Sunday
Jan022011
by 
Bishop Hill
Bishop Hill Supercomputers
Jan 2, 2011 Climate: MetOffice Climate: Models Autonomous Mind has interviewed Joe Bastardi and Piers Corbyn about the use of supercomputers in long-range weather forecasting. I liked Joe's last word on the subject:
It’s not the computer, it’s the limits of the computer in trying to adjust to what only men can understand and use. I dont think you need more money to arrive at the wrong answer faster. Should put it into fighting hunger, or giving men a chance to be free enough to dream and pursue that dream… much better causes in my opinion.
Reader Comments (33)
You should have also quoted Piers, just to emphasise how much of the near £200million a year that the met office are wasting on our behalf
In WeatherAction my Solar Lunar Action Technique (SLAT) does involve a number of equations and theoretical concepts (Weather action indicators) and calculations which are all performed on a pretty low level PC.The key thing to understand is that all weather circulation patterns have near enough happened before; the key is to find out when and how this time around they will be not quite the same as before.
"on a pretty low level PC......."
The Met Office will be able to come up with their worse-than-chance predictions so much more quickly with their shiny new supercomputer. Or did I miss something and they are actually going to correct their model?
As I posted in the Slingo on climate models thread:
The old saying about the workman blaming his tools for his crap results come to mind. "If only I had a better saw/supercomputer I could have made a door/weather prediction that was actually some use".
Piers has posted his January 2011 forecast for Britain and Ireland at http://climaterealists.com/index.php
He's forecasting one of the three coldest Januarys in 100 years. Bad up north!
As someone who studied Maths at university, and now uses computers all the time in my job, I am 100% behind the statements expressed by Bastardi and Corbyn.
I think Lindzen has said a lot on this topic too.
doubt if Joe Bastardi and Piers Corbyn will get any honours!
New Year Honours: Robert Napier
http://www.building.co.uk/news/2011-new-year-honours-robert-napier/5011000.article
I just wish the Met Office would just get their presentation of the current computer forecasts to vaguely match up. I often use my puter to check the latest Met Office predictions for where I live via the BBC: The 24 hour forecast, the 5 day forecast and the local TV weather bulletin. As per usual none of them match up. The 24 hour forecast tells me that tonight will get down to -1, whereas the 5 day tells me it will get down to -3. This lack of coherence is totally normal. The latest onlive TV bulletin says -5 for tonight, though that was recorded a few days ago. Do they expect us to not notice these discrepancies?
Corblimey, I thought that chaos theory had entered the body of science a decade or two ago. Wasn't it the flippin' weather forecasters who first realised that chaos placed mathematical limitations on the limits of their craft: butterflies' wings and all that? Were they not the first ones to say that, even with an infinite amount of initial-condition measurement and infinite processing power, their predictions were subject to hard limits?
So what's all this nonsense about bigger Crays?
These charlatans are cynically spinning out the global warming myth. Shame on them for so deceiving our governments into wasting a king's ransom on this folly. Shame on the Chief Boffins, master-schmoozers and apparatchiks par excellence, for giving our governments the very opposite of sound advice.
Brent Hargreaves: "Wasn't it the flippin' weather forecasters who first realised that chaos placed mathematical limitations on the limits of their craft: butterflies' wings and all that?"
Nice reminder, Brent; worth noting.
Agree with James about the Texas sharpshooting approach to weather forecasts as practised by the comedy duo of the "Not only the BBC but the Met office too!" show.
Yup, James, they're just "having a 'larf."
Weather forecasting is now an 0845 number, vote now, mobile calls may be more expensive than land-lines and, do, seek bill-payer permission before dialling, candy-floss service that passes for public information in a Nation which has rolled over to get its tummy tickled by sandal-suckers!
Weather is soo yesterday. Climate is the new black. Getting the weather of next week, next month or next Season wrong is too stressful. Get it wrong and the great-unwashed start moaning about how their water went away on a fortnights holiday or their heating bills turned into RBS pension plans, sans Gubment bailouts or just that some old bloke or lady just froze to death.
Weather is tricky but Climate is easy. Ten degrees of variation with forecasted weather
Temperature (units irrelevant) may cost a career while tenths of degrees, in 50 years keep the money coming!
Weather is now the poor relative of Climate. The BBC can't even be *rsed putting numbers on their charts. Colours are so much Cooler.
Post Normal Science, Post Numerical Society and Posted Missing.
RIP, UK
This comment by Piers also caught my eye:
'For a small fraction of the extra money they want to waste on supercomputers we could reliably forecast extreme events and general weather development details across the WORLD many months ahead.'
I believe him.
According to the people I've spoken to, they do not propose to predict weather, only climate, by averaging an ensemble of various models outputs. Precisely how the results are constrained to stop them shooting off into lala-land was not clearly explained to me, but I was assured that very good confidence indeed can be placed on them.
... but I was assured that very good confidence indeed can be placed on them.
Confidence in the sense of "likely serve the funding goals of the CAGW field in general, the funding aim of his/her 'research' establishment in particular and, most especially, the career aspirations of the assurance giver", that kind of confidence?
Where are the checks and balances? We chuck all this money at the MET office, yet their "climate" predictions are still as shakey as the 1970's. Don't they have to show some form of progress report?
I can save us a fortune, these climate forecasts will be good for the Next 10yrs: I can't see then going as far as Ice age predictions, I think they'll be talking about maunder/dalton, and reminding us there is heat in the pipeline (winter meme). While bemoaning the drought/wet summer as clear signs Co2 is disrupting climate (summer meme).
Yup, that's all the climate forecast we need, as much use as it ever was.
The problem with the computer models is incorrect optical physics introduced by Sagan's ex-students Lacis and Hansen at NASA/GISS. The 'two-stream approximation' predicts polluted clouds with smaller droplets transmit less light so cool the earth more than unpolluted clouds. This 'cloud albedo effect' part of 'global dimming' comes to 175% of the raw median AGW in AR4 and is the main reason why the IPCC claims three times higher future warming from CO2-AGW than the unamplified case.
However, next time you are out on a rainy day, look at the clouds. When they are about to rain, less light is transmitted. But hold on here, the droplets in rain clouds are larger. So, all the IPCC computer models are wrong. Instead of biased internal diffuse scattering [deduced by assuming constant 'Mie asymmetry factor', wrong because in his 1908 paper, Mie assumed a plane wave and that condition breaks down in a sol] there's probably substantial direct backscattering at the upper cloud surface, strongly dependent on droplet size, and symmetrical diffuse scattering.
So, 'cloud albedo effect' cooling in AR4 is imaginary for thicker clouds. But there's more: if it's not cooling, it can be heating, another form of AGW. Indeed, net CO2-AGW may be zero via a Miskolczi-type mechanism and no-one can prove otherwise. This abysmal failure to get the basic science right has lasted for over 50 years. For those who know climate science, Twomey didn't agree with Sagan and warned that his [Twomey's] analysis which correctly predicted 'cloud albedo effect' cooling for thinner clouds could not be extrapolated to thicker clouds.
In 2004, after experiment had shown no evidence of 'cloud albedo effect' cooling, Twomey was given a prize and NASA published a new 'reflection explanation' of the effect, fantasy physics apparently intended to purport that Twomey's observations could be extrapolated to thicker clouds. When AR4 was published, insiders probably knew it was incorrect. If true, this may be the biggest scientific fraud in History and probably explains the warming bias in the Met. Office model.
The problem is that in order to correct the models, the people connected to the WWF and its carbon-offset operation, who apparently control the Met. Office have to admit that the CAGW argument has absolutely no supporting theory or evidence. It'll be an interesting 2011 as the scientists in the Met. Office reassert their control over the activists.
But only a year ago didn't Slingo report to the Parliamentary inquiry into CRU that they had every confidence in the climate models because they used the same platform for the 5 day forecasts? What has changed in a year?
As Peter Taylor pointed out a few years ago, the Met Office urgently need an expert in the jet stream. Their desire for yet more super computers to run linear models [which apparently don't take account of a negative AO] just shows how little their understanding of what is driving the cold winters we are now experiencing in north-west Europe.
The position of the jet stream is historically-proven to be very much dependent on solar factors. It's probably simple heat transfer theory: low UV output means a thinner thermosphere and lower impedance for IR emission from the stratosphere to space. That in turn controls all the linearly-connected convective heat transfer process down to ground level hence the Hadley cells cool earlier and the jet streams move nearer the equator.
One interesting deduction is that the control system limiting GHG warming with increasing CO2, the earlier transformation of latent heat into precipitation and convection of drier air, may reduces the severity of storms elsewhere because of the southward shift of the downward convection at the extreme of the Hadley cell. The deserts will shift nearer the equator.
[I'm not a climate scientist or meteorologist. It seems the scientists in climate science know too little basic physics to be able to handle the complexities. Proper physicists can't handle the control systems which professional engineers know how to handle.]
If a bloke can't drive an Escort van safely and get where he's supposed to be on time, with contents undamaged, who would believe that putting him behind the wheel of an articulated lorry would make the slightest sense?
Joe Bastardi's quote is one of the best to describe the hubris of the AGW crowd.
Well I'm now going to play the BOF (boring old fart) with the comfortable homily that computers, super or not, will never replace human skill and judgement to any extent unless rethought out from top to bottom (and then only perhaps). The reason is simply that computers in their present state are "square" in the old nineteen fifties sense of the word when it meant the opposite to "cool"; unfortunately the latter is the only one of these two epithets to have survived devoid of its meaning amongst the young into modern times. As far as I can see, no matter what the resolution and sophistication of a computer it only works through linear "reasoning" and will never argue with your reasoning and only confirm what you've told it you want to hear. The problem is that the people who operate computers have tended to become as square as they and in placing their faith in them, once they get an inkling that they may possess little or no skill will always believe that the answer lies in ever increasing teraflops — I can't perceive what a teraflop is except a description of what the MO's long-range forecasts have recently been (sorry that's a bad Prime minister's question time snark I couldn't resist but somebody's going to pinch it, you'll see).
Both Joe and Piers have clearly implied that they neither want nor need these lavish Christmas presents from government. I don't know about Joe, but what Piers obviously needs, whether from govt. or private sources is funding to expand his premises, probably his personnel and certainly means to improve the dissemination of his results so they can be of immediate use to people.
More power to them both I say.
The quote from Bastardi resonates in two ways.
First, "the limits of the computer in trying to adjust to what only men can understand and use" evokes Michael Jackson's telling phrase about all software models struggling to cope with the "informality and unboundedness of the real world". That's Jackson the veteran UK software engineering thinker (and a friend and colleague in days past), not the one with the white gloves. Bastardi gets me thinking again that weather is of course a human construct - hence the informality as well as the unboundedness. Much to chew on.
If that was all, it would be immense. But there's something much more important. Instead of re-heating the last sentence, let me make my own New Year prediction. Unless and until the 'climate sceptic' community has this concern front and centre we will never change the world for the better. And that's what is needed, because colossal harm is going to ensue - in fact has already begun to ensue. But it won't get sorted until our values and our vision go much deeper. I've no doubt Bastardi's a real, live, breathing human expert on the complexity of weather. But in this moment, precisely as a human being, he without apology moves over to something far more important. We have to become like him.
lapogus,
"But only a year ago didn't Slingo report to the Parliamentary inquiry into CRU that they had every confidence in the climate models because they used the same platform for the 5 day forecasts? What has changed in a year?"
There's a long history of End of The World predictions. See "Extraordinary Popular Delusions and The Madness of Crowds" by Charles Mackay. E.g. Whiston's prophecy that the world would end on 13th October 1736. When they failed to come to pass, the prophets often brazened it out and put off the fateful day for a few years.
It seems to be a part of the human condition that we are periodically taken in by charlatans and crazes of various kinds.
"Extraordinary Popular Delusions and The Madness of Crowds" is essential reading for anyone trying to understand the CAGW phenomenon, cataloguing as it does the various mass brain holidays large numbers of people have collectively taken.
As for Slingo et al., does anyone seriously expect them to admit that the jig is up? It was put to me a long time ago that a large part of not losing is to insist that you haven't lost, whatever the evidence is to the contrary.
Of course the problem with the MET Office isn't their almost-brand-new £30M Supercomputer or its proposed replacement with one four times as powerful. They would be just as dogma driven and incompetent if they had the computer power of the Pentagon on one hand or an old Amstrad on the other.
The problem is that they are staffed up with hard line ecotards like Pope & Sligo, who take every opportunity to promote their AGW religion. And led by an even more hard line ecotard like Robert Napier - just awarded a CBE for "Public Service". (A bit like awarding one to Harold Shipman for "Patient Care".)
And it isn't just a cruel stroke of fate that led to him getting this post, it was because the Government and senior figures in the Civil Service chose him precisely because he is a hard line ecotard.
If in any doubt, check out the 2010 report "Zero Carbon Britain 2030", which has the MET Office (together with the University of East Anglia and other usual suspects) as "partners".
This is firmly in the pixie dust and away-with-the-faries end of the energy debate. Note that even LoonyHuhne only pretended that the UK could be "zero carbon" by 2050 in the LibDim election manifesto.
"Zero Carbon" by 2030? Really? And we have a publically funded body like the MET Office on board?
And you are surprised that their forecasts are crap?
"But only a year ago didn't Slingo report to the Parliamentary inquiry into CRU that they had every confidence in the climate models because they used the same platform for the 5 day forecasts? What has changed in a year?"
But Prof Sligo has form on major shifts of position over a 12 month period. Does anyone remember this from 2009
http://wattsupwiththat.com/2009/06/20/the-met-office-brings-doom-to-a-place-near-you/
Hello, it's me again. And if it's OK with the rest of you, I'd like to have another rant about the Met Office's fine forecasting work.
I'm a good boy - I have a very low carbon footprint. My chosen method of transport is a piddling little moped. As a result of that, when it's cold, the overnight temps are rather important to me. This is because, if conditions are good, a moped is a very cheap and efficient way of getting around. However if conditions are icey or snowy than a moped is essentially a death machine.
So I check the forecast for tonight. The 24 hour forecast, as provided to the BBC by the Met Office, is that tonight will get down to 0C. That's good news, because as I speak it is sleeting. 0C means that there's unlikely to be much freezing and I won't die. However, the newly updated 5 day forecast (available on the same BBC web page) tells me that tonight will get down to -2. That is bad news, as it means I will die. Just for clarity, I watched the online version of the latest local TV bulletin (also by the Met Office, also on the same BBC web page) and it tells me the temp will be -1. Incidentally, neither the 24 hour or 5 day forecasts mention anything about sleet. The TV bulletin mentions it as a possibility.
So will I die tomorrow or not? Who can tell.
BTW, I work as a gardener, so the weather reports in general are rather important to me for work. Or they would be, if it was possible to use them to actually know what the weather is going to be like. I work in a team of about 10 gardeners, and most mornings when we first arrive in we all have a chin-wag about what jobs we should do, given the weather. We all get our weather reports from the Met Office via the BBC. Some get it from the radio, some from the national TV weather forecast, some from the local TV forecast, some from online. I don't think there's ever been a day when more than two people actually agree on what the weather has in store for us that day. "The rain's arriving at lunch." "What? It's raining all day." "No it's coming over this evening." etc. etc. etc.
The only one of us immune from all this nonsense is Eric, who has completely given up on weather forecasts, and thinks we're all incredibly stupid for paying any attention to them. And he's the one who nine times out of ten is right about what the weather will do. He's not psychic, he just looks at the current conditions, compares it to what he's experienced before, and takes a guess.
To be honest, my feeling is that the cause of all this nonsense is very simply that all the new BBC/Met Office ways of presenting the weather give the firm impression that the numbers they come up with are far more accurate than they really are. In other words, what they really mean about the weather tonight is that it'll be somewhere around 1 to -3ish and maybe sleety, maybe not. But they never put it like that, do they?
End of rant.
James, does your moped have a two-stroke engine? If yes, not such a good boy then.
John in France
Oh shame on you! ;-)
Sod the engine. What a great comment from James Evans.
Perhaps I am being naive but I always wonder why weather forecasting is not done by comparing the weather conditions of the last 2 days to see if there is another 3 day period with the same first 2 days matching current conditions and then using this to predict day 3. This would be a simple pattern recognition problem and one which computers should be able to do without needing massive resources. The only problem is ensuring enough data - pressure maps, windspeeds, temperatures etc.... I suppose that the UK met office must have enough data e.g. say 50 years times 365 rolling 3 day windows gives about 18,000 historical scenarios. And if it does not match these closely enough, perhaps we should just interpolate the 3rd day from the closest matches. Perhaps this would do better than solving the Navier Stokes equation a few billion times on a system which is fundamentally chaotic.
James - I agree with much of what you say (and as a cyclist sympathise). I still look at the Met Office's 5 day forecasts, which I find are reasonably accurate for what might happen in the next 12-24 hours, especially if there is a standard Atlantic low pressure system coming our way. But I also check the Norwegian equivalent to see what their models suggest - e.g.
http://www.metoffice.gov.uk/weather/uk/gr/braemar_forecast_weather.html
http://www.yr.no/place/United_Kingdom/Scotland/Braemar~2654965/long.html
Significant discrepancies are not unusual, and both tend to over estimate winter temperatures as their models don't seem to take account of typography - most straths and glens in the Highlands don't get much winter sun, and are prone to snowfield-induced temperature inversions also, so unless there is a steady westerly wind, our temperatures are almost always about 4C less than Perth which is only 25 miles to the south.
By far the most useful page on the Met Office site is: http://www.metoffice.gov.uk/weather/uk/surface_pressure.html
But http://www.raintoday.co.uk/ is very useful for what's likely to happen in the next few hours, especially if showers are predicted.
p.s. zero is not cold - properly salted roads shouldn't get icy until it is -9C. I regularly take my son to school on by bike in such temperatures. (Though he was weeping with the pain of the windchill on his face when it was -17C a couple of weeks ago).
Dominic,
I have the impression that you are hoping (or assuming) that known days X,Y will uniquely predict Z for the third day. But that seems rather unlikely. For example, searching your 18,000 three day scenario history, we might find four that match the X,Y prefix: (X,Y,Z1), (X,Y,Z2), (X,Y,Z1), (X,Y,Z3). So, given X,Y, we might infer that day three is likely to be like one of Z1, Z2 or Z3, but we can't say which and we can't even say that Z1 is twice as likely as the other two (too little data to draw that conclusion).
I wonder if anyone has ever attempted to apply probabilistic finite state transition networks to short-term weather forecasting?
Yes, sod the engine. That's what I used to think when my Villiers two-stroke ran with stinky blue smoke most of the time, which it used to do no matter how much you shook the moped before starting - that is until you were about to run out of petrol and had burned up most of the lubricating oil - so what was my carbon footprint like? Then there was de-coking the silencer: you took it off, corked the ends, filled it with caustic soda, shook all it about and then what did you do with it? - Chucked it down the drain of course! I'm talking about the early Sixties, what a lot of sinners we were. But have they improved two-strokes since then?
James Evans' comment brought these nightmare memories flooding back and it's true I didn't properly read the rest, so duly chastened by BBD's reaction I just went back and read it right through.
And yes, where predicting the weather's concerned, it still seems to be every man for himself. I think I would generally trust Eric's observation, experience and judgement and try to get him to give me a few tips. For good measure I'd hang a thermometer on the wall outside my back door checking it especially first thing in the morning before leaving. That way you're likely to know the worst, as temperature's not likely to rise until well after sunrise this time of year
Jane - Yes I am hoping that it would be a good first-order predictor. Clearly I do not expect it to be exact. My point is why use an incomplete model based on fluid dynamics and other features of a system which is fundamentally chaotic when we could just use the past behaviour of the system itself as our predictor.
You are right to draw attention to the fact that if we find that we have a perfect match with 2 or more historical periods and that in these historical 3-day windows the 3rd day was not the same then we have a model which is incomplete. In my view this means that more data is needed to resolve this degeneracy, this could probably be seen quite clearly if we were to look at the data. Perhaps a 2-day window is too short and data from 3 days ago would break this degeneracy or we need another weather metric.
I was supposing that some sort of spatial interpolation (in weather parameter space) could be done based on the distance between the actual weather conditions on days 1 and 2 to interpolate the conditions on day 3 on the closest matches using some multidimensional interpolation scheme. This is of course an assumption but we have to start somewhere.
I would not characterise this as a rating transition approach - if that is what you are implying - since that is just another model requiring assumptions about probabilities and transition rates. It would also not be finite state since the weather parameters are continuous and the weather can therefore take an infinite number of states. It would only be finite in the number of state variables.
I would characterise this as a pattern recognition problem combined with some spatial interpolation algorithm.
Dominic,
Probabilistic Finite State Transition Networks (PFSTNs) are compactly described here. I was assuming that the continuous weather variables would be converted to discrete variables (e.g., "below freezing", "0-20C", "above 20C"). Ten parameters each with ten possible discrete values would yield one hundred states. A proposal like yours could then be implemented as a PFSTN with 10,200 states (100 for each of days one and three, 10,000 for day two [which needs to 'remermber' day one]) with the transition probabilities estimated from the 50 year dataset [I'm not implicitly claiming that this would be a sensible implementation of your idea]. To avoid the probabilities, just use an FSTN instead.
There's quite a large academic literature on PFSTNs (and FSTNs) under various names but I have never seen a weather forecasting application thereof.
An elaboration of your proposal would use the longest sequence available in the dataset. E.g. use a matching five day sequence if such exists; failing that, look for a matching four day sequence; failing that, look for a matching three day sequence; failing that, look for a matching two day sequence.