Abstract: With the development of InSAR technology and the richness of SAR data sources, experts and scholars from various countries have solved the problems of SAR data coherence, registration, orbit error and terrain error correction. It is very effective for observing small-scale deformations such as landslides(relative to tectonic movement and land subsidence).Nowadays, atmospheric phase error is still an important factor affecting InSAR applications. The use of spatial correlation to remove regional low-frequency atmospheric phase delay is a simple and efficient method, and we do not need any external weather data. However, in the existing commercial software for processing InSAR,such as GAMMA,the atmospheric model uses a linear curve model of elevation and phase, which cannot adapt to the complex atmospheric phase wet delay in the southwestern mountainous region of China, and often introduces additional errors. In this paper, I studied the distribution characteristics and optimal fitting function of atmospheric phase with elevation, and implements it on Matlab software platform. In this way, I can remove the simulated atmospheric phase from the unwrapped phase, and make the deformation information of the landslide clearer. We used the bank slope of the Qiaojia section of the Jinsha River Valley as the study area, and performed the identification verification of the creep deformation slope. Result shows that the use of elevation phase fitting polynomial is easier to identify the landslide range than a single linear fit, and reduce the omission of geological disasters. That is of great significance for the observation of landslide deformation on the D-InSAR reservoir bank in southwestern China.