Linear trends in temperature extremes in China, with an emphasis on non-Gaussian and serially dependent characteristics

Abstract

Record-breaking hot and cold extremes have occurred in China in recent years and, therefore, it is compelling to investigate the long-term trend in temperature extremes at individual stations to see whether they have become more frequent. Many previous studies on the linear trend analysis of temperaure extremes in China have used oridinary least squares (OLS) regression, without consideration of non-Gaussian and/or serially dependent characteristics, or nonparametric methods, again not considering the latter, thus leaving some uncertainty in the significance testing. The present study examines in detail these characteristics in eight commonly used extreme temperature indices, on the basis of both station data and gridded data across China. The results show that 71–100% of the stations or grids cannot directly use standard OLS regression to analyze the statistical significance of the linear trend, because of either non-Gaussian or Gaussian but serially dependent characteristics in the regression residuals. Also, more than 43% of the stations and more than 54% of the grid boxes for annual indies cannot directly use the original Sen’s slope estimator and Mann–Kendall test because of serial dependence. Based on a nonparamtric method that takes into account serial dependence, the spatial patterns of the linear trend on an annual basis, as well as in hot and cold extremes, are examined for the period 1960–2017. The results show that hot extremes at most stations have increased, more than 57% of which are statistically significant; whereas, cold extremes at almost all stations have decreased, more than 32% (85%) of which are statistically significant during daytime (at night).