Study Of Chinese Temperature Variation Trend Based On Detrended Fluctuation Analysis

As globing warming has raised increasing concern in these years, research on trends in climate has become a hot topic. Usually, the linear regression is employed for trend analysis. This detection method works on the basis of an assumption that the series is just first-order correlated or uncorrelated at all. However, in the real-life data, especially in the geo-environmental data, e.g. the temperature data, there generally exists a long-term correlation. Therefore, the direct appli-cation of linear regression analysis is inappropriate. This thesis employed the Detrended Fluctuation Analysis to investigate trends in temperature data in 590 meteorological stations of China in recent 46 years. The result shows that 555 stations out of 590 have insignificant trends. In contrast, only 49 stations display insignificant trends if using linear regression. Such difference in results should be due to the effect of long-term correlation. Furthermore, we tried to calculation what would be the most possible trend in long-term correlated temperature data. Our analysis demonstrates that traditional linear regression method, which does not take long-term correlations into account, significantly overestimates the trend in observed data in Tianjin by 6.52%, Hohhot by 54.98%, Nanjing by 20.35%, Haikou by 35.93%, and Xining by 59.12%. Long-term correlation is ubiquitous in the natural world. Therefore, the effect of long-term correlation has to be taken into consideration in trend analysis of the real-life data. The analysis in this thesis provides a general framework for the trend analysis of long-term correlated data.