Abstract
Land-based compilations of gridded monthly surface air temperature anomalies, averaged into hemispheric values for the last 140 years, have been available for climatological analyses for the last 10 years or so. The analysis techniques used in their construction, particularly the need for a common reference period, mean that it is difficult to include, retrospectively, any of the new temperature datasets now available for some countries. So, despite data availability improvements in some areas, the number of stations used has fallen since 1970, both in the hemispheric averages and in their constituent grid-box datasets.
The present study is a reanalysis of both the existing and the newly available temperature datasets to produce a grid-box dataset of 5°×5° temperature anomalies. The reanalysis not only uses over 1000 more stations (2961 in total), principally covering the period from the 1920s to about 1990, but also arrests the decline of stations incorporated in real time for the latest years. Two hundred and fifty-two more stations are used in this analysis for the 1991–1993 period, compared with earlier analyses, The purpose of the reanalysis, however, is not just to calculate hemispheric averages. The improvements in station numbers used mean that the grid-box dataset should better estimate time series for small subcontinental scales.
Despite the dramatic improvements in the numbers of stations used, the results change little from earlier analyses for the Northern Hemisphere average, indicating the robustness of the earlier time series. Similar results could have been achieved with as few as 109 stations. Over the Southern Hemisphere, comparisons of the results indicate larger (but still relatively small) differences with earlier analyses, particularly over continental-scale regions.
