See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/303960899

Watershed Models

Chapter · January 2009

CITATIONS READS
0 10,426

1 author:

Vladimir Novotny
Northeastern University and Marquette Univerity
135 PUBLICATIONS 3,971 CITATIONS

SEE PROFILE

All content following this page was uploaded by Vladimir Novotny on 14 June 2016.

The user has requested enhancement of the downloaded file.

Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

WATERSHED MODELS

V. Novotny, Northeastern University, Boston, MA, USA

Introduction
Since 1960s computerized watershed models have been used by scientists to simulate the
hydrological, erosion and deposition throughout the watershed, and nonpoint pollution loads
from the watershed. Recently, ecological watershed models and applications have been
developed and are finding their way into practice. The term “watershed modeling” implicitly
describes a category of geographical models that simulate movement of water and associated
processes that change the quantity and quality of water. This type of modeling may differentiate
from other geographical ecological modeling such as the models of forest ecology (Band et al.,
1991), impact of drought conditions on crops and crop yields, habitat suitability for fauna and
flora (Wigmosta et al, 1994), etc.
A watershed is a geographical unit contributing flow to a location on a receiving water body.
The watershed is bordered by a watershed divide that surrounds the watershed, connecting the
highest points. The area behind the divide contributes surface and shallow groundwater flows to
another receiving water body. However, in many watersheds deeper groundwater flows may not
follow the surface geography of the watershed and groundwater contribution to flow may
originate from recharge areas beyond the divide or the groundwater flow originating within the
watershed may discharge into another receiving water body (Novotny, 2003).
The category “ecological watershed modeling” includes mostly mathematical computer
models. Figure 1 presents general categories of models for watershed modeling. Physical
watershed models, common in the 1940s, still do exist and are used by laboratories but will not
be extensively covered herein. In most cases these models have small ecological relevance and
have been primarily used for scaled down development, testing, and simulation of floods,
overland flow, rainfall/runoff transformation and erosion. The largest physical model covering
an area of dozens of hectares was built during the 1940’s in Vicksburg (Mississippi, USA). This
model represented the Mississippi River Valley watershed from Minneapolis to New Orleans and
was used for predicting flood flows and flood propagation. Many smaller subwatershed models,
as small as a platform, were built and are still used by scientists to study overland flow, washoff
of particulates that accumulate on impervious surfaces, and erosion and elutriation of pollutants
from soils. Often some ecological components were added such as various types of vegetative
cover to study and model erosion or growth of plants. Physical model category should be
distinguished from physically based models which is a model category synonymous to
deterministic-mechanistic computer models.
Distinction between small physical watershed models and pilot small watershed plots may be
sometimes fuzzy but one can presume that the former physical models are built in laboratories
and are detached from the natural systems while the latter are a part of nature equipped by
monitoring instrumentation. Analog models are rare and have been typically used for modeling
groundwater aquifer processes.
Today, watershed models are mathematical and built for computers. The computers
themselves may range from the typical desktop or laptop computers to high speed large
computers; however, the capacity of the current desktop or lap top computers can accommodate
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

most watershed models. In an ideal scenario, ecological models should include three spheres: (1)
atmosphere; (2) terrestrial, including shallow underground and riparian zones; and (3) aquatic
fresh water and oceanic systems (Figure 2). In this chapter the focus is on modeling terrestrial
processes that affect water quality and the integrity of receiving aquatic systems.

Figure 1 Classification of watershed models

Types of Models
The computer watershed models can be separated into two distinct categories (Figure 1):
1. Deterministic, also called mechanistic, models; and
2. Indeterministic models that include probabilistic models and artificial
intelligence (e.g., Artificial Neural Net or Genetic Algorithm models).
Deterministic (mechanistic) models are generically a priori developed from an assembly of
known processes, for example, hydrologic processes included in rainfall/runoff transformation,
accumulation of pollutants in the watershed, vegetation growth and hydrological and pollution
impact, etc. The model components are then tested by special field monitoring and in the final
stage, the entire model is then calibrated and verified by field data for modeling a specific
watershed. Indeterministic models are typically developed a posteriori from measured data. The
data assemblies should be large enough because the basic premise of these models is that the
models are developed from the data (data mining) and the previous knowledge about the
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

processes is only basic. Hybrid models, however, combine the knowledge based deterministic
features with data mining and model creation by artificial intelligence models. Stochastic models
yield both the mean and probability distribution of outputs (Kendall et al., 1999; Singh, 1995).
The most common stochastic modeling is Monte Carlo application, where a deterministic model
is run thousands times with randomly generated (based on their probability distribution) inputs
and model parameters to yield a probability distribution of the model output.

Figure 2 Spheres of ecological watershed modeling

Deterministic models do not consider random variables and for each unique set of input data
they produce fixed repeatable results. A mechanistic model is a representation of the physical,
biological, or mechanistic theory governing the system; in contrast, a statistical model accounts
for the statistical fitting of equations to the available data. Stochastic models use distribution for
each variable and parameter to generate random variables of model inputs and system parameters
(Law and Kelton, 1982; Clarke, 1998).
Every watershed model can be represented by a box (the watershed) concept (Figure 3). The
term “black box” signifies that not much is known a priori about the system and the model is
developed from the monitored data.
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Figure 3 Black box representation of watershed models (after Novotny and Olem, 1993)

Deterministic (Mechanistic) Watershed Models

Watershed Loading Models

As soon as the computer memory and speed of operation reached higher levels in the 1960s,
scientists focused on putting the knowledge of hydrological processes constituting rainfall –river
flow transformations together in a watershed hydrological model. The most known and most
likely the first larger and comprehensive watershed model was the Stanford Watershed Model
conceived by Professor Linsley and his graduate students (Linsley and Crawford, 1960;
Crawford and Linsley, 1966). The model was so sound that the concepts and the basic structure
of the model are valid today and incorporated into modern hydrologic watershed models.
Hydrology and hydrological processes are the backbone of all deterministic watershed models.
However, ecological models represent and simulate more than hydrology. Deterministic
watershed models have been developed and are used for simulating erosion and sediment
movement, pollutant loads, crop and forest growth and resulting loads of nutrients, spread forest
fires, surface – groundwater interactions, drainage, etc.
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

A detailed deterministic or stochastic ecological model has five components (Jørgensen and
Bendoricchio, 2001): (1) Inputs or Forcing Functions, (2) State Variables, (3) Mathematical
Equations, (4) Parameters of the Models, and (5) Constants. Inputs can be controllable or
uncontrollable. The model, like the real system, produces outputs to various inputs and the
outputs also reflect the changes in the system itself. The variables describing the system are
called state variables. The distinction between inputs and state variables is sometimes fuzzy but,
typically, watershed size, soil composition and erodibility, land use distribution, watershed
configuration with slope are examples of state variables while rainfall, atmospheric deposition,
temperature, humidity and solar radiation are inputs. Most watershed models are driven by
precipitation which is an uncontrollable input, while fertilizer application is a controllable,
managerial input. The foundation of the model is in equations that describe the input to output
transformation. The parameters of each equation may have ranges and the proper value of the
parameter must be typically established by calibration and verification of the model. The
equations also contain constants and thresholds. A threshold is a constant that activates or
terminates a process described by a particular equation or a submodel. For example, many
biodegradation processes cease when the temperature is near or below freezing or a priority
pollutant is not toxic below a certain concentration.
There are two approaches to modeling watershed processes: lumped parameter and
distributed parameter models (Figure 4). Lumped parameter models can be both stochastic and
deterministic. Distributed parameter models are mostly deterministic. The lumped parameter
models treat the watershed (subwatershed) as a homogenous unit. Because of a significant
variability of the parameters, even within a relatively small area (e.g., less than one hectare), the
various characteristics of the watershed are lumped together by an empirical equation, and the
final form and magnitude of the parameter are simplified to represent the computational
watershed as one homogenous unit. The computational units can be vertically compartmentalized
to represent surface-ground zones interactions. Water and mass flow from one compartment can
overflow into another adjoining compartment. For lumped parameter models the input – output
relationship can be represented as

Y = Φ (X) (1)
where X = input vector or matrix (single or multidimensional)
Y = output vector or matrix (single or multidimensional)
Φ = multidimensional transfer function

According to this definition, a lumped watershed model is a multidimensional transfer

function. The transfer function Φ does not represent a simple multiplication. In most general
terms, a transfer function is a functional response of the system to a pulse (in dynamic models) or
a step (in steady state) models input. A well known unit hydrograph incorporated in many
lumped parameter hydrologic models is a watershed flow response to a short duration (pulse)
unit rainfall while the Rational (Lloyd – Davies) Formula is the watershed flow response to a
step (continuous after initiation) uniform rainfall (Novotny, 2003).
In the distributed parameter models the watershed is divided into computational subunits that
are much smaller than those of the lumped parameter. The size of the subunits may range from 1
hectare to more than one km square. The subunit shapes are either rectangular (square) or
triangular or they may follow some morphological or hydrological feature of the watershed. Each
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

unit may receive external and internal (from surrounding units) input and is homogenous. The
general mass balance equation for any component within the computational unit is

Figure 4 Lumped (vector) and distributed (raster) paramater concepts for watershed modeling

dS
X Y G (2)
dt
where S = mass storage of the compound in the computational element
G = sinks (losses) and gains of the compound within the element
X = external input to the element, and
Y = output
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

The major difference between the lumped parameter and distributed parameter models is the
mathematical formulation of the mass balance in each unit that for distributed parameter models,
is generally expressed by a differential mass balance equation. The differential mass balance
equations for each computational unit are then solved simultaneously in a small computational
time interval, Δt, which ranges from minutes for the hydrologic calculation of a small watershed,
to a day or more for quality constituents or crop modeling in larger watersheds. Another
difference between the lumped and distributed parameter models is the availability of the output.
In lumped parameter models the output is available only at the watershed or subwatershed outlet.
In distributed parameter models, results are available for each computational unit which makes
distributed parameter models attractive for multidimensional displays in the raster GIS modeling
environment.
Watershed models come in all sizes, from one hectare uniform experimental modeling plots
or hydrological subunits to regional large models with a size of thousands km 2. Large watershed
models are dominated by channel storage while smaller watersheds are sensitive to precipitation
evens and are dominated by overland flow.
Models can be designed or run on an event or continuous basis. Event models simulate the
response of a watershed to a single large rainfall or another major input (e.g., forest fire). The
output from event models is a hydrograph of loads of a constituent from the watershed. The
principal advantage of event modeling over continuous modeling is that it requires little
meteorological data. The disadvantage of even modeling is that it requires a definition of the
“design” storm and antecedent state of the system prior to the event.
Continuous (dynamic) process modeling sequentially simulates processes such as
precipitation, water and pollutant storage and movement within the watershed towards the
watershed outlet, runoff components, evaporation and transpiration, uptake of pollutant and
nutrients by vegetation and their release into aquatic systems, and erosion. The major output is
the time series of flow and quality constituents from the watershed. The principal advantage of
continuous modeling is that it provides time series of the output (flow, concentrations, soil and
vegetation contamination). These time series can be statistically analyzed but with caution and
warning. The simulated time series are not the “true’ time series because the deterministic model
inherently do not include random components that are contained in every time series of natural
data. The models can be calibrated to simulate local means but not extremes. Thus the statistics
of exceedances or comparisons of the time series of outputs generated by deterministic models
with ambient standards could be misleading. This problem could be overcome with Monte Carlo
modeling methodology which would then categorize such applications as stochastic modeling.
Traditional deterministic watershed models are mostly linear. Linearity means that the
equations in the model compartments are either algebraic or differential linear equations. If the
mathematical formulation describing the process is nonlinear, it is often linearized within the
short computational interval. Under the assumption of linearity, the principles of proportionality
and superposition apply. The principle of proportionality implies that if the input is multiplied by
a constant the output is also multiplied by the same constant. The superposition means that if two
separate inputs to the model are added, the output of the model is the addition of the separate
outputs. The linearity of ecological models has been recently criticized (Carpenter et al, 1999)
and nonlinear models are being developed (Folke et al, 2002).
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Components of the Deterministic Watershed Ecological Models

Deterministic watershed models are compartmental models, i.e., they describe and simulate
the transfer of water and biomass from one compartment of the ecosystem to another. The
ecological models are almost always dynamic; they calculate mass balances in specific time
intervals that do not have to be identical for each compartment. For example, water balance and
overland water routing can be done in time intervals of an hour or less, while the growth of
plants or mass balances of water quality and algae in a receiving lake can be accomplished in
time intervals of weeks or months. The models simulate the fate of water and its quality
(chemical) constituents and also their direct and indirect effects on the resident organisms. The
mass balance can be calculated sequentially or concurrently. Watershed ecosystem models
employ a compendium of submodels (Figure 5).

Hydrological Component
Hydrological watershed models are the backbone of most deterministic watershed ecological
models. These models perform transformation of precipitation into water and water flow. The
watershed water balance models can be accomplished in a lumped sub watershed format (vector
representation in GIS) where computational subunits have more or less uniform characteristics or
in a distributed format (raster representation in GIS) (Figure 4). The water movement throughout
the watershed is calculated in the Lagrangian coordinate system of fluid mechanics in which z is
vertical direction, and X and Y are lateral axes. X can be in the main directions of flow, i.e., X
direction may not be perfectly horizontal. The hydrological mass balance is carried out vertically
(Z - direction) in three zones: (1) on the surface, (2) in the top soil, and (3) in the groundwater
zone. Each zone represents a storage component and the mass balance according to Equation 2.
There are many literature sources dealing with hydrology and hydrological modeling (see
Additional Reading).
The surface hydrology modeling component performs the following calculations.
Snowmelt calculation. Based on the ambient temperature, this component makes a decision
whether the precipitation is in a form of liquid rain or snow. It then calculates snow
accumulation and melting.
Initial subtraction from rain or snowmelt. Water that is intercepted by vegetation, surface
roughness or terrain is returned back to the atmosphere by evaporation.
Infiltration. Infiltration into soil depends on the top soil texture, moisture content, and
freezing.
Excess or net rain occurs when the precipitation (rain or snowmelt) at the soil surface
exceeds the sum of infiltration and surface subtraction. The surface water component accepts
solids with associated contaminants liberated from top soil by erosion and leaching Surface
runoff is excess rain transformed by overland routing. Surface runoff is the most polluted
component of flow in the receiving water bodies.
Top soil layer water balance has infiltration as input and evaporation, vegetation water
uptake (transpiration) and groundwater recharge as outputs. Groundwater recharge depends on
saturation permeability of subsoil. Interflow in the top soil occurs when the top soil layer is
saturated and infiltration exceeds the downward movement of water through the subsoil.
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Figure 5 Suite of basic submodels in watershed modeling. Solid line arrows represent transport processes
for water and contaminants (including sediment and nutrients), dot arrows are impact effects.

Groundwater recharge is a residual of infiltration into top soil storage evapotranspiration.

Groundwater flow is a near horizontal gravity movement of water from the recharge area to
surface outlets into the receiving waters. A smaller portion of groundwater is lost to deep
groundwater zones and to anthropogenic withdrawals.
The flow from the watershed and the contaminant loads into the receiving waters is a sum of
the surface runoff, interflow and groundwater (base) flow.

Interaction of contaminants with soils and surface vegetation.

This component of the deterministic ecological watershed models simulates retention and
attenuation of chemicals on soil particles of top soil and on dust and dirt particles that
accumulate on impervious surfaces of street gutters and other impervious areas. Chemical
contaminants are either hydrophilic that are mostly dissolved or dissociated in pore water of the
soil and in groundwater, or hydrophobic that have a high affinity for adsorption on soil
particulate matter or precipitate in soil. Nitrate (NO3-) is an example of a strongly hydrophilic
compound that moves readily with soil water into groundwater recharge and with groundwater.
Many pesticides and phosphate fertilizers are mostly hydrophobic and remain adsorbed onto soil
particles or precipitate into solid forms such as most toxic metals at a higher pH. However, a
small portion of these pollutants can exist in the dissolved form in the soil pore water. The
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

relationship between the concentration of the chemical in pore water of soil and the total
concentration is
CT
Cs  (3)
   mss
where Cs = concentration of the soluble contaminant fraction in the pore water of soil, μg/L
CT= total contaminant concentration in the soil volume, μg/L
θ = volumetric water content of the soil
Π = partition coefficient of the contaminant on soil particles, L/g, and
mss = soil density in g/L
Other isotherms used in modeling specific pollutants are Langmuir isotherm for phosphorus
(Novotny and Chesters, 1981) or Freundlich equations.
The soil component separates the contaminants into dissolved and adsorbed (precipitated)
fractions. Also some contaminants are biodegradable or can undergo biochemical transformation
(e.g., nitrification, biochemical break down) that often depends on the redox status of the soil
which is affected by soil moisture.
Crop/vegetation growth. This component is important for agricultural or forest growth
models. Plants withdraw soil moisture, use for build up of organic plant matter and release some
of the moisture to the atmosphere as transpiration. The withdrawal is limited by the wilting point
which is soil moisture content held by soil so tightly that plants cannot overcome it by their
suction. With water, plants can get their nutrients and also pick up contaminants. Plants can only
take up the dissolved mobile fraction. The adsorbed immobilized fraction is not available for
uptake.

Soil erosion component

Erosion moves sediment and contaminants adsorbed on or associated with the sediment from
the source area to the receiving water body. In modeling, sediment is routed overland by surface
runoff; however, not all eroded sediment will reach the receiving water body. The attenuation of
the sediment loads and associated contaminants during overland flow is expressed by a fuzzy
parameter called the delivery ratio, DR. Then the sediment yield in the receiving water body
becomes

 DR
N
Y i E i Ai (4)
i 1
where Y = total sediment yield from the watershed
Ei = erosion rate from the segment I, and
Ai is the area of the segment.
Erosion rate depends on the energy of the rainfall to dislodge soil particle from the soil. In
addition, erosion is activated only when excess rain occurs that will become surface runoff.
Hence, erosion is linked to the surface hydrology component. The most widely used erosion
model was developed by Wischmeier and Smith (1965). The model is more than forty years old
and has been modified but the fundamentals are sound. The erosion equation is

E = (R)(K)(LS)(C)(P) (5)

where E = calculated soil loss in tones/ha for a given storm or season or a year
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

R= rainfall energy factor that has the same units as R

K= soil erodibility factor
LS = slope/length factor
C = cropping management or vegetation cover (protection) factor, and
P = erosion control factor
The magnitudes methods of calculating the factors are presented in many publications (e.g.,
Goldman et al., 1986; Novotny, 2003).
The well known dynamic watershed hydrologic and water quality model, Hydrocomp
Simulation Program – Fortran (HSP-F) use an older Negev erosion model developed at Stanford
University (Negev, 1967; Johansen et al., 1984).
The Universal Soil Loss Equation considers the impact of Best Management Practices on the
loads sediment and associated pollutants.

Examples of Watershed Loading Models

HSP-F. This lumped parameter model evolved from the Stanford Watershed Model. It is
now included in the EPA’s watershed modeling suite BASINS (Better Assessment Science
Integrating Point and Nonpoint Sources). The model has several components. The components
used for modeling hydrology and pollutant loadings are The Agricultural Runoff Management
Model (ARM) and urban Nonpoint Simulator (NPS). Both are driven by the hydrological
component and the main difference between the two submodels is the soil and groundwater
modeling in ARM, and impervious surface hydrology, pollutant accumulation and washoff
incorporated in NPS. By including a channel component (RCHRES) HSP-F has a capability of
integrating the land and receiving water quality modeling. The result of the modeling is a time
history of flow and water quality at any point in the watershed (Johansen et al., 1984).
Agricultural Nonpoint Source Pollution Model (AGNPS) was developed by the US
Department of Agriculture (Young et al, 1987, 1989). The primary emphasis of the model is on
nutrient and sediment, and on comparing of various best management practices on the pollutant
loading.
Soil and Water Assessment Tool (SWAT) (Arnold et al., 1998) is a distributed parameter
version of previous models CREAMS (Knisel, 1980) and GLEAMS (Leonard et al., 1987).
SWAT is spatially distributed, up to several hundred of subbasins, and the subbasins can interact.
The SWAT hydrology is based on the water balance equations and uses National Resources
Conservation Service (formally Soil Conservation Service) runoff equation) (Soil Conservation
Service, 1986) expanded to consider also soil and groundwater movement. It can also simulate
irrigation and channel transmission. Erosion is calculated by the Modified Universal Soil Loss
Equation (MUSLE) (Williams, 1975) and includes overland and channel sediment routing.
SWAT GIS linkage incorporates advanced visualization tolls and enables statistical analysis of
outputs (Arnold et al., 1999).

Receiving Water Models

In deterministic ecological watershed modeling, each compartment should be described by a
model. Models of hydrology, contaminant transformation and movement throughout the
watershed provide estimates of the water and contaminant loads to the receiving water body.
The format of the loads is time variable hydrographs or histograms of the contaminants (plots of
the pollutant contaminant concentrations vs. time), daily loads, or seasonal (annual) loads for
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

larger impoundments with a longer residence time. A riparian zone models can be also
considered to provide information on attenuation of pollutant loads (Figure 6) (Inamdar, 2006).
Because the endpoint of the modeling effort is the ecology of the water body that receives the
loads from the watershed and from the point sources, a receiving water quality model should be
the last component of the modeling effort. Figure 6 shows a schematic of the ecological
receiving water body model. In many applications, the receiving water model is a stand alone
model which receives the inputs from the watershed loading model by the file transfer. In some
large water quality models (for example, HSP-F) the receiving water body model is a part of the
overall modeling package.
In, general, an ecological model is a expanded version of a deterministic water quality model
with added flora and fauna components. The first ecological models linked with water quality
were models simulating growth of algal population which signifies eutrophication. These models
can be steady state such as QUAL 2-E (Brown and Barnwell, 1987) and its derivatives, or
dynamic, such as WASP4 (Ambrose et al, 1990). Deterministic modeling of aquatic flora is more
demanding, more complex, and less reliable. An extensive review of models for receiving waters
and wetlands is in a publication by Straškraba (1995).
The key water quality components depicted on Figure 6 that impact the aquatic flora and
fauna are organic matter (both particulate and dissolved), suspended sediment , nutrients
(nitrogen and phosphorus), toxic (priority) pollutants and dissolved oxygen. Nutrients and toxic
compounds in the water column are both particulate (part of suspended sediment) and dissolved.
Only dissolved and dissociated nutrients and contaminates are available for uptake by flora and
fauna populations.
An important component of the ecological receiving water model is benthic sediment.
Legacy pollution is stored in the sediments as well as organic matter that is slowly anaerobically
decomposed. The benthic decomposition releases methane, ammonium and phosphates. The rate
of release depends on the redox status of the interstitial sediment-water layer. If this layer is
aerobic oxides reduce the release of phosphorus and ammonium is oxidized to nitrate and
methane to carbon dioxide. Oxidation of methane and ammonium results in the Sediment Oxygen
Demand (SOD) (DiToro et al., 1990). Nitrate, on the other hand, because of it concentration
gradient, infuses partially back to the lower anoxic layers of the sediment where it is reduced to
nitrogen gas. This simultaneous nitrification/denitrification results in a loss of nutrient nitrogen
from the water body.
The processes described by the model equations and simulated by the ecological models are
numerous and described throughout the encyclopedia. They include biodegradation of organic
matter imposing demand on oxygen resources in the water body, adsorption desorption
equilibriums of toxic compounds with the particulate organic matter and suspended sediment,
uptake of dissolved nutrients and toxics by flora and fauna that impacts their growth, growth and
die-off modeling for flora and fauna, dissolved oxygen balance model, settling and scour of
cohesive sediments, and nitrification. Many reactions and processes are temperature dependent.
There are several links of water quality constituents to the flora and fauna that are not yet
quantitatively understood. For example, one of the parameters that most affect the integrity of
fish and the macroinvertebrate population is embeddedness that describes the amount of clay
embedding the gravel and rocks in the bottom substrate. High embeddedness interferes with
spawning of fish. This parameter is observational, i.e., predictive models were not available at
the time of writing this article. It is related to the amount of suspended solids load of the
receiving water body and its geomorphic and hydraulic parameters (Virani et al., 2005).
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Figure 6 Schematics of the water-sediment ecological model

Example of ecologic water body models

AQUATOX is the latest in a series of models developed by Richard Park, starting with the
aquatic ecosystem model CLEAN that was subsequently improved and released by the US
Environmental Protection Agency by the US EPA (2005). AQUATOX simulates transfer of
biomass, energy and chemicals from one compartment of the ecosystem to another in daily
intervals. The model includes several interrelated components of aquatic ecosystems, including
multiple species of phytoplankton, periphyton and submerged aquatic vegetation, planktonic and
benthic algae, forage, Game and bottom fish, nutrients and dissolved oxygen, organic and
inorganic sediments, and toxic organic chemicals. AQUATOX is now an extension of BASINS
(USEPA, 2005) that provides linkages to geographical information system data, HSP-F and
SWAT simulation models.

Regression Based Statistical Models

Regression allows modeling the dependence of (usually) one response variable on one or
more predictor variables. The simple models are based on a stepwise progression of univariate
and multivariate analysis. Simple linear regression involves discovering the equation for a line
that most nearly fits the given data. These models are easier to test in replication and cross-
validation studies. Furthermore, they are less costly to put into practice in predicting and
controlling the outcome in the future. In watershed ecological modeling, multi-regression models
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

have been used to find relationships between the key chemical (e.g., nitrate concentrations in the
receiving water bodies) or biotic (e.g., indices of biotic integrity, diversity indices) and key
watershed stresses such as percent of polluting land uses (imperviousness, agriculture),
landscape parameters (slope), habitat quality, etc. A growing number of studies have established
relationships between some landscape metric and a biological endpoint (Allan, 2004).

Principal Component Analysis Multi-regression Models

More recent improvements of multi-regression statistic such as Principal Component
Analysis (PCA) can alleviate problems with cross-correlations between multiple inputs. Principal
component analysis enables grouping of a large number of multiple input parameters that may be
cross-correlated into fewer independent variables (Beach et al., 2005; Yuan and Norton, 2005).
The work and paper by Yuan and Norton described extraction of a multiparemeter model from
measured biotic integrity (response endpoint) and physical/habitat and chemical stressors
(inputs) in watersheds in western Ohio.
As any multiple-regression method, PCA can only analyze single parameter output matrix,
e.g., an overall IBI value or a single nutrient concentration vs. multiple parameter input vectors.
However, Indices of Biotic Integrity (fish, macroinvertebrates) are composites of and calculated
from multiple metrics. Extracting simultaneously knowledge on the metrics relation to the input
parameters and to the overall IBI would be very tedious and would have to be done individually,
metric by metric.

Canonical Correspondence Analysis (CCA)

CCA and similar correspondence analysis models are also special cases of multivariate
regression. CCA is a direct gradient technique that can relate species composition directly and
intermediately to the input environmental variables. CCA combines Correspondence Analysis
(CA) with multiple regression, whereby the measured endpoints (dependent parameters) are
related to measured environmental, landscape and habitat stresses (Ter Braak, 1986, 1987; Palmer,
1993). CCA and CA are weighted average ordination techniques that provide simultaneous
ordering of sites and species, rapid and simple computation and very good performance when
species have nonlinear and unimodal relationships to environmental gradients. CCA allows
simultaneous plotting of output species variables and their site scores in an ordination diagram
known as joint plot (Ter Braak 1994; 1995; Palmer, 1993). Figure 5 shows the two axis plot of the
25 variables affecting the Index of Biotic Integrity in Ohio streams. The plotting of variables when
appropriate scaling is used is represented by an arrow (vector) in the direction of the output
variables or clusters and the length of the arrow represents a measure of significance of the
variable. Furthermore, the direction of the arrows indicates not only how closely the input variable
is correlated with the output species variable, but also how closely the input variables are correlated
to each other. For two arbitrarily chosen input variables, if their arrow vectors are close to each
other or on the same line they are closely correlated. If they are normal (90º angle) to each other
they are uncorrelated. CCA is a powerful method for the multivariate exploration of large-scale
data (Hill, 1974). CCA preserves the chi- square (χ 2) distance between the rows and columns of
the contingency table (Legendre and Legendre, 1998). CCA is a weighted average ordination
technique that provides simultaneous ordering of sites and species, rapid and simple computation
and very good performance when species have nonlinear and unimodal relationships to
environmental gradients (Palmer, 1993).
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

The vectors of stressors on Figure 5 were combined with the clusters of metrics of the Index
of Biotic Integrity (Karr et al.,1986) to determine the Cluster Dominating Parameters. Clustering
of metrics was accomplished by Self Organizing Mapping (Kohonen, 1900) by the unsupervised
learning of the Artificial Neural Network (ANN) modeling. It can be seen that the inferior Cluster
3 sites were dominated by pollution caused by pollutants while the superior quality Cluster 1 sites
were dominated by habitat parameters (Virani et al., 2005).

Figure 7 CCA two axis plot of the watershed stressors on the metrics of the Index of Biotic Integrity.
Cluster 1 contains sites with good integrity, Cluster 2 is intermediate and Cluster 3 has inferior sites
(from Virani et al., 2005). The two axis represent about 50% of variability.

The explosive growth of information technology has given a major boost to the development
and implementation of modeling strategies in watershed management and restoration.
Geographic information systems (GIS) are becoming important components of simulation
models and decision systems, increasingly being used to store both geo- referenced data and
associated attributes. GIS technology has played critical roles in all aspects of watershed
management, from assessing watershed conditions through modeling impacts of human activities
on water quality and to visualizing impacts of alternative management scenarios. GIBSI
(Rousseau et al. 2000) is one such integrated modeling system comprising of physically-based
simulation models (hydrological, soil erosion, agricultural-chemical transport and water quality),
a relational database management system and a GIS.
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

21st Century Developments in Watershed Modeling

Novel techniques such as fuzzy logic, artificial neural networks and genetic algorithms are
increasingly tested against ecological data (Recknagel, 2001). The advantage of these approaches
is the self selection of the critical model inputs based on their data-driven approach.
The Land Transformation Model (LTM) employs GIS and Artificial Neural Networks to
describe the influence of landscape changes on ecosystem integrity of large areas (Pijanowski et
al., 2000). The LTM currently employs a multiplayer perceptron (MLP) neural net topology with
one or two hidden layers; each layer has at least the same number of nodes as the number of
input vectors. ANNs are used to learn the patterns of development in the region and test the
predictive capacity of the model, while GIS is used to develop the spatial, predictor drivers and
perform spatial analysis on the results (Pijanowski et al. 2002).
New tools such as global positioning systems (GPS) and remote sensing are also being
developed to inventory and monitor watershed characteristics (Guertin et al., 2000). The tools
ArcPad, which are developed in the ArcPad environment, try to implement GISs, mobile
computing systems, satellite and aerial images, network interconnection in the frame of standard
ecological methodology (Matejicek, 2003). All these tools provide a better understanding of how
ecological systems are managed on various levels of the ecological research.
The reasons for the new look at watershed modeling and the new direction of modeling
development are:
1. Key processes of hydrological and ecological systems to be modeled are nonlinear
(Walker et al, 2002; Gunderson, 1997, Folke et al, 2005) while the current deterministic
watershed models are mostly linear (described by linear differential equations) or the equations
that originally were not linear (e.g., adsorption/desorption of pollutants on soils and particulate
matter) were simplified to linear representations. While a non-linear watershed model can be
deterministic, the nonlinearity complicates the model structure and solutions of the equations.
2. The typical impacts of contaminants and other stressors on flora and fauna have
thresholds, i.e., a minimal or even beneficial (e.g., nutrients, some metals) concentration effect
on the biota below the threshold limit and nonlinear detrimental effect on biota beyond the
toxicity threshold (Muradian, 2001). This leads to sudden shifts in species occupying the
terrestrial and aquatic environments and clustering (Virani et al, 2005). Ecological clustering
describes a similar and steadier response (resilience in species population or diversity) of the
aquatic and terrestrial biota to smaller perturbation in input stresses. Clusters may also be
sensitive to different stressors as shown on Figure 7 for the metrics of the fish Index of Biotic
Integrity. Most deterministic models, as stated before, are proportional, i.e., they result in a
corresponding change in populations when any input changes unless a threshold (an “if – then”)
block is incorporated into the model.
3. Large data bases of ecological (e.g., fish, macroinvertebrate, phytoplankton, periphyton
numbers, composition, and indices of biotic integrity), chemical, and habitat quality data have
been collected by the many agencies. The mega databases containing hundreds of thousands of
measurements over larger, often multinational regions cannot be modeled by classic
deterministic models that are mostly suitable for smaller watersheds. The models and the
relationship must be retrieved by “data mining”.
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

ADDITIONAL READING
Abbott, M.B., and J.C. Refsgaard (1996) Distributed Hydrological Modeling, Kluver Academic
Publishers, Netherlands
Chapra, S.C. (1997) Surface Water Quality Modeling, McGraw Hill, New York
Chow, V.T., S.R. Maidment, and L.W. Mays (1988) Applied Hydrology, McGraw-Hill, New
York, NY
DeBarry, P.A. (2004) Watersheds: Processes, Assessment and Management, Wiley, New York,
NY
DiToro, D.M. (2000) Sediment Flux Modeling, Wiley Interscience, New York
Krenkel, P. and V. Novotny (1980) Water Quality Management, Academic Press, New York
Jørgensen, S.E. (1994) Fundamentals of Ecological Modeling, 2nd.ed., Elsevier, Amsterdam
Legendre, P., and L. Legendre (1998) Numerical Ecology, 2nd, ed., Elsevier, Amsterdam
Lehr, J.H., and J. Keeley (2005) Water Encyclopedia, Wiley, Hoboken, NJ
Maidment, D.R. (1993) Handbook of Hydrology, McGraw-Hill, New York, NY
Novotny, V. (2003) WATER QUALITY: Diffuse Pollution and Watershed Management. J. Wiley
& Sons, New York, NY
Ramaswami, A., J.B. Milford, and M. J. Small (2005) Integrated Environmental Modeling:
Pollutant Transport, Fate, and Risk in the Environment, Wiley, Hoboken, NJ
Salomons, W., and U. Főrster (1984) Metals in the Hydrosphere, Springer Verlag, Berlin
Schnoor, J.L. (1996) Environmental Modeling: Fate and Transport of Pollutants in Water, Air,
and Soil, Wiley, New York
Singh, V.P. (2995) Watershed Models, CRC Press, Boca Raton, FL

REFERENCES
Allan, J.D. (2004) Landscapes and riverscapes: The influence of land use on stream ecosystems,
Ann. Rev. Ecol. Evol. Sys. 35:257-284`
Ambrose, R.B., Jr., T.A. Wool, J.P. Connolly, and R.W. Schranz (1990) WASP4, A
Hydrodynamic and Water Quality Model, Model Theory, User’s Guide, and Programmer’s
Guide, Environmental Research laboratory, US Environmental Protection Agency, Athens, GA
Arnold, J.G., R. Srinivasan, R.S. Muttiah, and J.R. Williams (1998) Large scale hydrologic
modeling and assessment, Part I: Model development, J. American Water Resources Association
34(1):73-89
Arnold, J.G., R. Srinivasan, R.S. Muttiah, P.M. Allen, and C. Walker (1999) Continental scale
simulation of the hydrologic balance, J. American Water Resources Association 35(5):1037-1052
Beach N.D. and V. Novotny (2005) Modeling In-Stream Nitrogen Concentrations Based on
Drainage Area Characteristics and Principal Components Analysis, Proc. AWRA National
Conference, Seattle, WA, November 2005
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Band, L. E., D. L. Peterson, S. W. Running, J. C. Coughlan, R. B. Lammers, J. Dungan, and R.

Nemani (1991) Forest ecosystem processes at the watershed scale: basis for distributed
simulation, Ecol. Model., 56: 151-176
Brown L.C., and Barnwell T.O., Jr. (1987) The Enhanced Stream Water Quality Models QUAL2E
and QUAL2E-UNCAS. Documentation and User Manual, Report EPA/600/3-87/007, U.S.
Environmental Protection Agency, Athens, GA
Carpenter, S., W. Brock, and P. Hanson (1999) Ecological and social dynamics in simple models
of ecosystem management, Conservation Ecology 3(2):4
Clarke, R. T. (1998) Stochastic Processes for Water Scientists: Developments and
Applications. John Wiley & Sons, Inc., New York, 183pp
Crawford, N.H. and R.K. Linsley (1966) Digital Simulation in Hydrology: Stanford Watershed
Model IV, Stanford Univ., Dept. Civ. Eng. Tech. Rep.
39, Palo Alto, CA

DiToro, D. M., P.R. Paquin, K. Subburamu, and D. Gruber (1990) Sediment oxygen demand
model: Methane and ammonia oxidation, J. Environmental Engineering 116(5):945-986

Folke, C. et al. (2002) Resilience and Sustainable Development – Building Adaptive Capacity in
a World of Transformation, The Environmental Advisory to the Swedish Government,
Stockholm
Folke, C., S. Carpenter, B. Walker, M. Scheffer, T. Elmquist, L. Gunderson, and C.S. Holling
(2005) Regime shifts, resilience, and biodiversity in ecosystem management, Ann. Review of
Ecology, Evolution and Systematics, 35:557-581
Goldman, S.J., K. Jackson, and T.A. Bursztynsky (1986) Erosion and Sediment Control
Handbook, McGraw-Hillm New York, NY

Guertin, D.P., S,N, Miller, and D.C. Goodrich,(2000) Emerging tools and technologies in
watershed management, USDA Forest Service Proceedings RMRS-P-13: Proceedings of the
Conference on Land Stewardship in the 21st Century: The Contributions of Watershed
Management, Tucson, AZ, pp. 194-204.
Gunderson, L. (1999) Resilience, Flexibility, and Adaptive Management—Antidote for Spurious
Certitude? Ecology 3(1):7
Inamdar, S. (2006) Challenges in modeling hydrologic and water quality processes in riparian
zones, Journal AWRA 42(1):5-14
Johanson, R.C., J.C. Imhoff, J.L. Kittle, and A.S. Donigian (1984) Hydrologic Simulation
Program-Fortran (HSPF): User’s Manual, Release 8, EPA 600/3-84-006, US Environmental
Protection Agency, Washington, DC

Jørgensen, E.S., and G. Bendoricchio (2001) Fundamentals of Ecological Modeling, Third

edition. Elsevier, 530 pp

Karr,J.; K.D. Fausch, P.R. Angermeier, R. Yant, and I.J. Schlosser (1986). Assessment of
Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Biological Integrity in Running Water: A Method and its Rationale. Illinois Natural History
Survey Special Publication No.5., Champaign.IL. 28pp.
Kendall, B.E., Briggs, C.J., Murdoch, W.W., Turchin, P., Ellner, S.P., McCauley, E., Nisbet,
R.M., and Wood, S.N. (1999) Why do populations cycle? A synthesis of statistical and
mechanistic modeling approaches. Ecology, 80: 1789-1805
Kohonen,T. (1990). Self-Organizing Maps.2nd edition. Springer Verlag, Berlin
Knisel, W.G. (1980) CREAMS: A Field Scale Model for Chemicals, Runoff, and Erosion from
Agricultural Management Systems, Conservation Research Report No. 26, USDA, Washington,
DC
Law, A. M., and Kelton, W. D. (1982) Simulation Modeling and Analysis. McGraw-Hill Inc.,
400pp
Legendre, P., and L. Legendre (1998) Numerical ecology. Elsevier Science BV, Amsterdam
Leonard, R.A., W.G. Knisel, and D.A. Smith (1987) GLEAMS: Groundwater Loading Effects of
Agricultural Management Systems. Transactions ASAE 30(5):1402-1418
Linsley, R.K. and N. H. Crawford (1960) Computation of a Synthetic Streamflow Record on a
Digital Computer, Int. Assoc. Sci. Hydrol. Publ. 51, pp. 526-538
Matejicek, L. (2003) Development of software tools for ecological field studies using ArcPad,
ESRI 24th Annual International User Conference, San Diego, California, pp. 12.
Muradian, R. (2001) Ecological thresholds: a survey, Ecological Economics 38:7-24
Negev, M., Sediment Model on a Digital Computer, Stanford Univ., Dept. Civ. Eng. Tech. Rep.
76, Palo Alto, Ca
Novotny, V. (2003) WATER QUALITY: Diffuse Pollution and Watershed Management. J. Wiley
& Sons, New York, NY
Novotny, V., and G. Chesters (1981) Handbook of Nonpoint Pollution: Sources and Management,
Van Nostrand-Reinhold, New York, NY
Novotny, V., and H. Olem (1994) WATER QUALITY: Prevention, Identification and Management
of Diffuse Pollution, Van Nostrand-Reinhold Publ., New York, NY, 1994, J.Wiley & Sons, New
York, 1997
Palmer, M.W (1993) Putting things in even better order: The advantage of Canonical
Correspondence Analysis, Ecology. 74: 2215, (1993).
Pijanowski, B.C., S.H. Gage, D.E. Long, and W.E. Cooper (2000) A land transformation model
for the Saginaw Bay Watershed. In:, Landscape Ecology: A Top Down Approach (J. Sanderson
and L. Harris, eds.). CRC Press., Boca Raton, FL, pp. 183-198.
Pijanowski, B., D. Brown, B. Shellito, and G. Manik (2002) Using neural networks and GIS to
forecast land use changes: A land transformation model. Computers, Environment and Urban
Systems, 26(6): 553-575.
Recknagel, F. (2001) Applications of machine learning to ecological modelling. Ecological
Modelling 146: 303-310.
View publication stats

Draft of a chapter in Encyclopedia of Ecology (SE Jorgensen and B. Fath, eds.), Elsevier 2009

Rousseau, A.N., Mailhot, A., Turcotte, R., Duchemin, M., Blanchette, C., M., Roux, Etong, N.,
Dupont, J., and Villeneuve, J.P. 2000. GIBSI: An integrated modelling system prototype for river
basin management. Hydrobiologia, 4223: 465-475.
Singh, V. P., 1995. Watershed modeling, In Computer Models of Watershed Hydrology (V. P.
Singh, ed.). Water Resources Publications, Highlands Ranch, Colorado. p. 1-22.
Soil Conservation Service (1986) Urban Hydrology for Small Watershed – TR55, USDA,
Washington, DC
Straškraba, M. (1995) Models for reservoir, lakes and wetlands, in Remediation and Management
of Degraded river Basins (V. Novotny and L. Somlyódy, eds.), Springer Verlag, Berlin
Ter Braak, C. J. F.(1986). Canonical Correspondence Analysis: a new eigenvector method for
multivariate direct gradient analysis. Ecology 67:1167–1179
Ter Braak, C. J. F.(1987). The analysis of vegetation-environment relationships by canonical
correspondence analysis. Vegetatio, 69: 69-77
Ter Braak, C. J. F. (1994).Canonical community ordination Part I: Basic theory and linear
methods. Écoscience, 1: 127-140
US Environmental Protection Agency (2005) AQUATOX (Release 2.1): Modeling environmental
fate and ecological effects in aquatic ecosystems, EPA-823-F-05-007, Office of Water,
Washington, DC (downloadable from USEA)
Virani,H., E. Manolakos, and V. Novotny (2005) Self Organizing Feature Maps Combined with
Ecological Ordination Techniques for Effective Watershed Management - Technical Report # 4,
Center for Urban Environmental Studies, Northeastern University, ww.coe.neu.edu/environment
Walker. B., S. Carpenter, J. Anderies, N. Abel, G. Cumming, M. Jansesen, L. Level, J. Norberg,
G. D. Peterson, and R. Prichard (2002) Resilience management in social-ecological systems: a
working hypothesis for a participatory approach, Conservation Ecology 6(1):14
Williams, J.R. (1975) Sediment routing for agricultural watersheds, Water resources Bulletin
11(5):965-974
Wigmosta, M. S., L. W. Vail, and D. P. Lettenmier, 1994. A distributed hydrology-vegetation
model for complex terrain. Water Resources Research, 30:1665-1680.
Wischmeier, W.H., and D.D. Smith (1965) Predicting Rainfall-erosion Losses from Cropland East
of Rocky Mountains, USDA Agricultural handbook No. 282, Washington, DC
Young, R., C.A. Onstad, D.D. Bosch, and W.P. Anderson (1987) AGNPS: Agricultural Non-Point
Source Pollution Model: A Watershed Analysis Tool. USDA-Agricultural Research Service,
Conservation Research Report 35, Department of Agriculture, Washington, DC
Young, R., C.A. Onstad, D.D. Bosch, and W.P. Anderson (1989) AGNPS: A nonpoint-source
pollution model for evaluating agricultural watersheds, J. of Soil and Water Conservation,
44(2):168-173
Yuan, L.L. and S.B. Norton (2003) Assessing the relative severity of stressors at a watershed
scale, Environmental Monitoring and Assessment, 98:323-349