The radar wind 3dvar has been changed to simplify and improve the
analysis of the radar data.
The earlier version used a linear fit of the model wind profile in
determining an estimate of the maximum and minimum wind in the profile.
This was done to simplify the gradient computation of the forward model,
but has the disadvantage that for deep layers (broad radar beam far from
the radar), internal maxima will be missed by the linear fit. The new
version eliminates fitting of the profile and just obtains the actual
maximum and minimum profile values and the corresponding eta level index
of maximum/minimum. This actually leads to a much simpler gradient for
the forward model and significantly improved fits of both guess and
analysis to the observations.
Two code changes have been incorporated related to bugs
found in anisotropic testing by Manuel Pondeca. These bugs don't
influence the current isotropic mode, but it was important to fix them
anyway so they don't creep back in later.
A diagnostic output file has been added to the radar wind 3dvar for use
by Ying Lin to make various graphic displays of the radar data.
The radiance processing code has been upgraded to the final global
analysis version which is about to be implemented in operations. Key
additions include a new land surface emissivity calculation, which
allows for use of more observations over land, and the turning on of
NOAA 16 HIRS data.
Since the new radiance processing has the new land surface emissivity
calculation, GOES moisture channels over land are usable. Therefore, assimilation
of GOES precipititble water data over land has been turned off.
Cycled EDAS restarted from ETAV control.
Fixed 3dvar code errors so that the new radiance channels would
be correctly processed. Bias corrections were spun up from
zero and the Cycled EDAS restarted from ETAV control run.
||Keyser, Parrish, Rogers
Modified 3dvar code so that it would use GOES radiances over
land. Added prepbufr generation job to the parallel with modified
parameter cards so that over-land GOES radiances would be turned on
in the prepbifr file.
A small bug was uncovered in the test 3dvar with new radiance
processing. An index was off by one, having a small impact on two of
the scan-angle dependent radiance bias correctors. It has been
corrected in all parallels currently testing the new 3dvar. This
problem is not present in the operational code.
Turned on use overland GOES precipitable water retrieval data
in the eta 3DVAR analysis.
Turned off use of GOES/TOVS radiances over land
Since the parallel is no longer using overland GOES radiances,
job reverts to using the production EDAS/Eta prepbufr files.
In order to successfully process radiance data for the regional
reanalysis, it was necessary to be more aggressive with the evolution of
the bias correction. The changes to the radiance processing make the
3dvar more robust, especially when starting a new experiment, when the
bias correction is unknown, and some spinup was required. The new
version does a spinup automatically every analysis, even though this is
redundant most of the time.
The principle change is as follows: the radiance processing subroutine
is iterated 5 times, with observation error and QC limits mutiplied by
factors of 5,4,3,2,1 for each iteration. The bias correction is
estimated after each pass without going through the entire analysis.
This correction is blended with a running average of the input bias
correction, with a 1/e time of 15 days. This becomes the new guess bias
correction. The analysis bias correction is then generated the same as
before. There is also a fixed bias file, which used to be the file
eta_satangbias_16_new.txt. This file is no longer used. The fixed bias
numbers are now allowed to evolve with a 1/e time of 30 days. They are
running means of a mean lapse rate of the background for each channel,
and mean difference of obs - guess brightness temperature for each scan
angle of each channel.
The format of the files "satbias_in" and "satbias_out" has been changed
to contain all the information necessary for doing running means of the
bias coefficients and the angle dependent biases. Each channel is
processed independently, and there is a "model id" which is a unique
number generated from resolution and position parameters, so that if the
model is changed, the file will be reset to zero automatically by the
The cycled EDAS was restarted from the 32-km control run (ETAV)