Matin et al. (2026) Hydrologically constrained genetic programming for interpretable rainfall–runoff model discovery: A process–informed machine learning approach

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-09-25
• Authors: Naila Matin, Herath Mudiyanselage Viraj Vidura Herath, Abhishek Saha, Lucy Amanda Marshall, Vladan Babovic
• DOI: 10.1016/j.jhydrol.2026.136493

Research Groups

• Department of Civil and Environmental Engineering, National University of Singapore
• School of Civil Engineering, Faculty of Engineering, The University of Sydney
• Hydroinformatics Institute, Singapore
• Delft Institute of Applied Mathematics, Delft University of Technology

Short Summary

This study introduces HydroXGP, a process-informed genetic programming framework for discovering basin-specific conceptual rainfall–runoff model structures. It automates structure discovery while keeping candidate hypotheses interpretable and grounded in hydrologic process components.

Objective

• Automate the discovery of conceptual rainfall–runoff model structures using process-informed genetic programming.
• Develop an open-source, modular framework that can be easily extended with new components and flux equations.

Study Configuration

• Spatial Scale: Catchment scale
• Temporal Scale: Daily to monthly time steps

Methodology and Data

• Models used: HydroXGP (Hydrological Model eXplorer using Genetic Programming)
• Data sources: CAMELS-US dataset, a large-sample dataset of catchments in the United States.

Main Results

• The framework successfully discovered parsimonious and mass-conserving model structures that achieved strong out-of-sample performance.
• Transferability was demonstrated by transferring learned structures between hydrologically similar basins.
• Automatically induced structures performed robustly across contrasting hydroclimatic settings relative to established conceptual and deep learning benchmarks.

Contributions

• This study contributes a novel approach to automating the discovery of conceptual rainfall–runoff model structures using process-informed genetic programming.
• The framework provides an open-source, modular platform for exploring alternative structural hypotheses in hydrology.

Funding

• National University of Singapore (NUS)
• Australian Research Council (ARC)
• Singapore Ministry of Education (MOE)

Citation

@article{Matin2026Hydrologically,
  author = {Matin, Naila and Herath, Herath Mudiyanselage Viraj Vidura and Saha, Abhishek and Marshall, Lucy Amanda and Babovic, Vladan},
  title = {Hydrologically constrained genetic programming for interpretable rainfall–runoff model discovery: A process–informed machine learning approach},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136493},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136493}
}