date: Tue Aug 30 11:25:59 2005
from: Phil Jones <p.jonesatXYZxyz.ac.uk>
subject: Fwd: sorry about this review
to: "Brohan, Philip" <philip.brohanatXYZxyzoffice.gov.uk>, "Kennedy, John" <firstname.lastname@example.org>
I had a more detailed look at the comments over the long weekend.
Here are a few thoughts.
1. If you want to know more about the structural uncertainty, that Tom P.
is talking about, then talk to either Peter Thorne or David Parker. This
term seems to have become widespread with the CCSP meetings on
the lower tropospheric/surface temperatures differences. Basically, it
is that errors don't include the effects of assumptions made in the
dataset construction. So, when you get new or modified data, such
as the revised SST data around WW2, the new data shouldn't change
things more than the earlier error estimates - if these were correct.
2. It would seem that the error bars appear too small for SST. I said
after the quick read, that I didn't think any more work is needed, but
I think you are going to have to add two things. First, it will be necessary
to produce HadCRUT3v a different way (the old HadCRUT2v way) of
merging the land and marine components - just to show that it
makes very little difference. Second, some comparison of the Smith
and Reynolds dataset and its error bars will have to be included.
The problem you'll find with SR, is that they have infilled data
everywhere, so their long-term trends are smaller, because
the infilling is mostly with near zero values, so the 1880-1920
period is warmer wrt 1961-90 than it should be.
As for Tom's points
His 4 is a good point.
6 will just need some additional words. This might need a sentence
saying how few stations these less good means relate to.
7 is mostly secondary points that won't be that important. r-bar isn't
calculated the way Tom thinks or maybe the paper implies. If you
look back and re-read Jones et al. (1997) you'll see that r-bar is
estimated from the earlier grid-box dataset. Probably just needs
a few extra sentences.
8. Is a reasonable point, but this will likely make the error bars
9. I agree on this !
10. If you compare with the old method merging this will show
that this isn't that important. All that seems to be required is
some more detailed explanation. The reason SST's appear
better is that the autcorrelation from day-to-day and measurement-
to-measurement is high (much higher than land).
11. A model could also be used to demonstrate this. Better than
A. My copy had these odd citations. There is an @ sign after each reference.
Probably came about when making the pdf?
B. WWR volumes for the 1991-2000 were not included. We've just received
these last week. Haven't had time to add them in. Harry will likely do this
but not till early next year, when we have some more funds for him.
D. Good point
E. More text needed, with a reference back. Maybe get David P to read through,
if he hasn't already.
F. We use anomalies
G Good point.
H. I thought I'd caught all these.
I. This is a good point. It would be useful to say that there
isn't a master dataset that we all draw from. We just use all
we can get. This will answer N as well. I can write this. I have
been in touch with Tom Peterson about this over the summer
wrt Roger Pielke.
J. Better explanation needed.
K. These sd's differ though?
P. Errors should be independent. Some rephrasing needed.
Q. Agree that this needs to be clearer.
S. More text needed.
U. The recent periods are too short. May stations/countries
haven't made the relevant simultaneous measurements.
X will get answered in one of the main points.
Z comparison of the different weighting schemes will show this
doesn't have much effect.
EE One of these is the average station variance.
Hope these are of some use.
Prof. Phil Jones
Climatic Research Unit Telephone +44 (0) 1603 592090
School of Environmental Sciences Fax +44 (0) 1603 507784
University of East Anglia
Norwich Email p.jonesatXYZxyz.ac.uk