Landslide susceptibility assessment in the Central Yunnan Plateau by assembling optimized statistical and machine learning models

Background Landslides are a significant geohazard in mountainous regions worldwide, with increasing occurrences due to the changing climate and intensified land-use activities. The Central Yunnan Plateau (CYP) in Southwest China is particularly prone to landslides due to its geological and climatic conditions. Methods This study employs a data-driven approach to optimize parameters related to rainfall, geology, and land use, using point-biserial correlation and decision tree methods. To verify the effectiveness of our method, we conducted a comparative analysis across three models: random forest, support vector machine, and logistic regression. The optimized random forest model was then used to sort out the importance of influencing factors. Finally, a landslide susceptibility map was generated by averaging the results from the three models. Results Key findings indicate that 24-hour and 30-day cumulative rainfall are primary climatic predictors in landslide susceptibility. Specifically, landslide susceptibility increases sharply with 24-hour rainfall up to 40 mm, while the effect of 30-day cumulative rainfall shows a slight initial decrease followed by a gradual increase. Decision tree analysis further reveals that landslide susceptibility is lower in forest and grassland compared to cultivated and constructing areas, and is highest in unconsolidated and carbonate rocks (30–70%). Proximity to roads within 1,000 m also present high landslide susceptibility. Model predictions suggest that the eastern and southern parts of the CYP with high rainfall and significant human activities have higher landslide susceptibility. Conclusion This study underscores the importance of improved rainfall monitoring, targeted infrastructure maintenance, and strategic land-use planning to mitigate landslide hazards and enhance safety for residents in vulnerable regions of the CYP.