The starting point for the creation of this dataset was HadCRUT.4.5.0.0, which is a 100-member ensemble of gridded temperature 'observations'. This was then infilled (i.e. expanded to give data everywhere) using multi-resolution lattice krigging with the uncertainty in the statistical model also sample. The R package used for the infilling and sampling was 'fields' (https://doi.org/10.5065/D6W957CT)