Spatial Distribution Patterns and Susceptibility Assessment Via PSO-BP Neural Network of Landslides Triggered by the Jiuzhaigou Earthquake in Sichuan on August 8th, 2017

Abstract

On August 8, 2017, a 7.0-magnitude earthquake occurred in Jiuzhaigou County, Ngawa Prefecture, Sichuan Province, China. This earthquake followed the Wenchuan earthquake in 2008 and the Lushan earthquake in 2013, both of which were strong earthquakes with magnitudes of 7.0 or above in Sichuan Province. The epicenter of this earthquake was in Zhangzha Town, Jiuzhaigou County. It is very important to interpret the spatial distribution of landslides quickly after an earthquake for disaster emergency rescue. The distribution of geological disasters in an earthquake area is obtained based on the interpretation of high-resolution remote sensing images after an earthquake. The main geological disasters are small landslides and collapses. In terms of the results of the spatial distribution of disasters, we analyzed the spatial distribution law and control factors of coseismic disasters (elevation, slope angle, slope aspect, peak ground acceleration, and proximity to faults). The study shows that the geological hazards are relatively developed along the gully, with an obvious fault effect, and concentrated within 2 km of the seismogenic fault. Based on the above foundation, a particle swarm optimization (PSO)–BP neural network model for predicting landslide susceptibility was established by optimizing the initial weight and threshold of a BP neural network using the PSO algorithm. The landslide susceptibility assessment considers eight factors contributing to landslide occurrence, including elevation, slope angle, slope aspect, proximity to stream network, lithology, density of geological boundaries, proximity to faults, and proximity to the road network. The validity and accuracy of the model was tested by calculating the area under the receiver operating characteristic curve, which was 0.918. The experimental results showed that the PSO–BP neural network model exhibited satisfactory accuracy in predicting landslide susceptibility.