利用多模式对中国气温序列中不连续点的检测

Detection of Discontinuities in Chinese Temperature Series Using A Multiple Test Approach

发展了一种对元数据依赖程度较小的气候序列均一化思路,在缺乏元数据的基础上,采取3种不同的方法,对中国东南部地区年平均最高、最低气温序列进行了均一性检验;并对其中不连续的气温序列进行了订正。对比表明,订正后序列较订正前更为合理。在检验过程中发现,对于最低气温检验可以明显得出比最高气温更多的不连续点。为了进一步认识这个问题,对此进行了进一步分析,认为最低温度对变化更为敏感主要是由于其物理特征最低温度一般出现在夜间,夜间大气较为稳定,因此一些变化(如仪器,迁站)可能引起明显的不连续现象,而最高气温往往出现在白天,因此大气混合比较充分,空间的均一性要明显强于最低温度。另外,热岛现象、台站密度等因素均是影响气温序列均一化的因素。

Annual average daily maximum and minimum in situ temperature time series from southeastern China were evaluated for artificial discontinuities using three undocumented change point test statistics. Breakpoints in each series were identified via comparison to a reference series based on values from a number of nearby stations. Relative change point adjustments were estimated from a procedure previously used only with respect to documented change points. Under the requirement of agreement between test statistics, a list of artificial change points was generated using the full collection of series. Each target series then was adjusted accounting for change points in series from nearby stations that otherwise could bias estimates of discontinuity amplitude. A spatial comparison of linear trends from adjusted series suggests that major artificial discontinuities were accounted for in the homogenization process. However, a larger number of relative change points were detected in annual average minimum temperature series than in average maximum temperature series. It is possible that minimum temperature fields are more sensitive to changes in measurement practice than maximum temperature fields. A likely explanation is that local climate effects are less evident when the atmospheric boundary layer is well mixed, which is more common during daytime. Fewer relative change points also were revealed when a larger number of nearby series were used to form the composite references. In this case, it is likely that the composite reference series is more insulated from change points in its component series than the larger number of component series. Consequently, the number of change points falsely attributed to the target series is reduced with a larger number of reference series components. Finally, although a zero slope model was assumed for the differences formed between target and reference series, the possibility of local trends was cursorily examined given previous evidence of urban heat island （i. e. , local） trends in this region.