Abstract: This paper comprehensively employs InSAR and machine learning techniques to conduct landslide susceptibility assessment in the key landslide-prone area in the northern part of Xiahe county,Gansu province.The deformation information obtained by SBAS-InSAR is incorporated as a dynamic evaluation factor into the 11 static factors.Three models,namely RF(random forest),LR (logistic regression),and XGBoost (extreme gradient boosting),are used for susceptibility assessment,and their evaluation performances are compared and analyzed.The results show that among the three assessment models,the XGBoost model has the best performance.The results indicate that the XGBoost model with the addition of surface deformation variables has a higher evaluation accuracy than the XGBoost model using only static factors.Its comprehensive performance indicators,such as AUC value,Recall,Precision,and F1,reach 0.93,0.896,0.894,and 0.898 respectively.Therefore,incorporating surface deformation variables obtained by SBAS-InSAR technology as landslide susceptibility evaluation factors can improve the accuracy of model prediction and enhance the effectiveness of the assessment.

Key words: landslide susceptibility evaluation, machine learning, SBAS-InSAR, evaluation factor, Xiahe county