English: Temperatures with respect to the 1951-1980 mean over the last 2 millennia (orange) and observational estimates ranging from 1850 to 2018 (black). The lightly shaded region contains the 95% confidence interval, while the darker area contains the 68% confidence interval. The temperature anomalies are computed in https://www.nature.com/articles/sdata201788 using a wide set of proxies that have been vetted by the Pages2k Consortium https://www.nature.com/articles/sdata201788, as. They include tree ring data, data from corals and ice core data.

The confidence interval incorporates two sources of uncertainty. The first one derives from the the method of translating information in the proxies into temperatures. Seven different peer-reviewed methods are employed and all considered equally likely in the figure. The methods range from a simple linear method where proxies sensitivities are fitted to the observational to determine past temperature, to more complicated methods in which information about spatial patterns is taken into account, or information about natural forcing is taken into account.

The second type of uncertainty is in the selection of proxies. Some proxies might not be entirely reliable and their inclusion might slightly bias the reconstruction. By randomly selecting a subset of proxies, this uncertainty can be quantified.

Code to create the figure is adapted from https://www.nature.com/articles/sdata201788. The figure is a simplification of Fig 1a from this paper.