A machine learning approach in spatial predicting of landslides and flash flood susceptible zones for a road network

In mountainous regions, landslides and flash floods are common and dangerous natural hazards. They often happen at the same time and seriously affect residential areas and transport systems. However, it still lacks studies on assessing landslides and flash floods simultaneously for the transport systems. Detailed maps of areas susceptible to flash floods and landslides on routes can provide managers with useful information on disaster management, adaptation, and mitigation activities. Therefore, this study aimed to build flash floods and landslides susceptibility prediction maps to measure the potential risk for a road network case study. The National-Highway-6 in Hoa Binh, Vietnam is selected as the study area. The collected data on landslides and flash floods on this route included 235 landslide locations and 88 flash flood locations. Thirteen influencing factors were applied for modeling landslide and flash flood susceptibility based on data and the association between historical landslide and flash flood sites and local geo-environmental conditions. In this study, three advanced hybrid machine learning models were developed for the modeling, including Decorate-SYS, Rotation Forest-SYS, and Vote-SYS with Systematically Developed Forest of Multiple Decision Trees (SYS) as a base classifier. The model's performance was validated using a variety of statistical criteria. The result showed that all these used models have good predictive performances in modeling the landslide and flash flood susceptibility. These proposed models can be applied effectively to other mountainous highways and contribute to better landslide and flash flood prevention, management, and mitigation.