Comparison of spatial interpolation methods for precipitation distribution in Xinjiang region

Based on the data from 154 meteorological stations in Xinjiang district and surrounding areas, five interpolation methods inverse distance weighting (IDW), ordinary Kriging (Kriging), Collaborative Kriging (Cokriging), Empirical Bayesian (EBK) and ANUSPLIN spatial interpolation were implemented for simulation and cross examination of monthly average precipitation from 1995 to 2004. The test results showed that the interpolation accuracy, from most to least accurate, was ANUSPLIN, Cokriging, EBK, Kriging and IDW. Accuracy varied between months. The Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE) showed lower value in winter and spring than in summer and autumn; the Mean Relative Error (MRE) was higher in spring and autumn. While comparing precipitation distribution in Xinjiang, ANUSPLIN considered both the accuracy and smoothness of the interpolation surface, which more precisely reflected the spatial distribution of precipitation. Interpolation results from the EBK method were much higher than the actual precipitation distribution in April and July, while the interpolation was better in January and October. Cokriging interpolation results were uneven and did not accurately reflect spatial distribution. Interpolation using the Kriging method was better in October, but less so for the other three months (January, April and July). The maximum or minimum value of the IDW method was mainly distributed around the station and differed with the actual precipitation distribution.