A comparison among spatial interpolation techniques for daily rainfall data in Sichuan Province, China

This paper focuses on testing appropriate methods to produce gridded rainfall surfaces based on the daily rainfall observed at 43 meteorological stations in the Sichuan Province during rainy seasons of 2008–2013. Rainfall is extremely variable in both spatial and temporal distribution. Therefore, three methods, namely Ordinary Kriging (OK), Inverse Distance Weighting (IDW) and ordinary CoKriging (CK), have been selected and examined. First, we choose a suitable model by adjusting relevant parameters including the power exponent, the Neighbours to Include and the Include at Least. Then, cross-validation is used to verify the prediction performance of the three methods. The results for spatial interpolation of daily rainfall show that: (1) both OK and CK methods recorded smaller mean absolute error and the root-mean-square error than IDW method. (2) A preferable combination of assigning 2 to power exponent, 15 to Neighbours to Include and 5 to Include at Least is identified for IDW method. As for the best combination of OK and CK methods, Neighbours to Include is assigned with 15 while Include at Least almost has no effect on their performance. (3) CK is the optimal method for spatial interpolation of the daily rainfall in the rainy season of Sichuan.