Bishop Hill From the least absurd models
Jan 20, 2014 Climate: Models A few days ago, I mentioned a paper that had looked at climate model simulation of clouds and found that the ones that did the best job of this narrow task produced the highest predictions of temperature rises. As I noted at the time this means that one can paraphrase the findings as "best cloud simulators are worst temperature predictors" but, as is normal in these circumstances, the headlines were all about global warming being "worse than we thought".
Yesterday Nature published a similar paper, this time looking at the El Nino phenomenon. It seems that if you take the models that best simulate extreme rainfall they predict that extreme Ninos will take place much more frequently in future, with all the floods and droughts and the like that accompany them.
The team identified 20 climate models — half of those available — that were capable of simulating extreme rainfall. They then used the models to compare the occurrence of extreme El Niños in a control period, 1891 to 1990, versus a warmer period extending from 1991 to 2090. Although the total number of El Niño events decreased, 17 out of the 20 models predicted more major El Niño events, with the average frequency increasing from once every 20 years to once per decade.
As an aside, it's notable that only half of the climate models were considered capable of simulating extreme rainfall, so one can reasonably wonder why we should base public policy on computer simulations that can't get even basic phenomena like heavy rain correct. One can also wonder at the paper's use of CMIP5 model runs, which don't incorporate the IPCC's latest estimates of aerosol forcings, and a perturbed physics ensemble, presumably the Sexton one, which has problems of its own. This is speculation though and I'll need to confirm when I get hold of the paper.
Of more immediate relevance to my point is the similarity in approach to the earlier paper - I think what we have here is a case of "the least absurd rainfall simulators predict a devastating increase in El Nino". Now you would think that, this being the case, the writers would have exercised a little restraint, perhaps mentioning uncertainties here and there. But not a bit of it. Here are a few of the headlines:
- Nature: Frequency of extreme El Niños to double as globe warms
- Guardian: Unchecked global warming 'will double extreme El Niño weather events'
- The Australian: 'Twice as many extreme El Ninos'
- Business Standard: Brace yourself for extreme El Ninos every 10 years
- New Scientist: Devastating El Niño events to double this century
Reader Comments (120)
if only these alarmists were like those millenarian cults...and commit mass suicide when/if the world warms 0.5 degrees or some other meaningless figure
Micky H Corbett
Its been done. Water vapour absorbs weakly at a large number of spot frequencies. In the OLR you can see this as a large number of spikes where water reduces the outward radiation by a few percent.
By contrast Co2 absorbs mainly at two wavelengths, with the 15 micrometre spike deep enough to saturate at the spot frequency.
Claes Johnson has good image of the OLR. You can clearly see the water spikes across most of the black body curve.
Overall water contributes more to absorption of the OLR than CO2.
diogenes
Welcome back. Have you seen your millenarian counterpart, DIOGENES, posting on Real climate?
Thinkingscientist
Challenge 1
Chlorofluorocarbons
Challenge 2
That's silly .You insist on constraints that would automatically make any model inaccurate. If you want an accurate hindcast it much include accurate forcing.
> Chlorofluorocarbons
The ozone hole in 2006 was the largest it has ever been, closely followed by the size in 2011. It had one of its smallest sizes in 2012. Overall the trend for the ozone hole is up, not down but it has a lot of annual variability.
So what, exactly has the CFC ban achieved?
Not only are there many, many people who fell good about themselves as a result but they also can preen themselves on having put one over on the chemical industry which, as evry (green) fule kno, is inherently bad because "chemicals" are evil and must be abolished.
Then there is the warm feeling of having done a good deed and done a bit of planet saving, albeit the planet wasn't actually in need of any saving but that's not really important, is it? It's the thought that counts.
And of course that was one step along the way to becoming important and being able to tell governments how to behave because they listened to us even though — as usual — we knew naff all about the science which we made up as we went along.
So the CFC ban actually achieved quite a bit. Most of it bad or counter-productive or pointless.
A yes, Claes Johnson. An authority on radiation physics. (EM 3.17)
Martin A
I linked to that article because it had a clear graph of the OLR. I agree with your opinion of the article itself.
Forgot the link!
http://claesjohnson.blogspot.co.uk/2013/03/the-fabrication-of-co2-alarmism-decoded.html
He has a number of articles which make entertaining reading. :-)
Entropic Man's answers to challenges
1. CFC. Nope, unproven as at no time was no ozone hole observed
2. You have answered "You insist on constraints that would automatically make any model inaccurate. If you want an accurate hindcast it much include accurate forcing.".
EM, if you are trying to predict the climate in 100 years time, the point is you don't know whats going to happen. And what has hindcasting got to do with forward prediction?
If you cannot make a reliable prediction of 2010 from a model initialised in 1910, why should anyone believe in the results of climate model predictions for the year 2110? We can't even have confidence in predictions made less than 10 years ago about temperatures now, or hurricane frequency or whatever.
A climate model starts with equations of state and change, plus a set of measurements which define you starting conditions. In your example that would be 1910. If this were all you needed, every run would come out the same.
Unfortunately there are unpredictable events. A volcano may erupt. A stronger or weaker solar cycle may change insulation. China may produce extra aerosols.
Since events such as Pinatubo cannot be anticipated they are randomised and the model run repeatedly. Your final dataset is the output of all these runs.
What this looks like is a temperature/time graph with confidence limits. Model runs with high net positive forcing come out warmer. Models with less forcing, perhaps with a lot of random eruptions, come out cooler. Over many runs you build up an idea of how the system behaves and how it responds to random variables. By starting the model at some past date you can assess its performance by comparing the model output with observation, giving the opportunity to tune the model to improve its skill.
The best model performance comes when eruptions, pollution etc are matched to reality up to the present. Kosaka and Xie demonstrated this.
http://www.nature.com/nature/journal/vaop/ncurrent/full/nature12534.html
Rather than using randomised ENSO data they ran their model with actual Pacific sea surface temperatures and found that the output was closer to reality.
"Challenge 2: Show me a model run, or suite of runs, initialised in the year 1910 and with no other information on aerosols, solar measurements etc, that can reliably predict the temperature response and internal climate variability out to 2010. You can have volcanic eruptions over the prediction period to help you, but nothing else."
Your criteria preclude including accurate information on past ENSO, aerosols, solar insulation and other random or unpredictable variables. This reduces the skill of the model, making your demand for reliability impossible.
The model runs which most accurately reflect current conditions have above average vulcanism, lower insolation and higher aerosols than the randomised average, giving lower forcing and less warming. This is because the real world has had above average vulcanism, lower insulation and higher aerosols than average.
"Your criteria preclude including accurate information on past ENSO, aerosols, solar insulation and other random or unpredictable variables. This reduces the skill of the model, making your demand for reliability impossible."
EM, thank you for highlighting why climate 'models' are worthless at predicitng the future.
Nial
Uncertainty about future events such as vulcanism means that future temperatures are uncertain. Knowing the likely variability of such variables allows one to define confidence limits for that uncertainty.
Being able to advise policy makers of the likely upper and lower limits for change is not worthless.
"Being able to advise policy makers of the likely upper and lower limits for change is not worthless."
Except without knowing the future 'variabiity of such variables' you've already admitted the predicitve powers of the models are worthless.
What are your error bars going to be +/- 10 DegC?
Rhetorical question.
Entropic Man, to be polite, you seem to be missing the point. If you cannot run a climate model from a start point in 1910 (and using any data you like before 1910) and make a reasonable prediction that would agree with the actual measured data out to 2010, then any prediction from now to 2110 is worthless.
You have effectively demonstrated the truth of my point - climate models cannot predict the future because they need way to many parameters about the future to be known already. In fact, without tuning and hindcasting, they cannot even accurately fit what we know and so far have failed to make an even remotely useful predictions out to just 10 - 15 years.
Richard Betts,
"You don't have to make predictions in order to assess risks - you just need to look objectively at possible outcomes, consider the implications of these, and make a judgement."
You forgot to mention "likelihood". Probably an oversight.
"Process-based numerical models can clearly represent atmospheric processes well enough to simulate the global atmosphere pretty well, in terms of the global atmospheric circulation, general rainfall patterns, etc, and this is why it is possible to provide useful weather forecasts."
I've yet to see a useful weather forecast beyond a few days. Is your incredible ability to roughly forecast the weather a couple of days in advance due to models, or simply due to the fact that satellites now allow you to watch the weather arriving from the Atlantic?
"Yes the models disagree widely on future changes due to increased GHGs, and also have a long way to go in being able to predict specific features of internal climate variability, but they can give a guide to the range of changes that could occur."
Sure, the models can "give a guide". So can the crazy cat-lady who lives down the road. What evidence is there to suggest that the models "give a guide" that is better than hers?
"I do not share your confidence that the future can be predicted so easily, especially on the basis of a couple of variables estimated over a relatively short period in the past."
I see what you did there. You turned it round, didn't you? You tricky little scamp.
You took a criticism that sceptics often use - and you used it against them! So clever. You win the debate!
That MO Tropical Storm link is priceless. Never mind the predictions even they had to admit were useless, look at the ones they claim as successful!
Forecast: 2-14
Actual: 2
MO: "values were at the extreme lower end of the range predicted." Model validated.
That's not a forecast, that's practically all possible outcomes, so it's hardly surprising that the actual was in there somewhere. It is not, however, useful for any practical purpose to know this.
Occasionally having a vague prediction come within shouting distance of reality does not provide a sound basis for predicting impacts.
James Evans
Thanks.
Yes, I should have included "likelihood".
And yes, improved weather forecasts are due to models, as they go beyond merely tracking existing weather systems. The St Jude's Day storm, for example, was correctly forecast by the models several days in advance before it had even formed out in the Atlantic.
Numerical models are better than "the cat-crazy lady who lives down the road" because, although they rely on assumptions and approximations, they are at least based on understanding of processes and provide an internally-consistent picture of the implications of this understanding.
Glad you spotted the turnaround of the overconfidence criticism! Yes indeed, I do genuinely find it highly ironic to see people here being so confident in predictions of the future, as a long as those predictions are of little change. Even with process-based calculations of future changes, there are enormous uncertainties due to the assumptions and approximations I mentioned above, not to mention chaos in the system. The challenge is therefore in deciding how to respond in the face of this large uncertainty. Simply ignoring the uncertainty (in either direction) does not, in my view, lead to informed decisions.
Yes, one of the many predicted storms actually happened. Based on the link, that will happen about 20% of the time.
We know it was right this time, because the BBC kept sticking MO people in front of a camera so that they could say "We predicted this, you know."
Given the BBCs documented position on CAGW, it would be interesting to know how this was arranged, and which organisation was the instigator. But it's our public broadcaster, so they would refuse to divulge anything on the basis of journalistic confidentiality.
Richard,
I am, as ever, genuinely grateful that you answer my comments. I can be an irascible tw*t at times. (Choose your own vowel there.)
"And yes, improved weather forecasts are due to models, as they go beyond merely tracking existing weather systems."
I really wish they'd do better though. Today, for instance, I spent the day in the glasshouse doing some potting-up, because the forecast was for rain all day. We had almost no rain. So I lost a day working in the garden. It really is a big deal for us. It might sound trivial to you, but for us it's incredibly important. We really notice how inaccurate the forecasts are. It impacts our work every day.
"Numerical models are better than 'the cat-crazy lady who lives down the road' because, although they rely on assumptions and approximations, they are at least based on understanding of processes and provide an internally-consistent picture of the implications of this understanding."
I completely disagree. I really couldn't give a flying feck (vowel choice again) what the models are based on. The climate models DON'T WORK. I don't care if they're based on the word of God, or the recipes of Mary Berry. THEY DON'T WORK.
"Simply ignoring the uncertainty (in either direction) does not, in my view, lead to informed decisions."
It's called "getting out of bed in the morning". If I had to resolve all the uncertainties in every direction beforehand, it wouldn't happen.
Thanks indeed for your time.