cc: Keith Briffa <k.briffa@uea.ac.uk>
date: Tue, 14 Jun 2005 14:10:25 +0100
from: Gerard van der Schrier <g.schrier@uea.ac.uk>
subject: fingerprinting
to: Tim Osborn <t.osborn@uea.ac.uk>

<x-flowed>
Hi Tim,

Before leaving for the Netherlands, I wanted to have shown you some 
results of the fingerprinting work. It turns out to be a bit of a 
headache, I'm afraid.

Following your suggestion, I use decadal averaged for the MOC/NAO and 
PDSI data. The amplitude of the MOC and the NAO-related signal is then 
estimated on the change in PDSI between consecutive decades. To get 
better statistics on the performance of the method, I used the trick to 
combine each decade with all other decades (rather than only the next in 
line). So the impact of MOC & NAO on the change in PDSI between e.g. 
decade 1700-1709 and e.g. decade 1810-1819 is then estimated. This gives 
1250 different testcases.

It seems that the percentage of getting the sign of the MOC right is 
about 50%......  Also other tests indicate that I'm not doing much 
better than guessing randomly. It works better for the NAO though.

Tricks like regressing out the (known) NAO-signal or smoothing the 
response patterns (hoping to further reduce the noise) don't help much. 
I expected this to work better, so I may have to look at this again.

There is a strong signal of the MOC on (summer) sea-ice concentration. 
But is mainly east of Greenland (and not near northern Scandinavia as I 
had hoped). Sea-ice has a strong impact on temperatures, so maybe we can 
use that link between terrestrial climate and MOC.

Any ideas?

[[[retracted 3 paragraphs: personal, family]]]

I'll be back wednesday morning 22st June.

Cheers, Gerard



</x-flowed>
