Integrating machine learning and geospatial data analysis for comprehensive flood hazard assessment.

Area of Science:

• Environmental Science
• Geospatial Analysis
• Machine Learning

Background:

• Flooding poses a significant global natural hazard, exacerbated by climate change.
• Robust flood hazard modeling is essential for disaster resilience and adaptation strategies.
• The Arambag region in West Bengal, India, faces substantial flood risks.

Purpose of the Study:

• To develop and evaluate an advanced machine learning framework for flood hazard assessment.
• To identify key conditioning factors influencing flood susceptibility in the study area.
• To map flood hazard levels for improved disaster management and urban planning.

Main Methods:

• Utilized multi-sourced geospatial datasets, including Sentinel-1 SAR and global flood databases for flood inventory.
• Incorporated fifteen flood conditioning factors: topography, land cover, soil, rainfall, proximity, and demographics.
• Trained and tested diverse machine learning models (RF, AdaBoost, XGB, etc.) with feature selection and interpretability methods (SHAP, Boruta).

Main Results:

• Machine learning models achieved prediction accuracy above 80% (AUC > 0.80), with Random Forest (RF) and AdaBoost showing strong performance.
• Precipitation and elevation were identified as the most significant factors contributing to flood hazard.
• An average of 17.2% to 18.6% of the study area was found to be highly susceptible to flood hazards.

Conclusions:

• The developed machine learning framework effectively assesses flood hazards, identifying critical influencing factors.
• Southern parts of the study area exhibit high flood susceptibility, impacting infrastructure and croplands.
• Findings support enhanced hydraulic and hydrological modeling for effective flood risk management and mitigation.