Urban Flood Susceptibility Assessment in Arid Environment Using a Novel Hybrid Deep Learning Approach

Urban flooding poses a growing threat to cities in arid and semi-arid regions, where rapid urbanization, limited drainage infrastructure, and scarce hydrological data exacerbate disaster risks. This study develops Hydro-TransformerNet, a novel hybrid deep learning framework for urban flood susceptibility mapping under data-scarce conditions. The model integrates a ResNet34-based convolutional encoder for fine-scale spatial feature extraction, a temporal transformer module for simulating sequential hydrological processes, and a hydrologically guided attention mechanism that embeds physical terrain and rainfall relationships into the learning process. The model uses nine remote sensing-based geospatial variables: Digital Elevation Model (DEM), Slope, Normalized Difference Water Index (NDWI), Normalized Difference Vegetation Index (NDVI), land use/land cover (LULC), Soil Type, Impervious Surface, Rainfall, and Topographic Wetness Index; as input features for the Sharjah region, UAE. Due to the absence of historical flood information, a synthetic flood mask was developed using expert knowledge of hydrological patterns and low-lying urban areas. The model demonstrates strong predictive performance, achieving an AUC of 0.945, with high alignment between predicted flood-prone areas and plausible hydrological zones. Quantitative evaluation using ROC, precision-recall, calibration, and confusion matrix metrics confirm the model’s reliability and robustness. Feature importance analysis via SHapley Additive exPlanations (SHAP)-based feature importance analysis reveals elevation, slope, and rainfall as dominant factors in flood susceptibility. The model’s predictions enabling integration with GIS platforms for urban planning and disaster mitigation. Hydro-TransformerNet represents a significant step in creating scalable, interpretable, and GIS-ready flood prediction tools for urban regions facing data scarcity. This study contributes a scalable, explainable nd data-efficient deep learning solution for early-warning systems, infrastructure planning, and climate resilience management in rapidly developing arid cities.