Detecting multidecadal predictability of the global ocean using a linear inverse model and its optimal initial condition

The decadal predictability of the global ocean in a coupled climate model (GFDL-CM2.1) is explored using a linear inverse modelling (LIM) approach. In this study, a cost-efficient method is introduced, in which a linear propagator matrix is constructed to fit best the principal components of the state vectors. By taking a two-step reduction of the state vector dimensionality, this method enables us to avoid the massive calculation while maintaining the essential features of the global temperature and salinity distribution. We focus on a global mode that grows the most in 25 years, since the Principal Oscillation Pattern (POP) analysis on the linear propagator suggests that only a few POPs amplify longer than 25 years. It is found that the maximum amplification occurs in the Southern Ocean near the Weddell Sea and the Southern Indian Ocean near the bottom following the local growth at the surface. The initial condition optimized for this growth is concentrated at the surface layers in the Tropical Pacific. In order to examine the robustness of this relationship, initial and final norm kernels are applied to isolate initial conditions and final states in various geographical locations.