Boundless Electrical Resistivity Tomography BERT the user tutorial

Thomas Gnther & Carsten Rcker u u March 16, 2009 version 1.0

In this tutorial we like to show the users of the BERT inversion how to work with our software. Some small but instructive examples, all real eld cases, are presented to show how the dierent options in the conguration le can be used to yield case-specic inversion results. The examples start from 2d inversion of surface measurements with and without topography. We show how to include structural information and how buried electrodes are handled. 3d inversion is done on a at surface and topography. For the latter two main techniques are presented. Also, measurements on closed objects, such as trees, humans, soil columns and model tanks are shown. Finally we show how to handle time-lapse resistivity measurements. The user is invited to follow by processing the data in the examples directory.

Leibniz Institute for Applied Geophysics, Hannover Institute of Geology and Geophysics, University of Leipzig

Contents
1 Introduction 1.1 BERT, DCFEMLib, GIMLi - Nomen est omen . . . . . . . . . . . . . . . . . . 1.2 Options and commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2D 2.1 2.2 2.3 2.4 2.5 2.6 surface measurements A at surface example . . . . Regularisation and data t . Mesh quality and renement Incorporation of topography . Structural constraints . . . . A sophisticated example: The 3 3 4 5 6 6 7 7 8 9 10 12 13 13 15 15 16 19 19 19 20 20 21 21 22 23 24 25 27

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . lake case

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

3 3D surface measurements 3.1 Flat surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 3D Topography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Closed geometries 4.1 2d example - tree tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 3d closed geometry - model tanks . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Buried electrodes 5.1 2D cross-hole data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 3D-Crosshole measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Topography and buried electrodes . . . . . . . . . . . . . . . . . . . . . . . . . 6 Time-lapse ERT 6.1 Crosshole timelapse measurements . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Soil column measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Concluding remarks and outlook A BERT for Windows users B Files and programs C Complete list of options and their default values D BERT using GIMLi

1 Introduction
1.1 BERT, DCFEMLib, GIMLi - Nomen est omen
Direct current electrical measurements are used in a wide range of applications such as medical imaging, geophysical surface or subsurface measurements or the investigation of trees and soil probes. This inverse problem is known under the terms ERT (electrical resistivity tomography), ERI (... imaging), EIT (... impedance tomography) or DC resistivity inversion. The aim of our software is to present a unique solution for all these cases. Main advantage is the possibility to work on arbitrary geometries. Therefore we decided to consequently use unstructured nite element meshes for forward calculation as well as for the parameter identication. By the use of triangles (2d) and tetrahedrons (3d) we can follow any geometry of the subsurface, probe or any structural information we have about the object to be investigated. Due to this generality we decided to call it BERT - Boundless Electrical Resistivity Tomography. BERT is a part of the software library DCFEMLib - Direct Current Finite Element Method Library. It is currently newly developed and will be called GIMLi - Generalized Inversion and Modelling Library. The latter is a much more general approach for arbitrary tomographic methods, as the functionality is growing BERT will more and more migrate to GIMLi1 . DCFEMLib is licensed under the GPL (GNU public license). Our aim is to give back to the academic community without letting people earning money by just clicking a button. Additionally to the library itself it comprises executables for ERT modelling and inversion, tools for generating mesh input (the so-called poly-tools) and tools applying the mesh generator. The theory and technology of BERT is described in our GJI paper (Gnther et al., 2006) u and bases on the nite element modelling techniques described by Rcker et al. (2006). It u is represented by a triple-grid scheme: Most inversion algorithm use a dual-grid scheme, i.e. the forward calculation is calculated on a ner mesh. We add another one in order to use a secondary eld approach and thus have a very fast forward calculation. Figure 1 shows the three grids: On a coarse and resolution-dependent grid the parameters are dened. On a globally rened and prolonged mesh the forward calculation is done. And a very ne primary mesh is used to calculate the primary potentials (for a homogeneous subsurface), but only once directly after the mesh creation.

Figure 1: The three meshs of inversion for a 2d example, from Gnther et al. (2006) u The overall scheme is visualised in Figure 2. It starts with the generation of the three meshes. Then the primary potentials are calculated and interpolated onto the secondary mesh. From this geometric factors are derived yielding the apparent resistivity and the sensitivity matrix is created for the homogeneous case. Finally the actual inversion is carried out: An inverse
1

See appendix for doing ERT inversions with GIMLi.

sub-problem is used to update the resistivity model, a forward calculation is carried out and checked against the data. The latter is done until the data are tted well or the process stagnates.
Topography Electrodes Data

Parameter mesh

Primary mesh

Configuration factor

Secondary mesh

Primary potential Inversion

Apparent resistivity

Constraints

Sensitivity

Forward operator

Inverse subproblem

Check

Solution

Figure 2: The BERT inversion scheme, from Gnther et al. (2006): the geometrical information u is used to prepare the actual inversion (rectangle). BERT is available under Linux and Windows2 , either from pre-compiled binaries or selfcompiled code3 . The paths to the binaries and the library must be known, e.g. by setting $ export PATH=$PATH:/path/to/dcfemlib/bin $ export LD LIBRARY PATH=$LD LIBRARY PATH:/path/to/dcfemlib/lib

1.2 Options and commands

The inversion itself is controlled by the program invert, which reads the so called conguration (cfg) le. In the cfg le all necessary information is stored in form of lines consisting of KEY=value type, as in bash everything behind the #-sign is ignored and can be used for comments. Note that the key must be uppercase. There is only one mandatory key: the DATAFILE key holding the name of the data le. Other important keys are DIMENSION (2 or 3) and TOPOGRAPHY (0 or 1, meaning false or true). We suggest to create a new directory for each problem or also for dierent strategies to solve it. The data le must be in the unied resistivity.net format (see www.resistivity.net?unidata), but so far without the topography part and the end. DC2dInvRes and DC3dInvRes (graphical FD-based inversion software also available on www.resistivity.net) can easily be used to convert from dierent formats, such as res2dinv or dierent equipment les (ABEM,Syscal,Geotom,Sting,4pointLight). Use the Save ohm le call for doing so. For list of possible options (with default values) see appendix C or call invert opts. However only few of them are of frequent use4 . In order to create a new project there are special
2 3

See appendix A for using BERT in Windows. See www.resistivity.net for information about how to obtain the binaries/codes and to compile the code 4 In order to change default options permanently, we suggest to create a le $HOME/.bertrc that will be read if present. Typical entries are, e.g., SENSMATMAXMEM=3000 (available memory in MB), LINSOLVER=2

commands for the individual tasks, namely invertNew2d, invertNew2dTopo, invertNew3d, invertNew3dTopo, invertNew2dCirc and invertNew3dCyl for the cases 2d/3d with or without topography and for circle/cylinder geometry. For example, $ invertNew2d datafile.dat > inv.cfg creates a new conguration le inv.cfg with the lines DATAFILE=datafile.dat, DIMENSION=2 and TOPOGRAPHY=0, but also adds a lot of possible options for this case with an explanation, most of them inactive/commented. The user can now (or later) change the options and run single steps or the whole of inversion by $ invert cfgfile commands where cfgle is the conguration le and commands can consist of the following: all makes all, that is probably the rst step in most (at least small) cases meshs just makes the meshes, so that these can be optimized (suggested for bigger problems) nomeshs do everything else but the meshes (after a successful mesh generation) primPot interpolate calculates primary potentials and interpolates it to the secondary mesh calcSensM actual inversion including sensitivity calculation (rst inversion) calc inversion with changed inversion options newPara after creating a new parameter model (keeps primary mesh and potential) save saves all important results (model&response for each iteration, log le, cfg le, meshes) in a directory called result<date> <time> clean cleans the directory from temporary results mrproper deletes all stu except input and result directories (releases disk memory fully) show creates a vtk le (can be used in the main directory or in a result directory)

1.3 Visualization
BERT creates a lot of les that can be used for post-processing, see appendix. You can visualize in-place or save the state rst using invert inv.cfg save. A result-directory is created out of the model le data and time and the les are moved in. Later you can change into the directory to do the visualization. There are several ways for doing it: 1. The command invSummary creates a pdf with all essential information and gures. This requires DCMatLab installed and is a easy method but only feasible for 2D. 2. You can use DCMatLab directly using >> Mesh=loadmesh(mesh/meshParaDomain.bms); %loads mesh into struct >> res=load(model_iter.final); %loads ascii resistivity vector >> tripatchmod(Mesh,res); % help tripatchmod for options

(for LDL instead of CHOLMOD or taucs), or favourite mesh options

3. There is a graphical tool called bert2dpost for doing it automatically, where you can specify some options. 4. For 3D but also 2D you can call invertinv.cfg show which will create a le showmodel.vtk and opens ParaView with it. After show you can specify a dierent model le. Paraview5 is a highly sophisticated and free tool for 3d (and 2d) visualisation. It handles both complete inversion results (obtained by invert cfgle show), mesh inputs (*.poly, to be converted by polyConvert -V -S to vtk or stl) and meshes (to be converted by meshconvert, if not already done). In order to have a quick look at 3d meshes when nding optimal parameters, we also suggest MEDIT6 , a free and fast 3d viewer. The usual proceeding is (i) create meshes and look whether they are suitable, (ii) run a full inversion with all, (iii) look at the results (show, ParaView), (iv) rene options and do a calc, save good results using save. The rst step is usually omitted for small projects. We now start to prepare and run inversions using the exemplary les in dcfemlib/examples/inversion, starting from easy-going to complicated ones.

2 2D surface measurements
2.1 A at surface example
The example in dcfemlib/examples/inversion/2dat/gallery was friendly provided by the University of Mining and Technology, Freiberg (F. Donner). It is a very small prole over a known mining gallery that is used for draining water out of the mines. It represents a perfect two-dimensional anomaly since it strikes perpendicular to the prole and is a 2x2m cavity. On a prole using 21 electrodes with 2m spacing, dipole-dipole measurements have been applied, the data quality was very good. The input in the data le gallery.dat is already the apparent resistivity. Standard deviations have been measured and 1% were added. We create a conguration le using $ invertNew2d gallery.dat > inv.cfg just holding the data le and the dimension. Figure 3 shows the result of invert inv.cfg all with default options that converged with a chi-squared value of 0.99 (data tted within error bounds). It clearly shows the cavity at about 20 m and another anomaly whose origin is not completely clear from 2d measurements. See section 3.1 for inversion of a 3d data set.
0 z/m 4 6 8 [email protected] 0 5 10 15 20 25 30 x/m 40

Ohmm 100 126 158 200 251 316 398

Figure 3: Result of the gallery example using default options.

5 6

see ParaView home page www.paraview.org INRIA Rocquencourt, see http://www-rocq.inria.fr/gamma/medit/medit.html

2.2 Regularisation and data t

We now might to change the characteristics of the model. The most important key for that is the regularization parameter LAMBDA. It controls the strengths of the smoothness constraints and thus denes how smooth the model will be. Therefore we test dierent values by commenting out the line with the parameter LAMBDA and setting it once to 200 and once to 2 instead of the default value of 20. After changing it we call $ invert inv.cfg calc and save the results as well using save.
0 z/m 4 6 8 [email protected] 0 5 10 15 20 25 30 x/m 40 0 z/m 4 6 8 [email protected] 0 5 10 15 20 25 30 x/m 40

Ohmm 100 126 158 200 251 316

Ohmm 79.4

100

126

158

200

251

316

398

501

Figure 4: Result for a regularization parameter of 200 (left) and 2 (right). Figure 4 shows the result for the two cases. The one for = 200 is clearly over-smoothed and cannot t the data appropriately (2 = 4.4, RMS=2.5%). The low value can t the data but produces too much small-scaled anomalies that are not necessary. Note that these data have a very good quality documented in the error values between 1 and 2.3%. An over-estimation of the errors will lead (if 2 = 1) to an over-smoothed result as well. Errors (in the meaning of how well can I t the data) should therefore be measured or estimated by reciprocal data or repetition. Always add a xed value that will account for error sources outside our measurements7 . In the absence of errors in the data le an estimation is made using a xed percentage (INPUTERRLEVEL) and a voltage error (INPUTERRVOLTAGE). If the current is not in the le, a value of 100 mA is assumed. For dierent current strengths the voltage error has to be adapted. Alternatively to a xed regularization it can be optimized using the L-curve by LAMBDAOPT=1. To decrease the regularization parameter in the course of iterations (cooling scheme) by, e.g., the factor 0.8 use LAMBDADEDREASE=0.8.

2.3 Mesh quality and renement

The depth of the modelling domain is by default automatically determined based on 1d sensitivity studies, but can be adjusted using PARADEPTH. The value of PARABOUNDARY denes how far (in % of the extension, default=5) the boundary is outside of the electrodes. You might of course think that the triangles in Figure 3 are quite coarse and that we can resolve the cavity better. Although we dont need it in this case according to Occams razor, we illustrate how to do it. Generally, all meshes must be ne near the electrodes (to achieve high accuracy in the forward calculation and because the resolution is high there) and coarse at the boundaries. The neness is achieved by placing additional nodes in the vicinity of the electrodes, PARADX holds the distance (in electrode spacings a). A value of 0.2 means two points are created 0.2a to the right and left of each electrode. By EQUIDISTBOUNDARY=1 several points are equally distributed.
7

Therefore DCxdInvRes ask for an error percentage to be added to the measured if those are too small.

Usually the points are linearly interpolated, SPLINEBOUNDARY=1 forces a spline interpolation, which is useful for round geometries or smooth topography. The coarse-ness towards the boundary is controlled by the mesh quality PARA2DQUALITY, which denotes a minimum angle. The higher the quality is, the more accurate are the results but with an increasing number of cells and thus run-time. In triangle8 (Shewchuk, 1996) version 1.6, our favoured 2d mesh generator, the range goes from 25-30 (bad quality) to 34-35 (good quality). Default value is 33.8, a trade-o between accuracy and run-time. Another way of avoiding a too coarse mesh is the maximum cell size by setting PARAMAXCELLSIZE (in m2 ).
0 z/m 4 6 8 0 5 10 15 20 25 30 x/m 40 0 z/m 4 6 8 0 5 10 15 20 25 30 x/m 40

0 z/m 4 6 8
0 z/m 4 6 8

10

15

20

25

30

x/m

40

0 z/m 4 6 8

10

15

20

25

30

x/m

40

10

15

20

25

30

x/m

40

0 z/m 4 6 8

10

15

20

25

30

x/m

40

Figure 5: Meshes with dierent parameters: default values dx=0.2 & q=33.8 (top left), dx=0.2 & q=33 (top right), dx=0.2 & q=34.5 (center left), dx=0.2 equidist & q=33.8 (center right), dx=0.3 equidist & q=33.8 (bottom left), dx=0.5 equidist & q=33.8 (bottom right) In Figure 5 the resulting parameter meshes for dierent settings are displayed. A lower quality is only recommended in case of topography. A too large quality value produces nicer triangles but can lead to heterogeneous mesh density even for equidistant electrodes. The EQUIDISTANT switch produces a nice near-surface triangulation that can yield too much parameters. The latter can be avoided by a larger PARADX value. Finally the user has to nd the optimum settings for his case (and taste).

2.4 Incorporation of topography

In 2d inversion, topography is easily integrated by setting the heights of the electrodes. All the rest should be done automatically, if necessary, additional electrodes must be inserted. However, rarely all electrodes will be measured topographically. Often it is sucient to have a few points. Note that in the current stage BERT requires the topographical information in the rst section, not at the end of the le. For this case we recommend the use of DC2dInvRes (Gnther, 2007) that will roll the positions along the surface. For this task, use Data:Save u Ohm le and specify whether the x values are along measure tape or real x. The slagdump prole The data in dcfemlib/examples/inversion/2dtopo/slagdump were friendly provided by the Federal Institute of Geology and Natural Resources (BGR), Hannover. It was measured by
8

see http://www.cs.cmu.edu/~quake/triangle.research.html

M. Furche and is one of several proles over a slag dump. A Wenner array with a = 2 m spacing was applied yielding 222 data points. The topography was measured at 8 points by levelling and appended to the original le before converting it with DC2dInvRes. We initialize the inversion with standard options using the command: $ invertNew2dTopo slagdump.dat > inv.cfg Note that this produces additional lines in the le for the primary mesh. Similar to the parameter mesh, the PRIMDX value species the (absolute, not relative!) renement at the electrodes and PRIM2DQUALITY denes the mesh growth. Additionally, we decide by PRIMP2MESH whether the primary potentials shall be computed by quadratic shape functions. As stated by Rcker u et al. (2006) those are more ecient. The necessary renement for a P2 mesh is about a/10 and a/100 for a P1 mesh. Figure 6 shows the topographic eect and the resistivity image. The upper pseudo-section shows the apparent resistivities using at-earth geometry factors. The middle one is the pure geometric eect, t = Gf lat utopo ( = 1S/m), and the lower image is based on the geometric factors out of the primary potential. Several anomalies can be explained solely by topographical undulations. The resistivity distribution shows a conductive interior and a resistive hard pan.
0 we2 we4 we6 we8 we10 we12 10 20 30 40 50 x/m 70
0 120 10 20 30 40 50 x/m 70

Raw Data in m 7.94 10 13 16 20 25 32

z/m 110

0 we2 we4 we6 we8 we10 we12

10

20

30

40

50

x/m

70
105 100

Topography effect 0.71 0.79

20

95

0 we2 we4 we6 we8 we10 we12

10

30

0.89

40

50

1.12

x/m

1.26

70

90 [email protected]

Corrected Data in m
Ohmm

7.94

10

13

16

20

25

32

6.31

7.94

10

12.6

15.8

20

25.1

31.6

39.8

Figure 6: Topographic eect and inversion result of the slagdump site.

2.5 Structural constraints

Often there is additional information about the subsurface. An incorporation into the inversion process is always to be preferred over a comparison of the results. In (Gnther and Rcker, u u 2006) we presented a more general minimisation approach that allows for arbitrary weights for each boundary between model cells. In existence of a known discontinuity this can be set to zero allowing for (but not enforcing) an arbitrary jump in resistivity. The following example (dcfemlib/examples/inversion/inversion/2dstruct) was measured and friendly provided by the K-UTec GmbH Sondershausen (T. Schicht). Aim of the study was bedrock detection carried out with resistivity and refraction seismics. The velocity structure showed to be a very clear 2-layer case. So the result (layer boundary) of the refraction study can serve as structural information. The le bedrock.xz contains the course of the boundary as x-z pairs. We now include this le into the conguration le using the INTERFACE option. In order to compare the result with

and without structure we call $ invertNew2d bedrock.dat > inv.cfg $ invert inv.cfg all save $ echo INTERFACE=bedrock.xz >> inv.cfg $ invert inv.cfg newPara save The latter (newPara) switch is used if we change parameterisation. So the primary mesh and potentials do not have to be calculated anew, but they have only to be interpolated onto the forward mesh and the sensitivities are re-calculated as well. Figure 7 shows the subsurface images without and with the structural information.
0 50 100 150 200 250 300 350 400 x/m 500 0 50 100 150 200 250 300 350 400 x/m 500

z/m

z/m

100

100

150

[email protected]

150

[email protected]

Ohmm 63.1 100 158 251 398 631 1000

Ohmm 63.1 100 158 251 398 631 1000 1580

Figure 7: Resistivity distribution without (left) and with (right) structural information. Obviously the additional information leads to a much clearer image of the subsurface. At most positions there is a sharp resistivity contrast at the boundary. However at some positions there is either a dierence to velocity or the refraction result is ambiguous.

2.6 A sophisticated example: The lake case

The last 2d example was measured by the Leibniz Institute for Applied Geosciences, Hannover (W. Sdekum and T. Gnther). Its aim was to delineate sedimentation structures beneath the u u Feldungel lake near Osnabrueck. Electrodes have been spread out from one shore along the lake bottom onto the other shore. The spacing was 2m and both Wenner-alpha and Wenner-beta were measured and combined. Since the lake resistivity is known (22.5 m) this is excluded from inversion by hand: 1. We put the electrode positions (0 to -2.6m height) and resistances into the input le 2. We start as for a topographic case and generate the meshing input $ invert inv.cfg domain 3. As a result we obtain the poly le mesh/mesh.poly which we copy to mymesh.poly9 4. We need to add the water surface by an edge between the left and the right shore. A view into the poly le shows these being represented by the points 3 and 138. So add another edge at the end of the edges (line 303) by inserting the line 151 3 138 -1 (number n1 n2 edgemarker) and increase the number of edges in line 152 from 151 to 152

see triangle page http://www.cs.cmu.edu/~quake/triangle.research.html for le description

10

5. Finally we add a region marker somewhere in the lake with marker 1 (not inverted) by appending the line 2 50 -1 1 0.0 (number x y marker maxtrisize) and increasing the number of regions from 2 to 3.
0 5 10 15 20 z/m 30 1 2 1

20

40

60

x/m

100

Figure 8: Representation of the input PLC for the lake case. Figure 8 shows a section of the input PLC (the outer space needed for forward calculation is much bigger). All nodes are denoted by a cross. The electrodes, which are rened by additional nodes on both sides, are marked by a circle. They form the topography as in the 2d topographic examples. The red rectangle denes the region to be inverted. Therefore the marker 2 (inversion region) is in this region and the marker 1 (no inversion) is outside. We use this altered poly le in the inversion by introducing into the cfg le PARAGEOMETRY=cp mymesh.poly mesh/mesh.poly This means that mesh.poly is created by copying mymesh.poly. Alternatively we can put here a shell script or matlab call that makes up our geometry. Furthermore we add the following options (obsolete in GIMLi due to better automatics and the region technique): TOPOGRAPHY=0 SPACECONFIG=2 RHOSTART=22.5 NOPROLONGATION=1 SURFACESMOOTH=0 # # # # # ensures that the primary potentials are really analytic ensures that mirror sources will be used (obsolete?) Start(=background) resistivity no resistivity prolongation (outside & lake stay const) prevent smoothing (edge swapping possible)

Furthermore we use the following options ZPOWER=0.3 # enhances layered (sediment) structures OVERRIDEERROR=1 # do not use the measured errors in file (optimistic), but: INPUTERRLEVEL=2 # 2% plus INPUTERRVOLTAGE=20e-6 # 20 microvolts and the inversion converges at 1 < 2 < 2. Since the data le also contains induced polarisation (IP) phases, they are automatically inverted at the end using a single inverse sub-step with the same options as for the amplitude. Figure 9 shows resistivity and phase distribution. The lake sediments show generally by low resistivities and phases. Particularly the youngest sediments are almost as conductive as the water. Below dierent sediment layers are visible giving hints how the historic topography was covered by sediments step by step. Note that for pure underwater measurements there is an additional option UNDERWATER which forms the sea bottom topography by the electrodes and adds the water surface at z = 0 automatically.

11

0 z/m 10 15

10

20

30

40

50

60

70

x/m

90

[email protected]

Ohmm 25.1

31.6

39.8

50.1

63.1

79.4

100

126

158

200

251

0 z/m 10 15

10

20

30

40

50

60

70

x/m

90

[email protected]

Figure 9: Inversion result of the water case: resistivity (top) and phase (bottom). Note that the water is not part of the inversion mesh and thus not plotted.

3 3D surface measurements
3D surface measurements can be carried out in several variants: 1. Layout of an electrode grid. However, due to the limited electrode number grids are restricted to small areas. 2. Parallel (and perpendicular) proles along the coordinate axes. 3. Proles in arbitrary directions due to accessibility limits. 4. Non-prole layout, e.g. large-scale dipole-dipole experiments. In any case, the electrode positions and measurements must be dened according to the unied data format. The data for the rst two types can be easily organized by hand. For number 3 (and 2) we suggest to prepare 2d les and to write a pro-le containing of lines with the 2d le name and x-y pairs of points where the line is going. This le can be read into DC3dInvRes Gnther (2008) and used to write the 3d le. In case of topography it is best to do the tape u correction on the 2d les before using DC2dInvRes and Export Ohm. The most exible element in 3d is the tetrahedron. The tetrahedralization is done by a mesh generator. Out choice is Tetgen (Si, 2008), a free and versatile quality mesh generator. The quality measure is dierent from 2d and describes a radius-to-edge ratio, note that small values point to higher quality. Appropriate values for (primary eld) forward calculation are 1.12 to 1.2, for the inverse (and thus secondary) mesh values of 1.2-1.5 are appropriate, the keys are called PRIM3DQUALITY and PARA3DQUALITY.

12

3.1 Flat surface

In dcfemlib/examples/inversion/3dat/gallery is a data set in the eld where the 2dat example (section 2.1) was measured. It comprises a grid of 9x14 electrodes. Dipole-dipole measurements have been measured on all x and y proles. In the data le is an error of constant 0.0 that will be overrided automatically. An inversion project le with default parameters is created by: $ invertNew3d rothschoen.dat > inv.cfg The inversion is then fully run (with command all) converging to a chi-squared mist of about 2 (rrms=4-5%). In order to t the data better, the regularization parameter is decreased using LAMBDA=5, which leads to an relative rms error of about 3% (2 = 1). Another way is to use an anisotropic regularization (CONSTRAINT=6) also leading to 2 1. (Note that constraint 6 is somewhat deprecated and replaced by the ZPOWER options, where we can control the atness continuously. Try out dierent ZPOWER values between 0 and 1.) The result is saved and converted to a vtk le10 using $ invert inv.cfg show

Figure 10: Inversion result of the 3d gallery data set using a smoothed iso-surface of 650m and a Plane Clip, the red spheres are the used electrodes. Figure 10 shows a Paraview visualisation that has been created by the following steps: i) Cell Data To Point Data, ii) Clip by Scalar 650 (m), iii) Extract Surface, iv) Smooth Surface, v) Another Clip based on Cell2Point with Plane, vi) representation of the input as Outline and Cube Axes. The color bar is logarithmic with a manual range of 100-1000m. The electrodes have been included as point vtk le and displayed by Glyph as Spheres of radius 0.05. After some exercise the reader will be able to create nice images, plots and calculate results such as extensions or volumes of geological bodies.

3.2 3D Topography
The denition of a 3d topography is much more complicated than in 2d, where every shape can be described by a simple polygon. The input PLC consists of faces instead of edges, the
10

can be displayed in 3d software ParaView, see http://www.paraview.org

13

resulting poly le has a similar but dierent format11 . Generally the proceeding is the following: i) create a at surface mesh, ii) interpolate heights from topographic information, iii) make a small (inversion mesh) and a large (forward mesh) box around it, iv) make renement, if necessary, and v) create the mesh using tetgen. For specifying topography, there are two dierent ways: the electrodes in the data le have an elevation and all other points are interpolated there is a digital elevation model (DEM) or at least a list of measured topo points (in a 3-column le containing x,y and z) Whereas the rst case is sucient for smooth topography and/or dense electrode coverage, the latter is more general. The topographic points are Delaunay triangulated. For every point of the meshes, also the electrodes, the elevation is linearly interpolated. Therefore electrodes with measured elevations should be included in the topo le as well to make sure their z values are correct. We specify this topographical list by the line TOPOPOINTS=filename. In dcfemlib/examples/inversion/examples/acucar there is a project measured by the Federal Institute of Geology and Natural Resources (BGR) Hannover (M. Furche12 ). The site is an old slag dump that comprises a topography reminding on the sugar hat in Rio. Two resistivity proles have been measured crossing the top of the isolated hill. Another prole was realised around the hill in a more or less constant elevation. Although this is not a dense sampling as an electrode grid it should be sucient to obtain a rough image. Additionally to the electrodes, some topographical points have been measured and put into the le points.xyz. So we create a new project using $ invertNew3dTopo acucar.ohm > inv.cfg and add the line TOPOPOINTS=points.xyz to the model. If we now call invert inv.cfg meshs we see the mesh does not show the hill, since the topography overrides the electrode elevation. Therefore we have to add the electrode denition (lines 3-230) to the topography le and see then the hill (Figure 11 left). However due to the point density the electrode line appears as a sharp edge that is not really the truth but sucient in this case. In other cases we might have a digital elevation model. In order to show this on the same example, we created one by cubic interpolation of the available points on a regular grid of 2m spacing. In order to avoid interpolation errors between the electrodes we created a polygon le poly.xyz for the three proles13 and introduce it by TOPOPOLY=poly.xyz. Figure 11 shows the surface mesh of both variants. The sharp edges are now disappeared. Finally the inversion result is visualised in Figure 12. It shows a conductive interior of the slag dump and dierent sediments at the surface, e.g. a resistive top. Of course the data coverage is low between the proles and at the model boundaries. Therefore the model becomes more or less interpolated by the smoothness constraints.

11 12

See http://tetgen.berlios.de Now at Leibniz institute of Applied Geosciences 13 Several polygons are separated by a blank line.

14

Figure 11: Surface mesh for the point-wise topographic information (left) and the digital elevation model (right), the electrodes are shown as red points.

Figure 12: Inversion result of the 3dtopo case.

4 Closed geometries
Closed geometries are actually easier than open ones since we do not need a mesh prolongation and two dierent regions. However since the whole boundary is of Neumann type, we must ensure two additional conditions that are not necessary in the open case: The current cannot vanish in innity, therefore we must use dipole sources, e.g. by a reference current node. Since only derivatives are present in the boundary value problem, we must make the forward solution unique, e.g. by adding a reference potential node, whose potential is forced to zero.

4.1 2d example - tree tomography

For 2d bodies the electrodes are usually on the boundary and the PLC can easily be formed by a polygon through it as in the 2d mesh generation. If additional points between the electrodes are needed, they can be introduced by fake electrodes. For tree (or human) geometry a dedicated

15

GUI named TreeBERT (before DC2dTree) was created making it easy to process the data visually. Tree tomography EIT on trees has been successfully established to investigate decay of trees. The example in dcfemlib/examples/circle/tree was measured and friendly provided by Niels Homann, HAWK Gttingen. It represents a lime tree, measured by 24 steel electrodes that are plugged into the o bark. Dipole-dipole measurements have been applied using a Geotom equipment. The conguration le reads as follows DATAFILE=hollow_limetree.ohm DIMENSION=2 TOPOGRAPHY=1 # activates the primary mesh CYLINDER=1 # defines a closed geometry SURFACESMOOTH=1 # makes a nicer surface EQUIDISTBOUNDARY=1 # equidistant refinement PARADX=0.2 # 5 segments between the electrodes PARA2DQUALITY=34.8 # very good quality, almost the upper limit SPLINEBOUNDARY=1 # round geometry PRIMDX_R=0.001 # refinement of primary mesh in radial direction LAMBDA=10 # regularisation strength BLOCKYMODEL=1 # enhance contrasts by robust (L1) methods For this case an equidistant renement, the use of splines and a high quality ensures a nice mesh with a round boundary. The primary renement is done in radial direction. Additionally we used the robust modelling in order to obtain a clearer contrast of the high resistivity.
0.2 0.2 0.1 0 x/m 0.2 0.2 0.1 0 x/m 0.2 0.2

z/m

0.1

0.05

0.05

0.1

0.15

0.2

0.25

126

158

200

251

316

398

Figure 13: Tree cut (left), inversion result (center) and overlay. After the measurements the tree was cut and revealed a cavity inside caused by decay. Figure 13 shows a photograph, the inversion result and an overlay of both. Clearly the cavity is marked by high resistivity that is in almost perfect accordance with the photo.

4.2 3d closed geometry - model tanks

In the Federal Institute of Geology and Natural Resources (BGR) Hannover, a cylindrical model tank was created in order to make inltration experiments with material from the slag

16

dump. The column has diameter of 30cm and a height of 80cm. In each of 5 rings with 5 cm vertical distance 24 steel electrodes of 2cm length were installed. Dipole-dipole measurements have been applied to all rings yielding a number of 320 data. The example is located in dcfemlib/examples/inversion/3dtank. Since the parameterization cannot be detected automatically from the le, we have to create the mesh input, i.e. the PLC in mesh/mesh.poly by hand using a script. There is a poly tool polyCreateCube creating a unit cube. With the option -Z is creates a unit cylinder instead, which has to be scaled appropriately. Then we put in the electrodes as points14 with the marker -99. We insert two additional nodes with markers -999 and -1000 that are used for current reference and potential reference. So the script reads: MESH=mesh/mesh # PLC name polyCreateCube -v -Z -s 48 -m 2 $MESH # create unit cylinder with 48 segments polyTranslate -z -0.5 $MESH # moves it such that top is zero polyScale -x 0.3 -y 0.3 -z 0.8 $MESH # scale to radius 0.15 & height 0.8 cat soil_column.dat | head -n 82 |tail -n 80 > elec.xyz # extract electrodes polyAddVIP -m -99 -f elec.xyz $MESH # add electrodes to mesh polyAddVIP -m -999 -x 0 -y 0 -z 0 $MESH # current reference node polyAddVIP -m -1000 -x 0 -y 0 -z -0.8 $MESH # potential reference node polyConvert -V -o $MESH-poly $MESH # convert to vtk to load it to paraview We create an empty cfg le (or use invertNew3dCyl) with the lines DATAFILE=soil_column.dat DIMENSION=3 TOPOGRAPHY=1 CYLINDER=1 # ensures the closed geometry We add our PLC script to the PARAGEOMETRY variable such that mesh/mesh.poly will be created by it. PARAGEOMETRY=mymesh.sh # make sure that mymesh.sh is executable A further renement can be achieved by quality improvement (PARAQUALITY), local renement (PARADX) or maximum cell size (PARAMAXCELLSIZE). In order to obtain an accurate we use a renement for the primary mesh of 1cm and quadratic shape functions ending in about 32000 nodes. PRIMDX=0.01 PRIMP2MESH=1 Figure 14 shows the course from the mesh input via the parameter mesh to the nal result.

14

Since the electrodes cannot be show a signicant extension compared to the column size, we put the points not onto the surface but moved it 1cm inside.

17

Figure 14: PLC (left), parameter mesh (center) and inversion result (right) of the soil column experiment.

18

5 Buried electrodes
5.1 2D cross-hole data
Of course cross-hole measurements can also be inverted using BERT. The height of each electrode must be set to the elevation minus depth. However, since we cannot distinguish whether it is topography or a buried electrode we must create the geometry by hand. Note, that for this reason we introduced the tokens h (height) and d (depth below surface) to the data format, but this is only recognized by GIMLi. The example in dcfemlib/examples/inversion/2dxh was produced by O. Kuras of the British Geological Survey (BGS) in the ALERT project (Kuras et al. (tted)). It represents about 1300 data obtained by cross-hole measurement between 5 very shallow (0-1.6m) boreholes. In order to create an inversion mesh we would create a small box with marker 2 (inversion) inside of a big box that is used for forward calculation (marker 1) by 8 points and 8 edges. This is more rigorously implemented by the script polyFlatWorld which automatically calculates the size of the model and the boundary around the electrodes from the parameters BOUNDARY, PARABOUNDARY and PARADEPTH. To pass the variables we use the Unix command source and introduce PARAGEOMETRY="source polyFlatPara 2dxh.ohm into the cfg le. Additionally we set SPACECONFIG=2 for subsurface sources and a pretty good quality for the parameter mesh PARA2DQUALITY=34.5. As before, we can use PARADX to rene the model at the electrodes. In this case we introduce a node between each of the 0.1m separated electrodes using PARADX=0.05. Note that, dierent from 2d surface measurements, it is treated by polyFlatWorld as an absolute value. We use a relatively large regularization parameter LAMBDA=300 for suppressing oscillations at the electrodes. Figure 15 shows the obtained resistivity distribution at the very beginning of a tracer experiment. See section 6.1 for time-lapse inversion.
0 2 2.5 3 3.5 4 4.5 x/m 5.5

z/m

1.5

10

12.6

15.8

20

25.1

31.6

39.8

50.1

63.1

79.4

100

Figure 15: Inversion result of the crosshole data set.

5.2 3D-Crosshole measurements

Crosshole measurements can of course be applied three-dimensionally. The example in dcfemlib/examples/inversion/3dxh was presented by J. Doetsch from ETH Zurich. The data le 3dhx.ohm comprises 753 data between 4 boreholes in the saturated zone (d=4-10m) and is part of a monitoring experiment. We create a cfg le using DATAFILE=3dxh.ohm DIMENSION=3

19

SPACECONFIG=2 # for mirror sources at z=0 PARABOUNDARY=15 # to get a bit more space around the electrodes PARAGEOMETRY="source polyFlatWorld $DATAFILE" By using ZPOWER=0.3 we can enhance the predominantly layered structures. The inversion converges then with defaults down to about 2 = 1. Figure 16 shows the nal result.

Figure 16: Inversion result for the 3D crosshole case.

5.3 Topography and buried electrodes

If both heights and depths below surface are given, we cannot use createSurface/createParaMesh and polyFlatWorld. In this case we suggest to create a data le with the surface electrodes (either with topo or with zeros and a topo le) and a list of subsurface electrodes including real. The normal PLC generation is done with the rst and the latter are then added using polyAddVIP. Importantly the order of electrodes must be set such that rst the surface electrodes and then the buries electrodes appear. The resulting script is introduced using PARAGEOMETRY=myscript.sh which can look as follows: cp inv.cfg surface.cfg # makes a copy with all options echo DATAFILE=datafile-without-electrodes.dat >> surface.cfg invert surface.cfg domain # creates mesh/mesh.poly polyAddVIP -m -99 -f borehole-electrodes.xyz mesh/mesh.poly Finally all electrodes are in the PLC in the right order and all should be well.

6 Time-lapse ERT
We are often interested in ongoing physical processes and use ERT for monitoring experiments. There are many dierent time-lapse inversion approaches. Here we use a very simple one that solves the linear subsystem for the resistivity ratio of two data sets after the inversion. An arbitrary number of subsequent data sets can be processed by writing their le names in a text le and pass it by TIMESTEPS=filename. Up to now, the data les must contain identical congurations in the same order.

20

6.1 Crosshole timelapse measurements

Lets go back to the crosshole case 2dxh (see section 5.1) and unpack the time data les in 2dxh-timelapse.zip. In 2006 the BGS injected a highly saline tracer in borehole number 8 and measured 36 data sets every 40 minutes such that a whole day was covered. The subsequent les are named 01.dat, 02.dat, ... and are assembled in timesteps.txt. By including TIMESTEPS=timesteps.txt and calling invert inv.cfg calc again. As a result we obtain a lot of model i res.vector containing the resistivity values for one time step each. By a show command they are transferred into paraview les ParaView recognizes this names as time steps and allows an easy scrolling through the times. Figure 17 shows some selected time steps that allow for seeing the tracer ow toward the left boundary. Note that these are only preliminary results that are used to present how BERT is working. With more sophisticated time lapse strategies the monitoring process can be traced more accurately.
0 2 2.5 3 3.5 4 4.5 x/m 5.5 0 2 2.5 3 3.5 4 4.5 x/m 5.5

z/m

z/m

1.5

1.5

10 0

12.6 2

15.8 2.5

20 3

25.1

31.6 3.5

39.8 4

50.1 4.5

63.1 x/m

79.4

100 5.5

10 0

12.6 2

15.8 2.5

20 3

25.1

31.6 3.5

39.8 4

50.1 4.5

63.1 x/m

79.4

100 5.5

z/m

z/m

1.5

1.5

10

12.6

15.8

20

25.1

31.6

39.8

50.1

63.1

79.4

100

10

12.6

15.8

20

25.1

31.6

39.8

50.1

63.1

79.4

100

Figure 17: Inversion results 3 hours (upper left), 7 hours (upper right), 12 hours (lower left) and 16.5 hours (lower right) after tracer injection.

6.2 Soil column measurements

We go back to the soil column example from section 4.2. After irrigating a certain amount of water, every 2 hours a complete data set was measured and included in the TIMESERIES le. Since the changes are relatively low, we take a look at the relative dierences in the les di i.vtk with respect to the initial resistivity. Figure 18 shows 5 selected time steps. We can see the water front intruding but at a certain stage the column is drying out again.

21

Figure 18: Relative resistivity dierence (in %) for the repeated measurements at about 2, 4, 6, 10 and 16 hours after irrigation.

7 Concluding remarks and outlook

We hope to having demonstrated that our software can be used for ERT inversion on very dierent geometries using dierent kinds of data. BERT is not a black-box dummy tool and probably it will never be. Thus a lot of experience is required to obtain excellent results. However, for special cases it is worth to branch out more user-friendly tools with immediate visualization and data analysis. We started this by TreeBERT, CylBERT and WallBERT. There are some disadvantages in the use of BERT that are due to the historic development. BERT as part of DCFEMLib will mainly stay the way it is now, only bugs are xed and minor issues are addressed. However, the technique BERT is included among other methods in the library GIMLi for geophysical inversion and modelling. The new and independent multi-physics library has some advances over DCFEMLib A much more exible style of programming using template techniques. Connection of dierent data by classical or structural joint inversion. Use of arbitrary transform functions. More rigorous techniques for time-lapse inversion. Application of a region technique for sophisticated constraints. See also appendix D for BERT within the scope of GIMLi.

Acknowledgements
We like to thank all the guys that provided the very instructive data: Folker Donner (formerly University of Mining and Technology, Freiberg), Markus Furche and Ulla Noell (BGR Hannover), Thomas Schicht (K-UTec GmbH Sondershausen), Niels Homann (formerly HAWK Gttingen), Oliver Kuras (British Geological Survey), Joseph Doetsch and Ilaria Coscia (ETH o Zurich). Furthermore we acknowledge all the users and testers of BERT that made the software what it is now, a powerful expert tool.

22

References
Gnther, T. (2002-2007). DC2dInvRes - Direct Current 2d Inversion and Resolution. resistivu ity.net productions, http://dc2dinvres.resistivity.net. Gnther, T. (2003-2008). DC3dInvRes - Direct Current 3d Inversion and Resolution. resistivu ity.net productions, http://dc3dinvres.resistivity.net. Gnther, T. and Rcker, C. (2006). A general approach for introducing structural information u u - from constraints to joint inversion. In Ext. Abstract, EAGE Near Surface Geophysics Workshop. 3.-6.9.06, Helsinki(Finland). Gnther, T., Rcker, C., and Spitzer, K. (2006). 3-d modeling and inversion of dc resistivity u u data incorporating topography - part II: Inversion. Geophys. J. Int., 166(2):506517. Kuras, O., Pritchard, J., Meldrum, P. I., Chambers, J. E., Wilkinson, P. B., Ogilvy, R. D., and Wealthall, G. P. (submitted). Monitoring hydraulic processes with automated time-lapse electrical resistivity tomography (ALERT). Compte Rendus Geosciences - Special issue on Hydrogeophysics. Rcker, C., Gnther, T., and Spitzer, K. (2006). 3-d modeling and inversion of dc resistivity u u data incorporating topography - part I: Modeling. Geophys. J. Int., 166(2):495505. Shewchuk, J. R. (1996). Triangle: Engineering a 2D Quality Mesh Generator and Delaunay Triangulator. In Lin, M. C. and Manocha, D., editors, Applied Computational Geometry: Towards Geometric Engineering, volume 1148 of Lecture Notes in Computer Science, pages 203222. Springer-Verlag. From the First ACM Workshop on Applied Computational Geometry. Si, H. (2002-2008). TetGen - a quality-constrained tetrahedral mesh generator. Weierstrass institute, Berlin, http://tetgen.berlios.de.

A BERT for Windows users

BERT is successfully applied on Windows platforms, however it can be slower for bigger problems. Most easily it is distributed under the name BERT4WIN-Suite comprising the binaries from dcfemlib. Additionally it holds dcmod and dcinv and the gimli library (see Appendix D). Since BERT is controlled on the command, line Windows users need a command environment as well such as the widely known Cygwin. We recommend MSYS (minimal system), a very small but nice bash-like command shell15 . Download MSys from http://www.mingw.org and install it. Install BERT4WIN, assuming under d:\software\BERT. Then this path must be known inside the shell, either by changing the environment variable Path under System Control - System - Environment Variables or in the shell by typing $ export PATH=$PATH:/d/software/BERT The latter can also be done automatically at startup by insertion into a .bashrc le in the home directory.
15

Since the scripts are written in bash syntax and in MSYS there is no /bin/bash, it has to be copied once from /bin/sh (cp /bin/sh /bin/bash)

23

Note that le names are in dierent from Windows, i.e. /c instead of c:, slash instead of backslash and names are case-sensitive. If you prepared your data and conguration le in a directory, you should go there with the shell by i.e. $ cd /c/data/profile1/trial and run the inversion using invert inv.cfg all and so on.

B Files and programs

Created les and their meaning
File types: *.poly *.bms *.vtk *.mesh *.vector *.collect triangle (2d) or tetgen (3d) PLC format binary mesh (house) format visual toolkit mesh or poly format (paraview) MEdit mesh format ascii vector of oats potential matrix of all electrodes

Directories and their content: mesh primaryPot/primary primaryPot/interpolated sensM result* meshes and mesh input (meshPara,meshSec,meshPrim) primary potentials on primary mesh primary potentials on secondary mesh sensitivity columns (smatrix.*) or rows smatrixCol.* saved result directory with most important les

Files in project or result directory: command.history *.data inv.log model iter.nal model iter.*.vector modelResponse.*.vector modelReciprocity.*.vector sensCov.vector mesh/mesh.poly mesh/meshParaDomain history of commands executed by invert ltered data le with apparent resistivities and errors inversion log le nal resistivity distribution model vectors of individual iterations model response vectors for each iteration model response reciprocity for each iteration coverage (sum of absolute sensitivies) mesh input PLC pure parameter mesh (use for visualization)

Program calls used for BERT

Inversion and parameterization: invert - BERT cfg le interpretation invertNew2d/2dTopo/2dCirc - CFG le generators for 2d cases (at,topo,circle) invertNew3d/3dTopo/2dCyl - CFG le generators for 3d cases (at,topo,tank) inversion - actual inversion routine paradepth - estimate appropriate model depth by 1D sensitivities dcfem - FEM modelling routine setsingvalue - set potentials at electrodes

24

collectdata - assemble potentials from dcfem lterdata - lter data yielding apparent resistivity and error estimate Vector functions: createStartModel va2b va2b interpolate create starting model vector convert ascii to binary vector convert binary to ascii vector interpolate quantity from one mesh to another

Mesh creation and alteration: createParaMesh createSecondaryMesh createSurface closeSurface prepareMeshRenement dctriangle meshconvert create parameter mesh from dat le create secondary mesh out of parameters create 3d surface mesh from xyz point list close 3d surface mesh by surrounding boxes insert renemeht points triangle call convert mesh between various import formats

Poly tools - creating PLC objects: polyCreateWorld polyFlatWorld polyCreateCube polyTranslate polyScale polyRotate polyMerge polyAddVIP polyAddProle polyReneVIPS polyConvert polyScripts.sh makes a world with 2 surface and interior boundary make inversion world (2 regions) around electrodes create (unit) cube around origin translate PLC scale PLC rotate PLC merge 2 PLCs into a new one add points (e.g. electrodes) to PLC add prole of electrodes rene points by local renement convert PLC to VTK or STL format various functions

C Complete list of options and their default values

# Global settings DATAFILE=datafile DIMENSION=3 TOPOGRAPHY=0 TOPOPOINTS= TOPOPOLY= TIMESTEPS= PARAGEOMETRY= INTERFACE= CYLINDER=0 ELECTRODENODES=1 # # # # # # # # # # defines the project filename (required) defines the dimension of the problem (2 for 2d or 3 for 3d) defines if topography is present (0 or 1) file with additional coordinates for topography (x y z) file with additional polygons for topography (x0 y0 z0) file with names of additional datafiles for timelapse inversion defines command for creating the PLC by hand file with known interface(s) (currently only 2d) defines cylindrical geometry (0 or 1) electrodes are represented as nodes (default)

25

SPACECONFIG=0 UNDERWATER=0

# geometric factor (0=half space, 1=full space, 2=mirror source) # defines underwater survey, sets SPACECONFIG=2

# Data settings INPUTOHM=0 # input data is in Ohm OVERRIDEERROR=0 # overrides given errors with INPUTERRLEVEL/INPUTERRVOLTAGE INPUTERRLEVEL=3 # sets input error level (in percent) if no error given INPUTERRVOLTAGE=100e-6 # sets input voltage error (V) if no error given FILTERVALS=0 # defines filter values (experimental) # Inversion settings RHOSTART=0 # resistivity of start model (0 = median of the data) NOPROLONGATION=0 # disables resistivity prolongation into non-inversion regions MAXITER=20 # maximum number of iteration steps SINGVALUE=-1 # potential value at electrodes, for sensitivity (internal) LAMBDA=20 # regularization parameter LAMBDAOPT=0 # optimze lambda by using l-curve (0 or 1) LAMBDADECREASE=1 # decrease lambda with each iteration CONSTRAINT=1 # order 0, 1, 2 (experimental) ZPOWER=0 # weight for vertical gradients BLOCKYMODEL=0 # iteratively reweighted model (L1) ROBUSTDATA=0 # iteratively reweighted data (L1) LOWERBOUND=0.0 # lower resistivity bound (logarithmic barrier) UPPERBOUND=0.0 # upper resistivity bound (0.0 = deactivated) SENSMATUPDATE=0 # update sensitivity matrix with each iteration step SENSMATDROPTOL=0 # only for very large problems (internal) SENSMATMAXMEM=800 # only for very large problems (internal) # Mesh settings PARAMAXCELLSIZE=0 # maximum cell size volume (m3) or area (m2) for para mesh PRIMMAXCELLSIZE=0 # maximum cell size volume (m3) or area (m2) for prim mesh PARADEPTH=0 # maximum depth of parameter domain in meter (0 = automatic) PARABOUNDARY=5 # boundary around electrodes in parameter domain (percent) SPLINEBOUNDARY=0 # use splines for additional surface points EQUIDISTBOUNDARY=0# equidistant refined space between electrodes BOUNDARY=500 # size of boundary area around parameter domain MESHGEN=tetgen # 3d mesh generator TETGENTOLERANCE=1e-12 # tetgen tolerance limit for face checks TETGENPRESERVEBOUNDARY=0 # prevents tetgen from splitting boundary facets PARADX=0.0 # refinement for para mesh (values>0.5 is forced to 0.5) PRIMDX=0.1 # ref. for prim mesh towards mesh center(2d: electrode) PRIMDX_R=0.0 # ref. for prim mesh towards electr. center (overrides PRIMDX) PARA2DQUALITY=33.0# parameter grid (from 20 (bad) to 33.4 (good)) PRIM2DQUALITY=33.4# primary grid (from 20 (bad) to 33.4 (good)) PARA3DQUALITY=1.5 # parameter grid (from 1.1 (good) to 2 (bad)) PRIM3DQUALITY=1.2 # primary grid (from 1.1 (good) to 2 (bad)) SURFACEQUALITY=30 # quality of topographical surface grid (20 (bad) - 33.4 (good))

26

SURFACEMAXTRISIZE=0.0 # maximal triangle area of paramatric surface grid SURFACESMOOTH=0 # improve quality of topographical surface grid ICDROPTOL=0.0 # if number of nodes 200k drop tolerance is set for ICCG solver LINSOLVER=1 # linear solver: 1=CHOLMOD, 2=LDL, 3=PCG (very slow), 4=TAUCS SECMESHREFINE=1 # refinement steps for secondary mesh (use with caution) SECP2MESH=0 # quadratic shapefunction for secmesh (experimental) PRIMP2MESH=0 # use primary p2 mesh (use with caution) # Directory settings MESHBASENAME=mesh # basename for mesh files DIRMESHS=mesh # directory name for mesh files DIRPOT=primaryPot # directory name for primary and interpolated potentials DIRPRIMPOT=potentials # subdirectory name for primary potentials DIRINTERPOLPOT=interpolated # subdirectory name for interpolated potentials DIRFEM=femM # directory name for finite element solutions DIRSENS=sensM # directory name for sensitivity matrix OLDPRIMMESHSTYLE=0# for internal use only

D BERT using GIMLi

DCFEMLib as a library for dc resistivity modelling and inversion is not extended anymore. Only a few improvements and bug xes will keep it updated as standard tool. However, the BERT technology is further developed within the open source library GIMLi - Geophysical Inversion and Modelling Library. It represents a completely new-written multi-method C++ class library for inversion of very dierent data. So BERT as a technology consisting of mesh generation, primary potential calculation and nally inversion, still persists. GIMLi::BERT has some advantages over DCFEMLib::BERT: mix of electrode nodes, node-free or nite electrodes faster runtime and easier build procedure more rigorous minimization approach, mainly timelapse improved options such as arbitrary transformations region technique for sophisticated regularization There are no mesh generation procedures in GIMLi, so we still use the invert poly and mesh functions. The DC FEM modelling tool is called dcmod, it can be used (besides pure modelling) for primary potential calculation and interpolation onto the secondary eld using the command line: $ dcmod -v -a rho.map -m mesh/secmesh.bms -o output mesh/primMesh.bms The actual inversion is done by dcinv, which has similar options (regularization parameter, resistivity bounds, timelapse steps) as used by invert. It is able to read meshes and primary potentials as prepared by invert. The mesh being used is mesh/mesh.bms, a mesh consisting of the original markers 1 (background) and 2 (inversion). Since the option list including mesh (-p ), potentials (-y) and sensitivity (-L) would become long, it is abbreviated by dcinv -1. A typical program call could be like this

27

$ dcinv -v -1 -b20 -u5000 -l12.5 -z 0.2 -t timesteps.txt mydata.data Finally, the command dcedit can be used to process and edit data. In order to build some sort of a complete environment we prefer the free open-source environment Python16 as a combination of shell programming and visualization. We are trying to develop some graphical user interfaces for specic purposes that allow easier control about whether meshes are appropriate.

16

see http://www.python.org

28