Saturday, April 14, 2012


date: Mon Jul 13 15:22:10 2009
from: Tim Osborn <>
subject: Re: cruts tmp to 2008
to: Ian Harris <>

Negative spikes fixed (see e.g. Mali on page 12 of new attachment).
Mean differences slightly improved, though not much.
BTW attachment shows last year's CRUTS3 in black and the latest in pink.
I haven't (yet) compared the 1961-1990 mean of last year's and latest CRUTS3 with the
correct normals from New et al. It might be that the latest version is the better. I'll
do that next!
At 10:30 13/07/2009, you wrote:

Hi Tim,
A new now in /cru/cruts/
It should fix the negative excessions, and I have a hunch that it
might fix the means too. You see, I noticed that the mean differences
were all negative.. you can probably guess the rest, given the other
fixes I've just made!
On 10 Jul 2009, at 10:55, Ian Harris wrote:

Hi Tim,
A slightly more in depth reply ;)
On 10 Jul 2009, at 00:19, Tim Osborn wrote:

Hi Harry,
finally had time to take a look at the latest cruts3 run through
to 2008
for tmp, picked up from /cru/cruts/
Two PDFs showing seasonal national means are attached.
Look at ...2008a_vs_2008b.pdf first. Black is your previous
update to
2008, pink is the latest one. Many very similar, some small
(presumably due to outlier 3/4 SD removal... note that as these are
national/seasonal means, outliers might be quite large, yet only
show up
small in the means if many other stations contribute).

Yup, as in the Mexico/Guatemala spike:

page 4. The hot spike in Guatemala SON has been removed in the new
version. That looks much better.
page 6 & page 9: the hot spikes in France, Italy and Austria in
JJA in
2003 have been reduce slightly too. Not sure if this is right or
could ask Phil what he thinks. Could Jul & Aug 2003 have been so
hot that
some observations validly did exceed the +3SD outlier check? Or
do you
use a +4SD check for TMP? Anyway, this is one to ask Phil about.

Nope, 4SD is for precip only.

There are various other erroneous hot spikes that have now been
removed, I won't list them all here.
However, there are some cold spikes in both previous and latest 2008
updates... see e.g. Mali SON on page 12. Have you turned on only
checking for +3SD, and not for -3SD? Some wrong-looking cold
spikes are
still present.

Yes, **sigh** - abs() now included, re-running.

Now look at ...2005_vs_2008b.pdf. Black is last years CRUTS3
through to
2005 (I know the files went to mid 2006, but I stopped at last
year). Note this isn't CRUTS2.1! :-) Pink is again the newest
version of
the update to 2008.
There are some early 20th century differences that I'm not too
about, though it would be nice to know why they arise. One
concern is
that the mean level is different between the versions... see e.g.
JJA for
various countries on pages 7 and 8. Seems to be a constant
offset. It's
too big to be a simple rounding error in my calculations (I may have
changed from 1 dec. place to 2 dec. place, but some differences
are about
0.5 deg C), and these are absolute values so there's no dependency
on any
anomalisation/reference period meaning as I'm not doing any.
Intriguing. Perhaps some normals have change in some regions/ seasons?

It's very worrying, as they really should be ~identical! Normals
are read from the original (sacrosanct) climatology files so they
shouldn't have changed at all. The gridding, etc are the same, too.
I will try running old and new anomaly programs to compare outputs..

(1) hot spikes have been corrected.
(2) cold spikes still there.
(3) some odd differences in mean level.

Of the seemingly-endless kind.
Cheers for your help with this.

Ian "Harry" Harris
Climatic Research Unit
School of Environmental Sciences
University of East Anglia
Norwich NR4 7TJ
United Kingdom

No comments:

Post a Comment