Khan et al. (2026) Flood Susceptibility Mapping of the Kosi Megafan Using Ensemble Machine Learning and SAR Data

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-04-13
• Authors: Khaled Mahamud Khan, Bo Wang, Hemal Dey, Dhiraj Pradhananga, L. Micaela Smith
• DOI: 10.3390/rs18081158

Research Groups

[Information not provided in the paper text.]

Short Summary

This study developed and validated an ensemble machine learning framework for flood susceptibility mapping in the Kosi Megafan, comparing its performance against established models and a 1D-CNN. The stacked ensemble model achieved the highest performance, identifying high-risk zones with strong agreement with observed flood data and assessing the exposed population.

Objective

• To develop and validate an ensemble machine learning framework for accurate and scalable flood susceptibility mapping in the geomorphologically complex and highly flood-prone Kosi Megafan.
• To compare the performance of the ensemble framework with established machine learning models and a one-dimensional convolutional neural network (1D-CNN).
• To assess the population exposed to high-risk flood zones.

Study Configuration

• Spatial Scale: Kosi Megafan, located in Nepal and India.
• Temporal Scale: Flood data for validation spans 1992–2022 (from Dartmouth Flood Observatory).

Methodology and Data

• Models used: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Support Vector Machine (SVM), 1D-Convolutional Neural Network (1D-CNN), and a Stacked Ensemble model.
• Data sources: Remote sensing datasets (for 13 flood conditioning factors, 8 retained), Dartmouth Flood Observatory (DFO) flood data, Sentinel-1 Synthetic Aperture Radar (SAR) data, and a created flood inventory for model training.

Main Results

• The stacked ensemble model achieved the highest performance among all tested models, with an Area Under the Curve (AUC) of 0.76, accuracy of 0.70, precision of 0.69, recall of 0.72, and an F1-score of 0.70.
• The resulting flood susceptibility map identified high-risk zones predominantly in the southern and southwestern parts of the Kosi Megafan.
• There was strong spatial agreement between the identified high-risk zones and both the Sentinel-1-derived flood inventory and the Dartmouth Flood Observatory (DFO) flood data (1992–2022).

Contributions

• Developed and validated an effective ensemble machine learning framework for flood susceptibility mapping in a data-scarce, hazard-prone basin.
• Demonstrated the effectiveness of combining SAR-derived flood evidence with ensemble machine learning approaches for accurate and scalable flood susceptibility mapping.
• Integrated human risk assessment by identifying the population exposed to high-risk flood zones, addressing a gap in previous studies for complex regions.

Funding

[Information not provided in the paper text.]

Citation

@article{Khan2026Flood,
  author = {Khan, Khaled Mahamud and Wang, Bo and Dey, Hemal and Pradhananga, Dhiraj and Smith, L. Micaela},
  title = {Flood Susceptibility Mapping of the Kosi Megafan Using Ensemble Machine Learning and SAR Data},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18081158},
  url = {https://doi.org/10.3390/rs18081158}
}