Sugahara et al. (2026) Slope-weighted cost distance for modeling flood dynamics: a physics-informed random forest approach

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-08-16
• Authors: Takumi Sugahara, Tatsuro Sato, Yasuhiro Mitani
• DOI: 10.1016/j.jhydrol.2026.136241

Research Groups

• Disaster Risk Reduction Research Center, Graduate School of Engineering, Kyushu University

Short Summary

This study proposes a novel physics-informed random forest framework that incorporates slope-weighted cost-distance features to model flood dynamics. The framework significantly outperforms static baselines and demonstrates reliable performance for hydrographs withheld from training.

Objective

• Develop a data-driven approach that replicates the behavior of physical models for rapid flood inundation forecasting

Study Configuration

• Spatial Scale: Local (Kinugawa River Basin, Japan)
• Temporal Scale: Real-time (instantaneous predictions)

Methodology and Data

• Models used: Random Forest (RF) and LightGBM
• Data sources: Physics-based flood model simulations (iRIC-Nays2DFlood), Digital Elevation Model (DEM) data from aerial laser surveys

Main Results

• The proposed framework achieved an F1-score improvement of 0.386, outperforming static baselines.
• Hydrograph feature-space sampling strategy improved predictive accuracy for withheld hydrographs by 11%.
• SHAP analysis provided quantitative evidence that the model learned statistical relationships consistent with flow-path selection and volume-related flood expansion.

Contributions

• Introduced physics-informed features (slope-weighted cost distance, relative elevation, elapsed time, and cumulative inflow volume) to standard ML models.
• Developed a data-efficient sampling strategy for withheld hydrographs within a predefined feature space.

Funding

• This research was funded by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.

Citation

@article{Sugahara2026Slopeweighted,
  author = {Sugahara, Takumi and Sato, Tatsuro and Mitani, Yasuhiro},
  title = {Slope-weighted cost distance for modeling flood dynamics: a physics-informed random forest approach},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136241},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136241}
}