A hybrid geospatial framework for flood hazard zonation using AHP and machine learning: Case study of Kaziranga Tiger Reserve, India

Flood hazard zonation is crucial for disaster risk reduction, land-use planning, and ecological conservation in flood-prone regions. The increasing frequency and intensity of hydro-meteorological disasters necessitate advanced geospatial approaches for disaster preparedness and mitigation. This study integrates Geographic Information System (GIS), remote sensing, and machine learning techniques to assess flood susceptibility in the Kaziranga Tiger Reserve, a UNESCO World Heritage Site. Five GIS and machine learning model like Analytic Hierarchy Process (AHP), Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), and Naïve Bayes (NB) were applied to derive flood hazard zones. The AHP model identified 35.94% (50,166.5 ha) of the area as "Very High" flood risk, whereas RF and SVM classified 27.01% (37,670.78 ha) and 25.27% (35,252.19 ha) in the same category, respectively. The DT model predicted the largest flood-prone area, with 54.94% (76,677.93 ha) under "Very High" risk. Model validation via Sentinel-1A SAR flood inundation maps from August 2017 and June 2024 corroborated ML outputs, with RF and DT achieving high predictive performance (AUC = 0.90 and 0.89, respectively). Statistical analysis revealed key predictors including low elevation (<100 m), drainage density exceeding 1.7 km/km², and proximity within 500 m to river channels strongly correlated with flood susceptibility. Land use dynamics indicated a 7% increase in sparse vegetation and 4% increase in flooded flats from 2010 to 2024, further exacerbating flood risk. ROC curve and synthetic minority oversampling technique (SMOTE) analysis confirmed ML superiority over AHP, especially under data imbalance conditions. This research provides a spatial decision support framework for policymakers and disaster management authorities to enhance early warning systems and mitigation strategies. Future work should focus on real-time hydrological modeling and deep learning-based flood forecasting for improved resilience planning in dynamic flood environments.