A semi-analytical model for predicting water quality from an aquifer storage and recovery system

Abstract

Aquifer storage and recovery (ASR) involves the injection of freshwater in an aquifer through wells for the purpose of creating a subsurface water supply that is recovered at a later time, often using the same wells, to meet seasonal, long-term, emergency, or other demands. In this paper a numerically efficient semi-analytical model is developed for predicting the quality of water recovered by an ASR system given data on the qualities of ambient and injected waters, hydraulic properties of the aquifer, ambient hydraulic gradient, and system operations. It is assumed the ASR well is installed in a stratified aquifer such that the semi-analytical ASR model (SASRM) simulates the fate of water injected under steady-state conditions into each stratum. It is also assumed that a sharp and mobile interface separates injected water from ambient groundwater such that in situ mixing of water within and between strata does not occur. SASRM assigns particles to define the location the interface in all strata and then follows the migration of these particles under ambient and induced flow conditions. During water recovery, the transient location of the interface is simulated in each stratum and this information is used to quantify the fractions of ambient and injected water extracted at the well-head and the quality of water recovered. To mimic the effects of dispersion, a Latin Hypercube sampling strategy is used to assign hydraulic conductivities according to a predefined probability distribution to the layers of a conceptually stratified aquifer. The volumetric fraction of water received or delivered from any given lithologic unit is assumed proportional to the transmissivity of the stratum normalized to the total aquifer transmissivity interrogated by the ASR well. SARSM is numerically verified against MT3DMS and then calibrated and validated using field data from an ASR system located in Boynton Beach, FL. The field demonstration shows SASRM is capable of predicting effects of apparent dispersion and rapidly executing 100-layer ASR simulations that are generally difficult to perform with typical numerical transport models like MT3DMS.