[Objective] The prediction accuracy of a regional landslide stability evaluation model was improved to solve the shortcomings of over-prediction caused by over-simplification of the landslide occurrence mechanism and the mechanical mechanism based on the static physical model of the traditional landslide stability analysis, and to determine the typical spatial-temporal variability and uncertainty of model parameters.[Methods] The data assimilation method of ensemble Kalman filtering was used to construct a regional landslide data assimilation scheme based on the TRIGRS model and SBAS-InSAR observation data in the area around the North Ring Road of Lanzhou City, Gansu Province. The coefficients of safety (F_s) in the model were assimilated, and the model parameters for the internal friction angle were updated. Then landslide stability was corrected and root-mean-square deviation (RMSD) was used to test the accuracy of the assimilated values.[Results] After assimilation, the landslide safety coefficient of the study area was significantly greater than the coefficient value predicted by the model, and the percentage of unstable area was reduced from 12 % to 7 %, which was closer to the actual observed value. The test gradually corrected the internal friction angle parameter towards the observed value, and realized the dynamic updating of the model parameters. The root-mean-square deviation decreased from 0.33 to about 0.04.[Conclusion] The data assimilation method based on the ensemble Kalman filter effectively corrected the model stability prediction results so that the actual situation of landslides in the current region was more accurately reflected with greater prediction accuracy.