INTEGRATING AIRBORNE GEOPHYSICAL SURVEYS WITH HYBRID NUMERICAL AND MACHINE LEARNING SURROGATE MODELING TO OPTIMIZE MANAGED AQUIFER RECHARGE IMPLEMENTATION (Invited Presentation)

Population growth, economic development and climate change have accelerated groundwater depletion throughout the 21st century. To combat this overdraft, water managers across the globe have rapidly expanded managed aquifer recharge (MAR) programs, through which excess surface water is used to replenish depleted aquifers. In the Central Valley of California, for example, some water districts hope to increase MAR capacity by a factor of 10 over the next 15 years. However, identifying suitable recharge sites remains challenging because sediment heterogeneity can strongly impact recharge efficiency, but it is difficult to assess. Recent airborne electromagnetic (AEM) surveys of the Central Valley fill this gap by providing an image of sediment texture within the underlying alluvial aquifer system. However, directly quantifying MAR site suitability from the AEM data is not straightforward, as textural contrasts within the vadose zone can create complex flow patterns that are difficult to predict without computationally expensive numerical modeling.

To address this challenge, we implement a hybrid modeling approach that uses a process-based hydrologic code (ParFlow) and machine learning surrogates to quantify recharge efficiency at prospective sites within the AEM survey. First, we generate stochastic realizations of sediment lithology across the Central Valley using sequential Gaussian simulation. Next, we simulate recharge at 1,000 potential sites using ParFlow. We then train a computationally efficient machine learning surrogate (a vision transformer) to reproduce these simulations and use the surrogate to quantify recharge efficiency across the remainder of the Central Valley. Results reveal that recharge potential is the highest at sites that overlie interconnected blocks of coarse-grained sediment, exhibit high water content through the vadose zone and have a shallow water table. These sites predominantly occur along fluvial fans on the eastern edge of the Central Valley, though small zones of high recharge potential can be found elsewhere across the Valley, each with different constraints on utilization.