Investigating Changes in Temperature Variability and Extremes Using Gridded In Situ-based and Reanalysis Daily Temperature Datasets

There is consensus among studies that both the global mean temperature and extreme temperatures are shifting to warmer temperatures. What remains in doubt are the characteristics of changes in the temperature distribution over the observational period, including changes in temperature variability. Differences in conclusions between studies regarding distribution changes are likely due to methodological choices and uncertainties surrounding the input datasets used. Temperature extremes generally occur on short timescales, making it critical to use daily datasets as a minimum to investigate changes. Currently, in situ-based daily temperature datasets that have complete global coverage are lacking. For this reason, reanalysis products that use assimilated observational data for the globe can be used to investigate changes in extremes and their variability. This paper investigates the sensitivity of global temperature variability over recent decades to the input dataset used. We assess the most globally complete in situ-based gridded observational dataset of daily maximum and minimum temperatures, HadGHCND, against several reanalysis products, including ERA-Interim, NCEP2 and JRA-55. Two methodological approaches are used to determine if analyses of temperature distribution changes are sensitive to both the statistical methods and the input datasets. Overall, while some reanalyses resemble HadGHCND more closely than others, depending on the region and time period assessed, ERA-Interim most closely resembles HadGHCND. NCEP2 displays the largest differences, particularly regarding the representation of temperature variance. In addition, we compare the tails of the distribution with overall mean changes in global temperature to investigate whether temperature extremes are warming faster than the mean. In general, parts of North America and Russia show that, for the cold tails of the distribution, extremes are warming in some areas up to 4°C faster than the mean. These same regions show an opposite signal for the warm tails for both maximum and minimum temperature, with the mean slightly warmer than the extremes. Other regions, such as parts of the Middle East and Australia, show extremes to be warming slightly faster than the mean for the warm tails, and slightly slower for the cold tails. These results are highly exacerbated in the NCEP2 dataset, while ERA-Interim shows similar trend magnitudes for most of the globe. The conclusions of this study will provide critical information for future work aiming to examine changes in daily temperature extremes over the observational period, and further, which products might be appropriate to use to evaluate climate models.