Discussion archives on Groundwater Modelling I have been compiling selected extracts on groundwater modelling discussion from various mailing lists for last many years, as given below. I understand that in view of random and discontinuous placing, many parts may not be easily comprehensible. Still I think that it could be helpful to groundwater modellers to some extent. Regards, C. P. Kumar ================================================== C. P. KUMAR Scientist 'F' National Institute of Hydrology Jal Vigyan Bhawan Roorkee - 247667 (Uttaranchal) INDIA http://www.angelfire.com/nh/cpkumar/ https://in.groups.yahoo.com/groups/cpkumar/ Groundwater Assessment and Modelling (Book) http://www.amazon.com/Groundwater-Assessment-Modelling-Mr-Kumar/dp/1511520493 ================================================== Discussion archives * In my experience, PCG is faster than LMG for small models but LMG is faster for large models. Of course, for small models, speed generally is not so much of an issue. * DAMP is used as a multiplier to scale the head change at each iteration, not RELAX. The DAMP parameter in the PCG solver package is analogous to the ACCL parameter in the SIP package. Although I cannot explain precisely what RELAX controls (since I'm no mathematician), the typical range for RELAX is 0.97 to 1.0. In my experience, setting RELAX=0.97 can help to achieve convergence, and may speed up convergence on a "well-behaved" model, but not always. * I'm sure you've heard "All models are wrong but some are useful." If you replaced all the wells in a model with drains, you would expect to get different results. How is this any different? You have replaced one package with a different one that is based on different assumptions so of course, they get different results. If you want to know which one is better, examine the assumptions behind the different packages and see which assumptions best match the conditions at the particular region to be modelled. If none of the available packages do what you need, write your own package to do what you need. * The newer LPF package is the way that most other models work, i.e., you enter Kz and the inter-cell conductances are computed internally. The way that MODFLOW88/96 did it with a pre-computed vertical conductance always struck me as a bit odd; however, that has been our happy MODFLOW world for about 20 years. So now we have a new option that is mathematically more correct, but it is still disconcerting when the computed heads are different. I guess you'd have to ask a lawyer about who would win. My sense is that the differences could never be explained to a jury. It does present the MODFLOW modeling community with a

problem, though. The way that you did it with MODFLOW-SURFACT is probably the way that the USGS should have done it as well - adding the input of Kz in the BCF Package. * Input of vertical hydraulic conductivity has its advantages, however, McDonald and Harbaugh got it correct in keeping a fixed thickness of a cell for the vertical direction conductance - for gravity-segregated vertical equilibrium (GSVE), you have a transmissivity that is conductivity times saturated thickness which is valid only for the horizontal direction and not for the vertical direction (in actuality, the vertical direction physics changes from GSVE to unsaturated flow which would lower the conductivity further and that would be closer to the "fixed thickness" answers that do not reduce separation distance between 2 cells vertically even for consideration of a homogeneous case, but that aside). Nor does the document state where or under what conditions the use of a "variable thickness for vertical conductances" option performs better. That's why I was asking to see if anyone knew why this was done in the first place. * I propose that the node location of a cell that contains the water table should be located at the water table instead of the center of the cell as is standard. Comparisons of this formulation to analytic calculations numerically implemented in the WATQ code indicate a much improved solution at shallow depths compared to the calculations of the current LPF formulation. This formulation does not, however, take into account the dynamics of drainage from the unsaturated region above the water table which can be a major influence. I want to repeat Richard Winston's point that "All models are wrong". A model is not intrinsically good or bad. That judgment depends on what the model is used for. What may be better in one situation may not be better in another. You have highlighted this concept in your discussion of GSVE. MODFLOW still exists for two contradictory reasons; 1) as the name implies it is modular and hence easily extensible to specialized situations. 2) It is widely known and accepted that the code has accurately solved groundwater flow problems. It is incorrect to extend the second point to expect MODFLOW to accurately solve all groundwater flow problems. Once again required accuracy like good and bad depends on the problem that needs to be solved. That MODFLOW produces two results for the same geometry and boundary conditions means that it approximates the physics of the flow situation differently. It proper care is taken, the ability to change the approximation of the physics of flow is a good thing because the code can be tailored to specific problems. * Yes, having flexibility in averaging schemes is a good thing. The LPF package provides one more averaging scheme in the vertical direction. It does a harmonic average of the conductances, as opposed to the BCF package's harmonic average of vertical conductivity times area divided by separation distance as provided in the MODFLOW documentation for the various fully three-dimensional, and quasi-three-dimensional alternatives. However, I am guessing that people are encountering drastically different and surprising or unexpected results is because of the vertical direction treatment. Specifically, the cell thickness varies as a function of its saturated thickness in the LPF package for vertical direction treatment. For a simple example, consider a homogeneous case with two cells where the cell below is getting drier. The vertical conductance quickly rises due to the cell below drying up, to such levels that drain and dry up the cell above (which also causes the conductance to increase in a cascading manner) which would be a very different result than with use of any of MODFLOW's BCF options. Like I stated earlier, in fact, when things dry up, it should slow down the conductance due to physics change to unsaturated flow, and even a gravity flow assumption will not speed it up. Yes, there is a dilemma in GSVE models for vertical direction treatment because GSVE by its definition, applies only to horizontal direction flow - I have seen this in saltwater-freshwater GSVE models with multiple aquifers as well. And there too, when you give it specialized treatment, it is due to certain physically motivated considerations. Otherwise, the way MODFLOW BCF did it (or a use of grid-block cell thickness in the LPF package) is the best you can do with the VE assumptions being used, since conductance may be smaller than saturated levels due to unsaturated zone physics - but you don't know by how much, unless you

simulate unsaturated zone physics. Using saturated thickness instead of cell thickness in the vertical direction leads to results that are more removed from this physics. * This is one of the most challenging groundwater modelling projects you can find in real-world modelling... Ideally, you should use a saturated-unsaturated code, such as FEFLOW or Modflow surfact or similar. Such codes are able of simulating perched water conditions. If you are using traditional MODFLOW, and assuming saturated conditions, it is a bit more complicated to simulate such a scenario. You will probably have some dry cell issues (and perhaps some convergence problems), but it is possible to simulate the open pit crossing multiple layers using drain boundary conditions. For the upper layers, where seepage faces may occur, set the drain elevation slightly above the layer bottom elevation (say 0.1 m above cell bottom elevation). If you use Visual MODFLOW as the interface, that's very easy to do using formulas that will assign the correct elevation even in irregular layers. In that case, if the head in the pit wall is higher than the cell bottom elevation, Modflow's drain package will remove the outcropping groundwater. The conductance value (another input to the drain package) is typically a calibration parameter, obtained through matching drain removal rates with real values observed at the mine site. I suggest that you also use a solver that is not too aggressive, such as PCG2, and that you play a bit with re-wetting parameters and dumping factors until you obtain a stable solution. We've done several mine dewatering modelling projects using this approach and it definitely works for most cases, as long as you understand the code and it's limitations. * Dry cells in a multi-layer model are not necessarily a bad thing unless they do not accurately reflect the conditions in the field. Perhaps I am stating the obvious, but just for the record, dry cells occur in MODFLOW models when the calculated water table is below the bottom elevation of the grid cell. Since MODFLOW only simulates saturated groundwater flow, it does not represent a head value in the unsaturated cell, so the cell becomes 'dry' and MODFLOW assigns a 'dummy' head value usually in the range of -1e-30 (or something else easily recognized as a dry cell). Although the presence of dry cells in a model is often a hassle because they create problems with the convergence of the numerical solution, dry cells do not necessarily indicate a 'problem' or error with the model. In your case, the dry cells don't appear to be causing a problem with the solution convergence (or atleast, you haven't mentioned this problem) and you are getting a good mass balance, which is a good sign but doesn't necessarily indicate 'correctness' of the solution. If the water levels in the second layer of your model match closely with the observed field conditions, then the occurrence of dry cells in layer 1 appear to be reasonable. However, if the dry cells are occurring where you would otherwise expect saturated conditions (based on water levels in the field) then you probably need to revisit your boundary conditions to make sure they are reasonable (conductance values are often the culprit), check your initial heads and make sure you are not starting the iterations with dry cells, and make sure you have enough recharge flux entering the system to accommodate flux leaving the system through wells, rivers, and other boundary conditions. If all this looks good, then Mark Wilsanac gave some very good suggestions on how to change the solver and rewetting package settings in order to deal with dry cells. - Patrick Delaney * If the river boundary cells are the only place in your model where water may leave the system, then one of the most likely causes of the mass balance problem is the conductance values you are using for the river cells. Check your conductance values to make sure they are within reason. The conductance values can be reasonably estimated using the following formula:

C = (L x W x Kv)/B where L = length of the river in a cell W = width of the river in a cell Kv = vertical hydraulic conductivity of the riverbed in a cell B = thickness of the riverbed in a cell Most of the popular graphical interfaces for MODFLOW allow you to enter these 'physical' parameters and will automatically calculate the conductance values for each river boundary cell. Another possible cause for mass balance problems could be the solver you are using (SIP often produces larger mass balance than other solvers like the PCG or LMG) or perhaps the convergence criteria for the solver is not 'tight enough'. Just as an afterthought, the fact that you have more water entering the system from the rivers than leaving the system through the rivers does not, in itself, indicate a problem with the water balance. This indicates you have 'losing rivers' because the water table adjacent to the river is lower than the river stage. The excess water entering the system from the rivers may be leaving through other avenues such as evapotranspiration, pumping wells, drains, specified head head boundaries, etc. - Patrick Delaney * Check out if you don't have any dry well cells, or any other dry boundary cell. If a cell goes dry during simulation, it becomes inactive for MODFLOW, and any boundary assigned to that dry cell will be ignored. That means that your wells may be inactive (even though you assigned them to be pumping all the time). That typically produces a noticeable change in the budget, but sometimes the heads may not change too much, unless you look at specific well cells very closely. My suggestion is to first look into the budget for changes in well pumping/injection rates. If the total rate is different from those you entered into the model, then you have a dry well cell. Zoom in all well cells to identify which ones have become dry during that stress period (look in all model layers). * With a model set up with cell-to-cell size changes below 50%, there can be instabilities or failure to converge. This can be due to other reasons, or cell size change coupled to highly variable conditions in a transient solution. If a geologic unit pinches out, that should be represented primarily by changes to cell properties, and secondarily by changes in cell dimension (aspect-ratio rules apply vertically as well as horizontally). Remember to preserve aspect ratios of each cell below 10:1, and below 5:1 to be safe. It is acceptable to have a layer with variable properties from cell-to-cell. It is rarely acceptable to use MODFLOW cell dimensions to closely simulate physical variations in layers when they pinch out. * I have not used the WHS solver sold by waterloo hydrogeologic. while i therefore cannot comment on this solver and its claimed superiority to SIP, i feel quite sure that SIP has been the best performing solver for almost all models i have constructed over the past few years, and never had problems with mass balance which could not be successfully overcome by using SIP. i do recall a paper published in Ground Water years ago when WHS proponents at software developer Waterloo Hydrogeologic compared their WHS to SIP and PCG (among other public domain solvers). i felt that sip had been shortchanged and misrepresented in their analysis in that they had claimed that sip yielded a much greater mass balance error than that obtained when using their whs solver. what they had failed to do in their analysis, was to accordingly decrease the HCLOSE with their corresponding decrease in the acceleration parameter. this is the equivalent of not closing tightly enough on heads, and therefore, ending up with an excess mass balance error. this was an improper use of SIP.

I have found that lowering of the acceleration parameter, and corresponding decrease in the HCLOSE will yield a very low (acceptable) to zero mass balance error, but one must be careful in that the acceleration parameter is not set so low that the model converges prematurely with insufficient head change. * Check all inputs in the flow model first. Make sure your flows are realistic. Make sure your K's and recharge are realistic. Refine your grid and use dispersivity values that will minimize numerical dispersion. Make sure your grid cell size complies with appropriate Peclet criteria (as a rule of thumb, for finite differences, Pe should be ~ 2; that means your grid size should be roughly 2* your dispersivity value, in the area where, for MOC, this can be higher). I suggest your read some transport modelling books that explain this in detail. Be careful with using constant concentrations, they can act as a source, but also as sinks (if he concentration in neighbouring cells is higher than the ones you specified in the constant concentration node). * When she is referring to inflow, in plane, and outflow, I think what she is doing is looking at the color coding for the velocity vectors or pathlines. When she sees that there are very few arrows or pathlines with a green color (indicating 'in plane' or flat horizontal flow in the layer) she gets concerned for some reason. Most likely the lack of 'in plane' flow has very little to do with anything she is concerned about, it just reflects the fact that she has slight vertical gradients in the model. WRT to the vanishing concentrations, you could also suggest her to make sure she didn't input a default degradation rate of 0.5 or something very high and then forgot about it, or the other possibility is she has assigned a sorption coefficient of 1 (mistaking it for a retardation coefficient), or perhaps just a very large sorption coefficient of 0.01 L/mg which would cause the plume not to move at all, thus giving the impression that it is disappearing from the model, when in fact it hasn't moved at all. Another possibility is that she initially assigned non-conservative values of sorption and decay coefficients when she set up the transport model scenario, and she is now trying to modify these values in the Setup/Transport Engine dialog instead of in the Input mode (I've seen this problem a few times) or Visual MODFLOW. Finally, the fact that the solution solves in two iterations seems to indicate that the total run time of the MT3DMS solution may have been left as the default of 1 day (perhaps she assumed it automatically picks up the end time of the flow simulation). She can change this by selecting Run/MT3DMS/Output Times and entering the desired Simulation Time. * The simplest way of defining a starting concentration is to use the Recharge Coverage and specify a polygon shape that denotes the contaminant concentration to be applied to the model. * What to do when your contaminant plume does not migrate as expected. When using MT3D or RT3D for a contaminant transport simulation, if the plume fails to move as expected (or does not appear at all in your Visual MODFLOW output), here are a few common causes and their solutions. Problem #1: Overlay is not active, or masked by another overlay Solution #1: As with Head Equipotentials, Particles, and even Base maps and other overlays, you must make the Concentration overlay active in order to see it. Click the F9-Overlay button at the bottom of your screen, and ensure the overlay has been checked off (to make it active). It is also possible that other overlays are masking your concentrations. To resolve this, change the Overlay Order setting to User Defined, then highlight the Concentration overlay, and use the arrow buttons to move it up the list.

Problem #2: Simulation Output time Solution #2: In Visual MODFLOW, the Simulation Output time can be different for your flow and transport simulations. Check to see that you have defined the correct simulation time(s) in the Run Menu, by selecting MT3D (or RT3D) / Output - Time Steps. Additionally, in the Output Menu of Visual MODFLOW, ensure that you have selected the Concentration overlay as the active overlay, then select the Time button in the left toolbar, to change output times as desired. Problem #3: No concentration input assigned, or not assigned properly Solution #3: Ensure that you have correctly defined at least one cell with an Initial Concentration, or a transport boundary condition of one of the following types: Constant Concentration, Recharge Concentration, Evapotranspiration Concentration, or Point Source. Please NOTE that Concentration values entered for Concentration Observation Wells are used for calibration purposes only; they do not contribute mass to a transport simulation. Problem #4: Inadequate flow gradient Solution #4: Before running your transport simulation, run just the flow simulation (MODFLOW). Using Pathlines or Velocity Vectors, ensure that there is a sufficient flow gradient to cause plume migration. Problem #5: Incorrect reaction parameters Solution #5: Check the reaction options in the Main Menu, by clicking on Setup / Numeric Engines / Transport. Check that the correct Sorption and Reaction options have been selected. First order decay rates (dissolved and sorbed phases) may have a tremendous impact on the plume mass, since during the simulation, some contaminant mass may be removed from the model. If the decay rate is too high, your plume will show very small concentrations, and/or will not be visible at all at later simulation times. Decay rates (due to biodegradation or other mass-removal processes) can be taken from literature values, but this should be done with caution, as this parameter is highly site-specific for most compounds. A good reference on decay rates used in natural attenuation studies can be found in: Wiedemeier et al., 1999: Technical Protocol for Implementing the Intrinsic Remediation with Long-Term Monitoring Option for Natural Attenuation of Dissolved Phase Fuel Contamination in Ground Water. Air Force Center for Environmental Excellence, Brooks, AFB. This document can be downloaded from: http://www.afcee.brooks.af.mil/products/techtrans/monitorednaturalattenuation/protocols.asp The decay rate (l, or sometimes called Kd, with units of 1/day), is typically obtained from half-life values, converted into appropriate units using the following relationship: l = ln(2)/t1/2 Where t1/2 = half-life of the compound

In addition, check that your Kd value is correctly defined in the Species Parameters tab. A very large Kd value will result in an extremely high retardation factor, which can result in lack of contaminant movement through your model. For typical organic compounds (such as TCE, DCE, PCE, BTEX, etc.) where linear, reversible sorption can be assumed, retardation will be calculated using the following formula: Retardation = 1+ (Bulk Density/porosity)*(Kd) Where Kd = Partition coefficient For hydrophobic organics, Kd can be determined using the relationship below: Kd = Koc*foc Koc - octanol-carbon coefficient foc - organic carbon fraction in the aquifer A more detailed overview on Kd's can be found in: US-EPA, 1999. Understanding variation in Partition Coefficient, Kd, values. EPA 402-R-99-004A&B. US-EPA Office of Air and Radiation. This document can be downloaded from: http://www.epa.gov/radiation/cleanup/partition.htm * From the Modflow standpoint, water that exits the GW system through of a drain cell is accounted for in the budget, but is otherwise not simulated. The situation for river cells is similar--Modflow does not simulate any flow in the river, it only simulates exchange between the GW system and each individual river cell. So there is no way to route water from a drain cell to a river. The Streamflow Routing (STR) Package, on the other hand, does simulate flow in surface-water streams, accounting for flow rates in the stream and calculating stream stage, which is then used as the external head in the GW/SW exchange calculation. If you want to simulate exchange between drains and rivers, you could use the STR Package for both types of features. * How to import modflow model files to gw vistas 4? Select File/New and click the MODFLOW button. On the next dialog, click browse to go find the name file (*.nam). If this is an older MODFLOW88 dataset, then browse to find the *.bas file. Some datasets (especially those that were created without the use of a preprocessor) can be troublesome. * I want to input separately 2 sets of recharge into a modflow model: (1) Effective recharge from rainfall, and (2) urban leakage (same format as the (1)), which will overlap. Since both components are fairly complex and over a long time (1200 Stress Periods), I would like to keep them as independent as possible (avoiding the data compilation option, if you see what I mean). It is possible to do what you want using just the Recharge package. Here is how you do it. Define each recharge distribution for each stress period using parameters. You can use one set of parameters for the rainfall and another for the urban leakage. With each parameter, you can use multiplier arrays and zone arrays to define the spatial distribution of the recharge. Then, for

each stress period, use the parameters for that stress period to apply recharge from both rainfall and urban leakage. MODFLOW-2000 will add the contributions from all the parameters you use in a stress period to determine the total recharge. You should be able to do this with any reasonably up-to-date GUI for MODFLOW-2000. If you are using a GUI and you can't figure out how to do this, contact the GUI developer for assistance. If you aren't using a GUI, you can check the input format in the Online Guide for MODFLOW-2000 at http://water.usgs.gov/nrp/gwsoftware/modflow2000/MFDOC/guide.html or in the original MODFLOW documentation as modified in "Time-varying-parameters.pdf" (distributed with MODFLOW-2000). * Since I know you are using Modflow VKD (based on MF96, not MF2K) and that the wells package is almost certainly being used for many abstraction wells, I suggest 2 routes: 1. A FORTRAN utility to combine two recharge files - a days work? 2. To use a MF package that you aren't already using - but with some manipulation e.g. the rivers package. By setting the stage of the river as above any elevation in the model (1000m?) and the conductance equal to the quantity you want to recharge for that SP - with a 1 m difference between stage and bottom elevation you will get the appropriate amount leaking (recharge) to each cell. You can achieve the same result with the drains package by specifying 2 drains in each cell. The trick is to have one with a positive conductance and the other negative, but the same value. Make the difference in elevation 1m and the magnitude of the two conductance values the quantity you want to leak/abstract. If the elevation is outside the range of variation in the model, you end up with a constant recharge/abstraction from the model. For composite/trick boundary conditions of this type I suggest drawing figures for them like those in the MF88 book, it will help to understand what I'm trying to say. * Design the shoreline? Depending if you are using FEMWATER or MODFLOW, it is handled differently. With FEMWATER, you want the boundary of your 3D mesh to follow the shoreline. With MODFLOW, you are working with a 3D grid. So, you need to mark the cells that contain the shoreline with a fixed head boundary condition, and the cells outside this area will be disabled. If you are attempting to model the interaction with the sea bed, then you will need to set the top elevations of the mesh or grid to match that of the sea bed. However, generally you end your model at the shoreline. * MT3D Performance What you need most depends on where your bottle-neck occurs. You need to monitor CPU and RAM usage on the computer during the MT3D run. If you run Windows 2000 or XP, you can right click (for a right-handed pointing device) on the Taskbar (grey bar at bottom of screen), then select Task Manager in the menu pop-up. Once Windows Task Manager opens, select the tab labeled "Performance". "Performance" shows CPU usage graphically and memory usage in numeric form. If the "CPU Usage History" (top graph) stays maxed at 100%, then a CPU speed increase should help. In parallel, observe the "Physical Memory" values. If the "Available" value becomes a small fraction of the "Total", AND the "Commit Charge, Total" exceeds the "Physical Memory, Total" then you are likely to have a RAM-limited condition. In this case, more RAM would help. With 300,000 cells, consider 1 Gbyte or more. If both things are happening (unlikely), then upgrading CPU and RAM should help. Be aware that upgrading one component sometimes results in the other becoming the limiting condition. That is why it is

"unlikely" both limits are happening simultaneously. Usually one tops out before the other. If you run Win98... consider upgrading the OS. * There are various things that you have to do to get Surfact running in Vista 3: 1. You can only have a limited number of packages. For example, if you are doing contaminant transport modelling then you have to turn some other packages off, otherwise Surfact will not run e.g. you can turn the cell by cell flow packages off (just type 0 into the packages number in Model - Modflow - Packages). Another pace you can turn packages off is in the output control: Turn off the drawdown unit. 2. You have to modify some things in the Model - Modflow options and some in the Modflow - Surfact options. These are: Model - Modflow - Packages: change modflow version to surfact; change solver to PCG4; and untick automatically reset package units Model - Modflow - BCF package: tick the box to use the BCF4 package Model - Surfact - Packages: add a unit number (one that you have not yet used) and tick the box for BCF4; add the unit number to PCG4 and tick the box (use same unit number as shown in Model - Modflow- Packages); add a unit number and tick box to RSF4 is you want the model to simulate surface seepages (and remember to turn this unit on in Model - Surfact - RSF4 package). Model - Paths to models: set modflowwin32 option to DO NOT USE; put in correct path for surfact in the modflow box. I can't think of anything else right now - I always forget one thing and spend a couple of hours wondering why surfact won't run. If you have any difficulties it would be helpful to know if the surfact Dos window displays and if so, what it says in the window. * The most common problem when using SURFACT is setting up the program file. Select Model/Paths to Models. Change the MODFLOWwin32 Option at the top to "do not use". That tells Vistas that you are not using one of our model DLLs. Then change the MODFLOW program from MFWIN32.DLL to whatever your SURFACT program is (usually it is msft.exe). * I believe that if you are using the latest version of SURFACT (V2.2) that problem 1. below is no longer an issue. It used to be that SURFACT could only open a limited number of files but that is no longer the case. Also, if you upgrade to Vistas Version 4 and SURFACT V2.2 the link between the two programs is easier to establish. The same holds true for all of the other versions of MODFLOW that Vistas supports (MODFLOW88, MODFLOW96 in double precision, MODFLOW2000, MODFLOWT, MODFLOW-SURFACT, SEAWAT, and SEAWAT2000). * VS2DI or VS2DT, being a Richards equation-based models have certain PROs and CONs when used for groundwater recharge estimation: Pros: the Richards equation models are the most theoretically based and allow representation of the flow processes in porous mediums more realistically than the water-balance models. They have been well tested against field and laboratory experiments and proved to provide good correlation with observed results. Cons: the major complicity of the Richards equation, comparing to the saturated flow models, is that its coefficients are non-linear functions of the pressure potential. To approximate these

functions, three different algebraic equations are usually used (named after their authors): Brooks and Corey's, Gardner's, and van Genuchten's. Richards equation can be solved with a very limited number of boundary conditions. The condition for water at the boundary can be specified either as the flux of water across the boundary, the head at the boundary or as a combination of specified head and specified flux. The Richards equation with these boundary conditions is a nonlinear partial difference equation that has no close-form or analytical solution. To solve the Richards equation, a finite-difference method of numerical approximation is applied in VS2DI. Searching for a best numerical method for the Richards equation became a special field in hydrologic science, particularly in 1980's, however, almost all implemented approaches cannot assure that the model will unconditionally converge and simulation results will be obtained. It is an art to get the model to work correctly (produce results for the whole simulation period with a good water balance) when you have varying flow boundary conditions in your profile. Our experience also proved that applications of Richards equation-based models to highly heterogeneous soils with variable hydraulic properties can be extremely difficult to execute and time consuming. These days there more and more examples when less sophisticated in flow equation but much more advanced in terms of weather/plant effects water-balance model HELP is used to estimate groundwater recharge. * I will suggest you keep the north and south boundary as variable head boundary. This I am suggesting you treating there is no river, lake etc. If these features are present on north or south side of model domain, then you have look for another type of boundary. Regarding limited observation well data and that too is limited to layer1, then it will not be possible to do any realistic modelling. If there is no scope of generating additional data, i would like to suggest that you confine your modelling activity to first layer only. * What makes a good correlation depends on what your expectations are. If you are conducting exploratory research, maybe 0.2 isnt bad. If you are testing a known process having some natural variability, 0.6 might be acceptable. But if you are calibrating instruments, maybe 0.9 isnt good enough. If you dont have any clear expectations, how do you tell if a correlation is good? The answer comes in three parts. 1. Value - Square the correlation coefficient (called the coefficient of determination or just R-square). R-square is the proportion of variation in the dependent variable (y) that can be accounted for by the independent variable (x). You might be able to decide how good your correlation is from a gut feel for how much of the variability you wanted your model to account for. 2. Significance - Every calculated correlation coefficient is an estimate. The real value may be somewhat more or somewhat less. You can conduct a statistical test to determine if the correlation you calculated is different from zero (or some other number if its relevant). The larger the calculated correlation and the greater the number of samples, the more likely the correlation will be significantly different from zero. For example, a correlation of 0.59 (R-square of 0.35) would be significantly greater than zero based on about 25 samples, but a correlation of 0.01 wouldnt be significant with 250 samples. 3. Plots - You should always plot the data used to calculate a correlation to ensure that the coefficient adequately represents the relationship. Specifically, the data should be linearly related and free of outliers. There are a few other things to look for (e.g., hidden trends, auto-correlation) that I wont go into here.

So, the reviewer who said that the "R square" value of 0.35 has no significance and is just as good or as bad as say 0.01 would only be correct if she looked at a statistical test of whether the value was significantly different from zero. R square does not need to be more than any particular value to draw a comparison. Remember, too, that no relationship may also be an important finding. * If you are considering GUI's for your project, you may want to look at Visual MODFLOW Pro. In particular, if cell splitting, or grid refinement is an issue for you. Visual MODFLOW includes two distinct features that help users with this. 1) Cell refinement tool 2) Grid smoothing tool 1) Cell refinement/splitting tool: You'll need no more than a few clicks to edit your grid in a very flexible manner. To split your cells, just click on 'Refine by ...", and specify how many times you want the cell to be split into (2, 3, 4...). Then, select the area you need to refine with your mouse.You can apply this to one row or column, or all cells in the domain - whatever you wish. The same applies for splitting layers and coarsening the grid. 2) Grid smoothing tool: After grid refinement, you may want to use the grid smoothing routine to produce a smooth transition between coarser areas to highly discretized ones. Again, just a few clicks. This is actually a perfect tool to help you better understand the effects grid size and refinement will have on your model results. Another feature that may be useful for you is Visual MODFLOW Pro's batch capabilities. It will allow you to prepare all data input files in different model projects and run them in 'batch mode'. * The interpretation method is different according to the geology. You have methods for porous media, others for fissured aquifer, others for double porosity media. Methods also vary according to the type of aquifer (confined or unconfined; semi-confined with a leaky aquifer...) and corrections can be made according to the type of well (partially penetrating well; pumping well with head loss; skin effect and so on). Finally, boundary effects (barrier or recharge) may affect significantly your interpretation. In order to remain pragmatic I would suggest you to start with the simplest method : the Theis formula, (or even the Jacob approximation if your pumping time is big enough) and check wether you get a good fit or not with realistic parameters. Even a fissured unconfined aquifer can be interpreted with Theis when your radius of influence is wide so that the fissure network can be assimilated to an homogeneous equivalent porous media and the depletion of the water level is moderate compared to the thickness of the aquifer. This should give you a rough estimate of your transmissivity, which is in many cases, sufficient. If you need more accuracy you can use specific softwares. You will find on the net many softwares proposing many different methods. You don't need Pest as they have their own fitting engine. AQTESOLV for instance proposes methods with well storage effect. However, as soon as a software proposes its own adjustment, it's becoming very difficult to know what's you're doing and you end up playing a video game. So be careful, especially if you lack experience in this domain, not to use complex methods with many parameters which will end in a perfect fit but with irrealistic results. In my company (French geological survey)we use ISAPE, a software which is no more commercialised, but whose philosophy is to let the user propose realistic values and where you can add or remove well effects or boundary influence. I wish this approach were more widespread.

* I think your scenario will represent two different geologic materials with peculiar intrinsic properties. Hence, I suppose this will require two separate simulations. To the best of my knowledge, MODFLOW 2000 only allow one K matrix values per layer per simulation and this is incorporated within the flow package. However, if the introduction of the permeable barrier is controlled along defined specific paths or unique relatively smaller area(s), then you can use one simulation with two stress periods and incorporate Horizontal Flow Barrier Package in the second period to account for the permeable barrier distribution. This is not suppose to be of much problem especially if you incorporate your modflow with GIS package like GRASS GIS. * Sophisticated modeling programs like Visual Modflow Pro, or any of a number of similar packages, will run properly with a variety of inputs that may not be physically realistic - and thus provide garbage for output. With that in mind, here are some thoughts on your questions: 1) You may divide a single aquifer into more than one layer if there is some reason to look at vertical stratification within the aquifer. For example, a thick aquifer may have different hydraulic or transport properties in different vertical zones. You may also divide a single aquifer into several layers if you want to examine the vertical flow patterns due to pumping from different zones within the aquifer. 2) The top of Layer 1 could be the ground surface, the water table (not as common because it can move vertically), or the top of the uppermost aquifer or aquitard of interest depending on the physical system being modeled. 3) You'll need to assign boundary conditions no matter what the size of the physical system is. If you don't have natural physical boundaries, one strategy is to assign boundary conditions at a distance far enough away from the area of interest to minimize the effects of boundary assumptions on the solution. * It is appropriate to use a single layer if you feel that it is reasonable to ignore vertical flow in your model. Assuming that you would be using the Layer Property Flow (LPF) Package of Modflow-2000, and if the aquifer represented by layer 1 is not overlain by a confining unit, then specifying the top of layer 1 as ground surface is reasonable. The reason that you would be better off this way than specifying the water table elevation as the layer top is that the LPF Package only provides two options with respect to the confined/unconfined status of a layer: Confined or Convertible. If the head in a cell goes above the specified top of a convertible layer, the cell is treated as confined. Every model needs to have at least one fixed head, either as a constant-head cell or as a fixed external head (such as the stage at a river cell), or the solver will not be able to reach a solution. You may need to expand the domain of your problem to encompass a fixed head. * 1- Model design depends greatly on your project goals. For example, you could have one aquifer, but an overlying layer (and/or underlying layer) as well. Or, you could be modelling just 1 aquifer, but split it up into several layers in VMOD to provide more detailed flow information (i.e. if the aquifer is 100 feet thick, having just 1 layer will not provide very detailed output. You can subdivide the 1 "real-world" layer into 10 "model" layers, all with the same hydrological properties, in order to obtain more detailed output). Or, you can simply design a 1-layer model if that is all your project requires. 2- In Visual MODFLOW, Layer 1 represents the uppermost layer of your profile, and the top of Layer 1 is therefore the ground surface. Layers in Visual MODFLOW represent soil layers (or conceptual soil layers)....the location of groundwater in these layers is determined by your model inputs, and the resulting calculations of the flow model.

3- The assignment of boundary conditions is a question that often goes beyond the scope of Technical Support, and enters the realm of Extended Modeling Support. However, to provide you with some guidelines: -If you have no "obvious" water inputs (streams, rivers, lakes, injection wells, recharge from rainfall, etc.), you can try assigning an upgradient equipotential line, and downgradient equipotential line, using Constant Head Cells at the upgradient and downgradient edges of your model, to represent the regional water table (other boundaries may be more appropriate, depending on your specific model domain and objectives). You can then add additional inputs (extraction wells, contaminant sources, etc.) to the model region, which will be influenced by your regional flow gradient. 4- I would recommend taking some time to run through the Tutorials included with your Visual MODFLOW software (the PDF instruction files are located on your installation CD-ROM, and the files are automatically copied to the Tutorial subfolder in your Visual MODFLOW program folder). These tutorials are designed to teach you how to work with your software, and how to design a model, import data for model inputs, run the various packages, calibrate your model, examine your output in 2D and 3D, and export data. * You can use the field interpolator program in the PMWIN package. Using the field interpolator you can interpolate the field data in txt file to the grid point of your model. On the other hand, you can import any dxf file which illustrate the changes in hydraulic properties to help you in the graphical interface. * Upwind discretization always has an artificial diffusion term no mater how refined your grid is (even thought it diminishes with grid refinement). Central discretization always produces an artificial dispersion term that causes "wigles" however you can't control it with grid refinement. For almost any Pellet number the upwind scheme will produce positive coefficients for your coefficient matrix, while that isn't true for central differences. However central differences has "second order accuracy" while upwind only has first order. You can visualize numerical diffusion and dispersion mathematically by deriving the "equivalent or modified equation" see http://widget.ecn.purdue.edu/~jmurthy/me608/main.pdf for some cool notes on this (as good as any book I've read). Physically you can visualize numerical diffusion by imagining a 1D discretization of a domain | : | : | : | where | are faces and : is the center of the cells. If your initial property is 1 on a given cell say i and 0 on all others it would look like this: | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 So using upwind with a courant = v dt / dx of say 0.5 and a velocity from left to right after the first time step the accurate solution would be something like (zooming on cell i): | 0 { 0.5 | 0.5 } 0 | Note that I have created a new cell delimited by the {} brackets this "new cell" that doesn't exist in our domain is exactly half of cell i and alf of cell i+1. The property's value will be 0.5 inside this new cell and 0 outside it (meaning that cell i and i+1 would have half with 0.5 and half with 0). This would be the accurate solution because a courant of 0.5 makes the property shift half a cell every time step. However we are bounded to our initial discretization so the solution we will obtain will be: | 0 | 0.5 | 0.5 |

so you can see than instead of having a 0.5 of property in a volume the size of a cell we have 0.5 in a volume the size of 2 cell thus lowering the concentration. This happened because the definition of concentration is C = lim vol ->0 DM/DVOL and we implemented it as DM/DVOL. If we would diminish the size of our grid so that currant = 1 we could obtain the exact solution. So the smaller your grid is the more accurate your solution will be. This is the physically correct way of solving the diffusion problem another way is to use higher order schemes witch usually maintain the sharpness of the gradients but usually remove mass from all the wrong places (even thought they conserve it). Looking at the same example a central differences discretization would derive something like: | -0.25 | 0.75 | 0.25 | Cell i-1 will produce a negative property because the concentration al the face between i-1 and i is 0.5 cell i +1 will obtain the accurate concentration of 0.25 (0.5 in half the volume of a cell). But we are producing the famous "wigles" close to the properties gradient. There are smarter second order schemes like qwick or TVD schemes. * In visual MODFLOW there is provision for importing MODFLOW files (*.bas files). In PMWin there is provision for conversion of MODFLOW88/96 (*.nam). I have tried to import *.bas file in Visual MODFLOW and found that model data is not correctly coming to model. These options may be available in GMS but I am unable to locate the exact sub-menu. Number of options is available for opening different format files In the FILE menu. But It does not permit opening of *.bas or *.nam files. This option must be in different menu. * To clarify the capabilities of Visual MODFLOW, the latest version (Version 4.0.0, released in April of 2004) is capable of importing any MODFLOW-based model. The import capabilities have been significantly enhanced since the release of Version 2.8.2 in 1999. The Version 4.0 User's Manual contains a detailed description of how to convert an older MODFLOW-88 or MODFLOW-96 model to MODFLOW-2000 format (using the USGS MF96TO2K program), then import the MODFLOW-2000 model directly into Visual MODFLOW. Although previous versions of Visual MODFLOW were able to import existing MODFLOW data sets, the import process often ignored important data such as Specific Storage and Specific Yield values, and it was not able to accommodate quasi 3D models containing implicitly represented aquitards. Visual MODFLOW 4.0 has a significantly improved importing process that supports both MODFLOW-2000 Groundwater Process data sets, and older data sets from MODFLOW-96 and MODFLOW-88. This process utilizes a modified version of the USGS utility (MF962MF2K.EXE) to convert MODFLOW-96 and MODFLOW-88 data sets to MODFLOW-2000 data sets, and then imports the MODFLOW-2000 data set into the Visual MODFLOW graphical environment. * Unfortunately, Version 3.1 was limited to DXF and BMP file formats. However, Version 4.0 released early in 2004 includes the following import possibilities. Plus, Version 4 also includes a handy layer management feature so you can adjust layer sequencing easily. - DXF - BMP - SHP - GIF - JPG - PNG

In terms of animating your heads, you can also do this using the built-in VMOD 3D-Explorer (part of Visual MODFLOW Pro). * You are write in pointing out that in Visual Modflow only DXF and BMP files can be imported as base map. Even in latest version Visual Modflow Pro 4.0 other than DXF and BMP import is not possible. I would suggest you that try to reduce the depth of BMP or save the original base in monochrome or less colour depth (grey scale). This will reduce your size of file considerably. This will also help you fast operation of the programme. You may import Visual Modflow generated heads to GMS. You import the super model file from open menu in GMS 4.0. * My suggestion would be to read the model directly into GMS. GMS supports multiple formats for image display (including the format that you have) and you can perform the animations like you need to. * The latest version of Visual MODFLOW 4.0 actually supports much more than BMP and DXF files. For example... - DXF - BMP - SHP - GIF - JPG - PNG * 1) If you have a point source boundary condition assigned to an injection well, the concentration of contaminant in the injected water remains constant, however the volume of water being injected into your flow model relates directly to the pumping rate. Therefore, the total volume of contaminant being added to the model also varies directly with the pumping rate assigned to the injection well. Therefore, I would anticipate that your model would react as you have indicated (reducing the injection rate results in a lower detected concentration level at your observation well) due to dispersion of a smaller volume of contaminant over the same area. 2) The following is an overview of the solution methods from the head of our Software department: Numerical dispersion and oscillations depend on a number of factors such as grid discretization, time step size, and solution method. Typically, when a solver is oscillation-free, such as Upstream Finite Differences (UFD), it is prone to high numerical dispersion. What you have described sounds like a typical breakthrough curve obtained by MOC. MOC can have mass balance issues, and I strongly suggest you look into mass balance errors before trying to reach any conclusions. A high mass balance error makes it very difficult to rely on the output data. There is no such thing as a "best solver" for all transport model cases. In order to determine which solver to use, you need to understand what the dominant transport process is in your project area (i.e. advection or dispersion). This is estimated by determining the Peclet Number (see the MT3D PDF Manual located in the Manual folder of your Visual MODFLOW installation CD-ROM for details). Pe (after some simplifying assumptions for most cases) is roughly equal to cell length/longitudinal dispersivity. So, if your dispersivity is 10m and your

cell length is 20m, your Pe=20/10 = 2. It is easy to see that large grid cells and small dispersivities will produce large Pe Numbers. For Peclet numbers that are smaller then 2, dispersion is the dominant process and finite differences (UFD) will work fine. For Peclet numbers larger than 10, advection dominates and MOC is a good option. TVD is a very good option in both cases. Every solver has pros and cons. Some are faster (e.g.: UFD, implicit in time) some are slower (MOC, explicit UFD). Some are more prone to numerical dispersion, but are oscillation-free (UFD, MMOC). Some are the opposite (Central-in-space Finite Differences, MOC, HMOC). TVD seems to be the most flexible solver available in MT3D. It does not typically present numerical dispersion nor oscillation, but is slower than Implicit UFD). Sounds confusing? I suggest Dr. Chunmiao Zheng's book on contaminant transport, a must read if working with MT3D. What we suggest in our courses is to use Implicit UFD in the early stages of the modeling project, and check results with TVD or MOC at the end. We then use Implicit UFD, TVD, or MOC for final simulations, depending on the results of initial simulations and objectives of the modeling project. We don't recommend using central-in-space finite differences in general, because it can produce oscillations that are difficult to overcome. However, it does not present numerical dispersion. To avoid the discrepancies your are describing, you could try using TVD (with Cholesky pre-conditioning) or Upstream Finite Differences (UFD) which is faster, but subject to numerical dispersion. A final note on grid size. The effects of numerical dispersion can be reduced by increasing your grid discretization (i.e. use smaller cells where contamination is being simulated). Actually, the grid cell size can be calculated using the Peclet Number. For example, if you want to use finite differences, make sure your grid size is small enough to achieve a Peclet that is smaller than 2. * The PEST program (developed by John Doherty) is a Parameter Estimation tool that can be used to optimize model parameters based on field observation data. Visual PEST is a graphical interface to the PEST program, and WINPEST is the Visual MODFLOW interface for PEST. * If you are looking for a GUI independent calibration tool, you can try Visual PEST-ASP. This version of PEST-ASP actually includes a user-friendly GUI and will allow you to automatically calibrate a complete range of parameters from your groundwater model. If you are using Visual MODFLOW, you can use an add-on program called WinPEST. This is a fully integrated version of PEST. Since it is part of the Visual MODFLOW GUI, calibrate parameters such as hydraulic conductivity is really quick and easy. * You will not be able to import *.vmf directly. You go file menu of GMS then click open sub-menu, then you search *.bas file in directory containing simulation made in visual modflow. You can import basic files of Visual Modflow directly into GMS. * I am guessing that the counting of the cells you want to do is for the estimation of the water logged area? If that's the case than you could quickly get this area using the following procedure: Save the simulation result you are using as a matrix file. In Search and Modify change the values in this matrix to integers. (For instance if you are interested in the area of the cells with values in the range of between 0 and -1 and nothing else than modify all these values to 1 and all the remaining values to 0.) - save this matrix as a separate file. Use program "Zone2dxf.exe" from John Doherty's set of "Groundwater Utilities" to create a dxf file with contours of the 0 to -1 zone. I think you can get Utilities from the internet. If not contact me directly. Use Autocad (or any other program that can calculate polygon areas) area

function/option to calculate the area of the polygon. All this can be done within just a couple minutes. * The RMS error is one method of judging calibration and you can not assume with RMS error that your model is calibrated. It is related to only obs. and cal. head. Normally in Visual Modflow 5-10 % RMS error is acceptable. You will find two parallel line above and below the 45 degree (i.e. 100 percent fit line) in visual modflow obs and cal. head display. I am not sure but It looks to me that you model is producing more than 60 % error. All depends on your target. I will suggest that you must modify your parameters so that you may able to minimize your dry cells in model. * I am using visual modflow for contaminant transport. while running the model i get error message as below : ERROR: MAXIMUM NUMBER OF PARTICLES ALLOWED [MXPART] IS 1000000 ACTUAL NUMBER OF PARTICLES NEEDED AT THIS POINT 1000004 INCREASE VALUE OF [MXPART] IN ADVECTION INPUT FILE ERROR: MAXIMUM NUMBER OF PARTICLES ALLOWED [MXPART] IS 1000000. I have already assigned maximum number of particles that could be assigned but still i get that error. Is there any way to increase the number? Try going to the Run Menu of Visual Modflow and click on the MT3D option at the top of the screen and select Solution Method. In the Solution Method window, select the Options button beside Advection Terms. In this window, increase the MXPART variable from 100000 to a number greater than that specified in the error message. You may alternately reduce the "Max particles per Cell" value from 30 to 20 or even 10 if this error continues to be a problem. * I would like to know if I can work SEAWAT with PMWIN 5.3 Also I would like to know if PMWIN 5.3 can be a graphical interface of SEAWAT and if so how? Looking at SEWAT manual you can use PMWIN to prepare your data files. However, you will have to make small modifications/adjustments to some of the input files (they are all in ASCII text format) outside PMWIN using a text editor. Once you have done this you will have to run SEAWAT from the DOS window. On the completion of the run you will be able to read simulation results using PMWIN again. Remember that SEWAT will give you simulated heads as equivalent freshwater heads. You will have to convert those to the "field" heads using a formula shown in SEAWAT manual. This formula converts heads using simulated fresh water heads and concentrations. Conversion will have to be done outside SEAWAT/PMWIN (e.g. in a spreadsheet) but once it is done and saved as an matrix you will be able to read it into PMWIN and export it from there as an image or dxf file. * 1. Typically PEST does not have an enormous RAM requirement - if you obtain the latest (from www.sspa.com/pest) there are a number of memory-conserving storage routines that help reduce memory requirements. The amount of RAM PEST requires - and also the relative speed of PEST execution at any particular CPU speed - is often a function of the number of parameters estimated as well as the number of observations. I have estimated a couple hundred parameters with a few thousand observations on a 3.2GHz/1MB RAM PC. However, this leads to the next aspect; 2. The forward model itself (MODFLOW) may be the determinant for RAM - the number of nodes you mention though is not extremely high, though the number of stress periods is fairly unusual. Beyond a certain point you will not gain any additional speed by having additional RAM. You can make some rough calculations of the RAM Modflow will use based on the size of the XARRAY when dimensioned (I forget if it is actually reported in the LST/GLO files?) 3. Finally - if you are estimating more than a handful of parameters and/or your forward model takes more than a few minutes to execute, I would strongly recommend using parallel PEST - I believe this comes with GMS, and if not you can get it from the same URL as above - this enables you to run the independent parameter perturbations (and sometimes

lambda runs) on different computers in a standard PC (office environment) network, and leads to enormous time savings. 4. I can't speak to the RAM requirements of GMS, but their help desk is extremely responsive and helpful. Summing up - naturally, if you can obtain a faster and bigger PC - why not(!) - but I would be inclined to focus on the processor speed, and would not go above 2GB RAM unless there is a very good reason to do so. And - I would use Parallel PEST for your calibration, from what I understand of it. You can make a lot more headway with 10 computers with 2.4 GHz processors and 1GB RAM in parallel than with one souped-up computer! * In Version 4 of Vmod there is an export tool for every single layer surface. You can find it in the input menu under file/export/data. Create a xyz file - choose your coordinate system and then reimport in your transient data set. * I am not familiar with Pmwin, but you could define head observations (and other types of observations) as part of the Modflow-2000 input and set up the Observation Process input file with the variable OUTNAM as something other than "NONE" to generate a series of output files, including one with the extension "_os". This file will list the model-calculated and observed values as well as the name you provide for each observation. * First you check the capability of the software you are interested to use it for modelling. If software is capable to manage the huge number (200,000 to 400,000) of cells (grids), then you need huge hard disk space as well as RAM. But at the same time it will be very difficult to manage the model. Your output modflow file will will more than 5 to 10GB. Opening such a huge output file will need more RAM. * You must first conceptualize the system. What are your boundaries? Is the system confined, unconfined, or both? What does the potentiometric surface look like? Are there any pumpage wells, streams, springs, or lakes within the system? Where and how does recharge occur and at what rate? Do you have any hydraulic property measurements? What does the structure look like? Is the system in steady-state? ... and most importantly, what is the purpose of the modeling effort? Once you have answers to these questions then you can begin to assess what your data needs are or will be. You can then determine the level of descretization and begin to organize the data into an appropriate format for model input. Then you calibrate to steady-state conditions and then to transient if necessary. Do a sensitivity analysis of your parameters and proceed to answering the questions being asked of the modeling effort. GW modeling is a very complicated effort which should be avoided if there is an easier way to solve your problem. * Waterloo Hydrogeologic, Inc. has developed a Non-Convergence Tip Sheet for Visual MODFLOW users. Even if you are not working with Visual MODFLOW, the suggestions contained in this Tip Sheet should be applicable to your modeling project. * I cannot give you any definite answers based on the information you provided, but you need to know why the cells are drying. Do they go dry due to the stresses imposed on the aquifer (this is what I think you alluded to), or do they go dry as a result of large head changes during the iteration process? In the former case, you can try using the Rewetting Package, but you should first decide whether or not your conceptual model can allow for a thicker top layer. This would be the easiest solution. If the cells are going dry due to unwidely iterations, try adjusting the seed parameter so as to slow down the iterations. * I assume that you are modelling only the unconfined aquifer system i.e. top aquifer. There is no problem in selecting partially penetrated open dugwells as observation wells. Only thing you

have to see that water level in the well do not go beyond the base of the well any time in the year. * Partial penetration for monitoring groundwater elevations in your unconfined alluvial aquifer should be OK. It might be a problem if you have significant vertical gradients within the alluvial system, but that is generally not the case. * You may try with different solver for convergence of your model. You may slightly increase your convergence criteria if there is large season variation in head or water table. This will not affect much in results. But without changing anything, you will able to solve your problem using different solver. You try to change convergence criteria only if you have only one solver with you. * What is your mass balance in the model? If it is reasonable than try to increase your convergence to say 0.05 or 0.1 re-run the model and check the mass balance. However, before you do it try to run the model using different solvers (and pre-conditioning methods if using PCG2), increase the number of iterations and try different relaxation and dumping parameters (say between 0.9 and 1). If your mass balance keeps below say 3% than you are doing well. I have heard the opinion that in some difficult cases mass balance as high as 10% may be acceptable and in general below 5% is OK. You may also try different time steps and PERLEN factor of up to 1.5. The idea is to have the very first time step, in the new stress period, quite short as this helps to stabilize the model. * The grid / mesh size must be the same for all layers. However, you can use no-flow cells to activate/deactivate different areas of the model domain by each layers to more accurately represent your lithology. This may be sufficient to meet your needs. * You can load up to 5 maps (I use usually dxf)at once. They will be shown as a background on all layers. You may have to switch them on and off manually if you require one particular map shown in a particular model layer. * The Recharge Concentration option will introduce leachate mass into the system through the recharge flux across the uppermost wet layer of the model. As such, the concentrations you will see in the model cells will most likely be much less than the concentrations you specify in your recharge, but this is probably the most realistic way to approach the problem from the point of view of getting a proper accounting of the leachate mass in the system. The Constant Concentration option will introduce the leachate into the system assuming a uniform specified concentration in the cell in which you assign it. If your model is discretized finely enough, this may not be a problem, but in many cases this approach will introduce much more solute mass into the system than is probable. However, if all you are interested in doing is matching the observed concentrations rather than the overall mass of the constituents, this approach may be acceptable. However, you have to be careful when using the Specified Concentrations because you cannot turn off this boundary condition at a later time. If you want to turn it off (so to speak) you have to set the concentration in the cell equal to zero for the stress periods when it is off. If you try to assign it only for a portion of the overall simulation time, it will automatically apply the final concentration for the remainder of the simulation. If you are using a two dimensional (one layer) model and the overall mass fluxes across a boundary are important, you will likely have to use the Recharge Concentration boundary. But in this case you will likely have a very difficult time matching any of your observed concentrations because the two dimensional model will effectively dilute the mass across the entire depth of the cell.

If you are using a two dimensional (one layer) model and you are only interested in the concentrations, then you can get away with using the Constant Concentration boundary condition. But we aware that the concentration is uniform across the entire depth of the cell, so you are likely significantly over-predicting the overall solute mass in the system. In general, the same applies for three dimensional models except if you have a fine vertical discretization at the source of the contamination. In this case it may be feasible to use a Constant Concentration boundary condition in the cell where the leachate is being introduced to the system. * Transport Model Considerations Volume 2: Properties Waterloo Hydrogeologic's Technical Support department presents the second of two articles highlighting the Contaminant Transport component of Visual MODFLOW. Along with defining the sources of contaminants in your model (see Volume 1: Boundary Conditions), it is important to consider the Transport Properties of your model: Bulk Density: The Soil Bulk Density is used to calculate the Retardation Coefficient for each chemical species. The Retardation Coefficient is used to calculate the 'retarded' flow velocity of each chemical species. The retarded flow velocity is used to calculate the advective transport of each species. The options available for customizing the Soil Bulk Density values will depend on the Transport Engine selected for the current transport variant. Model Parameters: This is only available for models where RT3D is the selected Transport Engine and the Reaction Parameters are Spatially variable (these settings are specified during the setup of the Transport Engine). Here you can define the soil Bulk Density and other reaction parameters (decay rates, reaction rates, yield coefficients, etc.) required for the selected Reaction model (the parameters required for each reaction model are described in Appendix C of your Users Manual). Species Parameters: Here you can specify the sorption and reaction parameters used by the selected Transport Engine. Each of the Transport Engines can handle a different set of sorption methods and reactions, so the options for customizing the sorption and reaction parameters will depend on which Transport Engine is selected for the current Transport Variant (see Appendix C of the Users Manual for a summary of the available sorption options for each Transport Engine). Some of the parameters that may need to be defined include: Distribution Coefficients (Kd), and first order reaction rates for the dissolved phase and for the sorbed phase. Initial Concentration: In many cases, the historical conditions of the site are unknown, and the contaminant source has been removed or remediated. However, the groundwater contamination is still present and the mass transport simulation must be run forward in time, starting from the existing conditions, to predict the potential downstream impacts. Here you can define the existing conditions (background groundwater concentrations) of each chemical species being simulated. Initial concentrations can be defined using the .ucn file from a previous simulation. Dispersion:

Dispersion is a physical process that dilutes, or spreads, the contaminant mass in the X, Y and Z directions along the advective path of the plume, reducing the solute concentration. Dispersion is caused by the tortuosity of the flowpaths of the groundwater as it travels through the interconnected pores of the soil. MT3D calculates the Dispersion tensor for the mass transport model using the following parameters: Longitudinal Dispersivity for each transport grid cell Ratio of Horizontal to Longitudinal Dispersivity for each layer Ratio of Vertical to Longitudinal Dispersivity for each layer Molecular Diffusion Coefficient for each layer * I am using visual modflow for contaminant transport. while running the model i get error message as below : ERROR: MAXIMUM NUMBER OF PARTICLES ALLOWED [MXPART] IS 1000000 ACTUAL NUMBER OF PARTICLES NEEDED AT THIS POINT 1000004 INCREASE VALUE OF [MXPART] IN ADVECTION INPUT FILE ERROR: MAXIMUM NUMBER OF PARTICLES ALLOWED [MXPART] IS 1000000. I have already assigned maximum number of particles that could be assigned but still i gt that error. To resolve your error message: Go to the Run Menu. Click on the MT3D option at the top of the screen and select Solution Method. In the Solution Method window, select the Options button beside Advection Terms. In this window, increase the MXPART variable from 100000 to a number greater than that specified in the error message. You may alternately reduce the "Max particles per Cell" value from 30 to 20 or even 10 if this error continues to be a problem. * When first creating a model in Visual MODFLOW, you can only set uniform thickness. Once the model has been created, you can import/assign variable elevation data from the Input Menu, using the Grid option. Detailed information about the process of assigning/importing elevation data is described in the Input chapter of your Visual MODFLOW user's manual. * As usual, the best software depends on the goals of the model. Hydrus can be used to solve the Richards unsaturated flow equation in either a 1D column or in a 2D vertical slice. Its 2D finite element approach allows it to easily handle both vertical and lateral flow and transport in the unsaturated zone. However, Hydrus cannot easily be used to simulate distributed nitrate pollution across, for example, a watershed. Nor, can it be used to simulate the uptake and interaction of nitrate pollution with vegetation or the movement of the nitrate in the groundwater zone and/or discharge to and subsequent movement in surface water bodies. MIKE SHE, on the other hand, assumes the unsaturated flow is purely vertical. This assumption is reasonable in most applications, except perhaps in small scale models of a farm field with substantial lateral flows in the unsaturated zone. MIKE SHE's strength is in its ability to simulate the areal distribution of nitrate movement and its interaction with the other important processes of the hydrologic cycle, including vegetation and surface water bodies (e.g. eurtophication). If you require a more sophisticated soil-plant-agro model, MIKE SHE can be combined with DAISY (www.agsci.kvl.dk/planteer/daisy/main.htm) in a GIS interface. MIKE SHE is flexible framework for hydrologic modeling. It includes a full suite of pre- and post-processing tools, plus a flexible mix of advanced and simple solution techniques for each of the hydrologic processes. MIKE SHE covers the major processes in the hydrologic cycle and includes process models for evapotranspiration, overland flow, unsaturated flow, groundwater flow, and channel flow and their interactions. Each of these processes can be represented at different levels of spatial distribution and complexity, according to the goals of the modeling study, the availability of field data and the modeler's choices. The MIKE SHE user interface allows you to intuitively build your model based on your conceptual model of the watershed. The model data is specified in a variety of formats independent of the model domain and grid,

including native GIS formats. At run time, the spatial data is mapped onto the numerical grid, which makes it easy to change the scale and spatial discretization of your model. Finally, MIKE SHE also now includes sophisticated tools for auto-calibration and monte carlo analysis, including distributed computing across your office network. * A hydraulic potential loss of 800 m is obviously not realist for 10 km. I think it would be well suited for a power plant. Please be careful with the units. When you have the right value, you can assign heads on the boundaries of your model to get the gradient (the absolute values of heads are not important, only the gradient is considered). You can also define your problem with an influx of water upstream if you want. It is the same in your case. But don't forget to allow your water to flow out of your model ! * As I know, you can delete any layer separately. It is also not possible with GMS and PMWin. Only way out is export the dataset of different layers and then import to new model. I hope Visual Modflow technical team is closely watching this group and they will be able to provide more appropriate answer. * Unfortunately there is no "simple" way to completely remove a layer. However, I will describe the process for you. If the layer you want to remove is the bottom layer of your model, you can simply import a new surface elevation file. For example, if your current model bottom is at -50 meters, and you want to move it up to -30 meters, just create a text file with the following (typical) coordinates: 0 0 -30 2000 2000 -30 Where the first line has the X and Y coordinates at the model origins plus the new elevation, and the second line has the Y and Y coordinates at the model extents plus the new elevation. The process for importing a surface elevation file is described in your Visual MODFLOW user's manual. If the layer you want to remove is in the middle of your model, you would need to remove the layer division (in cross section, select Edit Grid-Edit Layers-Delete), then use the Move option to move all other gridlines up (to resolve the issue of "merging" 2 layers into 1), and finally import a surface elevation file for the model bottom as described previously. Please remember that you need to make sure any boundary condition/well screen/layer property data is correctly modified to reflect the changes in layer surfaces. * The analytical solution assumes an infinite domain while MODFLOW simulation is always finite. That said MODFLOW can be set up to represent a quasi infinite domain using appropriate boundary conditions and model grids. * There are a number of approaches you can take for modelling groundwater flow in fractured rocks. The choice depends on factors such as what you are trying to get out of the modelling and availability of data. 1. Equivalent porous medium (EPM) approach. In other words, ignore the fractures and treat the rock as a continuum with average "whole rock" hydraulic properties. If you haven't got much data on the fractures then this is probably the best approach and can be done using MODFLOW or FEMWATER within GMS. Drawbacks are that the EPM properties may be scale dependent (the larger the area you look at, the more large but rare fractures you encounter). Also, averaging may work for flow, but not for contaminant transport. 2. Discrete fracture representation. If you know where your fractures are then they can be simulated. If you have a regular, orthogonal distribution then you can use MODFLOW, if not

then FEMWATER. The permeability of the fractures can be related to the aperture width. May have some stability problems due to the heterogeneity of the grid. 3. Stochastic fracture modelling. If you have lots and lots of data on your fracture distribution then you can take a stochastic approach. You need to build up probability density functions (PDF's) for fracture aperture size, fracture spacing and fracture orientation. You then use the same PDF's to stochastically build your fracture grid. Run the model in a Monte Carlo fashion with many different fracture patterns. Each will provide a different result but will be probabilistically correct. Present your results also as PDF's. This will require specialist software, such as FRACMAN, or others. * There is no best baseflow separation technique. There are some which separate only the lower, slow part of the hydrograph, others which separate a good part of the peaks as baseflow. The HYSEP methods belong to the second category, leading to high BFIs. For the first one you could simply pick for every month of your time series the lowest discharge value. The time series built using these low-flow values is a slow baseflow line. If you want to play around, use the Digital Filter method (Nathan and McMahon 1990 WRR), which has a simple iterative formula with coefficients that you can change. Baseflow separation methods are not only subjective. They are simply restricted to the information contained in the discharge hydrograph. There are a lot of modelling results showing that you cannot determine a unique solution for the groundwater discharge based on the hydrograph alone. So there is a fundamental weakness here, no best method is possible. * The analytical solution assumes an infinite domain while MODFLOW simulation is always finite. That said MODFLOW can be set up to represent a quasi infinite domain using appropriate boundary conditions and model grids. * MODFLOW calculates the "average" hydraulic head in each cell at each time step. It then calculates the drawdown by subtracting the head in each cell from the starting head for that cell. Your hand calculation and the calculations performed by typical aquifer-test software will calculate the drawdown at the exact point of interest, not a finite-difference cell with some dimensions around that point. The drawdown in MODFLOW is thus an average value for the cell that contains the point at 500 meters from the well, and will usually be less than the value you would calculate for the exact point of interest. The difference between the hand calculation and the MODFLOW result will become less as the cell size becomes smaller, but in general the two methods will never agree with precisely the same calculated drawdown. * This is easy to answer as many users probably will. Modflow calculates AVERAGE drawdown (and head, and other values such as concentrations) for one whole cell. If "your" 500 m distance from the pumped well is somewhere in a cell of large area, the average for the cell will not be the same as for the point exactly 500m away. Advice: if you need better precision than average for an area, make the grid finer near the point of interest. Thus you will reduce the size of the cell and averaging will be closer to your expectations. Again, you are right in your calculation (1.24 m), but only if the whole aquifer satisfies Theis conditions (homogeneous, horizontal, that is one value of T and Sy for the whole model, infinite extent, etc.). * In Visual MODFLOW you can assign Zone Budget zones for areas of interest. So, you may assign a Zone Budget zone to your drain and another zone or zones to the areas your water is coming from and discharges to the drain. When you run the model be sure to check the Zone Budget option in the Engines to Run pop-up window. In the Output module you may than go to Maps/Zone Budget and select Zone Budget from the left menu. Here you will have listed all the

flows (inflows and outflows) for the selected zone in a given stress period, time step or model time. * The .FLO file (which contains Cell by Cell flow terms) is not a readable file. To create a readable output file, go to your Run Menu, and click on Modflow-Output Control. In the Output Control window, check off all boxes under Print to LST. The LST file is an ASCII file that can be read with Notepad. Run your model and view the LST file. NOTE: This will ONLY produce flow terms for Boundary Condition cells. If you are interested in flow terms for "regular cells" you can assign ZoneBudget Zones to those cells of interest. * Your lowest model layer has its base set as a no flow (impervious) boundary by default. If you want for, whatever reason, introduce another impermeable boundary between layers inside the model you can set VCONT in the upper layer to a very low value (say 1e-8), which will effectively separate your model into two independent sub-models. However, I do not think it would be a good idea and in this case it would be probably better to build two independent models. * There is no need to explicitly assign a bottom-most layer as totally impervious. By default, all boundaries on the exterior of the MODFLOW model grid are simulated as no-flow. * You can not directly insert a new layer, you have to divide any one layer into two layer, and re-assign the elevation. * MODFLOW is a saturated zone model, so you cannot directly handle rainfall. An option is to use Visual HELP to estimate recharge to the groundwater from rainfall, and import the data into Visual MODFLOW as recharge boundary condition. Another option is to use MODFLOW-SURFACT. * Computation of T by tide-aquifer interaction involves two approaches: (a)direct use of 'Time Lag' model and/or 'Tidal Efficiency' model; and (b)inverse modeling. The first approach is very simple, and computation can be done using a scientific calculator. However, the second approach requires the use of a nonlinear optimization technique (e.g., Gauss-Newton or Levenberg-Marquardt). * The Recharge boundary condition could be used to simulate average rainfall (for example, monthly averages) being applied to the top layer of your model. It is important to note that depending on your model design (i.e. discretization of layers), if you have one of more layers above the water table, they will likely become unsaturated, or oscillate between unsaturated and saturated if recharge is being applied. Andras Jakab's suggestion of using MODFLOW SURFACT in this case is a good one, as SURFACT is designed to handle both unsaturated and saturated flow modeling. However, if your layers are designed so that the bottom of layer 1 is below the water table, this will not be a concern. * The simulation of a wetland could be done using either the Constant Head boundary condition (which can act as an infinite source/sink for water, to maintain the head value assigned) or the River boundary condition (which will add/remove water from surrounding cells through the river bottom, depending on the head value of the surrounding cells). * Visual MODFLOW software can be used to determine optimal pumping rates, volumes of water extracted, and drawdown cones of influence. 1) The Well Optimization component of Visual MODFLOW is designed to answer questions such as: -what is the minimal pumping rate required to maintain an assigned head level? -what is the minimal volume of water that must be removed to contain a contaminant plume? -what is the

maximum mass removal rate attainable? -what is the maximum pumping rate that can be used while maintaining a specified drawdown level? etc. 2) The Zone Budget component of Visual MODFLOW can be used to calculate the cumulative volume of water being extracted through a pumping well (or other boundary conditions) over time. 3) Visual MODFLOW provides drawdown contours as one of the output maps available. You can customize the contour interval, add specific contour values, and colour shade your contours. * Unfortunately there is no automatic way to find the number of active or inactive cells in your Visual MODFLOW model. However, you could take the active/inactive array from the .VMG file and put it into excel, then use the COUNTIF function to count the number of 1's (active cells) or the number of 0's (inactive cells). * Dewater Model You can download the model from the following zip file: http://www.pmwin.net/models/dewater/dewater.zip Here are what I have done: 1. Since PMWIN uses consistent time/length units, all parameters are converted to using meters and days (for example recharge = 0.000219 m/d) and then entered to the model. 2. In order to estimate the zone of influence, the extent of the model grid is constructed quite large (north-south width = 2275m; west-east width = 3080 m). I noticed later that this size is not large enough (see below). 3. The northern and southern boundary of the model are constant head boundaries. The southern boundary represents the sea and is set at h = 0 m. The northern boundary is set at 9.1 m so that the hydraulic gradient between these two boundaries is 1/250. 4. The groundwater recharge is entered, and a steady state flow simulation has been carried out. The trench is not yet modeled at this point. 5. The calculated steady-state head values are imported as initial heads (so that we can calculate the "real" drawdown cone). I got a problem here: When considering gradient of 1/250 and the distance of the mine pit from the sea (500 m), the groundwater level at the mine pit should stand at 2.0 m MSL (not 1 m MSL). When groundwater recharge is considered, the calculated groundwater level stands at 2.5 to 2.8 m MSL. Therefore, there is a disagreement between the given water level and gradient. However, I ignored this problem and continue my exercise. 6. The trench is modeled by using constant head condition with h = -1 m. 7. Finally a steady state simulation is carried out again. 8. I put the screen shot of the drawdown cone (zone of influence) here: http://www.pmwin.net/models/dewater/drawdown.jpg The image shows that the drawdown reaches the boundaries of the model. This means that either the extent of the model is still not big enough and/or the problem with the data disagreement (see step 5) needs to be solved first.

9. The amount of water that needs to be pumped out from the trench can be easily obtained by carrying out a subregional water budget calculation for the trench. The remaining question is whether the capacity of the aquifer/trench system is able to support the required pumping rate. Ignoring the capacity problem, question #2 can be solved by carrying out a transient flow simulation (with desired simulation length and number of time steps) at step 7 instead of steady-state simulation. We