234-6 Leveraging Surface–Subsurface Interactions for Water Table Prediction in Barrier Islands: A Machine Learning Approach Using Gradient Boosting

Session: Advance Ground Surface Modeling for Hydrological and Environmental Applications

Presenting Author:

Authors:

Abstract:

Understanding and predicting shallow groundwater fluctuations —specifically water table elevations— in barrier island systems is critical for managing groundwater resources but also for assessing the vulnerability of buried infrastructure under changing climate conditions, sea-level rise, and anthropogenic impacts. These low-lying, porous environments exhibit intricate surface–subsurface interactions shaped by precipitation patterns, soil moisture, atmospheric pressure, tidal dynamics, and geomorphic setting. In this study, we developed a data-driven framework to forecast water table elevations in a coastal barrier island using high-frequency well monitoring data, complemented by in-situ and remotely sensed environmental variables. Temporal aggregation and LOESS-based imputation addressed data gaps and enhanced integration of surface signals with subsurface behavior. Feature engineering included lagged predictors, statistical summaries, and hydrological indicators such as soil moisture, sea level pressure, precipitation, terrestrial water storage, sub surface runoff, and tidal stage. Spatial synchronization and inter-well correlation revealed subsurface connectivity between monitoring sites located near the Gulf and the backside of the island, reflecting geomorphic and hydrostratigraphic controls. We evaluated the performance of six hybrid machine learning models—combinations of gradient boosting algorithms (XGBoost, LightGBM, and CatBoost) with or without LOESS-based preprocessing—across multiple wells and hydrostratigraphic settings. FE+LightGBM and LOESS+FE+LightGBM consistently demonstrated superior generalization and predictive accuracy, with NSE values often exceeding 0.90 during validation. Case studies show the models captured delayed water table responses to cumulative rainfall and changes in root zone soil moisture, while also distinguishing groundwater behavior based on tidal proximity and landform orientation. This research underscores the potential of machine learning to uncover the linkages between surface observations and subsurface hydrologic processes in complex coastal systems. It provides insights into how geoscientists can better infer groundwater behavior in data-limited barrier island settings using surface-driven signals.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-10270

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.