Using sensitivity analysis in model calibration efforts

By:  and 



In models of natural and engineered systems, sensitivity analysis can be used to assess relations among system state observations, model parameters, and model predictions. The model itself links these three entities, and model sensitivities can be used to quantify the links. Sensitivities are defined as the derivatives of simulated quantities (such as simulated equivalents of observations, or model predictions) with respect to model parameters. We present four measures calculated from model sensitivities that quantify the observation-parameter-prediction links and that are especially useful during the calibration and prediction phases of modeling. These four measures are composite scaled sensitivities (CSS), prediction scaled sensitivities (PSS), the value of improved information (VOII) statistic, and the observation prediction (OPR) statistic. These measures can be used to help guide initial calibration of models, collection of field data beneficial to model predictions, and recalibration of models updated with new field information. Once model sensitivities have been calculated, each of the four measures requires minimal computational effort.

We apply the four measures to a three-layer MODFLOW-2000 (Harbaugh et al., 2000; Hill et al., 2000) model of the Death Valley regional ground-water flow system (DVRFS), located in southern Nevada and California. D’Agnese et al. (1997, 1999) developed and calibrated the model using nonlinear regression methods. Figure 1 shows some of the observations, parameters, and predictions for the DVRFS model. Observed quantities include hydraulic heads and spring flows. The 23 defined model parameters include hydraulic conductivities, vertical anisotropies, recharge rates, evapotranspiration rates, and pumpage. Predictions of interest for this regional-scale model are advective transport paths from potential contamination sites underlying the Nevada Test Site and Yucca Mountain.

Publication type Book chapter
Publication Subtype Book Chapter
Title Using sensitivity analysis in model calibration efforts
Year Published 2003
Language English
Publisher NRC
Contributing office(s) Office of Ground Water
Description 4 p.
Larger Work Type Book
Larger Work Subtype Conference publication
Larger Work Title Proceedings of the International Workshop on Uncertainty, Sensitivity, and Parameter Estimation for Multimedia Environmental Modeling
First page 53
Last page 56
Conference Title International Workshop on Uncertainty, Sensitivity, and Parameter Estimation for Multimedia Environmental Modeling
Conference Location U.S. Nuclear Regulatory Commission Headquarters 11545 Rockville Pike, Rockville, Maryland, USA
Conference Date August 19–21, 2003
Country United States
State California, Nevada
Google Analytic Metrics Metrics page
Additional publication details