In a recent post I gave several reasons to explain why having a ‘climate policy’ was a waste of time, energy and money for any country. In part I said this ‘…Fourth, but we can’t stop weather, or even predict it with any great success, because we lack deep knowledge about the basic components of weather (and climate). Fifth, it may be that we will never possess such knowledge…’ JoNova built one of her posts around this one of mine, which made me feel most honoured, since her website is on my Blogroll.

Over the weekend I receive another, this time less-than-gruntled, mention from another respected climate sceptic, Jennifer Marohasy, also on my Blogroll. She wrote that she was disappointed in those words, and that I was repeating statements that had been made frequently at the recent 9th International Climate Change Conference in Las Vegas (the annual climate sceptics conference). And she set out her objections in a long essay on her website, which is well worth reading.

Jen is a first-rate scientist herself, and she has published (yes, in a peer-reviewed journal), a weather-forecasting technique based on artificial neural networks that she claims provides more accurate medium-term forecasting of rainfall than the forecasts issued by the BoM. The paper is behind a paywall, but you can listen to Jen summarising it, with some slides as well, here. Artificial neural networks are a form of artificial intelligence, in which a model is trained to find patterns in observed data. I have no competence in this area at all, but if Jen and her colleague John Abbot have been able to produce forecasts that are systematically superior to those issued by the BoM we will all be in her debt. There are of course other people who issue medium and long-term weather forecasts. I think it fair to say that none of them has yet proved to be right all the time.

A quick word or two about weather before I move to her long paper. As anyone knows who is interested in weather forecasts and their accuracy, current official forecasts decline in accuracy quickly after three days. They are usually also given in terms of probability, which means that they are hardly ever completely wrong. My remark above I think stands as accurate: we do not know everything about the components of  weather and climate, and we may never possess that knowledge. But of course that might happen, if someone unlocks the regularities that some feel are there inside the chaos. I wasn’t consciously repeating statements made at the ICCC, since I wasn’t there, and I wasn’t following it online.

Let me now turn to Dr Marohasy’s paper. She did go to the 9th ICCC, and she came away feeling that the air of victory she observed there was premature. It was repeatedly suggested at the ICCC9 conference that those sceptical of man-made global warming have some how won the scientific argument. This is nonsense.

Why so? Because governments everywhere, even here, where the carbon tax has been thrown out, are still apparently committed to the consensus view that carbon dioxide is the villain — and they do so because all the official sources, like learned academies, say so. Well, I’m with her, in part. But the reasons governments behave this way is that they are conscious of the electorate, and there is a substantial body of passionate opinion out that ‘believes’ in ‘climate change’. She didn’t say so in her paper, but it is not going away at all quickly.

So government will do what they think they have to to balance the books, get the budget into shape, and so on. But they won’t frontally say that the science has moved on. They will throw out carbon taxes, and they will abolish subsidies for expensive alternative energy, and they will block demands for more wind turbines. But they’ll do so without saying that ‘climate change is crap’, as somebody once said — but that was before he became PM.

Her second point is that rebuttals don’t overthrown paradigms, and again I’m with her. There are abundant articles that are not ‘consistent with’ the current AGW orthodox, but the IPCC is able to ignore most of them. The retreat from the AGW scare will be a slow one, though I expect it to pick up speed. The paradigm might change, too, since so much of it is expressed in terms of measurements, as in the case of temperature itself, and ‘climate sensitivity’.

Her last point is that ‘natural variability’ is not a convincing explanation, and since I have said almost that myself, I can only agree with her. She writes: the claim is that ‘nature not human activity rules the climate’. But this tells us almost nothing. In many ways it’s a cop-out. It’s like a theory of electricity without any explanation of charge, voltage or magnetism. My position is a bit milder. While I would currently accept that the CO2 control knob is not a convincing explanation, and that there is something out there in nature which is more powerful, I would very much like to know what that something is. Not to know is intellectually unsatisfying.

Like Jen, I would like to see much more spent on real research into climate and weather than does not start with the propositions that carbon dioxide is a villain, or that fossil fuels are bad, or that we don’t need any more research.

I haven’t been to any of these ICCCs, so I don’t know exactly why Dr Marohasy was so nettled. And I do look forward to seeing her neural network approach to forecasting established as the best system now available.

  • Ray B

    Don, its Ray the Moderator from Jennifer’s blog. The 2 comments you recently submitted are stuck in moderation. They look disjointed & are actually posted on separate articles of Jennifers. Do you want me to trash them so you can start again?

    • Don Aitkin

      Ray,

      Thank you. Yes, please. Trash them and I’ll start again.

  • Doug H

    Thanks Don – to me the evidence is overwhelming that we don’t understand weather and climate well enough to make good long range predictions and may never do so. The fact that someone claims to be able to make better predictions does not change that fact and will not until the predictions are tested in the real world.

    As for whether or not the sceptics have won the scientific argument, they did so years ago and the facts continue to support them. The problem is that they have not won over enough of the press, academia and business, many of whom have reputations and money at stake and happily ignore inconvenient facts – like that we have only warmed by 0.7C in 130 years (according to the IPCC) and the only dangerous warming is in predictions, not the real world. In short, the sceptics problem is convincing the world at large that we don’t have a problem.

  • Jennifer Marohasy

    Don,

    I thought you might give us some explanation of why its not possible to forecast weather or climate?

    Some comparison of the methods used by the Bureau versus Artificial Intelligence?

    Perhaps make some comment about the Bureau moving to a method of seasonal forecasting last year that has limited if any skill as detailed in my open letter to Greg Hunt that can be read here… http://jennifermarohasy.com/questions-for-the-australian-bureau-of-met eorology/
    In particular… scroll to questions 6 and 7.

    As regards the one paper published… perhaps if you read my presentation to the Heartland conference you can see that I have significantly more than this published… http://jennifermarohasy.com/2014/07/the-need-for-a-new-paradigm-includ ing-for-rainfall-forecasting/ . I was simply quoting one paper in the blog post… the paper that I thought was most relevant because it gives an idea of our levels of skill.

    But the bottom-line is, as I explained in an article for the IPA last year, its not useful or correct for sceptics to keep saying we can’t forecast… https://ipa.org.au/publications/2216/competition-in-climate-science

    • Gus

      I believe that the IPCC forecasting methodology was roundly criticized for lack of compliance with scientific forecasting methods and principles. There have been papers published on this. Some distinguished Australian researchers are involved, e.g., Kesten Green (University of South Australia) alongside with Anthony Lupo (University of Missouri) and J. Scott Armstrong (University of Pennsylvania).

      And, yes, you can make mid-term weather predictions to some extent. The best way is to compare current weather patterns with known patterns and developments in the past. This is how medium weather forecast was usually produced in the past. This can work well, because effectively you rely on nature to carry out the computation for you: identical initial conditions should result in identical development, all else being equal.

      This is how the Farmer’s Almanac produced a perfect winter forecast for 2013/2014: everything panned out. In contrast, the prediction *computed* by NOAA failed woefully.

      Now, when you invoke your AI net, what you do, I presume, is what the Farmer’s Almanac had done: you train the net on past weather patterns and in this way you let it “predict” when it sees the current weather pattern unfold. Where this is bound to fail, is where all else is not always equal. For example, you may have a major volcanic eruption, that’ll disturb the weather patterns for years. Or you may have some unexpected development on the sun’s surface. Although the sun responds with quite good regularity to the pull of Jupiter and Saturn, it is, after all, a ball of chaotically boiling plasma. The cosmic radiation background that is modulated by solar activity and that affects cloud formation on earth is not constant either, so its response to solar activity may vary.

      • dlb

        Well the Australian BoM seem quite blinded by their new fangled GCMs. These GCMs do a great job for 5 days but like you I have serious doubts for any longer period. They remind me of dead reckoning navigation which suffers from cumulative errors.

    • Don Aitkin

      Jen,

      When I wrote the post I didn’t think it was necessary to give a full explanation. But my understanding is that with respect to weather, despite the great advances that have been provided through satellites, radar and computing, forecasting skill still drops off quickly after the first few days. Gus has set out well why mid-term forecasting also has its problems. As for climate, the GCMs still seem not to be able to deal with clouds, while ENSO is not included in the parameters, and cannot yet be because it too is not understood.

      None of that means that we should not try to do better, and I’ve never said so. Indeed, I have written several times that we need to devote more research to simple understanding of both weather and climate, and to better data. And to stop assuming that CO2 is the real cause of everything.

      In general, I use links to allow readers to follow the theme and find out for themselves. My reference to ‘a paper’ did not imply that you had only written one, and no such inference need be drawn. And I did, after all, direct readers to your own essay.

      I repeat that what you are doing is innovative and important. What will carry real conviction is continued success in mid-term forecasting that is plainly better than the current alternatives.

      • David

        “None of that means that we should not try to do better, and
        I’ve never said so.”

        Yes you have, many times. Your fallback position in many of the AGW
        debates on this blog is that “it is not up to you to have to propose alternative explanations for global warming”, which is your prerogative. But that is hardly consistent with a “..WE should not try to do better”.

        • Don Aitkin

          David,

          I’ll repeat the paragraph so you can see the context:

          ‘None of that means that we should not try to do better, and I’ve never said so. Indeed, I have written several times that we need to devote more research to simple understanding of both weather and climate, and to better data. And to stop assuming that CO2 is the real cause of everything.’

          ‘Doing better’ means obtaining a better understanding of the components of climate and weather.

          I have also said that while I don’t have to propose my own theory of the forces that cause changes in climate, I would be glad when something comes along that does give us a handle on what is meant by ‘natural variability’. In fact, I said as much in the post above:

          ‘While I would currently accept that the CO2 control knob is not a convincing explanation, and that there is something out there in nature which is more powerful, I would very much like to know what that something is. Not to know is intellectually unsatisfying.’

          In terms of the policy debate, it is enough to show that the CO2 control knob is so fatally flawed that it cannot be the basis of good policy. To do that one doesn’t have to be anything other than a decent critic. But I like good theories too.

          • David

            Ok lets look at the whole sentence.

            “None of that means that we should not try to do better, and I’ve never said so. Indeed, I have written several times that we need to devote more research to simple understanding of both weather and climate, and to better data.”

            Surely to “devote more research to simple understanding of both weather and climate” would include a willingness to ask questions and explore alternative possibilities to AGW, rather than try than strategically “protect your hypothesis” by refusing subject any of your own views to critical analysis.

            “it is enough to show that the CO2 control knob is so fatally flawed that it cannot be the basis of good policy’

            Prove it. Show me one climate model that is better for excluding CO2 as an explanatory variable!

  • Doug H

    Sorry Jennifer – I can and will say that we can’t accurately predict the weather more than a few days ahead and attempts to predict the climate these past 20 years have mostly failed dramatically.

    I used met forecasts for 20 years as an aviator and never met a meteorologist who had high confidence in forecasts for more than 3 to 5 days – and even then they talked percentages. I believe many today have some confidence as far out as 10 days, but past that it’s all flute music.

    As for climate predictions, I think we have enough knowledge from failed alarmist predictions these past 20 years to confidently state that they don’t know enough to make reliable predictions. They were not proved wrong by better predictions from sceptics, but by the earth (and possibly the sun) doing what what they have always done, whatever that is, to vary the climate.

    • Mike O’Ceirin

      Doug I agree entirely the long range weather (or was it climate) forecaster has a history of failure. To say it can be achieved by neural nets and AI plus some statistics is naive. When I studied both NN and AI at Uni they were the great hope to finally produce machines that would surpass the reasoning power of humans. As I remember there was even hardware neural network logic integrated circuits produced. The brain has neurons and axons which interconnect so with that bit of knowledge we should be able to program computers to divine the answer to the Ultimate Question of Life, the Universe, and Everything! The fundamental problem is that it is not understood how brains do what they do. Until that is the case attempts to produce AI with a computer are naive in the extreme. A similar folly is the building of something that looks like an aeroplane out of palm leaves as some primitive New Guinea tribes have.

      Now the forecaster of climate has the same problem. The fundamental problem is that it is not understood how climate or weather does what it does. Until that is the case attempts to produce a climate model with a computer are naive in the extreme.

      I have followed the BOM predictions by recording the forecast max/min temperature each night and comparing to the actual result. I was amazed to find up to 5 degrees variance and that the forecast is changed during the day. Those in the prediction industry have many ways to suck the gullible in and unfortunately none of us are immune.

      As to Jennifer’s belief that accurate forecasts are possible, well yes I would agree but only when we understand what we are dealing with and the same goes for AI. If that prerequisite is achieved I suspect then we will find computer hardware hopelessly inappropriate and inadequate for the purpose.

  • dlb

    The BoM were using statistical 3 month forecasting up to May 2013 when they changed to the GCM methodology. They still produce the statistical maps for comparison. When I get some time and patience I need to check their archived maps to see which method has provided the best forecast. It would also be interesting to know if the three month POAMA dynamic model uses the same water vapour and cloud feedbacks as the the decadal models of the IPCC?

  • Gus

    Have those “skeptical of man-made global warming” won the scientific argument? I’d say they have. They won it when nature sided with them by not warming over the past 17 years and a half, in spite of the atmospheric CO2 concentration increasing. They’ll win it even more when it gets colder. Our winter in the Midwest was unusually long and cold this year, although not unprecedented, but certainly not a sign of “global warming.” Our summer has been similarly cold so far.

    This alone is enough to say that man-made CO2 is irrelevant. If you want to be more cautious, you’ll say that “natural variability” may have countered any man-made contribution to “global warming,” but then you have to admit that the same “natural variability” may have been behind the slight warming over the last two decades of the 20th century. As it happens, there’s a perfect explanation for both: the sun was most active in 9000 years in the last two decades of the 20th century, whereas the activity ceased in the first two decades of the 21st and we see climate responding in perfect synchrony.

    The current prediction of solar scientists is that it’ll be getting progressively colder towards 2030.

    What about the rest of humanity then? What about the journalists, economists, businessmen, environmentalists, politicians, administrators, all those who put their bets on the horse of man-made “global warming,” and who are seeing their horse running slow? Are they going to cheat and shoot the leading horse in mid-race? Sure, they’ll try! Too much money is at stake. But the reason their horse is so slow is because it’s dead already, it moves forward, not because it’s got legs to run on, but because its dead carcass still has a lot of momentum.

    • dlb

      But Gus, who cares if the horse is dead, didn’t you know it was a one horse race? As far as most of academia is concerned there can only be one winner in this race, even if it takes 1000 years to show some form.

      • Gus

        No, it was never a one horse race. I live and work in academia, and interact with scientists daily, including earth scientists, atmospheric scientists (including meteorologists), ocean scientists, planetary scientists, astrophysicists, biologists and more. I also read scientific literature on the subject and, by this token alone, I have never seen this as being a “one horse race.” There was always vigorous discussion about this: in the literature, in seminars, in academic conversations “around the table.” The man-made CO2 story has never been accepted by geologists who simply knew better from the earth’s geologic history. It has never been accepted by physicists who did not see a proof of the idea, neither in observations, nor in theoretical considerations. On the contrary, physicists and engineers were the ones who pointed out fallacies in popular explanations of the supposed phenomenon and published critical papers on this in their own peer reviewed journals.

        Then you have scientists in countries other than the UK, US and Australia. Theirs is a different academic culture. They have their own journals. And so, you will find that the Dutch and the Norwegians have always looked at this with some skepticism, as have many Germans and Swiss scientists. And then you have Russia, China and India, whose scientists look at this in their own way and what they see is not at all what their US colleagues want them to see. This is also reflected in their publications.

        When you look at *some* of the US climate journals, you find that nearly every paper has to do with computer simulations. This is the Achilles heel of US science: there’s a diminished reliance on scientific observation and analytical reasoning and too much unwarranted trust and money invested in computer simulations. It is a self-perpetuating industry, driven by supercomputer centers (and their self-centered directors) and industry. The authors of the papers do not even have enough knowledge themselves to know what’s inside the programs their careers are built on. They just use what is given to them. Yet, the programs are highly deficient. They are physically incomplete and sometimes even physically defective. They use a variety of kludges, none of which are real science for starters and which introduce their own unphysical effects into the computed solutions.

        Only this year, in May, a paper was published in Quarterly Journal of the Royal Meteorological Society that showed how the ICON-IAP model *violated* the second principle of thermodynamics, because of its use of Smagorinsky closure. It is quite enlightening to observe that the paper was not published in a climate science journal, but in a meteorological journal, and not in the US, but in the UK!

        • dlb

          Guss, my comment was rather tongue in cheek, but very interesting to hear your observations. My only contact with academia in Australia is in ecological science. As far as I am aware they still accept the theory and are still publishing papers on what will happen to (pick species) under global warming.

  • David

    ….and as I have tried to explain to you many times, Don there is a big difference between using a statistical models to predict as opposed to explain. The differences are subtle but important. In my opinion you should undertake some practical steps to familiarize yourself with these concepts.

    • dlb

      David, are you saying statistical models are better at explaining than predicting? Do you think GCMs are better than statistical models for medium to long term prediction?

      • David

        I don’t know about GCM models. My concern is more fundamental. What I am saying is that statistical models can be used to both explain and predict. They are two different goals, which should
        not be conflated. Here is a standard OLS

        Y = a +b X_1 + cX_2 + error

        If you want to know how well this model can predict you look at the R_squared which will tell you how much of the variation in Y is explained jointly by all the explanatory variables on the left hand
        side of the equation, i.e. X-1 and X_2.

        If you want to know what effect X-1 has on Y then you look at the coefficient b plus a p-value for that estimate.

        It is incorrect to imply as Don, Gus and others do, repeatedly, that because Climate models “fail” (according to him but not Jennifer Marohsay) to reach some arbitrary level of predictive power that you cannot then draw meaningful conclusions about the relationship between an explanatory variable (e.g. carbon dioxide) and the independent variable (e.g. temperature).

        I will give you an example. The ability of a statistical model to predict accurately the severity of car crash injuries could be quite low. All sorts of factors may affect this. Even so, within the data we could still see that seat belts are correlated with a reduction in injury severity. So we make seat belts mandatory. But it would be wrong to dismiss this correlation just because the data can not predict with “sufficient ” accuracy the probability of a road crash
        injury.

        I hope this helps.

        • David

          sorry that should read

          by all the explanatory variables on the RIGHT hand
          side of the equation, i.e. X-1 and X_2.

        • dlb

          Thanks David. I think many statistical models suffer from the correlation vs causation dilemma.

          In your example the statistical significance of seat belts is justified as the introduction of seat belts were based on very simple physics and physiology, not to mention laboratory testing.

          Unfortunately the influence of CO2 on climate is not so simple as it has indirect influences on water vapour, clouds, the ocean and biosphere all of which are poorly understood. Despite what the “intellectual” main stream media might be saying, the earth crash dummy is barely raising a sweat.

          • David

            :) Yes you may be right

    • Don Aitkin

      I know you have, David, and with great respect, I think I actually know the difference. But I don’t know why you want to tell me again. GCMs, a form of statistical modelling, are used as one of the bases for weather forecasting, while the IPCC likes to talk about ‘scenarios’ and ‘projections’ rather than forecasts when it is referring to climate. But I don’t think playing with words counts for much there.

      Since I have actually devised and employed statistical models, I personally don’t think they ‘explain’. Rather, they provide clues to an explanation, which needs to be reached through other data and evidence. Their outcomes are suggestive, rather than evidence or proof.

      • David

        “Since I have actually devised and employed statistical models, ..”

        Could you provide a reference so I can evaluate them.

        • Don Aitkin

          No, can’t help you. My work appeared in a paper given to the 1969 APSA Conference, and was called something like ‘A Simultaneous Equations Model of Voting in Federal Elections, 1910 to [somewhere earlier than 1969]‘. It was based on work done in the US by Donald Stokes. The logic of it is to take every subdivision of every division in every election in every state in Australia and compare the outcome in terms of (e.g.) the Labor vote to the average for all subdivisions in that division, in that state, at that election etc.

          From those outcomes you can offer a proposition about how much local, regional, state and national effects have mattered in a given outcome. My interest was in whether or not the national effects grew over time. They did, but not in a regular way — that is, sometimes state and sometimes local or regional ‘forces’ depressed the growing tendency of national elections to mirror national issues.

          I did this in the early days of computing, and it was a lot of work. It was too technical to be published in the political science journals (that wouldn’t be the case today), and I was busy on a major work anyway— Stability and Change in Australian Politics, where it is referred to somewhere.

          I gave all my political science stuff away when I became an administrator, some thirty years ago. But if you are really interested, the ANU or APSA may have archives that include the paper.

          With respect to your response to dlb, the amount of variance ‘explained’ depends on the extent to which your variables cover all possibilities. The word ‘explain’ here has a technical meaning with respect to your constructed equation., and shouldn’t be conflated with the ordinary meaning. No statistical model, in the social sciences, anyway, can account for everything.

          • David

            “No statistical model, in the social sciences, anyway, can account for everything.” Agreed. That’s why they have an error term.
            I take your point about my use of the term “explain” How about identify a correlation?

            For interest take for example your Simultaneous Equations Model of Voting intentions. That model is designed to explain some aspect of voting behaviour. Specifically to explain how national effects grew over time. I am sure it was well designed etc.
            But it would be silly of me to start criticising your model because it could not predict voting intentions, because that was not your stated purpose. They are different goals which would require different models.
            So, it is fundamentally misguided to criticise a climate model that seeks to inform policy by quantifying a relationship between CO2 and temperature, on the basis that it does not “predict” temperature.

      • David

        “Rather, they provide clues to an explanation, which needs to be reached through other data and evidence. Their outcomes are suggestive, rather than evidence or proof.”

        Yes I agree.

  • Pingback: The history of climate « DON AITKIN

  • Mike O’Ceirin

    Beware fellow comments I was censored here I didn’t think what I wrote was offensive, but there you go. How about saying something rather than silent censorship.

    • Don Aitkin

      I’m mystified. What has been the problem?

      Don

  • Mike O’Ceirin

    Apologies I was mistaken sorry about that.