Investigations into the role of anthropogenic emissions in the occurrence of extreme weather often use a method that compares simulations of atmospheric climate models run under a factual scenario of historical boundary conditions observed during the period of the event against simulations run under a counterfactual scenario of what those boundary conditions might naturally have been over that same period in the absence of anthropogenic emissions. A particular requirement for this experiment design is an accurate estimation of ocean surface boundary conditions for use by the counterfactual natural simulations. Here we use output from the CMIP5 multi‐climate‐model archive to develop a robust estimate of sea surface temperatures and sea ice conditions for use in counterfactual natural simulations, intended as a benchmark estimate to facilitate comparison across climate models and across studies. This development includes tests to ensure that the final estimate is stable from year‐to‐year and stable against other perturbations to the methodology, as well as consideration of the strengths and weaknesses in comparison to other available attributable warming estimates. While this estimate is tailored specifically for the International CLIVAR C20C+ Detection and Attribution Project, it can be used by related projects as well.

Recent studies have examined the role of anthropogenic emissions in the probability of extreme weather events. These studies examine an event aggregated over a spatial domain, but the dependence on domain definition is unknown. Here we investigate this dependence for the frequency of daily weather extremes across South Africa using a climate model run under both a real-world and a nongreenhouse gas world scenario. Attributable changes in extremely hot and cold days are dominated by large-scale spatial structures, with sharp gradients at the 100 km scale arising for hot events because of the large magnitude of changes. The attributable probabilities of heavy precipitation events are spatially heterogeneous down to the 100 km resolution of the climate model. Therefore, while estimates of attributable probability for temperature events may often be considered valid within smaller and neighboring spatial domains, it appears that estimates for heavy daily precipitation events may be sensitive to the definition of the event.