cc: John Kennedy <john.kennedyatXYZxyzoffice.gov.uk>, firstname.lastname@example.org, david.parkeratXYZxyzoffice.gov.uk
date: Wed Apr 9 12:48:40 2008
from: Phil Jones <p.jonesatXYZxyz.ac.uk>
subject: Fwd: problem with trends in Europe in HadCRUT3
to: Geert Jan van Oldenborgh <oldenboratXYZxyzi.nl>
This stems from the way we combine the land/ocean datasets around coastlines.
In HadCRUT2 we did it according to the area of land/ocean, but not letting one
dominate, so if land or ocean were < 25% it was made at least 25%, with
the other changed accordingly.
In HadCRUT3 we did it according to the errors of estimation (see the
Borhan et al paper). This tends to bias the coastal areas to the SSTs as their
errors of estimation are smaller. This results from the way we deal with
errors in the land and ocean components.
Probably neither of these is right all the time, but we will reconsider this
when we think about HadCRUT4 - which is someway off! We might be doing
something sooner if some improvements to HadSST2 (incorporating new
SST data from WW1 and WW2) get completed soon.
Doing what you've done is essentially going back to what was done in
the earlier version. Combination around coasts has always been a problem.
Date: Wed, 09 Apr 2008 10:27:46 +0200
From: Geert Jan van Oldenborgh <oldenboratXYZxyzi.nl>
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:18.104.22.168) Gecko/20080313
To: Phil Jones <p.jonesatXYZxyz.ac.uk>
Subject: problem with trends in Europe in HadCRUT3
X-OriginalArrivalTime: 09 Apr 2008 08:27:46.0989 (UTC) FILETIME=[971995D0:01C89A1B]
X-Spam-Score: undef - message too big (size: 597818, limit: 153600)
X-CanItPRO-Stream: UEA:f028 (inherits from UEA:10_Tag_Only,UEA:default,base:default)
X-Canit-Stats-ID: Bayes signature not available
X-Scanned-By: CanIt (www . roaringpenguin . com) on 22.214.171.124
as you may know I am busy verifying climate models on the observed trend in Europe so
far. We submitted a GRL about this a few weeks ago. When cross-checking against
station data I found a very curious problem in the HadCRUT3 daatset that we used to
represent the real world. In summer, in northern Europe, the trends in HadCRUT3 grid
boxes are much higher than the station data (and CRUTEM3) indicate, whereas in southern
Europe the reverse happens.
Looking at the CRUTEM3 and HadSST2 trends I found large positive trends in HadSST2 in
grid boxes that are >90% land, e.g., western France (0-5E, 45-50N) and northern Germany
(10-15E, 50-55N). Apparently these get a similar or larger weight to the CRUTEM3 data
in the same grid box, and thus in HadCRUT3 these grid boxes have large positive trends,
which are due to a few SST observations near the beach and down-weigh the large amount
of good station data of CRUTEM3. The opposite happens along teh Mediterranean coasts.
A weird trend in winter in Finland (20-25E, 60-65N) is also due to this effect.
As a stop-gap measure I defined my own merged dataset by weighing CRUTEM3 and HadSST2 by
the fraction of the grid box covered by land and sea respectively (derived from CRU TS
2.1, so overestimating land a bit). I used this to redo all the plots in the GRL, and
will substitute them when I revise it. Fortunately, the main conclusions are not
The problem is illustrated in the attached page from my log book, in which I show trends
in HadCRUT3 (left), CRUTEM3, HadSST2 and my home-brewn CRUTEM3+HadSST2 combination
(right column). The first row shows the trend in the annual mean, next DJF, MAM, JJA
and SON on the bottom row.
I can understand that you give a disproportional weight to island stations to
characterize SST around it, but the opposite seems to give unphysical results in areas
with a convoluted coast line. If anyone is working on the weighing for HadCRUT4 this
information may be of interest.
Greetings from sunny & chilly Holland,
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