Variational Data Assimilation for the Global Ocean - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 334

easily. Existing variational ocean data assimilation technology and capabilities are
not lost. Second, when ensemble variances are imperfect the optimal error variance
estimate is a linear combination of a climatological covariance and an ensemble
covariance. The superiority of hybrids over conventional ensemble assimilation
schemes is particularly marked when the ensemble size is small or the model error
is large.
A static 4D ensemble covariance data base will be computed from an ensemble
of mesoscale anomalies using the long term integration of global HYCOM in
the 1993-2009 reanalysis product, which includes NCODA 3DVAR assimilation.
Covariances calculated in this way have clear physical meanings and represent 4D
model climate flow dependence and model variable interactions. Existing 3DVAR
initial covariances will be extended to 4D by assuming that the error covariances
between variables are a separable function of space and time. The computational
overhead of imparting this 4D aspect to the 3DVAR covariances is expected to be
very small. The 4D extension of the NCODA covariances will then be linearly
combined with the 4D localized HYCOM static ensemble covariances forming
a fully 4D hybrid data assimilation scheme. Optimum values for weighting the
ensemble and extended 3DVAR covariances in the hybrid are determined from
model statistics.

13.8 Summary

This paper describes the development, implementation, and validation of a new
oceanographic 3DVAR assimilation system. The system is unified and flexible
and a key component of many Navy ocean and atmosphere applications. It is
run globally or regionally, where it can be applied to nested, successfully higher-
resolution grids, providing analyses on a range of scales. NCODA 3DVAR provides
the assimilation component for both ocean and wave model prediction systems as
well as multiple atmospheric prediction systems, where it is used to provide sea
ice and SST lower boundary conditions. It assimilates a wide range of ocean data
types and it contains numerous diagnostic features for assessing and tuning the
statistics needed for the assimilation as well as quality control. The background error
covariance formulation permits considerable anisotropy with adaptive horizontal
and vertical length scales and error variances that vary with location and evolve with
time. It is shown to be efficient for very large scale, high resolution global ocean
model grids, assimilating millions of observations a day. The intelligent, adaptive
data thinning algorithm permits all sources of the high density surface data types to
be assimilated with minimal loss of information. The parallel implementation has
minimal communication overhead, with granularity of the code (important for load
balancing) easily controlled by the number and size of the observation data blocks.
The NCODA 3DVAR system is operational at the Navy oceanographic production
centers and is in the final phase of pre-operational testing as the data assimilation
component for the global HYCOM forecasting system.