Recently the Chief of the met office went on UK TV to say:
“OUR SHORT TERM FORECASTS ARE AMONG THE BEST IN THE WORLD.” (see video here)
Yesterday, the UK Met Office had to make a rare mea culpa, saying they had botched their own recent snow forecast, it is useful to point out that they aren’t the only one with egg on their faces.
In early October, the Arctic Oscillation (AO) took an unexpected dip into deeply negative territory, which led to the sixth snowiest October on record in the Northern Hemisphere and the snowiest on record in the US. If you look at the 14 day forecast at the bottom of the graph below, you can see that the dip caught NOAA forecasters off guard.
Source: NOAA Arctic Oscillation Forecast
According to Rutgers University Snow Lab, October, 2009 was the snowiest on record in the US.
| Contiguous United States | ||||
| Month | Rank | Area | Departure | Mean |
| 12-2009 | 1/44 | 4161 | 1292 | 2869 |
| 11-2009 | 39/44 | 585 | -512 | 1097 |
| 10-2009 | 1/42 | 538 | 385 | 153 |
| 9-2009 | 5/41 | 21 | 13 | 8 |
| 8-2009 | 12-41/41 | 0 | -5 | 5 |
| 7-2009 | 24-40/40 | 0 | -17 | 17 |
| 6-2009 | 32-42/42 | 0 | -64 | 64 |
| 5-2009 | 37/43 | 34 | -151 | 185 |
| 4-2009 | 17/43 | 859 | 106 | 753 |
| 3-2009 | 23/43 | 1964 | -18 | 1983 |
| 2-2009 | 17/43 | 3172 | 110 | 3062 |
| 1-2009 | 15/43 | 3696 | 185 | 3511 |
Source: Rutgers University Snow Lab
The director of NCAR captured the moment perfectly in this East Anglia Email – dated October 12.
From: Kevin Trenberth <trenbert@xxxxxxxxx.xxx>To: Michael Mann <mann@xxxxxxxxx.xxx>
Subject: Re: BBC U-turn on climate
Date: Mon, 12 Oct 2009 08:57:37 -0600
Cc: Stephen H Schneider <shs@xxxxxxxxx.xxx>, Myles Allen <allen@xxxxxxxxx.xxx>, peter stott <peter.stott@xxxxxxxxx.xxx>, “Philip D. Jones” <p.jones@xxxxxxxxx.xxx>, Benjamin Santer <santer1@xxxxxxxxx.xxx>, Tom Wigley <wigley@xxxxxxxxx.xxx>, Thomas R Karl <Thomas.R.Karl@xxxxxxxxx.xxx>, Gavin Schmidt <gschmidt@xxxxxxxxx.xxx>, James Hansen <jhansen@xxxxxxxxx.xxx>, Michael Oppenheimer <omichael@xxxxxxxxx.xxx>
Hi all
Well I have my own article on where the heck is global warming? We are asking that here in Boulder where we have broken records the past two days for the coldest days on record. We had 4 inches of snow. The high the last 2 days was below 30F and the normal is 69F, and it smashed the previous records for these days by 10F. The low was about 18F and also a record low, well below the previous record low. This is January weather (see the Rockies baseball playoff game was canceled on saturday and then played last night in below freezing weather).
Trenberth, K. E., 2009: An imperative for climate change planning: tracking Earth’s global
energy. Current Opinion in Environmental Sustainability, 1, 19-27,
doi:10.1016/j.cosust.2009.06.001. [1][PDF] (A PDF of the published version can be obtained
from the author.)
The fact is that we can’t account for the lack of warming at the moment and it is a
travesty that we can’t.
http://www.eastangliaemails.com/emails.php?eid=1048&filename=1255352257.txt
Once again, this begs the question – if the GCMs can’t forecast the AO two weeks in advance, how can they possible forecast snow and cold 70 years in advance? University of Colorado professor Mark Williams used climate models in 2008 to come up with a remarkable prediction (below) in a year when Aspen broke their snowfall record.
Study: Climate change may force skiers uphill
From the From the Associated Press
Tuesday, December 16, 2008
DENVER — A study of two Rocky Mountain ski resorts says climate change will mean shorter seasons and less snow on lower slopes.
The study by two Colorado researchers says Aspen Mountain in Colorado and Park City in Utah will see dramatic changes even with a reduction in carbon emissions, which fuel climate change.
University of Colorado-Boulder geography professor Mark Williams said Monday that the resorts should be in fairly good shape the next 25 years, but after that there will be less snowpack — or no snow at all — at the base areas, and the season will be shorter because snow will accumulate later and melt earlier.
If carbon emissions increase, the average temperature at Park City will be 10.4 degrees warmer by 2100, and there likely will be no snowpack, according to the study. Skiing at Aspen, with an average temperature 8.6 degrees higher than now, will be marginal.
Since the first of October, Colorado is averaging two to eight degrees below normal, as is most of the US:
Source : NOAA High Plains Regional Climate Center
In December 2009, Colorado averaged three to fifteen degrees below normal, once again correlating with a strongly negative Arctic Oscillation
Source : NOAA High Plains Regional Climate Center
Climate models are iterative through time, which means once they go off in the weeds they can not recover. If AO trends can not be forecast more than a few days in advance, it would seem problematic to make any sort of meaningful long-term climate projections using GCMs.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
I’d say Leif, that given our tiny human timeslice of the observation of processes that are active for billions of years, we can only ‘predict’ the current prevalent oscillation. Granted it doesn’t matter much that in a million years the solar cycle will be different but we have no way of knowing the solar cycles as we (assumingly) have observed are going to flip over today or in a million years.
Same goes for any other chaotic/entropic system.
A clear discussion of the skill of models.. skill in forecasting various parameters at various temporal and spatial resolutions would create a better conversation. If you did that, then you’d probably get Leif’s point in a more concrete fashion.
Good to see you again Dr. S.
‘University of Colorado-Boulder geography professor Mark Williams said Monday that the resorts should be in fairly good shape the next 25 years, but after that there will be less snowpack — or no snow at all — at the base areas, and the season will be shorter because snow will accumulate later and melt earlier.’
Actually, an element of that later accumulation happened in Europe from about the mid 1980s to the early 2000s – probably shifted back toward earlier snowfalls in about 2004/05. The melting comes when the melting comes, because in spring temperatures rise way up, so the melt pattern kicks in regardless. But the key thing is the cycles which compact the snow early in the season, which slows rates of thaw once the sun comes out later in the season. That’s why mountain guides say that ‘the snow which falls before the middle of January is the snow which lasts’.
I think such things happen due to PDO cycles, although I can’t confirm that.
Don’t know enough about Colorado climate patterns to know for them, but in europe the late autumn/early winter snows tended not to come for about 20 years, with snow really only starting properly after New Year.
Something to monitor perhaps?
Leif,
I understand your analogy, but Earth’s energy-budget contains factors analogous to rolling all the pebbles back up the hill again. You need to add a sort of Sisyphus to your analogy, who (in rolling the rock up the steep hill,) turns energy into potential energy.
When a La Nina’s strong trade winds push warm surface water west, and pile it up by Australia, and also drag heavy and cold water up from the deeps by Peru, the system is defying gravity, and creating potential energy.
Also any time ice turns to water, or water turns to vapor, available energy is turned into latent energy. It doesn’t show on thermometers, but it is there.
A positive AO creates a strong polar vortex which holds arctic air up at the poles, “against it’s will,” as the cold and heavy air wants to obey gravity and flow south.
I’m sure there are other examples you’d know more about, involving upper atmosphere chemistry and various sorts of radiation produced by the sun.
And of course there are the sloshings we haven’t studied much, involving thermohaline circulation and the thermocline, which may involve cycles of a thousand years.
In effect there are all these Sisyphuses running about, screwing up our calculations……and also our analogies.
Leif Svalgaard (23:49:57) : | Reply w/ Link
My examples are biased by what I do: I think we can predict the solar cycle [i.e. solar climate] 10 years hence, even though we have no idea how to predict next month’s sunspots. Now, there are people that think solar cycle prediction is impossible too [e.g. Tobias and Weiss] and there are people whose solar cycle prediction has failed, so perhaps those people would by more sympathetic to the post.
Well, if you had a computer program that could predict the number of sunspots for the next two months with an 80% success rate ( as the weather GCMs do for the next few days), would you trust it to predict what will happen on the sun in one hundred years, after fiddling with the parameters? That is the correct analogy between weather and climate programs.
In your gas and pressure example it is clearer that one has to go to a “meta” state. The kinematics that describe individual molecules though known, are useless in describing the total ensemble. A new mathematical description is needed, in this case with new variables and measures like temperature and pressure. ( Bohm’s implicate and explicate orders)
If we take this to the weather and climate setups, it means a new tool is needed and in my opinion it is only with the tools of chaos that one might be able to go into long term climate predictions.
Joe D’aleo gave a great forecast about the exceptionally cold conditions the UK would experience and I seem to remember he suggested a thaw about now, which is what is happening.
Has anyone come across a forecast from him for the UK for the rest of the winter? Should I huint the shops for salt for our drive or merely get my raincoat out as the mild wet weather reasserts itself?
tonyb
Leif Svalgaard (22:50:02)
Here is another for your list.
We cannot predict the average temperature for next summer. But we can predict it will be warmer than this winter.
Leif Svalgaard (22:50:02) :
Once again, this begs the question – if the GCMs can’t forecast the AO two weeks in advance
I think this logic is faulty. Let me illustrate with some examples:
1) Take a massive landslide. We cannot predict where each little pebble goes, but that safely predict that the whole mess will end up down in the valley.
2) We cannot predict where next month’s sunspots will be [the astrologers expected, of course], but we can predict with some success the size of the next cycle [at least roughly]
3) When a dam breaks and the water rushes downstream we don’t know where individual pieces of debris will end up, but can be sure that the water eventually flows to the sea.
4) We cannot predict where an individual molecule of a hot gas will strike the wall enclosing the gas, but we can predict quite well the pressure of all the molecules in toto
I think this logic is faulty. Climate models do work by setting off with initial conditions and summing the totals from the collection of parts (cells) going forward. So although a model might predict that the pebbles in the landslide will end up in the valley ratherthan the sky, they will probably not successfully predict how far down the valley the majority of the mess will end up.
It’s the weather that’s supposedly going to hurt us (hurricanes, storms, extreme temps etc) not climate. So predicting climate but not weather is not really as usefull as we’d like.
I’m not so sure AO is weather. I know it causes weather events, but is it weather itself?
There is a grey area between weather and climate. Some claim (like Piers Corbyn) they can predict extreme weather months in advance by understanding the sun. I believe if the IPCC were honestly trying to understand our climate, they would not have started by gathering temperature data. They should have started with the biggest, longest cycle then worked their way down so to speak. Then they may have had a chance of some accuracy regards predicting AO ENSO and PDO
Back to your drawing boards, lads and lasses, but buy your own pencils this time!
anna v (01:16:38) :
If we take this to the weather and climate setups, it means a new tool is needed and in my opinion it is only with the tools of chaos that one might be able to go into long term climate predictions.
My reply: I am only suggesting that we study the effects of the Lunar declinational tides in the atmosphere, because if I pull out the past three patterns of 6558 days, or 240 lunar declinational cycles, and plot them side by side, day by day, they show a resultant pattern that gives a better forecast than any of the models.
With compensation for the changes in the solar cycles, from the past three patterns of high activity, to this one of lower activity I would get the correct amount of cooling, and better representation of the depth, of the cold arctic air invasions that are arriving on time, just larger than before.
Leif – They run computer models (many thousand) on football games in the US to analysis the best odds for said match. Vegas odds makers have their money on this “line”, therefore no stone is left unturned. Yet the game still must be played in reality and the outcome from time to time differs from the models/odds.
The Earth’s climate mechanisms seem so much more complex and with events such as volcanic eruptions and external effects still to be understood I just can’t jump on any long term prediction wagon. You sure as hell wouldn’t want to commit policy and resources, which is what AGW proponents are demanding.
Leif Svalgaard (22:50:02) :
I think this logic is faulty. Let me illustrate with some examples:
I hesitate to question Leif’s logic as I have followed his comments for quite some time. That said, it’s unclear to me how Leif’s examples apply.
Three of the four examples are modeled by simple application of a single “law” of physics (gravity, Boyle’s Law) to predict an outcome quite precisely. One example (sunspots) requires several physical “laws” to model and (who would guess) the predictions are less rigid (“we can predict with some success the size of the next cycle [at least roughly]”). In all these examples we have a great deal of empirical evidence for the cause/effect relationships.
Since climate modeling currently seems to require the simultaneous solution of many physical “laws”, it seems reasonable to me that the predictive certainty of such models would necessarily be a good deal less than any of the above examples. Also, from reading this blog, it appears we are still trying to derive most of the cause/effect relationships of the empirical observations that are believed to relate to global temperature changes.
The weather=single particle vs climate=particle aggregate is an interesting way to describe this often contested relationship. However, based on the above paragraphs, I think this comparison itself actually undercuts Leif’s argument with a faulty premise that the climate models equate to a single formula or physical “law” that describe cause and effect of global temperature changes similar to the examples proposed.
Daniel L
I sometimes check (via the BBC) the Met Office weather forecast. The term is a misnomer. It is not a forecast, but a commentary. Say I look at the projection for Boscombe Down (nearest weather station) for 4.00pm, at 8.00am. I guarantee that the 4.00pm forecast will have changed when I look again at 10.00am.
Projections for coming days are even worse. Nice blazing sun symbols at the weekend usually turn to something less amenable within the space of a few hours. So I was very pleased to see the head of the Met Office skewered by Andrew Neil – notwithstanding I can’t stand Andrew Neil.
Our pathetic government, led by a bloke (Gordon “Prudence” Brown) who takes feedback for his speeches from his pet goldfish (just look at a clip of his speeches and you’ll know what I mean), would do better to scrap the whole thing and provide every household with a bunch of seaweed to hang outside the front door
Please moderate out all this weather/climate crap.
If you stand on a beach it’s difficult the forecast how high the next few waves will run up the strand.
But forecasts of height of high tide reamin highly predictable.
mkurbo (02:33:36) :
snip
The Earth’s climate mechanisms seem so much more complex and with events such as volcanic eruptions and external effects still to be understood I just can’t jump on any long term prediction wagon. You sure as hell wouldn’t want to commit policy and resources, which is what AGW proponents are demanding.
My repy:
The problem with what the CAGW proponents are demanding is we place all our bets on the health and performance of the kicker alone, with out looking at the usability or readiness of the rest of the team.
Then they want us to bet the whole farm, not just our milk money.
I would much rather take my time and evaluate all of the players till I can figure out who should even start in the game, and who to keep on the bench or in the locker room.
My bets would be on the method that could bridge the gap, between short term forecasts that work, and connecting the dynamics that make them functional, to the longer term influences that are just the extensions of those cycles interactions with the already known factors of climate variability.
Richard Holle
For once I would partially agree with the good old doc, perhaps he has initially chosen not entirely adequate example, but his subsequent notes did put the matter right.
Climate, sunspot cycle, stock market index etc, are all made of numerous smaller elements of apparently random (chaotic) movements, but if majority of them (not necessarily all) have the same or similar underlined tendency, than long term forecast is possible, by correlating past to one or combination of two or more plausible causing factors, even if the mechanism of it is not entirely clear. Most of the above examples are cyclical thus making longer term predictions feasible.
Although, ‘correlation is not causation’, but correlation is necessary, so ‘no causation without correlation’.
It was assumed that CO2 mechanism was understood, kind of misplaced ‘correlation’ was there, but now and in the past has failed, since there is no good reason put forward why it has failed, scientists have to try again and find true correlation which does hold.
To quote A.E.
“Occurrences in this domain are beyond the reach of exact prediction because of the variety of factors in operation, not because of any lack of order in nature.”
Or my own “the nature is adverse to a coincidence; it is ruled by a cause and its consequence”.
TFN Johnson (03:04:21) :
Please moderate out all this weather/climate crap.
If you stand on a beach it’s difficult the forecast how high the next few waves will run up the strand.
But forecasts of height of high tide reamin highly predictable.
My reply;
To an organic gardener even a turd, is useful. The heights may very some but the timing of the peak, of the effect is the known factor that repeats well.
The waves reaching the shore change character with the ebb and flow of the tide due to the compounding of the trends, is as true in the sand, as in the Rossby waves and jet streams.
Leif Svalgaard (00:26:32) :
Dave F (00:18:54) :
anything is even correct in the models.
I don’t believe the current models are correct [too many things are parameterized], but my beef was with the bland assertion that because we cannot predict two weeks ahead, all prediction further out is impossible. If the climate to a large extent is self-regulating [negative feedbacks] then a model may not necessarily ‘go off the rail’ in the far future, but may be kept within bounds by the regulator [whatever it may be].
Leif,
Predicting the AO within two weeks and the climate over 70 years is more akin to predicting what a bunch of pebbles will do in two weeks and 70 years during a hypothetical 70 year mudslide. In short term weather models, incorrect parameter inputs will lead to short term model failure for tomorrow’s weather. Oftentimes, the weather models will have a sense of something that may happen 7 days away, but when the inputs become more real (the low pressure short wave reaches a data collection point and then fed to the computer), we all of a sudden lose the potential storm entirely.
Getting back to your analogy, if I predict where those pebbles will be in two weeks, and they are collectively within a standard deviation of being correct, then I can say that my predictive model has merit to that point, and that my future prediction for 70 years hence has a higher likelihood of being correct than if my pebbles two weeks hence were largely in a different location than my prediction.
As far as I can tell, not one of the GCM’s a decade ago predicted the stalling of the temp rise this past decade. If so, then this failed prediction must have long term implications on the predictions being made for 2060, 2080, 2100, etc. Again, that is how it works in weather models, and climate models for this comparison are no different. If the model predicts the storm will be over Georgia this evening, and it ends up over Cuba, I can assure you that the actual weather in NYC tomorrow will be different than what the model had predicted based on a storm in Georgia.
Tom
TFN Johnson (03:04:21) :
Please moderate out all this weather/climate crap.
If you stand on a beach it’s difficult the forecast how high the next few waves will run up the strand.
But forecasts of height of high tide reamin highly predictable.
TFN,
Not entirely correct on your individual wave thought. One may not be able to predict the exact height of the next few waves, but wave sets do have predictive qualities as I learned in my oceanography studies in university (many, many days ago). You cannot arrive at highly predictable tide forecasts (your words) without the underlying wave propagation being predictable. No??
The highest AO variability (by far) is in winter.
Reconstructions based on PCA or factor analysis need to be conditioned seasonally.
1899-2002 SLP-based AO reconstruction residuals show very strong seasonal bias:
http://www.sfu.ca/~plv/AOresi.png
Reality is consistently lower than model in January.
The variance in the residuals can easily be reduced by a factor of 2, cutting errors in half and reducing systematic bias:
http://www.sfu.ca/~plv/AOresi_.PNG
If the reconstructions play a role in forecasting, there should be room for improvement.
While Met Office can’t predict the weather a few months ahead, you should not doubt or question their ability to predict it 50 or 100 years ahead.
As we all know, it’s “ settled logic” and “ consensus statistics” ”that it is much easier to predict the future the further you go out in time.
Yes, it may be impossible to exactly predict tomorrow’s sports outcomes, but you can easily predict the outcomes of sports events decades away when you use computers and software as powerful as those used at the omniscient Met Office.
The American great promoter, B.T. Barnum noted:
“There’s a sucker born every minute.”
He would have made a fortune in “carbon credits”.
“University of Colorado professor Mark Williams used climate models in 2008 to come up with a remarkable prediction (below) in a year when Aspen broke their snowfall record.”
I think the fault at the heart of this winter’s unexpected wintery-ness is climate models which predict average conditions being misinterpreted by the modellers themselves into weather conditions, and as a conseqence local authorities and national Governments not spotting this error for a variety of reasons.
We’ve seen that in the UK to such a deleterious effect – the constant squeaks from the Met Office of ‘milder winters’ has led councils to be ill-prepared for normal wintery conditions. Even with milder winters on average you should still expect times of bitterly cold weather, icy roads and snowfall. It is winter afterall and blasts of arctic weather can hit the UK with very short notice. The modellers are seeing what they want to see in the models, things that aren’t actually there, and councils and Governments are seeing the same things.
The landslide is an interesting analogy. Generally they can be predicted with simple mathematical models; however this isn’t always the case. Instances of landslides propelling themselves miles along a plain, often going up small slopes as they progressed in apparent defiance of gravity were written off as hearsay and tall tales because the modelled behaviour of landslides didn’t predict these events, and it seemed impossible that they could happen. Surely a landslide couldn’t slide up, could it?
Turns out that the mathematics used in modelling the behaviour of landslides was faulty. Originally it was assumed that a landslide was just rock and mud moving downhill. That’s the simplest model, but it was wrong nearly all the time. This very simple model wouldn’t be able to predict long tails, or the fluid-like behaviour of certain landslides, and absolutely and completely unable to predict a sturzstrom.
A sturzstrom is by its nature, and our current understanding, highly unpredictable. The reasons for its behaviour are not known with any real certainty. It can be modelled, but those models can’t be used for predictive purposes as the slide is chaotic. You could predict that it would flow along the valley, but that isn’t enough; Given their propensity to sweep up slopes, is the house on the south side of the valley safe from the slide or not? How far will it go? Will it revert to another sort of slide, or will it carry on flowing like a river? Will it go up that hill or down that creek? Will it even be a sturzstrom, or will it just be a “regular” landslide? A model of the dynamics of a sturstrom can’t predict this, even if based on a real potential or actual slide, because the model is necessarily chaotic.
Just like climate models. We can model climate in a limited way but they’re useless for prediction. And beyond a few days, weather models are useless for prediction too, for the same reason. The argument over weather vs climate is moot as in both cases we are trying to use models of chaotic systems that have little predictive merit. We can model climate out to X number of years but as a prediction it is meaningless.
Regarding GCM’s: Does a failure in predictive quality in the short term (ten years) require that the long term (50+ years) predictive quality must also fail due to propagation of the short term actual results??