Correcting for Amplification Bias and Sea Ice Bias in Global Temperature Datasets
To estimate changes in global mean surface temperature (GMST), one must infer past temperatures for regions of the planet that lacked observations. However, current global instrumental temperature datasets (GITDs) do not adequately account for the tendency of different regions of the planet to warm at different rates, creating a bias in their estimates, which this paper calls amplification bias. In addition, most GITDs do not adequately account for changes in sea ice, creating a bias in their estimates, which this paper calls sea ice bias. To estimate the impact of these two biases, a new GITD was created that used maximum likelihood estimation (MLE) to combine the land surface air temperature (LSAT) anomalies of HadCRUT4 with the sea surface temperature (SST) anomalies of HadSST4. The new GITD has improvements compared to the Cowtan and Way version 2 dataset, including an improved statistical foundation for estimating model parameters, taking advantage of temporal correlations of observations, taking advantage of correlations between land and sea observations, accounting for more sources of uncertainty, and better treatment of the El Niño Southern Oscillation (ENSO). Corrections for amplification bias and sea ice bias in the new dataset increase the estimate of GMST change from the late 1800s (1850-1899) to 2018 by 0.01°C and 0.08°C respectively, although tests suggest that there may be an overcorrection by a factor of two for sea ice bias. Overall, the median estimate of GMST change from the late 1800s to 2018 is 1.20°C, with a 95% confidence interval of (1.11°C, 1.30°C).