Accounting for uncertainty from internal variability in global-temperature based attribution of climate extremes with single realisations

Attribution of regional climate change to anthropogenic forcing within the single realisation available from observations is an important but challenging goal for statistical methods in climate science. Correlating regional conditions with global temperatures is a popular approach, especially for attributing downstream impacts on human health or the economy. However, the influence of internal variability on such approaches remains unquantified. Here, we use large ensembles from three climate models to quantify the role of internal variability for attribution of climate extremes. For temperature extremes, internal variability brings uncertainties exceeding 40% of the climate change signal across 50, 38, and 9% of the global surface area in the MIROC6, MPI-ESM1-2-LR and CanESM5 models respectively. We furthermore show that these uncertainties can be accurately inferred from individual ensemble members using a block-bootstrap procedure - offering potential for application to observations. For precipitation extremes however, relative uncertainties are substantially larger, exceeding 100% across 90%, 75% and 55% of the global surface area in the three models used, and the block-bootstrapping fails to replicate the uncertainty of the large ensemble - although spatial aggregation reduces the discrepancies between them to some extent. This work provides a basis for the attribution of temperature extremes from observations which can robustly capture the uncertainty driven by internal climate variability, but indicates the need for caution when applying such approaches to noisier variables such as precipitation.