Guo et al. (2026) A knowledge-guided machine learning framework for reconstructing climate-driven subsurface water dynamics from time-lapse electrical resistivity tomography

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-15
• Authors: Weiyu Guo, Hang Chen
• DOI: 10.1016/j.jhydrol.2026.136631

Research Groups

• School of Earth, Environment, and Sustainability, University of Iowa, Iowa City, IA, USA

Short Summary

This study develops a knowledge-guided machine learning framework using time-lapse electrical resistivity tomography (TL-ERT) and meteorological forcing to reconstruct climate-driven subsurface water dynamics on a hillslope. The framework successfully models the seasonal evolution of subsurface resistivity, providing a proof of concept for hydrologic interpretation under climate scenarios.

Objective

• To develop and evaluate a knowledge-guided machine learning framework for reconstructing the spatiotemporal evolution of subsurface electrical resistivity from daily meteorological forcing and continuous time-lapse electrical resistivity tomography (TL-ERT) observations.
• To interpret climate-driven subsurface water dynamics, particularly in heterogeneous mountain environments, by using reconstructed resistivity as a proxy for water content.

Study Configuration

• Spatial Scale: Monitored hillslope in a heterogeneous mountain environment.
• Temporal Scale: Daily meteorological forcing, continuous TL-ERT observations, reconstruction of seasonal patterns and evolution.

Methodology and Data

• Models used: Knowledge-guided machine learning framework, neural architectures (including PredRNN-based architecture), recursive framework with structured loss terms.
• Data sources: Continuous time-lapse electrical resistivity tomography (TL-ERT) observations, daily meteorological forcing. TL-ERT data were temperature-corrected.

Main Results

• All evaluated models successfully reproduced the dominant seasonal patterns of subsurface resistivity.
• The PredRNN-based architecture demonstrated the best performance, achieving a Root Mean Square Error (RMSE) of 9.5 Ω⋅m over held-out test windows.
• The framework maintained seasonally coherent trajectories within the observed record during rolling recursive reconstruction without subsequent ground-truth resistivity observations.
• Under a simulated +4 °C warming scenario, the temperature-corrected resistivity showed earlier shallow spring decreases (approximately 12%) and slight shallow summer increases, with weaker responses observed at depth.
• These resistivity changes, when converted to water content, are consistent with earlier spring wetting and shallow summer drying.

Contributions

• Development of a novel knowledge-guided machine learning framework that integrates continuous TL-ERT observations with meteorological forcing to reconstruct spatiotemporal subsurface resistivity dynamics.
• Provides a site-specific proof of concept for using this framework to model the seasonal evolution of subsurface resistivity and support scenario-based hydrologic interpretations.
• Addresses the challenge of understanding climate-driven subsurface water dynamics in heterogeneous mountain environments, where traditional physics-based models are often limited by sparse subsurface property data.

Funding

• Not specified in the provided text.

Citation

@article{Guo2026knowledgeguided,
  author = {Guo, Weiyu and Chen, Hang},
  title = {A knowledge-guided machine learning framework for reconstructing climate-driven subsurface water dynamics from time-lapse electrical resistivity tomography},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136631},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136631}
}