Bishop Hill Hulme on Nurse
Jan 25, 2011 BBC Climate: Hulme Royal Society Mike Hulme has published some thoughts on the Horizon programme, little of which will be disputed by sceptics. Here's a snippet:
I do not recognise [Nurse's] claim that “climate science is reducing uncertainty all the time”. There remain intractable uncertainties about future predictions of climate change. Whilst Nurse distinguishes between uncertainty arising from incomplete understanding and that arising from irreducible stochastic uncertainty, he gives the impression that all probabilistic knowledge is of the latter kind (e.g. his quote of average rates of success for cancer treatments). In fact with climate change, most of the uncertainty about the future that is expressed in probabilistic terms (e.g. the IPCC) is Bayesian in nature. Bayesian probabilities are of a fundamentally different kind to those quoted in his example. And when defending consensus in climate science – which he clearly does - he should have explained clearly the role of Bayesian (subjective) expert knowledge in forming such consensus.
Reader Comments (59)
I make that ten posts today! You on haggis today, Bish?
Thanks for what you are doing.
I was struck during that programme at how many of his (Nurse) assertions could be reversed and legitimately thrown back at him.
However one marvelous sequence was how he consoled Dr Jones over the numerous and coordinated FOIA requests Jones had received to reveal his data while never asking why the data was not freely available in the first place.
I very much doubt climate scientists explicitly exploit Bayesian probabilities - but maybe other readers can tell us where they are in the literature...
It is more likely that climate scientists are committing the prosecutor's fallacy with their p values.
Your Grace,
Forgive me for cluttering you pages.
There is a problem in science. Scientists maintain a certain courtesy to other scientists. If one thinks what another scientist is saying is total bollocks he keeps quiet since this is not his field of expertise and defers to the experts in that field (hence the idea of "concensus").
The second problem is that "peer review" is not rigorous analysis but "getting your friends to agree with it".
Both of these concepts were clearly present in the Horizon programme.
And Holland on Nurse:
'While waiting for the 00:01 am deadline I have just watched the BBC Horizon programme “Science under attack”, which was truly awful Climate Change propaganda....(more)'
http://www.thegwpf.org/opinion-pros-a-cons/2299-david-holland-comments-on-the-new-scitech-report.html
I'm not really sure what Hulme means by Bayesian (subjective) probabilities. As a maths guy, I understand the objective meaning of the term, but I wonder if it is being used in some Post-Normal context.
The Wikipedia page sheds some light:
Objective and subjective Bayesian probabilities
Broadly speaking, there are two views on Bayesian probability that interpret the 'state of knowledge' concept in different ways. For objectivists, the rules of Bayesian statistics can be justified by requirements of rationality and consistency.[1][4] Such requirements of rationality and consistency are also important for subjectivists, for which the state of knowledge corresponds to a 'personal belief' (rather than the objective state of knowledge in the world)
I was thrown by an early claim in the Horizon programme (some NASA bloke I think)
Humans annually add 7GT of "Carbon" to the atmosphere that completely swamps all other sources.
Is this correct? I've seen many "authoritive" claims that Man contributes about 3-5% of the total.
I've only got a slow Internet connection at the moment so can't check this for myself.
If this statement is incorrect or pea/thimble/move misleading then perhaps an apology from the BBC is due?
No, it's false. Natural CO2 is 20x greater.
How is climate uncertainty being reduced when we cannot model clouds? There is a divergence between instrumented temperatures in the 20th century and various climate proxies including tree rings. If this is so, how do we really know what temperature it was in the past if proxies could diverge in the past as well? Just two concrete examples.
Thanks Robinson, that's what I'd thought but was beginning to doubt myself.
Maybe I didn't hear the claim correctly but if, I did, then the BBC needs to be brought to heel for spreading false claims.
Can anyone corroborate that the statement was indeed made?
Sorry for making a nuisance of myself (again) but I'm finding it hard to believe that Nurse and the BBC would let such a (possibly) refutable "fact" intrude into a program that asked why the public is wary of establishment science.
andyscrase: "I wonder if it is being used in some Post-Normal context."
I too was very confused by that statement. They are clearly not using Bayesian methods to establish uncertainty, they are using simply p-values. I had no idea that the post-normalists had begun to distort this word as well. Thanks for pointing this out. To me, its inexcusable to use the term 'Bayesian" in a discussion of statistics and to *not* point out that you are using the term in a post normal religious sense. That is highly misleading.
I've now read Hulme's piece more carefully. I think it is nothing more than a justification for post normal religion supplanting the scientific process. BH says: "..little of which will be disputed by sceptics." I find the entire line of thinking to be illegitimate.
RoyFOMR
I think you will find that every other larger source e.g. ocean or vegetation is both a sink as well as a source of CO2. I have read that Human CO2 emissions in 2008, from fossil fuel burning and cement production, was around 32 gigatoones of CO2.
See e.g. this site
http://www.skepticalscience.com/human-co2-smaller-than-natural-emissions.htm
Thanks matthu but the quote made was about sources not sinks.
The quote was about where C/CO2 came from, not where it went.
Lets sort it out one at a time. First where, and how much, does it come from.
Second. Where does it go?
I enquired about the statement in the BBC Horizon program that related to genesis. IIRC, where it went to was not a consideration.
Many thanks for your input however. Much appreciated.
Judith Curry has a technical post on climate sensitivity here that has a number of references to Bayesian probabilities.
Andy,
Talk about Bayesian probabilities all you want.
You clearly think it's an issue and you are most certainly correct but that means about nowt to our esteemed leaders and the voting population.
Did the BBC mislead them, or not, that is the question that matters in the mindset of the more populous, less educated but more influential influencers.
We were told by the BBC, on Monday, that NASA man say we contributes more C stuff to the atmosphere than any other source. It's either right or wrong.
Which is it? Enquiring minds want to know.
Apologies for appearing to Hijack this thread. I didn't start out to do so but I'm really annoyed, rightly or wrongly, that the BBC has chucked a duck at us?
Sorry Bishop, if I'm out of order then please excise this post.
@RoyFOMR
I am just trying to understand what it is that Mike Hulme actually meant. I am angry about BBC bias too, but that can wait.
If we just dismiss everything as BS without investigating it, then it doesn't do much for the sceptic cause, IMHO.
I too am confused by the reference to Bayesian statistics; the systematic use of Bayesian methods in climate reconstructions appears to be extremely rare.
Hulme could be refering to predictions which are given in the style of a Bayesian posterior distribution (i.e. as a normalised probability weighting over mutually exclusive outcomes) but his emphasis of "... Bayesian (subjective) expert knowledge ..." seems to be more a criticism of the choice of "prior" probabilities in a Bayesian framework.
As far as I am aware, there is nothing inherently Bayesian about the vast majority of climate predictions which might be the subject of debate. Neither is there anything inherently subjective in the use of Bayesian methods: in fact, most, if not all, Classical methods have a Bayesian interpretation in which previously implicit assumptions are made apparent.
"Science is under attack", claims Sir Paul. I watched his programme expecting to see examples of the profession's falling reputation.
Instead he spoke at length about the loss of confidence in the pattern-extrapolating pseudoscience of Climatography. Good! To demonstrate that the hoi polloi have begun contradicting expert science, he presents us with a poor man who treats his AIDS with yoghurt. I wish Sir Paul had turned to camera and said: "See this bloke? You Global Warming deniers are him, you are!"
He is tacitly claiming that an attack on one dodgy branch of science is an attack on all science. Wrong! An attack on one crooked banker is NOT an attack on the entire banking profession. And, just as honest bankers seek to distance themeselves from a disgraced colleague, the genuine sciences must cut Climatography adrift and dissociate themselves from this dumb numerology.
Their idea of reducing uncertainty is making the output of the models match up. Eventually there'll be no uncertainty there, but it won't mean they'll be right.
My clear memory is that the NASA guy said man produces 7 gigatons of CO2 per year which swamps that from allother natural sources. It raised my eyebrows at the time as contradicting everything I had read on the web.
Thinking about quantities of CO2 in the atmosphere: according to one graph I've looked at CO2 has risen from 290ppm to 390ppm since 1880, about the first half measured by ice cores and the second measured at Mauna Loa. Historically, CO2 rise has always come after temperatures rise by a very approximate figure of about 800 years. So how can they possibly know which portion of the CO2 rise is anthropogenic and which is a result of the Medieval Warming? How can they even identify anthropogenic CO2? Just asking. Please correct any mistakes here.
Roy FOMR
You'r right about CO2. Our contribution is tiny in comparison with the total natural turnover. Which means that the natural sources are quite capable of sponging any excess CO2 so if CO2 is still rising, that's because the natural sources themselves are doing this. It's the oceans, the ultra-slow thermohaline, still recovering from LIA, outgassing on balance.
http://www.greenworldtrust.org.uk/Science/Scientific/CO2-flux.htm
Neal Asher:
"Historically, CO2 rise has always come after temperatures rise by a very approximate figure of about 800 years."
So, what was happening 800 years ago I wonder?
The MWP perhaps?
Peter Walsh
Nurse should be called out on his claim that anthropogenic carbon emissions are dominant over other sources. He is the President of the Royal Society and should not be allowed to make unsubstantiated claims which contradict the best data we have on carbon emissions. Trust only the data Sir Paul! Or does that only apply to other scientists?
I refer you to the IPCC Third Assessment Report (I know, admittedly not always a reliable source) http://www.grida.no/climate/IPCC_tar/wg1/pdf/TAR-03.PDF
Total annual take up of the Biosphere = 120PgC/yr, (over 15% of total atmospheric CO2)
Human emissions in the 1980s averaged 5.4 +/- 0.3 PcC/yr and in the 1990s averaged 6.3+/-0.4 PgC/yr
Increase in atmospheric CO2 due to Human activity = 3.2+/- 0.1PgC/yr in the 1980s and 3.2+/-1 PgC/yr in the 1990s.
The increase due to Human Emissions is calculated from Oxygen 18 isotope ratios of the atmospheric CO2
the original sources for these values are (Farquhar et al., 1993; Ciais et al., 1997)
On the subject of degassing, I was highly dubious about the claim on the Royal Society website that CO2 would remain in the atmosphere for "over a thousand years", and sought to challenge it, purely out of curiosity. The Mauna Loa data shows, yes, a year-on-year rise but a steep downward slope every northern summer, when the NH vegetation is hoovering it up. The peak rate of decay of that curve occurs between July and August. I get a half-life of 123 months. This assumes, of course, exponential decay, a reasonable assumption I think.
Admittedly we'd have to chock the planet in its August orientation - quite a hefty geoengineering task - er, that's mild humour - for atmospheric CO2 to actually halve in ten years. The point is that the ecosystem has a large capacity for absorbing this gas and the RS's claim that its residence time is a matter of millennia was flat wrong. The outrageous assertation has now been removed from their website; the tone of AGW hysteria has stopped, I'm pleased to say.
As a check, I found that the half life from the earliest data (late 1950s) matched the 2008 result nicely, 121 months and 125 months respectively. (If anybody's interested, the July 2008 CO2 concentration of 386.1 PPM would fall to 193 PPM 125 months later, and the July 1959 figure of 316.5 to 157 121 months later. Assuming that the world stops turning.;)
On CO2, AR4 WG1 Ch7 gives these numbers for the natural carbon cycle-
http://www.ipcc.ch/publications_and_data/ar4/wg1/en/ch7s7-3.html#7-3-1
Which has some uncertainties. The NASA guy may have been quoted out of context when comparing gross CO2 emissions from fossil fuel burning to residuals from natural processes, but if he was doing a like for like comparison then he seems to have been wrong.
My two cents on Nurse and Science Under Attack.
Let us not forget the sceptical Dr John Snow, and his fight against the consensus of miasma theory. Were Sir Paul Nurse somehow transported to 1854, into an area afflicted by an outbreak of cholera, would he as lief listen to the ninety-nine doctors who accepted that cholera was caused by foul odours, or the one doctor who asserted that contaminated water was to blame?
'Hulme on Nurse' - blimey - apologies for the expletive, Your Grace - but I thought for one delightful moment that it was CHRIS Hulme, our beloved Secretary of State for Energy and Climate Stuff - that had gone all sceptical..!
MIKE Hulme - oh, right...
Actually its Chris HUHNE, isn't it..? I really am having a senior moment...
Your emminence
As a scientist/technologist myself (BSc Chem Eng UMIST 1960) I found a necessary facet to my (reasonably successful) career supervising engineers and technicians to be an ability to recognise bullshit even when it was well camouflaged. AGW is just that - complete nonsense. The science is straw and it is promulgated by a small coterie of scurrillous pseudo scientists making loads of dosh from it. There must be thousands just like me - we write to our elected politicians, we write to papers, we blog ... but is government talking any notice? No ... . it just remains in denial and keeps spending on more windmills that deliver zilch. I could also write as a Yachtmaster (which I am) to pioint out that our climate is dominated during the winter months by anti-cyclonic conditions - high pressure, light and variable winds - no output. This is not supposition, the electricity output by sector is available in real time. Are they ever going to listen?
In the Nature.com article: "More knowledge, less certainty" (http://www.nature.com/climate/2010/1002/full/climate.2010.06.html), the prominent IPCC climatologist Kevin Trenberth says:
"So here is my prediction: the uncertainty in AR5's (next IPCC report) climate predictions and projections will be much greater than in previous IPCC reports, primarily because of the factors noted above.".
This flatly contradicts Nurse's claim that “climate science is reducing uncertainty all the time”. The reality is that as climate scientists learn more about the climate, they realize there is much more about the climate that is not understood than they had previously realized.
Anyone out there tell me I'm wrong?
dreadO: "Hulme could be refering to predictions which are given in the style of a Bayesian posterior distribution (i.e. as a normalised probability weighting over mutually exclusive outcomes) but his emphasis of "... Bayesian (subjective) expert knowledge ..." seems to be more a criticism of the choice of "prior" probabilities in a Bayesian framework."
That is the way I read it also -- with one important twist. An expert prior in a Bayseian method has a specific meaning. An 'expert' is a class that had demonstrated previous predictive power (it could be a model or in some cases a person using tools). Once predictive power is established, then that expert can be used for prior estimations. What I think these guys are saying is: "hey, we're experts -- we have PhDs, therefore we should be to pick any prior we'd like without challenge". That's not Bayesian method, that's pure post normal gibberish. Climate scientists, as a class, have not established predictive power in a Bayesian sense. For them to declare THEMSELVES experts under the Bayesian method is more an attempt to "embrace, extend then corrupt" the word.
It would be easy to splice parts of Nurse's propaganda piece that are false into a response video and follow each fallacy with the facts that contradict Nurse's fallacies. Alarmists such as Nurse are particularly vulnerable because they are very prominent and vociferous but actually have little knowledge of climate science so they are prone to gaffes.
After making the video, just post it to youtube and circulate the link on WattsUpWithThat, InstaPundit, DrudgeReport and other popular websites.
"I very much doubt climate scientists explicitly exploit Bayesian probabilities - but maybe other readers can tell us where they are in the literature... "
Here is an example of (mis)use of bayesian statistics in climate science. In this particular case they demonstrate the common misconception that bad data can somehow be turned into good data by bayesian statstics:
http://www.nature.com/nature/journal/v462/n7275/full/nature08686.html
If you look at Mike Hulme's research interests and biography at UEA (http://www.uea.ac.uk/env/people/facstaff/hulmem), you find that he has produced reports for WWF and written monthly columns for the Grauniad. But it is very difficult to find any traces of any real scientific work that he has actually done. All his publications seem like - well it is hard to find an adjective to describe the titles of the publications. Go and check out the titles and see if they make any more sense than a lot of the things you hear him say. There's nothing of any value to society. We pay for this claptrap disguised as science.
Would Puffball work as the word Phillip.
Looks good on the outside but no substance inside ;) .
Unlike Nurse's script, Hulme's comments do convey 'thinking man at work'. Nurse's script on the other hand conveys 'spinners be here'.
Breath of fresh air
Yes "puffball" is a good word. I also thought of "fluff" - light and airy and good for nothing.
Many thanks mpaul, I wasn't aware of the specific meaning of the word expert in relation to the selection of priors. I'm not a statistician or mathematician by training, just inclination, so all corrections or insights are most welcome.
What Mike Hulme is talking about is the distinction between "Bayesian Belief" and probability. Mathematicians working in expert systems (as in computer science) have tried to formalise rules for human knowledge, belief, and reasoning, so they can get computers to do it. One particular variety models knowledge and belief by borrowing the rules for reasoning with probabilities - especially Bayes theorem. There are some others - Dempster-Shafer Belief, Fuzzy Logic, etc. but the only one with any mathematical support is Bayesian Belief (it is at least consistent and always gives sensible answers to sensible questions). Bayesian Belief is *not* probability, although it looks and acts like it, and the same language is often used.
Bayesian Belief is sensitive to a person's state of knowledge, and so is a very subjective concept. An event can have only one probability, but can have several different Bayesian Beliefs, depending on who you ask and when. For example, they question "Did you just pick up the Ace of Spades?" has a probability of either zero or one, but because I don't know what the card was, as far as I'm concerned, the "probability" is 1/52. You just cheated and peeked and saw that it was a black ace but couldn't see which one, so as far as you're concerned, the "probability" is 1/2. How can an event that has already happened have two different probabilities? Because they're not probabilities, they're examples of Bayesian Belief.
A lot of what you learn at school that gets called probability is actually belief - for most practical purposes it makes no difference. The philosophers of mathematics do gets would up over the war between the Frequentist school and the Bayesian school, but the subtleties pass most people by. Probabilities can only be defined in theory - you can't actually observe them, so everything that comes out of perception or an experiment is actually a belief anyway.
In this case, Hulme is referring to an even more subjective use of the term: - expert judgement ("gut feel") is sometimes given a number and treated as if it were a probability according to Bayesian rules. This was often done by those computer scientists, who tried to get computers to reproduce the intuitive expertise of subject matter experts.
Or to put it in terms the layman might understand more easily - it's a a subjective expert opinion dressed up as mathematics, but with no mathematical basis for believing it. However, it is, remarkably, at least semi-respectable in science. Make of that what you will.
I like what you say, Nullius in Verba, although I am not sure about everything coming out an experiment being a belief. Some of it is mere data after all. Here is my shot at introducing Bayesian statistics in our current context:
With regard to Bayesian statistics, the key point is picking up on what some call subjective probabilities. These reflect the judgement of the observer, e.g. I think horse A will run with speed X, which is more than the others and I therefore believe it will win the 2:30. This is different from the coin-tossing, coloured-sock-selecting probabilities most encounter during elementary statistics classes.
I can say that I judge the impact of CO2 on, for example, mean temperature at location X over time period Y in epoch Z to be well-represented by such and such a distribution of temperature shifts, with my best 'guess' in the 'middle'. In the absence of CO2 increases, my distribution is such and such, somewhat different if I believe CO2 is an important factor. As real data comes in I can convert my 'prior distribution' into a 'posterior distribution' which reflects the new knowledge provided by the data. Now if I had what is called a strong prior, the data will have relatively little impact unless they are far from my guesses. If I had a very weak prior, my posterior distribution will be more strongly influenced by the data.
An example of a weak prior is to assume equal probabilities for all possible outcomes. I was reminded of weak priors when I saw the Met Office forecast for the UK winter revealed on this blog. For all practical purposes, they had allocated approximately equal probabilities to each of the 3 outcomes which they saw as spanning all the possibilities. This was an admission of a very low level of knowledge.
Now I suspect, but here I make an idle speculation, that Hulme finds Bayesian statistics congenial, as do perhaps some climate scientists, because they allow the language of statistics, the talk of distributions and so on, to be launched on no more than the judgements of the participants, perhaps as reflected by the computer models they have constructed to illustrate those judgements. In this way, the paucity of data about our climate under various circumstances is coped with, and expert judgements get another platform. It is a very respectable area of Statistics, and provides an elegant way of demonstrating, and conducting, the modification of our expectations as each new dollop of real data arrives.
Statistical problems have been classed (e..g. by Deming) into 'analytic' and 'ennumerative'. The latter can be answered by gathering enough data, e.g. what is the average weight of the socks in this drawer? The former may not be, e.g. why are there more red socks than any other colour? Many questions in climate are ennumerative, e.g. what was the rainfall level, suitably defined, here yesterday? Many are analytic, and require not just data but theory and insight into what causes what. e.g. why is rainfall higher here than somewhere else? Both types are important, and can be approached, if not always answered, by Bayesian and by non-Bayesian methods. The non-Bayesian approach pays no heed to prior beliefs, and just lets a data set speak for itself. The Bayesian approach would say this would unreasonably disregard what we already know, or believe to be true.
The two approaches, Bayesian and non-Bayesian, converge as data grows to dominate the analysis. The non-Bayesian approach merely differs in not admitting those 'subjective probabilities, or judgements. I suspect most of us are somewhat Bayesian whether we are aware of it or not, but I myself have never used the technique in my own toils as an applied statistician in industry.
A Rumsfeldian epistemic analysis:
The modal 'climate scientist' believes: (i) there's a lot that he knows that he knows; (ii) he knows that there's not a lot that he doesn't know; and (iii) there's virtually nothing that he doesn't know that he doesn't know.
The sceptic believes that the modal 'climate scientist' (i) doesn't know much of what he thinks he knows (either because it isn't true or because it isn't justified); and (ii) completely fails to appreciate how much he doesn't know.
[This formulation presupposes that the 'climate scientist' is honest. All bets are off for the crooks.]
John,
That's a good way of putting it.
For more thoughts, see the conversation starting here:
http://judithcurry.com/2010/10/17/overconfidence-in-ipccs-detection-and-attribution-part-i/#comment-4980
and here, where I try to explain it slower...
http://judithcurry.com/2010/10/24/overconfidence-in-ipccs-detection-and-attribution-part-iii/#comment-5874
You might also be interested in my thoughts on the distinction between confidence and likelihood, which I suspect is quite important for interpreting what the IPCC mean by their probability statements, discussed here:
http://judithcurry.com/2010/10/29/uncertainty-and-the-ipcc-ar5/#comment-6742
Jane Coles you have hit the nail on the head.
Indeed and I should add that analoguous to ignoring the Bayesian probabilistic setting of all climate change extrapolation, also the heterskedasticity of the crud data is ignored by warmists:1998-2010 data should get more weight in any regression analysis.
and unit roots analysis , of course.
I missed the nurseship elaborating on unit root analysis
yellowstone or a comet or nuclear war, or one of Paul Mason's abysses get very little probability points in the warmists extrapolations.
When tomorrow any of them would happen, suddenly we will see the climate change models substantially "adapted".
That is strange. We know they can happen today. Something is scientifically very wrong