MAPPING FLOOD SUSCEPTIBILITY WITH MAXENT AND SATELLITE-DERIVED PRESENCE-ONLY DATA

Abstract

Accurate identification of flood-prone areas is a cornerstone of disaster risk management, particularly in data-scarce regions characterized by a scarcity of detailed hydrological data. The research centers on Quang Tri province, Vietnam, a coastal region frequently impacted by extreme weather events and severe flooding. This study proposes an effective approach that integrates remote sensing with the Maximum Entropy (MaxEnt) model to generate flood susceptibility maps. To ensure reliability, 153 initial flood points interpreted from Sentinel-1 SAR imagery were refined to 129 independent samples using a spatial thinning process to mitigate spatial sampling bias. Flooded areas were extracted using the Otsu-based thresholding method, ensuring the accuracy of input data. The model was constructed based on the complex interactions of nine key environmental parameters: Elevation, Rainfall, Slope, Soil type, Distance to faults, Land Use/Land Cover (LULC), Height Above Nearest Drainage (HAND), Topographic Wetness Index (TWI), and Drainage density. The model’s performance was robust, achieving an Area Under the Curve (AUC) value of 0.837, indicating high predictive accuracy. Permutation importance analysis identified the primary environmental drivers as Elevation (36.6%) and HAND (31.5%). The resulting Flood Susceptibility Index (FSI) enabled a detailed classification of susceptibility zones across the province. This methodology provides a rapid, cost-effective tool for flood susceptibility mapping in data-limited regions, offering robust support for policymakers in developing adaptive planning and mitigating disaster-induced damages.