Global Warming Levels as a basis for merging different generations of CMIP and CORDEX ensembles

The development of new generations of climate projections for use in the IPCC assessments is a core part of climate science.  But while new datasets are always welcome, they also present challenges for the operationalisation of climate services. For example, how should the upcoming CORDEX-CMIP6 ensemble be incorporated into an existing climate service based on CORDEX-CMIP5? Should there be a 1:1 replacement of the older data? At what point is the new ensemble sufficiently large to justify the switch? And what do we do with the old dataset? Ideally we would blend the two ensembles into one larger super-ensemble, but this approach is hampered by the use of different generations of emissions scenarios (e.g. RCPs vs SSPs). Here we introduce an approach for blending ensembles from different generations, with differing emissions scenarios, based on global warming levels. We borrow the approaches employed in the statistical attribution community, where a local response variable (e.g. frequency of local extreme temperatures or precipitation) is modelled as a function of the global temperature change: the use of global warming levels is particularly relevant here, as it removes the effect of differing emissions-scenarios. We extend this approach further through the use of a mixed-effects modelling framework, where each individual model is considered as a deviation from a fundamental underlying response. We show using CMIP ensembles that our approach is able to reproduce the characteristics of a single ensemble, to merge multiple ensembles together, and to adequately predict “out-of-sample” ensembles that the model has not been trained against. The use of a statistical framework also allows statistical inference and testing to be performed, and we show how cases where the approach breaks down can be automatically highlighted. Finally, we show how the approach can be applied in a climate-service context, and propose a scheme that blends the existing CORDEX-CMIP5 ensemble with new members from the CORDEX-CMIP6 family to produce a climate service that uses both datasets to the fullest extent.