Do km-scale global models reshape our understanding of local changes in temperature and precipitation extremes?

Human-induced global warming manifests as a distinct spatial pattern of changes in temperature and precipitation extremes. IPCC assessments of such changes are primarily based on models of the latest Coupled Model Intercomparison Project (CMIP6), which are limited in their representation of local details due to their rather coarse resolution of 50-200km. Here, we test if the first multi-decadal simulations with two fully-coupled km-scale global climate models (ICON and IFS), project greater or smaller local changes in extremes in response to global warming, focusing on annual minimum and maximum temperature, as well as on extreme precipitation. 

Using spatially pooled rank histograms of changes, we find that IFS behaves remarkably similarly to the CMIP6 multi-model mean in many cases, indicating a very low range of local trends across the globe despite its high resolution. ICON, in turn, shows a much broader range with more strongly positive or negative local trends than any of the CMIP6 models. However, while this leads to ICON being more similar to the observation-based ERA5, further analysis also reveals that this behavior is, at least partly, caused by unrealistic change signals in some regions, where local extreme temperature changes can exceed 15K per degree of global warming even in the historical period. 

Notably, both km-scale models show a higher fraction of strong positive trends in extreme precipitation than CMIP6 models. This is a promising result as CMIP6 models have previously been shown to underestimate the area fraction experiencing a strong intensification in extreme precipitation. Both ICON and IFS also show considerably more spatial detail than CMIP6, in particular along coastlines and mountain ranges, and, in some cases, even capture the influence of large rivers on change signals. 

Our results clearly demonstrate the potential of km-scale models for resolving sharp gradients in change signals, but also reveal remaining shortcomings of this new model generation. In this first analysis, we, hence, find no robust evidence that changes in daily extremes are consistently different between CMIP6 and km-scale models, but our results highlight that more and longer model experiments are needed to robustly quantify extremes in this new generation of models. These findings are particularly relevant as km-scale models are envisioned to serve as the basis for Digital Twins of Earth, which, in turn, are supposed to inform impact assessments and support mitigation and adaptation decisions.