Evaluation of Landslide Susceptibility Based on GIS and WOE-BP Model

Susceptibility assessment of region landslides plays an important role in geological hazard risk management. In previous studies, few of them applied the combination of multivariate statistic model and machine learning method to assess landslide susceptibility. Taking Wanzhou District of Three Gorges reservoir as an example, nine index factors including slope angle, slope direction, curvature, terrain surface texture, stratum lithology, slope structure, geological structure, water distribution and land use, were selected as the evaluation indexes of landslide susceptibility. The state of each index was graded based on the contrast values calculated by weights of evidence (WOE) model, landslide area ratio and grading area ratio firstly. Then the BP neural network model optimized by particle swarm optimization (PSO-BP) was applied to obtain the weight of each index. The landslide susceptibility index (LSI) was calculated by the combining weight of states and weight of indexes determined by these two models (WOE-BP) and landslide susceptibility mapping was obtained based on the GIS platform. The results indicate that water distribution, stratum lithology and geological structure are the main index factors influencing the development of landslides in Wanzhou District. The accuracy of the WOE-BP model reaches 80.8%, better than 73.1% of WOE model and 71.6% of BP neural network model. The proposed model provides an effective approach for calculating the weight of index quantificationally and optimizing the landslide susceptibility evaluation.