Downscaling with AI reveals the large role of internal variability in fine-scale projections of climate extremes

Abstract The computational cost of dynamical downscaling limits ensemble sizes in regional downscaling efforts. We present a generative-AI approach to greatly expand the scope of such downscaling, enabling fine-scale future changes to be characterised including rare extremes that cannot be addressed by traditional approaches. We test this approach for New Zealand, where heterogenous regional climate effects are anticipated. End-of-century projected daily precipitation extremes become progressively more responsive to climate change as rarer events are considered, on average increasing at $$5.6\%^{\circ }\textrm{C}^{-1}$$ for annual extremes to $$9\%^{\circ }\textrm{C}^{-1}$$ for 100-year extremes. The internal variability of precipitation extremes increases with warming much faster than the mean and nearly twice that of Clausius-Clapeyron scaling, dominating other sources of uncertainty. This precipitation “tail stretching" phenomenon for internal variability is accompanied by similar stretching of the extreme 850 hPa specific humidity distribution. Overall, fine-scale precipitation changes are less predictable than widely assumed and require substantially larger ensembles for reliable assessment than previously recognised.