Quantifying Observation Impact for a Limited Area Atmospheric Forecast Model - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 129

It had also been used for the global DA system from October 2003 until October
2009, when it was extended to a four-dimensional variational system. NAVDAS
uses an incremental update cycle to create the initial conditions for generating
new model forecasts every 6 or 12 h. In the regional application, NAVDAS
uses conventional and aircraft observations, geostationary satellite winds, Special
Senor Microwave/Imager (SSM/I) winds speeds, satellite total precipitable water
retrievals, scatterometer and passive microwave derived surface marine winds and
satellite temperature retrievals. In addition, synthetic observations are used to define
the wind and thermal structure of tropical cyclones. Satellite sounding radiances
are currently under testing to replace the satellite temperature retrievals in near
future. NAVDAS includes a geostrophic balance constraint and uniform analysis
length scale.
The preprocessing and quality control software for the different observation types
is built into NAVDAS. Innovations are first computed by interpolating the 6 or 12 h
background forecast to the observation locations and then subtracting the result
from the observations. A complex quality control ( Collins and Gandin 1990 ; Gandin
et al. 1993 ) is used for checking the rawinsonde observations. For aircraft data, the
quality control includes sophisticated flight track checking and characteristic error
detection. Vertical profiles of temperature and wind from the surface to 400 mb
are created from the aircraft accents during takeoff for more efficient handling
by the analysis algorithm. Satellite feature tracked winds from the geostationary
satellites and polar-orbiting satellites, surface marine winds from scatterometer and
WindSat instruments, and wind speed from SSM/I are all checked and averaged
to create lower density “superobs” for each instrument type ( Pauley 2003 ). The
feature track satellite winds, scatterometer and WindSat winds are averaged in 1 ı
by 1 ı boxes, while wind speed from SSM/I, total-precipitable water estimates from
several instruments, and temperature retrievals are averaged in 2 ı by 2 ı boxes. The
total-precipitable water superobs are used to generate vertical retrievals of pseudo
relative humidity (ratio of observed mixing ratio to the saturation mixing ratio of
the background).
NAVDAS uses Eq. 6.4 to update the background field (although the observation
operator is not always linear as in Eq. 6.4 ). A preconditioned conjugate gradient
algorithm is used to invert HP b H T . C R / followed by a post-multiplication

step to obtain the analysis corrections at COAMPS grid points. For

increased efficiency, vertical profiles of observations from any single instrument are
transformed into coefficients of the eigenvectors of the background error correlation.
The number of vertical modes used in the analysis can depend on the type of
instrument used for the observation; ten modes are used for the low vertical
resolution of satellite temperature retrievals and a reduced set of modes can be
used for satellite humidity retrievals. Predefined functions are used to define the
horizontal and vertical variation of the background error covariance ( Daley and
Barker 2001 ). The horizontal and vertical length scales of the background error
correlations may also vary with any combination of height, horizontal location
and grid resolution. The covariances for the wind are based on those for the
streamfunction and velocity potential ( Daley 1991 ). For mesoscale analysis, a

. x a x b /