GeoFUSE: An Efficient Surrogate Model for Seawater Intrusion Prediction and Uncertainty Reduction

Abstract

Seawater intrusion into coastal aquifers poses a significant threat to groundwater resources, particularly with rising sea levels due to climate change. Accurate modeling and robust uncertainty quantification of seawater intrusion are crucial for effective groundwater management. However, traditional numerical methods are computationally expensive, often requiring days to weeks for ensemble-based uncertainty quantification. To address these challenges, we develop GeoFUSE, a framework that integrates the U-net Fourier neural operator for high-fidelity surrogate modeling, Principal Component Analysis for geological model dimension reduction, and the Ensemble Smoother with Multiple Data Assimilation for uncertainty quantification. GeoFUSE enables fast and efficient seawater intrusion simulations while significantly reducing uncertainty in flow predictions. We apply GeoFUSE to a 2D cross-section of the Beaver Creek tidal stream–floodplain system in Washington State, demonstrating its capabilities and adaptability to hydrogeological systems. The framework achieves a speedup of 360,000 times compared to high-fidelity numerical simulations, reducing simulation times from hours to seconds while maintaining predictive accuracy. By integrating measurement data from monitoring wells, GeoFUSE further reduces geological uncertainty and improves the predictive accuracy of salinity distribution over a 20-year period. Our results demonstrate that GeoFUSE significantly enhances computational efficiency, providing a robust tool for real-time uncertainty quantification in groundwater management. Future work will extend GeoFUSE to 3D models and incorporate additional factors such as sea-level rise and extreme weather events, broadening its applicability to a wider range of coastal and subsurface flow systems.