Effects of sampling intensity and non-slide/slide sample ratio on the occurrence probability of coseismic landslides.

Due to the influence of sampling strategy, the resulting probability of landslides using logistic regression (LR) can deviate considerably from the actual areal percentage of coseismic landslides. This study used the landslides data of the 2013 Lushan, China earthquake to further address this issue. Based on the Bayesian theory, we proposed a sampling method that selects the sliding samples and non-sliding samples based on the ratio of the stable area to the landslide area. Using this method, we tested 15 values of sampling intensities (1, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, 500, 1000, and 2000 grid cell km−2) and 12 values of non-sliding/sliding sample ratios (1, 5, 10, 25, 50, 75, 100, 125, 150, 175, 200, and 228) for the analysis of the LR model. Ten factors were considered in this analysis, including elevation, slope gradient, aspect, the topographic wetness index (TWI), peak ground acceleration (PGA), distance to the epicenter, distance to rivers, distance to roads, lithology, and annual precipitation. In terms of these 15 sampling intensities and 12 sample ratios, the samples were trained 200 and 150 times, respectively using the LR model, yielding 4800 predicted pictures of potential landslides in the study area. The results show that different sampling intensities have a certain effect on the total predicted landslide area – the higher the intensity of the samples, the more stable the prediction results. Especially when the sampling intensity reaches 1000 grid cell km−2, the total area of the predictive model is about 17.1 km2, which is close to the real area with 17.16 km2, with a difference within 2%. Different ratios of non-slide/slide sample greatly affect the occurrence probability of coseismic landslides. When the ratio is 1:1, the predicted landslide area (A p) is between 1265 and 1290 km2, with an average of 1280 km2, which is 75 times the actual landslide area. The functional relationship between the ratio of predicted area to real area (R pa) and the ratio of non-sliding samples to sliding samples (R ns) is R pa = 99.156 R ns − 0.826. It implies that the non-slide/slide sample ratio determined by the ratio of stable area to the landslide area permits us to construct real probability models to predict the areal percentages of landslides, and based on such models, the predicted probability is largely consistent with the actual areal percentage of coseismic landslides. • Real occurrence probability modeling of co-seismic landslides • The effect of sampling intensity on the resulting probability of landslides • The non-slide/slide sample ratio greatly influences landslide probability [ABSTRACT FROM AUTHOR]