Assessment of urban flood vulnerability using multi-criteria decision making and geospatial techniques in Chhatrapati Sambhajinagar, Maharashtra, India

Urban flooding poses a significant threat to lives, infrastructure, and sustainable development, particularly in rapidly expanding Indian cities. This study aims to evaluate urban flood susceptibility in Chhatrapati Sambhajinagar, Maharashtra, India, by integrating geospatial analysis and machine learning techniques. Eleven flood-conditioning parameters—elevation, slope, aspect, rainfall, distance to stream, distance to road, topographic wetness index (TWI), stream power index (SPI), plan curvature, normalized difference vegetation index (NDVI), and land use/land cover (LULC)—were derived from remote sensing and GIS datasets. The models employed include Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Tree (CART). Multicollinearity analysis (Tolerance > 0.7; and VIF < 1.5) confirmed the independence of predictors. Model calibration was performed using grid search-based hyperparameter tuning with tenfold cross-validation. Among the three algorithms, RF achieved the highest predictive performance (Kappa = 0.887; ROC–AUC = 0.988; PRC–AUC = 0.991), followed by SVM (Kappa = 0.839; ROC–AUC = 0.980; PRC–AUC = 0.984) and CART (Kappa = 0.817; ROC–AUC = 0.954; PRC–AUC = 0.961). The spatial distribution of flood susceptibility indicates that low-lying and river-adjacent areas, particularly along the Kham River, exhibit very high flood risk, whereas western elevated zones show minimal susceptibility. The integration of machine learning models with geospatial datasets effectively delineates flood-prone zones and enhances urban resilience planning. The findings provide valuable insights for policymakers to strengthen urban flood management, drainage planning, and sustainable development strategies in semi-arid Indian cities.