Model Changes
                                 ----- -------


In theory, the model used by reanalysis should not change.  However, some
changes are unavoidable.  For example, the model periodically needs to be 
recompiled whenever the operating system, compiler or even computer is changed 
(the project started with a Cray Y-MP and will later use a Cray J-16).  The
latest recompiled system (may 95) produced results that were different from 
the previous version.  These changes were attributed to small random 
differences caused by the compiler optimizing the order of operations
differently (different roundoff errors).  These small differences would tend 
to grow because of the unstable nature of the atmosphere.  This affects the 
data-poor regions because observed data keeps the analyses on the 'straight 
and narrow'.

Another change (April 95) involved eliminating a coding error that caused
the model to "blow up".  The effect of this change was found to be minor 
because the bug was rarely invoked (first time the boundary layer code crashed 
after years of use in operational forecasts and research applications.)

Another change (April 95) fixed a problem with the convection scheme.  The
modification fixes a problem with the atmospheric moisture budget. The change 
was considered to be minor and have little effect except to better conserve 
moisture.


------------------------------------------------------------------------

Jan 1986-Dec 1992: See PROBLEM.TMP.
  1/86 to 12/92 Avoid using Tmax.  Tsfc and T2m should be screen
      (as done on the Annual CD-ROMS).
  Other months: Tmax and Tmin are included on the Annual CD-ROMS.
      Data is not screened.

---

  PAOBS are Australian analysts estimates of sea-level pressure
based on satellite data, time continuity, ship observations and
some "rules of thumb".  I've seen analysts work at the South American
desk (NCEP) and they can create analyses in very data poor regions.
By the proper use of satellite imagery, one can get front locations,
cyclone position and estimated intensities.  They are using techniques 
that are not available to the data assimilation system.  On the other
hand, the analysis can sometimes appear to be "magical".  Fronts 
appearing out of thin air.  I would expect that the error bars on the 
PAOBS data to be very dependent on the analyst and location.  
  PAOBS are used in the operational analyses and got "grandfathered"
into Reanalysis (in my opinion).  


prior to Jan 1979:
  PAOBS not used

Jan 1979-Dec 1992:
  The PAOBS data were used with a 180 degree longitude shift in these
  pseudo-observations.

Jan 1993-present:
  Correct positioning of the PAOBS.