Attribution of climate extremes with global temperature regressions: accounting for uncertainty from internal variability

Maximilian Kotz , Markus G. Donat

Attribution of climate change to anthropogenic forcing is an important goal in climate science. Regression of local conditions on global temperatures is a popular statistical approach, but whether the uncertainty derived from observations alone can reflect the range of outcomes possible due to internal variability remains unclear. Here, we use large ensembles from three climate models as an idealized testbed to quantify the role of internal variability for such attribution of climate extremes with single climate realisations. Internal variability generates large uncertainty in the linear-regression slope between global temperatures and regional temperature extremes, exceeding 40% of the ensemble mean slope across 50, 38, and 9% of the global surface area in the MIROC6, MPI-ESM1-2-LR and CanESM5 models respectively. Nevertheless, we show that these uncertainties can be accurately inferred from individual ensemble members using a block-bootstrap procedure - offering potential for a robust application to observations. Block-bootstrapping also accurately captures the absolute uncertainty for precipitation extremes, despite a substantially smaller signal-to-noise ratio which leads to underestimated and overly noisy estimates of relative uncertainty at present-day warming levels.  This work provides a basis for the attribution of climate extremes from observations which can robustly capture the uncertainty associated with internal climate variability.