Yay, more nonsense about sea ice: the traditional "US Navy predicts summer ice free Arctic by 201x", where this time x=6. Does anyone actually believe this rubbish? If so, I have money just sitting around, bored, twiddling its little green fingers and waiting to take your bet. If "summer ice free" means "oh yeah, not actually ice free, but less than 1 million square km" then please form an orderly line. Even odds, let's say £1k. Who's first up?
Do I see Nafeez Ahmed there? Or Wieslaw Maslowski? No? How odd.
Found by the lost [*]:
* Ed Hawkins @ed_hawkins: Brave prediction by Maslowski for ice-free Arctic in 2016 (+/-3) | http://www.theguardian.com/environment/earth-insight/2013/dec/09/us-nav…...
* Gavin Schmidt @ClimateOfGavin: @ed_hawkins do you mean 'brave' as in unphysical, unpublished and even at the time demonstrably unlikely?
[*] You can have a point if you can find the source for that. Hint: google won't help you.
* When will the Summer Arctic be Nearly Sea Ice Free?
* Sea ice betting report for 2012. Yes, I know I haven't written 2013 yet, but it was dull.
*The Future of Arctic Sea Ice Annual Review of Earth and Planetary Sciences, Vol. 40: 625-654 (Volume publication date May 2012), First published online as a Review in Advance on March 8, 2012; DOI: 10.1146/annurev-earth-042711-105345, Wieslaw Maslowski,1 Jaclyn Clement Kinney Matthew Higgins and Andrew Roberts
Not that I am interested in taking up the bet, but is that 1m km^2 area or extent and by what measure? Average Sept extent is a lot more likely to stay above 1m km^2 than daily area.
[Ha, the link to a previous post was in terms of extent. Anyone interested in the bet is free to try to negotiate, though -W]
Well, looking at the PIOMAS minimum volume trend, 2016 seems quite plausible to me, even with the "60% recovery" that the potty media were screaming from the rooftops a couple of months ago.
Arctic Sea Ice Minimum Volumes 1979-2013
Without signing on to the conclusion, the argument appears to be that if the ice thickness in the winter continues to fall as it has in the last 20 years or so, pretty soon it will be so thin that it melts out in the summer. Oh yes, and that GC'ms suck at this issue
In the beginning Maslowski was pointing out (at least as early as spring 2006) that *if* the trend (volume loss) continued the arctic would be nearly ice-free by 2013. By “ice-free,” Maslowski meant a more than 80% drop from the 1979-2000 summer volume baseline .
One of the interesting facts that gets little attention is that the Naval Postgraduate School was predicting in 2001 that the northwest passage would be open to non-icebreaking ships by 2015. How many people would have taken *that* bet circa 2001?
[Got a ref for that? -W]
"Within 5-10 years, the Northwest Passage will be open to non-ice-strengthened vessels for at least one month each summer"
Naval Operations In An Ice-free Arctic Symposium, 17-18 April, 2001.
My understanding is the prediction was based on NPS modeling studies.
[Thanks. I didn't doubt you had it, BTW, I just thought you'd be quicker finding it than me.
I'm not sure the text you quote counts as a prediction. Its clearly one opinion. But Its prefixed by "see appendix A" which contains the usual stuff. For example, there's:
Future scenario: A conservative scenario is that by 2050 the observed trend will reduce summer minimum ice extent by 15%; this is an extrapolation of the satellite observations which are quite reliable and are not contradicted by climate model forecasts. For volume and thickness, a conservative estimate is obtained by extrapolating model forecasts that are not contradicted by sparse observations. By 2050, the end-of-summer volume can be expected to be down by about 40%, of which about 15% would be due to decreased extent and the remaining 25% would be seen in an end-of-summer thickness reduced by 25% to about 1.5 m.
"see appendix A" seems somewhat different to
"The major discussion items of this brief are contained in Appendix A"
Appendix A seems to be labelled "A scenario" and suggests it is a compilation of expert views which are expected to be honest expectations of the expert scientific community.
The summary in appendix A seems lacking in dates and in parts at odds with what is in the main part of report. eg 2 different figures for extent reduction roughly 30% and conservative consideration of models being 15% reduction in extent has changed to at least 20% in the main report. So perhaps the main report is 'a view'. OTOH volume reduction at 40% is the same.
If the appendix is intended to summarise all expert opinion, you can't really expect it to be making different opinions to the expert community.
Coming up with one accurate prediction is not all that spectacularly good if there are also lots of wrong predictions.
There is also 'a view' of
"Within 5-10 years, the Sea of Okhotsk and the Sea of Japan could potentially remain ice-free throughout the year."
which hasn't happened. (It was also in Appendix A.) While the 40% volume reduction over 20 years has easily been exceeded. So there are over predictions of change as well as under predictions.
My impression is of a document saying lets treat these guesstimates as likely occurrences and consider what effects they have and what planning we need to do. It doesn't seem the place for an out on a limb prediction that the scientists really believed and would bet on.
I am not quite sure what to conclude. Have they done quite well to have over-predictions of change as well as under-predictions despite all the models being terribly wrong? Or does this indicate poor understanding and with better understanding they should have under-predicted the change in all the predictions/views in line with models at the time?
Object lesson in how to get attention first then worry about the detail later. In response to the headline, simply, 'no, it doesn't'.
The paper is more closely addressed in the sub heading - it strongly suggests that GCMs can't capture sufficient detail to be reliable predictors of the timeline. But their model is better. So please spend more money on it.
On your predictions; we've had four record lows(in extent) since 2002. at this rate, the next record low will be between 2014-2016. Assuming an area of the Arctic Ocean of 15.046 Mkm2, Last year's extent (3.46) represents about 23% of the ocean area. If 10% is the 'target' definition of 'Ice Free', SSIE needs to halve again (roughly) either cumulatively or in one season. It's 'only' a loss of 1.94Mkm2, which would be covered by losses in the central Arctic Basin.
So, like the paper says, it might be the lower bound of what is possible, given ice age and thickness and the AO, this year's Siberian temp. anomalies; all it would take is one particularly 'suitable' season. Is it likely? Hmmm...
ISTM that people tend to forget the audience:
"Naval Operations In An Ice-free Arctic Symposium"
The Navy is interested in planning for future scenarios in terms of ships, budget, training etc. So their interest is in accurate predictions if possible, but I should think they'd prefer error be in the worst-case direction, rather than best-case.
More to the point take a look at the sea ice area, extent and mass curves up to 2000, and then see where it extrapolates to zero. There has been a huge fall off since 2000, an extrapolation to 2050 was not unreasonable.
[I don't think anyone was suggesting that extrapolating to 2050 was unreasonable then. Indeed, its still a reasonable prediction now -W]
Funnily enough, one of the commenters on this article in the Guardian challenged me to a £100 wager on this topic. My stance being that the arctic sea ice area for september 2016 will be *higher* than september 2007 and september 2012. A few commenters on that thread were advising me not to go ahead due to the certainty on their part that i will lose- but... i think i will win.We will see...!
Phil, was that area or extent? Daily minimum or monthly mean?
If you haven't completely worked out the terms of the bet, you might want to use that as an excuse to back out. Otherwise, your £100 is at great risk.
To demonstrate, I will use the NSIDC September monthly area data from here:
I am looking at the column labeled "area" rather than "extent". The Sept. 2007 value is 2.78. The 2012 value is irrelevant to your wager since the 2007 is the higher of the two.
The trend appears basically flat until the mid-1990s, then turns downward. Since 2000 the trend is close to linear. For 2000-2013 I get the following linear model:
b0 (intercept): 276.2847
standard error: 0.4367
This predicts an area of 2.28 million km2 for Sept 2016, which is 0.494 million km2 lower than 2007.
The probability of a value higher than 2.78 in 2016 would appear to be approximately 12% to 13%. You would seem to have an 88% chance of losing your £100 if you proceed, based on the past decade's observations.
As a disclaimer, I am not a statistician nor a financial analyst, just a random anonymous stranger on the internet, and the above may be riddled with errors.
What happened with the bet? Has it been taken up or are you trying to establish enough confidence that the other will pay up?
Are you interested in taking more bets on the same terms? Not sure if you are too close to Williams position but perhaps other counter-parties could be found if you are interested and sufficient trust could be established.
the bet is on informal terms but i will pay up (by paypal!) if i lose. the other chap will stick to his word as well,i hope.
the terms are area @ the september 2016 lowest (whichever day that occurs) and for me to win the sea ice area ,as quoted by the recognised authorities (NSIDC,JAXA),must be higher than the 2007 min. for the other chap to win, it must be lower than 2007 even if it is higher than 2012. i guess i didn't need to mention 2012.
i can afford to lose £100,but i wouldn't want to risk any more than that!
thanks for the kind advice and risk evaluation :)
OK, I haven't looked at the daily minimum data, but based on the NSIDC September monthly average:
(1) If 2016 follows the trend of the data 1995-2013, you have a greater than 80% chance of losing £100.
(2) The same is true (80% chance of losing) if you do that calculation with any other start year (1995, 1996, 1997...) up through a start year of 2004.
(3) If you think the trend began "bottoming out" after 2004, you might choose to ignore all the earlier years. In that case, the regression model is based on an increasingly small number of data points, and is thus increasingly unreliable. But, with that in mind:
(a) The trend 2005-2013 would suggest that you have a 78% chance of losing.
(b) From 2006, it is a 66% chance.
(c) From 2007, it is a 36% chance.
(d) From 2008, it is a 52% chance.
(e) From 2009, it is a 63% chance of losing.
After that, you are looking at trends based on fewer than 5 data points, so I didn't even bother calculating the trends.
So, to sum up -- if you ignore the data from before 2007 and focus only on the last 7 years, you might expect to have a pretty good chance of winning (almost 2 in 3!)
But if you look at any longer record (or, for that matter, a shorter record) then you would conclude you will probably lose. It's only if you cherry-pick 2007 as your start year that the statistics would reassure you about your chances of winning.
Thanks for running those numbers J.
It is of course possible to believe in something other than a linear trend. Before this year, quite a lot of people saw downward accelerating trend. Models predict a decelerating trend as we get to lower ice levels. There might be some people believing in solar cycles, galactic space-dust or who knows what.
So I am not sure that the one start year in 30? for linear trends means much but I guess it is as good a starting point as any.
What I am wondering is, how do you start to account for different possible beliefs about whether decline will be steeper or less steep than linear?
That would be harder if we were talking about projecting further out. But 2016 is only three more seasons. I don't think the differences among linear and various nonlinear models should matter as much over that short a time frame.
I remember on Neven's site last summer one commenter made a forecast of sea ice extent for the near future using simple linear regression. Someone else said basically "Hey, that's too unsophisticated, you need to look at something like a Gompertz model" until it was pointed out that over the short time period involved there was virtually no difference in predictions between the linear and Gompertz models.
If the question is "when will Sept extent drop below 1 million km2?" then the issue of what model to choose becomes critical. A linear model would say something like 2030. Quadratic or exponential models much sooner, other models later.
But why am I writing this? You probably know it better than me!
yep cheers for that analysis J.
It is a big risk for me ,but it will be interesting to observe what happens either way...
I want a piece of this action. put me down for 1000 at the 1 million minimum not being reached by said date.
[Alas, I fear its people on the other side of the bet we need -W]
I posted the trend in minimum volumes in #2.
Remember that when volume=0. Extent and area are also 0.
The trend is clearly exponential, even with the wiggles, but I'm not sure what volume would correspond to a 1m km² area.
In 2012 the ice was mostly a like a slush puppy.
One thing is sure for me, no way would I bet on a recovery!
I've got the PIOMAS numbers in a spreadsheet somewhere and don't feel like trying to find them, but it would be interesting to see how close Maslowski is to his prediction of an "ice-free" arctic. Based on his criterion of an 80% volume reduction from the 1979-2000 baseline, I'd guess we're somewhere in the 75% reduction area now. Anyone care to chime in with the real numbers?
[I think its quite interesting that there's been an "rebound" (careful with the language here) in PIOMAS this year; in that the monotonic decline has gone. e.g. http://psc.apl.washington.edu/wordpress/wp-content/uploads/schweiger/ic… -W]
Great graphics Andy!
>"The trend is clearly exponential"
That is further than I was prepared to go even before 2013. 2013 with it return to the linear trend in volume terms:
doesn't cause you to at least pause for thought before saying 'clearly exponential'?
If you really believe clearly exponential, then there would appear scope for betting against William though I would recommend trying to negotiate terms better than I got bearing in mind what happened in 2013 after I agreed my bets.
Been intending to say William's '2013 was dull' is impressively modest given how it makes it appear likely that he got good bets against me.
2013 is only one year and perhaps there are reasons to dismiss this year as a one-off eg lack of di-pole patterns and blocking events. OTOH perhaps models are now more believable with their gompertz shaped curves beginning to make an appearance with sufficient ice still left.
Whatever way you interpret it, there now appears more noise than might have appeared to be the case before 2013.
"Remember that when volume=0. Extent and area are also 0."
Can someone please explain that argument to me? Because I see it all the time over at Neven's and elsewhere, and I don't see the logic.
You have two series that ought to reach zero at the same time, but are following different trends.
So, yes, obviously, one (at least) of them must depart from its current trend to fall into line with the other.
But ... why the assumption that extent must speed up to match volume?
Couldn't Andy just as well have said:
"Remember that when extent and area are greater than 0, volume is also greater than 0."
Volume is a fairly good proxy for the energy balance because of the high latent heat capacity. Of course the energy balance will change as the ice gets thinner, but at least you are nearer to modelling some physics than with area and extent. William promised an explanation of why he didn't believe volume was better than area or extent when he had more time. Approaching Christmas may not be a good time to ask.
[Um, yes, I'm still with J on this one. Or at the very least, he is correct, in that there are two series and there is no reason to automatically assume that one is "correct" -W]
But isn't there a negative feedback involving heat loss at the ocean-atmosphere interface? As extent decreases, heat loss increases? There is no such negative feedback constraining volume loss at first -- going from 4m thick to 3m doesn't produce this negative feedback -- so volume starts by dropping faster than extent. But when thickness gets near zero the ocean-atmosphere heat flux increases and every winter new ice forms. So the volume curve adjusts slowly to something closer to the extent curve.
The above was a response to crandles, and was an attempt to provide a physical explanation for why one might prefer Curve #2 as an alternative for crandles's physical explanation for why one might prefer Curve #1.
I still think a lot of the people who say "You can't extrapolate the extent curve because the volume hits 0 first and at that point extent has to be 0 also" don't necessarily have a physical mechanism in mind; it often seems to be an intuitive idea that whichever line hits 0 first must be the most important one.
I agree there is negative feedback in the early part of the freeze season and this makes Gompertz curve shapes more likely.
There is also positive feedback in the melt season. Dealing with both a positive feedback and a negative feedback is complicated.
I therefore think it makes sense not to look at the annual minimum series but at the maximum volumes and the falls from max to min. I don't see any reason for an upward trend in the maximums at best it might level out due to the ice regrowth negative feedback. (Has levelled out based on last 3 years seems pretty dubious to me.) While the maximum continues to decline then the fall from max to min should increase due to the positive feedback.
So I think this gets us further than looking at trends in the september minimums.
The MIN/MAX numbers don't tell us a lot. Think of a high latitude lake: many will have zero MINs in summer and MAX out in winter. You can also have a process where summers get warmer and winters colder - an average increase does not mean it has to be uniformly distributed.
The question no one knows the answer to is: Was 2013 a year indicating negative feedbacks will now begin to slow the losses - or was it just normal weather variation?
I opened my PIOMAS spreadsheet and the 1979 average minimum was 13770 km^3
The 2012 minimum was 3263 km^3 or a reduction of 76.3%
The 2013 minimum was 4942 km^3 or a reduction of 64.1%
Volume at minimum would have to decrease from the 2013 minimum by 2188 km^3 to reach the 80% volume reduction level. 5 years in the satellite era have seen larger year over year losses than 2188^3, including 2007 and 2010.
Previous post should have said
"I opened my PIOMAS spreadsheet and the 1979-2000 average minimum was 13770 km^3 "
Since extent is defined as the region defined by ice cover greater than or equal to 15% (or sometimes 25%) ice, sea-ice extent will hit zero before either area or volume. They will not hit zero together.
However, extent tends to be larger than area when there is a core of well-packed ice, so there will be an interesting crossover in the late stages of the arctic ice pack
>"extent tends to be larger than area when there is a core of well-packed ice"
I thought when extent > 15% count whole of area of cell for extent but count %* cell area for area. I think that if extent is less than 15% then no area (or extent) in the cell is counted, but I could be wrong.
If I am right, extent is always greater than or equal to area and they both hit zero together.
Not sure about volume's minimum threshold, but discussing minimum thresholds seems a little lame.
Unless they explicitly defined their terms, I'd be inclined to bet the unstated part of the Navy paper's prediction reads
the "Arctic Ocean"* is expected to be "ice-free*"
*As defined for Naval surface and submarine operational purposes, the term "Arctic Ocean" means [redacted] and the term "ice-free" means [redacted]
Does salinity at under sea ice/water boundary matter? When looking at the column underneath the ice, How fast is the heat transferred? Do we measure it?
[Salinity certainly matters a lot. But its pretty hard to summarise the effects -W]
An observation on Extent, Area and Volume..
Extent can mean 15% sea ice area. Area just means ' some detectable ice'.
So, for a 50cm thickness, you could have:
250km3 Volume (within error of zero)
500,000 km2 area, and
3 million km2 'extent'.
You probably wouldn't get such extremes, but this does show how volume can get very close to zero whilst other metrics are a fair way from zero.
Looking at september volume minimums:
All of the first 7 data points are below trend, all of the last 7 data points are below trend and in-between, 19 out of 21 points are above the line. A linear trend plus uncorrelated random noise seems very unlikely to generate that pattern.
At what point does it become appropriate to dump a linear trend plus random noise? If/when that point is passed, what do you replace linear trend plus uncorrelated noise with? Correlated noise? Gompertz? Exponential? Models seem to suggest Gompertz shape.
Before 2013, there was negligible difference between exponential and gompertz shapes for 1 year ahead forcasting.
Now there is quite a difference even at just 1 year ahead:
see 1st graph here.
2016 at 3 years out, the difference forecasts from different curve shapes are quite different.
How about transforming the metric of the bet from ice area or thickness into a wager on circumpolar albedo averaged over an agreed summer interval?
Is there a good long record of 'circumpolar albedo'?
Different things are happening in arctic and antarctic so I wouldn't want to mix them.
Might be interesting to see the data - is albedo acceleration more clear than in sea ice volume data? Is that your reason for wanting to bet on albedo rather than sea ice? If not, why would you prefer to bet on albedo?
from the "Monterey County Weekly" -- a newspaper nearby the Navy Postgraduate School:
NPS researchers predict summer Arctic ice might disappear by 2016, 84 years ahead of schedule.
Lt. Dominic DiMaggio describes the area captured by NPS’s innovative Arctic climate model, a project involving 25 researchers and graduate students from 10 institutions.
.... an effort that will change the way the world thinks about the world. Their project is the Regional Arctic System Model (RASM), and it is arguably the most advanced – and accurate – Arctic climate model in existence.
[OK, so this is Maslowski again, but this time its clearly described as a "lower bound" not a prediction - effectively, they are guaranteeing that the Arctic won't be ice free before then -W]