Abstract: We introduced the exploratory spatial data analysis(ESDA) and Kriging theory,and proposed an improved spatial variability-oriented Kriging interpolating model for GPS height transformation.Five kinds of typical modeling scheme were put forward through combining Kriging interpolation factors with the ideas of step-by-step comparison.The error results shows that the No.4 scheme is the best one,which is called ESDA-K.We did an experiment with ESDA-K,neural network to simulate GPS height anomaly.Experimental results show that the simulating accuracy of ESDA-K is more excellent than that of traditional methods.The ESDA-K model's precision is stable,and the model resolves the questions that traditional methods must face,such as local undulation,over-learning,falling into regional maximum easily etc.