Abstract: Post-earthquake landslides are secondary geological hazards in areas adjacent to fault zones,posing a serious threat to human life and infrastructure.To address the limitation of traditional landslide susceptibility assessments,which generally fail to consider the spatiotemporal dynamics of post-earthquake landslides,this study employs SBAS-InSAR technology combined with a ConvLSTM neural network model to evaluate landslide susceptibility following the Ms6.4 Yangbi earthquake in the northern segment of the Honghe fault zone.Firstly,time-series surface deformation information is obtained from 204 Sentinel-1A ascending and descending orbit images acquired from January 2021 to January 2025.This is combined with high-resolution optical imagery for landslide identification.Secondly,environmental factors are analyzed using the geographical detector method and Pearson correlation analysis.Finally,the ConvLSTM neural network model is used to classify landslide susceptibility levels,and the results are compared with those of three other neural network models: BP,LSTM,and CNN.The results show that the root mean square error and mean absolute error of the ConvLSTM model are both lower than those of the other three models,and the model achieves an AUC value of 0.912,indicating higher accuracy and better performance.High-risk and above-risk areas are concentrated to the west of the northern segment of the Honghe fault zone,distributed within 10 km on both sides of the Weixi—Weishan fault.This method further improves the accuracy of landslide susceptibility assessment and can provide data support and technical reference for post-earthquake landslide disaster prevention and control in areas near fault zones.

Key words: InSAR, landslide hazards, artificial neural networks, landslide susceptibility assessment, Honghe fault zone