Analyzing centennial variability in the Southern Ocean using data assimilation

The agreement between simulated and reconstructed multidecadal to centennial climate variability has improved over the past decades. However, significant disagreements still exist, especially at regional scale. In the Southern Ocean, both reconstructions and climate models display large variability at those timescales but models fail in reproducing some key elements such as the centennial variability in the strength of the westerly winds inferred from various types of proxy data. Data assimilation combines in an optimal way information from proxy data and climate models. It can help in identifying the cause of such model-data mismatch by improving the reconstructions as well as by testing the compatibility of those reconstructions with model physics or between different types of proxy data. Two examples will be discussed here. The first one focuses on the shift in the westerly winds between the 14^th and 16^th century, showing that it is clear in reconstructions based on classical statistical methods and on data assimilation but it is not simulated in models without data assimilation. In the second example, we will discuss the deep ocean convection and open ocean formation in the Southern Ocean that induce large multi-decadal to centennial variability in some global models while it is totally absent in many others. We will check how data assimilation can be used to test the validity of the simulations and to determine which model behavior is the most realistic.