基于 SBAS-InSAR 技术及 LSTM 神经网络的席芨滩巨型滑坡形变监测及预测 postprint

Abstract: This study examines the surface deformation characteristics and deformation rate prediction of large-scale landslides in the upper regions of the Yellow River between the Longyang and Jishi Gorge riverbanks. The study area was the Xijitan giant landslide within the Guide region of the upper Yellow River. The Small Baseline Subset Interferometric Synthetic Aperture Rader（SBAS-InSAR）technology was employed to monitor the surface deformation of the Xijitan giant landslide and analyze, its deformation rates and variation characteristics for the period 2019—2022.The results show that the following. (1)The maximum surface deformation rate of the land- slide body was -96 mm·a−1,with a maximum cumulative deformation of 464.71 mm. Distinct deformation zones were observed along the front and rear edges of the landslide body, with surface deformation rates ranging across−96~16 mm·a−1.(2)The cumulative deformation of characteristic points on a landslide body, determined using SBAS-InSAR technology, exhibited a maximum cumulative deformation of -140.50 mm. (3)The long short-term memory (LSTM) neural network model was used to predict the cumulative deformation of these points, and the results were compared with those obtained using support vector machine（SVM）and back propagation（BP）neu-ral network models. The LSTM model demonstrated high prediction accuracy, with an absolute error within 5mm and a goodness-of-fit (R2)greater than 0.8.This confirmed the effectiveness of the LSTM model in predicting the cumulative surface deformation of landslides. Thus, the findings of this study provide data support and practical guidance for the enhanced monitoring of giant landslide deformation in the upper Yellow River regionand the early detection of potential landslides.