Date:

Monday, April 15, 2002

Speaker:

Dr. Mary Hill

Affiliation:

USGS, Boulder, CO.

Contact Information

mchill@usgs.gov

Title:

Guidelines for Effective Model Calibration (Any Model!)

Abstract:

Models are used extensively to evaluate many kinds of systems and to predict their response in a variety of situations. Because many aspects of many systems are unknown and cannot be directly measured given present technology, many models need to be calibrated. In many fields, such as ground-water modeling, calibration has historically been achieved by trial and error alone, but these methods provide less insight than can be achieved. This talk focuses on how nonlinear regression and associated statistics can be used to dramatically improve how data are used to calibrate and test models. For example, parameters which cannot be estimated accurately and uniquely with the available data and the likely utility of potential new data can be clearly and quickly identified. Parameter values that produce the best fit between simulated and observed values can be determined using nonlinear regression. Measures of prediction uncertainty are a natural consequence of regression methods. Examples are drawn from ground-water modeling, including modeling of the Death Valley regional ground-water system, which underlies the Nevada Test Site and the proposed US high-level nuclear waste repository at Yucca Mountain.