Abstract: To take advantage of geometric and physical information of slopes and restrain the influence of the observation outliers on the estimates of deformation parameters,a filtering model handling unknown systematic errors is developed.An adaptive fitting algorithm for the systematic errors based on moving windows is presented and the estimation method for covariance matrices of the predicted states is given.The presented algorithm utilizes the statistical information of observations as well as landslide related information such as mechanics status and geological conditions.The results of the GPS monitoring network show that the algorithm may reduce the effect of abnormal observation by fitting the geophysical information,and therefore improve the precision of deformation parameter estimates.