Effects of raster resolution on real probability of landslides

Abstract

Raster resolution of DEM is an important parameter in probability prediction of coseismic landslides using statistic methods. Taking the 2013 Minxian, China Mw5.9 earthquake that triggered massive slope failures as an example, this paper attempts to further address this issue. Based on the Bayesian theory, we conduct analysis of the logistic regression (LR) model under DEMs of different resolutions (2.5 m, 5 m, 10 m, 20 m, 40 m, and 80 m) for the study area. Eight factors are used in the LR modeling, including elevation, slope angle, profile curvature, topographic wetness index (TWI), distance to fault, distance to epicenter, distance to roads, and lithology. The samples are trained 20 times on the LR model with these six DEMs, yielding 120 predicted results. The resultant predicted landslide area (Ap), maximum predicted probability (Pmax) and AUC/AIC for the different resolutions are compared. The results show that Ap at the different resolutions is roughly same, and the best are medium (10 m, 20 m). It means that when the raster resolution is closest to the average area of the landslide, the LR model has the best prediction accuracy. Then, we use cross-scaling technique to examine the applicability of above six resulting models on different resolution datasets. Finally, 36 probability results are obtained. The results show that the satisfactory result could be achieved with high resolution DEMs but the accuracy of landslide prediction becomes increasingly worse with coarser resolutions.