Threshold prediction model for the occurrence of shallow soil landslides in red beds triggered by heavy rainfall

In this study, a threshold prediction method for shallow soil landslides in red bed areas is presented. A heavy rainfall event occurred in Nanjiang County on September 16–18, 2011, which triggered more than 600 shallow soil landslides. Another heavy rainfall event that occurred in Yunyang and Fengjie counties from August 30 to September 1, 2014, triggered hundreds of shallow soil landslides. We analyzed the slope angle, upslope drainage, cross-section, terraced fields, and free-face of a (potential) shallow soil landslide. A machine learning model (a support vector machine, SVM) was used to create a topographical factor. The higher the value of the topographical factor is, the higher the probability of soil landslide occurrence is. A normalized rainfall factor was used in this study. The higher the value of the normalized rainfall factor is, the higher is the probability of soil landslide occurrence is. A probability indicator for the occurrence of shallow soil sliding was obtained by establishing the relationship between the topographical factor and the normalized rainfall factor in the red bed area. The prediction model applies to soil landslides in the red bed areas in Nanjiang, Yunyang, and Fengjie counties. It is assumed that the prediction model is also suitable for other red bed areas.