Quantifying Observation Impact for a Limited Area Atmospheric Forecast Model - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 130

horizontal length scale of 385 km is currently used for the autocovariance of
the streamfunction. The geopotential/wind background error covariances are then
related by the geostrophic assumption. A latitude dependent coupling parameter
(smaller in tropics compared to poles) is used to control the degree of geostrophic
coupling between the streamfunction of the wind and the geopotential.

6.2.3 COAMPS Adjoint

In addition to the dynamical core, the COAMPS tangent linear and adjoint
models include all of the respective components of the nonlinear model's physical
parameterizations, with the exception of the radiation parameterization and the
cumulus scheme (a less complex option is available in the tangent linear and adjoint
models). More information on the adjoint COAMPS atmospheric model can be
found in Amerault et al. ( 2008 ). The convective and moist parameterization adjoint
operators are not robust for the length of the adjoint integrations (12 h) due to
discontinuities and nonlinearities inherent in the corresponding nonlinear schemes.
Therefore, current results are obtained with a “dry” adjoint model (although the
nonlinear forecasts are run with all available physics options). The observation
impacts calculated with a dry adjoint model will not capture all of the information
in the nonlinear model's error. In practice, the estimate of

e f

in analysis or
observation space is roughly 80-90 % of the actual value (similar to global systems).
Furthermore, COAMPS can be configured with multiple nests where the horizontal
grid spacing is less than 10 km. The adjoint model can also be run with nests, but this
option is computationally expensive due to the extra trajectories and not currently
configured for the observation impact system. Therefore, all nonlinear forecasts and
adjoint integrations are performed on a single domain for this work.

6.2.4 NAVDAS Adjoint

The adjoint NAVDAS adjoint operator is performed by reordering the operations
of NAVDAS (Sect. 6.1.2 ). The adjoint operator does not involve any of the code
responsible for quality control of the observations or creating the innovation vector.
Therefore, the adjoint of NAVDAS for COAMPS was able to incorporate many
of the components of the global system ( Langland and Baker 2004 ). However,
NAVDAS and COAMPS differ in their state variables and vertical coordinate.
Additional code was needed to transfer the gradient field produced by the COAMPS
adjoint model into NAVDAS analysis space.
For example, NAVDAS creates analyses of pseudo relative humidity which are
converted and output as fields of dewpoint depression to be read by COAMPS. One
more conversion to mixing ratio takes place before the COAMPS integration begins.
In the adjoint observation impact system there is corresponding code to transfer the