From predictions to projections: A large ensemble of initialised predictions for the end of the century

In the past the fields of initialised decadal climate predictions and uninitialised climate projections have emerged as separated areas of research. They are used to provide important information on the future development for stakeholders on time scales from a few years to centuries. Furthermore, with so-called SMILEs (single model initial-condition large ensembles) the community provides ensembles in the hundreds of members allowing a better estimation of developments of extremes and uncertainties.

With our new experiment we combine all these research fields to learn more about the future and about our model systems. In this study we have extended all members of the decadal prediction system with the MPI-ESM-LR, initialised between 1960 and 2024, to the end of 2100. Besides using it as a prediction system for up to 40 lead years, we can view this system as a large ensemble. The 1040 ensemble members does not only allow us to better estimate the trends of extremes, but show us also some extremes, which are up to now hardly seen with large ensembles.

In our analysis we focus on summer temperature extremes in Central Europe at the end of the century. We show how extremes are changing over time and what advantages a large model with more than thousand members has compared to a smaller uninitialised large ensemble created with the same model. Furthermore, we show the effect of initialisation within the model on variables like surface temperature and AMOC and investigate how long we can detect the initialisation compared to an uninitialised model. Finally, we will in light of these findings discuss the consequences we can draw for how we do and interpret initialised predictions and uninitialised projections in our community.