HR: 0800h
AN: H11C-0310 INVITED [Abstracts]
TI: Testing Data Collection Strategies for Improving Ground-Water Model Predictions
AU: * Tiedeman, C R
EM: tiedeman@usgs.gov
AF: U.S. Geological Survey, 345 Middlefield Rd. MS496, Menlo Park, CA 94025
United States
AB:
Calibrated numerical models are powerful tools for guiding collection of data to improve model predictions. In previous work,
"value of improved information" ({\it voii}) and "observation-prediction" ({\it opr}) statistics were developed for
identifying, respectively, the model parameters and potential system-state observations that are most important to the
predictions of interest. These methods were then applied to a complex regional ground-water flow model to identify potential
hydrogeologic and system-state data beneficial to the predictions. However, it is difficult to test the validity of the
results for a field application because the true predictions are not known. Thus, to test whether data collection strategies
identified by the statistics can actually improve the predictions, the {\it voii} and {\it opr} methods are applied to a
synthetic ground-water flow problem with advective-transport predictions. First, observations generated from the true
synthetic model are used to calibrate an incorrect model. Second, the {\it voii} and {\it opr} statistics are applied to the
incorrect model, to identify the parameters and potential observations that are most important to the predictions. Third,
several updated incorrect models are constructed and calibrated by improving the value or model feature associated with one
or more parameters, or by adding one or more new observations to the calibration data set. The consequent increase in
prediction accuracy is then assessed. Finally, the parameter improvements and additional observations that produce the
greatest increases in prediction accuracy are compared to the parameters and observations that rank as most important by the
{\it voii} and {\it opr} statistics. Preliminary results of testing the {\it voii} method show that for a majority of the
model predictions, improving the values of the parameters identified as most important by the {\it voii} statistic actually
causes the greatest increases in prediction accuracy.
DE: 6309 Decision making under uncertainty
DE: 3210 Modeling
DE: 3260 Inverse theory
DE: 1829 Groundwater hydrology
SC: Hydrology [H]
MN: 2004 AGU Fall Meeting