Latest PIOMAS Volume update

[Andreas T]

What I find striking is the difference in the cryostat thickness plot compared with the piomas gridded plot above (difference between april / may is probably not large in arctic basin). Do you think with icebridge backup cryostat is more convincing than piomas thickness? Hycom http://www7320.nrlssc.navy.mil/hycomARC/navo/arcticictnnowcast.gifhas much larger thickness in Beaufort can that be disregarded? Do we have to wait for the melt to show which has got it right?

[Wipneus]

Updated, graphs are in the top post.

(gridded data updated also, I will post the graphs later)

[Wipneus]

Apologies for still comparing 2014 with 2013, at the moment I only have time to update the existing graphs.

Here are the thickness graphs of May 2013 and 2014 and the difference. No surprises I think.

[Wipneus]

Difference in April-May thickness change (positive means more growth/less melt in 2014).

[iceman]

The slope of the anomaly trace has been shallower over the past two months than in recent years.  My guess is that this coming month it will be steeper (excepting 2012), because many areas that show an increase in thickness for May over April will experience rapid melt during June.

[Nightvid Cole]

Neven,

Thanks for reposting my post. One thing I didn't state, but is probably clear from the graph, is that this is based on the monthly data. The results probably would not differ greatly if I used the daily data instead.

wili,

A couple of points in response. The reliability of predictions is dependent on the goodness of fit of the model, and unfortunately, this model isn't that good.

The model assumptions of a SARIMA model are that the next month's value can be written as a weighted sum of all previous months, plus a noise term. The monthly noise is assumed to be constant variance and uncorrelated with the noise from all previous months.

The best model I can find (I used a (3,0,0)x(1,1,1) model) has noise with an increasing variance and no short term correlations, but longer term correlations are nonzero. Bigger models don't do much better, and this model is already bigger than I would like anyway.

What this means in practice is that the predictions out to a year or so probably aren't terrible, but the reliability goes down beyond that, so I'm not sure it makes much sense to look at the predictions for 2016.

If you do look at the prediction for September 2016, the minimum value for the confidence interval is 1.3 x 10^3 km^3. Keeping in mind that the confidence intervals are probably too small, this means that there is some non-zero probability that the ice minimum in 2016 will be less than 1000 km^3.

(But this interpretation is a Bayesian interpretation of a frequentist model, which is kind of sketchy from a statistical standpoint. Coming from a frequentist model, the correct interpretation of the confidence interval is a probability statement about the model. The interpretation above is making a probability statement about the ice volume.)

On the other hand, the model also states that there is an equal probability that the ice minimum in 2016 will be above 8500 km^3. Based on the record up to this point, I do not think these two probabilities are in fact equal.

You're conflating two issues: The first is that there is some uncertainty in the fit parameters due to weather noise behind the data currently available (For which we only have likelihood or odds ratio distributions, not probabilities absent a Bayesian prior), and the second is that, even given a set of fit parameters, there is a proability distribution of deviations from the fit in a given future year due to weather in *that* year (For which we can assign a "true" probability distribution even from the frequentist school of thought).

[OSweetMrMath]

Warning: philosophy of statistics.

I've done some searching and haven't found any references which directly address the interpretation of prediction intervals. The predictions are structured as future value = deterministic term + random term. If we assume we know the exact deterministic term and probability distribution for the random term, then the prediction interval is a statement about the distribution of future values. In the language I used previously, this is a Bayesian interpretation of the prediction interval. In fact, the confidence intervals I have computed are constructed under this assumption, which implies that Nightvid Cole is correct. Contrary to my qualms, the confidence intervals define probabilities that the ice volume will fall in certain ranges.

But I'm not sure that I buy that interpretation. For linear models, there are two confidence intervals associated with the model, which are sometimes called the fitted confidence interval and the predicted confidence interval. The fitted interval shows the error in the estimate of the mean value for the function. This is clearly a frequentist confidence interval, and shows the uncertainty in the estimate due to the random nature of the data used to generate the estimate.

The predicted confidence interval is an interval for new data values. This includes both the variability due to the estimation errors and the variability due to the random term in the original equation. Since the estimation error is a statement about the precision of the parameter estimates, and the noise error is a probability distribution, I'm not convinced that the prediction interval has a single coherent interpretation.

As a possible way out, we can go full frequentist. The claim then becomes that the September ice volume is already a fixed number, and we just don't know what it is yet. The prediction formula then becomes

future value = known deterministic term + unknown deterministic term

The known term represents stuff that we know about and can model. The unknown term is stuff which we can't model. Since we can't model it, effectively we assume that all the relevant parameters are zero, and then the random noise term from the previous model can be interpreted as a statement about the accuracy of the claim that all the unknown parameters are zero.

Under this interpretation, the prediction interval is a frequentist confidence interval, making a statement about the prediction accuracy, not a probability distribution. I think this interpretation is more consistent with the construction of the underlying model and with the different kinds of intervals that are used for linear models.

Like I said, I checked various textbooks and did some searching online, and I didn't find anywhere directly addressing this interpretation question. If there is a source which argues one way or the other, I would be interested in reading it.

[jdallen]

Following OSweetMrMath, it follows to *me* as a result that extent and area have a near impossible number of terms to rationalize sensibly.  Volume is what I'd try to focus on. To wit:

Future ice volume equals Current ice volume in CC
Plus/minus (total regional insolation in joules (predictable value)
                  Minus albedo loss
                  Minus re-radiation loss
                  Minus sensible heat loss (energy expended raising the temp of other stuff)
                  Plus heat import from other sources in joules
                 )
                 Divided by 334 (heat of fusion of ice in joules/CC)

I'd argue from this that extent and area are simply how the deck chairs are arranged.  They may affect re-radiation and albedo, and have a limited affect on heat uptake otherwise, but are not as key.

[OSweetMrMath]

jdallen,

On the other hand, one of the virtues of a noise term is that you can throw all the terms you don't want to (or can't) manage into the noise, and as long as the noise is well behaved, you don't have to worry about those terms any more. In fact, I've been working with models to forecast both the PIOMAS volume and the NSIDC extent, and based on their histories, the extent model is likely to be more reliable than the volume model. We'll have to wait and see how they actually play out this year.

[crandles]

I may well be talking rubbish, but I thought that I understood that

Confidence intervals represent the probability of data being in the range but only on the assumption that our model is in fact correct.

What you often want is a credible interval that only comes from Baysian probability. This is inherently subjective (requires a prior) and give your degree of confidence that the data will be in the range.

If you have a physics today subscription, maybe this might help
http://julesandjames.blogspot.co.uk/2007/04/data-analysis-frequently-bayesian.html

[Wipneus]

Updated, graphs are in the top post.

(gridded data updated also, I will post the graphs later)

[epiphyte]

Delurking for a moment...

Look at the metrics we have for area, extent and volume. Is the volume estimated by the
PIOMAS model in part dictated by the other two... which would make any attempt to impute the health of the overall pack using volume/area an exercise in circular reasoning?

My own (totally unvalidated) hypothesis is that the health of the ice is better informed by surface:volume ratio than area/extent. This might help quantify the notion of ice "quality".... the larger the surface for a given volume in a given area, the more fragmented and vulnerable the ice...

[Wipneus]

Delurking for a moment...

Look at the metrics we have for area, extent and volume. Is the volume estimated by the
PIOMAS model in part dictated by the other two... which would make any attempt to impute the health of the overall pack using volume/area an exercise in circular reasoning?

epiphyte, PIOMAS brings in the thickness of the ice. With only extent and area there is no direct way  to establish thickness and therefore volume.

[Wipneus]

Here are the thickness graphs of June 2013 and 2014 and the difference.

[Wipneus]

And the difference in thickness change, the reds means less melt/more growth in 2014.
I think this shows that PIOMAS thinks large portions of the had less melt in June 2014.

[DavidR]

The thickness variation diagram for 2013 and 2014 is a good pointer to the reason PIOMAS is higher in June this year but still irrelevant to  final  extent. 

The ice is thinner in most of the areas likely to melt out but much thicker than 2013 in areas where the ice is already too thick to melt out.  It would be good to see a calculation of the volume of ice in the area that was above 2.5 m thickness last year, with a comparison for the same area in 2012 and 2014.  I suspect  the additional thickness in that  area alone would account for most of the additional volume this year. Thickness is above 1m greater for most of this area.  If this area covers 2M Km^2 that amounts to  an extra 2000 Km^3 in PIOMAS.

The extra thickness on the Atlantic edge of the pack in the Atlantic occurs from a very low thickness last year.  This ice is still thin enough to melt out and may be nothing more than an aberration caused by more ice drifting out of the central arctic than last year.

[slow wing]

Wipneus, so is that last plot the thickness change for: (June 2014 - May2014) - (same for 2013)?



I recall you asking for suggestions for the plot title; you could indicate the formula in the title...


"PIOMAS ice thickness trend: (June2014-May2014)-(June2013-May2013)".



Another option might be to write the formula on the legend.


"ice thickness change [m]: (June2014-May2014)-(June2013-May2013)".

[epiphyte]

Delurking for a moment...

Look at the metrics we have for area, extent and volume. Is the volume estimated by the
PIOMAS model in part dictated by the other two... which would make any attempt to impute the health of the overall pack using volume/area an exercise in circular reasoning?

epiphyte, PIOMAS brings in the thickness of the ice. With only extent and area there is no direct way  to establish thickness and therefore volume.

That's understood. What I'm trying to get at is that PIOMAS imputes volume using (in part) the area, + discrete measurements + gridded estimates of thickness, and some here go on to impute the average thickness from the PIOMAS volume, and then use that as an indicator of the robustness of the ice. Doesn't this just reduce the information content of the measurements + model output?

[crandles]

Delurking for a moment...

Look at the metrics we have for area, extent and volume. Is the volume estimated by the
PIOMAS model in part dictated by the other two... which would make any attempt to impute the health of the overall pack using volume/area an exercise in circular reasoning?

epiphyte, PIOMAS brings in the thickness of the ice. With only extent and area there is no direct way  to establish thickness and therefore volume.

That's understood. What I'm trying to get at is that PIOMAS imputes volume using (in part) the area, + discrete measurements + gridded estimates of thickness, and some here go on to impute the average thickness from the PIOMAS volume, and then use that as an indicator of the robustness of the ice. Doesn't this just reduce the information content of the measurements + model output?

I think the modelled thickness is improved by assimilating actual data but largely because there are hindcasts where parameters are tweaked so that the minimum adjustments to assimilated information is necessary. 

I think this ensures the modelled thickness is more consistent to actual over time. It could well be consistently wrong if the model is poor but when compared to icesat2, volumes seem very similar at minimum but PIOMAS understates volume most at maximum.

Under assimilation, PIOMAS is adjusted to actual extent and area. Apart from this, thickness comes from the PIOMAS model. These seem nearly* independent so I don't see any circular reasoning.


*nearly: differences between model and actual extent/area used to tweak parameters seems close to independent from actual extent/area to me. Other people may disagree about this.

[A-Team]

Wipneus, I notice sometimes your palettes are measured gray scale steps with a hue overtint. Other times, not. In the first case, I can drop them into PovRay freeware for a 3D perspective view (for what that is worth). Script:

include "colors.inc"
  camera{
    location <-15,20,-31>
    look_at 0
    angle 35
  }
  light_source{ <1000,1000,-1000> White }
  height_field {
    png "t2gray.png"
    smooth
    pigment { White }
    translate <-.5, -.5, -.5>
    scale <17, 2, 17>
      texture{pigment{
                     gradient y
                     color_map {
   [ 0.00000 rgb < 1.00000, 0.01111, 0.05829> ]
   [ 0.05000 rgb < 0.99728, 0.68188, 0.62282> ]
   [ 0.10000 rgb < 0.99457, 0.43569, 0.12356> ]
   [ 0.20000 rgb < 0.99457, 0.65239, 0.12521> ]
   [ 0.30000 rgb < 0.99457, 0.87889, 0.06625> ]
   [ 0.40000 rgb < 0.83409, 0.99457, 0.02736> ]
   [ 0.50000 rgb < 0.38010, 0.99457, 0.07204> ]
   [ 0.60000 rgb < 0.05287, 0.99457, 0.64586> ]
   [ 0.65000 rgb < 0.46341, 0.97697, 0.89598> ]
   [ 0.70000 rgb < 0.04840, 0.95938, 0.99457> ]
   [ 0.75000 rgb < 0.43931, 0.82767, 0.99457> ]
   [ 0.80000 rgb < 0.09890, 0.49815, 0.99457> ]
   [ 0.90000 rgb < 0.16623, 0.08811, 0.99457> ]
   [ 1.00000 rgb < 0.47403, 0.02386, 0.99457> ]
} }}  }

[Wipneus]

Wipneus, I notice sometimes your palettes are measured gray scale steps with a hue overtint. Other times, not.

A-Team, I guess that is because thickness asks for a "sequential palette", going from 0->max. For changes a "diverging palette" is appropriate going from thinner->neutral->thicker.

"sequential" and "diverging" definition taken from the color brewer man page:

     There are 3 types of palettes, sequential, diverging, and
     qualitative.
     1. Sequential palettes are suited to ordered data that progress
     from low to high.  Lightness steps dominate the look of these
     schemes, with light colors for low data values to dark colors for
     high data values.
     2. Diverging palettes put equal emphasis on mid-range critical
     values and extremes at both ends of the data range. The critical
     class or break in the middle of the legend is emphasized with
     light colors and low and high extremes are emphasized with dark
     colors that have contrasting hues.
     3. Qualitative palettes do not imply magnitude differences between
     legend classes, and hues are used to create the primary visual
     differences between classes.  Qualitative schemes are best suited
     to representing nominal or categorical data.

For the thickness maps I use a R-builtin palette called "terrain colors", for the changes the color brewer palette named "Spectral".

Attached are the R-builtin palettes and those from color brewer.

[Wipneus]

Updated, graphs are in the top post.

[Andir]

Volume anomalie isn't so big this year. How large is the influence because of the slow Fram export in your opinion?

[Wipneus]

Volume anomalie isn't so big this year. How large is the influence because of the slow Fram export in your opinion?

It did not help, but low Fram export in the summer is not unusual.

[Wipneus]

Here are the thickness graphs of July 2013 and 2014 and the differences.

[Wipneus]

And the difference in thickness change in July (red is morelessmore thickening or lessmoreless thinning).

[slow wing]

Fascinating! Thanks, Wipneus!



The volume has split right away upwards from 2013 and is now hanging out with 2009 and 2007!



It's going to take a lot of melt in the next month or so to get back to the post-2009 years at minimum. Otherwise, we will end up back around the minimum volume levels of the 2007-2009 years!

[crandles]

PIOMAS         
Year, fall to min _31/7 __minimum
2005   3.094   12.253   9.159
2006   2.946   11.939   8.993
2007   2.714   9.172   6.458
2008   4.360   11.432   7.072
2009   2.974   9.813   6.839
2010   2.823   7.405   4.582
2011   2.677   6.979   4.302
2012   3.003   6.676   3.673
2013   2.403   7.795   5.392

Year   _ Fall __ If follows this fall   
2005   3.094   6.481   
2006   2.946   6.629   
2007   2.714   6.861   
2008   4.360   5.215   
2009   2.974   6.601   
2010   2.823   6.752   
2011   2.677   6.898   
2012   3.003   6.572   
2013   2.403   7.172   

         
07+Average____   2.993   6.582   6th lowest
07+Average+3sd   4.900   4.675   4th lowest
07+Average+2sd   4.265   5.310   4th lowest
07+Average+sd   3.629   5.946   5th lowest
07+Average-sd   2.358   7.217   8th lowest
07+Average-2sd   1.722   7.853   8th lowest

2008 looks a bit of an outlier having a large fall and it was from a high quantity at 31st July. Excluding that, I get

07+9-13Avg___   2.766   6.809   6th lowest
07+9-13Avg+3sd   3.430   6.145   5th lowest
07+9-13Avg+2sd   3.208   6.367   5th lowest
07+9-13Avg+sd   2.987   6.588   6th lowest
07+9-13Avg-sd   2.544   7.031   7th lowest
07+9-13Avg-2sd   2.323   7.252   8th lowest

So currently 6th lowest and at minimum seems likely (2sd range) to be 5th to 8th lowest.

[Nightvid Cole]

Fascinating! Thanks, Wipneus!



The volume has split right away upwards from 2013 and is now hanging out with 2009 and 2007!



It's going to take a lot of melt in the next month or so to get back to the post-2009 years at minimum. Otherwise, we will end up back around the minimum volume levels of the 2007-2009 years!

There's no such thing as "the post 2009 years", apparently, as far as PIOMAS volume is concerned. Just fluctuations superimposed on a long term decline, even though the latter still appears significantly nonlinear.

[iceman]

I'm quite surprised at the upturn in the anomaly trace: steeper this past month than in any recent year.  Judging from your last graph, much of the anomaly increase comes from slower thinning in the central part of the ice pack. (more so than in the comparable graph from a month ago)  Is this attributable to the unusually high compactness in 2014?

[Wipneus]

I'm quite surprised at the upturn in the anomaly trace: steeper this past month than in any recent year.  Judging from your last graph, much of the anomaly increase comes from slower thinning in the central part of the ice pack. (more so than in the comparable graph from a month ago)  Is this attributable to the unusually high compactness in 2014?

Yes of course. Directly because the thickness of a grid box is really the average thickness ( the modelers call it effective thickness) of the ice. If concentration goes up, the average goes up because the box contains more ice and less open water.

Indirectly because compacter ice absorbs less incoming solar.

Further I am convinced that part of the increased compactness as far as it is measured with data like the NSIDC sea ice concentration data is explained by the low temperatures resulting in less melting, dryer surface. Obviously that is also going along with less thinning.

[NeilT]

Help me out here, but didn't we see, earlier in the year, buoy data which showed low temperatures on the surface and significant bottom melt at the same time?

Wasn't that around where the "thick ice" is now vanishing in the Beaufort?

I know that PIOMAS has been accused of overestimating in certain circumstances.  Could it be that if there is a cold snap and we see 5cm ice in an area of 1M ice that PIOMAS will average it all as 1M?

Just thinking out loud.

[edit]

I also recall Dr Barber talking about this phenomena when he did his presentation on rotten sea ice. Where areas of very thick but disintegrating ice was being averaged as all the same thickness when it was, in fact, largely thinner ~10cm ice with thicker ice holding it all together.

[jdallen]

Help me out here, but didn't we see, earlier in the year, buoy data which showed low temperatures on the surface and significant bottom melt at the same time?

Wasn't that around where the "thick ice" is now vanishing in the Beaufort?

I know that PIOMAS has been accused of overestimating in certain circumstances.  Could it be that if there is a cold snap and we see 5cm ice in an area of 1M ice that PIOMAS will average it all as 1M?

Just thinking out loud.

[edit]

I also recall Dr Barber talking about this phenomena when he did his presentation on rotten sea ice. Where areas of very thick but disintegrating ice was being averaged as all the same thickness when it was, in fact, largely thinner ~10cm ice with thicker ice holding it all together.

I think you are on the right track, if not the exact detail.

While we are increasing the data we have available about volume, we are limited by scale and resolution.  When Icebridge flew over the Beaufort, ESS and Chukchi, it showed large amounts of MYI in place. It did not clearly describe its construction, or how in August it would come to look like this:

(Chukchi/Beaufort north of Bearing strait)
http://map2.vis.earthdata.nasa.gov/imagegen/index.php?TIME=2014217&extent=-2076192,-121792,-1138208,699456&epsg=3413&layers=MODIS_Terra_CorrectedReflectance_TrueColor,arctic_coastlines_3413&format=image/jpeg&width=1832&height=1604

I think we are struggling greatly to understand the effect structural changes in the ice will have on both its behavior and regional weather.

[SCYetti]

The PIOMAS validation page states that their model has a bias that under estimates thick ice and over estimates thin ice. What is the effect of this when the Arctic ice is overwhelmingly thin.

[jbatteen]

Looks like a lot of the thickening is in areas of thin ice on the periphery that shouldn't be too hard to get rid of.  While a lot of thinning is in the reserve of MYI that's holding everything together.

[ChasingIce]

The PIOMAS validation page states that their model has a bias that under estimates thick ice and over estimates thin ice. What is the effect of this when the Arctic ice is overwhelmingly thin.

apparently, it would over-estimate thin ice, and under-estimate thick(er) ice. 

[epiphyte]

SCYetti writes:

The PIOMAS validation page states that their model has a bias that under estimates thick ice and over estimates thin ice. What is the effect of this when the Arctic ice is overwhelmingly thin

I've been wondering about that too. W.r.t the August update I find it especially intriguing that from wipneus' chart : "PIOMAS modeled ice June-July thickness change 2014-2013" it looks at first glance (as jbatteen alludes to above) as though as almost all of the
reddest areas (i.e. more thickening/less thinning) are at the periphery (i.e. - where the ice is probably thinnest... and where PIOMAS is most likely to overestimate). It also appears from comparison with MODIS that as of Aug 04. almost all of these deep red areas are over open water... which causes me to question the practical value of this model in predicting the seasonal fate of first-year ice.

[ktonine]

It also appears from comparison with MODIS that as of Aug 04. almost all of these deep red areas are over open water... which causes me to question the practical value of this model in predicting the seasonal fate of first-year ice.

Yes, it seems rather odd that the ice has gained thickness precisely in those areas where there is no ice.

I have been confused a number of times over the past two melting seasons.  I'll just add this to the growing list of things I simply cannot comprehend.

[Wipneus]

Help me out here, but didn't we see, earlier in the year, buoy data which showed low temperatures on the surface and significant bottom melt at the same time?

Let me hasten to say that most what I said is relative. I have no doubt that if the temperature on the surface were not "low", the melt would have been even higher.

Wasn't that around where the "thick ice" is now vanishing in the Beaufort?

Did it? I have looked at the images from february and it seems a good portion of that thick ice has moved on with the Beaufort/Arctic gyre.

I know that PIOMAS has been accused of overestimating in certain circumstances.  Could it be that if there is a cold snap and we see 5cm ice in an area of 1M ice that PIOMAS will average it all as 1M?

That is not how PIOMAS works. It is a physical model that result in modeling data. When comparing the model data with real data, the thinner ice is found to be overestimated (and thicker ice underestimated) at least in the comparison period up to 2010. That is not explicitly built in, it is an "emerging" feature of the model.

[Wipneus]

Yes, it seems rather odd that the ice has gained thickness precisely in those areas where there is no ice.

I have been confused a number of times over the past two melting seasons.  I'll just add this to the growing list of things I simply cannot comprehend.

Sorry guys, I must have been confused. The reds are more thinning in 2014.

So where there is ice in 2014, but none in 2013 (eg Barents Sea) you will see red because the thinning in 2013 was zero.
And where there is no ice in 2014, but in 2013 there was (eg Beaufort) you will see blue because thinning in 2014 is none.

[Neven]

I have put up the latest PIOMAS update on the ASIB: PIOMAS August 2014.

From the conclusion:

After the lessons of 2012 and 2013, showing the importance of the start of the melting season, it's fascinating to see the theory confirmed again. A lack of melt ponds during May, and very litle movement, even during times when high-pressure areas dominate, seem to have serious implications for how much the ice pack recedes and thins during the melting season.

Equally fascinating will be to see how high volume can end up this year, perhaps even doubling the rebound from last year. And of course, what it will all mean for melting seasons to come. But there's still a couple of weeks of melt to come before drawing any preliminary conclusions.

[NeilT]

Thanks wipneus.  I just read stuff and get impressions. Then I ask questions.  Nice to know.

I'm just watching to see what is happening right now.  The pack appears, from the modis images, to be in a worse state every year.  Yet the volume keeps going up.

Something doesn't match.

Looking at the Modis images The entire eastern side of the pack, from Svalbard through to the Chuchki, looks more like a flotilla floating in concert than an ice pack.  In such a scenario I would have thought that the ice would have had to be much greater thickness than is actually evident to give such high volume figures.

That's just an observation mind you.  No science behind it at all and could quite likely be wrong.

As for the thick ice moving on.  Where would it have gone?  As we see from the thickness graphs, the entire eastern side of the pack is thin or first year, even trending back into the pole.  Has they gyre moved it back into the CAA?

[Steven]

Sorry guys, I must have been confused. The reds are more thinning in 2014.

Wipneus, I think the original (unedited) version of Reply #275 upthread was the correct one.  The red color means that there was less thinning in 2014 than in 2013.

where there is no ice in 2014, but in 2013 there was (eg Beaufort) you will see blue because thinning in 2014 is none.

Replacing the word "blue" by "red" in the quoted sentence, it becomes:   

"where there is no ice in 2014, but in 2013 there was ice (e.g. in southern Beaufort/Chukchi) you will see red because thinning in 2014 is none."

[iceman]

Sorry guys, I must have been confused. The reds are more thinning in 2014.

Wipneus, I think the original (unedited) version of Reply #275 upthread was the correct one.  The red color means that there was less thinning in 2014 than in 2013.

That's the way I read it as well.  If an area was mostly melted out by June 2014, it will show red as "less thinning" as of July 2014.
   The combination of month/month and year/year comparisons makes this a little tricky to interpret: previous sentence is predicated on 2013 having more July thinning in the area being compared.

[Wipneus]

red color is less thinning in 2014

Let me see. What I am displaying is the difference between two subtractions:

  ( thk(2014,7) - thk(2014,6) )   -     ( thk(2013,7) - thk(2013,6) )

thinning means that thickness in July is less than in June, so these subtractions have negative results.

Less thinning means the number on the left side is less negative than on the right.

So the result is positive.

Look at the legend: red is positive.

So it is true that red is less thinning.

Now what did I write, hmm, grrm. I need a vacation.

[iceman]

Now what did I write, hmm, grrm. I need a vacation.

                                                    ... or a nap.

[ChrisReynolds]

Now what did I write, hmm, grrm. I need a vacation.

                                                    ... or a nap.

It's reassuring to find that Wipneus sometimes finds these things confusing, I may be in good company in that respect, but for the record - Wipneus's mathematical ability beats mine any day.

[Frivolousz21]

Thanks wipneus.  I just read stuff and get impressions. Then I ask questions.  Nice to know.

I'm just watching to see what is happening right now.  The pack appears, from the modis images, to be in a worse state every year.  Yet the volume keeps going up.

Something doesn't match.

Looking at the Modis images The entire eastern side of the pack, from Svalbard through to the Chuchki, looks more like a flotilla floating in concert than an ice pack.  In such a scenario I would have thought that the ice would have had to be much greater thickness than is actually evident to give such high volume figures.

That's just an observation mind you.  No science behind it at all and could quite likely be wrong.

As for the thick ice moving on.  Where would it have gone?  As we see from the thickness graphs, the entire eastern side of the pack is thin or first year, even trending back into the pole.  Has they gyre moved it back into the CAA?

It's the MYI.  It's just in shambles and piomas thinks it's thicker than it is. That simple.

[OSweetMrMath]

I usually work from the monthly data from PIOMAS, but it doesn't seem to have been updated yet, even though the daily data is clearly out for July. Does anyone else use the monthly data? Have you gotten the updated numbers? Anyone have any info on why it's not updated yet?

[ChrisReynolds]

OSMM,

I don't use it, but you're right, it's not been updated. Don't know why, daily and gridded is OK.