Integrating Remote Sensing, Machine Learning, and Local Knowledge for Innovative Flood Susceptibility and Vulnerability Mapping

ABSTRACT This study develops a comprehensive framework for mapping flood susceptibility and vulnerability in the Cheshmeh‐Kileh forest watershed in northern Iran by integrating remote sensing (RS), local knowledge, and machine learning (ML) algorithms. This was accomplished through the application of various MLs, such as K‐nearest neighbor (KNN), random forest (RF), support vector regression (SVR), and Naive Bayes. In this study, flood susceptibility refers to the physical propensity of an area to experience flooding, influenced by geo‐environmental factors, while flood vulnerability captures the socio‐economic and institutional dimensions that determine a community's ability to cope with and recover from flood events. This research first identified critical geo‐environmental factors influencing flood susceptibility and utilized remote sensing to locate areas prone to runoff generation. Flood risk zoning was then implemented using machine learning techniques in Python. To assess flood vulnerability, data were collected from local residents via questionnaires, focusing on economic, infrastructural‐physical, institutional‐policy, and social‐cultural aspects. The flood vulnerability map was created by integrating these survey results with population density data to identify areas where high social exposure coincides with high physical susceptibility. Findings indicated that the combined remote sensing‐SVR model was the most effective for sensitivity classification, identifying sub‐watersheds 2 and 8 in the Sehezar River (a major basin within the study area) as the areas with the highest and lowest flooding susceptibility, respectively, with sub‐watershed 10 in the Dohezar River (another major basin) being the most vulnerable. The estimated values for Mean Absolute Error (0.041), Mean Square Error (0.042), Root Mean Square Error (0.205), and Area Under the Curve (0.980) demonstrated high model accuracy. The Friedman statistical test showed that the average scores for the different dimensions of vulnerability decreased in the order of: economic (0.48), social‐cultural (0.44), infrastructural‐physical (0.34), and institutional‐policy (0.28). Consequently, the economic dimension was prioritized for its highest score. Flood vulnerability mapping revealed that sub‐watersheds 5, 11, 14, and 15, which had higher population densities, were naturally more vulnerable to floods. This finding reflects a direct relationship between population density and flood vulnerability. Overall, this study underscores the urgent need for effective planning and preventive strategies to mitigate flood risks and enhance resilience in the region.