Ensembles of global ocean analyses for seasonal climate prediction: impact of temperature assimilation


*Corresponding author.
e-mail: rogel@cerfacs.fr

†Now at Centre de Recherches Halieutiques Mediterranéenne et tropicale, Sète, France.


Two sets of ensembles of ocean initial conditions for seasonal climate prediction have been constructed following a strategy that preserves the dynamical balance of the ocean. One set has been constructed by perturbing the wind stress and sea surface temperature forcing of an ocean model. The other set is constructed by assimilating perturbed in situ temperature data using a fully multivariate three-dimensional variational system, in addition to using the perturbed forcing. Ensemble statistics are computed for the period 1990–1999.

The ensemble mean temperature has a significant subsurface cold bias when no data are assimilated, indicating that some part of the DEMETER hindcast systematic error may be explained by the way the ensembles are generated. Constraining the temperature through data assimilation contributes to a reduction in this subsurface bias.

The dispersion of the initial conditions is supposed to sample the uncertainty associated with the ocean initial state estimate. Therefore, a smaller spread in the assimilation case than in the forced-only case is consistent with the idea that ocean uncertainty has been reduced by constraining the ocean state through data assimilation. However, comparing ensemble spread and innovation statistics suggests that ocean uncertainty may be underestimated by the ensemble method in the assimilated case. It is also shown that wind stress perturbations mainly control the spread amplitude in the non-assimilated case, but that perturbed subsurface temperature observations have a stronger relative impact in the assimilation case.