Research on early warning for meteorological risks of rainfall-induced landslide hazards in typical areas of southwest Yunnan

The area of Longyang district–Mangshi section is located in the middle and lower reaches of the Nujiang river and the Daying river. The landform in this area is dominated by medium mountains, wide valleys, and basins. This area is characterized by a southern subtropical monsoon climate and a southern subtropical mountain monsoon climate. The overall structural features within the area are composed of a series of near north-south faults, and their derived secondary transverse tensile faults as well as tight folds. Lithology is mainly composed of metamorphic rock strata in the Proterozoic Gaoligong mountains, strata from the Paleozoic to the Mesozoic, and clay, fragments, gravel from the Quaternary. The study area is typically concentrated and developed with rainfall-induced landslides, and geological disasters mainly include collapses, landslides, debris flows and ground subsidence. Among these disasters, there are 1,175 landslide events, with rainfall-induced landslides accounting for over 95%. This type of disaster is the most significant in the study area. Water systems are intensively distributed in this area, with large depths of river valleys, and a pronounced variation in terrain. Landslide hazards are mainly distributed along valleys on both sides of rivers. Based on the assessment of vulnerability and the analysis of rainfall thresholds, an early warning model for rainfall-induced landslide risks has been established. This model can effectively support early warnings for regional landslide risks.

The study area is located in the section of Longyang district–Mangshi section in southwest Yunnan Province, which is prone to frequent landslide hazards. This study employed Pearson correlation coefficient to analyze the relationships among various evaluation factors. A total of 13 evaluation factors, including elevation, slope and slope direction, were selected, and grid units measuring 100 m×100 m were divided. The random forest model was employed to evaluate the vulnerability of landslide hazards and the area under curve (AUC) was utilized to verify the accuracy of the model. Then, by analyzing the coupling relationship between landslides and rainfall, the early effective rainfall intensity (EI) was calculated. The EI and rainfall duration (D) were used as the horizontal and vertical coordinates respectively, to create a double-logarithmic coordinate system. This system illustrated the scatter distribution of the probability of landslide occurrence time and allowed for the fitting of the EI-D threshold curve according to the classification standards for landslide warning grades. Subsequently, a rainfall threshold model was created. Finally, the susceptibility classification region and the EI-D rainfall threshold were superimposed and combined to establish early warning levels for meteorological risks of rainfall-induced landslides based on the EI-D rainfall threshold.

Research findings indicate that the AUC value of the training results for the random forest model is 0.84. This suggests that the model selection is appropriate and that the susceptibility evaluation results are reliable. An EI-D rainfall threshold model has been developed, and four EI-D rainfall threshold curves have been fitted. In conjunction with the evaluation results of the vulnerability of the study area, a dynamic evaluation model for landslide risks based on the EI-D rainfall threshold has been established. This model can serve as a reference for early warning evaluations of meteorological risks in the study area.