Research on the method of determining landslide threshold based on accelerating point dynamic identification

This article employs dynamic recognition of acceleration points to determine velocity thresholds for different deformation stages of landslides. It proposes a dimensionless time parameter, t_w, and analyzes over 60 slope cases using probability statistical methods to establish landslide threshold levels. Finally, based on various threshold determination methods and the dimensionless time threshold method, it establishes multiple early warning threshold indicators for rocky slopes along National Highway G312. The study validates the applicability of the dimensionless time threshold method by using landslide cases outside the statistical slope dataset. The research indicates the following: (1) The “two-sample t-test” method determines t_w = 0.24 and t_w = 0.39 as the first-level and second-level early warning thresholds for slopes, respectively. (2) The “ROC curve” method calculates an AUC of 0.85, confirming the high reliability and accuracy of the dimensionless time threshold method. (3) Using the dimensionless time threshold method, the analysis determines the first-level and second-level landslide thresholds for a landslide in Zhenggeer Banner, Ordos City, Inner Mongolia, as 84.03 h and 91.53 h, respectively. The second-level threshold provides an early warning time only 2.7 h ahead of the actual landslide time, demonstrating timely and accurate landslide prediction.