用Bivariate-Test方法分析温度资料的不均一性

Ｂｉｖａｒｉａｔｅ－Ｔｅｓｔ方法（以下简称Ｂ－Ｔ方法）是一种基于二维正态分布随机矢量的假设检验方法，可以在一个分量的均值不改变的前提下，检验另一分量的均值有无改变。一些气象研究工作者最早用这种方法来分析年降水量的均一性。从温度资料不均一性产生的原因看，一般可分为两类：突变型不均一及渐变型不均一。对两种类型的不均一性，我们分别推导了理想情形下Ｂ－Ｔ方法的检验结果．并用随机试验的方法分析了检验结果对资料的敏感性。以上两方面的结果表明。渐变型及突变型不均一分别对应明显不同的Ｂ－Ｔ检验结果，在实际分析中这种差异可以用来定性判断不均一性的类型。东北区２０站３０年年平均温度资料的试算结果表明，有７站资料不能通过信度９５％的Ｂ－Ｔ检验，有３站资料不能通过信度９０％的Ｂ－Ｔ检验。其中有４个站的检验结果与理想情形下渐变型不均一的检验结果相似，有４个站的检验结果与理想情形下突变型不均一的检验结果相似。有一个站存在个别点的资料错误，有一个站可能有一次以上的不均一性发生。计算结果也表明，尽管Ｂ－Ｔ方法按照理论推导只是用来检测资料的不均一性，但对个别点上的资料质量问题也有一定的检测效果。

To Detection of the Inhomogeneity of Yearly-mean Surface Temperature Data Using the Bivariate-Test Method

Bivariate-Test method (B-T method) is a statistical method based on the hypothesis of normal distribution of two-dimensional random vector. The B-T methodcan be used to detect a system,atic change of mean in one variable under the hypothesis that the mean of the other variable does not vary. Researchers of meteorology have already used this method to detect the inhomogeneity of yearly total precipitation.The inhomogeneity of temperature data may fall into two classes: abrupt inhomogeneity and gradual inhomogeneity. The abrupt inhomogeneity may be caused bystation move, change of the method used. to compute the monthly-mean temperature,etc. The gradual inhomogeneity, on the other hand, may be caused by the change of environment around the station, such as urban heat-island effect, and so on.In this paper, we analysed the results of B-T method of the ideal abrupt. inhomogeneity and the ideal gradual inhomogeneity. Our results show that there is significant difference between the B-T method result of the ideal abrupt inhomogeneity and that of the ideal gradual inhomogeneity, and the difference can be usedto fix the reasons of temperature inhomogeneity.The yearly-mean surface temperature data of 20 stations in Northeast Chinaover 30 years Were used to analyse their inhomogeneity. The results show thatthere are 7 stations that are inhomogeneous on a significance level of 95 percent,3 stations that are inhomogeneous on a significance level of 90 percent. The B-Tmethod results of 4 stations are very similar to that of an ideal abrupt inhomogeneity. And the results of 4 stations are very similar to that of an ideal gradualinhomogeneity. There exists one station with data error. One station may haveseveral inhomogeneities. Our results also show that though the B-T method is theoretically used to detect the systematic change, of mean,it is also effective in the detection of dataerror.