Estimating Uncertainty of Predicted Seven Rivers Augmentation-Field Drawdown

The Roswell Artesian Basin Groundwater Model grid was refined in the area of the Seven Rivers Augmentation Well Field and the model recalibrated using results from the Seven Rivers 73-day pump test. Model calibration was performed using PEST, with singular value decomposition (SVD) as the method of regularization. Simulated levels of drawdown from the calibrated model provide a very good match to measured. The level of uncertainty associated with model predictions was not specifically evaluated as part of the recalibration effort, but an evaluation of uncertainty may prove helpful in interpretation of future data and management of the augmentation system. As a preliminary investigation, this work formulates a constrained optimization problem to explore the postcalibration predictive uncertainty associated with simulated drawdowns. Using drawdown in wells as the prediction of concern, maintaining fit similar to the calibrated model’s and constraining the range of transmissivity, maximum levels of simulated drawdown were determined. Maximum drawdown levels per well provide insight to variability of uncertainty. Constraining the transmissivity limits the SVD tendency to sometimes “overcalibrate” and produce unreasonable results. Results provided insight to practical issues associated with application of prediction-uncertainty methods to the model and the amount of uncertainty associated with the calibrated model’s predictions of drawdown.