Measuring Uncertainties in Global Temperature Changes

The first quantitative estimates of uncertainties in global warming calculated global and hemispheric annual temperature anomalies by combining land surface air temperature and sea surface temperature (SST) through an optimal averaging technique. The results have been incorporated into the Third IPCC Assessment Report (Folland et al., 2001).

To assess the uncertainties in annual global and hemispheric average surface temperature anomalies due to data gaps, random data representivity errors and measurement errors, a two-step optimal averaging (OA) method was employed. The OA method provides a better estimate of the true mean than does a simple average and a consistent framework on which to add independent uncertainties. We present the first analysis of global and hemispheric surface warming trends that attempts to quantify the major sources of uncertainty. We calculate global and hemispheric annual temperature anomalies by combining land surface air temperature and sea surface temperature (SST) through an optimal averaging technique.

The more robust RSOA (reduced space optimal averaging) method used here minimises errors. In particular, RSOA seeks to produce the most accurate estimate of the average by applying smallest weights to the most uncertain values, and is therefore expected to produce smaller errors than other methods. As a result, the global RSOA average may not be exactly equal to averages calculated by different methods, e.g. taking the average of the two hemispheres.

The credibility of the optimum averages before 1942 depends considerably on the accuracy of the bias corrections to SST. The largest known source of bias occurred before late 1941 after which time SST began to be sampled mostly via ship engine intakes, common to this day, rather than using uninsulated or partly-insulated buckets (Folland and Parker, 1995). The global annual mean bias correction increases steadily from 0.17°C in 1872 to 0.30°C in 1900 and 0.39°C in 1920, remaining around 0.4°C until 1941.

European temperatures are well reproduced after 1895; earlier observations may be biased cold by thermometer exposures (Parker, 1994). The cause of the discrepancy in Australian data between 1910 and 1930 is unclear; it might be due to less accurate SST data south of Australia. Newly adjusted LAT data for three stations in southeast Australia are less warm in the late nineteenth century so that the corrected-SST simulations for the average of these southeast Australian stations agree better.

Without bias corrections to SST, ensemble mean LATs are significantly too cold between 1872 and 1941 in all major regions except extratropical S. America



Acknowledgements. This work was supported by the UK Ministry of Defence and the UK Dept. of the Environment, Transport and the Regions. PDJ is supported by the US Dept. of Energy. Through the Met Office authors, the work is Crown Copyright. The authors thank R.E. Livezey and M.R. Allen for valuable discussions.



Global temperature change and its uncertainties since 1861
C. K. Folland1, N. A. Rayner1, S. J. Brown1,2, T. M. Smith3, S. S. P. Shen4, D. E. Parker1,
I. Macadam1, P. D. Jones5, R. N. Jones6, N. Nicholls7 and D. M. H. Sexton1
1Hadley Centre, Met Office, Bracknell, UK
2Now at Dept. Environment, Transport & Regions, London, UK
3National Climatic Data Center, Asheville, NC
4Dept. Mathematical Sciences, Univ. Alberta, Edmonton, Canada
5Climatic Research Unit, Univ. East Anglia, Norwich, UK
6Commonwealth Sci. and Indust. Res. Org., Aspendale, Australia
7Bureau of Meteorology Research Centre, Melbourne, Australia