Uncertainty of the Analyses
                         ----------- -- --- --------


    The uncertainty of the analysis depends on location, time (year, 
season, synoptic situation) variable and elevation.  Monthly means 
will have a different uncertainty than a single analysis.

    One way to estimate the uncertainty is to gather an ensemble
of analyses for the same time.  You want to get analyses produced
by different centers because they would have different assumptions,
model parameterizations, quality control, observed data (use of
satellite data).  The spread of the different analyses would be
an estimate of the uncertainty of the analyses.

    One want analyses from different centers because you don't
want common errors influencing the results.

    NCEP & NCAR have produced a DVD with an "ensemble" of analyses for 
the year 2006.  The 2006 ensemble memembers includes, Canada, ECMWF, 
FNOC, GDAS, CDAS, R2, UKMET.  There appears to be too many NCEP
models in the ensemble but the operational GDAS is much different
from the CDAS-R2 generation system.

    You can use the RMS of the ensemble members from the DVD as an 
estimate of the uncertainty of a modern analysis.  Of course, the 
uncertainty in 1950 would be much higher than in 2006 because of all 
the additional observations used in the current analyses.  The DVD 
can be ordered from

   http://dss.ucar.edu/pub/reanalysis/annual_dvd.html

At the time of writing the 2007 ensemble-of-analyses is in preparation.