Threshold assessment of rainfall-induced landslides in Sangzhi County: statistical analysis and physical model

This paper proposes a multidimensional landslide warning method based on statistical and physical models. Firstly, the least square linear fitting (LSF), quantile regression (QR), and logistic regression (LR) methods are used to establish the cumulative rainfall-duration-mean intensity ( E - D - I ) threshold models based on the rainfall landslide events in Sangzhi County, Hunan Province, from 1987 to 2007, which improve the threshold precision compared with traditional I - D analysis. Then, the thresholds are quantitatively compared using list skill scores and receiver operating characteristic (ROC) curves. The results show that the skill score of T LR for E - D performed the best, indicating that the corresponding threshold equation is the most suitable for Sangzhi County. Further, the improved SINMAP model is used to analyze the slope stability for Liyuan, Sangzhi County, under four rainfall return periods of 5, 10, 20, and 50 years. By the relationship between rainfall and landslide instability probability, the rainfall threshold inducing the landslide in Liyuan Town is determined to be 120 mm/day. Finally, the landslide events from 2008 to 2017 in the SINMAP model are extracted to verify the appropriateness of the threshold equation. At least 75% (in fact, 100% of T LR for E - D ) of the landslide events are above all thresholds of 0.5 percentile. The converted I - D thresholds share similar trends compared with similar working areas globally. The multidimensional threshold proposed can provide a theoretical basis for preventing and managing landslide disasters in Sangzhi County, which is of significant academic value and has practical implications.