Biased Estimates of Changes in Climate Extremes From Prescribed SST Simulations

Large climate model ensembles are widely used to quantify changes in climate extremes. Here we demonstrate that model‐based estimates of changes in the probability of temperature extremes at 1.5 °C global warming regionally differ if quantified using prescribed sea surface temperatures (SSTs) instead of using a fully coupled climate model. Based on the identical climate model used in two experimental setups, we demonstrate that particularly over the tropics and Australia estimates of the changes in the odds of annual temperature extremes can be up to more than a factor of 5 to 10 larger using prescribed SSTs rather than a fully coupled model configuration. The two experimental designs imply a different perspective on framing projections. If experiments conditional on prescribed observed SSTs are interpreted as unconditional real‐world projections, they project changes in extremes that are systematically biased high and overconfident. Our results illustrate the importance of carefully considering experimental design when interpreting projections of extremes. Plain Language Summary There is great interest in understanding the likelihoods and associated risks of potential future climate extremes, especially at the Paris Agreement global warming targets of 1.5 and 2 °C warming above preindustrial conditions. In this study, we assess the implications of the model setup for the quantification of changes in the odds of temperature extremes between different global warming levels. Our analysis illustrates the strong sensitivity in the outcomes of such analyses related to the use of different model experiments. We demonstrate that despite using the exact same global climate model the projected changes in the probability of extreme annual temperature anomalies for a climate consistent with a 1.5 °C warming target are in some cases much larger if sea surface temperatures are prescribed over a decade rather than if the model is run in a fully coupled configuration. If prescribed sea surface temperature experiments are interpreted as a projection for the real world at the end of the 21st century independent of ocean variability, they regionally lead to estimates of changes in extremes that are systematically biased high and overconfident. Our results illustrate the importance of carefully considering experimental design when interpreting projected changes in extremes. Key Points Model‐based estimates of changes in probability of temperature extremes at 1.5 degrees Cels