On the robustness of near term climate predictability regarding initial state uncertainties

A set of four ensemble simulations has been designed to assess the relative importance of atmospheric, oceanic, and deep ocean initial state uncertainties, as represented by spatial white noise perturbations, on seasonal to decadal prediction skills in a perfect model framework. It is found that a perturbation mimicking random oceanic uncertainties have the same impact as an atmospheric-only perturbation on the future evolution of the ensemble after the first 3 months, even if they are initially only located in the deep ocean. This is due to the fast (1 month) perturbation of the atmospheric component regardless of the initial ensemble generation strategy. The divergence of the ensemble upper-ocean characteristics is then mainly induced by ocean–atmosphere interactions. While the seasonally varying mixed layer depth allows the penetration of the different signals in the thermocline in the mid-high latitudes, the rapid adjustment of the thermocline to wind anomalies followed by Kelvin and Rossby waves adjustment dominates the growth of the ensemble spread in the tropics. These mechanisms result in similar ensemble distribution characteristics for the four ensembles design strategy at the interannual timescale.