Characterising the range and outliers in CMIP6 multi-model climate projections of extremes

Abstract Climate projections made using ensembles of model simulations typically present spatial information using ensemble-average changes. However, quantifying and understanding spread in model projections, along with building a picture of different possible climate futures across variables, is critical for meeting decision-makers’ needs. Here we explore projected ensemble ranges in mean and extreme temperature and precipitation metrics and their relationships using simulations from the sixth Coupled Model Intercomparison Project. The range is characterised by ensemble minimums and maximums, along with intermediate percentiles, and heatmaps display the contribution of individual models to the overall global pattern of changes at these points in the range. Furthermore, we examine if the framing of changes, both in terms of a future time slice (2070–2099) and a future global warming level (GWL) of 3 K, relative to a historical period (1850–1900), affects interpretation of results. The 2070–2099 time slice exhibits known patterns of change such as Arctic warming, and drying and wetting regions, exacerbated according to the ensemble percentile considered. For extreme temperature metrics, while 25th, median, and 75th percentile changes are relatively evenly contributed to by all models, ensemble-minimum and -maximum changes are instead dominated by only one or two models. In particular, the CanESM5 dominates ensemble-maximum changes by over 40%. Furthermore, the NorESM2-LM dominates ensemble-minimum changes for the coldest night of the year (64%) and contributes largely (23%) to ensemble-maximum changes for the warmest day of the year—which are opposite ends of the most extreme measures we consider. Moreover, for the 3 K GWL, the CanESM5 domination ceases yet the NorESM2-LM domination persists. This has implications for the expression of ensemble ranges in multi-model projections, depending on the framing of the future period, and for the interpretation of what constitutes an outlier model. Thus, we encourage multi-model studies to understand and communicate ensemble spread in their climate projections.