Arguments
2011 Sea Ice Minimum
Posted on 27 September 2011 by Rob Honeycutt
If you enjoy Peter Sinclair's YouTube video series here is his latest installment which includes footage of James Hansen, ice pilot Arne Sorensen, Stefan Rahmstorf, and Julienne Stroeve from the NSIDC.
I remember you saying, Rob, on an earlier thread that the melting of the polar ice caps would be the big wake up call about the realities of AGW, and indeed you were correct.
What astounds me is how those in denial avoid mentioning ice decline at the higher lattitudes and coral atoll inundation at the lower lattitudes.
How much longer can the likes of Bastardi go on predicting that all the current trends will reverse next year for no apparant reason?
If you had predicted something in the order of a 20% drop in the minimum extent over the next 4 years in May 2007 the skeptics would have gone spare calling you all the names under the sun for such an unthinkable and outlandish suggestion. Now if you suggest that that drop was in any way unusual you are lambasted for trying to make a mountain out of a molehill. Its all natural variability and to be expected you see.
The unthinkable becomes the mundane. Denial flourishes.
I suspect indefinitely. The opening of the NW passage and the Northern sea route should have been a wakeup call. But it's already become the new normal. We have short memories.
And also in contrast to Tamino, Anthony Watts submitted a forecast (based on a poll from his readers) to the SEARCH project on 31 May 2011 (so a 3.5 month forecast versus 11 months for Tamino). Here is the WUWT forecast:
"PAN-ARCTIC OUTLOOK – WUWT (acronym for WattsUpWithThat.com)
1. Extent Projection: 5.5 million square kilometers
2. Methods/Techniques: web poll of readers
3. Rationale: Composite of projections by readers, projection bracket with the highest response is the one submitted.
4. Executive Summary: Website devoted to climate and weather polled its readers for the best estimate of 2011 sea ice extent minimum by choosing bracketed values from a web poll which can be seen at: http://wattsupwiththat.com/2011/05/19/sea-ice-news-call-for-arctic-sea-ice-forecasts-plus-forecast-poll/15.64% chose 5.5 million km2 or greater, with 13.09% choosing 5.0 to 5.1 million sq km2 as the second highest vote.
5. Estimate of Forecast Skill: none"
Pay attention to point number 5. On that we agree ;) Out by about one million km^2.
Readers do not buy into this meme that the Arctic sea ice is recovering that is being perpetuated by 'skeptics' and those in denial about AGW.
I think I may have a go at statistical prediction for next year, looks like fun.
You are welcome.
What is kinda interesting is that WUWT readers actually predicted the sea ice extent minimum in 2011 would be higher than that in 2005, and that is very similar to Bastardi's erroneous prediction. So I wonder how much he influenced their thinking? Is it an example of group think, or group denial about the plight of the Arctic sea ice? I mean the loss of Arctic ice is accelerating, yet they keep trying to tell themselves and that there it is going to recover soon, "No really, this time it will recover". I wonder when will they will be saying that, "Oh don't worry, the ice extent next year will be back to the 2007 extent next year".
Dikran @9,
Yes, strictly speaking, I would have to agree with you.
"Will they learn something from their poor Arctic sea ice prediction? "
Very likely not. They will likely continue to delude themselves and others, or rationalize that it is not a big deal, or eventually concede that it is a big deal but argue that human emissions of GHGs are only a bit player. There are unfortunately, and sadly, far too many ways people can rationalize their denial of something, including AGW.
Delusion indeed. See the comments at Peter Sinclair's Sea Ice Min video posting:
"I’m (one of the ones) siding with Steven Goddard. He was right."
It's funny because the video has the NSIDc's scientist saying "being the 2nd lowest vs the lowest is not really recovery."
So, every time you see those guys plotting a trajectory for the next season off of 2007... they just don't know what they're looking at.
NSIDC have called the absolute minimum for 2011. The graph in the vid is of September monthly minimum. The 2011 value on that chart should reflect the monthly average, but September isn't even over.
I remember posts at realclimate criticising 'skeptics' for giving annual values for global temperature before the year was even finished. Same thing.
Add to that the observations of a mean ice thickness of ~0.9m, compared to ~2m 10 years ago (Alfred Wegener Institute, Polarstern). Can anyone think of a reason or a mechanism that will see anything other than a record low level of ice next year? Weather patterns weren't 'great' for ice loss this year, and ice thickness is on a frightening trajectory.
I'd like to know what they think the sea ice volume will do next year, and the year after, and the year after that...
I obtained from data for Arctic sea ice extent from 1979-2009 (the NSIDC data archive appears to be down at the moment, that was the best I could find). I then fitted a Gaussian process model, using the excellent MATLAB Gaussian Processes for Machine Learning toolbox (the book is jolly good as well). I experimented with some basic covariance functions, and chose the squared exponential, as that gave the lowest negative log marginal likelihood (NLML). The hyper-parameters were tuned by minimising the NLML in the usual way.
Here is a pretty picture:
Note the credible interval gets wider the further you extrapolate from the data, which is a nice feature of Bayesian models. Other highlights include:
prediction for 2010 = 4.927226 (+/- 1.069078)
prediction for 2011 = 4.772309 (+/- 1.096537)
prediction for 2012 = 4.614637 (+/- 1.128915)
ice free summer unlikely prior to 2027
ice free summer probably after 2041
I haven't checked to see how accurate the first two "predictions" actually are. This isn't really a serious attempt at a prediction, I just wanted to try out the regression tools in the GPML toolbox, but when I can get some up-to-date data, I can update the projections for 2012 and onwards. Hopefully I won't end up being the subject of a lessons from predictions post. ;o)
Caveat lector: This is a purely statistical prediction, so it is less reliable than physics, but hopefully better than chimps & buckets.
Or perhaps: SkS now using chimps and buckets for climate science!
As my predictions are more accurate that the WUWT committee projection, it would be a poor reflection on them indeed if the skeptics say we are using a chimps'n'buckets approach!
I hope Dana will be kind to you ;)
Interesting game.
What happens when you plug in the sea ice volume values?
I don't mind saying what the model says about when the Arctic will be ice free in Summer, but I wouldn't make a serious prediction about it.
This gives rather poor predictions for the last two years:
prediction for 2010 = 5.414904 (+/- 1.124377)
prediction for 2011 = 5.338500 (+/- 1.130803)
prediction for 2012 = 5.262056 (+/- 1.137584)
However Tamino's model had more data to work with, so it is possible that with more data the model will decide that there is more curvature and less noise in the data than at present.
I still think Tamino's method is better; I suspect the problem here is that the model has a couple of hyper-parameters that have been optimised, whereas a proper Bayesian would integrate them out. It may be there are two ways of explain it the data, high curvature/low noise and low curvature/high noise and the optimisation approach only finds the latter, whereas the simpler frequentist approach finds the former.
Self-skepticism is vital in statistical analysis, which is why I greatly prefer a physical model over a purely statistical one.
The graph is by Larry Hamilton, who explains it thus:
Graph and quote from Neven's blog.
With the gompertz model, it should be noted that once the average sea ice extent falls below about 0.5 million km^2, the Arctic will be ice free on some summers (but not all) due to variations in weather, so on this estimate the first zero ice minimum should be around 2030 even though there will be minimums with above zero ice for several years, even a few decades later.
Do feel free to include the 2010-2012 predictions, the model assumptions are not too unreasonable for short term predictions, but do also include the ones from #34 to show how sensitive projections can be to the assumptions on which they are based, which is a useful lesson.
It would be nice to use a method that handles the fact you can't have a negative extent properly (in the credible/confidence interval as well as in the projection). It might be a nice application for Beta regression (where the Gaussian noise assumption is replaced with a Beta distribution with the lower limit at zero and the upper limit learned from the data (or set to the surface area of the globe). Shame the GPML toolbox doesn't support it (yet).
Dang. You just took the wind from my sails. Being an ecologist and all, Gompertz is one of my favourite sigmoids. I was about to plug the data in, and whaddaya know, I refresh and discover that you've pipped me!
Pretty cool trajectory though...
Not that I would in anyway dispute Larry Hamilton's final sentence.
Bob Grumbine scooped you with a logistic curve fit:
By my estimation method, there's about a 50% chance (54%) that 2035 or some year before that will show zero ice extent for September. It's only 6% that we'd see zero ice in 2029 (or before). And rises to 96% that we'll see zero ice (for the month) in 2042 or before. The 'or before' is important.
Similar shaped graphics from model runs in Stroeve et al 2007, although the earliest ice free point in that paper was 2050.
Of course, this is not to say that the shape of a Gompertz is not itself an artifical fit to certain sets of data (usually it is), but in various ecological contexts at least I've found Gompertz nearly always seems to better describe the nature of trajectories than do logistics, von B's, or other of the stable of sigmoids.
Whether or not melting ice follows a Gomperts path, I suspect that it will follow one of the asymmetric sigmoids...
...for the sort of reason that Dikran Marsupial mentions in the first paragraph of #38.
43, Dikran,
I believe, however, that the "lower angle of incidence" theory presumes that incoming sunlight is a major factor in local warming/melt. I thought that it's been determined that the water temperature and not the sunlight itself was the main factor.
Admittedly, the increase in local warming due to more exposed water due to more ice melt will still be lower (due to the lower angle of incidence), but if the main temperature of the water comes from heat being transported in from outside of the Arctic, then I would think that would low angle of incidence would be a minor factor.
The bigger factor, in the end (I believe) will be flat out warmer water temperatures. Ten years from now, with the water that much warmer, everything will melt faster and more thoroughly, from start to finish. Eventually it will be so warm that it just overwhelms whatever ice is there.
For that reason, I think that the end will come surprisingly quickly.
[On the flip side, I do think the complete disappearance of the sun, dropping air temperatures to the point where things start to refreeze, is a major factor, and the fact is that the melt season is much shorter at the actual pole, so it may be a long time before the seas are warm enough to melt everything there before the sun sets.]
This gives the general shape of the projection:
Focussing in on the near future:
The periodic covariance function looks like a neat way to model the seasonality.
prediction for 2012 = 4.342434 (+/- 2.071584) 10^3km^3
prediction for 2013 = 3.306950 (+/- 2.227936) 10^3km^3
prediction for 2014 = 2.211567 (+/- 2.426253) 10^3km^3
1. As far as predictions and "tipping points", Eisenman and Wettlaufer (PNAS 2009), in their publication "Nonlinear threshold behavior during the loss of Arctic sea ice" state: Emphasis added.
The article also contains this graphic:
3. Another fascinating paper which I regret to not having internalized yet: "Influence of Initial Conditions and Climate Forcing on Predicting Arctic Sea Ice" by Blanchard-Wrigglesworth et al GRL 2011 The Death Spiral yet lives...
As a statistical exercise, those plots warm the cockles of my heart.
As a hint of what
maywill come, they churn my innards.Of course, as you and others of us have noted above, they are not parametric modellings of the phenomena, and thus at some point the differing ice-free-summer projections of extent versus volume are going to knock one (or each other) over.
It's interesting to consider how it will play out. For the trend in extent to hold into the future, something major would have to put a brake on the trajectory of (and thus on the causative factors for) loss of summer ice volume (= mostly thickness). That's a lot of 'inertia' with which to tangle. For the trend in volume to hold into the future, all that really needs to happen is for extent to suddenly collapse in magnitude, which can easily be anticipated once thickness reduces beyond a critical threshold.
I think all wise money would be placed on the latter scenario: at some point between 2015 and 2050 (and probably very much in the first half of that period) there will be a spectacular plummetting of the values for sea ice extent. I'm still of the inclination to consider that it will likely occur before 2020.
Whenever it does eventually happen, it will be instructive to see how humanity responds...