Abstract:

Based on the monthly mean temperature in 2015 from 105 weather stations in Xinjiang and digital elevation model (DEM), the spatial interpolation of monthly, seasonal and annual mean temperature in Xinjiang were done by using the vertical correction method and multiple linear regression plus residual error correction method. And the precisions of temperature interpolated by two kinds of methods were evaluated with the observations at testing stations by using the cross-validation method and correlation analysis. The results show that the root mean square error (RMSE) and mean absolute error (MAE) of temperature interpolated by the vertical correction method in Xinjiang were 0.58-1.79 ℃ and 0.46-1.17 ℃, respectively, while those interpolated by the multiple linear regression plus residual error correction method were 1.47-2.89 ℃ and 0.98-2.23 ℃. The correlation coefficient between the observed and the predicted by the vertical correction method was 0.90-0.98, while that between the observed and the predicted by the multiple regression plus residual error correction method was 0.52-0.91. Compared with the multiple regression plus residual error correction method, the error between the observed and the predicted by the vertical correction method was smaller and the correlation was better. So, the vertical correction method was more suitable for temperature interpolation in Xinjiang.

Key words: Xinjiang, temperature, vertical correction method, multiple regression plus residual error correction method

摘要：

以新疆为研究区域，运用气温垂直订正和多元回归+残差订正2种插值方法对研究区105个气象站点2015年月、季、年均温进行空间插值，并通过交叉验证和相关分析方法对其精度进行评价。结果表明：新疆地区气温垂直订正插值法的均方根误差和平均绝对误差分别为0.58~1.79 ℃和0.46~1.17 ℃，而多元回归+残差订正插值法分别为1.47~2.89 ℃和0.98~2.23 ℃；垂直订正插值法得到的气温预测值与实测值的相关系数达0.90~0.98，而多元回归+残差订正插值方法为0.52~0.91；与多元回归+残差订正法相比，气温垂直订正法的误差更小、相关性更好，更适于新疆区域气温的插值。

关键词: 新疆, 气温, 垂直订正法, 多元回归+残差订正法