Artificial Intelligence Modeling for Groundwater Environments across Spatial Scales.

Abstract

Groundwater is threatened by climate change and human activities, with depletion and contamination emerging as critical risks, necessitating the development of models to estimate its response to changes. Artificial intelligence (AI) is gaining increasing traction in groundwater applications on multiple scales. This review evaluates the potential of AI to model groundwater systems including flow and transport problems across various scales. AI has been leveraged to identify contamination sources and optimize remediation strategies at site scales. The substantial promise of AI in predicting groundwater levels and groundwater quality and conducting risk assessments has been evidenced from regionally to globally. AI has demonstrated potential in cross-scale modeling, with initial progress achieved in upscaling hydraulic parameters and downscaling groundwater levels and quality predictions. Quantifying uncertainties in input data, model structures, and predictive outcomes has been used to enhance model reliability. In addition, physics-consistent explainability decreases as the simulation scale expands due to the increasing challenges in describing boundary conditions and the limited applicability of governing equations. Establishing an evaluation system for multisource uncertainties, enhancing data accessibility, and integrating various post hoc techniques and physical constraints to enhance model explainability present opportunities for the applications of AI in groundwater modeling.