RELIABILITY IN ESTIMATING URBAN GROUNDWATER RECHARGE THROUGH THE VADOSE ZONE

Abstract

Reliance on vadose zone models to estimate groundwater recharge in arid and semiarid regions of the United States is increasing due to limited groundwater supplies and continued urbanization. The success of vadose zone models in providing reliable estimates of urban recharge and other fluxes depends on the information content used to constrain the calibration process. In this study, a numerical experiment of artificial recharge through a three-layered vadose zone system revealed several findings related to coupled model calibration. First, the extension of vadose zone model calibration to three dependent variables added information content that enhanced parameter sensitivities. Second, predictive analysis using the calibration-constrained Monte Carlo approach was time-prohibitive because of the tendency toward local minima when using the gradient algorithm. Third, despite a perfect match to historical data, the affect of alternative starting calibration parameters sets on condition number illustrated the limitations of information quality on model uncertainty. Fourth, perfect observed / simulated profiles correlation coefficients for all calibration parameter sets, were, by themselves, poor indicators of model success. Fifth, the range of predictive recharge uncertainty, and uncertainty in energy and solute mass entering the groundwater system, estimated using the likelihood-type approach, is attributed to parameter non-uniqueness due to limited calibration information. For a calibration of a field system, the estimated range of predictive uncertainty would be larger because of additional uncertainty from errors in measurements and/or the conceptual model.