Model Sensitivity Analysis, Data Assessment, Calibration, and Uncertainty Evaluation
Utrecht University, 1 - 3 June 2005
 |
Model Sensitivity Analysis, Data Assessment, Calibration, and Uncertainty Evaluation Utrecht University, 1 - 3 June 2005 |
|
|
Scope of the Short Course
Course Description
Building models of complex natural systems that are sufficiently accurate to be useful generally requires that the models be calibrated. Yet model calibration can be difficult. Groundwater model calibration, for example, is exceptionally difficult because of system inaccessibility, importance of the details of simulated results to groundwater users, and mathematical difficulties of the nonlinear calibration problem.
This course teaches
- nonlinear regression and associated methods
- fourteen guidelines that describe how the methods can be used to
dramatically improve how data are used to develop models
|
Students will learn how to
- build parsimonious models that do not over-fit the data
- identify parameters that can be estimated accurately and uniquely with the available data
- assess the likely utility of potential new data
|
Students will also learn how to use nonlinear regression to estimate parameter values that produce the best fit between simulated and observed hydraulic heads, flows, concentrations, and so on, and to use associated methods to evaluate model fit and prediction uncertainty. The cause of and potential problems related to the nonlinearity of groundwater calibration are emphasized in class.
The methods and guidelines are taught using hands-on computer exercises based on a groundwater model that is simple enough to clearly understand the performance of the methods taught. The exercises start with sensitivity analysis and parameter estimation for a steady-state model. The calibrated model is used to evaluate possible contamination of a pumping well by a proposed landfill and to evaluate the likely utility of proposed additional data that could be collected during a planned pumping test. Although the exercises used in class focus on a groundwater system, the methods and guidelines are applicable to a broad range of problems.
Course Software
The exercises use the well-documented, USGS, public-domain computer programs MODFLOW-2000, UCODE_2005, MFI2K, and GW_CHART. UCODE_2005 can take advantage of the MODFLOW-2000 parameter definition, observation, and sensitivity capabilities, or can be used with any other application model(s). MFI2K and GW_CHART provide limited graphical interface capabilities. Commercially available graphical interfaces are discussed.
Course Handouts
Participants will receive course notes and associated USGS reports, including the MODFLOW-2000 documentation and the UCODE_2005 manual. Electronic versions of the manuals and the computer codes can be obtained for free from:
http://water.usgs.gov/software/ground_water.html.