Personal Science

Daniel Kahneman (Nobel Prize winner) has done some great work on the psychological disrupters to decision-making. One of his points is that hindsight tricks the mind into thinking it now has a situation figured out. You think you understand cause-and-effect because in retrospect you can create a believable model that explains that one situation.

It's comforting, but it might still be wrong (missing parameters needed to predict other situations). It's the reason that so often people don't necessarily really learn from mistakes.

If the model can't consistently predict outcomes in an objective test, then it's not ready for use.

Sometimes the right answer is "I don't know" and we shouldn't be ashamed to say that- and keep working on it.

As a current Ph.D. student in Atmospheric Science, I have to say that you are misinformed about the nature of climate models. I could go on for pages going in detail, but I'll keep it relatively short and simple.

1) You cannot just reproduce anything in a climate model because it is constrained by the laws of physics. They are unlike statistical models in which you can just twiddle with the numbers to show any correlation you want. Since the laws of physics don't change, it is perfectly reasonable to treat reproductions of the past/present climate as predictions, and therefore can be used as tests of the model's performance.

2) When you tune a parameter one way or the other in a climate model, it is extremely difficult to predict what effect it will have on the ultimate outcome of the model run. This is because there are many complex interactions and feedbacks in the system...that parameter will affect some other physical property, which affects several others, etc. Some interactions are coupled, meaning they are both cause and effect.

3) There are many examples of climate models making successful predictions that have been verified, even if you don't agree with #1 above. For example, it was predicted by climate models that the lower atmosphere would warm, but the upper atmosphere would cool, and this was confirmed later by satellite measurements. They also predicted cooling due to the eruption of Mt. Pinatubo, and that indeed happened over the ensuing couple years.

There are certainly many unknowns that remain regarding climate change, as evidenced by many in the field that are actively research various topics. The significant warming due to human greenhouse gas emissions (2-4.5 degrees Celsius per doubling of CO2) is well-agreed upon in the community, however.

If this is what a graduate student in the field thinks, no wonder these models don't hold up. The only way to test a complex model such as climate models is to find out whether they predict correctly. Their complexity (your point 2) is irrelevant. Of course they are constrained by physics (your point 1). But all climate models are gross simplifications. They obviously leave out many things, from the effects of co2 on ground cover (which isn't physics) to effects on cloud formation, etc. Anyone can look at any climate model and point out a hundred things that have been left out. Whether the omitted things are important or not can only be determined by testing the predictions of the model. The models were tweaked (e.g., by adding this and that) until they fit past data -- that fit means very little.

So your important point is #3, where you say that there are several predictions that have been verified. The first verified prediction you give isn't about global temperature -- so it makes little difference. It is entirely possible the models are correct about the relation of upper and lower atmosphere but wrong about everything else. The second one you give (about the effects of a volcanic eruption) I could have made. You don't need a complex model to predict that. So that too is uninteresting. To be interesting a prediction has to improve on common sense.

Perhaps you know that in 2001, several scientists, based on nine climate models, predicted that Antarctic ice would increase over the next ten years. In fact, it decreased. Now the models have been adjusted so that they "predict" the decrease. So you can see how "predicting" the past means very little. And if these models have been adjusted as recently as 2010, we clearly need another 10 years to judge if these adjustments were good or bad.

A bit rich considering psychology is much more of a pseudo science than meteorology.

Your entire arguement is based on an observation that is not even wrong. Fact is temperature rise has been and is tracking the upper bound of climate models.

Sad this is the only way for some people to get attention.

Actually, Seth, with the exception of the Antarctic Peninsula, which is melting rapidly, the Antarctic continent itself has accumulated snow/ice. The models have been correct.

Some individuals are contrarian by nature. It's useful to science, even when they're incorrect, because it prods those doing the work to continue their inquiries and improves our long-term understanding.

However, relying on bogus sources, as is often the case with this subject, undermines you. You're simply neither well-informed nor credible on this topic.

Anonymous, my comments about Antarctic ice come from this talk

http://www.youtube.com/watch?v=VbR0EPWgkEI

by Richard Muller, a UC Berkeley physicist. He is a mainstream well-respected scientist in the field, not a "bogus source". It is possible that he is utterly confused and completely misled his audience, but I doubt it. If you think he made such a huge error I would appreciate a source so that I can contact Muller (whom I have met) about the discrepancy.

Leonardo, you confuse knowing something -- which climate scientists certainly do, as reflected in their models, which sometimes make accurate predictions -- with knowing enough to make useful predictions that the rest of us should trust. There is a difference. Consider cars. We use cars. They work. They are safe enough. Long before cars there were tiny things, things that grew into cars. Small engines, for example. The makers of these tiny things knew something. They tested their ideas. Enough of them were true that they could make tiny things. Fine. It took a long time to make cars because being able to make tiny impractical things was different than being able to make a car. Sure, climate models are "useful" just as those tiny things were useful. The tiny things, research about them, trying to improve them, led to cars. But these tiny things were not enough to make cars, just as climate models have not yet demonstrated enough predictive power to make the rest of us trust them. Sure, climate scientists are impressed by the sort of prediction you give. But if you want me, a non-climate-scientist, to take predictions of 10 years in the future seriously, you have to show that the model makes similar predictions correctly (and better than common sense or much simpler models). If climate models have done this, it has been kept a secret. Which is hard to believe (and you give no examples of such predictions) so I conclude they haven't.

Leonardo, you again refer to fitting past data as if I should be impressed. Do scientists publish their failures? Rarely. The models that didn't fit the past weren't published. In this way, at least, the models we learn about have been selected to fit the past. To repeat, these models have lots of adjustable aspects so fitting the past means little or nothing. Perhaps they could have been selected (adjusted, tweaked) to fit any plausible past.

The volcano prediction would have been a lot more impressive if it was the first volcano. It wasn't. As far as I know, modelers knew from previous volcanoes, such as Krakatoa, roughly what to expect. So they could adjust their models to produce that result. For a verified prediction to impress us, to influence our beliefs, it must be at least a little surprising. Your description of the volcano prediction omits the "surprising" aspect. When your professors or books told you about the volcano prediction, did they make clear how surprising or not surprising it was? It sounds like they didn't. Moreover, the volcano prediction is about the effect of dust in the atmosphere. It sheds little light on whether the models can predict the effect of co2 in the atmosphere. Which is what we are told they can do.

I agree that the models are not hopeless -- they have made some predictions correctly. But the predictions I care about are about future global temperature: What will be the temperature in 5 years? In 10 years? That is what those models have been used to make claims about. I have not seen such predictions verified.

The climate change debate is about the effect of CO2. To say you believe a climate model because it predicted what happened after a dust event is not terribly convincing. I agree that such verified predictions, assuming they were surprising, support the part of the model that made them. But predicting the effect of dust is different than predicting the effect of CO2, in spite of overlap.

"The models that accurately reproduce past climates do so because they reproduce the physical processes involved in producing that climate." A model can do a good job of fitting past data for two reasons: 1. It is a reasonable description of the underlying mechanism. 2. It is too flexible. There is not enough degrees of freedom in the data relative to the degrees of freedom in the model. You seem unaware of possibility #2. A climate modeler has thousands of choices. What to put in, what to leave out, plus adjustable parameters. These create flexibility in the model, in the sense that the model that emerges could have been quite different. Too much flexibility? Could the modeler have fit any plausible past data? We don't know. That's why predictions of the future are essential.

Seth,
It's sad how little some "scientists" understand about the discipline of scientific research. You're supposed to objectively test your hypothesis against new data- that means to apply your model to predict future outcomes and then test to see how well your prediction fit reality.

Establishing a model based on past data and then validating it using past data (or a weak form of predictive data) is utter garbage and intellectually dishonest.

It's the same garbage that the finance industry spews every day. Each day they tell us that the stock market went up because of oil prices or this, that and the other- there's always a solid hindsight-based characterization. And that's good for Wall St. because it makes the hapless day trader think they now understand the markets and should trade more (even though in truth they know nothing about future events). But it's BS. None of those explanations hold up to scientific scrutiny- what about the days when oil prices went up and the stock market stayed even or fell (ignore those).

The sad part is that now scientists are losing the discipline. I hope you can impress upon people to not dumb down scientific thinking and the scientific method.

I mean really, why do people feel qualified to have such strong opinions on something they don't have a clue about? I'm done.

Thanks for the lengthy chat, Seth. Even though we disagree, hopefully it's been informative and engaging on your end. I only leave with this humble advice: Feel free to tread into subjects that you have limited knowledge in, but be aware of those limits.

Leonardo, you should learn about Ockhams razor, the principle of simplicity,
because it is one of the two main principles of science.
The other one is testing.

Your climate models are not science,
because they are not tested sufficiently against new data,
and because they are too complex.

I agree with Seth 95%, but I will concede if the climate simulations really made actual bona fide, interesting, accurate predictions about atmospheric temperatures I would consider a point in their favor. But I doubt that they really did.

Every time I have had this debate with warmists, the predictions they cite turn out not to be bona fide predictions (i.e. back-casting); not to be interesting predictions (i.e. the world will cool after a big volcano); or not to be accurate predictions (e.g. Hansen's various predictions).

So Leonardo, if you are still around, please give me a cite, quote, and link to these predictions about atmospheric temperatures.

"Modelers do. not. ever. tweak. to fit past data. "

Well what happens if somebody constructs a model; runs it; and discovers that it does not fit past data? If they don't tweak it, their choices are (1) quietly discard it; or (2) send it out for publication.

Choice (1) amounts to the same thing as tweaking, since there will be a selection process for models that fit past data.

If (2) is what happens, then we would expect there to be lots of publications of models which don't fit past data.

I have a hard time believing that this happens.

Also, how do you know that modelers don't tweak their models to fit past data? There's really no way to verify this claim.

I am unfamiliar with Richard Muller. He may be a respected physicist. But he is seemingly unfamiliar with the data on this topic.

The National Climate Data Center should be able to shed some light on this for you. With...data. It's generally considered a more credible source than youtube.

Leonardo, I've written a paper about how to test complex theories, which has been highly cited in my field (experimental psychology). I am using the ideas of that paper here.

Anonymous, the YouTube clip is a talk Richard Muller gave at UC Berkeley, where he is a professor. He has given a whole series of talks on the topic (available on YouTube). Perhaps he is wrong. But the fact that you haven't heard of him suggests you are wrong.

If you are interested, why not examine the data first-hand? There is no need to rely on second-hand interpretations from anyone at all.

I am on good terms with a few who are the most frequently presented as "the" sceptics, who have expertise, whose arguments contain verifiable data and are significantly more nuanced than anything presented here - and who do receive significant funding support, despite your assertion in another post that they're subject to a vast conspiracy to deny them such.

Dr. Muller's expertise simply does not lay in climate. He may discuss it - he may discuss anything he likes. That does not make him an expert. He is incorrect. The data are clear.

Since you're certain they tell a different story, though, why object to having a look yourself?

Awesome article! I have gradually become fan of your article and would like to suggest putting some new updates to make it more effective. I am unfamiliar with Richard Muller. He may be a respected physicist. But he is seemingly unfamiliar with the data on this topic.
Incredible information about upcoming production of Mattresses and I am sure people must have get impressed with this post. Thanks for it very much...

Hostess torino

Would that be the Prof. Muller who's just completed a huge study that concludes "global warming is real"? That Prof. Muller? That "respected physicist" who was a known skeptic who's turned denialism on its ear? (I know, I know, we have to wait for the peer review, but you maybe don't believe in that either.)

Models or no models, ask the people already horribly impacted by the destabilization of their climate ... people losing their livelihoods and their loved ones, their water sources and food security, their homes and their entire homelands. Ask *them* whether the models matter.

You know, it's not considered "doomsaying" if the doom is actually happening. And it's not "alarmist" to raise the alarm if there is actually something to be alarmed about.