(edited 12:15 p.m. 7/26/09 for clarity)
(3:25 p.m. DOH! Hawaii IS part of the U.S…)

I’m getting a lot of e-mails asking about a new study by Clement et al. published last week in Science, which shows that since the 1950s, periods of warmth over the northeastern Pacific Ocean have coincided with less cloud cover. The authors cautiously speculate that this might be evidence of positive cloud feedback.

This would be bad news for the Earth and its inhabitants since sufficiently strong positive cloud feedbacks would have the potential of amplifying the small amount of direct warming from our carbon dioxide emissions to disastrous proportions.

The authors are appropriately cautious about the interpretation of their results, which are indeed interesting. The very fact that the only 2 IPCC climate models that behaved in a manner similar to the observations were the most sensitive AND the least sensitive models shows that interpretation of the study results as proof of positive cloud feedback would be very premature.

But how could such a dichotomy exist? How could what seems to be clear evidence of positive feedback in the observations agree with both the climate model that predicts the MOST global warming for our future, as well as the model that predicts the LEAST warming for our future?

In my view, the interpretation of their results in terms of cloud feedback has the same two problems that previous studies have had. These problems have to do with (1) the regional character of the study, and (2) the issue of causation when analyzing cloud and temperature changes.

Problem #1: Interpretation of Feedbacks from a Regionally-Limited Study
Back in 1991, Ramanathan and Collins published a paper in Nature which indirectly argued for negative cloud feedbacks based upon the observation that regional warming in the deep tropics is accompanied by more convective cloud (thunderstorm) activity, which then shades the ocean from solar heating. This was the basis for what became known as the “Thermostat Hypothesis”, an unfortunate name since there are many potential thermostatic mechanisms in the climate system.

But as subsequently pointed out by other researchers, cloud feedbacks can not be deduced based upon the behavior of only one branch of vertical atmospheric circulation systems — in their case the ascending branches of the tropical and subtropical atmospheric circulation system known as the Hadley and Walker circulations. This is because a change in the ascending branch of these circulations, which occurs over the warmest waters of the deep tropics, is always accompanied by a change in the descending branch, and the two changes usually largely cancel out.

The new Clement et al. study has the same problem, but in their case they studied changes in the strength of one portion of the descending branch of an atmospheric circulation system, generally between Hawaii and Mexico, where there is little precipitation and relatively sunny conditions prevail. So, even if the regional cloud response they measured was indeed feedback in origin (the ‘causation’ issue which I address as Problem #2, below), it must be lumped in with whatever regional changes occurred elsewhere in concert with it before one can meaningfully address cloud feedbacks.

Unfortunately, further complicating feedback diagnosis is the fact that these atmospheric circulation cells are interconnected all around the world. And since cloud feedbacks are, strictly speaking, most meaningfully addressed only when the whole circulation system is included — both ascending and descending branches — we need global measurements. Except for our relatively recent satellite monitoring capabilities, though, we do not have sufficiently accurate cloud measurements over all regions of the Earth to do this over any extended period of time, such as the Clement study that used ship observations extending back to the 1950s.

But there is a bigger – and less well appreciated — problem in the inference of positive cloud feedback from studies like that of Clement et al.: that of causation.

Problem #2: The Importance of Causation in Determining Cloud Feedbacks
I am now convinced that the causation issue is at the heart of most misunderstandings over feedbacks. It is the issue that I spend most of my research time on, and we have been studying it with output from 18 of the fully coupled climate models tracked by the IPCC, a simple climate model we developed, and with satellite data.

[As an aside, as a result of reviews of our extensive paper on the subject that was submitted to the Journal of Geophysical Research, we are revamping the paper at the request of the journal editor, and working on a resubmission. So, I am hopeful that it will eventually be published in some form in JGR.]

Using the example of the new Clement et al. study (the observation that periods of unusual warmth on the northeast Pacific coincided with periods of less cloud cover)…what if most of that warming was actually caused by the decrease in clouds, rather than the decrease in clouds being caused by the warming (which would be, by definition, positive cloud feedback)? In other words, what if causation is actually working in the opposite direction to feedback?

It turns out that, when a circulation-induced change in clouds causes a temperature change, it can almost totally obscure the signature of feedback — even if the feedback is strongly negative. This is easily demonstrated with a simple forcing-feedback model, which is what one portion of our new JGR paper submission deals with. It was also demonstrated theoretically in our 2008 Journal of Climate paper.

In other words, a cloud change causing a temperature change gives the illusion of positive feedback – even if negative feedback is present.

In the case of the new Science magazine study, one of the major changes seen in temperature and cloudiness (and other weather parameters) occurred during the Great Climate Shift of 1977, an event which is known to have been accompanied by changes in atmospheric circulation patterns, such as immediate and prolonged warming in Alaska and the Arctic. And since circulation changes can cause cloud changes, this is one example of a situation where one can be fooled regarding the direction of causation.

This issue of the direction of causation is not easy to get around. While I’ve found some researchers who think this is only a matter of semantics, and who claim that all we need to know is how clouds and temperature vary together, our 2008 J. Climate paper demonstrated that this is definitely not the case.

The bottom line is that it is very difficult to infer positive cloud feedback from observations of warming accompanying a decrease in clouds, because a decrease in clouds causing warming will always “look like” positive feedback.

Based upon the few conversations I have had with other researchers in the field on this subject, the issue of causation remains a huge source of confusion among climate experts on the subject of feedbacks. Even though our 2008 Journal of Climate paper on the subject outlined the basic issue, the climate research community has still failed to grasp its significance. Hopefully, the more thorough treatment we provide in our JGR resubmission will help the community better understand the problem — if it ever gets published.