Model-consistent ocean data assimilation for seasonal to decadal climate prediction

In Earth's climate system, the slowly varying ocean represents an important source of memory for predictions on the seasonal to decadal time scale. The ocean picks up atmospheric variability on a broad range of scales and feeds back on the large-scale atmospheric circulation. While today’s comprehensive Earth system models (ESMs) used in climate prediction are able to simulate this atmosphere-ocean feedback in a broad sense, data assimilation - which brings the climate model close to the observed state – allows the use of ESMs for climate predictions. We propose that the quality of climate predictions can be improved by initializing the ESMs using a model-consistent assimilation of observations resulting in (1) an initialization of the ESM with a model state close to the observed one, while (2) minimizing a potential initialization shock resulting from a mismatch between the simulated climate state and observations.
Here we demonstrate our approach towards a model-consistent assimilation of two ESMs used in climate prediction at Universität Hamburg and Deutscher Wetterdienst: MPI-ESM and ICON-ESM. Central to our approach is a weakly coupled assimilation setup, consisting of an Ensemble Kalman filter assimilation scheme in the ocean component (MPI-ESM, ICON-ESM) and a nudging assimilation scheme in the atmospheric component (MPI-ESM). We show that our approach facilitates a large part of atmosphere-ocean interaction already within the assimilation, allowing for a quick adaption of the assimilation in case of unrealistic behaviour of key processes. For two key large-scale oceanic processes, Atlantic meridional overturning circulation and oceanic Rossby waves, we analyze how sensitive they are to the degree of atmosphere-ocean interaction allowed for during assimilation and what this implies for the respective climate predictions.