Volcanoes, Climate and the Limits of Computer Modeling
Scarce a week goes by without some scaremongering headline about climate change, premised on apocalyptic conclusions drawn from a computer-generated climate model. Modeling lies at the heart of the whole vast climate-change industry, an industry sparked by the big government-backed computer modeling centers in the U.S. and U.K. To understand the frail connection between these models and the realities of world climate today and tomorrow, consider the crisis in world travel and aviation prompted by the Eyjafjallajokull volcano in Iceland.
The current eruptions began on March 20; the plume of ash from the larger, ongoing eruptions that began on April 14 led to systematic grounding of international and local European flights, losing airlines billions in revenues and paralyzing the travel industry.
There are very well-documented cases from the 1980s of volcanic ash—i.e., microscopic jagged particles of pulverized rock—bringing jumbo jets within minutes of disaster. The U.S. leaves the airlines to decide whether it's safe to fly, whereas European governments say Yea or Nay, based on computer models from the Volcanic Ash Center in London and Eurocontrol, an organization that coordinates air travel.
But as red ink spread across the airlines' balance sheets, and passengers bunked down for days at hubs like Frankfurt, questions about computer modeling of the extent of the potentially lethal plume became more insistent. Exactly how far had the plume extended? How come monitoring planes were reporting safe conditions in areas the models were identifying as no-fly zones?
Computers at the Met Office, which earlier forecast a "barbecue summer" last year and a mild winter for this year, produced a stream of maps predicting the ash would cover a vast area, eventually stretching from Russia to Newfoundland. But across almost all of it, there was virtually no ash at all, and none visible to satellites. (It didn't help that the main monitoring plane was laid up for a paint job.)
"We never understood why a blanket ban had been imposed—something that would not have happened in other parts of the world," a senior airline executive said to The Mail on Sunday. "Safety is always our paramount concern, but this seemed like over caution gone mad. As the days went by without the restrictions being lifted, we became more and more concerned that the policy was based on theoretical models which had little grounding in reality."
I called Pierre Sprey, a defense analyst with a background in statistics and a healthy skepticism about climate modeling, and he gave a dry laugh. Back in the 1970s, Sprey had done some environmental consulting and speedily learned firsthand the insuperable difficulties of a seemingly elementary assignment in air pollution: modeling the behavior of a plume drifting downwind from a single smoke stack. "It was a vastlier simpler problem than some generalized climate model, but still hopelessly intractable" when it came to predicting the downwind dispersion of the plume and its toxic constituents.
Sprey found, to his surprise, that the useless air pollution models he was dealing with in the early 1970s were actually based on WWII models developed to predict the behavior of chemical warfare weapons being tested by the British at Porton Down back in the 1940s. What emerged with finality from those tests was that there was no knowing where the poison gases might head, and indeed one powerful inhibition against the use of chemical weapons has always been the ease with which, amid a sudden shift in the wind, some act of stupidity by the gassers can end up killing one's own troops, as unforgettably described by the poet Siegfried Sassoon in his WWI memoirs.
Contrast the demonstrated impossibility of computer modeling the simple downwind dispersion of a plume from a single smokestack or volcano with the mind-boggling scientific hubris of trying to model the climate of the entire globe.
Here we start with endlessly faulty data—from instruments parked on urban "heat islands" to severely massaged data bases of daily temperature readings, from sketchy numbers for the vast reaches of the planet where there are almost no readings, to expurgation of decades of inconvenient data. Then these are meshed with models constructed around bad thermodynamics, baseless suppositions about the hugely dominant heat effects of water vapor and clouds, and hopelessly inaccurate quantifications of carbon uptake by the earth's forests and oceans.
These quack science models are further skewed by the modelers' doctrinaire anti-carbon passions, the vetting of their results by the corrupt bureaucracy of the U.N.'s IPCC and the dependence of their salaries on the expectations of funding agencies.
Small wonder, then, that the modelers' computer "reconstructions" of the planet's past climate conveniently wiped out the well-documented three-centuries-long Medieval Warming Period, as well as the subsequent 500 years of Little Ice Age—nor is it surprising that their terrifying computer prognostications in the IPCC's 2001 Third Assessment failed to predict the next decade's absence of any global warming trend at all.
COPYRIGHT 2010 CREATORS.COM
Entries(RSS)
Most refreshing and enspiriting, and thoroughly credible, I think, from what I have read of the shortcomings of the methods used to predict events in the "earth sciences." How do all the knee-jerk Cockburn detractors feel about him in this case?
Yup, well done. In addition to the concerns listed by Mr. Cockburn, there is also the simple fact that scientists have just been bamboozled by the computer itself, a still new invention to science -- a mirror with a razor edge. The Hodgkin-Huxley model, I believe done in 1952, of the giant squid axon was essentially done by hand and slide rule, and was comparatively simple to today's models (4 diffy q's).
It would take them weeks of hand calculation to get one experiment done, and they did dozens of experiments to test and validate the model. They also probably thought about it a lot more and didn't publish something until they were satisfied. The importance of this model was that it was *predictive*, meaning it didn't just model well some known observations, it predicted previously unseen behavior that was later validated by experimental evidence.
Now, modelers have become more like decorated hackers. All too often, they just throw something into the computer and get some result as fast as possible, and then do it again. This happens remarkably in sync with the funding schedules of the NIH.
I don't know if any of these large scale weather models has even come close to the success rate of the farmer's almanac, though. It is so often lost that computers are mere tools or, in some cases, just catalysts. They are often portrayed as solutions.
good article!
The two best-known global-warming scientists--James Hansen and James Lovelock--have both expressed significant reservations about climate modeling. They say that the crucial evidence for man-made climate change is to be found in an examination of the oxygen isotopes trapped in small bubbles in ice cores that date to millions of years ago. And after 2007, when unexpected Arctic melting surprised scientists, it's believed that the computer models have understated the problem because they have trouble accounting for all the feedbacks--most of the known feedbacks accelerate rather than mitigate global warming.
Cockburn's last paragraph refers to the global warming of the last decade. That slight relative warming might be due to aerosals from China's air pollution; most aerosals cause a cooling. And it was undoubtedly partly caused by the solar minimum in the 10- to 12-year solar cycles that affect global temperature. Now that we are entering a new cycle, there' a fair to good chance that this summer will be the hottest on record (globally, with local variations).
As James Hansen says in his book, even the Great Enlightened Liberal known as President Obama just doesn't "get it." If the Great Enlightened Liberal is clueless, how can paleos be expected to understand that we are facing a completely unprecedented planetary crisis?
"4 diffy q’s." I haven't come across that terminology for years.
For an explanation of the Hodgkin-Huxley model, see:
http://www.ces.clemson.edu/~petersj/BioEng/SimpleGates/SimpleGates002.html
For some interesting videos of the experiments leading to the model, see:
http://www.science.smith.edu/departments/NeuroSci/courses/bio330/squid.html
Mmmmm. My previous comment is "awaiting moderation." Request an explanation if I've been immoderate.
Mr. Van Sant, our software automatically throws comments that contain links into moderation.
Thank you for the information, Mr. Richert. I better pay closer attention. I don't recall that happening before now, but it probably has because I often (too often?) respond with a link. In this case, although I am reasonably well-educated and also work for an organization that does modeling, I had never heard of the Hodgkin-Huxley model mentioned by Mr. McCabe. I thought others might be interested in what I found during a quick search. If you prefer that I not add links in the future, please let me know. My intent is to make a useful contribution, not to offend.
Creating models or simulations that accurately reflect reality and, thus, have predictive value is extremely difficult. A model does not have to include all factors. (If it did, it wouldn't be a model, it would be a description or definition.) A model does have to include all of the essential factors, those essential to the outcome the model is aimed at. How to identify those essential factors is a problem. A simple test of a model is to see if it can predict a known outcome using the known conditions as input. If the model's predicted outcome deviates from the actual outcome, then the model is not properly taking all essential factors into account. (Some essential factors may be missing or some may not be properly understood and be modeled incorrectly.) My understanding is that the current climate models cannot accurately predict known (past) climate outcomes; therefore they cannot predict future climate outcomes. They apparently do not take all essential factors into account.
"They apparently do not take all essential factors into account."
That is an overstatement at best. Don't you find it flat out funny that people are taking these models seriously? Think of the incredible complexity of what they claim to be doing, it should clear your sinuses. It's a Modest Proposal more so than a serious grant proposal, yet they are doing quite well in the funding department -- I don't believe the irony has settled in as yet.
I have also heard stories about the Government exhaling after WWII and trying to harness the considerable, collected scientific brains used to fight the war into something more productive. The goal was long-range weather forecasting. After years of work, lots of tries, and all the best computing power at their fingertips, they concluded they were still no more accurate than a dartboard. FYI, I think they defined long-range as more than 7 days out, and these were local predictions.
Mr. Van Sant is right about the climate models. They have value in the limited way he suggests--and bear in mind that there's a difference between predicting weather and predicting climate. Climate is essentailly longterm weather, and so is easier to predict (general states of the system are easier to predict than the specific behaviors of the system.)
It's a myth, however, that climate scientists rely mainly on modeling. The key evidence is considered to be paleoclimate data, which has nothing to do with climate modeling via computers. The information about past climate from ice cores and ocean sediments and the stomata of tree leaves (the Gingko tree is 200 million years old and is a mirror to past climate) provides the basis for predicting Earth's reaction to increased carbon dioxide put in the atmosphere (and cycled to the ocean) by humans.
To science-respecting skeptics, I'd recommend David Archer's *The Long Thaw*. To anyone who enjoys disaster scenarios, may I recommend *Six Degrees*, a rather terrifying book under the imprint of National Geographic.
KDZ, thank you for making the distinction between weather and climate. Very true. I don't know if the rest of your description is accurate though. It's not that climate is easier to model just because it's at a more general level. For example, how many known data points are there to build a general climate model? We've only been measuring global temperature levels for just over a century, and if our granularity is a year, then that's not many data points, especially given the nature of using time-sensitive data analysis (presumably, there's a relationship between one year of climate data and the next -- data dependency is a huge boogeyman). There is also a great deal of error introduced into the analysis when controlled, measured data (from the past 100+ years) is complemented with extrapolated, estimated data. I can't say much about that process other than I am very skeptical of it in my blissful ignorance of the specific subjects of ice samples and old trees.
You are the first person I've heard say that modeling is not at the heart of the research. The question is not really what is happening to earthly temperatures (the only specific "consensus" opinion of the scientific community -- it really is trending hotter!), which I believe can be fairly reliably estimated even from hundreds of thousands of years ago; it's what is the cause of these temperature changes. That is a modeling question, regardless of the use of computers or data. And certainly computational models are used to predict what will happen in all these cases where we skeptics continue to not listen to the advertisers.
Although, I am slightly intrigued by the idea that my SUV will help put Miami under water.
Thanks also for the book reco's.
In M&S we have a saying. "All models are wrong. Some models are useful."
Mr. McCabe--The following quote deals with the causes of climate change:
"To put it simply, observed data from the climate system are compared to what we would expect from natural internal variability (such as random weather fluctuations), from changes in solar activity, from [non-anthropogenic] greenhouse gases, or from possible other drivers of climate change. Different drivers of climate change cause different tell-tale patterns of change, so-called "fingerprints." . . . For example, greenhouse gases tend to trap more heat in the lower parts of the atmosphere, in contrast to a change in solar activity. The effect of greenhouse gases also equally works at night or in winter, while a change in solar activity is obviously more pronounced when the sun actually shines, namely during daytime and summer. . . . Global temperatures have already risen beyond their natural range of internal variations, and the observed pattern of changes can only be explained when the effect of rising [man-made] greenhouse gases is included. That is, the observed patterns are inconsistent with changes in solar activity or any other natural driver of climate change."
If you want more, Google "Stefan Rahmstorf" and at his homepage click on the book image for "The Climate Crisis" (co-authored with David Archer), and then click on chapter 3 and look for the section titled "Causes of Observed Climate Change."
I forgot to respond to your first point. The problems with climate models are not as large as you seem to think. For example, although it's true that observed temperatures go back only 120 years, so-called proxy measurements--notably from ice cores--go back hundreds of thousands of years. Such proxy measurements are not mere "estimates." They are an empirical substitute for more straightforward measurements. Proxy data is more objective than modelled estimates and somewhat less objective than measurement with thermometers. When different proxy data agree, the reliability of the data is enhanced. And as far as I know, there are no serious discrepancies in the proxy data, as between ice cores, ocean sediments, and tree rings.
KDZ, I appreciate your comments. As I mentioned earlier, I'm not a subject matter expert on the sources of proxy data you mention, but they are a source of error in each case. I wasn't implying they were model estimates, but they are in fact temperature estimates and are used to build (or inform) the model. Since there's no way to go back in time and directly observe temperature, they do it this way.
I am also pretty sure proxy data are accurate enough to piece together the temperature part of our ancient climate into the present. However, temperature is just one -- and possibly a very small -- piece of what makes up our climate.
To look at a very simple system model of global climate, let's say between sun storm activity, atmospheric CO2 levels and temperature levels, you'd have to have accurate (in time and space) readings of each of those three, and then you could build some sort of model of how they affected each other. Of course, you'd be leaving a lot out as well, like oceanic CO2 levels, and levels of all the other greenhouse gases, and tons of other stuff like wind patterns, magnetic fields, the Cubs' last championship, volcanoes, CO2 sinks, etc.
So how do you suppose that these scientists have been able to connect the simplest of data (like this 3 headed monster) from hundreds of thousands of years ago? How would they have been able to observe sunspot activity from today's perspective for even 1,000 years ago, let alone hundreds of thousands? They can get an estimate of it through proxy data, but then you have almost no reliable interplay regarding the specific timing back then, which as you can imagine is the level we want to make sure we're actually capturing in the model. It makes a big difference if CO2 is leading temperature or vice versa, and by how much time, and it makes a big difference what factors you are leaving out, and it makes a big difference if there is correlation among your input factors (as well as time dependencies).
Even if scientists can accurately estimate data from hundreds of thousands of years ago via proxy data for each of temperature and CO2 levels in the atmosphere, I don't think there's any reliable way to link their specific relationships together. It's proxy data. They assume the proxy data is the most reliable they can get, then they build the relationship from it.
It's just not there. Even a slight logarithmic shift over time, differing in those two dimensions would throw the thing off completely (one proxy's error increases more quickly than another's backwards through time).
I do stuff like this all the time. We estimate patient's "adherence" levels by looking at the timing of their prescription fills. It's useful to help guide our estimates of how many drugs they're actually taking in order to guess what the outcomes should be over time and vs. what a doctor is telling them to do. There's no way to directly observe patient adherence (patients tend to lie horribly about what they do, and our estimates are actually more accurate than their own accounts).
So I'm not saying proxy data is wrong or impossible to use, just saying it's another source of error, and it's incredibly difficult to combine disparate sources of proxy data reliably.
I'm not saying we can't do something with these models, but as some of the other posters pointed out "All models are wrong" and you have to define the exact purpose of the model before you build it, then you have to validate it against data, preferably third party data not used to build the model, etc.... that isn't always happening in this gold rush. I think of my dad's model airplane of his dad's WWII fighter. Beautiful model, down to the detail. Doesn't fly. Wasn't built to.
“…one [student] once told me I was a model professor. I thought this was high praise until I realized that a model is a small imitation of the real thing.” -- John D. Sterman
"A complex system that works is invariably found to have evolved from a simple system that works." (this has deep implications for Occam's razor, which also often gets overlooked)
John Gaule
"Today's scientists have substituted mathematics for experiments, and they wander off through equation after equation, and eventually build a structure which has no relation to reality." (even more true today with super computers)
- Nikola Tesla
Dr. McCabe (I'm assuming you're an M.D.): Thanks for your thoughtful response. I'm a lawyer by training, not a medical professional, so I know less about modeling than you do. My "faith" in AGW (anthropogenic global warming) depends on the basic reliability of some kinds of climate modeling, but it depends far more on proxy data and direct contemporary observations. We know that Arctic ice is melting in a way not seen in the Holocene era, and that glaciers worldwide are melting; we know that recent summers seem hotter than those of our childhood, and the winters milder; and we know that Al Gore is an idiot. Okay, the last one was a joke. We know, also, that humans are putting a lot of carbon dioxide in the atmosphere, as a result of industrialization. We know that carbon dioxide causes a greenhouse effect. We also know that the other greenhouse gases, specifically aerosals, can't explain the temperature rise since the '70s (thanks to sulfur filters on smokestacks, aerosals no longer have an overall cooling effect). What we don't know is what effect the clouds are having on all this. Despite near-total uncertainty about what the clouds are doing, which is freely acknowledged by the "warmist" scientists that I've been reading, we have very good reason to think that the measurable rise in mean global temperatures since the late 19th century (the thermometer and industrialization happily came along at the same time) is due to the increased carbon dioxide in the atmosphere--and humans put it there. The only plausible alternative explanation for rising temperatures is solar variation--the sun has been getting hotter for millions of years. Unfortunately for climate skeptics, the degree of warming since early in this century is much too high to be explained by solar causes.
If this is so, then someone who wants to deny AGW (of a potentially catastrophic kind) has to explain why we should think the chauffer did it when the butler was caught with a smoking gun and the chauffer has the alibi that he was at the theater when the murder occurred. (The analogy is David Archer's--he puts it a lot better than I can in his book on "The Long Thaw.") The skeptic has to shoulder the double burden of explaining not only why the butler (human-caused greenhouse effect due to excess carbon dioxide) didn't do it, but also why we shiould think the chaueffer (solar variation or whatever) did do it. That burden of proof is very hard to meet. It's simpler, and more defensible, to assume that the historically unprecedented level of carbon dioxide in the atmosphere is what the vast majority of climatologists are saying it is--dangerous anthropogenic global warming.