A combined estimate of global temperature

Abstract Recently, several global temperature series have been updated using new data sets, new methods, and importantly, assessments of their uncertainties. This enables us to produce a timely estimate of the annual global mean temperature with a smaller combined estimate of uncertainty. We describe the hierarchical model we propose, and a Bayesian scheme for fitting the model, allowing for dependence between the data sets, which all use some of the same observations. The discrepancy between individual data series and the combined estimate illustrates potential sources of deviation between them. In addition, we test the sensitivity of the results to each of the series, using a leave‐one‐out approach. This is a way of combining all the data sets in a way that improves on the straight or precision weighted ensemble mean, thus providing a more authoritative global temperature series with corresponding standard errors, which are smaller than that of individual products. Using the combined estimate of the global temperature series, we estimate that the global temperature has increased 1.2°C with a standard error of 0.03°C over the 1880–1900 average. By taking into account the uncertainties of the estimates rather than just comparing the estimates, we find that the probability that 2020 was the warmest year on record is 0.44, while the years 2015–2020 are virtually certain to have been the six warmest years in recorded history. We show that our estimate performs similarly to the reanalysis product ERA5, and that the satellite record from University of Alabama does not agree very well neither with ERA5 nor with our product.