Python-based Automated Tool for Flood Susceptibility Modelling in Kerala, a part of Ecologically Sensitive Western Ghats, India

Understanding spatial patterns of flood susceptibility is essential for targeted mitigation and resilient land-use planning, especially in ecologically sensitive zones. We present a comparative flood-susceptibility modelling framework that integrates a multi-criteria AHP (analytic hierarchy process) weighted criteria-based overlay and a data-driven neural-network (NN) classifier. The classification models are trained on a binary flood inventory map (0=No flood, 1=Flood) in Kerala, a coastal state in western India, and part of the ecologically sensitive zone of the Western Ghats. The flood inventory was developed using the microwave remote sensing data (Sentinel-1 SAR of 2018 and 2020) through Google Earth Engine (GEE) and validated through ground-based event (Actual Flood Occurrence). The study compiles an extensive set of 18 conditioning factors spanning climate and hydrology (annual precipitation, drainage density, flow accumulation, stream power), topography and morphometry (elevation, slope, profile curvature, TPI, TRI), soil wetness and permeability (soil type, soil moisture, TWI, erodibility), vegetation dynamics (NDVI, SAVI), and anthropogenic influence (built-up index, population density, built-up/impervious indices, distance to road, distance to river). Feature preprocessing included resampling, scaling, and inversion (where needed), and stratified random sampling 10 million labeled pixels (train: test = 8:2). AHP pairwise comparisons produced λmax ≈ 5.2, CI ≈ 0.05 and CR ≈ 0.05, indicating acceptable consistency. Model outputs comprised hydrological, morphometric, permeability, LULC, anthropogenic susceptibility maps and composite flood-susceptibility zonation maps from both AHP and NN workflows. Validation was performed using ROC-AUC and confusion-matrix analyses to assess predictive skill and class-level accuracy. Comparative analysis reveals that the NN approach improves predictive discrimination and spatial detail compared to the expert-driven AHP map, while AHP offers more interpretable insights of the factor weights. A Python-based application has been developed to automate flood-susceptibility mapping using dynamic precipitation and vegetation data, supporting long-term prediction and the development of mitigation measures. We discuss implications for operational flood risk mapping, targeted adaptation measures, and how combining knowledge-driven and data-driven methods can provide robust, actionable susceptibility maps for decision-makers.