warmist - someone who doesn't reject modern climate science
evolutionist - someone who doesn't reject modern biology
I've noticed that some people who wouldn't contemplate wholesale rejection of geology or biology are completely dismissive of climate science: "it's all models and simulations", "it can't be tested", "it's dependent on complex computer programs", "it's all too uncertain", and so forth. And this extends beyond the denialists, who have no interest in the science and are largely just trying to avoid having to accept its implications, to some who have been misled by their rhetoric.
This seems to be driven by some misunderstandings about how science works, so some comparisons may be useful.
First of all, climate science operates within the same institutional and social frameworks as the other sciences. It is carried out by scientists working at the same universities, publishing within the same system of peer review (in some cases in the same journals, with broad science journals such as Nature or Science), funded by the same or similar research bodies. (Obviously biology gets more medical funding, geology more mining industry funding, etc.) So there's no sociological reason to expect climate science to be more biased, less reliable, or otherwise inferior.
Every science is different, obviously, and "biology", "geology", and "climate science" are actually complexes of different disciplines. But there's nothing foundationally different about climate science. The use of complex computer programs, simulations and models extends throughout the sciences, the interpretation of data is always dependent on existing theory, there are many sciences that are historical rather than experimental, and many disciplines have to deal with large uncertainties.
Computing is now ubiquitous in the sciences. Rejection of everything dependent on complex computer analysis or simulation would mean throwing away huge slabs of high energy physics, astrophysics, chemistry, genomics, and so forth. (Even in mathematics there are proofs of results such as the Four Colour theorem which rely on computer programs and are justified by indirect arguments about the validity of those programs.) Bridges, aircraft and other engineering constructions are tested in simulations and hence have to deal with simulation errors in addition to model specification risks.
Model assumptions in science are often untested, or hard to test, and model validation is always heuristic, with models often used as theory discovery tools or predictive frameworks. Attempts to use phylogenetics to probe the early parts of the tree of life, for example, involve choices of substitution models, selection of parameters, and other assumptions for which there are only ad hoc justifications, and produce conclusions which we have limited ability to test against other lines of argument (from geochemistry, say).
Theory dependence of observation and model-data interactions are standard themes in the philosophy of science, and this extends even to the "hard" experimental sciences. Raw data from something like the Large Hadron Collider is hardly self-explanatory (my understanding is that it is pretty much incomprehensible without the aid of a bucketload of auxiliary theories and complex computer analyses). Fossils are searched for and interpreted in the context of an existing understanding of the history of life. And so forth.
There are many sciences which are, largely or in part, historical rather than experimental. The reconstruction of the history of life is one example. Geological models of plate movements rest on different physical foundations to paleoclimate models, but operate on the same spatial and temporal scale. (And also perform relatively poorly at predicting short-term phenomena such as earthquakes and volcanos.)
As far as large uncertainties go, the obvious parallel is with other environmental models. Hydrological forecasts of flood levels, for example, have large uncertainties, but that's hardly an argument for not performing them. The scientific response to uncertainty is to try to reduce it, or to understand and track it, not to throw up one's hands in despair.
What is unique about climate science? Perhaps the scale of the data collection and computation involved. Perhaps the "density" of the model-data interactions. Nothing, it seems to me, that places it on a qualitatively different epistemological footing to the rest of science.
The IPCC, much-hated by denialists, actually makes climate science stand out. Few, if any, other disciplines make such an open, international attempt to synthesise key results and present them in a format accessible to policy-makers and laypeople.
There are all kinds of unsolved problems, poorly understood areas and lacunae within climate science, just at there are in other sciences, but I see no sign that the discipline as a whole is in crisis. Perhaps the biggest problem it faces is in educating non-scientists in the face of well-resourced external attacks. (It is no coincidence, surely, that among developed nations the United States has by far the largest proportions rejecting both evolutionary biology and climate science.)
The political implications clearly explain why people attack some sciences rather than others, but don't have any direct bearing on the strength of the science itself. And political pressure is just as likely to distort science towards conservatism.
http://www.smh.com.au/opinion/society-and-culture/climate-change-the-black-white-and-grey-in-the-science-20101227-198ey.html
This is the latest article by Kurt Lambeck which I read on the smh aome weeks back. He is also highly critical of the way Ian Plimer refutes the hard facts of changing weather patterns world-wide. On Plimer's book, Lambeck says if it were an honours student's submission, he would send it back for a re-work.
Aren't you neglecting one thing : citing your source. Or is it your own etrapolation? You are opening yourself to attack by doubters ( though not in this blog.)
You're right, I should attribute that. It's from the NASA Goddard Institute for Space Studies http://data.giss.nasa.gov/gistemp/graphs/
That was a substantial, meaty post, Danny, and one that deserves considerable thought.
I also believe it's an honest view of science, with all its limitations.
Some of the points you make are indeed hard to argue against. In particular, I agree with this:
"So there's no sociological reason to expect climate science to be more biased, less reliable, or otherwise inferior."
I, for one, wish one could say the same about economics or the social sciences.
I also agree with your point that indiscriminate rejection of simulation would make advances in many sciences at least hard, if not outright impossible. This clearly implies that a wholesale rejection of simulation can’t be considered a good thing.
But I don't think the following has had a thorough consideration:
"Model assumptions in science are often untested, or hard to test, and model validation is always heuristic, with models often used more as theory discovery tools or predictive frameworks."
The fact that model assumptions are often untested means that we cannot be sure they are "true" (or even "approximately" true) in any sense. And the consequence is that we can’t know a priori that the conclusions derived from them are true, either.
But even if we could ascertain a posteriori that the conclusions are indeed true (and maybe that’s what you meant by heuristic validation?), we still wouldn’t know why they are true: the model could be right, but for the wrong reasons.
I am not sure what “theory discovery tools” means (I suppose it may have an specific technical meaning), but judging by the everyday meaning of the phrase, I would say that the above observation about untested assumptions severely constrains the usefulness of simulation models as “theory discovery tools”.
If I am mistaken, please let me know, because this is probably my most fundamental objection.
With qualifications, I have similar reservations about the use of models for detailed predictions. The qualification is that I do believe some simulation methodologies are more useful than others: for instance, discrete simulation is a much more tried and tested methodology with clear and useful applications for forecasting.
As other people have already mentioned the social implications, I will not touch them here.
Anyway, that’s my two cents.
Magpie, you're right, that bit of what I wrote wasn't so clear.
Perhaps a better way of putting it is that models are not theories. They can be components of theories, products or implementations of theories, tests of theories, tools for developing new theories, and so forth, but they don't constitute theories by themselves. Even in physics, something like Maxwell's Equations do not make up a theory of electromagnetism in themselves. (This is where I think some economists and social scientists go wrong, when they start reifying their models and separating them from their theoretical contexts.)
So the question about models is not whether they are "correct" - models are invariably simplifications, and hence known to be "defective" before they are even constructed! - but whether they are useful. Models can be useful for prediction, for rejecting hypotheses, for finding new hypotheses, for integrating data for use in other theories, etc. etc.
Returning to climate science, climate models are useful predictively (yes, the uncertainties are high, but they have performed much better than throwing our hands in the air and yelling "too hard"), useful in uncovering phenomena, useful in testing hypotheses, useful in making climate knowledge accessible for other applications (in palaeontology, geochemistry, hydrology, ecology) and so forth. If we were to find some kind of big anomaly or unforseen result in our climate models, that would be really interesting, as it might bring to light new phenomena. (Or it could result in model revision or rejection.) If we didn't know about the solar cycle, for example, we would have discovered it from our modelling. Similarly, the temperature trend is incompatible with our models in the absence of carbon dioxide greenhouse forcing, so if that isn't happening then there would have to be some other, as yet unknown process involved. (Or we could reject the data, or the idea that the atmosphere behaves as a fluid, or the conservation of energy, or any number of other auxiliary hypotheses. Science is complex!)
In some sense all predictions are model predictions. And all predictions have uncertainties; the important thing is to keep those uncertainties visible and not hide them.
I'm not sure what you mean by "discrete simulation". Climate models are discrete, as is anything run on a digital computer. There are certainly issues with the use of finite-difference methods to handle differential equations, but this is now relatively well understood. It's also worth noting that the continuous equations which are handled by numerical approximations in climate models (and aerodynamic and fluvial modelling) are themselves simplifying models of underlying molecular movements which could be considered "discrete".
Just a short note to thank Magpie ( though I don't really know you) for posting Gerry Rafferty's Baker Street. It is one of many '70s hits I like, yes, the Abba as well. In the seventies I was well into adulthood, too busy making a living and later fatherhood. I had no time for the pop scene.
Danny, please bear with me on this.
You said: "They [i.e. models in general] (...) don't constitute theories by themselves."
Further: "This is where I think some economists and social scientists go wrong, when they start reifying their models and separating them from their theoretical contexts."
In principle, I tend to disagree with the above. But I'll admit this is an entirely new idea to me.
So, before engaging in an argument that may yet prove unnecessary, let me explain what I understood, so that you can check if that corresponds with what you have in mind: models are either (1) a particular instance of a theory ("implementations of a theory"), which is a more abstract object; or (2) models can be implementations of a partial aspect of a theory ("Maxwell's Equations do not make up a theory of electromagnetism in themselves").
Or maybe it would be best if you could provide a simple concrete example of what you mean.
===================================
Discrete simulation or more precisely discrete event simulation is a kind of simulation methodology (I know only of applications to engineering and business administration, like waiting lines, scheduling and such).
The difference between it and continuous time simulation is that in continuous time simulation the simulation clock advances by arbitrary, but equal, time steps (one day, one hour, one month, etc.). Think of what JW Forrester and the Meadows couple did.
In discrete simulation the simulation clock only advances when an event (normally of random nature) occurs: say, when a customer places a phone call in a call center waiting line.
Thus, the natural output of a continuous time simulation is a time series; not so for a discrete simulation (you can have one phone call at 9:00 am; no phone calls between 9:00 and 9:05, three phone calls from 9:05 and 9:06, 10 phone calls in the next two minutes, etc.).
Discrete simulation is useful to predict, say, customers' waiting times as a function of the number of attendants (in the call center example). And, if properly done, using hard data and all, it can be surprisingly accurate, too. The downside is that the universe of applications appears limited to quite structured situations.
Have a look at:
http://en.wikipedia.org/wiki/Discrete_event_simulation
@DL,
You're welcome!
Magpie, I understand what you mean by discrete simulation now. My honours thesis involved implementing a discrete event simulation for comparison of different minimum spanning tree algorithms.
I don't think discrete event simulations are necessarily any more
reliable, robust, or easy to interpret than other kinds of simulations. With my minimum spanning tree algorithms, for example, the behaviour was hard to predict and heavily dependent on the inputs (the network structure) and the parameters (for the distribution of message delays). The only way I found for getting a handle on the uncertainties involved was to run multiple simulations with different inputs and parameters (and different seeds to the pseudo-random number generator).
Of course MSTs are a hell of a lot simpler than the climate -- and there are some nice analytic results for them (correctness proofs, maximum time bounds, and so forth). But there's no reason a discrete event simulation couldn't be just as complex as a finite-difference based climate model, in which case it's likely to suffer the same problems with precision and uncertainty estimation. A possible example might be a model of a cell.
Thanks for responding Magpie. I need to clear the air a little.I am Danny's long "lost" cousin. I met up with him again after an absence of about thirty years or so. Before then Danny was just a little primary kid. As you can guess there is an age gap of over a generation between him and me
I do not have a science or maths. background, don't have the right stuff to go on. I like logging on the blog and post a bit of my trivia as a diversion and relief to the hard core subject matters you lot delve in.
I have my fun.
My pleasure, DL!
I suspect I am older than Danny (and most certainly better-looking, too), as well.
Oh well, you can see I'm very modest, too.
Magpie, point 1, you are older, of which I am certain ( as per your home page). Better looking? Let me be the impartial third judge when the time comes.
Point 2, in my book modesty and tastiness go hand in hand. Love to give magpie chow mein, magpie chop suey or sweet and sour magpie a try. ;-)
@Danny,
"Of course MSTs are a hell of a lot simpler than the climate -- and there are some nice analytic results for them (correctness proofs, maximum time bounds, and so forth). But there's no reason a discrete event simulation couldn't be just as complex as a finite-difference based climate model, in which case it's likely to suffer the same problems with precision and uncertainty estimation. A possible example might be a model of a cell."
Now you see what I mean by hyper complex computer simulations, then.
Granted, if that's the best that can be done, then nothing better can be done and one needs to live with that. But that, at least in my view, means one needs to be extremely cautious when it comes to accepting results.
Pardon my ignorance, do the laws of thermodynamics have anything to do with climate change? The rising greenhouse gases must constitute a lot of heat energy in the atmosphere, it just can't disappear in "thin air."
@DL
Although Danny is probably much better qualified to answer that, I'd risk to say the laws of thermodynamics surely have a lot to do with climate change.
Rising greenhouse gases, however, by themselves do not constitute heat energy, I think.
As I understand it, those gases keep the heat coming from the Sun from irradiating back into space.
Thanks Magpie. I maybe guilty of "name-dropping", quoting thermodynamics laws, which is something completely out of my depth. It is that I had never read or heard of any climate scientists bringing it up. And I am curious to know if the laws have a part in explaining global warming, thus climate change.
Don't you worry! The important thing is not how much one knows, but to have a disposition to learn and an open, but critical mind. And to me it looks like you have them.
So keep it up!
PS: BTW, I myself mention them, but I don't know what specific role they play in the models Danny discusses. ;-)
Danny,
Let me ask you something related to climate change and I apologize in advance because the question might be incredibly stupid.
I understand that the causality involved in weather events is a complex one, and that a great many things can be behind a single weather event (this, I believe, can be translated into the statement that individual weather events are random events). Fine.
But here in Australia, during the last few weeks we had a number of extreme weather events (infrequent events, because of their large magnitude): floods in QLD, followed by a week of record high temperatures in Sydney (over 10 C above average), plus a category 5 cyclone in QLD, and floods in VIC.
Further afield, there also were catastrophic floods in Brazil, and country-wide snowfalls in the US, all happening at the same time with the Australian events.
My questions are these:
(1) Is there a point at which one is justified in REASONABLY attributing any such sequence of extreme and unusual events to climate change? Or,
(2) On the contrary, the usual "no single extreme weather event can ever be attributed to climate change", that one sees in the media, is invariably the right answer?
Let me justify my question this way: if one were betting against a coin toss, and one assumes the coin to be a legal one, one can expect in the long run a similar (although not necessarily equal) number of heads and tails.
However, any two equal-length sequences of tosses from a legal coin are equally probable. In particular, a long sequence of uninterrupted heads is equally probable to an equally long sequence with a similar number of heads and tails mixed, even if coming from a legal coin.
In practice, no one I know would keep betting on the legality of a coin if they observed, say, H,H,H,H,H,H,H,H even if it's equally probable to, for instance, H,T,H,H,T,T,T,H. And, personally, I would not blame them!
And lately, we've had a really long sequence of heads in Australia and overseas. From where I am sitting, this should mean something, maybe not with certainty, but with HIGH LIKELIHOOD.
This, I trust, clearly leads to our previous discussion at The Stubborn Mule on deductive vs inductive science.
It may be best to reformulate this:
"Let me justify my question this way: if one were betting against a coin toss, and one assumes the coin to be a legal one, one can expect in the long run a similar (although not necessarily equal) number of heads and tails."
In the following terms:
"Let me justify my question this way: if one were to judge on the legality of a coin, based on the evidence provided by the results of coin tosses, one SHOULD expect in the long run a similar (although not necessarily equal) number of heads and tails."
Magpie, I'll have a stab at answering this.
If you ask something simple like "can changes in climate induce changes in weather patterns?" then the answer clearly has to be yes. The question is how one quantifies this, and that's going to depend on regional specifics and will be different for each of the examples you give: we might, for example, predict a particular change in the frequency and intensity of Queensland cyclones resulting from the Coral Sea having warmed.
If you're asking whether weather events can be evidence for hypotheses about climate (say global warming), the answer is also yes. In this case, though, it might be more straightforward to measure the temperature of the Coral Sea and use that to update our climate models (and the strength of our belief in our climate models) than to rely on cyclones as a proxy for that. Rainfall and temperature averages certainly get used in climate models, but I'm not sure whether their variability is.
But you should ask a specialist about this, I don't know much at all about how climate and weather models interact.
Thanks for the answers, Danny.
My question was more addressed towards the second possibility (evidence for hypothesis about global warming).
There are then two considerations here:
(1) If those events can be considered evidence for climate change, then, it seems to me, denialists should be able to explain them without the hypotheses of global warming.
(2) If the default answer to unusual climate events were to be invariably "no single extreme weather event can ever be attributed to climate change", then climate change would be by definition "unprovable". This, I think, is an unfair burden of "proof" (I am using the term "proof" loosely, not in the deductive-logic sense).
Magpie, about "betting on the legality of a coin": this is the province of probability and statistics. There are lots of tests that can estimate the likelihood of an outcome (e.g. your sequence of coin tosses) based on (say) the percentage of heads. You would normally expect around half the tosses to be heads but not exactly. If you had 1000 coin tosses, you could reasonably expect it to be in the range 470 to 530 or so. (This is the 'margin of error' that public opinion pollsters like to quote). Someone who has more recent knowledge of probability could probably answer it better.