Pawar et al. (2026) Hydrologically enhanced machine learning framework for urban flood inundation mapping using multi-sensor remote sensing data: a case study of Mumbai, India

Identification

• Journal: Natural hazards and earth system sciences
• Year: 2026
• Date: 2026-10-20
• Authors: Ankush S. Pawar, Gayatri M. Phade
• DOI: 10.5194/nhess-26-5119-2026

Research Groups

• Department of Electronics and Telecommunication, Sandip Institute of Technology and Research Centre, Nashik, Maharashtra, India
• Department of Electronics and Telecommunication, PVGCOE&SSDIOM, Nashik, Maharashtra, India

Short Summary

This study developed a hydrologically enhanced machine learning framework for urban flood inundation mapping in Mumbai using multi-sensor remote sensing data, demonstrating that incorporating hydrologic-topographic predictors significantly improves model performance and physical interpretability. The optimized XGBoost model achieved an overall accuracy of 71.7 % and an AUC of 0.803, with hydrologic predictors increasing discrimination from approximately 0.66 (SAR-only) to 0.80 (complete framework).

Objective

• To develop a hydrologically enhanced machine learning framework for urban flood inundation mapping in the Mumbai Metropolitan Region by integrating multi-sensor remote sensing observations with physically meaningful hydrologic-topographic predictors.
• To systematically evaluate the contribution of these hydrologic-topographic predictors to model performance and physical interpretability.

Study Configuration

• Spatial Scale: Mumbai Metropolitan Region (approximately 18.9–19.3° N, 72.7–73.0° E). Data resolutions include 10 meters (Sentinel-1, Sentinel-2), 30 meters (SRTM), and 0.05° (approximately 5 kilometers) (CHIRPS).
• Temporal Scale: June–September 2019 (monsoon season).

Methodology and Data

• Models used:
  • Machine Learning Classifiers: Random Forest (RF), Extreme Gradient Boosting (XGBoost), and an RF-XGB ensemble model.
  • Hydrologic Predictors: Relative Elevation Model (REM) and Rainfall-Normalized Index (RNI).
• Data sources:
  • Satellite: Sentinel-1 C-band Synthetic Aperture Radar (SAR) (VV, VH polarizations), Sentinel-2 Surface Reflectance Optical imagery (B2–B11).
  • Elevation: Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM).
  • Precipitation: Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) Daily Rainfall.
  • Ancillary: HydroRIVERS vector dataset (for river mask and REM computation).
  • Platform: Google Earth Engine (GEE) for data acquisition and preprocessing.

Main Results

• The optimized XGBoost model achieved the highest predictive performance with an overall accuracy of 71.7 % and an Area Under the Receiver Operating Characteristic curve (AUC) of 0.803.
• The inclusion of hydrologic-topographic predictors (Relative Elevation Model - REM and Rainfall-Normalized Index - RNI) significantly improved model discrimination.
  • AUC increased from approximately 0.77 (using traditional SAR-optical features only) to 0.803 (with REM and RNI), representing a relative improvement of 5–6 %.
  • The complete predictor framework, including hydrologic predictors, increased AUC from approximately 0.66 (SAR-only predictors) to approximately 0.80.
• Statistical evaluation using the DeLong test confirmed that these improvements were statistically significant at the 95 % confidence level (e.g., Full Model vs. NoREM Model: p = 3.97 × 10⁻⁶; Full Model vs. Default Predictor Set: p = 2.37 × 10⁻⁴).
• Feature importance analysis highlighted REM and RNI as crucial variables, enhancing discrimination beyond traditional spectral and SAR predictors.
• The resulting flood maps exhibited hydrologically plausible spatial patterns, consistent with major drainage networks and low-lying areas across the study region.

Contributions

• Integration of multi-sensor remote sensing observations with hydrologically informed terrain-conditioned predictors (REM and RNI) to enhance the representation of flood-generating processes within a machine learning framework.
• Systematic evaluation of the contribution of hydrologic-topographic predictors through feature importance analysis, feature ablation experiments, and statistical significance testing.
• Assessment of model robustness and internal stability using cross-validation, sensitivity analysis, and comparative evaluation of multiple machine learning classifiers.
• Provides a reproducible, scalable, and physically interpretable approach for urban flood inundation mapping, demonstrating potential for operational flood monitoring and decision support in data-limited tropical urban environments.

Funding

The authors acknowledged the European Space Agency (ESA) for providing Sentinel-1 and Sentinel-2 data, the United States Geological Survey (USGS) for the SRTM Digital Elevation Model, and the Climate Hazards Group for the CHIRPS precipitation dataset. Geospatial processing was carried out using Google Earth Engine. No specific funding projects or programs were explicitly listed.

Citation

@article{Pawar2026Hydrologically,
  author = {Pawar, Ankush S. and Phade, Gayatri M.},
  title = {Hydrologically enhanced machine learning framework for urban flood inundation mapping using multi-sensor remote sensing data: a case study of Mumbai, India},
  journal = {Natural hazards and earth system sciences},
  year = {2026},
  doi = {10.5194/nhess-26-5119-2026},
  url = {https://doi.org/10.5194/nhess-26-5119-2026}
}