Model bias reduction and the limits of oceanic decadal predictability: importance of the deep ocean

Florian Sévellec and Alexey V. Fedorov

Ocean General Circulation Models (GCM), as part of comprehensive climate models, are extensively used for experimental decadal climate prediction. Understanding the limits of decadal ocean predictability is critical for making progress in these efforts. However, when forced with observed fields at the surface, ocean models develop biases in temperature and salinity. Here, we ask two complimentary questions related both to decadal prediction and model bias: (1) Can we temporarily reduce the bias and improve prediction by perturbing the initial conditions? (2) How fast will such initial perturbations grow? To answer these questions, we use a realistic ocean GCM and compute temperature and salinity perturbations that reduce the model bias most efficiently during a given time interval. We find that to reduce this bias, especially pronounced in the upper ocean above 1000~m, initial perturbations should be imposed in the deep ocean (specifically, in the Southern Ocean). Over 14 years, a ~0.1 K perturbation in the deep ocean can induce a temperature anomaly of several degrees in the upper ocean, partially reducing the bias. A corollary of these results is that small initialization errors in the deep ocean can produce large errors in the upper-ocean temperature on decadal timescales, which we interpret as a decadal predictability barrier associated with ocean dynamics. To study the mechanisms of the perturbation growth, we formulate an idealized model describing temperature anomalies in the Southern Ocean. Our results indicate that the strong mean meridional temperature gradient enhances the sensitivity of the upper ocean to deep-ocean perturbations in this region through nonnormal dynamics inducing pronounced stationary-wave patterns.