NSIDC 2018 Arctic SIE September average: July poll

[DavidR]

ATTENTION: There are two polls on the ASIF. This one is for NSIDC sea ice extent monthly minimum or September average (which is also used for the SIPN sea ice outlook), the other is for JAXA sea ice extent daily minimum (provided by ADS, previously by IJIS). Make sure you are aware of the difference before voting.

These are the September averages for the last 13 years (in millions km2,):

    2005: 5.50
    2006: 5.86
    2007: 4.27
    2008: 4.69
    2009: 5.26
    2010: 4.87
    2011: 4.56
    2012: 3.57
    2013: 5.21
    2014: 5.22
    2015: 4.62
    2016: 4.51
    2017: 4.80

[Brigantine]

You can't change your vote before polling closes?

And the top bin isn't open ended.

[DavidR]

You can't change your vote before polling closes?

And the top bin isn't open ended.

Neven will fix the poll so votes can be altered.  The creation system is a bit touchy about  that.

The top bin is now open ended.

[Neven]

I've changed a few small things (2018 instead of 2016), made it possible to change votes, and made the top bin closed again. The problem with the open-ended stuff, is that it makes it more difficult to calculate averages and medians. And besides, the last time September average was over 5.5 million km2, was in 2006. I might re-consider if it happens again, but until that time I'll take my chances for the sake of accuracy.

I'll be voting in around 10 days. I expect/hope NSIDC extent to be up and correct again, by then.

[Sterks]

I have gone up one semi-bin to 4.5 M - 5 M. But it is just to say the same, all points to an "average", slightly cool, post 2007 (or 2010s is you wish) year.

[Tetra]

Agreed with Sterks. Except went a bin lower as I think 2018 will be normal, but just one of the lower years in record.

[Paddy]

I've also gone up half a bin to 4.5 to 5.0. Somewhere similar to the past three years, in short.

[Richard Rathbone]

I've changed a few small things (2018 instead of 2016), made it possible to change votes, and made the top bin closed again. The problem with the open-ended stuff, is that it makes it more difficult to calculate averages and medians. And besides, the last time September average was over 5.5 million km2, was in 2006. I might re-consider if it happens again, but until that time I'll take my chances for the sake of accuracy.

I'll be voting in around 10 days. I expect/hope NSIDC extent to be up and correct again, by then.

10 days time is in the middle of another testing period. I'd either wait a week, or leave it to the deadline. I need to see momentum indicators for my July vote, which means PIOMAS has to be out, and I'd like to have heard something from Schroeder too.

The bottom end of the range goes down well below credible values, but the top end of the range clips the low frequency but still could happen range out of the poll. Its one of the reasons these polls bias low, the polls themselves are structured with the expectation that the only low frequency events that can happen are below rather than above trend and the voters pick up that message. I reckon they should go to at least 6.0 and that 7.0 is as credible as 0.0.

[Neven]

The reason there are all these bins below 3 million, is that it makes it easier to calculate medians and averages, without relying on some arbitrary number the statistician ascribes to 'below 3 million'. The reason there aren't any bins above 5.5 is because it's more work.

Even with perfect polls the predictions would come in too low most of the time, because this is a forum filled with people who worry about AGW.

Just be happy there aren't any bins below zero. 

[EgalSust]

4.75-5.25

Up a bin from last month. Noted that NSIDC Slater team came up with 4.91 in their SIPN June prediction.

Slater Probabilistic extent prediction seems to be at 5.71 for August 19th. Don't know how much that is affected by the satellite testing issues.

Good to hear about the background for the bin choice. I think I've read that the range of possibilities does have a proven psychological effect on the responses, but don't have the source for that now.

[Steven]

I voted for 4.75-5.25 million km2.

This is based on linear regression, using the NSIDC sea ice area in June as a predictor of the NSIDC September extent.  The calculation gives 5.0 +/- 0.8 million km2 (95% prediction interval).

Moreover, I tried a few variants of the calculation:

• Use only the last 15 or 25 years of data, rather than the entire dataset 1979-2018
• Use the June sea ice area for the Arctic Basin proper, rather than the entire Arctic (i.e., exclude the Hudson, Baffin, Kara and Greenland Seas)

But it turns out that the calculated values don't change much for those variants and are somewhere between 4.7 and 5.1 in all cases.

[icefisher]

4.64 +- .15.  Dispersion due to thinner ice may increase SIE but result in less volume.  Time will tell.

[Juan C. García]

Moving half a bin up, to 4.25-4.75M km².

[Richard Rathbone]

For the July poll, I take momentum indicators into account as well as the long term trend. These are awful for melt, the CPOM June SIPN entry, based on May melt pond modelling is calling for the highest September extent since 2006, 5.3 (+- 0.5)

I'd prefer to see their June analysis, but my crude June indicators (NSIDC area anomaly cliff and PIOMAS volume anomaly cliff) also point to low momentum as does the late appearance of general surface melt on SMOS and high NSIDC compactness.

So, I have to go up a lot, the question is just how many bins. For the moment, just 3, but if the CPOM SIPN July entry is out before the closing date (it normally just misses this) and their June analysis backs up their May analysis, it'll be 4.

4.75 - 5.25

[Richard Rathbone]

The reason there are all these bins below 3 million, is that it makes it easier to calculate medians and averages, without relying on some arbitrary number the statistician ascribes to 'below 3 million'. The reason there aren't any bins above 5.5 is because it's more work.

Even with perfect polls the predictions would come in too low most of the time, because this is a forum filled with people who worry about AGW.

Just be happy there aren't any bins below zero. 

When one of the methods with the best track record (CPOM) is calling for 5.3 in its June SIPN entry, and saying 6.3 (+2sd) isn't out of the question, 5.5 is too low a cut off.

[Brigantine]

The reason there are all these bins below 3 million, is that it makes it easier to calculate medians and averages, without relying on some arbitrary number the statistician ascribes to 'below 3 million'. The reason there aren't any bins above 5.5 is because it's more work.

Even with perfect polls the predictions would come in too low most of the time, because this is a forum filled with people who worry about AGW.

Just be happy there aren't any bins below zero. 

And my 2c is that, without any way for participants to register a prediction of >5.5, we're not collecting a verifiably representative dataset. There might be data above 5.5, we just haven't included it in our sample.

It might be easier to calculate *an* average, but it might not be an honest average.
Ideally, there would be an open-ended top bin, and it would be high enough that no one votes for it.

As a workaround in August I'd suggest open-ended placeholder(s) with a notice that your vote is only valid if you post your prediction in a comment. (in an appropriate format, containing a specified ctrl-f anchor, without edits after the poll closes). It's still more work, but the extra work is shared by more people

[Ned W]

That "workaround" sounds like a bit too much work for me, but YMMV.

I do agree that in the future the NSIDC poll should have more bins at the upper end.  Right now it has the same set of bins as the JAXA poll, but NSIDC monthly is consistently higher than JAXA daily.  And sure enough, if you look at the distribution of votes, it does look like the upper tail of the distribution is getting truncated on this one.

FWIW, the NSIDC Monthly version of the predict-o-matic currently predicts a September mean of 5.05.  So the topmost two bins actually contain the current statistical best estimate.  And the 95% CI extends up to around 6 Million km2.  So it would not be particularly far-fetched for someone to want to choose a "5.25 to 5.75" bin or even a "5.5 to 6.0" bin.  At least, no more far-fetched than choosing a bin below 4.0 ... of which we offer a plenitude.

All that said, I am voting 4.75-5.25 in this poll, at least for now.  I am guessing that the predict-o-matic is going to stop its upward creep (like the JAXA predict-o-matic, the NSIDC one was very flat up until June, then began a slow but steady rise).

[Ned W]

All that said, I am voting 4.75-5.25 in this poll, at least for now.  I am guessing that the predict-o-matic is going to stop its upward creep (like the JAXA predict-o-matic, the NSIDC one was very flat up until June, then began a slow but steady rise).

Here's how the NSIDC monthly extent predict-o-matic has evolved this year:



Looks rather like a hockey stick, doesn't it?   June really was remarkable. 

[DavidR]

I do agree that in the future the NSIDC poll should have more bins at the upper end.  Right now it has the same set of bins as the JAXA poll, but NSIDC monthly is consistently higher than JAXA daily.  And sure enough, if you look at the distribution of votes, it does look like the upper tail of the distribution is getting truncated on this one.

This is difficult to do a with the current poll as there is no way of inserting options above the top option without eliminating all the votes.  As there hasn't been a September minimum above 5.26  since 2006 anything above 5.5 seems unlikely but then so does anything below 2.0.  If the extent is still  high  towards the end of the month I will create the next poll with extra bins.

[RikW]

I still have a feeling the ice pack has changed so much last years that it is really hard to make a prediction based on recent years. There are large areas that are lagging behind and will still melt out completely. Seeing the PIOMAS volume for the CAB, which is still 2nd lowest, I think there is a possibility extent will drop faster than normal in the coming 2 months. So I go for the 3.75-4.25 bin

[Ned W]

So I go for the 3.75-4.25 bin

Of the top eight bins in the poll, that's the only one that has never occurred. 

Probably not this year either, but it's got to happen some time...

[Juan C. García]

Moving half a bin up, to 4.25-4.75M km².

I am going up another half bin, to 4.5-5M km².
Hopefully, I will go up again in the near future.   

Surprise! There is not a "Remove vote" option now on this poll.   

[Neven]

Should be there now, Juan.

[Rob Dekker]

I entered 4.75 - 5.25, based on this method, which uses land snow cover, sea ice area and sea ice concentration in June as predicting variables :

https://forum.arctic-sea-ice.net/index.php/topic,292.msg162415.html#msg162415

[Paddy]

The concentration map, to my mind, is currently looking more icy in the areas that matter than 2016, but less icy than 2017: https://sites.google.com/site/arcticseaicegraphs/concentration-maps/sic0707

So I'm sticking with 4.5 -> 5 for now.

[Richard Rathbone]

For the July poll, I take momentum indicators into account as well as the long term trend. These are awful for melt, the CPOM June SIPN entry, based on May melt pond modelling is calling for the highest September extent since 2006, 5.3 (+- 0.5)

I'd prefer to see their June analysis, but my crude June indicators (NSIDC area anomaly cliff and PIOMAS volume anomaly cliff) also point to low momentum as does the late appearance of general surface melt on SMOS and high NSIDC compactness.

So, I have to go up a lot, the question is just how many bins. For the moment, just 3, but if the CPOM SIPN July entry is out before the closing date (it normally just misses this) and their June analysis backs up their May analysis, it'll be 4.

4.75 - 5.25

Its out (at least to Neven), and it says there were rather more June melt ponds than May suggested there would be, so I switched to 4.5 - 5.0.

[DavidR]

All that said, I am voting 4.75-5.25 in this poll, at least for now.  I am guessing that the predict-o-matic is going to stop its upward creep (like the JAXA predict-o-matic, the NSIDC one was very flat up until June, then began a slow but steady rise).

Here's how the NSIDC monthly extent predict-o-matic has evolved this year:



Looks rather like a hockey stick, doesn't it?   June really was remarkable.

Ned, Is it possible to include the graphs from the past 4 years.  2015 and 2016 both had, like this year, very low melt rates for June whereas 2014 and 2017 had fairly normal June melt rates.  It  would be interesting to compare a few of the predict-o-matics' predictions with final performance.  2016 had a late run whereas 2017 had a late slump; we might find that June 1 is a better predictor than July 1.

                               2014      2015      2016    2017      2018
Jun 1st     Mkm2        12.052   11.568   10.998   11.78   11.361
July 1st   Mkm2          9.324     9.842     9.179   9.244     9.746
Loss       Mkm2          2.728     1.726     1.819   2.536     1.615
Sept        Min              5.220     4.620     4.510   4.800   

[Steven]

The July SIPN report is out:

https://www.arcus.org/sipn/sea-ice-outlook/2018/july

The median of the 39 projections for total September 2018 Arctic extent is 4.7 million square kilometers with lower and upper quartiles values of 4.4 and 4.9 million square kilometers. ... In contrast to last month's (June 2018) report, the spread in the dynamical models is much reduced. Predictions of September extent from statistical and heuristic approaches are generally below the median value observed in 2017, whereas the modeling contributions predict a September extent similar to that from last year.

[Rob Dekker]

All that said, I am voting 4.75-5.25 in this poll, at least for now.  I am guessing that the predict-o-matic is going to stop its upward creep (like the JAXA predict-o-matic, the NSIDC one was very flat up until June, then began a slow but steady rise).

Here's how the NSIDC monthly extent predict-o-matic has evolved this year:

Ned, do you have an update on the NSIDC monthly extent predict-o-matic ?
With the recent stall in melting, it looks to me that around 5.0 for the NSIDC monthly extent is reasonable.

[Steven]

do you have an update on the NSIDC monthly extent predict-o-matic ?
With the recent stall in melting, it looks to me that around 5.0 for the NSIDC monthly extent is reasonable.

Here is an updated version of Ned's graph.  The latest prediction using that method is 4.86 million km2.

[Ned W]

do you have an update on the NSIDC monthly extent predict-o-matic ?
With the recent stall in melting, it looks to me that around 5.0 for the NSIDC monthly extent is reasonable.

Here is an updated version of Ned's graph.  The latest prediction using that method is 4.86 million km2.

Nice!  Thanks, Steven.

[Rob Dekker]

NSIDC published the results of the 2018 Arctic SIE September average : It's 4.71 million km^2.

ftp://sidads.colorado.edu/DATASETS/NOAA/G02135/north/monthly/data/N_09_extent_v3.0.csv

Looks like we were spot-on this year, folks !

[Stephan]

Not exactly for me, I voted for the 4.75-5.25 mio. bin

[Rob Dekker]

Not exactly for me, I voted for the 4.75-5.25 mio. bin

Me too. I was expecting something close to 5 million km².
But September weather went really weird, so it never made it there.

I think anyone voting between 4.25 and 5.25 was in the right ballpark, and that seems to be the majority of the votes in this poll.

Overall, we did much better than last year and the years before.

[FishOutofWater]

I didn't vote but that was in the range I discussed in my comments. Most people this year realized that the cold July weather would keep the September extent substantially above the 2012 record. We did well.