Prediction of Landslide Susceptibility Based on Neural Network Model and Negative Sample Selected by Information Value Model
Year: 2024
Type: article
Abstract: Landslides occur frequently in the Chishui River Basin under the interaction of the geological environment and local human activities, negatively impacting the safety of people and properties, and social order; thus, landslide-prone areas must be analyzed. Here, based on field research and data collection performed in the Chishui River Basin, we identify 13 landslide conditioning factors to construct a landslide susceptibility identification system through principal component analysis by comprehensively considering the geological environment, topography and geomorphology, climate and hydrology, human engineering activities, vegetation cover, and other factors. The information volume model was used to select non-landslide points, and the back-propagation (BP), long- and shortterm memory (LSTM), and convolutional neural network (CNN) models were selected to predict the landslide susceptibility zoning in the study area; the area under the curve values of the three models were 0.981, 0.984, and 0.997, respectively. The CNN was significantly more valid in predicting landslide zones than BP and LSTM and could better predict landslide susceptibility. CNNs have a promising future in landslide susceptibility analysis. These findings provide a basis for landslide susceptibility assessment, which can aid in developing appropriate pre-disaster prevention and post-disaster relief programs to decrease the threat posed by existing or future landslides.
Authors: Zixuan Wang, Jingru Tang
Institution: Guilin University of Technology
Language: English
Subfield: Management, Monitoring, Policy and Law
Field: Environmental Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4404334295