Comprehensive urban flood risk assessment using machine learning algorithms and GCM projections

Urban floods have become a pressing concern as cities worldwide face unprecedented flooding events that severely impact the lives and livelihoods of millions in densely populated areas. This issue is particularly alarming in developing countries, where rapid, unplanned urbanization often outpaces the development of adequate infrastructure. Climate change exacerbates this challenge by intensifying the frequency and severity of extreme rainfall events, further straining fragile urban systems. Given the growing vulnerability of urban areas to flooding, it is crucial to develop targeted mitigation strategies grounded in comprehensive urban flood risk assessment. The present study aims to quantify flood risk and leverage Global Climate Models (GCMs) for predicting future flood scenarios. A sensitivity analysis is performed on spatial layers, including land use/land cover (LULC), elevation, slope, rainfall, stream density, distance to roads, distance to rivers, population, population density, literacy rates, and building footprint, to evaluate their influence on flood risk. Machine learning (ML) algorithms—support vector machine (SVM), random forest (RF), gradient boosting (GB), and artificial neural networks (ANN)—are employed to generate urban flood risk zones (UFRZ). The UFRZs derived from these algorithms are validated using the area under the curve – receiver operating characteristic (AUC-ROC) metric to ensure accuracy. The optimal UFRZ model is then used to predict future urban flood risks based on GCM outputs. Fifteen downscaled, bias-corrected GCMs are evaluated against observed rainfall data for the historical period (1985–2020) to identify the best-performing model for the region. Future flood risk predictions are made for three time periods: 2025–2050, 2051–2075, and 2076–2100. Identifying high-risk flood zones will aid in formulating effective mitigation strategies, providing a roadmap for flood resilience that can be adapted for similar regions globally.