C.A., Depressional storage on tilled soil surfaces, Trans. ASAE., 27, 729-732, 1984. 313

Appendix A Perrier, A., Land surface processes : vegetation, ed., Eagleson, P., Land Surface Processes in Atmospheric General Circulation Lodels, 395-348, Cambridge Univ. Press, 1982. Pfister, L., Humbert, J. & Hoffmann, L., Recent trends in rainfall-runoff characteristics in the Alzette river basin, Luxembourg, Climate Change, 45(2), 323-337, 2000. Ragab, R. & Cooper, J.D., Variability of unsaturated zone water transport parameters: implications for hydrologicalal modelling, J. Hydrol., 148, 109-131, 1993. Rawls, W.J. & Brakensiek, D.L., Prediction of soil water properties for hydrologic modelling, In: Watershed Management in the Eighties, eds., Jones, E & Ward, T.J. 293-299, ASCE, New York, 1985. Rawls, W.J., Brakensiek, D.L. & Saxton, K.E., Estimation of soil water properties, Trans. Amer. Soc. of Agric. Engin., 25(5), 1316-1320, 1328, 1982. Rowe, L.K., Rainfall interception by an evergreen beech forest, J. Hydrol, 66, 143158, 1983. Saghafian, B., Julin, P.Y., & Ogden, F.L., Similarity in catchment response 1. Stationary rainstorms, Water Resour. Res., 31, 1533-1541, 1995. Sharma, K.D., Runoff behaviour of water harvesting microcatchment. Agric. Water Manage., 11, 137-144, 1986. Sheaffer, J.R., Wright, K.R., Taggart, W.C. & Wright, R.M., Urban Storm Drainage Management, Marcel Kekker, New York, 1982. Shuttleworth, W.J., Micrometeorology of temperate and tropical forest, Phil. Trans. Roy. Meteorol. Soc., 110, 1143-1162, 1989. Simons, D.B., Storm water and sediment runoff simulation for a system of multiple watersheds, Volume II, Sediment routing and yield, USDA Forest Service, Rocky Mountain Range Experiment Station, Flagstaff, Arizona, 1981. SINCE, WatfloodSPL8, Flood Forecasting System, Developed for Surveys and Information Branch Ecosystem Science and Evaluation Directorate, Environment Canada, 1972. Singh, P., Kumar, N. & Arora, M., Degree-day factors for snow and ice for the Dokriani Glacier, Garhwal Himalayas, J. Hydrol., 235, 1-11, 2000. Smakhtin, V.Y., Sami, K. & Hughes, D.A., Evaluating the performance of a deterministic daily rainfall-runoff model in a low flow context, Hydrol. Process., 12, 797-811, 1998. 314

WetSpa Extension: Documentation and User Manual Szeicz, G. & Long, I.F., Surface resistance of crop canopies, Water Resour. Res., 5, 622-633, 1969. Thornthwaite, C.W. & Mather, J.R., The Water Balance, Laboratory of Climatology, Publ. No. 8, Centerton NJ., 1955. Ullah, W. & Dickinson, W.T., Quantitative description of depression storage using a digital surface model, I. Determination of depression storage, J. Hydrol. 42, 63-75, 1979a. Ullah, W. & Dickinson, W.T., Quantitative description of depression storage using a digital surface model, II. Characteristics of surface depressions, J. Hydrol. 42, 77-90, 1979b. USACE, HEC-1 Flood Hydrograph Package: Users Manual. U.S. Army Corps of Engineers, Davis, CA., 1998. Vieux, B.E. & Farajalla, N.S., Capturing the essential variability in distributed hydrological modelling: Hydraulic roughness, Hydrol. Process., 8, 221236, 1994. Vijay, P.S. & Woolhiser, D.A., Mathematical modelling of watershed hydrology, J. Hydrological Eng., 7(4), 270-292, 2002. Wang, Z.M., Batelaan, O. & De Smedt, F., A distributed model for water and energy transfer between soil, plants and atmosphere (WetSpa), Phys. Chem. Earth, 21(3), 189-193, 1997. Wittenberg, H., Baseflow recession and recharge as nonlinear storage processes, Hydrol. Process., 13, 715-726, 1999. Wittenberg, H. & Sivapalan, M., Watershed groundwater balance estimation using streamflow recession analysis and baseflow separation, J. Hydrol., 219, 20-33, 1999. Wu, Y., Woolhiser, D.A. & Yevjevich, V., Effects of spatial variability of hydraulic roughness on runoff hydrographs, Agric. Forest Meteorol., 59, 231248, 1982. Yen, B.C., Channel Flow Resistance: Centennial of Mannings Formula, 43, Water Resources Publications, USA, 1992. Zinke, P.J., Forest interception studies in the United States, International Symposium on Forest Hydrology, eds., W.E. Sopper & H.W. Hull, 137-161, Pergamon Press, Oxford, 1967.

315