38 Geoscience and Hydrogeology in the AI Era: From Predictive Models to Real-Time Applications

This session offers a unique platform to showcase advancements in Artificial Intelligence (AI), Machine Learning (ML), and physics-informed methodologies, focusing on topics in hydrogeology, subsurface flow and transport, real-time geophysical inversion, critical minerals, and energy applications.

Sunday, 19 October 2025
1:30 PM - 5:30 PM
HBGCC, 210AB

Prediction of Karst Aquifer Recharge Through a Hybrid Explainable Artificial Intelligence and Hydrological Modeling Approach

Paper number: 38-12
Program: Technical Sessions
Time: 4:35 PM – 4:50 PM

Authors

Abstract

Reliable prediction and long-term projection of aquifer recharge are essential for evaluating the sustainability and climate resilience of groundwater resources. However, recharge estimates are inherently uncertain due to the absence of direct measurements. To address this challenge, we developed a serial hybrid modeling framework in which historical recharge estimate from the U.S. Geological Survey's hydrological model were used to train and evaluate the prediction performance of Artificial Intelligence (AI) models.  The framework was applied to two hydrologically distinct basins in the Edwards Aquifer Region: the Bexar basin and the Nueces basin. Historical climate data were derived from local meteorological stations and remote-sensing products, while statistically downscaled climate data from global climate models with intermediate- and high-emission scenarios informed recharge projections through 2100.  

Among the AI models tested, the Extremely Randomized Trees (ERT) achieved the best prediction accuracy on the test data for both basins.  Notably, ERT and other AI models consistently predicted nonzero recharge during three instances in the test dataset for the Bexar basin when the hydrological model estimated zero recharge. These AI model predictions were corroborated by multiple independent indicators, including precipitation records, groundwater levels from an index well, GRACE-derived groundwater storage anomalies, and recharge estimates from the Hydrological Simulation Program—FORTRAN model, providing compelling evidence of the AI model’s ability to learn intricate hydroclimatic relationships and predict a critical and challenging aquifer recharge.

To enhance interpretability, we coupled the ERT model with the Shapley Additive eXplanation (SHAP) to form a comprehensive XAI model.  The XAI model revealed that previous-month recharge and current-month precipitation emerged as the most influential predictors of aquifer recharge in the Bexar and Nueces basins, respectively, accounting for 32% of the variability in model predictions. XAI-based aquifer recharge projections through 2100 suggest a potential reduction in large recharge events under future climate scenarios. This study underscores the value of XAI as a robust and interpretable tool for assessing long-term aquifer sustainability in the face of a changing climate.

Geological Society of America Abstracts with Programs. Vol. 57, No. 6

doi: 10.1130/abs/2025AM-414511

© Copyright 2025 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.