Abstract: The Jinsha river is the upstream section of the Yangtze river located in the mountainous region of western China. Due to complex topography and frequent disasters, it often endangers people's lives and properties as well as cause serious damage to infrastructure, Timely identification of the feature and patterns of landslide deformation is conducive to the prevention and early warning of landslide disasters. Firstly, the one-dimensional line-of-sight deformation results were obtained based on the small baseline subsets (SBAS) technique for the Sela landslide in Gongjue county, Tibet, and then, the deformation rates and time series deformation of the two-dimensional deformation in the east-west direction and vertical direction were calculated with the multidimensional small baseline subsets (MSBAS) technique. The results show that the Sela landslide mainly produces in east-west deformation and the cumulative deformation of the feature point on landslide front edge can exceed 100 mm. Comparing and analyzing the time series deformation and regional rainfall, the deformation rate of the landslide front edge was found to be accelerated due to heavy rainfall from June to September each year, pulling the middle and rear parts of the landslide accelerated deformation. Finally, a long short-term memory (LSTM) model was used to predict and analyze the time series deformation of landslide, providing the appropriate prediction results with both periodic and slow-moving deformation case, and the points in the severe deformation and also in accordance with their motion trends. The obtained results can provide a reference for the early warning of similar landslide disasters in the valley of Jinsha River.