Wednesday, May 16, 2012


date: Tue, 29 Apr 2008 12:09:19 +0200
from: Stefan Rahmstorf <>
subject: Re: nature paper / ocean model as short-term regional climate
to: Andrew Revkin <>

Dear Andy,

thanks for asking. I think this an interesting paper and this kind of
decadal forecasting will become increasingly important. On the other
hand, it is still early days, this is pioneering work and many aspects
of this are not yet properly understood, as Richard Wood rightly
cautions in his N&V. Not least, nobody knows what the MOC really has
done over the past decades.

So what does this mean for the forecasts?

The prediction of European cooling: I'd take that in a Bayesian sense as
some evidence for possibly cooler temperatures, but not enough to make
cooling more likely than warming for me. So if I had to bet money at
equal odds on warming or cooling, I'd still bet on warming, although
with less confidence than previously. I'm not sure their error bars give
the full error - note for example that their forecast error bar for 2015
in Fig. 4 is almost zero, so even without having had time to properly
look at what this error bar encompasses, I suspect that it is not the
full uncertainty. And of course they predicted the 1994-2004 period to
be quite a bit cooler than before, and it turned out to be warmer (Fig. 3c).

Prediction of global temperature: The authors claim that this "may not
increase over the next decade". However, their standard A1B run has a
higher correlation with global temperature than their hindcast run, and
their hindcast run predicted cooling over the past 5 years (the period
1994-2004 was hindcast as being colder than the decade centered five
years earlier) when in fact warming happened, and it predicted flat
global temperatures in the past 10 years (in this sense) when warming

This is very interesting when you compare it to the time evolution of
the MOC. Almost the entire verification period, where their method shows
some skill, the MOC has been increasing. Since about 1990 it has stopped
to increase and started to decrease (in their analysis, nobody knows for
real), and consequently their method predicts cooling over Europe and
globally since then - but up until now, for the past two points in their
series, this prediction turned out to be wrong. That's why I'm not sure
whether I believe it for their next two points. If the method works only
while the MOC is increasing, but not while it is decreasing - maybe the
past forecast skill has nothing to do with the MOC? Certainly it has not
been well-validated for times of decreasing MOC. They predicted a slight
cooling for Europe for 1965 and 1970, but the observed trends here are
closer to the scenario run without the MOC forecast. The early phase of
weak diagnosed MOC, before 1970, may also not be weak MOC at all
(according to some work by Mike Mann), it may be at least partly an
effect of aerosol cooling over the North Atlantic region, which has the
effect of cooling regional SST and thereby looking like weak MOC - it
would be hard to disentangle the two effects.

So I think this is good work, I do not want to denigrate this in any
way, but as I said it is very early days with this type of forecasting
and many questions are still not properly understood, so I would not
start to draw strong conclusions at this point. Of course the basic
message is always correct: some natural variability is always
superimposed on the greenhouse warming trend, so ten years of less
warming or more warming don't tell us anything about this greenhouse
trend - see the discussion we had on realclimate on this point:

You could test how serious the authors are: tell them that a prominent
climatologist is offering them a bet of $10,000 at equal odds that
global mean temperature will be warming over the next decade. Are they
prepared to bet against this?

Cheers, Stefan

Stefan Rahmstorf


No comments:

Post a Comment