Coastal Flood Vulnerability and Spatial Risk Assessment for Small Island Urban of Cat Ba Town, Northern Vietnam

Abstract

This study develops an integrated geospatial framework to assess coastal flood vulnerability in Cat Ba town, northern Vietnam. A binary flood inventory was generated from Sentinel-1 SAR imagery, while indicators describing the built environment were extracted from Google Earth and topographic datasets. Flood susceptibility was modeled with three decision tree algorithms: Classification and Regression Tree (CART), J48 and Reduced Error Pruning Tree (REPTree). Model performance was evaluated using the overall accuracy, AUC and Cohen’s Kappa. To capture the spatial distribution of built environment exposure, a Built Environment Vulnerability Index (BEVI) was developed using the AHP-based approach. The BEVI composes 05 indicators, including built-up area density, road network density, building height, site elevation and distance to major infrastructures. The composite flood risk map, formed by overlaying BEVI and flood susceptibility outputs, revealed that low-lying areas near the town center and harbor are highly prone to flooding. Among the tested models, CART achieved the best performance, with 94.84% accuracy, an AUC of 0.925 and a Kappa value of 0.874. The resulting risk map offers a practical tool for land-use zoning, drainage planning, and emergency management, directly supporting the 2021–2030 Vietnamese national strategy for disaster prevention and climate adaptation. In particular, targeting the high-BEVI zones for adaptation will help fulfill the strategy’s goal of proactive flood mitigation and climate adaptation.