Evaluation of landslide susceptibility in Ya’an City based on depth neural network model

Accurate evaluation of landslide susceptibility results are the basis of landslide risk assessment and are of great significance to disaster prevention and reduction. This paper focuses on Ya’an City as the study area and selects various factors, including elevation, slope, aspect, plane curvature, profile curvature, topographic wetness index, sediment transport index, runoff intensity index, normalized difference vegetation index, annual rainfall, peak ground acceleration, topographic relief, distance from fault, stratum lithology, distance from river, and distance from road, to construct a landslide susceptibility evaluation index system. Based on field geological survey data, a deep neural network model is used to evaluate the landslide susceptibility. The study area is classified into five categories based on susceptibility index, including landslide extremely high-prone area (12.2%), landslide high-prone area (7.0%), landslide moderate-prone area (9.8%), landslide low-prone area (17.0%) and landslide extremely low-prone area (54.1%). The accuracy of the DNN model was tested with an AUC value and compared to the artificial neural network (ANN) model. The results show that the DNN model has a higher evaluation accuracy AUC (0.99) compared to ANN (0.96). Thus, the DNN model has a better fitting degree and prediction ability in the study area than the ANN model. The extremely high-prone area and high-prone area of landslides are primarily distributed in the low altitude areas with significant human engineering activities in Ya’an City, along the roads and water systems. The main control factors affecting the development of Ya’an landslide are distance from the road, elevation and annual average rainfall.