Model Climatology
                            ----- -----------


In regions of no observations, the analysis is extremely close to the first 
guess.  Since first guess is produced by the forecast model, any biases in the 
model's climatology could show up in the first guess.  This problem is more 
severe in large unobserved regions as the air parcels have more time to 
equilibrate to the model climatology before observations bring them back
to reality.

As a first step in understanding the systematic errors of the analyses, one 
needs to know the climatology of the forecast model.  A multi-year simulation 
using the Reanalysis forecast model and observed SST and sea ice was done
by S. Saha (NCEP).

Another approach that is used at NCEP is to compare the 'analysis' climatology 
with the climatology of the 5-day forecasts.  Such forecasts are part of the
full Reanalysis archive.

Determining the analysis errors from either the climatology of the 5-day 
forecasts or a GCM simulation is very difficult.  However, one might expect 
that the systematic error of the analyses to have the same sign as the model 
bias with an amplitude related to the magnitude of the bias and number of 
observations.