Using Calibrated Models to Guide Collection of Field Data for Improved Predictions

Calibrated numerical models are powerful tools for guiding collection of new field data to improve model predictions. Two convenient measures that facilitate using models in this way are the “observation-prediction” (OPR) and “parameter-prediction” (PPR) statistics, both available in U.S. Geological Survey public-domain software (Tonkin et al., 2007, USGS Report TM –6E2, http://water.usgs.gov/software/OPR-PPR). The OPR statistic evaluates the relative importance to model predictions of existing or potential system-state observations, and can be used to guide monitoring network design. This statistic calculates the percent change in prediction uncertainty that is produced when observations are omitted from or added to an existing observation data set. The PPR statistic evaluates the relative importance to model predictions of potential new system information related to model parameters, and can be used to guide collection of new hydrogeologic field data. This statistic calculates the percent decrease in prediction uncertainty that results when potential information about parameters is added. The OPR and PPR statistics are linear local-sensitivity measures and require minimal computational effort. Initial testing suggests that they can perform well for nonlinear models despite the linearity assumption.

 The statistics are applied to a model of groundwater flow through a contaminated fractured sedimentary rock aquifer, to help identify observations and parameters most important to predictions of flux through a rock volume in which bioaugmentation has been implemented. In this system, dominant flow paths are along bedding plane partings, yet the most important parameters identified by the OPR and PPR statistics include a vertical hydraulic conductivity parameter controlling flow across beds, and important observations include hydraulic heads at locations where simulated heads are sensitive to this parameter. These results point to the importance of characterizing hydraulically active cross-bed fractures in sedimentary rock aquifers, even when the primary flow paths are in bed-parallel directions.