Flash flood and landslide susceptibility analysis for a mountainous roadway in Vietnam using spatial modeling

Abstract

Flash floods and landslides are dangerous natural hazards in hilly areas. They often occur extensively and potentially cause widespread destruction to agriculture, infrastructure, roads, houses, and human beings. This research aimed to analyze the hazard susceptibility on a mountainous roadway using advanced Machine Learning (ML) models. We conducted field surveys to collect data on flash flood and landslide locations in 2017, 2018, and 2019 on a particular roadway in Vietnam, National Highway 6, consisting of 88 flash flood sites and 235 landslide sites. The state-of-art ML models were utilized for the predictive modeling, including AdaBoost-RBF, Bagging-RBF, MultiBoostAB-RBF, and Random Sub-spaceRBF, with Radial Basis Function (RBF) serving as the primary classifier. The AdaBoost-RBF model outperformed all others in predicting landslide and flash flood vulnerability. The resulting map showed that 44.89% or 14,183 ha is in very high susceptibility zones, 15.55% or 4914 ha is in high susceptibility zones, 10.37% or 3.275 ha is in moderate susceptibility zones, 13.69% or 4324 ha is in low susceptibility zones, and 15.50% or 4899 ha is in very low susceptibility zones. A detailed map of the areas where landslides and flash floods are most likely to occur on the roadway might provide local authorities with crucial information for disaster management.

Keywords

• Flash flood
• Landslide
• Transportation
• Geospatial modeling
• Machine learning