Spatial geo-information includes global linear trend, local temporal-spatial linear, cycles, nonlinear components, and white noise. The global linear trend is filtered out by ordinary linear least square; Considering the possible temporal-spatial correlation, the local temporal-spatial signal is filered out by 7 temporalspatial linear regression equation; The cycles are filered out by Lomb frequence method;the nonlienear components are filered out by error propogation neural network model.The components are integreted sequencely or simulteouly. With training dataset and testing dataset, and model diagnostic and prediction indices, the control parameters in the model are adjusted adaptively and the model with the hast generation is found and is used for spatial prediction.