Low frequency variability and trends in surface air temperature and humidity from ERA5 and other datasets

The ERA5 reanalysis of atmospheric and surface observations provides new estimates of low frequency variability and trends from 1950 onwards. Near-surface atmospheric values of temperature and humidity are considered here. They are compared with corresponding values from two earlier reanalyses, ERA-Interim and JRA-55, and with other observationally based estimates. Topics covered include global trends and the fits of ERA5 and ERA-Interim to land-station data, sea-surface and marine-air temperatures, estimates of actual global temperatures, regional variations of actual temperatures over land and sea-ice, trends and variability of monthly temperature anomalies from multiple datasets, and some local issues for ERA5. The discussion of relative humidity is shorter, but covers many of the same lines.
The trend and low frequency variability of global-mean temperature from ERA5 are largely consistent with values provided by other datasets from 1950 onwards. ERA5 is biased cold over the majority of the land surface prior to 1967 because the cold bias of its background forecasts is less well constrained by analysed observations then than in later years. The effect is smaller, however, than the uncertainty in all datasets that arises from differences in sea-surface temperature (SST) analysis, and no larger than is estimated to arise in some datasets from using SST rather than air temperature over sea. ERA5’s use of marine air temperature rather than SST is also one reason its global temperatures are a little higher than those from other datasets for the latest few years.
ERA5 performs relatively well for Europe from 1950 onwards, but uncertainty is somewhat larger in the mid-1960s, when there are significant gaps in observational data coverage. Its worst persistent temperature bias is over Australia prior to the 1970s. The issue in this case is not only lack of surface observations but also an unusually large warm bias of the background forecasts. Aside from this, the ERA5 surface analysis scheme does not cope well with the preponderance of observations from Australia that are for non-standard times. In addition, agreement over Australia between several reanalyses and monthly climatological datasets tends to be poorer during occasional wet spells. This stems in part at least from different definitions of daily average temperature. A number of issues elsewhere, of a more-local nature, have been identified. They relate to data gaps, questionable representation of fractional sea-ice cover, inconsistent coastal SSTs and erroneous temperatures of the Great Lakes.
ERA5 generally agrees best with GISTEMP and HadCRUT5 when its temperature anomalies are compared with those from monthly climatological datasets. Agreement is particularly close over the land masses of the extratropical northern hemisphere. Differences in the tropics and southern extratropics are more pronounced earlier in the period, when ERA5 suffers from an absence of synoptic data that is more acute than the absence of monthly averages of climatological observations. ERA5 and other reanalyses provide a more physically sound calculation of temperature over ice-covered sea than the extrapolation of land values used in the production of several of the monthly climatological datasets.
Reanalyses and the HadISDH datasets give similar depictions of interannual variability and longer term changes in moisture from the early 1970s onwards, including a net increase in specific humidity but decrease in relative humidity over land. Interannual variability is similar over sea and land, with changes over sea preceding those over land by a month or so. ERA5 is nevertheless moister over sea than HadISDH for recent years. It also appears from data fits and comparisons with later years to be biased dry as well as cold over land for the 1950s and 1960s. Long-term average values are drier over south-east Asia for ERA5 than for JRA-55 and HadISDH, but both reanalyses are drier than HadISDH over the Arabian Peninsula.
Much of the trend and low frequency variability in the reanalyses studied here is captured in the background fields of the data assimilation. Aside from helping to improve analyses of land-surface conditions, the surface air analyses of temperature and humidity act largely to reduce biases in the products provided for these variables. Issues arise where bias in the background values is large and observational coverage varies in time. A number of potential improvements have been identified.