NSIDC 2015 Arctic SIE September minimum: June poll

[Neven]

ATTENTION: There are two polls on the ASIF. One is for NSIDC sea ice extent monthly/September average minimum, the other is for Cryosphere Today sea ice area daily minimum. Make sure you are aware of the difference before voting. You can discuss various extent/area data sets in this dedicated thread.

-----

This NSIDC extent poll will run for 10 days (until June 14th). Until then you can change your vote. There will be a new poll next month.

Here's how things are currently looking based on data up to June 2nd:



These are the September minimums for the last 8 years (in millions km2, found here):

    2005: 5.57
    2006: 5.92
    2007: 4.30
    2008: 4.73
    2009: 5.39
    2010: 4.93
    2011: 4.63
    2012: 3.63
    2013: 5.35
    2014: 5.28

You can use the comment thread below to motivate your choice, but discuss various SIE/SIA data sets in this dedicated thread.

[Neven]

Oh, and there's a meta-vote going on too: I'll only open a next vote in July if at least 100 people vote. EROEI and all that. 

[Neven]

Based on my interpretation so far, as just posted on the ASIB, I'm going to go for something somewhat below the last two rebound melting seasons: Between 4.75 and 5.0 million km2. I'd go one bin higher, but that's no fun. 

[Peter Ellis]

Wild-ass guess, somewhere between 2007 and 2011

[Nightvid Cole]

I think it will be similar to 2012 levels, due to fast-retreating snow.

[jdallen]

I'm using total annual melt, and expecting between 10.5 and 11 million KM2 to melt out by late September. I also think the melt season may extend into early October. 3.25-3.5.

[seaicesailor]

A lot more potential than past 2 years at this time of season, but a lot more volume than 2012. I go with 2007 extent.

[Paddy]

Gone for a relatively conservative "similar to 2011" vote for now, but if the forecast arctic heatwave materialises I may need to revise this down...

[OSweetMrMath]

I have been intending to do some data analysis of my predictions from last year, and to update my method for this year, for the last month.

Unfortunately, I've been absolutely clobbered by other responsibilities and haven't had time to work on it. Qualitatively, the observed May extent was well below what my model would have predicted. However, September is still sufficiently far away for the current extent to not have a large impact on my predicted extent.

Without taking the time to update the data and run the calculations, my best guess is that my model's extent prediction would be somewhere in the range of 4.5-5.0 million sq km. I will hold off on an "official" prediction until I can update my model.


My model for extent does not take area or volume into consideration, but it's very interesting that the behavior of the extent has diverged from the behavior of the other measures this year. My guess, and this is completely unsupported, is that the measures will become more consistent with each other by September. Furthermore, I think it's more likely that extent will start behaving more like volume than the other way around.

Also, 100 votes is a high hurdle. Looking back at previous years, 60-70 seems more typical.

[Nightvid Cole]

How do I change my vote here? I want to go down to 2.75-3.0 M km^2 now since I was basing my previous prediction on a comparison of snow cover for June 19th, 2012 instead of June 12, 2012.

[plinius]

I'd first wait how much snow remains, since forecasts there are chronically unreliable...

[Neven]

How do I change my vote here? I want to go down to 2.75-3.0 M km^2 now since I was basing my previous prediction on a comparison of snow cover for June 19th, 2012 instead of June 12, 2012.

I could've sworn I ticked the box that allows voters to change their vote. Done now. There should be a button somewhere that says 'remove vote'.

[greylib]

A bit of ultra-scientific coin tossing gave "Between 0.75 and 1.0 million km2" 

[Peter Ellis]

Worth noting: the Yamaguchi lab have produced their forecast for this summer, which can be applied as an overlay when looking at the AMSR2 data here:
https://ads.nipr.ac.jp/vishop/vishop-monitor.html?N
http://www.1.k.u-tokyo.ac.jp/YKWP/2015arctic_e.html

Their prediction is for the summer minimum to be essentially identical to 2011 in extent and shape/distribution.

[Buddy]

3.25 - 3.5 New record low....or close to it.

Why?

1)  Warm SST continues
2)  Snow melt ahead of past years at this time of year
3)  Already ahead of record low 2012 extent as of now by 500K or so
4)  Ice condition continues a long deterioration in condition of the ice
5)  Expectation that "June cliff" will occur

[Neven]

Worth noting: the Yamaguchi lab have produced their forecast for this summer, which can be applied as an overlay when looking at the AMSR2 data here:

Thanks for that, Peter. Great explanation on their web page. Does this team participate in the SIPN Sea Ice Outlook?

[Nightvid Cole]

How do I change my vote here? I want to go down to 2.75-3.0 M km^2 now since I was basing my previous prediction on a comparison of snow cover for June 19th, 2012 instead of June 12, 2012.

I could've sworn I ticked the box that allows voters to change their vote. Done now. There should be a button somewhere that says 'remove vote'.

Thanks.

[Tensor]

After pulling out my thread and protractor and measuring the decent angle of all the previous year's, I put the string down on the graph at what appeared to be the correct decent angle.  My problem was I couldn't add all the wiggles in the thread, that the other lines, on the graph had.  The problem was the damn thread kept moving every time I tried to add a wiggle.  So, finally I just rolled my 22 sided die and came up with 4-4.25 million k²

[DavidR]

3.5-3.75 M as its the September average which is usually a couple of hundred thousand km^2  above the daily minimum.

Lots of confusing indicators this year with no real explanation of why we are so low. 
My  plateau hypothesis says go low so I'm going low.
Something near the record and hedging my  bets on whether a record is set.

[anthropocene]

3.5 - 3.75M for me.  Basically around 2012 value. Reasons for this (in no particular order):
- Beaufort melting early means lots of warm water at the high point of the arctic so it can flow down through the ice (as compared to early Laptev melt in the last two years)
- Heat from 2014 "almost" El-nino
- Lack of snow cover on surrounding land
- Change of weather pattern from High pressure (sunshine heating melt ponds) to  frequent low pressure storms (rain, waves, import of warm air ).
- State of the ice is not as solid as ice thickness indicates - ice is made up of ice floes glued together by thin FYI.

[ChrisReynolds]

Using the same method used to calculate grid box effective thickness from PIOMAS gice, and the centre values of each extent range...

The effective extent prediction for the board is 3.93 million kmsq

So far...

[Vergent]

The arctic albedo is at least 20% below 2012. The area is the same. That corresponds to 100 watts in full sunlight. Discounting it to 50 watts for cloud cover, that will still melt 0.8 meters of ice over 60 days. Over 9 M km^2, that corresponds to 8 kkm^3. Eyeballing PIOMAS, that would put us in negative numbers. The poll didn't have negative numbers, so I picked the bottom bucket.

Verg

[Pmt111500]

 Based on North Pacific (and partly S.Atlantic) temperatures Feb 2014.  . I know my two poll answers aren't compatible, yes.

[slow wing]

The arctic albedo is at least 20% below 2012. The area is the same. That corresponds to 100 watts in full sunlight. Discounting it to 50 watts for cloud cover, that will still melt 0.8 meters of ice over 60 days. Over 9 M km^2, that corresponds to 8 kkm^3. Eyeballing PIOMAS, that would put us in negative numbers. The poll didn't have negative numbers, so I picked the bottom bucket.

Verg

Tbf that assumes the abedo will remain 20%+ below 2012 for the next 60 days.

Wouldn't the albedo instead have fallen off soon afterwards in 2012 due to the melting of most of the remaining snow cover?

[Nightvid Cole]

The arctic albedo is at least 20% below 2012. The area is the same. That corresponds to 100 watts in full sunlight. Discounting it to 50 watts for cloud cover, that will still melt 0.8 meters of ice over 60 days. Over 9 M km^2, that corresponds to 8 kkm^3. Eyeballing PIOMAS, that would put us in negative numbers. The poll didn't have negative numbers, so I picked the bottom bucket.

Verg

Tbf that assumes the abedo will remain 20%+ below 2012 for the next 60 days.

Wouldn't the albedo instead have fallen off soon afterwards in 2012 due to the melting of most of the remaining snow cover?

Once snow cover melts, this does not mean albedo drop is over, because the melt ponds left behind by the lost snow continue to expand as they absorb more energy and melt ice at their edges and bottom.

By the time 2012's snow would be gone, this year's melt ponds will have grown significantly bigger, while 2012's will still be newer and smaller. By the time 2012's get bigger, this year will have even bigger melt ponds. I'd venture to say the albedo difference will likely remain for at least a few weeks, though perhaps somewhat subdued.

I do think Verg's estimate is too aggressive, however, since areas that melt earlier in the season will then start using the energy to warm the water locally rather than to melt more ice, which is a negative feedback. Put another way, the thickness of ice at any given pixel on PIOMAS or at any given point in the real world, cannot drop below zero, so as the ice cover shrinks, there is less and less energy being absorbed to actually melt ice and more is absorbed to simply warm the surface water in the ice-free regions.

This probably won't be a significant effect until around August 1, so it is still proper to expect that 2012's record will be smashed.

[Tor Bejnar]

My guess is a tie with 2012: 

• We're ahead right now (influence: Vergent - watch the horses run!)
• There is more ice volume (influence: ChrisReynolds - hold your horses (for now)!)

[Vergent]

Cold snow: Before melt onset the snow albedo is0.85.

Melting snow: Starting with the onset of snow melt,there is a linear decrease to 0.6 in 7 days

Pond formation: For the next 7 days, decrease from 0.6 to 0.32.

Pond drainage: For next 7 days, increase from 0.32 to 0.54.

Pond evolution: Decreases by 0.0083 d1 until albedo reaches 0.2.

Then the ice is assumed to rapidly melt and albedo drops in a single day to 0.07.

Open water: Albedo of 0.07

http://onlinelibrary.wiley.com/doi/10.1029/2012GL051432/pdf



7 + 7 + 7 + (.34  / .0083 = 41 ) + 1 =  63 days

The onset of snow melt was universal by June 1 (JD 152)

June 1(J.D. 152) + 63 days = August 3, (JD 215)

That leaves 42 days of melting to take care of the ice that was tardy because it was thick. But it will be attacked and scattered by wind and wave from the storms that will surely blow.

Vergent

[Nightvid Cole]

I used the same source, MyOcean, with a snow depth threshold of 0.05 (which is 5 cm.) I set the date to June 13th again for each of the 4 frames, and changed the color scale and axes so that the snow would be a single color. I then grabbed the downloadable PNG file and re-processed as a pixel array in MATLAB, in order to extract the pixel count. I also crudely re-plotted the pixel array, using green for land, blue for water, and red for snow with depth >= 5.0 cm (See the above image). The forecast used was today's.

Linear regression of NSIDC September ice extents*, against the snow pixel count, yields an expected September 2015 extent of 3.26 M km^2. I shall proceed to change my vote to this figure.

*as retrieved from their reports issued in October, which are 3.61 M km^2 for 2012, 5.35 M km^2 for 2013, and 5.28 M km^2 for 2014.

[jdallen]

I used the same source, MyOcean, with a snow depth threshold of 0.05 (which is 5 cm.) I set the date to June 13th again for each of the 4 frames, and changed the color scale and axes so that the snow would be a single color. I then grabbed the downloadable PNG file and re-processed as a pixel array in MATLAB, in order to extract the pixel count. I also crudely re-plotted the pixel array, using green for land, blue for water, and red for snow with depth >= 5.0 cm (See the above image). The forecast used was today's.

Linear regression of NSIDC September ice extents*, against the snow pixel count, yields an expected September 2015 extent of 3.26 M km^2. I shall proceed to change my vote to this figure.

*as retrieved from their reports issued in October, which are 3.61 M km^2 for 2012, 5.35 M km^2 for 2013, and 5.28 M km^2 for 2014.

It is gratifying (even if a depressing conclusion...) when someone reaches the same answer you did, by way of different data.  I'm also in the 3.25-3.5 bucket, by way of total annual melt figures.

[Vergent]

September Arctic sea-ice minimum predicted by spring melt-pond fraction.

http://www.micheltsamados.co.uk/september/

Verg

[epiphyte]

I think there's a possibility of something approaching the biblical this year - but don't think it's by any means certain at this point. Going for 3.0-3.25 based simply on 1) the fact that right now there's no significant snow anywhere and none on the horizon, 2) there is melting everywhere, and 3) the assumption that if it's melting everywhere in early June, it's likely to continue to do so because there's almost nowhere the weather can come from which isn't going to cause more melting.

If those are true everything less than 3 years old is gone. I'm guessing that so is about half of the rest - just on the bet that the models are overestimating the condition of the MYI.

[Lord M Vader]

I voted for 4,75-5,0 million km². Given the two last cold seasons together with no persistent negative AO I think we'll be lower than both 2013 and 2014.

//LMV

[ChrisReynolds]

With 12 more votes the effective forum vote has dropped ~0.1 million kmsq to 3.83 million kmsq.

[Steven]

September Arctic sea-ice minimum predicted by spring melt-pond fraction.

http://www.micheltsamados.co.uk/september/

Schroeder et al.'s forecast for September 2015 sea ice extent is 5.1 +/- 0.5 M km^2:

https://twitter.com/micheltsamados/status/608582411773952000

[ChrisReynolds]

Thanks Steven,

That's very close to my prediction which uses April PIOMAS volume, but it's tighter (having the advantage of over a month's extra data. My prediction is 5.15M km^2 +/-0.64M km^2.

[Nightvid Cole]

September Arctic sea-ice minimum predicted by spring melt-pond fraction.

http://www.micheltsamados.co.uk/september/

Schroeder et al.'s forecast for September 2015 sea ice extent is 5.1 +/- 0.5 M km^2:

https://twitter.com/micheltsamados/status/608582411773952000

Their main big region that enters their calculation seems to be a large area north of the New Siberian Islands (look on their map which shows points of significant correlation between pond fraction and the final September extent - a huge chunk shows up there...)

I kind of had been suspecting since late May that their projection would be rather high, because the early melt this year was in the Alaska sector, not the Russian sector so much, which gets the highest weight.

I suspect there is something a bit unphysical about this though as their map shows very irregular distribution of points of significant correlation - with no rhyme or reason whatsoever - (especially Baffin Bay - WTF?)

I suspect over-fitting here.

EDIT: I looked at the map again, and it seems that the single high-melt year, 2012, is heavily influencing the distribution of those grid points. If this is so, then in effect, the study assumes that any year with a different pattern of spring surface melt from 2012 is going to bias the estimate high. Or in other words, the only way to get a very low projected extent is to have an exact repeat of 2012. This does not seem right, simply put.

[Neven]

Dr. Schröder sent me a couple of maps yesterday for 2012, 2014 and this year that I'll use for my short blog post on Melt pond May at the end of this week. He also wrote: "Be aware that we do not know how accurate the locations of our pond maps are and that the relative and absolute numbers for May pond fraction are very small."

[ChrisReynolds]

Nightvid,

It is worth pointing out that in 2014 Schroeder et al predicted 5.40 +/-0.5M km^2, that was a success as actual September extent was 5.28M km^2. In 2013 they forecast 5.55 +/-0.44M km^2, closer than any other entry that year to the actual (5.35M km^2). In 2014 Schroeder were the third closest to the actual.

Their hindcasts are impressive.
http://www.micheltsamados.co.uk/wp-content/uploads/2014/04/HindcastForecast.jpg
Note panel C.

Overfitting or not, the hindcasts since 1979 are very close, and the forecasts of 2013 and 2014 were very close. I think it is unwise to dismiss the approach given its proven success.

[Paddy]

Adjusted my vote upwards to 4.75 -> 5 million square km.  Partly because predictions of really low snow cover around the Arctic don't seem to be matching events so far.

[sedziobs]

Paddy's link seems to indicate 100cm snow depth above 75N, unless I am reading the charts wrong.  That is definitely at odds with the forecasts being posted here, but also at odds with buoy observations.

I have gone with 4.5-4.75.  The character of the ice seems most similar to 2010 in my view, with the El Nino providing another analog between the two years.  I think there is far too much uncertainty with snow depth right now for it to be a useful indicator of extent minimum.  Though for the same reason, I am glad that so many have an eye on it to test the hypothesis and assess the validity of the various snow depth data sources.

[Lord M Vader]

Interestingly, 90 persons have voted so far!! Only 10 more to the magic goal 100 set by Neven!!

[anotheramethyst]

cool!! now i'm glad i voted, i almost didnt because my vote is an uneducated guess haha

[OSweetMrMath]

I've been intending to do a more detailed analysis of my statistical model, but I just won't have time for the next several weeks. So I have to settle for just computing a prediction. My predictions for both April and May were off by less than 0.1 million sq km, so my model continues to be more or less reliable. My predictions for June, July, August, and September are 11.0, 8.1, 5.7 and 4.8 million sq km. I want to recompute my confidence intervals, so I'm not posting intervals for now.

[Nightvid Cole]

Nightvid,

It is worth pointing out that in 2014 Schroeder et al predicted 5.40 +/-0.5M km^2, that was a success as actual September extent was 5.28M km^2. In 2013 they forecast 5.55 +/-0.44M km^2, closer than any other entry that year to the actual (5.35M km^2). In 2014 Schroeder were the third closest to the actual.

Their hindcasts are impressive.
http://www.micheltsamados.co.uk/wp-content/uploads/2014/04/HindcastForecast.jpg
Note panel C.

Overfitting or not, the hindcasts since 1979 are very close, and the forecasts of 2013 and 2014 were very close. I think it is unwise to dismiss the approach given its proven success.

Their f(pond31) forecast residuals are actually very poorly aligned with the data (see the red curve in Figure "b").

Since it is still mid-June, they cannot be basing their forecast on either f(pond56) or on f(pond86). Thus we are concerned with the accuracy of a forecast based on f(pond31), which has a poor track record.

Also, we must consider the forecast accuracy, not the hindcast accuracy, since the projection isn't based on a fit to any data that includes September 2015!

This is why I don't think it's very compelling.

Now don't get me wrong, their method is absolutely a good one, but only once the data is in through June 25th, in order to establish f(pond56).

[ChrisReynolds]

Poorly aligned? You're looking at anomaly difference which will increase differences, instead look at panel c.

I'd love to have such a method, I've been playing around with prediction methods for over five years and I can't do better, even with May data.

Once again:
2013 - SIPN closest to actual,
2014 - SIPN third closest to actual,
Skillfull hindcasts.

It's like criticising a horse for the condition of its teeth when it won the Grand National over the last two years.

Vergent is right, considering your method of ice pack snow cover prediction, the TOPAZ data for 2012/13/14 and the snow cover you use for 2015 implies a virtually ice free Arctic for 2015.

Steven was right to point out what the Schroeder result was in light of Vergent referencing Schroeder's web page. Your method relies on melt ponds, so does Schroeder's, yet the considerations based on three years of data here are wholly at odds with the prediction by Schroeder.

Note that Arctic Ocean compactness remains average, as it was in May.

[Nightvid Cole]

Poorly aligned? You're looking at anomaly difference which will increase differences, instead look at panel c.

It's detrended data, not a difference between the anomaly values for two single years.

Panel c is irrelevant, because it is based on pond56, which requires data up through June 25th (The "56" in "pond56" refers to the number of days averaged for pond coverage, and "56" is the number of days from May 1 through June 25 inclusive.)

I'd love to have such a method, I've been playing around with prediction methods for over five years and I can't do better, even with May data.

Once again:
2013 - SIPN closest to actual,
2014 - SIPN third closest to actual,
Skillfull hindcasts.

Forecasts, not hindcasts.

Yes, they did well for both 2013 and 2014, however, the forecast based on pond31 for the September extent in 2012 (in Fig. b) was a total bust - off by well over 1 M km^2.

It's like criticising a horse for the condition of its teeth when it won the Grand National over the last two years.

What if the horse passed out in the Grand National in 2012, before reaching the finish line, due to heart problems?

Vergent is right, considering your method of ice pack snow cover prediction, the TOPAZ data for 2012/13/14 and the snow cover you use for 2015 implies a virtually ice free Arctic for 2015.

Steven was right to point out what the Schroeder result was in light of Vergent referencing Schroeder's web page. Your method relies on melt ponds, so does Schroeder's, yet the considerations based on three years of data here are wholly at odds with the prediction by Schroeder.

Note that Arctic Ocean compactness remains average, as it was in May.

As I said on the 2015 melting season thread, I expect compactness to plummet by June 20-22 (and on your own blog, you said your July SIPN submission will be based on June 23 IIRC, so if this comes to pass, you'll be be joining me!  )

[ChrisReynolds]

Did anyone get 2012 right for the June SIPN? excluding those who always go low?

As that plot shows, in 2007 and 2012 compactness was falling by now. Compactness will fall, but I don't think it will be a much greater drop than the long term average, that's why I am tracking anomalies.

CTarea by around 23 June should give a reliable refining indicator.

I'm going to leave you to it now, the situation will be clear within about 2 months, and the final answer will be in 3 months from now. With a clear answer so readily available arguing seems a bit pointless.

[Nightvid Cole]

Sometimes I just wish for more near-real time data - snow, melt ponds, etc. and also for existing data to go further back in time. Alas, this isn't available.

Does Schroeder et al. typically give a refined estimate once melt pond info through June 25 is available? Though they didn't make a prediction in SIPN, the forecast based on pond56 (May 1 - June 25) would indeed have been better than the vast majority of the others in 2012...

[ChrisReynolds]

I quite agree, we need more data. For me, it's Bering Strait and Atlantic inflow temperature. By the time a lot of these questions are answered it will be over for summer sea ice. 

Ocean influx of heat is cited by Bitz as having a role in modelled rapid ice loss events.

As for melt ponding. compactness is a poor substitute for proper daily index of melt ponds, but I think it is still useful.

I suppose Schroeder et al will update for July, but I was too busy with work last year to properly follow it. We should have all the SIPN predictions next week I'm keen to see them. But the overall spread means it has little predictive value as a set.

[slow wing]

Have to predict now before poll closes but large uncertainty.

Imo current extent is more or less irrelevant as it is dictated largely by seas on the periphery, e.g. Hudson Bay, whereas the CAB determines the melt season minimum.

Frustrating to not know the volume very well, with different determinations - PIOMAS etc - disagreeing.

Each year it is a race as to whether any particular area melts out before the end of the melt season. The CAB appears to be off to a good start. The following factors support a big melt out this year:
1) the snow cover has already disappeared over most of the ice in the CAB - appears to be earlier than in at least most of the recent years;
2) Both Alaska & Siberia have been relatively hot recently & the winds & rivers will continue to transport this heat & fresh water to the CAB;
3) winds are currently transporting ice out the Fram Strait.

Given all of the above, and unaware of any compensating negative indications, I think it will be a big melt year.

So I went with same as 2012: 3.66 million sq km, which is the 3.50-3.75 bin.
With one sigma uncertainties, I would guesstimate about a 2/3 probability of end up in the range 2.7-4.7 million sq km.