ANN-Based Predictors of ASR Well Recovery Effectiveness in Unconfined Aquifers

Abstract

In this study, we present artificial neural networks (ANNs) to aid in a reconnaissance evaluation of an aquifer storage and recovery (ASR) well. Recovery effectiveness (REN) is the proportion of ASR-injected water recovered during subsequent extraction from the same well. ANN-based predictors allow rapid REN prediction without requiring preparation for and execution of solute transport simulations. REN helps estimate blended water quality resulting from a conservative solute in an aquifer, extraction for environmental protection, and other uses, respectively. Assume that into an isotropic homogenous portion of an unconfined, one-layer aquifer, extra surface water is injected at a steady rate during two wet months (61 days) through a fully penetrating ASR well. And then, water is extracted from the well at the same steady rate during three dry months (91-day period of high demand). The presented dimensionless input parameters were designed to be calibrated within the ANNs to match REN values. The values result from groundwater flow and solute transport simulations for ranges of impact factors of unconfined aquifers. The ANNs calibrated the weighting coefficients associated with the input parameters to predict the achievable REN of an ASR well. The ASR steadily injects extra surface water during periods of water availability and, subsequently, steadily extracts groundwater for use. The total extraction volume equaled the total injection volume at the end of extraction day 61. Subsequently, continuing extraction presumes a pre-existing groundwater right.

Keywords

• aquifer storage and recovery
• unconfined aquifer
• MODFLOW
• MT3DMS
• artificial neural network