Predicting Maximum Groundwater Elevation Using Machine Learning to Support Site Investigation Planning and Development

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

Presenting Author:

Authors:

Abstract:

Elevated groundwater levels can reduce a soil's effective stress, potentially leading to foundation instability and structural failure. In high-risk infrastructure design—such as for nuclear power plants—inaccurate estimates of maximum groundwater elevation (MGE) can either result in catastrophic failure if underestimated or lead to costly over-design. Therefore, an accurate method for estimating MGE is essential for responsible engineering and planning. The goal of this study is to develop and validate a machine learning (ML) model for predicting annual/seasonal MGEs. We will use a comprehensive dataset of geospatial, climate, hydrogeological, and other relevant site characteristics from a location with extensive long-term monitoring data (e.g., USGS NWIS). The ML model's predictions will be evaluated against historical data to evaluate its accuracy. Using SHAP (SHapley Additive exPlanations) values and feature importance analysis, we will also identify the key drivers of MGEs. This study's findings will be highly applicable to the preliminary planning stages of risk-sensitive infrastructure projects, including not only nuclear power plants but also tunnels, dams, and underground storage. The resulting workflow will offer a data-driven approach to estimate MGEs, serving as a valuable tool for early-stage site screening and guiding more focused, cost-effective field investigations.