Peeples et al. (2026) A hybrid machine learning and physics-based approach to model high-resolution inundation

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-08-13
• Authors: Amelia Peeples, Elena Leonarduzzi, Laura E. Condon, R. M. Maxwell
• DOI: 10.1016/j.jhydrol.2026.136246

Research Groups

• Department of Civil and Environmental Engineering, Princeton University, USA
• Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Switzerland
• Department of Hydrology and Atmospheric Sciences, University of Arizona, USA

Short Summary

This study proposes a hybrid machine learning and physics-based approach to model high-resolution inundation. The method combines coarse-resolution hydrologic modeling with machine learning downscaling to produce accurate 100 m resolution hourly binary inundation status at a significantly lower computational cost.

Objective

• Investigate the feasibility of using machine learning to downscale surface water products from physics-based models for large-scale applications.
• Develop and test a hybrid methodology that combines coarse-resolution hydrologic modeling with machine learning downscaling to produce high-resolution inundation maps.

Study Configuration

• Spatial Scale: 1800 km2 watershed in Colorado, USA
• Temporal Scale: Hourly timesteps over the full 2019 water year

Methodology and Data

• Models used: ParFlow-CLM (PF-CLM) integrated hydrologic model, Random Forest (RF) machine learning model
• Data sources: CW3E precipitation and temperature data, NASA DEM, PriorityFlow R package for watershed boundary and slopes

Main Results

• The hybrid RF-PF model achieved 97.6% accuracy in predicting high-resolution inundation status compared to the 100 m PF-CLM model.
• The baseline RF-PF model was more accurate than the coarse PF-CLM model alone, with a higher F1 score and IoU.
• The temporal ablation RF-PF model applied to unseen timesteps showed a small decline in accuracy from the baseline RF-PF model.

Contributions

• This study demonstrates the applicability of hybrid machine learning and physics-based approaches for large-scale high-resolution surface water modeling.
• The proposed methodology can produce accurate 100 m resolution hourly binary inundation status at a significantly lower computational cost than traditional high-resolution hydrologic modeling.

Funding

• Not specified in the provided text.

Citation

@article{Peeples2026hybrid,
  author = {Peeples, Amelia and Leonarduzzi, Elena and Condon, Laura E. and Maxwell, R. M.},
  title = {A hybrid machine learning and physics-based approach to model high-resolution inundation},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136246},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136246}
}