青藏高原气温分布的空间插值方法比较

Comparison Study of Spatial Interpolation Methods of Air Temperature over Qinghai-Xizang Plateau

使用反距离平方、趋势面、Kriging插值、Cokriging插值和综合方法对青藏高原1961—1990年30年平均1月气温进行空间插值比较研究,其中后两种方法能够把影响青藏高原气温分布的关键因素———高程置于插值算法之中。反距离平方和趋势面插值的结果都与实际情况相差较远;普通Kriging插值能够反映出青藏高原气温分布的一定的空间结构,但结果依然不好;由于考虑了高度变量,Cok riging插值表现出一定程度的性能改进,但因为台站海拔高度偏低,插值结果依然不理想。前4种方法效果不好的原因主要是因为青藏高原上气象台站稀少且高原西北地区无气象台站。综合方法把气温分解为结构化分量和随机分量,使用直减率把气温订正到同一海拔高度后,再对它们做Kriging插值分析,其结果较为正确地反映了青藏高原气温空间分布的特征,误差远小于其它方法。研究结果表明:样本本身的空间分布是影响插值精度的重要因素,合理的采样设计是必要的前提;对于台站稀少的地区,必须把随机插值方法和确定性方法结合起来估计气候变量的空间分布。同时给出了青藏高原1月的气温空间分布状况。

The spatial interpolation methods of inverse distance weight, trend surface, Kriging, Cokriging and combined method are utilized for the comparison study on spatial interpolation of January mean air temperatures from 1961 to 1990 in the QinghaiXizang Plateau (QXP). The results of the former four methods display a nearly similar spatial structure. Moreover, the center of low temperature in northwestern QXP can not be revealed due to uneven spatial distribution of meteorological stations and limited obsevations. In the combined method, monthly mean air temperature is divided into structural and stochastic components, and, with the help of Digital Elevation Model (DEM), the structural component is simulated by the lapse rate of different latitude and altitude zones in the QXP, the stochastic component is treated by the Kriging method. The results of combined method correctly represent spatial distribution of monthly mean air temperatures in the QXP. Its errors are far less than ordinary Kriging method that depends only on samples. Furthermore, the errors have been averaged, and they are evenly distributed in the study area. The comparing study of these interpolation methods on the monthly mean air temperature in the QXP indicates that the best spatial interpolation method does not exist, the best interpolation method only exist in given conditions. The best results of data interpolation can only be achieved by exploratory spatial data analysis and various experiments.The interpolation results showed that spatial distribution of monthly mean air temperatures in the QXP is controlled by altitudinal and latitudinal zonations, displaying a structure of high temperature in eastern and southern parts and low temperature in western and northern parts. The January mean air temperature in the QXP is lower than -13℃ and increases southeastward. The biggest center of low temperature is located in northwestern QXP north of 34°N and west of 93°E, with January mean air temperature lower than -16℃ and the lowest interpolated value of -25℃; another lower temperature area is in Qilian Mountains. Secondary level low temperature centers with smaller areal extent have been found in the regions around Wudaoliang, Tuotuohe and Maduo in the interior of the QXP, with January mean air temperature lower than -15℃. High temperature center of January is located in southeastern QXP, the high temperature ridge extends northward to Changdu and westward to Lhasa and Xigaza along the 29°N (Yaluzangbu River). The air temperature increases southeastward from -3℃ to 6℃.