Tuesday, May 8, 2012


cc: taylor13@llnl.gov, santer1@llnl.gov, wigley@meeker.ucar.edu, p.jonesatXYZxyz.ac.uk
date: Tue, 17 Aug 1999 16:30:58 -0700
from: "Karl E.Taylor" <taylor13atXYZxyzl.gov>
subject: to mask or not
to: mmaccracatXYZxyzcrp.gov


I thought maybe I could contribute a few comments to your concern
over using a common coverage mask for surface and MSU temperatures.
(Copy of your relevent paragraph copied below.)

Whether or not to mask depends on the question being addressed. If
we wanted the best estimate of global mean MSU temperatures, then
clearly we wouldn't want to mask. The issues we address, however,
are largely based on an expectation (from models and observations)
that over large portions of the globe strong vertical coupling tends
to lead to large positive correlations between surface and lower
tropospheric temperatures. There is a further (model-based) expectation
that any warming trend at the surface should be slightly amplified
higher up in the troposphere. These expectations seem to be
contradicted by the MSU data (at least for global mean trends).

Masking makes most sense if there is in fact strong coupling between
the surface and troposphere. Suppose the CO2 warming signal were
one with relatively strong warming over land areas and weaker
warming over ocean. Suppose further that we only had surface
temperature measurements over land, but had MSU retrievals over all the
globe. Also assume a case of perfect coupling (1K rise in local upper
air temperature for every 1K rise in local surface temperature).

In this case the unmasked global mean MSU temperature increase would be
less than the "global" mean surface temperature increase, falsely
indicating a damping with height of the CO2 signal. If we masked
the MSU temperature (sampling only over land), then the global
means would be computed over the same area as the surface temperature
and the MSU temperature change would equal the surface temperature
change, indicating no damping of the response with height. This
second conclusion would be the correct one. Note, however, that
the true global mean temperature change (both at the surface and
aloft) would be best estimated using the MSU unmasked data (under
the conditions of this hypothesized case).

Under different conditions, and again depending on what question is
being addressed, it might be best not to mask the MSU data. In our
paper we wanted to determine whether the apparent discrepancy between
the MSU trend (very small) and the surface trend (positive, and larger)
could be explained by coverage differences. This makes sense since
models seem to indicate that the trends should be comparable. One
explanation for the discrepancy is that in models true global
means had been considered until now, whereas in the data the MSU mean
was computed from global coverage, but the surface changes were
computed from data covering about 70% of the globe. In our study
both model data and observations were treated with the same mask
so we rule out different sampling as a full explanation for
the difference between surface and MSU temperature trends.

Hope this doesn't confuse things further.

Mike wrote (in part):

I think one needs to be very careful about this coverage
argument--basically becuase the atmosphere can move anomalies around
compared to the surface. One would just not expect therir spatial patterns
to be the same, so taking a common spatial maskwill not resolve this (even
if it seems plausible). To illustrate, take an extreme example of there
only being sfc msmts for the equatorial eastern Pacific (the El Nino
region). There, the MSU and sfc temp go in opposite directions for quite
plausible physical reasons. Doing a mask and comparing for that small
region would make no sens and give negative correlations, etc. Now, in that
sfc obs cover most of the globe, the problem will not be so severe, but it
persists (it was for this reason that I was suggesting extrapolating to the
global value for sfc temp based on changing coverage--not sure how to do
that however). In any case, I believe taht MSU and sfc should only be
compared, if at all, for the globe as a whole.

No comments:

Post a Comment