Humidity
                                 --------


                             Negative Humidity
                             -------- --------

Like many other spectral models, the relative humidity can go negative in the 
Reanalysis model.  Some models add or redistribute moisture in order to 
eliminate the negative RH.  Rather than add a source of water vapor, the 
Reanalysis model ignores the negative RH (except for the radiation code which 
uses a zero relative humidity).  

The post-processor sets the relative humidity to zero whenever it encounters a 
negative RH.  The only place where the negative RH is present is in sigma 
spectral, sigma grid, and the precipitable-water calculations.  This implies
that a calculating the water budget should be done using the sigma files.


                         How good is the Humidity?
                         --- ---- -- --- --------

 The humidity analysis is believed to be the weakest of the primary
atmospheric analyses; i.e., Z, T, U, V, and Q.  The other primary
variables (Z, T, U and V) and their gradients have to be internally
consistent.  One can not change T without changing U, V, and Z. As
a result, the internal consistency provides a check and constraints
on these fields.  The humidity analysis, however, is unconstrained 
(except for q <= q-sat) and is effectively produced by a univariate 
analysis with no dynamical constraint on the gradients.  Then there 
is the sampling problem.  The humidity has many small-scale features 
and a single measurement may not be representative of a grid box average.  
By the way, did I mention instrument errors?


                         Near Surface Humidity
                         ---- ------- --------

Unlike temperature and winds, q is assumed to be well mixed in the boundary 
layer (0 meters to lowest model level).  Two-meter q observations are 
assumed to have been made at sigma = 1.  In addition, the post processor 
uses the q from the model's lowest sigma layer as the 2 meter q.

                                SSM/I
                                -----

SSM/I (satellite) estimates of precipitable water were not used in
Reanalysis which was fortunate in hindsight.  Early experience (sep '96) 
with SSM/I in the operational system (GDAS) produced large increases in 
the tropical precipitation.   Obviously more research is needed before 
NCEP can use the SSM/I precipitable water in its assimilation system.


                          RH Round-Off Errors
                          -------------------

When the temperatures are cold, calculating the relative humidity
from the temperature and specific humidity leaves much to be
desired.  The problem is that we only allowed our GRIB files to 
have 4 digits of precision in the specific humidity fields (kg/kg).  
Over the Antarctic, the q is so low that the conversion to GRIB makes
the values equal to zero.  The corresponding RH is zero which is
obviously an erroneous result.  One user noticed that the RH variance
became too large near the North pole during winter.  This was a
result of the quantization errors.  In addition, the user noted
that the RH could be greater than 100%.  This was probably caused
by the grib encoder rounding the value of q upward.