A novel model order optimization method based on semivariogram of geostatistics(GS) was proposed. With the combination of this method with support vector regression(SVR), we constructed a new non-linear forecasting model of multidimensional time series analysis named GS-SVR that can show the dynamic characteristics of sample set as well as the effect of environmental factor. To evaluate the performance of GS-SVR, two sets of ecology data were predicted by one-step method, the results showed that GS-SVR had the highest accuracy and had the a...