ALBERT

Description: Projections of precipitation extremes over land are crucial for socioeconomic risk assessments, yet model discrepancies limit their application. Here we use a pattern-filtering technique to identify low-frequency changes in individual members of a multimodel ensemble to assess discrepancies across models in the projected pattern and magnitude of change. Specifically, we apply low-frequency component analysis (LFCA) to the intensity and frequency of daily precipitation extremes over land in 21 CMIP-6 models. LFCA brings modest but statistically significant improvements in the agreement between models in the spatial pattern of projected change, particularly in scenarios with weak greenhouse forcing. Moreover, we show that LFCA facilitates a robust identification of the rates at which increasing precipitation extremes scale with global temperature change within individual ensemble members. While these rates approximately match expectations from the Clausius-Clapeyron relation on average across models, individual models exhibit considerable and significant differences. Monte Carlo simulations indicate that these differences contribute to uncertainty in the magnitude of projected change at least as much as differences in the climate sensitivity. Last, we compare these scaling rates with those identified from observational products, demonstrating that virtually all climate models significantly underestimate the rates at which increases in precipitation extremes have scaled with global temperatures historically. Constraining projections with observations therefore amplifies the projected intensification of precipitation extremes as well as reducing the relative error of their distribution.