Improving global temperature datasets to better account for non‐uniform warming

Abstract

AbstractTo estimate changes in global mean surface temperature (GMST), one must infer past temperatures for regions of the planet that lacked observations. While current global instrumental temperature datasets (GITDs) estimate different rates of warming for different regions of the planet, this non‐uniform warming is often modelled as residuals relative to underlying trends of spatially uniform warming. To better account for spatial non‐uniform trends in warming, a new GITD was created that used maximum likelihood estimation (MLE) to combine the land surface air temperature (LSAT) anomalies of non‐infilled HadCRUT5 with the sea surface temperature (SST) anomalies of HadSST4. This GITD better accounts for non‐uniform trends in warming in two ways. Firstly, the underlying warming trends in the model are allowed to vary spatially and by the time of year. Secondly, climatological differences between open‐sea and sea ice regions are used to better account for changes in sea ice concentrations (SICs). These improvements increase the estimate of GMST change from the late 19th century (1850–1900) to 2023 by 0.006°C and 0.079°C, respectively. Although, for the latter improvement, tests suggest that there may be an overcorrection by a factor of two and estimates of SICs for the late 19th century are a significant source of unquantified uncertainty. In addition, this new GITD has other improvements compared to the HadCRUT5 Analysis dataset, including correcting for a small underestimation of LSAT warming between 1961 and 1990, taking advantage of temporal correlations of observations, taking advantage of correlations between land and open‐sea observations, and better treatment of the El Niño Southern Oscillation (ENSO). Overall, the median estimate of GMST change from the late 19th century to 2023 is 1.548°C, with a 95% confidence interval of [1.449°C, 1.635°C].