The notion that climate is chaotic tends to be taken as a given, with little supporting argument.

No, complex dynamic systems with more than 3 independent variables are taken to be chaotic.

Clearly, if you turn down the sun, the temperature drops. Clearly, if you throw a bunch of SO2 into the stratosphere, the temperature drops. Clearly, if you turn the surface completely white, the temperature drops. And clearly, if you double the amount of an important GHG in the atmosphere, the temperature rises.

But these are single independent variables.  y=f(x),  Temp = f(SO2).    Yes, you can strip away single variable relationships, but they don't add up to "climate".

I see no problem with speaking in a meaningful way about future expectations.

I therefore challenge you:

Present to us, the best of the best "climate models" here on Grist.   But not models that explain the past, but models that predict future behavior.

I would like to see 50 and 100 year  projections, with 5 year intervals.   These models should show the relationship of anthrogenic CO2 gases to Global Mean Temperature.  

In fact, why not a year by year model?   There should be a definitive measure of how antrogenic CO2 is defined.

Then we can bet a beer or something on whether (pun intended) the model fits the results or not!

The Texeme Construct offers international text memetics construction and textcasting services.

Sigh,
Clearly you are not familiar with complex systems. Presence of more than three variables does not mean chaos. It means that our brains are less capable of understanding these variables and their interactions, with the effect of appearing chaotic to someone less knowledgeable in the field. This is due to us thinking in three dimensions along a timeline. Clearly, there are plenty of systems (including anthropogenic) that experience more than three variables at one time. How do you think a plane flies?

What you are saying with y=f(x) is that we cannot create multiple regressions. Seems that one semester of statistics for me has just been completely erased by your post. Wow, you must be a real powerhouse in theoretical statistics. Models that 'explain the past' have to exist, that is how models are validated and if validated, then projections for the future can be made using these same models.

Come on now, if that were so, the motion of a particle experiencing no forces wouldn't be predictable.

I think you're confused.  There's a result--a clever one, at that--in the theory of discrete-time dynamical systems that says that the existence of a cycle of period three implies the existence of a cycle of every other possible period, sometimes called the "Period Three Implies Chaos" theorem.

"The ocean-atmosphere climate system is certainly a complex system, and capable of some surprising behaviors, but there is no evidence that it is chaotic in the formal sense."

Interestingly, the link embedded in this comment leads to a Wikipedia article on chaos theory that actually includes the following comment:

"Everyday examples of chaotic systems include weather and climate."

The idea that the climate, "is an inherently chaotic system, and as such its behavior can not be predicted." is important and needs to be addressed, but your article could clearly use some work.

I would try something along the lines of:

The scientific term 'chaotic' has a very specific meaning.  The behavior of chaotic systems are not unpredictable or random.  Chaotic systems develop in a deterministic way and are therefore, in principal, totally predictable.

In practice however, chaotic systems are very difficult to predict because their behavior is extremely sensitive to their initial conditions.  Make a slight error in measuring the initial conditions and observed behavior tends to depart very rapidly from predicted behavior.

The atmosphere is the classic example of a chaotic system, which is why predicting the weather more than a few days in advance is so frustratingly difficult.

However, if we were able to precisely measure the current state of the atmosphere and we fully understood all rules governing its development over time, we would be able to perfectly predict the weather.

As it happens we can only roughly measure the current state of the atmosphere and we don't fully understand all the rules that govern its operation.  However, this does not mean that weather is completely unpredictable.  Our approximations are good enough that we can make usably accurate predications over limited time horizons.

This leads to the obvious question, if weather predictions are next to worthless more than a couple of weeks out, how can we seriously make predict climate decades or even centuries in advance?

The answer is that 'climate' implies a much less specific prediction than 'weather'.  The less specific the prediction, the longer the time horizon over which we can predict with some degree of confidence.  

We can predict with reasonable accuracy the high temperature in Baltimore a few days in advance.  We can also make a pretty reasonable prediction of the average summer high temperature in Baltimore in February.  What can't predict in February, at least not any better than a random guess, is the high temperature on the 4th of July in Baltimore.

Average annual global temperature is about the least specific environmental measurement immaginable, so it is, unsurprisingly, something we can predict with reasonable confidence over very long timescales.

Having said that, I'm not particularly enamored with long range climate models.  They inherently include all kinds of unconfirmed assumptions about how the atmosphere-ocean-ice system operates and about future economic activity (which is not famously easy to predict).  It doesn't matter how much computing power you throw at the problem, a wild ass guess is still a wild ass guess.

The fact that the models can back predict historical environmental behavior proves absolutely nothing since that was exactly the data used to generate the models in the first place.  You can buy any number of software packages that back predict stock prices beautifully.  Unfortunately this does not make them effective predictors of future.

In my view we shouldn't be addressing global warming because the models are scary. We should address global warming because we don't know what the hell is going to happen, which is really scary. If you are driving blindfolded, taking your foot off the gas is a pretty good idea.

That's correct! Chaos does not imply that we can't make any general predictions.

Interesting that the author hand waves chaotic dynamics away as unimportant and then claims the average is taken when the mean value is more important to prediction of repetitious pattern behavior.

Regardless, the author's assertion does not measure up to the observed behavior. Weather is chaotic, and where recurring patterns continue, the parameters observed in a time frame that appear stable are in what's known as a subcritical region. Those parameters that are observed in a time frame with wildly varying and unpredictable results tend to be within a supercritical region, which can be the result of variable perturbations of the system under observation.

To get the picture of predicting chaotic systems on different scales, imagine a candle that emits smoke.  The behavior of individual smoke particles is indeed part of a chaotic system, BUT it can be easily understood that a slight wind will deviate the stream of smoke in one general direction.

Likewise, while anthropogenic carbon emissions will undoubtedly cause the world to warm up, the very nature of Earth's climate makes predicting the weather on any given day several years from now impossible.

All physical systems obey the laws of quantum mechanics.  One of those laws, the Heisenberg Uncertainty Principle, states that it is impossible to know, even in principle, the exact speed and location of a particle.  The minimum uncertainty is really really tiny, but has a big effect on small things.

A chaotic system is one that obeys exact laws, but is highly sensitive to initial conditions.  Since Heisenberg says that exact specification of initial conditions is impossible, no chaotic system is 100% predictable.

Another law of quantum mechanics says that all movement is essentially random and is predictable only in terms of probability.  Predicting the behavior of an individual particle is pretty much impossible.  But when bazillions of them get together, you can use probability to predict the average behavior to amazing precision.

To continue the previous post, trying to predict the pattern of a bunch of particles at any given moment is a pipe dream.  But if you look at those particles over a long period of time, the average behavior is quite predictable.

Weather and climate are similar.  Trying to figure out where all the clouds will be on Feb. 23rd, 2009 is nuts.  However, predicting the average temperature for the whole month of February in a given location will not be off by very much, even though weather is chaotic.

Excellent point! If you DO NOT look long enough at the smoke from a candle, it will surprise you. So, the real question is: for how long do you need to "look" at the smoke, or better, make measurements of its position, distribution, density, and temperature (remember, we are talking about a model with about 1% accuracy, about 3K over 288K)? An obvious answer is that you need to look at it until the smoke passes through most of its statistically possible shapes. Therefore, for the problem at hand, climate change, you need to "look" at it at least for a couple of glaciations and deglaciations, and have all measurements (including historical data about cloud cover with 1% accuracy). When you will have all these data handy, then we can talk about fitting them into a model of average weather, and predict averages for the next turn of global climate change. Until then, sorry.