A non-Bayesian nonparametric model for characterization of basin-scale aquifers using groundwater level fluctuations

Abstract In this study, a non-Bayesian nonparametric model for characterization of the hydraulic properties of a basin-scale aquifer is proposed based on the natural stimulus of recharge from precipitation via a geostatistical principal component adaptation evolution strategy. We modified an existing model to use a low-dimensional projection onto the principal component space as well as improve computational efficiency and memory requirements. To illustrate the scalability of the proposed model, we estimated about 10 million unknown hydraulic conductivities (HCs), which required only 35 min or less on a modern workstation. To test the estimation accuracy, the proposed model was employed to estimate several aquifer HC fields using timeseries data of groundwater levels of more than 60 days at 16 monitoring wells with the recharge rate provided. Two cases were considered—when HC measurements at the monitoring wells were available and unavailable. In addition, inversions were performed using perfectly known and erroneous recharge rates. Using various test configurations, we found that the estimation was better when HC measurements at the monitoring wells were available. For inaccurately determined recharge rates, the estimation accuracy generally deteriorated, but the structural features of the estimated HC fields were similar to those obtained using the correctly assigned recharge rate. Regardless of the availability of HC measurements at the monitoring wells, the proposed model estimated the distribution of HC reasonably well, even when incorrect prior statistics of the mean and correlation scales were provided. By demonstrating that an aquifer can be characterized at the basin-scale using recharge from precipitation, the relevance of the proposed model, which had high computational efficiency, was evident.