28 Geoscience and Hydrogeology in the AI Era: From Predictive Models to Real-Time Applications (Posters)

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
8:00 AM - 5:30 PM
HBGCC, Hall 1

Forecasting Karst Spring Discharge with a Hierarchical Spatiotemporal Graph Attention Network: A Case Study of Barton Springs

Paper number: 28-0
Program: Technical Sessions
Time: 8:00 AM – 8:00 AM

Authors

  • Renjie Zhou Presenting Author
    Sam Houston State University, Huntsville, Texas

Abstract

Karst aquifers are critical freshwater sources, supplying nearly one-quarter of the global population. However, their intricate hydrogeological architecture characterized by dual-flow regimes and high heterogeneity poses significant challenges for discharge forecasting and sustainable management. Physics-based hydrological models often struggle to capture the complex spatial interactions and multi-scale temporal behaviors inherent in these systems, particularly the interplay between rapid conduit flow and diffuse matrix flow. In this study, a novel hierarchical spatiotemporal forecasting framework that combines a multi-scale dynamic Graph Attention Network (GAT) with a Long Short-Term Memory (LSTM) architecture is specifically designed and developed for karst spring discharge prediction. Key innovations of the model include: (1) a dynamic graph neural network that updates spatial dependencies using both geographic and data-driven hydrological relationships; (2) a multi-head attention mechanism to capture diverse spatial patterns; and (3) a hierarchical temporal module that integrates monthly and seasonal discharge dynamics through an adaptive fusion strategy. These components enable the model to effectively represent the dual-flow behaviors in karst aquifers. The model’s performance is demonstrated on Barton Springs in the Edwards Aquifer (Texas, USA), showing that it consistently outperforms traditional spatial and temporal deep learning baselines (including GAT, GCN, LSTM, and GRU models) across multiple time horizons. Further, the ablation study and permutation feature importance analysis are conducted to evaluate the critical roles of key input variables and spatial nodes. This research highlights the value of spatiotemporal deep learning approaches for complex groundwater systems and demonstrates the importance of integrated monitoring strategies to support accurate, data-driven water resource forecasting in karst terrains.

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

doi: 10.1130/abs/2025AM-411987

© 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.