GMS:Automated Parameter Estimation: Difference between revisions

From XMS Wiki
Jump to navigationJump to search
No edit summary
(No difference)

Revision as of 16:34, 21 September 2009

MODFLOW
Pre-processing
MODFLOW Commands
Building a MODFLOW Model
Map to MODFLOW
Calibration
Packages Supported in GMS
Saving a MODFLOW Simulation
Importing MODFLOW Files
Unsupported MODFLOW Features
Run MODFLOW
Post-processing
MODFLOW Display Options
MODFLOW Post-Processing Viewing Options
Reading a MODFLOW Simulation
Tutorials
Packages
Flow: BCF6, HUF, LPF, UPW
Solvers:

DE4, GMG, NWT, PCG,

PCGN, LMG, SIP, SOR,

SMS
Other:

BAS6, BFH, CHD1, CLN,

DRN1, DRT1, EVT1, ETS1,

GAGE, GHB1, GNC, HFB1,

HUF, LAK3, MNW1, MNW2,

OUT1, RCH1, RIV1, SFR2,

STR1, SUB1, SWI2, WEL1,

UZF1

Calibration,
Parameters,
Stochastic Modeling
Calibration
Model Calibration
Automated Parameter Estimation
PEST Dialog
PEST
Run Options
Observations
MODFLOW-USG Observations
Plot Wizard
Calibration Targets
Parameters
Parameters
Parameter Dialog
Pilot Points
Multiplier Arrays for Parameters
Standard MODFLOW Parameters
Stochastic Modeling
Stochastic Modeling
Gaussian Field Generator
Risk Analysis Wizard
T-PROGS

One of the tools provided in GMS for model calibration is automated parameter estimation. With automated parameter estimation, an external utility, sometimes called an "inverse model", is used to iteratively adjust a set of parameters and repeatedly launch the model until the computed output matches field-observed values. Parameter estimation is used in conjunction with the point observations and the flow observations.

Automated parameter estimation is supported in GMS for the MODFLOW simulations using PEST a general purpose parameter estimation utility developed by John Doherty of Watermark Computing.

Inverse models should only be used carefully and with a full understanding of the assumptions, equations, and methods involved. It is suggested that the user read the available documentation on the inverse model being used. Only the steps involved in setting up an inverse model are described in this document.

Basic Steps

The basic steps involved in using an inverse model for parameter estimation are follows:

1. Create a Working MODFLOW Model

The first step is to create your MODFLOW model and run a simulation. Before launching the inverse model, you need to have a MODFLOW model that successfully converges and you need to determine a good set of starting values for your parameters. Once you have a solution it is also a good idea to copy the computed heads from your solution to your starting heads array. This ensures that as the inverse model modifies the parameters and runs MODFLOW repeatedly, it is more likely that MODFLOW will quickly converge each time it is launched.

2. Enter the Observations

Once you have a working MODFLOW model, you should enter your head and flux observations. Head observations are entered as points using an observation coverage in the Map module. Flow observations are assigned directly to arcs and polygons in source/sink coverages. Each of the observations is assigned a weight that is saved to the inverse input files.

3. Turn on the Inverse Model

You must select an inverse model. Bring up the Global Options dialog and select either the Parameter Estimation or Stochastic Inverse Model button depending on whether a stochastic simulation is being run.

4. Parameterize the model

The next step is to parameterize your model. See the Parameters page for more details.

5. Create a Parameter List

The next step is to create the parameter list. See the Parameters page for details.

6. Set parameter estimation options

Once the parameter list is set up, you may wish to edit the general Parameter Estimation options. These options include the output control and convergence criteria.

7. Edit the Group Weight Multipliers

The group weight multipliers can be edited to adjust the relative weight of the head and flux observations.

8. Edit the Convergence Options

Edit the MODFLOW Convergence Options if necessary in order to ensure a stable solution.

9. Save and Run MODFLOW Model

Once all of the inverse model options have been set, the next step is to save the MODFLOW model using the Save/Save As command in the File menu. Next, run MODFLOW and the inverse model will run with MODFLOW. The inverse model will then be launched in a separate window or the model wrapper in which you will see information relating to the MODFLOW runs and the status of the objective function. Depending on the problem, the inverse model may take anywhere from several minutes to several hours (or days) to run to completion. When the inverse process is completed successfully, GMS automatically launches a MODFLOW forward run with the optimal values computed by the inverse model. Thus, the solution will reflect the optimal values computed by the inverse model.

10. Viewing the Optimal values

When the inverse model is finished, it writes out a text file containing the set of parameter values corresponding to the minimum calibration error. These values can be viewed with the Import Optimal Values button. This copies the optimal parameter values to the Starting Value field in the Parameter List.

Sensitivity Analysis

At each PEST iteration, PEST computes the sensitivities of each of the parameters. This information is available in the model.sen file (where model is the name of the MODFLOW model). PEST records the composite sensitivity and the relative composite sensitivity of each parameter in this file. This information is useful in determining which parameters have the greatest effect on the model as well as which parameters have the least effect on the model. Thus, the "insensitive" parameters can be removed or held constant in a subsequent PEST run.

For a more detailed description of parameter sensitivity see section 5.3.2 of the PEST manual.

Parameter Estimation Dialog

Options affecting parameter estimation can be changed via the Parameter Estimation Dialog.

See also

Model Calibration