ABSTRACT
Risk assessment and management of mountain torrent disasters are important means to reduce disaster losses. Mountain torrent disasters occur frequently in Nanping City. This paper takes Nanping City as the research area, through multicollinearity analysis and indicator importance measurement. Nine indicators are determined, the mountain torrent disaster risk assessment model is built using a random forest algorithm, and the risk assessment results are visually expressed and zoned by GIS. The results showed that: (1) the AUC value of the stochastic forest model after parameter optimization reached 91%, and the accuracy, recall, accuracy, and F1-score of the model were higher than 80%; (2) according to the verification of two flood inventories, the classification accuracy of mountain flood disaster risk level has reached more than 90%; (3) the results show that the vegetation index, the nearest distance to the river network, and the elevation are the important disaster factors of mountain torrents in Nanping; (4) according to the statistics of mountain flood disaster risk zoning, Songxi County in Nanping City has the highest risk of the mountain flood disaster.
