Initializing climate predictions using climate states from an atmosphere- ocean coupled assimilation system

Initialized climate predictions are designed to align model simulated climate variability with those of observations and also aim to correct for forced model response. Significant efforts have been made in developing these climate prediction systems during the recent years with some success in predicting certain aspects of climate on annual to multi-annual timescales. However, the prediction skill on decadal timescales remains limited. Several issues have been identified with most prominent being initial shock due to different mean states of the observational data (i.e.  observationally constrained assimilations) and the model, resulting in climate drift towards the model's own attractor usually after a few months of initialization.

In this study we present results from a new initialization approach, in which the assimilation is generated by nudging both the ocean and atmospheric component of the model towards observed SST anomalies and sea level pressure respectively using the coupled model EC-Earth3. The initial evaluations suggest that the coupled ocean-atmosphere nudging results in assimilated atmospheric and ocean states that correlates better with observations both over ocean and land regions compared to ocean only nudging. The combined nudging also improves the representation of the North Atlantic Oscillation (NAO) in the assimilated data. Further assessment of different climate components (such as sea ice extent and volume) of the assimilations are ongoing. In this work we will present evaluations carried out for these two assimilations (i.e. from ocean only and coupled ocean-atmosphere nudging) and preliminary assessment of the skill of decadal predictions initialized from the combined assimilations. Furthermore we investigate the impact of the length of nudging to generate the initial state on the prediction skill on annual to decadal time scales.