Landslide susceptibility evaluation based on FR-DNN coupling model: A case study on Yanyuan County

Abstract

Yanyuan County is located on the southeastern edge of the Qinghai-Tibet Plateau, with strong regional tectonic activities. Under the internal and external dynamics actions, landslide disasters are extremely developed in the country, which has caused huge casualties and economic losses. It is necessary to carry out landslide susceptibility assessment and then control the regional landslide disasters scientifically. Based on the 1∶50000 geological disaster survey in Yanyuan County, this study selected 13 landslide susceptibility evaluation factors, including elevation, slope, slope direction, terrain curvature, distance from fault, stratigraphic lithology, distance from water body, average annual rainfall, topographic wetness index, stream power index, normalized difference vegetation index, distance from road, and land use type. Based on 27596 grid points of landslide disaster data, combining the traditional Frequency Ratio (FR) model with the advantages of quantitative analysis and data quantification and the emerging Deep Neural Network (DNN) model with the powerful nonlinear learning and fitting ability, The FR-DNN coupling model was constructed to evaluate landslide susceptibility. The study area is divided into five levels: extremely high susceptibility area, high susceptibility area, medium susceptibility area, low susceptibility area, and extremely low susceptibility area, with an area percentage of 11.90%, 18.38%, 18.34%, 9.13%, and 42.25%, respectively. The accuracy was verified by the AUC value of the ROC curve. The AUC values of the FR model and the FR-DNN coupling model are 0.754 and 0.859, respectively. The prediction accuracy of the FR-DNN coupling model is improved by 10.5% compared with that of the FR model, indicating that the FR-DNN coupling model has better prediction ability and is more suitable for landslide susceptibility evaluation in the study area.

Keywords

• landslide
• frequency ratio method
• deep neural network
• susceptibility