November 6, 1997

MEMORANDUM TO:  Record
FROM:           Dennis A. Keyser -- W/NP22
Subject:        Changes to OI/SSI-Data Preprocessor
                (May 1, 1997 Version -- UPDATE # 1)

The May 1, 1997 version of the data preprocessor program
(PREPDATA) was updated for the first time.  UPDATE #1 was
implemented on 1200 UTC 20 May 1997.  All changes now affect only
OBUFR = TRUE switch.  Except where noted, the changes below
affect both the GLOBAL and RGL/ETA models. PREPDATA now
exclusively uses JBUFR file input.

*********************  U P D A T E   # 1  ***********************

I.  GENERAL CHANGES

      1) The arrays NOBS, DAT, and TMP were expanded to hold four
more words on each level, namely the reason codes for pressure,
temperature, moisture and wind.  For NOBS, the parameter MXWRDL
was increased from 7 to 11.

      2) Reason codes are now stored for certain situations (see
II.A.1 below).

      3) All references to the NAS/HDS were removed.

      4) The standard output print for reports was expanded to
include reason codes on each level (GETUPA, SMERGE, FILLX,
SATBFR, GOESDG).
     
      5) In the processing of winds-by-height in subroutine
WNDBYZ (for upper-air data), reports are always sorted by height
in order to insert the category 4 levels into the report
properly.  It was always assumed the height increased as pressure
decreased, so a sort by height would not cause pressure to no
longer monotonically decrease with height.  This may not now be
the case, because raob reports no longer undergo any form of cqc
prior to PREPDATA.  Reports with erroneous heights may violate
this.  In order to assure that all reports in the PREPBUFR file
always have pressure that decreases monotonically, a final
pressure sort is now made in WNDBYZ after the category 4
processing.  This will, for cases of "bad" data, result in some
reports with heights that no longer monotonically increase with
decreasing pressure.  The feeling is that CQCBUFR will correct
these heights, or at least flag them as bad.



      6) In testing the change above, the new executable aborted
at the beginning of the processing of ADPSFC data, in W3LIB
routine w3fs22.  This abort had nothing to do with the change
above, but rather was the result of an incompatibility in the
PREPDATA code which was causing memory to be clobbered. 
Apparently this incompatibility was present for some time in the
code, but was never in a place that caused any problem when the
program ran until now.  The position of the clobbered memory must
have changed to a "critical" location when PREPDATA was
recompiled with updated versions of both the W3LIB and the
BUFRLIB.  The PREPDATA source was carefully examined, and some
"possible" problems related to equivalences between variables in
COMMON and longer non-COMMON variables were discovered.  The
source code was modified to remove this and the new executable
then ran fine.  Hopefully this memory clobbering problem has been
corrected, not just shifted to a non-critical memory location by
the fixes to the source code. 

       7) The original subroutine STOROB was removed because it
is no longer called.  The previous subroutine STORXX was renamed
to STOROB.

       8) Reason code are now encoded into the output PREPBUFR
file if they are non-missing (W3FIZZ, FIZZ01).


II.  OPERATIONAL CHANGES

A. CHANGES IN THE PROCESSING OF UPPER-AIR DATA

      1) Reason codes 100, 101-103 and 105 can now be stored with
report levels for all variables.  (See Reason Code documentation
for their meaning.)  Currently reason codes 101-103 and 105 can
only be placed on radiosonde, pibal, dropwinsonde, multi-level
reconnaissance, and wind profiler data.  The default reason code
of 100 will appear in all cases for all other data types. 

       2) The ON29 category value for "special levels" was
changed from 6 to 7 to avoid confusion with the Cat. 6 aircraft
and satwnd data levels.

       3) Tropopause levels in the PREPBUFR file are now ALWAYS
assigned ON29 category 5.  This was not the case before.

       4) Levels in the PREPBUFR file are now assigned the proper
ON29 category.




       5) The baseline check has been turned off (Namelist switch
IDELEV set to 99999 in data cards).  The downstream program
CQCBUFR already performs a more sophisticated baseline check, and
in fact, the PREPDATA baseline check may have led to erroneous
corrections later by CQCBUFR.