A variational data assimilation system for the global ocean

Anthony T. Weaver
CERFACS
Toulouse, France

CERFACS is leading the development of a variational data assimilation system for the OPA ocean general circulation model. The system (OPA_VAR) is designed to produce global ocean analyses through the assimilation of in situ (temperature and salinity), sea surface temperature and altimeter observations. This work is part of a collaboration between CERFACS, ECMWF and LODYC (Paris), and is a contribution to the European ENACT project, to develop advanced data assimilation methods for climate research and climate forecasting.

Some of the key features of the current OPA_VAR system will be described. A particularly important feature is the availability of incremental versions of both three- and four-dimensional variational assimilation (3D-Var and 4D-Var). Maintaining a well-tuned 3D-Var is argued to be crucial for a number of scientific and practical reasons, and ultimately beneficial to 4D-Var since most of the system components are shared by the two algorithms.

The main scientific component of the variational algorithm is embodied in a sequence of transformations from a vector of mutually uncorrelated model control variables to a vector of variables that represent the model estimate of the observed quantities. The transformation sequence involves: 1) smoothing via a square-root of a univariate background-error covariance operator; 2) application of a static balance operator; 3) integration of the OGCM to the measurement times of the observations; and 4) application of an observation (interpolation) operator. The principal difference between the incremental 3D-Var and 4D-Var algorithms is in the linearized counterpart of Step 3), which is used to describe how increments at initial time evolve within the assimilation window. In 3D-Var the approximation is made that changes in the model initial state persist for the duration of the assimilation window, while in 4D-Var changes at initial time are translated to changes at later times in the window by a dynamical model based on the tangent-linear of the OGCM.

As part of ENACT, sets of global 3D-Var and 4D-Var analyses are being produced for the periods 1987-2001 and 1963-2001. Several of the experiments have been designed to test the sensitivity of the analyses to different components and formulations of the system. Some of the key findings from these experiments will be presented.