« Peeking through the clouds 4 | Main | First Petermann Ice Island photos »


Feed You can follow this conversation by subscribing to the comment feed for this post.

Peter Ellis

Paul: I acknowledge your thanks for my having pointed you at the IMS product from the NIC. I'd really like it however if you addressed the rest of the post rather than tunnel-visioning your pet theory. To make it simple for you, I'll boil it down to one question.

What is your interpretation of the MODIS pictures from the 13th of August, showing sea ice floating near Point Barrow?

I'd like it if you didn't insult the hard-working analysts by claiming "mis-dating" until you've discussed this point. Different sensors have different properties. THIS DOES NOT MEAN THAT ONE OF THEM IS GETTING THE DATA WRONG!

If anything, the difference tells us something interesting about the kinetics of melting ice. That is, it tells us that it takes a little over ten days for ice to fully melt out, once it has been rotted/submerged to the point it no longer shows up on passive microwave imagery.

Artful Dodger

Hi Peter.

Good analysis upthread. Of course, there is a easy, diagnostically certain way of determining when part of the Arctic ocean is 100% sea ice-free in Summer: SSTs.

Temps will always hover around the freezing point, until the last of the sea ice is gone. Only then will solar heat go into warming the water, rather than the heat of fusion being consumed as ice melts.

If it's still freezing cold, there's still some ice!


Interesting stuff that about IMS. I have their graph on and off on the ASI Graphs page (I use it to fill up space when necessary) and was wondering why they were the odd man out.

Paul Klemencic

Seke Rob: IJIS uses two days, but since using the new sensor, sometimes uses three days. However, I was never able to find a set of numerical values that fit simple two day averaging, so…

Anyway, suffice to say that IJIS doesn't show even one 200k loss day during the GAC-2012, whereas the NSIDC report says that they saw two consecutive 200k loss days.

From the IJIS site there is this discussion of averaging:

Averaging period and the update timing of daily data:

In general, sea-ice extent is defined as a temporal average of several days (e.g., five days) in order to eliminate calculation errors due to a lack of data (e.g., for traditional microwave sensors such as SMMR and SSM/I). However, we adopt the average of latest two days (day:N & day:N-1) to achieve rapid data release. Only for the processing of WindSat data (Oct. 4, 2011 to the present) the data of the day before yesterday (day:N-2) is also sometimes used to fill data gaps.

Paul Klemencic

Peter Ellis: I really don't want to argue with you about the accuracy of the different sensors. I just don't find it plausible that sensor accuracy could lead to ice extents that disagree by over one million sq km. This is a huge error, and nullifies your conclusion.

MASIE also uses a tighter grid of 4 km, and actually should read lower than the bigger gridded products during the melt season because they can pick up open water amongst the pack. I know you believe otherwise because of the different sensor methods, but I disagree. The other tighter grid product, the Bremen SIE uses a 6.25 km grid (although now with a less accurate sensor), and has usually shown lower SIE during August and September. I believe that MASIE, which was designed to show accurate daily SIE, will show extents close to the Bremen results, when the dating problem is resolved.

I do look at the Arctic Mosaic, and I don't see million sq km errors in the passive microwave product maps, as you claim. What I do see, is very close agreement of the MASIE/IMS maps with the Bremen/CT maps for two week earlier.

I don't think you have ever bothered to check the agreement between today's MASIE map, and the Bremen map for two weeks ago. Do it, and really examine the two maps. There are differences in areas with low concentration ice, but not much…. almost certainly less than 100k sq km. Furthermore the distribution of the ice matches. For example, the MASIE map for Friday, still showed a lot of ice at the entrance to Hudson's Bay. Not likely, that. But two weeks ago, there was ice there.

Your repeated hypothesis of sensor "errors" of up to one million sq km doesn't hold up to observations.

Paul Klemencic

Peter Ellis: One more observation to clinch it. We can look at the Bremen map and predict the future!

Tomorrow (or following day at the latest) the MASIE map will show the extent of the pack for August 8th, and will show the detachment in the E. Siberian region.

Check the MASIE map dated August 20 tomorrow, and see the detachment for yourself.


Paul, Peter,
Let’s not get into the trenches over this difference of interpretation.
Looking at MODIS the Healy cam images perfectly reflect there still is ice in a large sector that looks like icefree on UniBremen.
It led me to assume that the final melt period (stretching into ’13) is about spread/fragmentation. An Ocean filled with debris, pieces.
As Lodger writes, most of the -1.5 dC area on NCEP/NCAR SST stands for area where bits and pieces remain.
Together, the total volume by mid-september will be frightfully low.


Peter has explained this in a way that makes sense a 100 times now.

Paul Klemencic

Ouch… can't add 14 to 8 and get 22…

The map for August 22 will show complete detachment, the map for August 21 might show it, depending on time of day observational differences. The detachment on the Bremen map was almost complete on August 7th, and done by August 8.

Paul Klemencic

Downil1: Yes, seems to make sense, but then doesn't fit all the observations. Science is funny that way. The hypothesis must be rejected if contradicted by observations, and there are a slew of observations inconsistent with the "sensor difference can measure SIE with differences of 500k to 1000k sq km" hypothesis proposed by Peter Ellis.

Paul Klemencic

Werther: The Healy sailed north, found a big area of ice floes right around 15%, and has been criss-crossing the same patch of ice. Check the ship's log.

Don't conclude too much from the number of conning tower shots showing ice. Its far from an random sample, as you might get from a straight course north.

Paul Klemencic

Here is the course info for the Healy:


Using the double zoom feature of Arctic.io, one can see that snow is developing again in the Canadian Archipelago.

Peter Ellis

*shrug* All I know is that I can look at the Terra/Aqua pictures and see ice in areas that IMS calls as ice, and that Bremen called as ice-free a fortnight ago. Not just the area near Barrow, that just happened to be a nice clear picture. If you look between the clouds, you can make out sparse ice all the way from the "island" the the main pack, throughout that last fortnight.

As well as the data of our own eyes, I've also pointed you at a published paper documenting a well-known deficiency of passive microwave methods in picking up rotten/submerged ice. Moreover, we ALL know (and have discussed extensively) the way melt ponds affect the microwave signal, lowering the apparent concentration substantially in heavily-ponded areas - by around 50% in the "blue ice" zones of the Laptev and CA this very year.

Tell me, what happens when the actual concentration is around 25%, then you have melt ponds on top of that, and then you finally submerging the whole lot under a storm surge?

Or no, don't tell me. I'm done arguing with you on this. We'll see what IMS say in regard to Florence's enquiry. And you can write up your paper saying that the scientists are too stupid to count to 30, and thus introduce a randomly-varying delay that's different each year. Perhaps you can use that to refute the published literature on passive microwave properties in sensing sea ice boundaries.

Peter Ellis

Paul: I'm not sure where precisely you mean by "the entrance to Hudson Bay". Can you be more specific?

If you mean Foxe Basin, then the Canadian Ice Service are indeed showing ice there right now, ~90% concentration in an area Bremen says is no more than about 40%



That's kind of the point - MASIE is multisensor, and will call an area as ice if any of the data sources show ice there.

Paul Klemencic

Peter Ellis: Please don't put words in my mouth. I never said any of the things that you wrote in your last paragraph. Nothing about "scientists are too stupid to count to 30", or "randomly-varying delay thats different each year", or that I can "refute the published literature on passive microwave properties in sensing sea ice boundaries".

None of those statements come even close to describing my observations, or the hypothesis that there is a delay in the IMS/MASIE reports. Please, lets not go to that level of hyperbole, and that level of distortion of the other's thinking and analysis.

Last year, when the delay (lag) was 5 days for the extent data, and the MASIE map seemed seven days old, the difference was only several hundred thousand sq km, and I didn't have decisive data demonstrating the problem. So I did what most scientists would do; wait for more data.

When the storm came, I knew that this was going to be a good test, because of the quick extent damage to the pack, and centered in the Beaufort, Chukchi, and E. Siberian regions at that. I didn't know the delay had grown to two weeks on the maps, and 12 days on the data.

But now we have a set of observations that are clearly inconsistent with the "sensor inaccuracy " hypothesis. The difference in SIE is over one million sq km, if the reports from IMS/MASIE are correctly dated. You continue to attribute that to sensor inaccuracies and large areas of low ice concentration. We haven't seen that kind of difference before, and its extremely unlikely that is the case.

Then there is the issue, that all these shredded ice floes hung around for two weeks, with very little SIE change, then suddenly 588k sq km melted out in two days! What is the possible explanation for that? There wasn't any obvious weather situation that would have caused this massive, unprecedented (at least in August) drop in SIE. Your hypothesis completely fails to explain this observation.

Then there is the obvious clear synchronization of the shape and distribution of the SIE shown in the IMS/MASIE maps, and the Bremen/CT maps from two weeks earlier. What possible explanation can explain why the ice hung around for two weeks, then melted out in an almost identical pattern to the PM maps of two weeks earlier?

I can see why you would want to ignore these observations. I think that you gradually got trapped into an untenable position. Over the last year, the position may have been defendable, but as the reporting delay increased, and then we encountered a megaweek melt event, the original argument became impossible to defend.

So now you must resort to different methods to defend it. I suggest that you think about it for a while, let several days, or a week of data roll in, then re-visit your position. You will likely find that you want to modify it.

I waited a year; you can wait a week or so.

Rob Dekker

Paul, Peter,
When there is a significant difference between two data sets, both resulting from different systems, trying to find the cause is often worse than figuring finding out why your laptop locks up every two days, while your desktop does not, when they use different versions of the same OS.

Paul, I don't envy you. Even if there is a system problem with that date fed to the Masie algorithms, what you are essentially trying to do is reverse-engineer a defect in a complex system. That's very hard to do. Even if you got the main problem correct, the system response to the data that follows could behave slightly different depending on the different sensor inputs and system resolution, and algorithms executed.

If your hypothesis is correct, and the difference in data sets is (mostly) as simple as a delay in Masie data set versus Bremen, then why don't you collect the Masie and Bremen data, and put them both in the same SIE graph and present it, and check which delay shows the best correlation, and how high that correlation really is. At least that would put some mathematical muscle to the 'delayed data input' assertion.

Even if it does not match exactly, it would still say something about a possible 'delay' issue with one of the two systems.

Peter, it's nice of you to propose an alternative hypothesis : different sensors picking up different ice melt stages, which may explain the 5-12 day difference. However, it is not so nice to shrug off Paul's hypothesis and skeptical analysis with "scientists are too stupid to count to 30" and "rather than tunnel-visioning your pet theory" and other unconstructive and unscientific remarks.

All that Paul is trying to do is explain the difference between Masie and other data sets. And that we are (should) all be interested in.


Rob Dekker wrote:

All that Paul is trying to do is explain the difference between Masie and other data sets. And that we are (should) all be interested in.

Au contraire, mon ami!

Before, we only should be interested in reliable data. Before all en everything.

Now what is it that Uni-Bremen is telling abouthis data?

A note of caution: In the diagrams below, the Arctic sea ice extent after mid-August, 2012, is below last year's minimum. While the drastic decline since early August is confirmed by other sources, the absolute value is not. It can be a consequence of the rather simple time series adaptation method used. While the result was in agreement with other data over most of 2012, the discrepancy has grown since early August. The issue is being investigated. This is probably not a record (yet)!

Bottom line, Uni-Bremen clearly states the date from August on aren't correct at all.

So, as I told Neven already in the very beginning of this theme, the desperate tries to purge domino or records or whatever out of these clearly wrong date are out of order.

Elementary my dear fellow. elementary.

Peter Ellis

I apologise for the intemperate tone of my remarks above. I agree that it would be helpful for Paul to actually plot some Bremen / MASIE data and show the correlations, but I don't think that's possible since Bremen don't publish their numbers at all, let alone the same area-by-area breakdown as MASIE.

The other thing I would add is that when you start allowing yourself a variable delay (5 days last year, now 12 days, or 14 days...), you are getting into the realms of epicycles. You can fit anything to anything, pretty much. Moreover, a fixed delay could perhaps be explained by a programming bug or systematic error in integrating data into MASIE. A variable delay can't so easily be explained without serious error on someone's part.

I guess perhaps there is not so much difference between us in our actual description of the curves, just in the interpretation. I say that Bremen sees the melt before MASIE (i.e. there's a delay) due to real, known physical characteristics of the detector systems. Paul says there's a delay for unknown reasons presumably related to data processing.

I think it's useful to look at the weather behind the evidence Paul's using to claim a processing delay. Both last year and this year, he's mainly looking at the "flash melt" events from storms, since these give blips in the melt record - both at the map level and at the numerical data level - that you can then use to align one record to the other. Last year had less severe storms, which showed up on MASIE after a few days' delay. This year had an unprecedentedly severe storm, which didn't show up on MASIE until a fortnight later. This is entirely compatible with my hypothesis. A weak storm will fracture and submerge comparatively thin ice, which will then only last a few more days. The Great Arctic Cyclone was able to overturn / fragment / submerge substantially thicker floes, which then take longer to finally disappear. Does Paul have any explanation for why storms of different strengths produce different degrees of delay on MASIE?

Finally, Paul's offered a couple of predictions - none of which I feel really distinguish between his hypothesis and mine, since I too expect MASIE to show the "island" forming within the next day or two.

In the spirit of hypothesis-testing, here are some of my predictions which I hope will distinguish the two. I realise that technically, most of them are retrodictions, but since I haven't actually looked yet, from my point of view, they're predictions!

1) In the first half of June, where microwave methods showed extremely rapid declines in the Canadian Archipelago and Laptev sea (the "blue ice" phenomenon), I predict that MASIE didn't show any changes and a coverage that stayed near 100%. Note that collectively this was a good few hundred thousand square kilometres.

2) MASIE will show much less of a delay in areas that were not touched by the Great Arctic Cyclone - these areas are less obscured by cloud, and the ice was not submerged by storms. Bremen shows that the ice edge by Severnaya Zemlya has retreated from about ~82 degrees to ~80 degrees in the last fortnight. I predict that MASIE shows much the same. This prediction also applies to the ice edge in general in other areas. When melt is occurring "naturally", without storm disturbance, the apparent delay will be a couple of days. Melt from storms will not show up on MASIE until later - and in general the stronger the storm, the longer the delay.

3) MASIE will not be a fortnight late in showing the September minimum <-- look, a genuine, simple prediction about the future!

4) In the areas of the Beaufort, Chukchi and Canadian Archipelago where MASIE shows ice remaining beyond the time Bremen says it's ice-free, then the charts compiled by the Canadian and Alaskan ice services (which are done by radar, visible wavelengths and eyes-on-the-ground observation) will show ice in areas that agrees with MASIE, and not with Bremen. This includes, but is not limited to:
* a small strip across the centre of the NWP around Resolute
* the north coast of Alaska near Point Barrow
* the area between the "island" and the main pack
* Foxe basin <-- I already checked the Canadian charts above, so this one is not a real prediction
* Any similar areas forming between now and the minimum

Paul Klemencic

First off, a comment on the TypePad system.
It is important that readers be able to read the full set of comments from commenters. Prior to this week, a reader could click on the highlighted name, get to their page, and read all their comments. This weekend, I did exactly that with Peter's profile, to make sure I understood his position, and his concerns. I re-read all his comments back to last August. I also did that with mine.

Unfortunately today, I can't do this anymore. Now I can only access his (or my) recent comments. This is a shame, because its harder to check the history and understand the full logic of the discussion. Does anyone know how to get access to the full run of comments?

If anyone did this, they could read the comments from last August/September, and realize why I am interested in MASIE (it has to do with the process characteristics of the melt process, particularly the heat transfer coefficients and heat transfer rates).

It would also prevent a lot of regurgitation of previous commentary, since we could simply link to prior comments.

If anyone knows how to turn access to the full comment stream back on, could they tell us (me) how to do it?

Paul Klemencic

Secondly, lets put up today's MASIE numbers for August 20. Another spectacular day of melt, with 143k sq km melting out to make the 3-day total 731k, an average of 244k sq km per day. These are unprecedented losses, especially for August.

The E. Siberian (72k) and the Chukchi (36k) led the way, with losses around 10k each in the Laptev, Greenland, CA, and CAB. The Beaufort actually gained 4k.

The proposition that we had over a million sq km of low concentration ice that isn't shown on the Bremen map, or in the IJIS/Bremen data, and that suddenly melted 731k sq km out in three days is impossible, due to the process mechanisms determining heat transfer. Lower concentrations of ice melt out faster, much faster, than higher concentrations of ice. The idea that one million sq km of ice almost melted to below 15%, then stopped melting for several weeks, then 75% melted out in three days, simply doesn't agree with known melt mechanisms. There must have been some other unusual cause for this massive melt loss (I think it was GAC-2012).

Next up, the MASIE map for August 20, greatly resembles the Bremen map for August 6, except that a lot of the dark blue areas on the Bremen map don't show up on the MASIE map.

Please be aware that the Bremen map shows ice concentrations down to 10% as dark blue (look at the color legend on the map). The MASIE map only shows 15+% ice concentration grids. Both use fairly tight grids of 6.25km for Bremen, and 4 km for MASIE. It shouldn't be surprising that a lot of the blue areas on the Bremen map, don't show up on the MASIE map.

If I can easily establish links to prior comments, then I can address the comment and suggestion by Rob Dekker above. I can also address and show as incorrect, the conclusion reached by Peter Ellis, that over one million sq km of low concentration ice exists (15-30%) that is being picked up by MASIE, and missed by IJIS/NSIDC/Bremen passive microwave products, and not displayed on the Bremen map. Poor Werther has now appeared on one of the other threads, repeating this incorrect conclusion.

Espen Olsen


The discussion about the various graphs, will of course always be there, but I think the Bremen map is what is showing the real concept of the actual ice conditions are like in the Polar Sea, and for those interesting in more conservative maps down 10 % ice look for this map :
That map looks more, but only looks like what Joe Batardi, "studies" found:


Paul Klemencic

Peter Ellis: Please think through the theory you have proposed. Do you honestly believe that one million sq km of SIE existed in the Beaufort, Chukchi, and E. Siberian regions, and completely missed detection by the passive microwave products that entire time, then suddenly over 70% melted out in three days?

One or two days for some it, maybe, even probably. But 12 days, and not picked up at all by the PM products? That is a pretty strong claim.

Peter Ellis

Do you honestly believe that one million sq km of SIE existed in the Beaufort, Chukchi, and E. Siberian regions, and completely missed detection by the passive microwave products that entire time?

Why not? It happened for over half a million square kilometres of sea ice area in the Laptev and CA due to melt ponding in June. MASIE showed not a blip, while all the passive microwave sensors showed concentration dropping to 50% or so. So we have a benchmark for the effect of waterlogging the ice - apparent concentration is halved. Now, apply that to a huge area under the GAC that started at ~40%...

then suddenly over 70% melted out in three days?
I would say rather that there was a vast expanse at ~50% concentration, which got smashed and submerged by the GAC over the course of a couple of days. Likely there was some immediate melting, but not as much as you credit.

Submerging ice cuts the apparent concentration (measured by microwaves) by ~50%, that means that the PM-measured concentration dropped down to ~20% (= 50% x 50%), leading to the DMI graph dropping like a stone. They then went back and tweaked that, likely by playing with the algorithm parameters to help spot submerged ice.

Over the next few days, this dropped off 40/30/20/10% (real concentration). When it went below 30%, Bremen and the other sites using a 15% threshold stopped seeing it. Bremen noticeably before the others, because their algorithm is more aggressive and they regularly show the ice edge retreat before JAXA and NSIDC do.

That is a pretty strong claim.
Your claim is far stronger. You're claiming the equivalent area dropped from ~50% to <15% during the course of the GAC. That's thermodynamically far less feasible than dropping it a few percent that happens to push it over the 15% borderline.

Moreover, I'm quite happy to accept that cloud-affected regions will show an additional delay in MASIE - if they had ice before clouding over, that stays the case through "persistence" until the clouds part and uncover several days' worth of melt at once.

Note that if you look at the MASIE/IMS pictures, a huge chunk of ice apparently grew back last night in the northern Beaufort. That's not thermodynamically possible at this part of the season (and certainly doesn't correspond to anything at all in the Bremen record, whether time-delayed or no. However, rapid fluctuations of the apparent ice edge are very likely when looking at a vast sparse ice field that's hovering at the threshold of being counted for extent.

Finally, I implore you once again to look at the actual visible pictures of the area. Here it is, on the 19th. There is a great swirl of melting ice that's visible on IMS, but not on passive microwave sensors.

The resolution is 250m, so the IMS's 4km resolution corresponds to at 8x8 grid of MODIS pixels. 15% of an 8x8 grid is 10 pixels. Overlay an 8x8 grid on that "swirl" area, and check for yourself what proportion of grid cells contain at least 10 white (ice) pixels. I think you'll be surprised.

Peter Ellis

Also, remember that Bremen are using an algorithm design for AMSR-E on Windsat data, without any parameter optimisation. It's not surprising that their results are a bit out of sync even with the other microwave data, let alone visible wavelength data.

Peter Ellis

Poor Werther has now appeared on one of the other threads, repeating this incorrect conclusion.
*shrug* Maybe he found my arguments more convincing than yours?

Peter Ellis

Espen: Great find! That's consistent with IMS (and having now checked them, the Canadian Ice Service charts are consistent with IMS maps along their coast).

Paul Klemencic

Peter Ellis: I don't think your estimates are creditable, but lets go through your claims.

You claim that the existing PM SIE products (NSIDC, IJIS, Bremen) have massive measurement errors. That they mis-calculate the ice concentration routinely, and repetitively day-after-day for several weeks, by a factor of two. You claim that their calculated ice concentration of 25% is actually 50%, and 15% is actually 30%, and 10% is actually 20%. And your basis for this, is the melt pond area impact on SIA measurement??? I don't really get that at all, but we will use your measurement error estimate.

And you claim that there was over one million sq km of ice extent in the Arctic with concentrations less than 30%, that the PM products repeatedly missed for two weeks.

Hmmm, I don't think so. We have been observing the maps for some time, and although we can see variations over a period of several days or so, in general, errors and reversals exceeding 50k sq km are rare, and 100k even rarer. Nothing I have seen ever came close to a million reversal.

And the Bremen maps show ice down to 10% concentration (which could be as high as 20%, according to your claim), and the maps don't show this missing million sq km.

So this missing ice must be below 20% concentration (actual) to not show up on the Bremen maps (repeatedly missing day after day) as 10% concentration ice. And to be counted in MASIE the missing ice must be over 15% (actual).

This means that you are claiming, using your assumptions for measurement error, that over one million sq km of ice extent lasted two weeks, that had a maximum concentration of 20%, and of course, the minimum of 15%. If the million sq km was higher concentration, we would see some of it appear, at least periodically, on the Bremen map. And if lower, than it doesn't count as SIE.

This isn't a plausible answer.

Paul Klemencic

I looked at the MODIS image for August 19, and that is not 15% concentration ice. And if you step back day by day, there wasn't appreciable ice in that area into you get back to August 11, and finally August 10 to see significant ice in that area (which showed up on the Bremen maps, and was caught by the PM products).

If you step the other way, August 20 and August 21 still show the very scattered floes… no big change. Whereas MASIE reported SIE in that region collapsed. Not very convincing evidence; it actually proves the opposite point.


Paul K and Peter Ellis, a very interesting debate you're having here. I've followed it by lurking over quite a while (last year too), and I would have to say at the moment I find Peter Ellis' explanation more convincing. A few thoughts below:

If I read you right Paul, you're trying to say that there's a variable delay in data input into MASIE, and frankly that's the least plausible thing said by either of you. Maybe I have missed it, but can you think of a good reason why these data analyses would have a significantly variable input time at MASIE but not in other products? It's the very weakest part of your hypothesis that you've clearly put a lot of effort into, and so it's important to have robust support for that idea.

The delays, to my eyes, appear coincidentally largest with different weather patterns - big arctic storms leading to periods of longer delays. Is this correct? If so, this tends to lean the explanation towards a physical cause, be it variations in ice concentration, or variations in weather/cloud cover. I see no reason why data inputs would vary in step with Arctic weater - is that reasonable? It's plausible to me that there may be large areas of weak ice that first disappear in the UB microwave datasets, and then go sometime later from a system using multiple observation platforms. I would also expect the patterns of accelerations/slowdowns to be roughly in step, for example, a storm breaks up a large weak area causing it to disappear on UB, and subsequent gradual melt-out takes that large area eventually below the MASIE threshold a week or so later. Cloud cover, often associated with deep low pressure systems, may help exaggerate steps too.

I would suggest to you guys a fairly strong test of this. if the delay is physically caused by detection thresholds, cloud cover and different registering of storm-washed ice, then the differences between the datasets should be at their smallest under a regime of strong compaction (most extreme would be 2007). Under these conditions, both datasets should have the ice edge close to the same place. Under periods when there is very thin and broken ice, or very cloudy conditions, the "delay" difference should be largest.

Do either of you have data that can test your hypotheses under such conditions?

Paul Klemencic

Rob Dekker:
This is
the work I did last year to compare MASIE and although I thought it showed some very interesting similarities to the Bremen SIE, it wasn't conclusive for most people.

MASIE did however show three big century breaks in September; on September 6,7 and September 11, and hit the minimum around 4.34 M sq km. The other SIE products didn't show these big century breaks that late.

With the two week delay this year

Paul Klemencic

I didn't have that last comment completely finished before I posted it by mistake. Anyway, with a two week delay this year, and if the minimum isn't hit until mid-September, then MASIE could show a SIE minimum in October, long after the other SIE products turn up. This problem won't be able to remain hidden for much longer.

Paul Klemencic

skywatcher: look at the graph of various SIE from last year that I linked to.

And I am not proposing a "variable delay"… I am suggesting that the delay (lag) is getting worse with time. Last year I thought it was due to missing data days…. But now, I suspect a different problem with dating the observational data at the NIC, and then used as input to the MASIE project. I got caught up in other work last September, and couldn't keep the graph going.

The reason I think its a dating problem, is that the MASIE map is two days older than the data… the same difference in dates was apparent last August/September.

The storm damage is very useful to diagnosing the delay (lags), because the big drops stick out in the data. Unfortunately, last year I only had spotty Bremen data to compare with, and "averaged IJIS" data. This year we have one day NSIDC data, so we have another PM product to compare to. I haven't had the time to plot NSIDC data versus MASIE yet.


Paul, I'm afraid that graph is not convincing to me either ... is it not possible to get a more complete dataset (more than one month, and perhaps over the whole melting season)?

From what point are you suggesting teh "delay" is getting worse? Just last year to this year? Over several years? The reson I say this is that since 2009, the ice pack has been visibly thinning and becoming more spread out, especially at this stage of the melting season. So it's not a diagnostic test to say that the 'delay' has been progressively getting worse year-on-year, if the actual ice pack has been more spread out, and so more conducive to lags between the different observing systems. If you can demonstrate that a data delay has been present since 2007 or earlier, then that would be much more interesting, as the pack was much different in '07/'08 than the progressive rotting from '09 through to present.

Would you accept that it is plausible that two observing platforms with differing thresholds or measurement systems for determining ice/open water will show similar trends, with a lag between the two?

Please don't take my view to be dismissive of yours, but in order for our ideas to survive, they need to survive challenges to their logic.

Paul Klemencic

skywatcher: MASIE is a recent dataset, only initiated in 2010, and has suffered from lack of support. Currently, the response to the big SIE losses from the storm on August 6,7,8 showed up on MASIE on August 18, 19, and 20, showing a 12-day delay in dating the data. The MASIE map matches the Bremen (and CT) map from about 14 days earlier.

The combination of the graphs, matched to the events, matched to the Healy sailing north right through the area of the flash melt on August 22 just after the storm that caused it, and not showing ice where the MASIE map showed ice… the combination triggered my suspicions.

Then six days after the flash melt in the Chukchi and Beaufort…. bam! MASIE showed big regional drops in SIE in the Chukchi and Beaufort… The same thing has been happening time after time, but now with a much longer lag time.

Paul Klemencic

skywatcher asked:
Would you accept that it is plausible that two observing platforms with differing thresholds or measurement systems for determining ice/open water will show similar trends, with a lag between the two?

The problem is that the explanation of why MASIE, and in particular, the MASIE regional extent graphs, take this regional graph for the Chukchi , doesn't show SIE losses comparable to actual events, and other SIE measures, or MODIS imagery.

A huge storm moved into the Arctic on August 4 and 5, then on Stormy Monday August 6th hammered the ice pack, with huge SIE losses on the 6th, 7th, and 8th. Over the week starting August 4th, the storm took out 900k to 1000k sq km of SIE according to the PM systems. The Chukchi region was right in the center of the storm circulation.

Look at the MASIE regional graph. Where is the imprint of this massive storm? Can you see it in the week between August 4 and August 11? No. Not at all.

The damage began to show up until 12 days later. The same thing for the Beaufort. And for the E. Siberian, although a little less conclusive.

The explanation for this has become increasingly convoluted, to the point where it makes no sense at all (a million sq km of low concentration SIE hanging around for almost two weeks, to suddenly disappear in just a few days).

So to answer your question, it is not plausible that the two separate measurement platforms could show the same trend impact of this storm with a 12-day lag.


So hypothesis tests have to come form the last couple of years, fair enough! I've had a look at the Healy images for that date (best accessible here) - certainly interesting; unfortunately the Aug 2011 cruise report track map has a disabled link. If you're right, it's cetainly interesting. What's Peter's take on that?

Paul, did you miss my question in the 3rd paragraph of my last comment? What do you think?

Paul Klemencic

skywatcher: The Healy photos start on page 8 with the conning tower shots of the storm (Healy was right in the middle of it). The Latitude can be made out at the top of the picture, if you zoom in on the image. The Healy was around 71N at the time, and should have been running into the ice. As the Healy sailed north, you will notice the big ice wasn't hit until around 78N and the solid pack around 80N.

Here is a shot from August 23 at almost 78N, whereas MASIE showed the ice pack extending down to 72N… The photo should be filled with ice.

Espen Olsen

I dont know what happened to the DMI domino page:

But here is a new ice report from DMI who expect a new record this season. Ice thickness within 100 kms from the Pole is around 1 meter, much less than earlier years!

Use Google for translation from Danish:



Espen, I'm posting the DMI domino post this morning. I want a bit more space between posts.

Peter Ellis

Paul: That Healy photo is from 2011. Here is 2012.

Healy has been showing scattered sparse ice in the areas called as ice by IMS.

Steve Bloom

Looks like the next four days or so should be enough to do for all of them, NORSEX extent looking likely to be the final domino.

Peter Ellis

Oh hang on, you're referring to last year's flash melt. Will check and report back. This year at least, Healy appears to me to be more in sync with IMS.

(re: last year - Bremen was once again ahead of even the other passive microwave data in calling areas as ice free. The one that particularly sticks in the mind is the Northwest Passage, where IJIS continues seeing ice blocking it for weeks longer than Bremen. The Canadian Ice Service didn't call it open when Bremen did.)


Paul, there are also photos from Healy from last year showing ice further south than 78N.

For example http://icefloe.net/Aloftcon_Photos/albums/2011/20110822-1301.jpeg

By the way, isn't it likely they try to avoid ice if they are going from one point to another? They travel much faster on open water than through ice.

Espen Olsen

Downill 1 ,

That amount of ice wont do any difference to speed, with this vessels.

Peter Ellis

Are the MASIE images archived anywhere? Can't find them to check. The IMS image from that date is here.

Although it looks to have the ice edge substantially south of Bremen for the same date, there's a definite loose area around where the Healy was. Remember that the Healy's field of view is substantially less than a single pixel of that map. You can only really judge ice coverage in the area immediately around the ship: around ~500m squared or so, so about four times the size of a MODIS pixel, 1/64 the size of a MASIE pixel, and 1/150 the size of a Bremen pixel. And remember that currently Bremen are artificially oversampling to get that resolution - the actual sensor they're using has a 13.2 by 15.5 km footprint.

When I look at the Healy archive, I see that they first encountered ice around a day earlier and over three degrees further south than the picture you linked.

Here are the entire page from that period so people can make their own judgement.
http://icefloe.net/Aloftcon_Photos/index.php?album=2011&page=94 (Aug 22nd, Healy meets ice at 74 degrees North)

Going back through pages 93, 92 etc. shows the ice getting more extensive as the Healy moves North.

Paul: I think perhaps you're getting misled by what you expect 15% concentration to look like on Healy photos and on the MODIS images. If you see a single 80-100%-covered photo (like the third link in this post), then in an area of 15% ice, you expect around ~5 time as many completely ice-free photos.

Did you actually overlay a grid and count ice pixels in the MODIS image I linked? Come to think of it, it should be a 16x16 grid (4km resolution for MASIE, 250m for MODIS). In that swirly area of broken floes, there really are not many 16x16 clear areas. This really isn't negotiable. We can have different opinions on interpretation, but data is data.

To be honest, I think much of the argument would just disappear if MASIE / IMS showed concentration levels like Bremen / IJIS etc, rather than just the boundary for extent.

Peter Ellis

(Want to know the really frustrating thing? I'm having this same flipping argument at The Other Place, only I'm on the other side. Steven Goddard's pointing at the IMS map and the loose swirls on MODIS and claiming they're high-coverage thick multi-year ice, and I'm patiently pointing out that it's highly-decayed ice on its last legs that's about to disappear, and which has almost no impact on ice volume. About the only people who'd care about it are people in non-ice-hardened surface ships: which is the audience the IMS product is prepared for! In both cases it boils down to explaining that just because different sensors show different things, it doesn't mean one is wrong, so long as you understand what each of them can see.)

Espen Olsen

This is a re post:
Here is the new ice report from DMI who expect a new record this season. Ice thickness within 100 kms from the Pole is around 1 meter, much less than earlier years!

Use Google for translation from Danish:


Seke Rob
To be honest, I think much of the argument would just disappear if MASIE / IMS showed concentration levels like Bremen / IJIS etc, rather than just the boundary for extent.
Think this captures the situation succinctly for MASIE... enter the zone and you'll need an Arctic rated vessel to get through.

Peter Ellis asked me to post this (too many links, so TypePad didn't accept):


Here are some charts from the Canadian Ice Service, based on microwave radar (active rather than passive), which is much more sensitive to sparse ice. It tracks the evolution of the ice in the Chukchi from Aug 6th through to today. The area covered is the eastern side of the region between the "island" and the main pack.

Concentrate on the area from 74-78 north and 140-170 west. That's north and slightly east from the Bering strait, and Bremen reckons this disappeared almost immediately the storm hit, and certainly by the 10th of the month.

Here it is at
around 40-60% on the 6th Aug before the GAC hits in earnest:

And at 20% on the 7th August:

Bounces back to 30% on the 8th (presumably even radar gets confused by that big a storm):

Ice swirls around and clumps up as high as 50% over the next 6 days (9th-14th Aug), as some more concentrated stuff moves eastward from the "island" proper:

It melts out further over the next week (15th-21st) and is now all at 20% and about to disappear for good. Note that the graph key overlay moves from left to right on the 17th, but the actual region of coverage stays the same:

This is EXACTLY what I said was going on based on the MODIS images. I think at the point the ball is firmly in Paul's court to provide some actual evidence of a processing delay that isn't just down to wiggle-matching graphs (a process fraught with error) or eyeballing maps.

Final thought: I take Paul's point that it seems odd for it to drop suddenly during the storm - by ~20-30% according to the Canadian ice service - and then for the last few percent to hold on much longer. Remember that this area is a mix of thin first-year ice formed last year and thick multi-year ice advected over the winter. Possibly what's going on here is that the first-year stuff melted during the storm, and left the embedded nuggets of rotted MYI to melt more slowly over the next couple of weeks. Or maybe the remaining chunks are the remnants of ice ridges between the floes?)


Peter Ellis

Thanks Neven. Alas, it seems Typepad will not even bow to its legitimate master and has mangled the links. Hopefully it will be clear enough how to get to each of them.

For those wanting to check it out, you need to copy-and-paste each link, starting from "http://dynaweb.cis.ec.gc.ca/.." and continuing as far as "...&imageId=nnnnn"

Seke Rob

Whilst, CT SIA pushes on:

2012.6329 -2.2681708 2.8442657 5.1124363
2012.6356 -2.3587317 2.7419870 5.1007185

Artful Dodger

Hi Peter & Neven:

I think Typepad mangles your html code because your quoted URL includes a 2nd "http" string. But using this simplified html to point directly to the image should work:

<a href="http://ice-glaces.ec.gc.ca/www_archive/AOI_01/Charts/sc_a01_20120806_WIS45CT.gif">Chukchi Sea - WIS45CT - 2012/08/06</a>

Applying that html code, we get this result:

Daily Ice Chart color WMO CT - Chukchi Sea - WIS45CT - 2012/08/06.

Cheers, and have fun!

Peter Ellis

Thanks Lodger! Rather than spam the entire post again, I'll just point out the base URL in a hopefully-human-legible form


If you then edit the "20120806" part to be 07, 08... etc., you can get the full set of pictures through to today.


‘…left the embedded nuggets of rotted MYI to melt more slowly over the next couple of weeks. Or maybe the remaining chunks are the remnants of ice ridges between the floes…’
Peter, this is exactly what I had in mind. When you mentioned the small volume involved, were you thinking of max. 50 km3, like me?

Paul Klemencic

Peter Ellis:

It might help you to follow the process I followed to reach the conclusions I have reached. The big change in my view was caused by carefully studying the effects of the August 22, 2011 storm in the Beaufort/Chukchi where the Healy was at the time.

To start, here I caution about reading too much about the ice pack edge in the Bremen preliminary map (Bremen used to put a preliminary map that showed the side near the Fram fairly accurately, but the side facing Alaska, not as accurately; after the final map came out, the Alaska side of the pack would usually show significant revisions compare to the earlier map, likely due to more satellite passes.)
August 21, 2011 at 20:36

Then the Bremen preliminary map came out (note the comment from Chris Biscan directly before mine) :
August 22, 2011 at 19:41

Here I agree with Neven, that melting out this much extent in one region in one day is impossible (little did I know what was coming.):
August 22, 2011 at 20:16

Here I begin discussing heat transfer rate:
August 22, 2011 at 21:30

And here I put some actual heat transfer rates on the table, and end by saying "On this issue, call me a skeptic."
August 23, 2011 at 03:01

But then the final Bremen map came out, and showed big pullback of the ice pack edge:
August 23, 2011 at 07:07

Here I discuss the next day's map, expecting to see a lot of the missing extent fill back in (didn't happen; some did the next day, but nowhere near the 100 km rebound in pack edge). The photos from the Healy shows very little ice in some areas that even the Bremen map shows ice, let alone the MASIE map. At this point I am beginning to believe a lot of the overnight loss was real… and that we had really observed events that demonstrated the very high heat transfer rates that I thought unlikely.
August 24, 2011 at 01:01

Here I begin to get suspicious, and a couple of comments down, I compare the MASIE map extent in the Chukchi with the Bremen map, and end by saying "Confusing."
August 26, 2011 at 21:10

And then a couple of comments later, I admit I don't understand why the maps show substantially different ice pack edges. And we had the Healy photos showing the main ice pack edge was around 78N, somewhat consistent with the Bremen map but far to the north of the MASIE pack edge.

A couple of confusing hours later, I first express the hypothesis that the MASIE data has been delayed by a week:
August 26, 2011 at 23:00

And a couple of comments below that admit:
"Golly guys, I am going to look like an idiot if I am wrong on this, but here goes:"
…and then I suggest that MASIE, if dated correctly, should (and does) compare closely to the Bremen SIE data, but not the IJIS data.

A couple of comments later, I forecast that we should see a big drop in the MASIE reported extent of over 100k centered in the Beaufort/Chukchi by August 31. And later I revised this to September 1st. I was right the first time; on August 31 MASIE recorded a 135k sq km drop, and just as interesting, reported rebounds of 58k and 19k the next two days. These reports were driven by regional drops in the Beaufort and Chukchi, whereas earlier big extent drops in August were attributed mostly to other regions.

Since the storm took out the ice in those two regions, the August 31st report was the only report that matched. And the rebound the next two days was centered in the Beaufort, which seemed to match the Bremen map results.

Since then, I am convinced that:
- dispersed floes on the open sea melt at much faster rates than in a higher concentration pack;
- that storms can take out a lot of ice by boosting the heat transfer rate high enough to take out 100 cm of ice thickness literally overnight;
- and that the MASIE maps have a dating issue, which if corrected, show the MASIE SIE matches the other small gridded product, the Bremen SIE, very closely.

A closing comment from last year:
August 27, 2011 at 22:47

Lucia, I put a lot of time and investigative time into my comments.

You want the simple answer: No, the Bremen numbers aren't available all the time, but one commenter here who knows what he is talking about, has access to the data, and shares occasionally.

But you are really out in the corn field on this one. You are busily crunching numbers with no idea of what they mean, how they were measured, how they relate to the physical system, or even when the measurement was taken. YOU should read every comment of mine on this thread, starting from the top. When you find the comment that starts, "OK guys, if I am wrong on this, I am going to look like an idiot." you should really start reading carefully.

You could learn a lot of how a real skeptic works, instead of a fake skeptic. I am not afraid to admit I am wrong, and correct my mistakes. Real skeptics MUST have this attitude to be effective.

Posted by: Paul Klemencic | August 27, 2011 at 22:47

Paul Klemencic

Peter Ellis: Try focusing on just one thing to start with. We know that major storms have an impact on the ice pack. We have the August 22, 2011 storm, and Neven has found evidence of a major storm on the Arctic ice pack from a 2005 storm (on one of the other threads), and clearly the GAC-2012 had an impact.

Then why doesn't any of the MASIE regional graphs for the Beaufort, Chukchi, or E. Siberian show the impact of this storm during the megaweek SIE loss of August 4 to August 11. These regional graphs actually show significantly less extent loss for that week than in a normal melt year?

This is the HUGE glaring "elephant in the room" problem with the MASIE reports.

Ned Ward

Paul, you're comparing apples and oranges, and interpreting the difference in color as proof that the orange-seller is selling you unripe fruit.

The different sea ice products are totally different. It's not that one is "wrong" and one is "right". It's also not the case that a difference in reported extent implies that one is lagging behind the other.

I've worked in remote sensing -- processing of earth observation satellite image data -- for over 20 years. What you're trying to do is natural, for an enthusiastic amateur, but totally wrongheaded. You cannot diagnose the kind of methodological problem you are trying to diagnose based on the information you're working with.

There will be inherent and large differences in the reported ice extent in different products based on the type of sensor, and based on the processing methodology, and even based on the map coordinate system used.

If you just compare the MASIE and Bremen output maps, they use slightly different map projections -- both are polar stereographic projections, but the Bremen maps have a latitute of true scale at 70 N, while the MASIE maps have that latitude at 60 N. Reprojecting the MASIE maps into the coordinate system used by Bremen -- with no other changes at all -- changes the reported MASIE ice extent by over 350,000 km2 (approximately 8% of the total).

That's just the trivial difference associated with the choice of map projection. The differences associated with the choices of sensor and analytical methods could easily result in much larger differences in the resulting ice extent.

The kinds of evidence you present for your claims about a 5 (?) day delay last year, or a two-week delay (?) this year, are just totally unconvincing to anyone who actually works in this field. The potential for confirmation bias is just way too high. In essence, you've come up with a fixed idea and now you're fitting the evidence to match your idea.

Last year, when you were proposing a much smaller time lag, I pointed out that the correlation coefficient between the two daily time series gave no evidence to support your position. Until you can either point to a specific step in the NIC/MASIE processing sequence that would create a multi-day time lag, or a convincing, objective, quantitative indicator of a time lag (e.g., based on the correlation coefficient) all you're doing is casting unjustified aspersions on the hardworking analysts responsible for the remote sensing sea ice products.

That's the kind of thing I'd expect to see at WUWT, not here.


Paul, what's your comment on the Canadian Ice Service charts?

Peter Ellis

Reprojecting the MASIE maps into the coordinate system used by Bremen -- with no other changes at all -- changes the reported MASIE ice extent by over 350,000 km2 (approximately 8% of the total).

You doubtless know more than I, but I'd be surprised at this. My understanding is that the polar stereographic views are for display only, and that internally they work with data in forms that correctly preserve area. What I think is more likely is that the projection will change the apparent area of the display image by 8%, so that people counting directly from the image will be misled.

Paul Klemencic

Downil1: I don't know how the Canadian Ice Service chart information is collected, updated, and used. Why don't you tell me how that is done?

What does 6/10 ice mean? Areas of 6/10 ice, or patches, or average? Is the data collected by pixel count?

I am assuming the ice charts are for navigation purposes, and swaths of thicker ice is what they might be interested in.

But I am ready to learn… How does the ice chart info compare to PM ice extent measurement?


Espen Olsen


The Canadian Ice Service charts, is probably like DMIs Ice Servvice Chart, and when it comes to ice free areas around Greenland there is "no ice free area", as long there is glaciers pulsing ice out fjords!

Peter Ellis

Paul: The manual is here:


Essentially it's a manual process done by trained observers, using predominantly visual imagery backed up with ship reports, radar imagery and microwave imagery where applicable. Pretty much exactly the same type of processing pipeline as MASIE/IMS, in fact.

The edges of the various regions ("polygons") are hand-drawn by the analysts. The egg code relating to each region describes the average conditions across the region as a whole. There are strict rules as to when a polygon boundary should be drawn, in particular:

1) To distinguish between concentration classes as follows:

* Open water/bergy water: < 1 tenth
* Very open drift: 1 to 3 tenths
* Open drift: 4 to 6 tenths
* Close pack: 7 to 8 tenths
* Very close pack: 9 to 9+ tenths
* Compact or consolidated: 10 tenths

2) To distinguish between ice age areas - so a given area of 10-30% ice could be further subdivided into a area of 25% first-year ice and an area of 25% multi-year ice)

* No old ice
* Trace of old ice
* 1 - 3/10
* 4 - 6/10
* 7 - 8/10
* 9 - 9+/10

Note that while the borders are drawn based on concentration classes as described above, the egg codes are generally more precise - if it falls in the "very open drift" class, the analyst will decide whether to code it as 10%, 20% or 30%.

You may at this point be tempted to dismiss it as subjective - I wouldn't do that, as human judgement is far superior to automated sensors in this kind of scenario. Moreover, this misses the point that the IMS product is based on exactly the same trained manual calling - the only difference being that the IMS analysts simply call "> 15%" or "< 15%" rather than breaking it down further. That's why I'm not surprised to see quite a close agreement between the two metrics.

For an even more cautious estimate, check out the charts from the National Weather Service in Alaska. They're still calling large parts of the region as 60-90% coverage! They have, however, downgraded it to "first year / young" ice - a classification that at this time of year seems to be based on thickness rather than actual ice age. So what this chart tells us is that the area's covered with floating shrapnel less than about a foot thick.

Paul Klemencic

Ned Ward: I checked my computer, and found the MASIE spreadsheet from last September. MASIE hit its minimum on September 14 at 4.303 M, lower than the IJIS minimum 2-day average reported 5 days earlier on September 9 at 4.527 M by 224k sq km.

Your theory is interesting. Does your projection error of 350k apply in mid-September? If so, then MASIE actually shows a minimum of about 574k lower than IJIS.

Is this your interpretation?

Paul Klemencic

Oh, and BTW, at the minimum on September 14, MASIE showed 4.303 M and on that date IJIS reported an extent 352k higher than MASIE.

Explanations are welcome.

Ned Ward

Peter Ellis writes: My understanding is that the polar stereographic views are for display only, and that internally they work with data in forms that correctly preserve area. What I think is more likely is that the projection will change the apparent area of the display image by 8%, so that people counting directly from the image will be misled.

Maybe, maybe not. There's really no way to know without seeing the details of the data handling process. Alternatively, they could be doing all the analysis in a polar stereographic coordinate system but adjust the numbers to account for the area of each grid cell, when adding them up to get the total area.

All we really know is that MASIE is based on a 4km gridded product from NIC, so somewhere along the line they're resampling the data from the various sensors' swath-oriented geometry (in which the ground dimensions of each pixel vary as a function of its distance from the satellite's nadir line) into some projected coordinate system.

There are lots of ways one could handle this. If two data providers don't deal with this in the same way, it can introduce a difference in the reported areas.

I'm not saying this *is* happening here -- my whole point is that speculating about technical details of the processing systems without adequate information is a fool's errand.

Last year, Paul went down a completely pointless path of trying to reconstruct daily IJIS numbers based on fiddling with the two day averages. The problem was that what Paul thought of as a "two day average" (calculating extent on Day 1 and Day 2, then averaging them) was totally different from the way IJIS actually handles the data (where each day's extent measurement is based on processing all image swaths acquired within a two-day window).

The bottom line is that it's very difficult to reverse-engineer these kinds of products and diagnose a problem in the processing sequence, even for someone with a lot of expertise working in this area.

Ned Ward

Paul writes: Your theory is interesting. Does your projection error of 350k apply in mid-September? If so, then MASIE actually shows a minimum of about 574k lower than IJIS.

Is this your interpretation?

No, no, no. Where the heck did you get that from?

I was comparing the nominal area from two different coordinate systems. We don't know that MASIE actually calculates area based on these projected grids, and even if they do, I didn't say anything about which direction the change went when one switched between the coordinate systems.

You're leaping to conclusions left and right. Just *stop*.

Ned Ward

The MASIE numbers are very noisy. The fact that the date of the 2011 minimum was five days later in MASIE than in IJIS isn't really meaningful.

The 2011 maximum occurred on the exact same day in both data sets. No lag at all! But that's mostly just coincidence.

You cannot match up individual dates' numbers from MASIE and IJIS, or MASIE and Bremen, and try to draw conclusions about time lags.

Paul Klemencic

Ned: Please bear with me a bit.

First, I apologize for the snarky comment about the 350k projection, I shouldn't have put that up.

From when I first came to this site, I have disagreed with many who have tried to forecast the minimum SIE using trend analysis. From my work on heat transfer in these kind of heat transport problems, the ice pack wasn't going gradually melt away into a big ocean of slush. The heat transfer rate picks up substantially once the pack is weakened, fractured, and mobile. The heat transferred from the seawater during the bottom melt season accelerates, and the pack weakens quickly. If combined with a persistent wind pattern, or rough weather, then the pack collapses (basically, what we are seeing this year).

In order to forecast this, understanding how the regions of lower latitude "guard" pack that protects the Central Arctic Basin are faring is very important. The guys on this site showed me the MASIE regional data, and I immediately fell in love with it. Being able to track regional ice extent, allows comparison to weather events, pack movement, wind patterns, and even regional insolation changes. Unfortunately the MASIE regional graphs didn't match well with these factors. But yet MASIE showed large drops from time to time, that seemed due to the factors I listed above.

What was going on? This is what motivated me to study MASIE regional data.

Second, last year I felt the IJIS reported data didn't match pack observations very well. But MASIE agreed with IJIS better on the total extent than the other PM products, seemed to confirm IJIS. The Bremen map, and Bremen SIE seemed like the odd man out.

But after I realized that MASIE seemed to be following the Bremen reports and map (with the lag), then I could make the outrageous prediction that the minimum on the MASIE would match the Bremen SIE minimum, and be below the IJIS minimum. This is the comment beginning, "Guys, if I am wrong on this, I am going to look like an idiot." At the time (August 25 data), MASIE was about 500k sq km higher than Bremen, and 200-300k above IJIS. It was a gutsy call that MASIE would blow past IJIS and end up with a minimum near Bremen SIE. There wasn't a lot of time to make up more than 500k sq km difference.

But that's what happened. The 2011 MASIE minimum was within 2% of the Bremen minimum.

You focused too much on the lag issue, and not enough on my point that MASIE should match Bremen near the minimum, since the smaller grids should capture the minimum better.

This year I believe it will happen again. Up until four days ago, MASIE was about 1.5 M sq km higher than Bremen, and still is over 1.0 M higher. So in a way, my forecast this year is even more gutsy. But I believe that MASIE will close the gap, fall below IJIS again, and record a minimum SIE very close to the Bremen minimum SIE. And I say this, knowing that is already August 22, and that forecasting this rapid a decline in SIE seems rash and ill-advised.

I was correct last year… so lets see what happens.

Paul Klemencic

Yes, the MASIE numbers are noisy, but they also match Bremen map observations well, once the lag is taken into account. And since we have the MASIE regional data to look at, then we can match the regional declines to weather events that affect only one or two regions. Thats how I knew that eventually the Beaufort, Chukchi, and E. Siberian regional extents were going to record serious declines associated with GAC-2012.

And that's what happened.

Now what if we have a prolonged melt year such that Bremen / IJIS hit minimums after September 18th this year? Then if I am correct, then MASIE will report the minimum extent for an October date. Then this could this lag could turn into a real problem. The skeptics will be all over the MASIE reported minimum, like rats on cheese.

They can say: "The ice is still melting, even a week after the sun went down? " And the temperature is minus 10 degrees?

Then this issue could turn into a real WUWT imbroglio. Better to get this resolved, and cleaned up if necessary, before the public is really aware of it.

Peter Ellis

The Bremen map, and Bremen SIE seemed like the odd man out.
It always has been. It's a lot more aggressive at filtering out thin ice.

...my point that MASIE should match Bremen near the minimum, since the smaller grids should capture the minimum better.
This factor should be totally absent this year, as Bremen is using a lower resolution satellite and simply projecting the data onto a higher resolution grid. You're arguing against yourself.

Note that if (as I suspect), the main confounding factor is water on top of the ice, then we automatically have an explanation for why they agree better at the time of the actual minimum. This is the point at which re-freeze has started, and there is thus very little remaining water on top of the ice.

Yes, the MASIE numbers are noisy, but they also match Bremen map observations well, once the lag is taken into account.
No they flipping well don't, and if you'd ever tried to actually align one on top of the other rather than eyeballing it, you'd realise it. Even bearing in mind that it's a different projection, I tried to superimpose IMS day 20 and Bremen day 6 and it's just not good. I mean, very not good. Steven Goddard (spit) aligned IMS to NSIDC for the same day - i.e. no delay - and got a MUCH better alignment, allowing for the fact that IMS picks up loose fringe ice better.

Just look at the IMS animation (and remember that IMS = MASIE)! The ice edge is flickering inwards and outwards all over the place, which Bremen just does not do. Look for example at the way the ice edge surges southwards past Severnaya Zemlya between August 4th and 7th. Did Bremen show a large area of ice gain in this region some time in the second half of July? No it bleeding well didn't. The only way you can possibly be claiming these maps match is if you tunnel vision the areas you think are the same and ignore the rest of the pack.

I'm done with this argument.


I mostly agree with Peter.

The weekly maps of the NIC are interesting because they are very similar to those of IMS, but with a more detailed concentration of each area.


Peter Ellis

Whoa, that's a very conservative picture, and it really is all over the shop. Total extent and area ping-ponging up and down all over the place, likewise the boundary between 8/10 and the marginal zone. I'm guessing that chart's is made for shipping where even the hint of ice is a no-no. Weekly maps seem more consistent, but again very conservative even relative to IMS.

Artful Dodger

From the NIC FAQ:

When was National Ice Center established?

The National Ice Center was established in 1995, when the U.S. Coast Guard joined the Navy/National Oceanic and Atmospheric Administration (NOAA) Joint Ice Center.

Why are you called the National Ice Center / Naval Ice Center?

The National Ice Center was established as an interagency comprised of three components: the Naval Ice Center, NOAA, and the U.S. Coast Guard.

Does NIC do any research on Global Warming?

No. The primary mission of the NIC is to provide strategic, tactical, and operational ice products and services to meet requirements of U.S. national interests and U.S. government agencies. Our products, available on the World Wide Web, are used by outside organizations to derive or interpret information of scientific value.

Seke Rob

The 80>100 concentration zonage is educating. The Russian icebreakers must be working real hard to get through leading those cargo vessels. Great find, Diablobanquisa

Paul Klemencic

I agree. Great site. I have been exploring it, and it has already yielded some very interesting information. And not all of it supporting either side of my discussion differences with Peter.

The Beaufort/Chuckchi storm of August 22, 2011 shows up late on August 28. But the GAC-2012 shows up on time, but with some very interesting ice pack movements.

I need time to study and digest this. I want to compare to IMS and MASIE maps and data a bit. It seems this site may have been updated to fix a problem…. too soon for me to know.

Paul Klemencic

I suspect all of you have seen the 2007 melt season movie, but I will put a link up to my earlier comment on it.

For anyone who hasn't watched the impact of storms on the ice pack, and melt process, watching the movie closely can yield some fairly interesting observations. If you click and hold the mouse key down over the cursor at the bottom, you can reverse or forward the movie at will. Two of the storms are discussed in my comment. Please note the June 9th storm in the Beaufort scattered the ice pack, but it seemed come back together again.

Anyway, I thought I'd put it up. It makes an interesting reference point for observing ice pack movements.


Ned, your post at August 22, 2012 at 15:12 covers a lot of my concerns with Paul's idea. Paul, you didn't answer my question yesterday - do you think it is plausible for two datasets, with differing thresholds for ice detection, to show a similar extent loss curve (down to the 'kinks' caused by large melt events), but with a lag between the two? If not, why not?

These conditions, with lots of thin and spread out ice, are exactly the conditions under which you would expect the different monitoring platforms to record large discrepancies.

I'm convinced more than ever that all we are seeing is a difference in observation capacity and observation methodology between multiple datasets, observing the same system in slightly different ways. And of course the end result is scary however we look at it!

Paul Klemencic

skywatcher: I answered your question (before either of your two recent requests to answer the question). I even quoted your question right at the beginning of the comment.

If you are having trouble seeing this comment, I can re-enter it. Let me know, I aim to please.

Peter Ellis

Relevant to this discussion:

Money quote:
"Compared with the Canadian regional ice charts, the NASA Team algorithm underestimates the total ice-covered area by 20.4% to 33.5% during ice melt in the summer and by 7.6% to 43.5% during ice growth in the late fall."

As to why the minima themselves coincide reasonably well, it seems likely that there may be a "sweet spot" where the rotten/melting ice has frozen over and is being detected properly by PM, but there is not yet extensive coverage of nilas/young ice.

Peter Ellis

From your reply to Skywatcher:

The problem is that the explanation of why MASIE, and in particular, the MASIE regional extent graphs, take this regional graph for the Chukchi , doesn't show SIE losses comparable to actual events, and other SIE measures, or MODIS imagery.

In order:

i) "comparable to events"
The actual events don't produce as much melting as you say, instead they have an additional effect which is to swamp ice with water, hiding it from passive microwave imagery.

ii) and other SIE measures
See (i). SIE measures are inconsistent because they measure different things

iii) or MODIS imagery.
I flat-out disagree. The MODIS images show ice in places where IMS shows ice, and Bremen does not. I don't think you realise just how sparse 15-20% coverage actually looks, and I certainly don't think you've bothered to try and analyse the MODIS images on a pixel-by-pixel basis. Remember that in order to generate an open water pixel in IMS, you need a 16x16 square of MODIS pixels without any ice in (or rather, less than 39 ice-containing pixels within the 16x16 grid.

Ned Ward

OK, I wanted to try to assess Paul's hypothesis (of a multi-day time lag in the MASIE data) quantitatively and objectively.

So I borrowed crandles' spreadsheet of MASIE extent data (thanks for providing that!), and compared it to the IJIS extent data. I looked at the data from 1 January to 4 August (the last date included in the version of the spreadsheet that I had). A few days are missing in the MASIE record, so I infilled them using linear interpolation. I then fit a LOESS model (a=0.3) to each data set to detrend them, and did a correlation analysis of the detrended residuals at a variety of lags.

The highest correlation between IJIS and MASIE is at a time lag of ... 0 days. The correlation drops off rapidly with increasing length of the time lag.

Here are the correlation results, as a function of time lag in the MASIE data (negative lags mean MASIE is ahead of IJIS, positive lags mean it's behind IJIS):

Lag: -3 days: 0.286
Lag: -2 days: 0.358
Lag: -1 days: 0.476
Lag: 0 days: 0.555
Lag: 1 days: 0.537
Lag: 2 days: 0.497
Lag: 3 days: 0.461
Lag: 4 days: 0.392
Lag: 5 days: 0.259
Lag: 6 days: 0.112
Lag: 7 days: 0.048
Lag: 8 days: 0.03
Lag: 9 days: 0.023
Lag: 10 days: -0.013
Lag: 11 days: -0.046
Lag: 12 days: -0.011
Lag: 13 days: 0.05
Lag: 14 days: 0.09

I think this suggests that it's highly unlikely that there is a 10-day or 5-day time lag in the MASIE data.

Seke Rob

Big wink, this picture "leads" the eye heavily, if one sees the offset of that big uptick between MASIE (22nd) and DMI after GAC'12 subsided:


Paul Klemencic

Seke Rob: Yes, the "classic signature" response of sea ice extent to a large storm. The initial big drop, as ice is dispersed and pushed around, increasing the heat transport into the ice floe/block area, and the increased heat transfer to the ice by water disturbing the normal stagnant colder water directly next to the ice, and disturbing the protective stagnant film layer of water (laminar layer) that slows the heat transfer. And of course the floes break up, and are submerged, increasing the surface area for heat transfer. As pointed out in some of my comments from last August linked to earlier in this thread, the amount of heat transfer to melt out 100k sq km in one or two regions is enormous. When an event like that occurs, there MUST be a causal factor that increased the heat transfer by a factor of 10x, and probably 30-50x, normal heat transfer rate.

So for a one day storm, the SIE falls like a rock on day 1, and maybe into day 2, but after the storm abates, many of the dispersed and flooded floes re-emerge and collect together, or re-flash as Neven puts it, to show up in the count again. Typically the rebound is 20-50% of the melt.

In this case the rebound in the MASIE of 279k today, after dropping 731k during the storm shows a rebound of about 38%. This is the kind of SIE response one would expect from a major storm lasting several days.

But wait! Where was the storm? We didn't see a major storm in the regions showing the drop and rebound in SIE. So What caused the massive increase in heat transfer to melt out a net of 452k so quickly?

Peter and Ned are completely ignoring this, and offer only that the measurements are noisy and vary that much. I don't buy it. The amount of heat to cause this kind of SIE loss requires a spectacular weather event.

And the evidence should be even clearer in the regional graphs?
Lets check…
Beaufort region.


and the E. Siberian all show the classic storm signature response.

As I have pointed out over and over, the regional SIE data response show this was the GAC-2012 storm effect showing up in the data, with a lag of 12 days! It is impossible to transfer enough heat to get this kind of SIE loss, without weather systems playing a key role.

Peter Ellis

But there was no uptick in the Bremen data. What's your hypothesis now, that some time in the last few days, the MASIE analysts decided to start shadowing DMI instead of Bremen, and increase the lag to 15 days instead of 14? Any more variables you'd like to add?

Paul Klemencic

And today the detachment almost showed up in the MASIE map, showing an event that occurred on Sunday. Tomorrow the detachment should be complete.

MASIE is about 1.0M behind the SIE extent, and will need to turn on the afterburners to catch the other.

Peter Ellis

The Canadian Archipelago also showed an uptick yesterday - quite a long way away from the GAC.

If this "uptick" is always a storm signature, how do you explain the Laptev graph from 4th to 7th August? Look at the ice edge surging out past Severnaya Zemlya on the IMS archive. Was there another GAC on 21st July around the Kara/Laptev border, which magically escaped detection by every weather satellite and every passive microwave satellite, only to show up in IMS 14 days later for an unknown region?


If your evidence from the maps is so strong: present it. Make an overlay or side-by-side animation showing Bremen and IMS. Not just for selected regions on selected days, but the full time period and the full picture. I've tried for a couple of days, and it just does not fit your hypothesis. Not even close.

Peter Ellis

That should be "... for an unknown reason", of course!

Artful Dodger

Hi Ned.

Good Job. About time some one computed the correlations. They also suggests that IJIS uses an average of Day 0 and Day 1 data (the well known 2-day average ;^)

Lag: 0 days: 0.555
Lag: 1 days: 0.537


Paul Klemencic

Peter Ellis: You are starting to twist and nitpick the data… not a good sign. You focused on the Canadian Archipelago in the MASIE data, and ignored the other regions.

Lets look at the whole picture, shall we?
The biggest gain was not in the CA, it was in:
E. Siberian +98k
Which with the other regions most affected by the storm, Beaufort +36k, Chukchi +36k, amounted to +170k of the +279k.

The other big regional gains came in the Greenland +50k and your selected favorite datapoint of the day, the Canadian Archipelago +43k.

Lets take a good look at the archipelago reports. Two days before the CA had 307131.25 k, and the next day… whoa, what's this? Exactly the same number, 307131.25. And now the latest report, shows a gain of +43k. Looking at the records, there were four days with 306k to 307k, then the jump by 43k.

Lets go back and look at the Bremen maps and examine the CA during the storm days of August 5-August 10.

What do we see?
Prior to the storm only the mouth of the McClure strait is blocked, and this continues until some ice moves from the Beaufort into the strait on August 7, with continued big moves into the strait on August 8th and 9th. By August 10, some portions of ice pack extended almost 200km into the Archipelago. And elsewhere in the CA, some ice reappeared on the Bremen map, as the massive storm cold front blanketed the region with snow.

Clearly the uptick in the CA data reported yesterday is consistent the storm days impact on the CA.

I don't have the time to chase down every red herring you throw at me. If you were interested in the CA MASIE report for this region, why didn't you check it out? I just can't investigate every little dimple in the data.

I am focusing on the enormous heat transport during the GAC-2012, and the high heat transfer rates necessary to wipe out approximately 400k sq km of mostly 70%+ concentration meter plus thick ice, and another 100k of 30% + concentration ice, primarily in three regions, and in just a few days. These events don't happen very often… in fact, never before in August during the satellite era.

And the MASIE data for these regions for the days this incredible event hit the Arctic, shows no sign of it! How is that possible?

You have been trying for days now, to come up with a plausible explanation, and you've failed. You need re-evaluate your position.

Ned Ward

Paul, the correlation coefficient between MASIE and IJIS is maximized at a time lag of 0 to 1 days.

At the absurdly long time lag you're proposing, the correlation coefficient is near zero.

This looks like pretty conclusive evidence against a multi-day delay in the MASIE data. Unless you're proposing that IJIS data *also* have the wrong date-stamp, and coincidentally are delayed by exactly the same number of days as MASIE....

It's time to drop that hypothesis and move on to some more productive line of inquiry.

Paul Klemencic

Now lets talk about the storm signature on the SIE data. The SIE reported by DMI showed a sizable uptick after the storm abated; the Bremen report shows at least one "level day" which we can't quantify without the actual numerical report; NSIDC daily data, shows a slight uptick on August 10th after the 240k decline the day before, and before the decline continued with another century break the next day.

So yes, the broader measures (that don't give us the regional impact of the storm on the three regions hit worst), show leveling or upticks after the storm abated; this is just as we would expect from the melt/measurement processes, and consistent with prior melt events caused by storms.

MASIE also usually shows a big uptick after storm drops, like the upticks after the August storm last year, but also upticks after a big drop in early September.

So the storm signature big drop - uptick - continued stronger than normal decline until the damage pack becomes more stable… this signature is present in all the SIE data for the storm, and in the MASIE total Arctic reports, and in all three regions most impacted by the storm, in the MASIE regional reports. Even the Canadian Archipelago data is consistent with the storm impact there.

This adds more observations that clearly support the hypothesis that the recent MASIE reported data were actually observed and collected during the storm dates.


Does it really matter ?. I think this inquiry can wait after the september refreeze commences...

Ned Ward

By the way, many thanks to crandles for compiling the past two years' worth of MASIE data from the rolling four-week CSV files.

Paul Klemencic

Ned, you are correlating single day data reports from MASIE against an average data set from IJIS… further, the time frame you selected is too large to test the hypothesis I am proposing. Sometimes the MASIE report is higher than IJIS and sometimes lower. Furthermore, the two SIE measurement systems use significantly different grid sizes. Finally, I don't find your correlations very strong.

Try plotting NSIDC single day data for the last several months, and compare to the MASIE single day data; both should show very close SIE readings for the same date, but don't.

If you shift the MASIE data 10-12 days you will find a much closer match with NSIDC numbers. I suggested this earlier, but you just repeated the same exercise you did last year.

I will take physical system observations coupled with actual heat and mass transfer calculations, particularly if available in a specific area or region… anytime… over an inconclusive statistical correlation on different average period data from different measurement systems over a timeframe where differences can be averaged out and obscured.

In process control, we use a step change to measure the system response, and the GAC-2012 gave us that step change. The MASIE data shows a serious transportation lag (the term process control engineers use for a delayed response to a step change, caused by the measurement and reporting system), to the GAC of 2012.

Paul Klemencic

Dr Tskoul: If Bremen and NSIDC hit minimums after September 18, MASIE may not show the SIE minimum until October, so it won't show the refreeze in late September.

You are correct that this hypothesis will stand or fall by the end of September, unless NSIDC pulls the MASIE reports before then.

Paul Klemencic

Ned, the link to the NSIDC single day data is on Neven's Daily Graphs page.

Ned Ward

OK, Paul. Over "the last several months" (May-August), the NSIDC daily data are most closely correlated with MASIE daily data with a 1-day lag. Correlation coefficients for 10, 11, and 12-day lags are effectively zero:

10-day lag: -0.0163
11-day lag: -0.0898
12-day lag: -0.0770

No, there isn't a ~10 day error in the date-stamp of the MASIE data ... unless the same error is also present in the IJIS data *and* the NSIDC data.

Seke Rob

Am afraid that MASIE had a [what storm] inversion:

2012234, 4999105.23
2012235, 5278249.56
2012236, 5012112.93

http://bit.ly/MASDMI refreshed. A conspiracy by forces to keep the Shell rig out :P

Ned Ward

Paul, your comment at 15:46 is an embarrassment.

This paragraph in particular is hilarious:

I will take physical system observations coupled with actual heat and mass transfer calculations, particularly if available in a specific area or region… anytime… over an inconclusive statistical correlation on different average period data from different measurement systems over a timeframe where differences can be averaged out and obscured.

You have two noisy time series. You think one of them is offset by 10+ days relative to the other.

The right way to test that is with some kind of objective, quantitative comparison of the two time series, both with and without that 10-day offset.

Not an ad-hoc process of selecting this or that wiggle in series A and then seeing if there happens to be a vaguely similar wiggle somewhere around 10 days later in series B. The potential for selection bias and self delusion is far too high.

This has nothing, nothing to do with "physical system observations coupled with actual heat and mass transfer calculations". You're cherry-picking wiggles in a noisy signal to fit your pre-existing ideas.

The comments to this entry are closed.