Flood Susceptibility Mapping: Scenario-Based Multi-Criteria Decision-Making Versus Random Forest Model

Floods are among the most destructive natural hazards, posing a significant risk to lives and infrastructure worldwide. Effective flood risk management demands precise, multidimensional approaches. In this direction, the current study aims to evaluate and compare two methods for flood-risk assessment: the scenario-based Ordered Weighted Averaging (OWA) and the data-driven Random Forest (RF) method. To this end, the Great Karun watershed in Iran was chosen due to its complex hydrological and climatic conditions. Hydro-climatic, hydrological, topographic, land-cover datasets, and actual flood observations were applied and analyzed based on fifteen influencing factors recommended by expert opinion. In the OWA approach, while factor weights were determined using the Best-Worst Method, flood-risk maps were produced based on five scenarios: very optimistic, optimistic, intermediate, pessimistic, and very pessimistic. In the RF approach, factor importance index was calculated via the mean decrease impurity algorithm, and the model was trained to generate flood-risk maps. Results showed that distance from rivers and slope were the most influential factors in OWA, while precipitation and flow accumulation dominated in RF. Prediction rate for OWA scenarios ranged from 1.4 to 3.5%, while RF achieved 16.0%, and the Area Under the Curve (AUC) was 0.984. Optimistic scenarios overestimated, and pessimistic scenarios underestimated risk, with the OWA intermediate scenario most closely matching RF results. RF demonstrated superior performance for flood-risk classification, highlighting its applicability for precise flood management. Overall, this study presents a novel comparative framework by integrating scenario-based OWA-BWM and Random Forest approaches to investigate the effects of decision-maker preferences and data-driven learning on flood susceptibility mapping.