Comparison of Different Interpolation Method for Calculating Spatial Distribution of Crop Water Deficit Based on Canopy Temperature

【Objective】 Canopy temperature varies with leaf water content and can be used as a proxy of crop water deficit. In this paper, we compare different interpolation methods for calculating spatial distribution of crop water deficit based on canopy temperature. 【Method】 The calculation was based on the 10 interpolation modules in ArcGIS. We studied two experimental sites cultivated with maize - winter wheat rotation. For each site, we analyzed the accuracy and zoning effect of canopy temperature, as well as the normalized relative canopy temperature (NRCT). The accuracy and robustness of each interpolation method was evaluated based on its characteristic value, normalized root mean square error (nRMSE) and Pearson correlation coefficient between predicted and ground-truth values, as well as spatial distributions of the predicted canopy temperature and NRCT. 【Result】 The spatial distribution of canopy temperature of the winter wheat and summer maize both has a strong autocorrelation on the two sites. The canopy temperature estimated using local polynomial and universal Kriging interpolation is spatially abnormal. The nRMSE and NRCT between the measured canopy temperature and that predicted using global polynomial interpolation are the highest, being 5.9% and 28.6% respectively; their associated Pearson correlation coefficient is 0.33. The ordinary Kriging method is most accurate in that the difference between the predicted and measured canopy temperatures is less than 0.5 ℃; its associated nRMSE (3.6%) and NRCT (17.5%) are the least with a Pearson correlation coefficient 0.8. The spatial distribution of canopy temperature calculated by the simple Kriging method, disjunctive Kriging method and empirical Bayesian Kriging method is similar to that by the ordinary Kriging method; their overlapping percentage is greater than 90%. 【Conclusion】 Considering accuracy and spatial distribution of canopy temperature and NRCT, the interpolation methods is ranked in the following order based on their accuracy: ordinary Kriging method > simple Kriging method > disjunctive Kriging method > empirical Bayesian Kriging method > radial basis function method (tension spline function) > radial basis function method(regular spline function) > inverse distance weight method. Overall, the ordinary Kriging method is most accurate for estimating crop water deficit from canopy temperature.