388
PIERRE BAHUREL
restoring term). The model equivalent to this SST is the temperature
at the first level of the model (6m for the 1/3° North Atlantic
system). The Reynolds product is used with a non-Gaussian error so
as not to damp the model’s meso-scale features but rather to correct
locally the model’s large drifts. As a consequence, only innovations
larger than 2V (V~0.6°C r.m.s) are taken into account in the
analysis. The model state is updated by the sum of the contribution
of each selected EOF to the gain multiplied by the innovation. This
scheme is fully multivariate, since the covariances between the
errors of different variables are taken into account in the EOFs
calculation. The effect is that the assimilation of a single
observation has an impact on all the variables in the state vector
through the multivariate statistics.
2.3.2
SAM-2
The next generation of multivariate assimilation system, referred as
SAM-2, is being developed from Reduced Order Kalman Filters using 3D
multivariate modal decomposition of the forecast error covariance (See
Brasseur [2005] for a complete description of these methods). The use of 3D
modal representation for the error statistics is intended to overcome some of
the limitations of SAM1v2 in highly inhomogeneous, anisotropic, and non
separable regions of the world ocean such as shallow areas, as well as in the
surface layer. SAM-2 includes two versions differing by the formulation of
the analysis kernel. The final scheme derived from the SEEK (Singular
Evolutive Extended Kalman) algorithm [Pham et al., 1998] (LEGI,
Grenoble) has an inversion in the modal space, whereas the intermediate
kernel has an inversion in the observation space.
2.3.3
SAM-3
Advanced data assimilation techniques, such as the variational approach,
are also investigated in MERCATOR. The variational approach is based on
the minimization of the model-observation square misfit over a time period
with their given a priori errors. The solution given is the closest trajectory to
the observations dynamically consistent with the model equations in a least
square mean sense. At given time, the solution is constrained by past and
future observations available on the assimilation window. This technique
will be used first with the coarse resolution (~2°) global ocean configuration,
ORCA2, to assimilate both in-situ and altimetric data.
2.4
Products dissemination
Dissemination of MERCATOR products is made through www and FTP
automated tools, both for real-time and archived products. Dissemination
PIERRE BAHUREL
restoring term). The model equivalent to this SST is the temperature
at the first level of the model (6m for the 1/3° North Atlantic
system). The Reynolds product is used with a non-Gaussian error so
as not to damp the model’s meso-scale features but rather to correct
locally the model’s large drifts. As a consequence, only innovations
larger than 2V (V~0.6°C r.m.s) are taken into account in the
analysis. The model state is updated by the sum of the contribution
of each selected EOF to the gain multiplied by the innovation. This
scheme is fully multivariate, since the covariances between the
errors of different variables are taken into account in the EOFs
calculation. The effect is that the assimilation of a single
observation has an impact on all the variables in the state vector
through the multivariate statistics.
2.3.2
SAM-2
The next generation of multivariate assimilation system, referred as
SAM-2, is being developed from Reduced Order Kalman Filters using 3D
multivariate modal decomposition of the forecast error covariance (See
Brasseur [2005] for a complete description of these methods). The use of 3D
modal representation for the error statistics is intended to overcome some of
the limitations of SAM1v2 in highly inhomogeneous, anisotropic, and non
separable regions of the world ocean such as shallow areas, as well as in the
surface layer. SAM-2 includes two versions differing by the formulation of
the analysis kernel. The final scheme derived from the SEEK (Singular
Evolutive Extended Kalman) algorithm [Pham et al., 1998] (LEGI,
Grenoble) has an inversion in the modal space, whereas the intermediate
kernel has an inversion in the observation space.
2.3.3
SAM-3
Advanced data assimilation techniques, such as the variational approach,
are also investigated in MERCATOR. The variational approach is based on
the minimization of the model-observation square misfit over a time period
with their given a priori errors. The solution given is the closest trajectory to
the observations dynamically consistent with the model equations in a least
square mean sense. At given time, the solution is constrained by past and
future observations available on the assimilation window. This technique
will be used first with the coarse resolution (~2°) global ocean configuration,
ORCA2, to assimilate both in-situ and altimetric data.
2.4
Products dissemination
Dissemination of MERCATOR products is made through www and FTP
automated tools, both for real-time and archived products. Dissemination