Monday, May 21, 2012


date: Fri, 03 Dec 2004 16:20:34 +0100
from: Stefan Rahmstorf <>
subject: [Wg1-ar4-ch06] response to "threshold" question of Ron Stouffer
to: Ronald Stouffer <>

Dear Ron
on the risk of abrupt climate change and thresholds I have three things to contribute which
I hope will be useful.
(1) Results of THC expert elicitation.
Attached is a ppt file with first results on what a dozen experts think the risk of THC
changes due to global warming is. We hope to publish this soon so it can become a citeable
source in time for the AR4.
(2) Review editorial in press with Climatic Change
Attached is a paper which argues what I think is an important point sometimes lost: given
the large uncertainties but high potential impacts, dealing with abrupt climate change is
an issue of risk assessment, not one of prediction, and it should be discussed as such.
This has a number of important implications - e.g., doing a few "best guess" scenarios with
models is *not* a risk assessment. Think of the risk of a nuclear power accident - looking
at a few "best guess" scenarios would only tell you that the most likely thing is that the
power station will work just fine, without accident. It is also senseless to try and
predict an accident in the sense of running a model that will forecast that Chernobyl will
blow up in may 1986. What can and should be done, however, is that we try to work out what
could go wrong, and how likely this is. Another implication is that relatively low
probabilities do not imply "nothing to worry" - or would you board a plane with 1% chance
of crashing? I say this since one still sometimes finds simplistic statements about abrupt
climate change risk in the style: "didn't happen in my GCM, so let's not worry about it".
(3) I want to comment on one of David Rind's points.

4) The model response was in all senses linear - the less freshwater added, in Sv-Yrs, the
smaller the percentage reduction in NADW. Thus if only 25 Sv-Yrs of freshwater were added,
NADW reduction was only 50%.

This implies that the model did not see a "threshold" - and that there were really no
'surprises', since the reduction developed over time, and would have been observable had
anyone been looking for it.

I respectfully disagree with the conclusion here. Reason: it is not clear that you would
find a threshold in this type of rapid transient experiment - in fact, I would have
expected that you don't. This view was confirmed by Andrey Ganopolski, who has repeated the
same experiments as David in the CLIMBER-2 model. Everything is linear, just as in David's
model. However, to conclude that there was no threshold is clearly wrong, since we know in
our model a threshold and hysteresis behaviour exist from doing the equilibrium
experiments, which is what you need to do to clearly identify thresholds. In other words,
the type of experiment done by David is not the right type to show the absence of presence
of thresholds, and nothing about the presence of thresholds can be concluded from these
There are simple physical reasons for thresholds. (1) There is a threshold in the physical
properties of water, i.e. the freezing point. (2) There is a threshold for surface density;
if it drops below that, no convection will be possible any more since the water column is
stable - i.e., convection is a threshold process. (3) There is the Stommel bifurcation,
resulting from the large-scale positive salt advection feedback (Stommel 1961). This is
also found in ocean GCMs and in all 11 EMICs participating in our intercomparison
(including those with 3d GCM ocean component). Whether it is found also in fully coupled
GCMs or not is not proven yet since no experiment I know of has yet been done which would
be able to demonstrate or refute it. Until then, I think there is no reason to expect this
bifurcation is not found in coupled GCMs, since the physics is simple and comes from the
ocean component, and is found in ocean GCMs. Hard to see how atmospheric feedbacks, not
found in the range of EMICS we tried, would wipe out this behaviour.
Another (and important) question then is: do thresholds matter in a transient global
warming situation, or are they only relevant as a theoretical concept that helps in
understanding equilibrium, but not transient, behaviour? I would argue that this is still
very much an open question that has hardly been investigated systematically. The
first-order idea of a threshold implies that when crossing it, something changes. The
"something" could be the qualitative difference between DWF recovering (e.g.
Manabe&Stouffer for 2xCO2), versus it stopping (for 4xCO2 - it doesn't matter here whether
it stops for good, or just for some centuries). Such a threshold is also seen in Rahmstorf
& Ganopolski 99 for differing amounts of freshwater input in the context of a transient
global warming run.
Thus my summary would be: some transient global warming scenarios do support the idea that
thresholds matter, not only for equilibria. Some model runs do not show signs of
thresholds, but in my view this is no evidence for their absence, it is rather absence of
evidence, related to the experimental design not suitable to show thresholds.
All the best, Stefan
_______________________________________________ Wg1-ar4-ch06 mailing list

No comments:

Post a Comment