AI-driven high-resolution flash flood susceptibility mapping and early warning in Son La, Vietnam Quang Binh Bui
Institute of human geography and sustainable development, VASS, Hanoi, Vietnam
Abstract
Flash floods pose severe risks to rural communities in Northwest Vietnam, particularly in Son La province, due to its rugged terrain and intense monsoon rainfall. This study introduces an innovative framework for high-resolution flash flood susceptibility mapping and early warning, leveraging artificial intelligence (AI) and machine learning (ML) integrated with Geographic Information Systems (GIS). By combining multi-source remote sensing data (Landsat-8, Sentinel-1, ALOS-2 PALSAR) with topographic and meteorological inputs, we developed a 10-m resolution spatial model using Random Forest (RF), Artificial Neural Networks (ANN), Decision Trees (J48), and Logistic Regression (LR). The RF model achieved superior performance, with an Area Under the Curve (AUC) of 0.95, identifying 25% of the study area as highly flood-prone. An AI-supported WebGIS platform and mobile application were deployed, delivering real-time warnings with an 86% detection rate during 2020 monsoon trials. This research enhances disaster resilience and supports sustainable rural development, offering a scalable solution for flood-prone mountainous regions.
Keywords: Flash flood susceptibility, Artificial Intelligence, Machine Learning, Random Forest, Early warning system
