气候变化趋势分析中自相关的检验与去除

受资料本身、分析方法及未来排放情景假设等因素影响, 气候变暖幅度尚存在较大的不确定性。从分析方法入手, 探讨气象观测序列可能存在的自相关及其对气候变化趋势分析的不确定性影响。引入了Durbin-Watson一阶自相关检验方法对气象观测序列进行检验, 并用Cochrane-Orcutt方法去除存在的自相关。分析发现:浙江省平湖市气温序列存在的自相关放大了该站气温的升温趋势, 并且虚高了气温变化趋势的显著性水平。因此, 对资料序列进行自相关检验与去除是十分必要的。

The Checking and Removing of the Autocorrelation in Climatic Time Series

It is well known that global climate is warming over the past decades. And there are great uncertainties for the assessment of warming magnitude caused by meteorological data itself, different methods adopted, hypothesis of future emission scenario, and other factors as well. Ordinary least squares (OLS) is usually used to analyze the trend of climate change by fitting linear regression to the time series of meteorological observation. And the hypothesis of independence should be met by the random error of the time series, otherwise autocorrelation exists in the time series and uncertainties will appear in the results.Unfortunately research on the autocorrelation of climate change is weak. So taking the temperature observation (including mean temperature, maximum temperature and minimum temperature) of Pinghu meteorological station in Zhejiang Province during 1954—2004 as an example, the autocorrelation possibly existing in time series and its influence on the trend analysis of climate change are studied. And the method of Durbin-Watson test is introduced to check whether the autocorrelation exists or not in the time series during the utilization of OLS in the trend analysis of temperature change. And if so, the method of Cochrane-Orcutt is adopted to remove the autocorrelation. Analysis indicates that the warming trend is magnified by the existence of autocorrelation in Pinghu temperature time series, and the confidence level of warming trend is improperly increased, by which uncertainties of clim ate change is added. In a word, it is indispensable to check and remove the autocorrelation of climatic time series, and much more work should be done further to test the autocorrelation and its possible impacts on the uncertainties of warming forecasting at regional and global scale.