Monday, October 05, 2009

The Arctic sea ice prognosticator of the year is...

Hypocentre!

The JAXA minimum arctic sea ice extent value for 2009 was 5249844 km2, which was recorded in September 13.

The final score for each contestant is the value of their Gaussian curve at 5249.844, and is listed in table 1.

Table 1: Contestant final scores and original guesses.

The results are plotted in figure 1.

If I recall correctly from the original comments, Hypocentre is a hard rock geologist, and the 2nd through 5th place winners were all climatologists.

Figure 1. Contestant curves and final 2009 minimum sea ice extent (red line).

Hypocentre, as the winner of the contest, may now nominate a topic for a future blog post (post nomination in comments, please).

Everyone else, tune in next (northern) summer solstice for the 2010 version of this contest.


Commentary, corrections, and sour grapes are welcome in the comments section.

10 comments:

hypocentre said...

Ha ha, a seismologist beats the climatologists at predicting the minimum Arctic sea ice extent!

The reasoning I used was that the 2007 extreme minimum was caused by unusual weather (particularly wind direction) rather than climate. In 2008 the recovery was only partial with relatively thin new ice. I reasoned that 2009 would have continued recovery, probably back to 2006 levels which proved to be the case.

Looks like too many climatologists were believing all the global warming hype:)

I'm tempted to set our genial host the future blog post topic of "Global warming ended in 1998 - discuss".

Hypocentre (with tongue firmly in cheek and big smile on face).

C W Magee said...

1998 AD, BC, or MA?

Silver Fox said...

Congrats to Hypocentre on your status as prognosticator!

crandles said...

Congrats Hypocentre.

Since you say sour grapes are welcome, I'll have a go and try and continue with my earlier comments:

"It would seem to me that only Penguin and possibly jyyh have put in suitable levels of uncertainty. Everyone else has reduced their uncertainty estimate to silly levels with the aim of claiming their spot at their favoured estimate (or claiming a wide range knowing the uncertainty is high)."

"Anyway, I suggest a new competition where people have to submit their central estimate and uncertainty range for each of the next three years before 31 Oct 2009."

"Crandles, three years is a long time in the land of blogs, and I certainly don't want to commit to doing this through 2012.

I'll probably do this again next year- but with an earlier closing date. I'm thinking solstice to perihelion."

I was vaguely trying (and completely failing) to say that the requirement to state a sigma seemed pretty redundant if people have to go for stupidly confident sigma to give themselves a sensible winning range. If most people are trying to give themselves a large winning range then abandoning the requirement for a sigma makes more sense to avoid gaming the system.

Obviously switching to score being number of km^2 from central estimate would give me a higher ranking and posting this is just sour grapes.

I was hoping there might be a competition from each soltice to nearest max or min. I also think it would be interesting to elicit next three years minimum estimates even on the basis that you are not committing to blog a result in 2012 on the basis it is more interesting to see the differences in estimates than a result.

Chris Phoenix said...

I'm not sure what "MA" stands for (million years ago?), but Chuck's comment got me thinking: Could we base a date standard on the scientific decision that we are now in the Anthropocene era, in which humans affect the earth's geography? That's a pretty significant decision / recognition, and it should remain relevant until most of us are off-planet.

C W Magee said...

Chris:
"present" is 1950 for 14C people, so why not pick that?

crandles:
I see where you are coming from, but people taking 'realistic' guesses would be uninteresting in that most of them would all lie within error of each other.

However, some of your real statasticians should be able to estimate a narrow guesser's estimation of the true standard deviation based on the stated st. dev., its position relative to other guesses, and the total number of guesses made. Sadly, I don't have the mad math skillz needed to do that.

crandles said...

Chuck,

I don't believe that, but I could be wrong.

Suppose two people, before James Annan submitted his guess, had guesses of 5050 one with sigma 300 and one with sigma 350. The optimum guess for maximizing the winning range may be something like 5050,80. In this case it would be impossible to tell from a 5050,80 submitted guess whether the person who got there first and made that guess believed the sigma to be 300 or 350.

Now maybe that is a special case and I suppose it is possible that if two people had guesses of 5010,300 and the other 5040,350 you might be able to tell them apart because the former would go for 5047,78 while the latter would go for 5049,79. However I would have thought there would have to be different beliefs that would result in the same submitted guess.

Not that this makes much odds - who would want to be bothered trying to calculate it even if it was possible to arrive at unique answers.

James Annan said...

I think there is a basic difficulty with evaluating a probabilistic forecast on the basis of one event :-) It seems that considering the uncertainty didn't add anything on top of just picking a best guess and seeing who is closest.

crandles said...

Not too sure about the 'Not add anything'. Surely the ability to take the mickey out of Chris Town and Alistair al la

'How many scores of 1E-78 does it take to make you realise you are being excessively alarmist?'

Doesn't that add something?

(My 5E-32 isn't too clever either.)

C W Magee said...

Something else y'all need to remember is that some of us had our formal education back in the days where experiments were too time consuming and expensive to repeat. Thus our formal statistical knowledge is poor and this sort of thing is good for familiarization (or, in my case, for figuring out how to plot Gaussians correctly).