Sea Ice Prediction Network 2014 Submissions

[DavidR]

Crandles,
The thickness measure used does not tell us anything about the thickness loss in ice that  has not melted out.  We can say fairly confidently that the extent loss up to July 22nd requires that all the extent that was less than 1.7m thick according to the April figures has now melted out.  In 2012 the July 22nd figure was only 1.66m.

This doesn't mean that  the ice that  was thicker than 3 m in the original measure has lost 1.7m as well.

The thicker, the ice at the start of the season, the less the thickness will reduce. So ice that was 4m thick might only  lose 50cm over the entire season. Chris's figures show that there was a lot  more of this thicker ice at the start of this season  than in 2012.  As a consequence we could see an extent record low without seeing a volume record low.

We have now moved into the 1.7-1.8m range in the histogram. In this range as well as the next two ranges there is more than 2 1/2 times as much extent as in 2012.  That's 3.356 M km^2 compared to just  1.281 M km^2 in 2012.  In the past three years all of this range has melted out by the end of the season.  I  expect it  to  happen again this year.  That  takes us down to  4.350 M km^2.

The question at this stage is how much the weather contributes to losses greater than this..

[crandles]

Crandles,
The thickness measure used does not tell us anything about the thickness loss in ice that  has not melted out.  We can say fairly confidently that the extent loss up to July 22nd requires that all the extent that was less than 1.7m thick according to the April figures has now melted out.  In 2012 the July 22nd figure was only 1.66m.

This doesn't mean that  the ice that  was thicker than 3 m in the original measure has lost 1.7m as well.

The thicker, the ice at the start of the season, the less the thickness will reduce. So ice that was 4m thick might only  lose 50cm over the entire season. Chris's figures show that there was a lot  more of this thicker ice at the start of this season  than in 2012.  As a consequence we could see an extent record low without seeing a volume record low.

Theoretically I do accept that is possible. Generally ratio of area to average thickness at minimums has stayed reasonably constant, and we generally see more volume minimum records than extent area minimum records but it might happen the other way around. For this year I think in declining probability we could have:
a year with neither records, or
a year with both records, or
a volume record but not an extent record, or
an extent record but not a volume record

The last three seem low probabilities compared to the first but even the least likely is not something I would completely rule out even if I think it is very unlikely. Maybe I would give subjective probabilities something like 78% 15% 4.5% 2.5%.

We have now moved into the 1.7-1.8m range in the histogram. In this range as well as the next two ranges there is more than 2 1/2 times as much extent as in 2012.  That's 3.356 M km^2 compared to just  1.281 M km^2 in 2012.  In the past three years all of this range has melted out by the end of the season.  I  expect it  to  happen again this year.  That  takes us down to  4.350 M km^2.

The question at this stage is how much the weather contributes to losses greater than this..

Thanks for confirming that you do believe the 2.5* faster volume melt that your model predicts. You don't seem to have answered if you would believe it if your model predicted 20*faster volume melt rate. What if I asked if you wanted to bet £100 on a record low extent occurring this year?

[greatdying2]

We have now moved into the 1.7-1.8m range in the histogram. In this range as well as the next two ranges there is more than 2 1/2 times as much extent as in 2012.  That's 3.356 M km^2 compared to just  1.281 M km^2 in 2012.  In the past three years all of this range has melted out by the end of the season.  I  expect it  to  happen again this year.  That  takes us down to  4.350 M km^2.

Thanks for confirming that you do believe the 2.5* faster volume melt that your model predicts. You don't seem to have answered if you would believe it if your model predicted 20*faster volume melt rate. What if I asked if you wanted to bet £100 on a record low extent occurring this year?

How does this mean more volume melt? Let's say you have two blocks of ice each covering the same area.  Let's say one is 10 cm thick over the entire area, while the other is thicker (by any amount). If both melt by 10 cm, they have both lost the same volume, but one block has lost 100% of it's area and extent, while the other has lost 0%. Am I misunderstanding something?

[DavidR]

Crandles,
My assessment at the moment is

1.  an extent record but not a volume record - 85%
2.  a year with neither records - 10%
3.  a year with both records - 5%
4.  a volume record but not an extent record - 0.0%

If I was a betting man, I would be happy to take your money.

As to your question about 20 times the melt. Let  me put it this way, if the April PIOMAS.  histogram showed there was 20 times more ice between 1.0 and 1.1 m thick than in 2012, I would expect all that ice to melt out in the last week of June.  The scenario is unrealistic however because it  would represent more than half the total extent at that time.  If it said the same at 2.1 - 2.2 I  would have less confidence because that melt is heavily impacted by August  weather.

However that  wouldn't have any significant impact on the melt of the thicker ice. This year there was about  2.4 M km^2 of ice above 2.2 m but the average thickness was about 3m.  In 2012 there was 4 M km^2 of ice above 2.2 m but the average was only about 2.6m. At the end of the season, when the storm hit, that ice was a lot more fragile than it is this year.  2012 finally  melted out to 2.37 m so there is a lot  of scope for getting well into that  last 2.4 M km^2 if conditions are conducive to melt. 

[crandles]

How does this mean more volume melt? Let's say you have two blocks of ice each covering the same area.  Let's say one is 10 cm thick over the entire area, while the other is thicker (by any amount). If both melt by 10 cm, they have both lost the same volume, but one block has lost 100% of it's area and extent, while the other has lost 0%. Am I misunderstanding something?

Sorry that is my mistaken wording. I believe the ratio of area to average thickness will remain roughly constant but it is now clear that DavidR doesn't. I should have said

Thanks for confirming that you do believe the 2.5* faster extent reduction that your model predicts.

[crandles]

Crandles,
My assessment at the moment is

1.  an extent record but not a volume record - 85%
2.  a year with neither records - 10%
3.  a year with both records - 5%
4.  a volume record but not an extent record - 0.0%

If I was a betting man, I would be happy to take your money.

That seems pretty strong self belief - 90% chance of extent record when you are the only one: 27 out of 28 search outlooks disagree:

[crandles]

As to your question about 20 times the melt. Let  me put it this way, if the April PIOMAS.  histogram showed there was 20 times more ice between 1.0 and 1.1 m thick than in 2012, I would expect all that ice to melt out in the last week of June.  The scenario is unrealistic however because it  would represent more than half the total extent at that time.  If it said the same at 2.1 - 2.2 I  would have less confidence because that melt is heavily impacted by August  weather.

Fair enough about August/2.1-2.2m.

Regarding 1-1.1m all melting by end of June. I do note that you have said it is unrealistic. Also that you mentioned 1.35m normally by end of June which is now being reduced to 1.1m in these exceptional circumstances. So the volume reduction is not quite 20 times as fast as normal for the present but presumably the volume reduction would be more than 10 times faster.

You do realise that you are taking a very small sample of years when there is not a lot of change in the thickness distributions or total quantity of thin ice and extrapolating your 'no change in date' conclusion way outside the range of your very small sample? It seems obvious to me that this is asking for trouble yet you claim 90% confidence when all the other experts disagree. Sorry but I am not sure whether to conclude this is a sad case of not understanding what you are doing or if there is any alternative to that.

 

[DavidR]

crandles.

I  am completely  aware of the simplicity of the model I  am using.   It  addresses only  two  variables in a highly  complex system.  However it has been a good predictor of melt  up until July 16th  and I  have been presented with no  evidence that  suggests I  am wrong.

The fact that others have alternative views has never been a big factor  in determining my  views.

Most of the models are understating the severity of the loss because they  are not placing enough weight on the most recent  years.  I  am doing the reverse, I  am only  considering the most recent  years and making  a prediction based on the average of those years. If  in fact extent is declining my  method is biased against the trend.

On the other hand 2012 could be considered exceptional, losing far more thickness that  2011 and 2013.

My  prediction was that  the July 1 figure would show between 1.27 and 1.33 m of thickness loss in fact it  was 1.37 m. On July  16th the prediction was for 1.57 - 1.67 of thickness loss, where the actual  was 1.64.   

The actual June loss in 2014 was 400K km^2 more than predicted compared to  2012 but still  300 k km^2 less than the actual loss in 2012. 

The variation in the thickness distribution between 2012 and 2014 was the driving factor in making this an interesting prediction.

In my  response I  didn't suggest that  the 1.0-1.1 would be only  ice to  melt out  in the last week of June just that  it would definitely be gone during that  week.  I  was merely trying to  make the point that  if you  told me the point in the distribution that  was 20  times greater I  could tell you when it  would disappear to within a few days.

So far the actual loss in 2014 is well  ahead of the prediction compared to  2012, even with the relatively  slow melt  so far  in July.

 

[DavidR]

Sorry that is my mistaken wording. I believe the ratio of area to average thickness will remain roughly constant but it is now clear that DavidR doesn't.

Crandles, I  am not sure what  you  are saying here.  If there is a volume  of ice that  is between 0.95 and 1.05 m thick and constitutes 1 km^3 of ice it  will equate to about 1000 km^2 of  extent if the same volume of ice is between 1.95  and 2.05 m thick it  will cover only about 500 km^2 of extent.  The equation is   Extent (km^2)  = Volume ( km^3)  / Thickness (km ) .  If the thickness doubles the extent halves.

What  do you  think I  don't understand?

[crandles]

Sorry that is my mistaken wording. I believe the ratio of area to average thickness will remain roughly constant but it is now clear that DavidR doesn't.

Crandles, I  am not sure what  you  are saying here.  If there is a volume  of ice that  is between 0.95 and 1.05 m thick and constitutes 1 km^3 of ice it  will equate to about 1000 km^2 of  extent if the same volume of ice is between 1.95  and 2.05 m thick it  will cover only about 500 km^2 of extent.  The equation is   Extent (km^2)  = Volume ( km^3)  / Thickness (km ) .  If the thickness doubles the extent halves.

What  do you  think I  don't understand?

It was Great Dying that said he didn't understand. This was because I made a mistake with my wording. I am not saying you don't understand anything there.

Your method and beliefs seem to imply that you expect the remaining ice to have small extent but relatively high volume because of the thickness distribution of ice in April. That might not be unreasonable given that there was a lot of extent of meltable thickness ice in April (but only if we were having a high rate of volume loss year, which we are not).

You don't seem able to perceive that your model predicts unreasonable rates of volume loss in cases where the thickness distribution is significantly different from the sample where you derived the no change in date for each thickness to melt out. Where do you think the extra heat comes from to allow those much faster rates of melt?

[Steven]

Crandles, thanks for mentioning that the July report of the 2014 Sea Ice Outlook is available.

http://www.arcus.org/sipn/sea-ice-outlook/2014/july

Since many comments in this thread are about predictions based on PIOMAS volume, perhaps it's of interest to compare with the Zhang and Lindsay prediction.  Their prediction is 4.8 +/- 0.4 million km² for the NSIDC September 2014 extent.  This is based on the PIOMAS model, using a numerical ensemble consisting of 7 ensemble members, described in more detail here:

http://www.arcus.org/files/search/sea-ice-outlook/pdf/zhang_lindsay_1.pdf

The ensemble consists of seven members each of which uses a unique set of NCEP/NCAR atmospheric forcing fields from recent years, representing recent climate, such that ensemble member 1 uses 2007 NCEP/NCAR forcing, member 2 uses 2008 forcing …, and member 7 uses 2013 forcing. These seven years of the reanalysis atmospheric forcing fields are used to represent the climate variability expected for 2014.

Figure 1. (a) Ensemble median prediction of September 2014 mean sea ice thickness and edge location [...] The white line represents the satellite-observed mean September 2013 ice edge defined as the line of 0.15 ice concentration, while the black line is the model predicted September 2014 ice edge.

[Adam Ash]

crandles.

I  am completely  aware of the simplicity of the model I  am using.   It  addresses only  two  variables in a highly  complex system.  However it has been a good predictor of melt  up until July 16th  and I  have been presented with no  evidence that  suggests I  am wrong.

The fact that others have alternative views has never been a big factor  in determining my  views.

Most of...

With respects DavidR, that your model has produced the 'right' answers (luck?) for the wrong reasons (vast simplification of a complex and now unstable system) does not actually validate your model.

Good luck for the next run tho - its all a bit of a lottery as even the best models are not good at handling the current rates of change and the proliferation of known and unknown unknowns Mother Earth is revealing to us.

[DavidR]

Your method and beliefs seem to imply that you expect the remaining ice to have small extent but relatively high volume because of the thickness distribution of ice in April. That might not be unreasonable given that there was a lot of extent of meltable thickness ice in April (but only if we were having a high rate of volume loss year, which we are not).

You don't seem able to perceive that your model predicts unreasonable rates of volume loss in cases where the thickness distribution is significantly different from the sample where you derived the no change in date for each thickness to melt out. Where do you think the extra heat comes from to allow those much faster rates of melt?

Crandles.
Sea ice is essentially two  dimensional, the forces that act  to melt  it  operate on the extent, not the volume. The amount of solar insolation and bottom melt are essentially  the same regardless of the thickness of the ice.  So if there is 2.5 * as much area to  melt there is also 2.5* as much power to  melt.

My  model is based on the smallest  possible volume loss to acheive each  level of extent loss.  It  doesn't start to  address the volume loss in areas that don't melt  out.   

I  am not suggesting that  the 1.1 M km^2 of extent that  was 1.7-1.8 m thick in April  is suddenly  going  to lose all that  thickness.  I  believe is has already  lost all but the last 0.2 m of thickness and that is what is melting  out  now.  After that  the same thing applies to the next  10 cm  and the next 10 cm.

My model predicts unreasonable rates of extent loss, the volume loss expected is directly proportional to the extent  loss. I  am not aware of any research that  claims that there is an upper limit to the amount  of volume loss in a given period.

[DavidR]

With respects DavidR, that your model has produced the 'right' answers (luck?) for the wrong reasons (vast simplification of a complex and now unstable system) does not actually validate your model.

Adam,
If you think the model has achieved its results, using invalid factors you  should explain what you think are the invalid factors.  Personally  I  think the thickness of the ice is likely to  be causal factor in the rate it melts out. It isn't the only causal factor but I  can't imagine it  isn't an important  one.  Simplification doesn't mean one is wrong, just  that one is trying to isolate the independent  variables in the melt.   
The biggest  argument  I  can think of against my  model is that the albedo  of ice is much  higher than the albedo of water.  Therefore early  melt  allows greater heating of the water and thus greater melting of the ice.  By  ignoring that  factor in my  model I  am obtaining a measure to  test my  prediction against.

[ChrisReynolds]

I  am not aware of any research that  claims that there is an upper limit to the amount  of volume loss in a given period.

Have you tried graphing the rate of change of PIOMAS volume and comparing to insolation? There are very fundamental limits to the amount of ice that can be lost in a given period, past behaviour gives us a guide to this.

[crandles]

I  am not suggesting that  the 1.1 M km^2 of extent that  was 1.7-1.8 m thick in April  is suddenly  going  to lose all that  thickness.  I  believe is has already  lost all but the last 0.2 m of thickness and that is what is melting  out  now.  After that  the same thing applies to the next  10 cm  and the next 10 cm.

My model predicts unreasonable rates of extent loss, the volume loss expected is directly proportional to the extent  loss. I  am not aware of any research that  claims that there is an upper limit to the amount  of volume loss in a given period.

Now you are slicing the histogram horizontally at least in the parts that you expect to melt out during the melt season. Have you calculated these volume losses and compared to PIOMAS volume losses? I think I suggested you do this a while back. It is more realistic to slice on a diagonal and calculating the volume loss in the parts that have melted out (slicing histogram vertically) and if you slice horizontally and comparing these to actual volume melt might give an indication of the typical slope of the diagonal.

+ what Chris Reynolds said.

I keep thinking I am going to give up on posting on this thread. 

[DavidR]

Crandles and Chris,
My model translated the April PIOMAS figures to an extent.  It then relates the actual extent to the modelled extent.  It has nothing to say about volume at this stage in the season. 

Insolation is measured in w/m^2, so it is directly related to area. The volume of ice that  can be melted out by  a given level of  insolation  is closely related to the extent.  Thickness is largely  irrelevant when you  are considering a force (insolation + bottom melt)  based on area where the area is 1,000,000 m^2,  and the thickness is less than 1 m which is the ratio of area to  thickness in one square km of extent   If the  extent is 2.5* greater, the volume of ice that  can be melted out should be 2.5* greater.

The model is based solely  on the statement:
If the NSIDC extent is X M km^2 the modelled thickness loss is Y m. 

That is, the minimum volume of ice that must melt out to acheive that extent loss, is the volume that was less than Y m thick at the start of the season.

At July 22nd the model shows:

July 22nd:         2011   2012   2013   2014
NSIDC Extent   7.046   7.197   7.576   7.745
Model loss (m)    1.72        1.66     1.75     1.70

The variation between the years after three months of melt  is remarkably small and I  expect the variation to  become even less in the next  few days.  The model has been close to  reality for the last three months.

Given that the next  month is the period when the model will be tested it is almost irrelevant to discuss it unless you can provide an alternative prediction that shows why it is incorrect.

[crandles]

We have now moved into the 1.7-1.8m range in the histogram. In this range as well as the next two ranges there is more than 2 1/2 times as much extent as in 2012.  That's 3.356 M km^2 compared to just  1.281 M km^2 in 2012.  In the past three years all of this range has melted out by the end of the season.  I  expect it  to  happen again this year.  That  takes us down to  4.350 M km^2.

The question at this stage is how much the weather contributes to losses greater than this..

On the same day, I wrote

CT area with Wipneus' updates:
5.292 - 0.077 + 0.032 = 5.247
...
To 8th 7th lowest for day (from 6th lowest).
Up to 822k above 2012.
      
07+Average   1.853   3.394 (6th lowest)
07+Average+sd   2.173   3.074 (5th lowest)
07+Average-sd   1.534   3.713 (8th lowest)

Further CT losses of 1.3 to 2.2 is still quite a range. Typical reductions like years 2007 onwards leads to 5 to 8th lowest area. A drop of 3.013 to become lowest on record seems well out of the range and seems exceedingly unlikely.

from this simple persistence model, becoming lowest on record looks like a 3.6sd event.

today I wrote

CT area with Wipneus' updates:
5.173 + 0.011 - 0.043 = 5.141
...
Remains 9th lowest for day.
Up to 1056k above 2012.
...
07+Average   1.485   3.656 (8th lowest)
07+Average+3sd   2.193   2.948 (4th lowest)
07+Average+2sd   1.957   3.184 (6th lowest)
07+Average+sd   1.721   3.420 (6th lowest)
07+Average-sd   1.249   3.892 (8th lowest)
07+Average-2sd   1.013   4.128  (10th lowest)

Further CT losses of 1.0 to 1.9 is still quite a range. Typical reductions like years 2007 onwards leads to 6th to 8th lowest area. Dropping to become lowest on record is just about a 6sd event and therefore seems exceedingly unlikely.

So a move from 7th lowest for the day to 9th lowest for the day and dropping from a mere 3.6sd event to a 6sd event to become lowest area on record.

Only 6 days later, maybe it is too soon to expect to see the rapid decline you seemed to believe was coming? How much does it have to move in the wrong direction to challenge your belief in your model?

[DavidR]

Crandles,
I  have to admit the last fortnight, in fact all of July, has not been good for the prediction. This is looking like the lowest rate of melt in July in the past 4  years.  I'll review the season at the end of the season and see what refinements need to be made. 

At this stage there is no point in moving the goalposts, if the hypothesis is wrong it's wrong.   The point was to make a prediction based on April data and see how good it was.  We'll see what  happens as the rest of the season pans out. 

The year to  date melt for the 4 years on 27th July varies from 1.73 - 1.79 m with 2012 making  a run to pass 2014 and catch  2011.

One possibility is that the fragmented ice we can see now is more mobile and therefore that  is maintaining the extent.  That's an issue to  address over the arctic winter.

[DavidR]

The metric for my  model is the April PIOMAS thickness estimate.
As at August  1st, the extent loss indicates that all the extent out to  1.77m in the original assessment has now melted out.

This is comparable to the original assessment of between 1.77 and 1.85 m however July 1st 2014 had the greatest melt figure of 1.37 m whereas August 1st is at the bottom of the range.

According to this metric only an additional 40cm of melt occurred in July  compared with 47 cm in 2011, 52 cm in 2012 and 55 cm in 2013.

Using the median figure for melt after August  1st , 26 cm , 2014 would still reach a final extent   below 2007, however with no apparent power in the system,  a melt  closer to  2013's 18 cm seems more probable.  This would put the minimum  at about 4.8M km^2.