Assessing landslide susceptibility based on the random forest model and multi-source heterogeneous data

Landslides pose significant threats to human lives and property. The usefulness of mapping susceptibility in predicting these events, by providing early warning and implementing preventive measures, cannot be overstated. This study relies on the well-established grid management mechanism in Dengfeng and utilizes historical data collected from field surveys. Combining multi-source heterogeneous data obtained by aerospace technology, and using random forest model and feature selection, a landslide sensitivity assessment system based on spaceground collaboration was constructed. The study categorized areas according to their susceptibility to landslides using probabilistic forecasts. Specifically, regions were classified into four susceptibility categories: very low (landslide probabilities below 0.26), low (0.26 to 0.38), moderate (0.38 to 0.50), and high (above 0.50). This resulted in a comprehensive landslide susceptibility map that was rigorously analyzed and assessed. The findings indicate that: (1) high, moderate, low, and very low susceptibility areas for landslides in Dengfeng are 127.36, 527.49, 385.67, and 184.03 km2, respectively, accounting for 10.40 %, 43.08 %, 31.49 %, and 15.03 % of the total area. (2) The spatial distribution of landslides in Dengfeng exhibits a "multi-core clustering-radiate distribution" pattern, from the peripheral areas to the center. (3) The spatial distribution of high, moderate, low, and very low susceptibility areas for landslides is primarily influenced by factors such as distance from faults, elevation, and distance from rivers. The study has constructed a landslide sensitivity assessment system based on space-ground collaboration that can provide replicable research methods, provide scientific support for disaster reduction and prevention work, and reduce casualties and property losses caused by landslides.