A New Method for Temperature Spatial Interpolation Based on Sparse Historical Stations

Abstract

AbstractLong-term grid historical temperature datasets are the foundation of climate change research. Datasets developed by traditional interpolation methods usually contain data for a period of less than 50 yr, with a relatively low spatial resolution owing to the sparse distribution of stations in the historical period. In this study, the point interpolation based on Biased Sentinel Hospitals Areal Disease Estimation (P-BSHADE) method has been used to interpolate 1-km grids of monthly surface air temperatures in the historical period of 1900?50 in China. The method can be used to remedy the station bias resulting from sparse coverage, and it considers the characteristics of spatial autocorrelation and nonhomogeneity of the temperature distribution to obtain unbiased and minimum error variance estimates. The results have been compared with those from widely used methods such as kriging, inverse distance weighting (IDW), and a combined spline with kriging (TPS-KRG) method, both theoretically and empirically. The leave-one-out cross-validation method using a real dataset was implemented. The root-mean-square error (RMSE) [mean absolute error (MAE)] for P-BSHADE is 0.98°C (0.75°C), while those for TPS-KRG, kriging, and IDW are 1.46° (1.07°), 2.23° (1.51°), and 2.64°C (1.85°C), respectively. The results of validation using a simulated dataset also present the smallest error for P-BSHADE, demonstrating its empirical superiority. In addition to its empirical superiority, the method also can produce a map of the estimated error variance, representing the uncertainty of estimation.