The 2015 melting season

[Jim Pettit]

An early date for SIE maximum doesn't guarantee that the SIE minimum will be low! The SIE minimum in 1996 was the fourth (correct me if I'm wrong!) highest minimums seen in the period of 1979-2015. 2010 on the other hand....

Best, LMV

That is absolutely correct, and can't be said enough. And by the same token, a late maximum is no harbinger of a high minimum. In 2012, SIA peaked on march 30, the latest on record--yet that year saw, quite memorably, the lowest area minimum, and by a huge margin. As the stock brokerage disclaimers say, past results are no guarantee of future performance...

[Jim Hunt]

Now we need to cross our fingers and hope that O-Buoy 10 wakes up as the sun gets higher in the sky!

Lo and behold! Our old friend O-Buoy 10 has powered up and (mostly) defrosted. Co-located ice mass balance buoy 2013F is still functioning too, but ITP 70 stopped acquiring temperature/salinity profiles many moons ago:

[plinius]

And the criticism of Cryosat is that it's "measurements" are not accurate enough to be assimilated into the models! See the 2015 paper above, which doesn't include Cryosat 2 because "there are currently few publicly available ice thickness data that are not preliminary products".

Mostly agreed on that. Actually surprised that they haven't gone over to publishing all data real time and fully free.

[viddaloo]

Yesterday's 2525 ppb is the 2nd highest CH4 level since tracking start on Nov 1st 2014.

[viddaloo]

Now even Guardian calls it: Arctic sea ice extent hits record low for winter maximum

[jplotinus]

By CT area we are currently over 250k below average in Bering sea.  Near max in 2012 we were as high as 400k above average in Bering sea.  Currently weather looks just about perfect for expansion of ice in this area.  The maximum potential expansion looks to be a good 650k in this sea, its just a question of how close we can get to this roughly calculated maximum, and whether we have enough time.

The other big player this time of year, Okhotsk is about 500k below 2012, but the next week looks mild and stormy in this region, so I doubt there will be much if any ice growth there.

At the same time GFS is showing a warm blast at the end of its 7 days run which is strong enough to start noticeable melt (at least on the surface) in the Hudson bay which would strike me as unusual if it happens, and highly likely to freeze back over once the warm burst is over.

I wouldn't say conditions are ideal over the Bering sea. While there are currently cold temperatures there, they're are only cold enough to form sea ice over a relatively small area, and without strong northerly winds (as happened at the end of 2014) there will be nothing to spread out the ice.

2014
March 6th


10th


13th



This year, the forecast shows stable winds and no northerlies until after 5 days but they are northerlies coming more off Alaska and so increase the temperature, so nothing like 2014.
I think we will see some growth in the Bering sea, maybe 100k or 150k, but not enough to take us away from the lowest area coverage on record for the region.

Basis the above and similar weather forecasts covering the middle 1/3 of March, there were indications it might be cold enough in several Arctic regions to surpass the February "mad-max" by, say, now (Equinox). That has not happened. What may have happened, however, is that cold temperatures have remained fairly constant in some parts--much of the Canadian Archipelago, for instance (-20s at Iqaluit since forever ago), yet not much ice growth.

Are there any hints or clues as to what brought on the "mad max"?

[Siffy]

So, I'm kind of confused about something on this image, in some of the graphs the standard deviation lines for the positive half of the graph go beyond the maximum level of ice possible? Is there an explanation for why this happens?

[Pmt111500]

So, I'm kind of confused about something on this image, in some of the graphs the standard deviation lines for the positive half of the graph go beyond the maximum level of ice possible? Is there an explanation for why this happens?

Statistical (numerical) mirroring of the true deviation under the max line, they could be eliminated from the graph, but probably it's just easier to use the same code for areas having a fixed border and those having not.

[ktonine]

Siffy writes: "<i>So, I'm kind of confused about something on this image, in some of the graphs the standard deviation lines for the positive half of the graph go beyond the maximum level of ice possible? Is there an explanation for why this happens??</i>

I haven't dug into the details of what's being done here, but I see this fairly often in my line of work.  It's usually the result of compiling uncertainties using RSS (root-sum-square) and just applying them symmetrically without checking to see if either the upper or lower bounds defy common sense.

But even when they do defy common sense, if you're stating that you are using RSS as your method of calculation, then you probably should just use those numbers (as done here) even if they do seem to defy common sense.  Otherwise you're violating the method described and perhaps giving a false impression of the actual uncertainties.

[cesium62]

Basis the above and similar weather forecasts covering the middle 1/3 of March, there were indications it might be cold enough in several Arctic regions to surpass the February "mad-max" by, say, now (Equinox). That has not happened. What may have happened, however, is that cold temperatures have remained fairly constant in some parts--much of the Canadian Archipelago, for instance (-20s at Iqaluit since forever ago), yet not much ice growth.

Are there any hints or clues as to what brought on the "mad max"?

The discussion in this thread has suggested that a string of storms running up through the north atlantic was instrumental in clearing ice out of the Berents.  Meanwhile, in the pacific, we're in that el nino phase or almost el nino phase that is letting warmer waters into the Okhotsk and Bering seas.

[jdallen]

Basis the above and similar weather forecasts covering the middle 1/3 of March, there were indications it might be cold enough in several Arctic regions to surpass the February "mad-max" by, say, now (Equinox). That has not happened. What may have happened, however, is that cold temperatures have remained fairly constant in some parts--much of the Canadian Archipelago, for instance (-20s at Iqaluit since forever ago), yet not much ice growth.

Are there any hints or clues as to what brought on the "mad max"?

The discussion in this thread has suggested that a string of storms running up through the north atlantic was instrumental in clearing ice out of the Berents.  Meanwhile, in the pacific, we're in that el nino phase or almost el nino phase that is letting warmer waters into the Okhotsk and Bering seas.

Reasonable summary.   Most of  January, February and through much March there was a strong dipole in the circulation that pretty much exported Siberian air across the Arctic into North America.  The "Cyclone Canon" along the Eastern seaboard was a side effect of that, as the cold flow hit Gulf Stream water that is running 5-8 degrees warmer than typical.

The cold being shoved out of the arctic was replaced from inflows over the relatively warm waters of the Bering and Norwegian Sea/Barents. That heat flow wove back and forth a bit, but was also expressed in the very warm winter Europe and the Pacific NW of North America had. (We've had fruit trees blooming over six weeks early here in Seattle...)

We had a strongly positive AO as well, so a feature of the season - FRAM export - was quite a bit more active than we have seen in quite a while.

It all added up.

Right now, through the end of the week, weather looks relatively neutral, but we are already well ahead of most previous melt seasons.  Right now I wonder if we will get conditions like 2012, or 2013.  The former would give us a new and very scary low minimum. The latter would likely see us close but probably above 2011. 

[Wipneus]

(...)

So, I'm kind of confused about something on this image, in some of the graphs the standard deviation lines for the positive half of the graph go beyond the maximum level of ice possible? Is there an explanation for why this happens?

Siffy, your question has been mostly answered by others. The grey bands show the average cover with error bands of 1 and 2 standard deviations.
That is ignoring a geographical maximum cover in some region, but also the very skewed behavior of ice cover deviations: negative swings are much larger than positive ones even with no physical maximum's.
Note that just cutting the grey's off is not a sound solution. The possibilities that have been cutoff will have to appear somewhere else, that is in other regions. Before you know it you get into modelling so complicated that it cannot possibly be useful anymore. 

For the sake of simplicity and because such issues are commonly ignored in the field I did choose not to do anything about it. I am open to suggestions though.
 

[seaicesailor]

What about using 16th - 84th and 2.5th - 97.5th percentiles (equivalent to +/- 1 and +/-  2 standard dev. respectively in normal distribution). Not to say how to do ur job, just saying.
Though at the end will provide very similar info (but within the limits)

[viddaloo]

Ouch! With a humiliating 82% smaller melt than 2011 in week 12, I guess all we can hope for is a merciless revenge in week 13: Then 2011 actually had a gain of 41847 km².

[cesium62]

I am open to suggestions though.

In hopes of a stimulating off-topic discussion of math, I created a new thread in the forums.

[Peter Ellis]

Don't think it needs a new thread.  I concur that a sensible way of graphing data with very asymmetric error distribution would be to use box-and-whisker plots to show median, quartiles and total range.

[viddaloo]

[Espen]

Tomorrow (today) will be a crucial day

[viddaloo]

Tomorrow (today) will be a crucial day

Yup, sort of. Just keep in mind that those 23 days of lowest extent for 2006 can be punctured and split at any point during the next 23 days, by 2015 being lower. That means 2015 may very well end up having the longest unbroken series of days in the first 6 months with the lowest extent on record.

[jdallen]

From a thread just (Justifiably) closed as redundant.

Cracking in the Beaufort.

Link to CAS GOES 11nm IR

http://weather.gc.ca/data/satellite/hrpt_dfo_ir_100.jpg

Ice is pulling away from the coast.

(edit - added jpg which won't change over time)

[jdallen]

And here's NightVid's image, which started me off.

[jdallen]

A better shot of the Beaufort from Yesterday showing the extent of the cracking.

[viddaloo]

Have to say the 'first week of melt' was a huge disappointment for anyone who's rooting for 2015 to melt more ice than 2011. A disappointment to such a degree that even I am confused about my own plot and what it is telling me:



«1% of 2015 max extent gone.» --> OK, fine. I get it. It was 2% before this week.
«2011 so far lost 289745 km² more than 2015.» --> Well, that's certainly a lot, and not at all what I hoped for.
«Total Week 12 melt: 138% less than 2011 = -80328 km².» --> Saying we lost 138% less than 2011 is really rubbing it in. 138? How's that even possible? Well, 2015 had a net *gain* of 80328 km² during week 12, so having a *melt* that is 138% smaller probably means it was absolutely no melt at all (100% smaller) + 38% in the wrong direction (gain). The 80328 km² gain figure is probably about 38% of the 209417 km² net melt of 2011 for week 12.

Let's hope we can do it better this week! As we recall, 2011 this week had a net *gain* of 41847 km², or +5978 km²/day.

[Peter Ellis]

Title:  "Daily changes in ice extent for week 12"

The graph appears to show the daily deltas in ice extent for 2015 and 2011.  I'm not clear on why those were chosen for comparison.

[Lord M Vader]

An interesting thing is what will happen to the older ice if the current weather conditions continues.. With deep low pressure areas in Barents Sea and a decent high pressure dome situated at Greenland there should be a really good transport of ice through Fram Strait..

In the next couple of days this pattern will be replaced by southerlies which will act to compact the ice. Looking at forecasts in 6-7 days and until 10 days ahead is always risky. But however, both GFS and ECMWF are supporting this idea even if they are not having it starting simultaneously... In addition, GFS has this weather pattern to continue for another week but I don't expect that to happen in the first case.

Another thing worth to point out is that ECMWF foresees a High Pressure dome to evolve over the Central Arctic. The temps in this part of the Arctic will be significantly lower than zero so there is no hope for us to see some melt ponds developing yet.

[Gray-Wolf]

Watching the images of ice melting as it is driven south into Bering make me wonder just what scale of impact those warm waters down the US coast will have on Beaufort as it pushes into the basin through Bering straights? I think 2012 saw record levels of water flowing into the basin from the Pacific and the triple R looks to be pushing quite a large jolt of pre warmed waters to the north. With Bering near ice free the water will not be much modified as it pushes through the straights so will we see early melt of the coasts/straights as warm water eats into the ice even before the sun is having impact?.

I'm also of the opinion that a good proportion of the water entering via Bering exits via Baffin so will we see rapid melt out of the NW Passage this year ( by early Aug?) as the warm waters flush through on their way to Baffin?

[LRC1962]

Another possible major impact is the fact that the AMOC is slowing considerably and a lot colder before reaching the Arctic. How will that impact the under melting of the ice? How will that impact the weather systems traveling over it? Could that mean less cloud coming in from the Atlantic side? Where will the warmer Atlantic salt water end up going?

[DavidR]

Here's something interesting that I would have thought would have delivered the opposite result.

I've plotted the loss in ice extent from the NSIDC 5 day average against the number of days that the 5 day average stayed within 200K km^2 of the maximum calculated. Measuring from the first day it got to within 200 K of the max to the last day it was within 200 K. 2011 and 2013 both had double peaks with days in between falling outside the range.  2015 has also had a double peak.

I thought that a short sharp peak would lead to a larger ice loss over the melt season however it turned out to be highly correlated the other way. 2015 is already the second longest period within 200 K and has every every prospect  of becoming the longest period.

Year           Days     Fall(M km^2)
2006     10              8.908
2009    30         10.016         << This figure previously was wrong at 11 days
2014     12              9.936
2013     23            10.113
2007     26            10.610
2010      32            10.688
2012     33            11.908
2008     34            10.711
2015     40   
2011     45            10.322

This suggests to me there is strong  possibility of a significant melt this year. The trend suggests about 11 M km^2 which would put us just above 2012

[Nightvid Cole]

Here's something interesting that I would have thought would have delivered the opposite result.

I've plotted the loss in ice extent from the NSIDC 5 day average against the number of days that the 5 day average stayed within 200K km^2 of the maximum calculated. Measuring from the first day it got to within 200 K of the max to the last day it was within 200 K. 2011 and 2013 both had double peaks with days in between falling outside the range.  2015 has also had a double peak.

I thought that a short sharp peak would lead to a larger ice loss over the melt season however it turned out to be highly correlated the other way. 2015 is already the second longest period within 200 K and has every every prospect  of becoming the longest period.

Year           Days     Fall(M km^2)
2006     10              8.908
2009     11            10.016
2014     12              9.936
2013     23            10.113
2007     26            10.610
2010      32            10.688
2012     33            11.908
2008     34            10.711
2015     40   
2011     45            10.322

This suggests to me there is strong  possibility of a significant melt this year. The trend suggests about 11 M km^2 which would put us just above 2012

Be careful about overfitting. The more arbitrary parameters you have the more likely you are to erroneously see a pattern in what is in fact random noise. 200K - the size of the drop - is one such parameter. Why 200K, not 100K or 300K, or 1 M? Why extent, and not area or volume? Why drop from max to min, and not absolute/detrended minimum extent?

This is several arbitrary things you have chosen here, so there are probably at least 50 independent parameter combinations. What is the P-value for the significance you think you detected? I would not read ANYTHING into this supposed correlation unless it's less than .05/N ~ .001 where N is the number of independent hypotheses. It's a crude approximation of what is known in statistics as a Bonferroni/Holm method correction.

With all the crazy and ad-hoc fitting techniques I see in this thread, no doubt multiple hypotheses have been tested like this!!!

I require P < .001 before I will take this as anything other than simple overfitting of random noise.

[DavidR]

Be careful about overfitting. The more arbitrary parameters you have the more likely you are to erroneously see a pattern in what is in fact random noise. 200K - the size of the drop - is one such parameter. Why 200K, not 100K or 300K, or 1 M? Why extent, and not area or volume? Why drop from max to min, and not absolute/detrended minimum extent?

With all the crazy and ad-hoc fitting techniques I see in this thread, no doubt multiple hypotheses have been tested like this!!!

I require P < .001 before I will take this as anything other than simple overfitting of random noise.

I've checked the data and as noted above found 1 significant error. However it doesn't significantly  change the trend. By the time you  get  out to the 200-300k range the increases and falls in the 5 day averages are becoming consistent.
I  also tried NSIDC extent with 300 K it points to almost the same figure, a drop of about 11.2 M
CT Area with 200K range shows a similar trend with a prediction of a 10.6 M fall.

All three figures suggest a minimum about 1/3rd of the way between 2012 and 2007. I have also posted on the IJIS thread to  see if they  come up with a similar correlation.

[Nightvid Cole]

Be careful about overfitting. The more arbitrary parameters you have the more likely you are to erroneously see a pattern in what is in fact random noise. 200K - the size of the drop - is one such parameter. Why 200K, not 100K or 300K, or 1 M? Why extent, and not area or volume? Why drop from max to min, and not absolute/detrended minimum extent?

With all the crazy and ad-hoc fitting techniques I see in this thread, no doubt multiple hypotheses have been tested like this!!!

I require P < .001 before I will take this as anything other than simple overfitting of random noise.

I've checked the data and as noted above found 1 significant error. However it doesn't significantly  change the trend. By the time you  get  out to the 200-300k range the increases and falls in the 5 day averages are becoming consistent.
I  also tried NSIDC extent with 300 K it points to almost the same figure, a drop of about 11.2 M
CT Area with 200K range shows a similar trend with a prediction of a 10.6 M fall.

All three figures suggest a minimum about 1/3rd of the way between 2012 and 2007. I have also posted on the IJIS thread to  see if they  come up with a similar correlation.

It's not the drop you get (10.6 M km^2 or 11.2) that matters, it is the P-value for the null hypothesis. You have to calculate this value to argue that your correlation is meaningful and not just random noise. You also have to figure out what the error bars on the drop would be.

[Ned W]

The p-value is not high, and the uncertainty in predictions is correspondingly wide. 

Which is ... unsurprising.  We already know that what happens between March and September is important.  If that weren't the case, Neven and all the rest of us could close up shop next month, as the results would be a foregone conclusion.

[cesium62]

I fully agree with Nightvid, but, just for grins, I correlated a normalized version of average winter sea ice extent with average summer sea ice extent.  The correlation is reasonably strong, but negative.

The raw data is NSIDC extent data from 1979 through 2014.  For each day (leap year's ignored), I computed the standard deviation for that day.  For each day, I took (extent minus mean) divided by standard deviation.  This more-or-less removes the seasonal wave from the data.  I then computed the average deviation for each year, and subtracted that from each day's deviation.  This more-or-less adjusts for the fact that it's getting warmer from year to year.

I then plotted, for each year, the average normalized winter value (1/1 through 3/31), with the average normalized late summer value (8/1 through 10/31).  2012 is the bottom-right dot.  1996 and 2006 are the upper-left dots.

This suggests that when winter sea ice is extent is large, summer extent will be small.  And vice versa.  2015 currently looks a lot like 2006, and both those years look a lot like 1996, but slid down on the graph.

[Peter Ellis]

I suspect that's an inevitable consequence of the fact that the seasonal cycle is getting wider, and that the distribution during the summer min is asymmetric (several recent low outliers, no equivalent high outliers).

Your protocol doesn't remove the seasonal cycle, it removes the _average_ seasonal cycle. The cycle has been getting wider over time because summer ice is collapsing faster than winter ice, which means that:
(a) Recent years will still have a residual cycle there (i.e. high in winter, low in summer),
(b) Early years in the data set will be over-corrected for the seasonal cycle (and thus end up low in winter, high in summer)

This is all you need to generate the observed correlation.  Dividing through by the standard deviations probably helps to some degree, but won't remove the effect.

You see this in every single long-term plot of daily anomalies - which is what this is, albeit well-disguised.

[DavidR]

I realise that statistically my observation is not significant.  There are far too few data points to make any long term claims.  Alternatively there is too much change in the long term conditions to  rely on the long term trends without  much more detailed analysis
I  started by  addressing an issue raised on Dosbat http://dosbat.blogspot.com.au/2015/03/sea-ice-maximum-2015.html regarding the fact that most record years coincided with a late maximum. 
My thought was that  this could be related to the fact that late maximums could be caused by the rapid development of a lot of thin ice that would then melt away quickly.  What I  found was the exact reverse. Most of the late maximums got close to the maximum early and then drifted up to the maximum over a long period. Generally, the longer they stayed near the maximum the greater the decline.

[DavidR]

I fully agree with Nightvid, but, just for grins, I correlated a normalized version of average winter sea ice extent with average summer sea ice extent.  The correlation is reasonably strong, but negative.

This suggests that when winter sea ice is extent is large, summer extent will be small.  And vice versa.  2015 currently looks a lot like 2006, and both those years look a lot like 1996, but slid down on the graph.

Cesium62,
This would appear to  agree with  my observations , if the average extent is high that implies a longer period close to the maximum, which is what I  am observing as a predictor  of a larger drop in extent.

[viddaloo]

DavidR, if we assume that extreme minima are caused by anomalous ocean heat content, then it may be logical that the same anomalously high ocean heat would prevent proper ice regrowth in the Feb–Mar period, and thus cause a long drawn–out plateau around the time of the yearly max.

I'm sure these shocks and other changes are happening so fast that waiting for 'enough' data to arrive for a correlation to be 'significant' using strict standards for such is comparable to waiting for the last person on Earth to agree that humans are causing warming before a change of policy.

That's not to say that 'anything goes', only that waiting for perfect statistical significance may not be a luxury we can afford in this century.

[Michael Hauber]

Watching the images of ice melting as it is driven south into Bering make me wonder just what scale of impact those warm waters down the US coast will have on Beaufort as it pushes into the basin through Bering straights? I think 2012 saw record levels of water flowing into the basin from the Pacific and the triple R looks to be pushing quite a large jolt of pre warmed waters to the north. With Bering near ice free the water will not be much modified as it pushes through the straights so will we see early melt of the coasts/straights as warm water eats into the ice even before the sun is having impact?.

I'm also of the opinion that a good proportion of the water entering via Bering exits via Baffin so will we see rapid melt out of the NW Passage this year ( by early Aug?) as the warm waters flush through on their way to Baffin?

The warming of waters off the coast of western US is not the result of warm water actually moving from the equator, but the result of the Pacific gyre being slowed down due to warm ENSO activity.  This results in less cold water moving from the North Pacfic and less upwelling of cold water along the coast as the NE Pacific high also weakens.  And so the water becomes warmer along the coast.

The Bering and Arctic are separated from the North Pacific gyre by the sub polar gyre.  As this gyre is in part a reaction to the North Pacfic Gyre I would expect it could slow down, meaning less warm transported into the Bering from further south along the Alaskan coast.

I'm not sure what the potential implications are for upwelling within this region are.  But suspect that upwelling off Alaska may be linked with upwelling off California.  If upwelling along the entire North America coast is surpressed, this will warm water along most of the coast due to less cold water from below.  However when we get into the Arctic the opposite will occur as the water underneath the sea ice is actually warmer.  With less upwelling and less current flowing out of the Bering Strait the ice expands less, but becomes thicker and is then harder to melt in summer.

I've long been struck by the fact that winter sea ice in the Bering region has been generally high since 2008 with 2012 being the highest in the 30 year Cryosphere today record.  And winter ice in this region has dropped off since 2012 as summer ice has increased.  It could be a coincidence.  In particular 2007 did not have particularly high winter ice in Bering, and 2006 was moderately high.  But I have suspected for a while that there may be a correlation between high winter ice in this region during winter and low summer ice in the Arctic.

[Michael Hauber]

I fully agree with Nightvid, but, just for grins, I correlated a normalized version of average winter sea ice extent with average summer sea ice extent.  The correlation is reasonably strong, but negative.

The raw data is NSIDC extent data from 1979 through 2014.  For each day (leap year's ignored), I computed the standard deviation for that day.  For each day, I took (extent minus mean) divided by standard deviation.  This more-or-less removes the seasonal wave from the data.  I then computed the average deviation for each year, and subtracted that from each day's deviation.  This more-or-less adjusts for the fact that it's getting warmer from year to year.

I then plotted, for each year, the average normalized winter value (1/1 through 3/31), with the average normalized late summer value (8/1 through 10/31).  2012 is the bottom-right dot.  1996 and 2006 are the upper-left dots.

This suggests that when winter sea ice is extent is large, summer extent will be small.  And vice versa.  2015 currently looks a lot like 2006, and both those years look a lot like 1996, but slid down on the graph.

If I understand correctly this is comparing the average during summer during the average for that entire year, and similar for winter.  This would tend to imply a negative correlation with winter simply because if summer is higher than average, then the rest of the year must as a whole be lower than average.

I do see three factors that may be relevant to summer vs winter:

1) Co2 is warming the world and reducing both as a long term trend.  De-trending should remove this effect.
2) ENSO causes significant year to year variations in global temperature.  However ENSO does not seem to be correlated in any meaningful way to Arctic sea ice.
3) Remaining effects would seem likely to be redistribution of heat from one area of the world to another.  It does not seem unreasonable that if the north Pacific and North Atlantic are warmer than normal  - leading to low winter ice that the Arctic would be colder than normal - leading to high summer ice.  Of course its also possible that the warmer than normal region could swap from the North Atlantic and/or North Pacific from winter to summer.  Or that the whole Arctic + North Pacfiic + North Atlantic could be warmer at the same time as somewhere else is colder. 

Considering the Arctic Oscillation it looks like cool conditions on the North Atlantic winter ice edge correspond to cool conditions in the Arctic, whereas warm conditions in the North Pacific ice edge correspond to cool conditions in the Arctic.

[cesium62]

I suspect that's an inevitable consequence of the fact that the seasonal cycle is getting wider, and that the distribution during the summer min is asymmetric (several recent low outliers, no equivalent high outliers).

Your protocol doesn't remove the seasonal cycle, it removes the _average_ seasonal cycle. The cycle has been getting wider over time because summer ice is collapsing faster than winter ice, which means that:
(a) Recent years will still have a residual cycle there (i.e. high in winter, low in summer),
(b) Early years in the data set will be over-corrected for the seasonal cycle (and thus end up low in winter, high in summer)

This is all you need to generate the observed correlation.  Dividing through by the standard deviations probably helps to some degree, but won't remove the effect.

You see this in every single long-term plot of daily anomalies - which is what this is, albeit well-disguised.

Dividing each day by the daily standard deviation removes the seasonal cycle.  The wavy mean is converted to a horizontal line.  Variations in width across the seasons are normalized away.

On the other hand, it's quite plausible that the data for any particular day is not normal, and using a model that assumes the data is normal produces meaningless results.  And, as Nightvid points out, without some measure showing the correlation is highly significant, the correlation is meaningless.  And the date ranges chosen for comparison might be silly.  October is a weird month.

I agree that the graph is a toy solely for amusement purposes only; your mileage may vary.  However, the graph does suggest looking at 2006 as a possible model for what could happen in 2015.  I find that model at least as likely as the model that we take 2012 and shift it down so that it's winter curve is near the 2015 winter curve.

[cesium62]

If I understand correctly this is comparing the average during summer during the average for that entire year, and similar for winter.  This would tend to imply a negative correlation with winter simply because if summer is higher than average, then the rest of the year must as a whole be lower than average.

Yes, I believe you have verbalized this flaw of mine well.

[crandles]

An alternative possible explanation - probably overdone parts and lot of other missing parts to it or at least needs a lot of tweaking:

If the ice pack near the edges is thick and healthy, then at some point near where maximum occurs there will be colder weather/winds from north for those edge regions and extent will grow quickly and recede quickly when the weather turns warmer/winds from south.

This leads to a short period near the max extent and the thick healthy ice near edges impedes a rapid start to the season so that albedo feedback can't get going quickly resulting in a small drop in extent during the melting season.

The opposite of this is thin ice near the edges with warm water flowing towards those edge regions. The warm water causes the thin ice to break up. By such small pieces of ice turning to float with largest possible surface area, this occupies more extent even before winds spread it out. As it spreads out concentration well over 15% you get a further increase in extent but the warm waters tend to melt these pieces before getting to the ice pack leaving little heat to attack the thin but solid pack. If concentration falls to under 15% then you get a period of low extent but the warm waters then start attacking the thin pack again which breaks up and spreads increasing the extent again for a late max.

Above might over-describe what has happened to extent this year. However a weak pack and warm water might mean lots of ups and downs so a long time with extent near the maximum. Weak pack and warm inflowing water is likely to persist such that season can get off to rapid start which starts the albedo feedback so that there is a large drop in extent during the melting season.

Of course if there is a correlation, it need not have any mechanism at all and if there is a mechanism, it might be nothing like the above.

Feel free to tear it apart.   

[Michael Hauber]

The opposite of this is thin ice near the edges with warm water flowing towards those edge regions. The warm water causes the thin ice to break up. By such small pieces of ice turning to float with largest possible surface area, this occupies more extent even before winds spread it out. As it spreads out concentration well over 15% you get a further increase in extent but the warm waters tend to melt these pieces before getting to the ice pack leaving little heat to attack the thin but solid pack. If concentration falls to under 15% then you get a period of low extent but the warm waters then start attacking the thin pack again which breaks up and spreads increasing the extent again for a late max.

I would expect that if warm water is flowing towards the pack from the south it will tend to compact the pack and reduce extent and area together.  I also note that recently we had a major storm blow some fairly serious southerlies into the Atlantic ice pack edge, and then the storm moved towards Russia pushing some fairly serious northerlies and cold onto this ice edge.  The southerlies caused a retreat of the ice edge.  Some of us speculated that we could see the extent and area rebound and possibly threaten the early max, but this seems to have not happened.  I suspect that the warm water has prevented the pack from expanding in this region as it has melted faster than it has expanded into the warm waters.  I suspect that the Atlantic ice edge has been dealt a significant 1-2 blow by this pattern.  Combined with lots of apparent transport out of the Laptev season and the Navy Hycom model showing very thin ice in this region I expect that the Atlantic sector will be the one to watch this melting season, and wait with interest to see if we get an ice free north pole.  A lot to happen between now and then though.

In contrast to warm water flowing towards the pack consider what happens if upwelling increases below the pack.  This will tend to push the pack apart and will introduce slightly warmer water from below to either cause bottom melt, or at least slow the freezing rate.  The surface conditions may be cold and stormy as upwelling usually follows low pressure (or offshore winds) so the cold air can freeze over the gaps that appear with thin ice and so the whole pack has a higher extent and area, but is thinner and more vulnerable.

[crandles]

The opposite of this is thin ice near the edges with warm water flowing towards those edge regions. The warm water causes the thin ice to break up. By such small pieces of ice turning to float with largest possible surface area, this occupies more extent even before winds spread it out. As it spreads out concentration well over 15% you get a further increase in extent but the warm waters tend to melt these pieces before getting to the ice pack leaving little heat to attack the thin but solid pack. If concentration falls to under 15% then you get a period of low extent but the warm waters then start attacking the thin pack again which breaks up and spreads increasing the extent again for a late max.

I would expect that if warm water is flowing towards the pack from the south it will tend to compact the pack and reduce extent and area together.  I also note that recently we had a major storm blow some fairly serious southerlies into the Atlantic ice pack edge, and then the storm moved towards Russia pushing some fairly serious northerlies and cold onto this ice edge.  The southerlies caused a retreat of the ice edge.  Some of us speculated that we could see the extent and area rebound and possibly threaten the early max, but this seems to have not happened.  I suspect that the warm water has prevented the pack from expanding in this region as it has melted faster than it has expanded into the warm waters.  I suspect that the Atlantic ice edge has been dealt a significant 1-2 blow by this pattern.  Combined with lots of apparent transport out of the Laptev season and the Navy Hycom model showing very thin ice in this region I expect that the Atlantic sector will be the one to watch this melting season, and wait with interest to see if we get an ice free north pole.  A lot to happen between now and then though.

In contrast to warm water flowing towards the pack consider what happens if upwelling increases below the pack.  This will tend to push the pack apart and will introduce slightly warmer water from below to either cause bottom melt, or at least slow the freezing rate.  The surface conditions may be cold and stormy as upwelling usually follows low pressure (or offshore winds) so the cold air can freeze over the gaps that appear with thin ice and so the whole pack has a higher extent and area, but is thinner and more vulnerable.

Storminess can certainly be part of this story of causing the pack to break into fragments. I don't expect storminess near max to continue through melting season so that doesn't at first appear to be the cause of the correlation being discussed.

However, warm Atlantic water tends to flow in sub-surface. If there is larger than normal volume of this flowing towards the arctic, then surface waters may be flowing away from arctic more than normal helping to scatter fragments rather than your 'compact the pack and reduce extent and area together'. Storminess can now add to this effect by mixing the water to efficiently lose the heat to the ice. The salt doesn't disappear but more heat lost means more cold salty water sinks and this draws in more warm Atlantic water.

Having said this, I am not sure that the story entirely holds - if the pack was more fragmented than usual we would expect area to be at record lows while extent was a little above record low. However this year has been more the other way around with extent at record low much more than area has been.

[DavidR]

Having said this, I am not sure that the story entirely holds - if the pack was more fragmented than usual we would expect area to be at record lows while extent was a little above record low. However this year has been more the other way around with extent at record low much more than area has been.

Crandles.
 it  may be that the area is fragmented rather than dispersed. The ice could be very  fragmented but  not  dispersed.  Area is not 100% ice! A fragmented pack could be highly  concentrated but susceptible to dispersion. 

 I think the changes we are seeing are because the pack is structurally  different from 10  years ago. Therefore the extrapolations from 10 years ago are not relevant to the current scenario. 

There is a difference between discussing  short term variations that  are consistent with the science and claiming that  short term variations that  are inconsistent with the science refute the science

[crandles]

It wouldn't have been scientific not to mention a possible problem with my explanation. That doesn't mean that I think that negates any possible effect of my explanation and as you point out the pack may be fragmented more but not dispersed.

If my explanation or something like it is a possible explanation for the correlation you found, do you think it might be better to look at total duration from first time 5 day average is within 200k of peak 5 day average value until the last time rather than excluding some days during this period than are more than 200k below?

If you change the 200k value to say 150k, 250k, or 100k does the correlation persist?
If it disappears with other values than it is more likely to be a spurious correlation occurring by chance. If it persists with other values it is more likely to be a real effect.

Before testing it, can we make any predictions about whether extent near peak duration will be a better predictor of extent decline, area decline or volume decline?

Does extent staying near the max for a long time work better than using whether area stays near the max for a long time? Given my explanation, I think extent should probably work better than area.

If there is still an effect but weaker using area then I think this would somewhat weaken the argument that it arises by chance. Likewise if it also predicts area and volume declines it looks more like a real effect than a spurious correlation.

[viddaloo]

46528 km² ahead of 2011! 

[viddaloo]

Icelook mar25: Average extent 4^th lowest in 46 days, average volume will be 6^th in 10 days, and p1k (Piomas minus 1000) will be 5^th in 7 weeks. Opinions differ, but I believe even more interesting than the chair & coffee routine this time of year is extent. Since last week we've crossed the 10.29 million line and we're now at 10.286, the lowest since record–low year 2013. Daily and weekly deltas have already surpassed everything in 2013 and 2014, and now even monthly deltas compare themselves to November 2012 as the closest match: The current monthly delta is down 36032 km². Current weekly delta at 10854 km²/week enables us to lose about 40 grand a month, which takes us below 10.24 by the end of April. And before that: Lower than 10.28 by Mar 30. According to the graph, this line is crossed on Mar 29. The forecast is of course more reliable in the short term, but now has us crossing into 4th lowest territory on May 10th:


[chart faq]

PS: DavidR's excellent work on the forum has prompted me to test out his Plateau hypothesis in relation to JAXA sea ice extent. My findings are astonishing, as post–2007 the 5 longest plateaus surrounding the yearly maximum ice extents are: 2007, 2010, 2011, 2012 and 2015. The 3 longest plateaus among these also have the biggest New Year's Eve to New Year's Eve drops in annual average extent, which is what you'd expect if the ocean heat content is exceptionally high and causing both the plateau around the max and an overall higher melt & slower freeze all through the year. The 4th long plateau is that of 2012, which everyone knows had the record lowest yearly minimum for both extent and volume in September. Now, the Top 2 longest plateaus are both 42 days long. One of them belongs to 2011, which had the record for lowest maximum before 2015, and which also lost more annual average extent than any other year, and finished 3rd lowest in both minimum volume and extent. And the last one is the current 2015 plateau, which looks set to get even longer than 42 days, perhaps lasting as long as 50 days? IMO it makes sense to attribute a long plateau — ie a long period within which 5–day averages within 200 k km² below the yearly max occur — to warm ocean water, hampering a proper refreeze and instead causing pulsing back and forth for 1 ½ months.

[DavidR]

Now, the Top 2 longest plateaus are both 42 days long. One of them belongs to 2011, which had the record for lowest maximum before 2015, and which also lost more annual average extent than any other year, and finished 3rd lowest in both minimum volume and extent. And the last one is the current 2015 plateau, which looks set to get even longer than 42 days, perhaps lasting as long as 50 days? IMO it makes sense to attribute a long plateau — ie a long period within which 5–day averages within 200 k km² below the yearly max occur — to warm ocean water, hampering a proper refreeze and instead causing pulsing back and forth for 1 ½ months.

Vid,
based on that calculation can you estimate the likely position of 2015 in relation to other years. ie is it lowest, half way  between lowest  and second lowest, fifth lowest or what. Just  based on the existing trend line.  It would be interesting to know if it matches the other predictions.

[viddaloo]

That's a tough call, and I'm absolutely no expert on sea ice, just looking at the numbers of it.

But if ocean heat related and wider than the 2011 previous record plateau, we could at least lose the same amount of extent as 2011 did from its max to min. That would take us from 13942060 max to 13942060-9858530 = 4.08 million km2 min. 3rd.

Code: [Select]

2007: 14209677 30 days (-269894)
2008: 14774776 27 days (494895)
2009: 14657047 27 days (-30429)
2010: 14688540 33 days (-219301)
2011: 14127729 42 days (-277725)
2012: 14709086 30 days (-86637)
2013: 14523635 26 days (459327)
2014: 14448416 10 days (-94278)
2015: 13942060 42 days (?)