[Dart-users] latest updates to the subversion repository

[Nancy Collins]

dear dart-users,

see below for brief notes about the latest updates to the
dart subversion repository.    all dart users should check out
items #1-7.  WRF/DART users should also see items #8-11,
and AM2/DART users should see #12.

thanks -
nancy

-----

#1.  tim has committed a program to convert our obs sequence
file format into pretty standard netcdf, for ease in post-processing.
note that this will do all obs types, but it will *NOT* maintain
any additional metadata in the obs, such as the gps or radar obs.
but for plotting and diagnostics, this should be a much simpler
format for people to handle.   these programs are under the
$DART/observations/utilities/threed_sphere directory.
there's a quickbuild.csh script, and some nice matlab plot
routines which understand the new format, including ones
for plotting wind obs.

#2. ryan has contributed updated MADIS and SSEC converter
programs, and i've updated the NCEP converters.  see the
README in the new $DART/observations/obs_errors dir for
a slick way to change the default obs errors when generating
new obs_seq files.

#3. it was possible for utility programs (like trans_sv_pv or
wrf_to_dart) to call the restart file read and write routines without
having called the initialization code, which among other things reads
in the namelist and sets the default for binary or ascii restart files.
the code now keeps track of whether the init routines have been called,
calls init if not yet called, and therefore honors the namelist settings.
IF YOU ARE DOING PRODUCTION RUNS, please check your
setting of: &assim_model_nml :: write_binary_restart_files.
for performance and numerical accuracy, it should be .true.
for debugging, it can now be .false., so you can examine your
restart files with a text editor.   note you can explicitly set binary
as the restart file format in the code, if you are writing translation
routines where you always want to use binary.

#4.  i've just committed a pervasive change which makes the default
behavior be to only echo the namelists to the .nml file and NOT
to the standard output of the job.  several of our namelists are
now long character arrays and they overwhelm the output so
you cannot find useful status and error messages.  there is a new
item in the utility_nml namelist if you want to change where
the namelists are written.   note that you will need to update
not only your model directory but the utilities mod.  best approach
here is to update your entire svn tree.  any local mods WILL BE
preserved.

#5.  i've finally removed the obsolete merge_obs_sequence source and
documentation.   all models should now build obs_sequence_tool,
which has many more features and bug fixes. 

#6.  one of our older compilers gives an error if tab characters are
found in the source files.  it seems like any white space should be
white space, but...   so several files which seem to have no real
changes have been committed solely to remove tabs.

#7.  the quickbuild.csh scripts now remove the auto-generated
input.nml.*_default files, as well as all the .o and .mod files after
everything is built.  it should be easier now to see what executables
are actually there.  if you want to keep any of these, feel free to
comment out or edit the remove (rm) lines.

wrf specific:

#8. i've committed an updated version of the wrf model_mod which
is patched to be faster.  josh intentionally left a slow routine in the code
as an incentive to finish the model_mod overhaul, but for now, ryan
has contributed a patch which makes the code faster.  josh, we promise
not to let it linger in this condition.  people who have remarked to me
that the newest model_mod seems slower should recheck and let me
know if the performance is closer to what they expect.

#9. if you did not have an 'namelist.input' file in the current directory,
or you had an empty one, you might have gotten a mysterious error
message from the netcdf routines about trying to reuse the unlimited
dimension.  this latest update fixes that error.

#10.  if any code tries to call either the model advance subroutine or
the initial conditions routine inside the model_mod, they will now
error out instead of silently returning without doing anything.

#11. if you selected the 'write diagnostics out in state vector form'
instead of divided up into variables, you might have gotten an error
when the netcdf routines tried to call enddef mode twice.  this is fixed.
thanks ryan.

note that if you get a new wrf model_mod, you also need to update the
rest of your dart tree because the new handling for the namelist output.

AM2 specific:

#12.  the updated model_mod with height interpolation and wrap has been
committed, and the updated trans routines as well.  trans*pert0 and time0
have been mostly commented out, since they don't appear to be compatible
with the updated model mod, so my guess is they aren't being used.   if 
called,
they will simply error out.  i also added the new nml code while i was 
there.