DESIGNING SAMPLING STRATEGIES TO SUPPORT GROUNDWATER MODELING IN THE UPPER KLAMATH BASIN, OREGON AND CALIFORNIA

Demand for groundwater in the upper Klamath Basin has increased in recent years due to drought and changes in surface-water management aimed at improving habitat for fish listed under the Endangered Species Act. Water and environmental managers in the basin need tools to assess the impacts of groundwater pumping on the basin’s already over-allocated water resources. To address this need, a transient groundwater flow model that simulates regional-scale groundwater/surface-water interactions has been developed using MODFLOW. The model was calibrated using a time-series data set that includes approximately 5,600 water-level measurements from 662 wells, and more than 300 groundwater discharge measurements for 10 major discharge areas. The calibrated model is able to reproduce the observed variations in water level and groundwater discharge at a range of time scales throughout most of the upper Klamath Basin. When coupled with uncertainty analysis, the calibrated model can be used to evaluate data worth and guide data collection. Model-prediction uncertainty analysis serves to measure the contribution of data, individually and collectively, to improving model predictions; data that provide a greater reduction in model uncertainty are assumed to have greater value for model calibration.

The work presented here applies the uncertainty-analysis method to evaluate data worth for the upper Klamath Basin groundwater model. The method is applied in three stages. In the first stage, the data worth analysis was applied to the existing calibration data set to identify the data that were most informative for modeling. The analysis assessed the contribution of each water-level and discharge measurement to reducing uncertainty associated with drawdown and discharge predictions. The results show that both data types provide substantial information for reducing model uncertainty. The analysis identified those data that are informative along with those data that contribute very little to uncertainty reduction. In the second stage, the data worth analysis was used to assess the prospective worth of potential new data and identify future measurements that would best supplement the existing data set. The results indicate that model uncertainty can be further reduced with the addition of new water-level and discharge measurements. The third stage couples the data worth analysis with a genetic algorithm to evaluate the trade-offs between varying costs of different types of data and the contribution of those data to improving model reliability. The methodology can identify the mix of data (water-level measurements at existing wells, water-level measurements at new wells, and measurements of groundwater discharge to streams) that will minimize model prediction uncertainty for a given data collection budget.