Landslide susceptibility evaluation based on landslide classification and ANN-NFR modelling in the Three Gorges Reservoir area, China

Abstract

The Chongqing section of the Three Gorges Reservoir area (TGRA) is a high-risk geological catastrophe warning zone in the Yangtze River Basin. This study aimed to categorize landslides and improve the accuracy of landslide susceptibility evaluations by combining the artificial neural networks (ANN) and the normalized frequency ratio (NFR) models. Considering that the indicator factors have different effects on various landslides, the landslides were separated into four types, that is, giant, large, medium, and small. Thirteen indicator factors were selected through correlation analysis, including elevation, lithology, precipitation, land use, population density, and so on. Combined with 7777 historical landslide events, an ANN-NFR coupling model was proposed. The susceptibility grade prediction of landslides for the entire region was conducted using a GIS platform. Compared to the NFR model, the ANN-NFR model could improve the accuracy of landslide susceptibility evaluation by approximately 4%. This indicates that the ANN-NFR model is an effective method to calculate the weights of indicator factors. The FR value of large landslides is 9.201 in the very susceptibility region, and the model evaluation effect is superior to that of the other three types of landslides. The success rate of the ANN-NFR model after landslide classification was 78.9%, and the prediction rate was 78.6%, both of which were greater than the unclassified ANN-NFR model. Therefore, the ANN-NFR (landslide classified) model could improve the prediction accuracy and can provide a scientific basis for disaster prevention, mitigation, and management in the TGRA.