Machine learning and AHP for pluvial flood risk assessment in Toliara City, Madagascar

Abstract In the face of increasing urban flooding exacerbated by climate change and urbanization, developing flood risk maps is vital for evidence-based decision-making to enhance population adaptation and resilience. The reliability of flood risk maps depends on the quality of the available data and the methods used to produce them. This study combines a hybrid approach to determine population vulnerability and characterize hydrological hazards in Toliara, Madagascar. The Analytical Hierarchy Process (AHP) was used to evaluate the consistency of risk managers’ judgments and determine the weights of socioeconomic factors affecting the population of Toliara. Machine learning models, namely binary logistic regression (LR) and random forests (RF), were used to identify flood-prone areas in the city. Simulation results indicate that the RF model achieved the highest predictive performance (AUC = 0.85) for identifying flood-prone areas. Moreover, the results show that the most important parameters influencing flood risk are proximity to roads, terrain morphology, curvature and land use. The flood risk maps show that most of the city (over 50% its area) is at a very low level of flood risk, whereas around 10% of the city area is at a high to very high level of flood risk. The resulting risk and vulnerability maps are valuable tools for local managers and decision-makers, providing practical outputs for urban planning, disaster risk reduction and climate change adaptation in the region.