Ray,
 
I am looking at a very fascinating contour plot.  (I extracted it from
the original file and attached it as part of a work document.)  I wonder
how much of the variability you can explain.  The variability also
raises questions about the best way to make comparisons between
observation and model.
 
Ben has sent me several files.  We zeroed in late last week on daily
profiles from 7 UT (essentially local midnight) at Bear Lake.  I have so
far only gotten as far as looking at the contour plot between 45 and 90
km altitude.  
 
What I am struck by is the amazing amount of variability.  You must have
inputs that vary on a daily basis.  Near May and September, I am clearly
seeing spikes that appear with something like a 5 day "period" and that
grow with altitude from 55 to 80 km.  I presume these are the response
to planetary waves, which means you have a planetary wave input.  Where
does that come from?  
 
Going into October, November, and December, the peaks in the variability
appear further apart.  What is happening there?  In contrast, there is
very little variability in January and February.  What is so different
between the winter at the beginning of the year and at the end of the
year?  Then, in late December there is a feature over a week long that
grows incredibly with altitude.  What is happening there?  
 
Between 80 and 90, km the appearance is very different.  Many features
one to two weeks in duration are isolated in time and do not appear to
reflect any activity that is rising from below.  Are these features
reflecting changes in solar input or are they responding to auroral
inputs?
 
Between 45 and 55 km, a number of features appear that are a few days to
a couple of weeks wide.  They also are most closely identified with
winter.  Are these minor stratospheric warmings?
 
Is there any aspect of this temperature variability that is clearly
identified with gravity waves in one way or another?
 
As I said above, the plot is fascinating.  It also raises interesting
questions as to how to compare an observed multi-year climatology with
model output.  As you see here, we simply looked at the local midnight
value from the model.  Does it make more sense to make nighttime
averages (solar zenith angles greater than 96 degrees) of the model
data?  (That would be the closest to what we observe.)
 
Then, another question arises.  If I remember correctly, this model run
was to simulate 2002.  Is there much variability in your results from
year to year?  Our climatology covers the period from late 1993 through
much of this summer, with some notable holes.  We can easily compare
with the simulation from one given year.  From out data, we can get an
idea of the variability.
 
This is a great start.
 
All the best,
Vince