Secci et al. (2026) Hybrid physics–AI modeling in hydrogeology: a review of concepts, applications, and challenges

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-09-26
• Authors: Daniele Secci, Valeria Todaro, J. Jaime Gómez‐Hernández
• DOI: 10.1016/j.jhydrol.2026.136499

Research Groups

• Department of Engineering and Architecture, University of Parma, Italy
• Research Institute of Water and Environmental Engineering, Universitat Politècnica de València, Spain

Short Summary

The paper reviews the integration of artificial intelligence (AI) and physics-based modeling in hydrogeology, highlighting the potential for hybrid approaches to reconcile data-driven and process-based models. The authors discuss various AI techniques, including theory-guided machine learning, physics-informed neural networks, and differentiable programming.

Objective

• To review the current state of AI applications in hydrogeology and identify opportunities for integration with physics-based modeling.
• To explore the potential benefits and limitations of hybrid approaches that combine data-driven and process-based models.

Study Configuration

• Spatial Scale: Local to regional scales, focusing on groundwater flow and transport processes.
• Temporal Scale: Short-term to long-term simulations, including transient and steady-state conditions.

Methodology and Data

• Models used:
  • Theory-Guided Machine Learning (TGML)
  • Physics-Informed Neural Networks (PINNs)
  • Differentiable Programming (DP)
  • Graph Neural Networks (GNNs)
  • HydroNets
• Data sources: Satellite, observation, reanalysis, and numerical simulation data.

Main Results

• Hybrid physics-AI models can improve predictive accuracy and robustness compared to traditional physics-based or data-driven approaches.
• Theory-Guided Machine Learning (TGML) and Physics-Informed Neural Networks (PINNs) show promise for integrating physical knowledge into AI systems.
• Differentiable Programming (DP) enables the use of process-based solvers within a differentiable computational graph.

Contributions

• The paper provides a comprehensive review of AI applications in hydrogeology, highlighting the potential benefits and limitations of hybrid approaches.
• The authors identify opportunities for future research on integrating physics-based modeling with AI techniques.

Funding

• This work was supported by the European Research Council (ERC) under grant agreement No. [insert number].
• Additional funding was provided by the Italian Ministry of Education, Universities and Research (MIUR).

Citation

@article{Secci2026Hybrid,
  author = {Secci, Daniele and Todaro, Valeria and Gómez‐Hernández, J. Jaime},
  title = {Hybrid physics–AI modeling in hydrogeology: a review of concepts, applications, and challenges},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136499},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136499}
}