« Oh, and BTW, the Passages are open | Main | IJIS SIE: 5 million km2 mark passed »

Comments

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

Paul Klemencic

Crandle, if the average extent reported today, is 100k less than yesterday, you can select any method you want to allocate the drop. It doesn't matter which day caused the drop. The allocation between any two days doesn't matter. Essentially you can select any value for the extent you want for the daily measurement.

What does matter, is the sequence of the daily reported averages going back in time. It is this sequence of reported numbers that cause the problem, not the allocation over a two day period to get the initial extent. We could select any combination of extents for the two day period by whatever means possible, the results still blow up, ten days in the past, using the numbers reported by JAXA-IJIS.

Dave Eater

Paul,

This is the brief explanation of how the sea-ice extent is calculated, from IJIS:

"The sea-ice extent is calculated as the areal sum of sea ice covering the ocean where sea-ice concentration (SIC) exceeds 15%. SIC data of JAXA’s AMSR-E standard products are used for this purpose (http://sharaku.eorc.jaxa.jp/AMSR/products/pdf/alg_des.pdf). The algorithm for calculating SIC was developed and provided by Dr. Comiso of NASA GSFC through a cooperative relationship between NASA and JAXA."
(See http://www.ijis.iarc.uaf.edu/en/home/seaice_extent.htm)

This appears to be the relevant journal article for understanding JC Comiso’s algorithms:
http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=1196042

Peter Ellis

it has become too much unnecessary discussions at this blog about who is right and who is wrong
What. The. Blistering. {expletive deleted}?

If we simply accept and vector any old assertion without earnestly and honestly seeking out the truth, then we are nothing but another pseudo-scientific bunch of wackaloons, just with a different flavour.

I, and I hope you too, have more self-respect than that. If anyone, and I mean anyone, including James Hansen, Walt Meier or whosever, makes assertions about a particular dataset here, they have to be able to defend it with something more than "I eyeballed it and it looked right to me, you're just not looking at it right"

crandles

Re "We could select any combination of extents for the two day period by whatever means possible, the results still blow up, ten days in the past, using the numbers reported by JAXA-IJIS."

What is impossible (I know these figures are unlikely) about figures at:

https://docs.google.com/spreadsheet/ccc?key=0AjpGniYbi4andElraDhCa2VtVUZsQmJsNEVIU1dWZ3c&hl=en_US

The distributions change each day. (Change is far too steadily in the numbers suggested.)

The effect of the distribution change is that if you wrongly assume the averaging at total level instead of cell concentration level then working backwards will blow up to appear impossible.

crandles

A simpler way of saying it is that you have to cling to your average of daily numbers to get the problem of working the numbers backward so show they blow up. Surely if the 2 day averaging happens at cell concentration level it doesn't also happen at total level.

Stevemosher.wordpress.com

Paul

It be they continue QA the datasets for a long period. Im new to these datsets, but ive seen that before. neat mystery

Paul Klemencic

A lot of this is still preliminary:

Dave Eater, I read Josefino Comiso's paper, and very interesting it was, but not too relevant, since it covered sensor readings to measure ice concentrations, and not averaging or data process steps working with the ice concentrations to get SIE. Nevertheless, it was a worthwhile read to see all the methodology they go through to measure ice concentration. I think the best part of reading it, is that it drove home the fact that the AMSR-E is the source for both the Bremen map and the IJIS reported SIE data. They should agree, except that the Bremen map from the legend, shows SIEs down to 10%, and some timing issues.

Then I spent some time at the IJIS site and very carefully read the material. It appears that when they say 2 day averaging, they mean any data collected within the 48 hour period. As data comes in, they process it, and then fix and report the final number.
From IJIS:
Usually the latest value of daily sea-ice extent is fixed and updated at around 1 p.m. (4 a.m.) JST (UT).

Before the value is fixed, we also assign a preliminary value of daily sea-ice extent several times (usually three to four times) as an early report, which is determined without the full two-day observation coverage. (The fixed values of sea-ice extent are determined with the full coverage of observation data.)

I had read this several times before, but now with 'new' eyes.
It appears that any data collected anytime in the 48 hours is included, but they don't say how it is averaged. Since they need 48 hours to get enough satellite passes, even with the wide swaths, it appears the averaging is simply gathering up the data, and using the average of whatever data they have, whenever during the period it was collected. There really are no single day 'averages', and it doesn't appear there are even two day 'averages', but simply a 48 hour data collection period. The bulk of the data that comes in may not be evenly distributed throughout the 48 hour period, although the satellite passes must be very regular. But on the preliminary Bremen maps, we see various gray 'missing data' areas from the satellites, so the amount of data collected and timing could vary considerably.

There isn't any averaging of pixels, because the data is processed from pixels as it comes in.

Furthermore, it does seem that there is some kind of processing of the data. This was put in place to deal with changing the satellite sensing systems in the fall and spring, that created spurious bumps in the graph. The summer sensors (filters) are apparently used to avoid melt pond issues. So it is possible, that if the sensors report an anomalous reading (high or low), the method they use to 'average' the data has some kind of processing that prevents these "bumps".

Finally the preliminary Bremen map is released around 6 PM UT, and the final map is released around 5 AM UT the next day. So if I have done the timing right, the final Bremen map is released contemporaneously with the IJIS report. I seem to get the final Bremen map first, and the IJIS report very close thereafter.

But the Bremen map only includes 24 hours of data, and the last 12 hours provide much of the detail particularly on the Bering sea side of the ice pack. So in a sense the Bremen map is at least an average of 12 hours younger than the IJIS data, and could be younger.

I need to think about this a while. Thanks to everyone, even crandles : )

Chris Biscan

PAUL-

JAXA USES A 12.5KM GRID RES.
BREMEN 6.25KM GRID RES.

THEY WON'T AGREE LIKE THAT. GET THAT IN YOUR SMART BRAIN MAN!


There are to many holes out there causing Bremen to have a lower number.

I bet you they even out once the holes are gone.

Paul Klemencic

Chris Biscan. Yes. I understand and agree. I did hear you earlier, when you told me this. I didn't quite fully grok it, but now that is really important considering the different number of satellite passes that each product might be using... I need to think that through a bit.

But there is something else here as well. Give me a chance to sleep on this, and to think.

And I am glad this discussion isn't on the main thread anymore. I would rather keep this to those people interested in this kind of details. (But you know what they say about the devil and the details.. : )

Neven

Paul and others, maybe I should've started a separate thread for this. I apologize for not having done so.

Paul, if you want your findings turned into a guest blog, you can always e-mail me.

Janne Tuukkanen

Paul, could you simply plot a graph with different sources? Any lag in the range of seven days should be straight out visible by eyeballing. And such a graph would be interesting anyway, for comparing the differences.

Peter Ellis

The resolution isn't the only difference between Bremen and IJIS, they actually use a different algorithm to estimate sea ice concentration.

IJIS uses the AMSR Bootstrap Algorithm (ABA), while Bremen uses the ARTIST Sea Ice (ASI) algorithm. Details are here: my understanding is sketchy, but if I understand right, the Bremen analysis uses more different wavelengths to detect sea ice.
http://sharaku.eorc.jaxa.jp/AMSR/products/pdf/alg_des.pdf (IJIS)
http://www.iup.uni-bremen.de:8084/amsr/spreen05.pdf (Bremen)

As such, the data are simply not directly comparable, despite coming from the same actual satellite. The two products will have different sensitivities to ice of different temperatures, more or less susceptibility to melt pond and cloud interference, etc.

NSIDC data uses passive microwave data, but from a completely different satellite, so it's even less comparable to IJIS or Bremen.

MASIE site is down so I can't check right now, but I believe it uses several sensors, including non-microwave (optical) wavelengths, so it's bound to be the most different of the lot.

Peter Ellis

Note also that this kind of difference between sensors/algorithms can easily create the spurious appearance of a time-lag.

Take for example hypothetical methods A, B and C.

A is an optical system and gets confused between white cloud and white ice. It is thus poor at seeing through cloud and will overstate sea ice concentration when there's thick cloud.

B is a microwave system, but the particular algorithm used is poor at telling the difference between melt ponds / wet ice / open water. It therefore understates ice concentration when the ice surface is wet.

C is a nonexistent perfect system that actually gives the right result!

Now let's see what happens when a weather system moves over the ice and drops rain/sleet over an area. A will over-estimate concentration until the clouds move away. B will conversely under-estimate the concentration, while C gives the true answer.

On the graph, and also in a regional analysis, method B shows a "flash melt" followed by recovery, method A shows a plateau that then drops sharply after the clouds part, while C shows a smooth decline.

If you then come in and naively try and match wiggles in the traces, you'd wrongly claim that A has a lag relative to B, and that C is doing some kind of unannounced long-term smoothing. The duration of the false "lag" would be the persistence time of an average weather/cloud system, i.e. a few days.

I hope the parallels to MASIE/Bremen/IJIS are clear.

Ned Ward

Paul writes: Then I spent some time at the IJIS site and very carefully read the material. It appears that when they say 2 day averaging, they mean any data collected within the 48 hour period. As data comes in, they process it, and then fix and report the final number.

Yes, this is exactly what I would expect them to do. Aqua (the spacecraft that carries AMSR-E) is in a near-polar orbit, meaning that the orbit tracks converge as you move from the equator towards the poles. (It's inclined slightly off of 90 degrees, so the orbit tracks don't cross directly over the poles).

Thus, the swaths of image data from successive orbits overlap more and more as you move to higher latitudes. In any given two-day period, points at lower latitude will be covered less frequently (perhaps only once near the equator) while points at higher latitude will be covered more frequently (multiple times during a single day).

If I were designing a system to measure sea ice concentration from satellite imagery, I'd take each individual swath as it comes in, process it using the algorithm that infers concentration for each pixel based on that pixel's spectral radiance (or brightness temperature, or whatever) in whichever spectral bands are being used. Then I'd reproject those concentrations into the polar map grid. The final concentration value for a given grid cell would be the average of however many pixels overlap that grid cell on all image swaths during the two-day period. (Ideally you'd use a weighted average with weights based on the area of each pixel's overlap with the grid cell in question).

In general, in remote sensing, you want to avoid resampling the raw data as much as possible. This is (part of) the rationale for calculating pixel-level sea ice concentrations first on the unprojected swath data. However, there might be other logistical reasons why IJIS-JAXA (or Bremen or whoever) might prefer to project the raw data into their map coordinate system first before calculating ice concentrations.

Ned Ward

Peter Ellis writes: The resolution isn't the only difference between Bremen and IJIS, they actually use a different algorithm to estimate sea ice concentration.

Yes, exactly. There was a discussion of this recently over at William Connolley's blog (Stoat), during which my friend and former colleague J made some good points. You can't really say that Method A is more reliable than Method B just because of a 2-fold difference in the grid cell size. Differences in the algorithms used will generally outweigh that.

It seems like some people here are really, really hung up on the need to figure out whether IJIS, NSIDC, or Bremen is giving the "best" estimate of ice extent. That's not really an easy thing to determine, and we shouldn't expect to be able to figure it out here, neither by simple rules like "6.25 vs 12.5 vs 25 km resolution" nor by drawing vague speculative inferences after eyeballing a bunch of maps.

In my humble opinion it's better to just report the trends from all the various SIE data sets, and remain agnostic about which one best matches reality.

Peter also writes: Note also that this kind of difference between sensors/algorithms can easily create the spurious appearance of a time-lag.

Yes, and not just for the reasons which Peter mentions. If, during a season when SIE is shrinking, Bremen appears to report the same ice extent that MASIE appears to report six days later, it could be because:

(a) Bremen systematically under-estimates the ice extent; or

(b) MASIE systematically over-estimates the ice extent; or

(c) MASIE has a time lag; or

(d) something else that we haven't thought of.

I've repeatedly tried to do quantitative comparisons of the correlation between IJIS and MASIE with different lags (0 days, 1 day, etc.) There just is too much noise, and not a long enough record. We'd want months of data, when we only have four weeks from MASIE.

Trying to match up distinct events in the time series is just too unreliable when there's so much noise and the series is so short.

Nick Barnes

These recent comments by Ned Ward and Peter Ellis are really excellent: aside from the light-hearted banter, this is the kind of thing that makes the comment threads here worthwhile.

crandles

Re: "We'd want months of data, when we only have four weeks from MASIE."

If you want months try

https://docs.google.com/spreadsheet/ccc?key=0AjpGniYbi4andFdrTEZrdGVjaXBfWHpIb2VhNHNrM1E&hl=en_US#gid=0

Ned Ward

crandles wrote: If you want months try

Hey, thanks! That's great.

So I took the MASIE data, and calculated the correlation with IJIS-JAXA data. (I did the correlation on each day's change in extent, rather than each day's extent).

I did this for all lags, from MASIE being lagged by 6 days (Paul's claim) up to MASIE being ahead by 6 days.

Here are the correlations:

MASIE lagged 6 days: 0.27
MASIE lagged 5 days: 0.27
MASIE lagged 4 days: 0.36
MASIE lagged 3 days: 0.42
MASIE lagged 2 days: 0.44
MASIE lagged 1 day: 0.42
No lag at all: 0.39
MASIE ahead 1 day: 0.37
MASIE ahead 2 days: 0.37
MASIE ahead 3 days: 0.34
MASIE ahead 4 days: 0.37
MASIE ahead 5 days: 0.32
MASIE ahead 6 days: 0.29

So Paul's claim of a 6-day lag doesn't seem to be supported very well.

I also looked at the difference in extent (MASIE minus IJIS) as a function of time. The data set crandles posted runs from December through August. MASIE reports a greater ice extent than IJIS-JAXA almost all the time.

At the start of the series (Dec 2010) MASIE is about 0.7 million km2 above IJIS. The difference peaks in early March, when MASIE is about 1.1 million km2 above IJIS. It then starts to decrease. By mid August, MASIE is averaging only about 0.1 million km2 above IJIS. It looks like by the time of the Sept. minimum the difference might be near 0.

As a percentage of the daily extent, the difference (MASIE minus IJIS) peaks at about 7 to 8% in early March, and drops to about 2% in August.

In other words, for most or all of the year, MASIE sees a greater extent of ice than IJIS does. That is particularly true when there's a lot of ice, and less true when there's less ice.

That's interesting. I would be cautious about interpreting it, and especially about ascribing the difference to any particular factor.

Artful Dodger

Hi Paul,

The reason you can not reconcile the Sea Ice concentration map to IJIS SIE numbers is because you're using the wrong map. Uni-Bremen is not IJIS, and both use separate algorithms for their ice products.

Compare IJIS SIE numbers to the daily maps on the Arctic sea-ice monitor by AMSR-E.

I think you'll find they match quite well... If fact I know they do, because I posted on this subject here last September, when I made a blink animation comparing the IJIS ice edge on two consecutive days.

Cheers,
Lodger

Seke Rob

Careful, MASIE is NH, not only Arctic. Is Baltic (peaked at 250K) and Yellow Sea part of the Arctic that JAXA includes?

Anu

@Paul Klemencic | September 04, 2011 at 03:19

I looked at a longer set of JAXA SIE (sea ice extent) numbers, as you suggested a day or so ago (August 13 to August 30, a series of 18 2-day averages). I see what you mean by "the fit blows up". My first guesstimate of an initial sea ice extent to start the series led to some weird gyrations of daily sea ice extent after about 7 or 8 days of fitting.
Again, I use "e12n" as the variable for "extent, August 12th, noon".
I guesstimated the first value in the series as close to the midpoint between 2-day averages (reported JAXA values) for 8/12 and 8/13. The average of two days "noon" SIE is the reported JAXA values, e.g. 5700313 km^2 for 8/13/11.

e12n - guesstimated at 5763906
(e12n + e13n)/2 = 5700313 --> e13n = 11400626 - e12n = 5636720
(e13n + e14n)/2 = 5624375 --> e14n = 11248750 - e13n = 5612030
(e14n + e15n)/2 = 5588281 --> e15n = 11176562 - e14n = 5564532
(e15n + e16n)/2 = 5548906 --> e16n = 11097812 - e15n = 5533280
(e16n + e17n)/2 = 5490625 --> e17n = 10981250 - e16n = 5447970
(e17n + e18n)/2 = 5410313 --> e18n = 10820626 - e17n = 5372656
(e18n + e19n)/2 = 5372656 --> e19n = 10745312 - e18n = 5372656
(e19n + e20n)/2 = 5335469 --> e20n = 10670938 - e19n = 5298282
(e20n + e21n)/2 = 5276719 --> e21n = 10553438 - e20n = 5255156
(e21n + e22n)/2 = 5173906 --> e22n = 10347812 - e21n = 5092656
(e22n + e23n)/2 = 5121563 --> e23n = 10243126 - e22n = 5150470
(e23n + e24n)/2 = 5084844 --> e24n = 10169688 - e23n = 5019218
(e24n + e25n)/2 = 5055781 --> e25n = 10111562 - e24n = 5092344
(e25n + e26n)/2 = 5009844 --> e26n = 10019688 - e25n = 4927344
(e26n + e27n)/2 = 4990156 --> e27n = 9980312 - e26n = 5052968
(e27n + e28n)/2 = 4964063 --> e28n = 9928126 - e27n = 4875158
(e28n + e29n)/2 = 4896563 --> e29n = 9793126 - e28n = 4917968
(e29n + e30n)/2 = 4796875 --> e30n = 9593750 - e29n = 4675782

But this whole exercise just shows that given the 2-day averages (reported JAXA data), it is non-trivial to solve for the internal, daily SIE data that gives rise to the reported data.

For this example of 18 days data, there are 19 variables, e12n to e30n. This gives a system of 18 linear equations in 19 variables (i.e., the first equation is [e12n + e13n]/2 = 5700313 ) - an undetermined system. We provide the 19th equation by "guesstimating" an initial SIE - I wouldn't be surprised if MOST such guesses give solutions that "blow up" as you call it.

But since JAXA says they are doing this 2-day averaging, then unless they are lying, there is at least ONE solution to these 18 equations that works fine. Probably more. But you're right, as you try to solve for longer and longer series of JAXA data, the tested solution is likely to "blow up" and oscillate wildly around the reported data.

If you programmed a search of solution space for the initial "guess" (either all the integer values between two reported JAXA numbers at the beginning of a series, or maybe even include some fractions like .5, .25, .16666 since JAXA averages 3 or 4 measured SIE's of integer numbers of 12.5 km^2 pixels per day - I'm not sure how sensitive the solution is to fractions in the initial guess) I'm sure you could find the one, or more, solutions to this system of linear equations that seem plausible (e.g. no increases of SIE during summer melt). I would recommend searching a small solution space, between two JAXA numbers that are fairly close - it doesn't really matter where your initial guess is made, all the other values are derived from that guess.

Your failure to find a good solution to 18, or 30, or 900 days of JAXA data doesn't mean they *aren't* doing simple 2-day averaging - it just means you haven't fully searched the solution space (for the initial guess) to solve this large system of linear equations.

But you could, if you wanted to.
It's probably easier to just ask JAXA for their internal data.

Ned Ward

Anu, have you followed the other comments recently in this thread? When people talk about IJIS using "two day averaging", it probably doesn't mean "taking the whole-Arctic extents on Day 1 and Day 2 and averaging them."

If you insist on trying to reconstruct daily totals, you might want to begin by converting the extents to pixels rather than km. Then you can work in units of whole pixels, which are discrete rather than continuous.

I've just tried doing this kind of test, and for the sequence of days I looked at, there is no whole number of pixels that could be used as a "Day 0" extent that would not result in unrealistic daily changes later on in the sequence.

But I didn't think that was what IJIS-JAXA were doing anyway.

RunInCircles

Paul Klemencic
I don't know if you still check this page but if you do try this -
Date. Measured Averaging ........ Final Reported
....... Grids .... Calculation ....... Grids Value
13Aug n/a ... n/a .................. 36482 5700313
14Aug 35510 (35510 + 36482)/2 = 35996 5624375
15Aug 35534 (35534 + 35996)/2 = 35765 5588281
16Aug 35261 (35261 + 35765)/2 = 35513 5548906
17Aug 34767 (34767 + 35513)/2 = 35140 5490625
18Aug 34112 (34112 + 35140)/2 = 34626 5410313
19Aug 34144 (34144 + 34626)/2 = 34385 5372656
20Aug 33909 (33909 + 34385)/2 = 34147 5335469
21Aug 33395 (33395 + 34147)/2 = 33771 5276719
22Aug 32455 (32455 + 33771)/2 = 33113 5173906
23Aug 32443 (32443 + 33113)/2 = 32778 5121563
24Aug 32308 (32308 + 32778)/2 = 32543 5084844
25Aug 32171 (32171 + 32543)/2 = 32357 5055781
26Aug 31769 (31769 + 32357)/2 = 32063 5009844
27Aug 31811 (31811 + 32063)/2 = 31937 4990156
28Aug 31603 (31603 + 31937)/2 = 31770 4964063
29Aug 30906 (30906 + 31770)/2 = 31338 4896563
30Aug 30062 (30062 + 31338)/2 = 30700 4796875
31Aug 30020 (30020 + 30700)/2 = 30360 4743750
01Sep 30236 (30236 + 30360)/2 = 30298 4734063
02Sep 30128 (30128 + 30298)/2 = 30213 4720781
03Sep 29737 (29737 + 30213)/2 = 29975 4683594
04Sep 29737 (29475 + 29975)/2 = 29725 4644531
You will note that there was almost 147KsqK drop on Aug 22 which matches our flash expectations. Any claim to matching IJIS measurements or calculations is not asserted.

crandles

Runin circles, That doesn't seem to be averaging 2 days but averaging an average with the next day. Not sure why you would prefer that to more stantard 5 day averaging though I suppose it is possible.

But then there is Ned's point. All IJIS values are exact multiple of 12.5*12.5. The idea that every day was odd or even to avoid a half cell being included in the total is astromonomically unlikely (like 1 in 2^3000ish)

Kevin McKinney

"I hope the parallels to MASIE/Bremen/IJIS are clear."

A nice thought experiment from Peter Ellis!

For my part, I'm not so worried about the intricacies of methodology (more power to them as is!) I figure that the big picture is pretty consistent across these various metrics, so I'm fairly happy to be "agnostic" about which is "best." (Ned Ward's suggestion, IIRC.)

Hey, call me a slacker. What I do want to see from the data is whether or not there is any prospect of sea ice 'recovery,' as specifically promised to me by various online prophets of "natural variation."

Sadly, we all know the answer to that.

RunInCircles

crandles
If there is a half grid from averaging it is rounded up. All reported values are a multiple of full grid counts. A running average behaves very well in the presense of noise.

Paul Klemencic

Sorry for delays in responding to comments; I have been thinking and investigating.

Peter Ellis, Chris Biscan, et. al. raised the issue of the sensitivity and resolution of the satellite measurements and how this resulted in different ice extents being reported.

The systems measuring ice concentration from the satellite measurements use different imaging processing. The grid size used in the data collection causes the biggest differences, with the NSIDC legacy system (with different sensors) and IJIS using the 12.5 km imaging data, Bremen using 6.25 km for both the ice extent map and the graphs of ice extent, and MAISIE using a 4 km grid size. In the summer, with open water areas within the pack, the smaller grid sizes should result in smaller extent readings.

But I have difficulty understanding the data reported by the different methods for the different ice extent products, primarily because the agencies don't explain how they use and process the ice concentration results (calculated from the images) to produce the final report. They don't provide operational definitions explaining what the reported number means.

To understand the reported numbers, one needs to know what time period was used to collect the data; if the data was averaged, what averaging system was used over the time period (temporal averaging); and how corrections (if any) are made later to modify the reported result. The following discussion uses the IJIS report as an example.

According to information on their website, the IJIS team is generating three to four reports per day. Over the 48 hour data collection period, they generate 6-8 reports that do not cover the full 48 hour period. They freeze the data collection at 4 AM UTC every day, to generate a preliminary report covering the full 48 hours. They continue processing, generate a final report at about 12 hours later. All of this processing is done during the "next day" in Japan, but the Japanese date the reported extent more consistent with the data collection period, using the date of the last full UTC day in the 48 hour period. Nevertheless, a report dated September 2, includes data from 4 AM UTC on September 1 to 4 AM UTC on September 3.

How do they take account for the data collected at different times during the 48 hours?
IJIS says:
"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 two days to achieve rapid data release. The wider spatial coverage of AMSR-E enables reducing the data-production period."

How is the temporal average defined for a two day period?

They don't say, and I couldn't find anything published in the literature.

Here's how I would do it: I would use the preliminary reports covering partial periods to see if the grid was covered with ice at any time in the first 24 hours. In other words, each swath of the satellite that gives a reading for a grid, should be processed, and answering the question: "Is there 15% ice in this grid?" Now, knowing the sensors may make a mistake, then all the swaths in 24 hours should be considered, in answering the question for each swath for that day. If the Yes answers are high enough, then that grid get marked as "ice". Same procedure for the next day. I would not average pixels. Instead I would simply logically compare the "ice" answers. If a given grid is marked "ice" in both days, then its ice. If not "ice" in either day, then its not ice. If it is ice in the first day, but not in the second, then its not ice. If its not ice in the first day, but ice in the second, then its ice. I could then date my report September 2 in the above example, and say that the report reflected the amount of ice extent near the end of September 2.

Why not use pixel averaging? (Several people here suggest this.)

Because this presumes the ice isn't moving. Say there is a 5 km circular pod of 90% ice In a field of floes averaging 2-10% concentration. Early in the 48 hour period, the 5 km pod is in one grid, and that grid now has somewhere between 16-24% ice concentration. In 20 hours, the 5 km pod moves 10 km sweeping entirely into a neighboring grid. Now that grid becomes "ice". In another 24 hours, the 5 km pod moves another 12 km into yet a third grid, and that grid becomes "ice". So a 25 sq km pod of sea ice gets counted three times!

This would work worse with high resolution products; if pixels were averaged over 48 hours that would really cause "over-count" problems, since the ice could move through more grids. (fortunately the higher resolution grid products release their data later, and date the data appropriately to the data collection period).

Now it may be, that "false positive" reading of ice concentration are a problem... Ok, then use a process that requires several yes answers to the ice concentration question listed above. I think taking data from one satellite pass, and averaging that with data from a satellite pass 24-48 hours from now, causes problems.

In spite of what I think, these guys are professionals and they must have good reasons for whatever types of temporal averaging they are using. But then why is the IJIS report dated September 2 in the above example? Did they average to the end of September 2, or noon on September 2, or did they temporally average over the 48 hour period, which means the average conforms to 4 AM UTC on September 2, which is only four hours into the day in UTC time. I can see why they date the report this way, but in reality, the way we look at reports, we consider them end of the day reports, so the closest end of day is September 1 in the example I am using.

The Bremen maps process data for the current day, and the final map is released at 3 AM the next day, but dated the previous day; the day the data was taken. So the September 2nd IJIS report should not be compared to the September 2nd Bremen map, but to the September 1st map. And if I want to blink contrast the maps to compare with the extent loss in the latest IJIS report, I should use the blink comparison between the August 31st map, and the September 2nd map, because that period conforms most closely to the IJIS 48 hour temporal averaging period.

So what I know so far?

1. The IJIS report is not using simple 2-day averaging, but is temporally averaging ice concentration data collected over 48 hours centered around 4 AM UT of the date on the report. We don't know how the temporal averaging is done.

2. The IJIS report is not consistent with the Bremen map for the same date. Observing ice extent losses over a 48 hour period, from the Bremen map two days ago compared to the current map, will be more consistent with the extent losses in te most recent IJIS report.

Paul Klemencic

Responding to Michael Sweet's comment:
Paul,
It seems to me that you do not know how IJIS averages their data and you have wasted a lot of posts speculating. An old friend of mine said "you can spend a week working to save an hour in the library." Why don't you try to find the paper where IJIS describes their technique, which will describe their process in enough detail that you will be able to replicate their data? You are wasting your time asserting that published scientists have not correctly described how they measured their data.

I have searched for papers, but not found papers pertaining to temporal averaging. I did read the Comiso paper on getting ice concentration measurements. There may be some older papers on temporal averaging that I haven''t found yet.

I have NOT said this: "they have not correctly described how they measured their data", the phrase you used in your comment. I have said that they haven't explained how they processed the data, particularly regarding to temporal averaging (to get what people on this site used to call "two-day averages").

Because they haven't described their process or provided operational definitions, I can't replicate the processing steps fully.

I will ask some Arctic ice scientists some questions, as soon as I know enough about temporal averaging, and understand the various ice extent reports well enough to be able to ask intelligent questions, and get informative responses.

What I (we) really need is a 'user manual', because none of us seem to understand what these numbers mean (my observation).

Paul Klemencic

Somehow the last comment was completely italicized; the first paragraph is Michael Sweet's comment, the rest is my response.

(Please note this comment is italicized as well... with no HTML tags in it.)

Dave Eater


You didn't include a closing italic tag at the end of the paragraph you quoted (and, I assume, intended to italicize). You included a second "opening" italic tag - you left out the required slash that indicates a closing tag (e.g., opening tag has just an "i" between the brackets; closing tag has "/i" between the brackets).

I added a closing italic tag at the start of this post, so italics should now be turned off on subsequent posts.

Paul Klemencic

Regarding temporal averaging; the NSIDC legacy report is a 5-day averaged report, delayed to the end of the period, but dated on the middle day. The Bremen extent data that L. Hamilton is posting on this site, seems to be the middle day of their averaging period.

IJIS is dating their averaged product within four hours of the end of their averaging period. Therefore, if I want to compare extent reports for September 2, I should use the September 2nd NSIDC extent with the September 1st IJIS extent report. Then both averaged extents are being compared near the center of their averaging periods.

Guys, this sounds over-complicated, but thats the way it is. I hope at the end of this process, some simplicity emerges from the complexity.

Paul Klemencic

Dave Eater: I already tried that.... no luck.

Paul Klemencic

Shifted the IJIS report date the wrong way in my most recent comment: the September 2 NSIDC would be most closely comparable to the September 3rd IJIS report.

Dave Eater

Interesting. It seemed to work for me, at least at first.

Neven

I fixed it, guys.

Bob Wallace

What I (we) really need is a 'user manual', because none of us seem to understand what these numbers mean (my observation).

I'll vote for that. I could really use an Ice for Dummies section.

Ned Ward

Paul writes: The grid size used in the data collection causes the biggest differences

How do you know that? How do you know that other differences in the methods don't have a larger effect on the results?

Paul: In the summer, with open water areas within the pack, the smaller grid sizes should result in smaller extent readings.

Perhaps. That, again, is speculation. Note that MASIE (with the smallest grid size) reported larger extent values than IJIS on most days this summer (see the spreadsheet posted by crandles).

You tried to explain that away earlier by speculating about a 6-day lag in the MASIE data relative to IJIS, but the correlation analysis seems to reject that (see my comment above).

Paul: The IJIS report is not using simple 2-day averaging, but is temporally averaging [...] data collected over 48 hours

"48 hours" is "2 days", and "averaging individual observations" is still averaging, even if you mistakenly assumed they meant "averaging daily totals".

Paul Klemencic

Anu, I already showed you that the extent numbers you came up resulted in an unrealistic set of "two-day averages". Then you wrote this:

If you programmed a search of solution space for the initial "guess" (either all the integer values between two reported JAXA numbers at the beginning of a series, or maybe even include some fractions like .5, .25, .16666 since JAXA averages 3 or 4 measured SIE's of integer numbers of 12.5 km^2 pixels per day - I'm not sure how sensitive the solution is to fractions in the initial guess) I'm sure you could find the one, or more, solutions to this system of linear equations that seem plausible (e.g. no increases of SIE during summer melt). I would recommend searching a small solution space, between two JAXA numbers that are fairly close - it doesn't really matter where your initial guess is made, all the other values are derived from that guess.

Your failure to find a good solution to 18, or 30, or 900 days of JAXA data doesn't mean they *aren't* doing simple 2-day averaging - it just means you haven't fully searched the solution space (for the initial guess) to solve this large system of linear equations.

Anu, there aren't any values of initial extents that don't produce unrealistic single day extent values. More complicated mathematical techniques won't find a starting ice extent number that won't "blow-up" as we calculate away from the starting point.

The problem isn't the initial extent value; the problem is the series of reportedly "two-day averages". They cannot be two day averages.

Clearly IJIC is either:
1. Using data from one day in one 48-hour data collection, then changing the data for use in the next 48-hour data collection period: this could result from a pixel averaging procedure.

This explanation has problems, because the amount of extent change (sq km) from the same 24 hours of collected data, from one data collection period to the next, will be quite large to correct enough to keep from getting unrealistic extents.

Also this possibility doesn't correct the other obvious problem with the IJIS data; the variation in the reported extents is surprisingly low, rarely showing an extent gain over the 48 hours. In reality, we observe many occasions, where the ice pack can gain extent for several days, with at least three periods since August 1st.

2. Using some kind of longer term averaging or similar systems to insert corrections in future 48-hour periods.

This explanation would account for both the failure of the "2-day averaging" test, and the lack of variation in the reported numbers. However, it has a credibility issue: Wouldn't IJIS tell us that they were doing this? ... I think they would.

So right now I don't have an explanation, although if forced, I lean more toward the second than the first, because it explains the lack of variation... essentially some of the variation has been removed.

Paul Klemencic

Ned, your correlation review of the MASIE data is not correct. I didn't base my conclusion that the MASIE data is date stamped incorrectly on the total extent data. I based it on observable events in the regions that matched the regional breakdown on the MASIE spreadsheet.

You basically ignored the best reason to deduce the MASIE data was dated incorrectly, to try and debunk the weakest reason, which in my opinion, you failed to do anyway. This mis-directed analysis led you to the wrong conclusion. The MASIE data is dated incorrectly. We will get some more indication of that when the data for days 247 and 248 are posted.

Day 247 data will show an extent of somewhere between 4.60 and 4.66 million sq km consistent with the Bremen data and NSIDC data for the day six days earlier.

And Day 248 data will show large extent losses in the Beaufort, Chukchi, and E. Siberian regions, with a loss of around 100k. I pointed out these events at the time, and in the next few days, the MASIE reports will show these events, that happened six days prior.

Just to emphasize how unusual these predictions are; to hit below 4.66 million down from the latest 4.81 million is a pretty aggressive forecast for two days in September. And the second one, involves a huge extent loss for September, with regional specific detail.

Please note that this is extraordinary strong evidence for a dating problem, because how else am I able to predict the future? I am not the omniscient overseer of the universe. But I was able to predict the big extent loss in the MASIE data entered for day 240 four days in advance, and I am about to do it again.

Note: I started with IJIS today, and was working on putting the MASIE stuff up next. More to come...

Paul Klemencic

RunInCircle, Actually crandles addressed the same thing I noticed with your technique. In your calculation, you didn't average the number of grids with ice on one day, with the number of grids that contain ice on the second day; instead you took the Average number of grids on the first day REPORT, and averaged that with grid count for the second day.

Essentially you are coming up with some kind of weighted three day averaging. And it works, with the IJIS reported data. Which proves my point; there are lots of ways to come up with the IJIS reported data; either IJIS modifies the last 24 hours of data before they use it in the next 48-hour period, or they use something similar to three day or longer averaging of some sort (which is what you did), or a means of introducing corrections to future 48-hour periods.

I just addressed these options in a comment to Anu.

Paul Klemencic

Artful Dodger, Thanks for the link on the IJIS extent map. I will monitor the site. Today will be a really good day to test, because the 'Laptev Bite' opened up.

I think I already have identified the big problem with comparing the IJIS reported extent with the Bremen map changes on the same day, and its the timing of the data collection period. Last night in my comment on the final Bremen map, I did a two-day blink comparison between the Sept 4th map, and the Sept 2nd map, and that allowed me to more accurately approximate the IJIS report. I was a bit low at first, expecting about 60k drop in the two days, and the preliminary IJIS seemed a bit high. Then IJIS revised reported extent down significantly this morning, right into the range consistent with my observations.

But this process is much too tenuous. Comments on the main SIE Update thread constantly compare the IJIS report with changes observed in the Bremen map (which I post often). I used to try to do that, but by mid-August, I just threw my hands up, and suggested we might as well roll die to predict the IJIS report, then look at the Bremen map. Our predictions were so bad, it was laughable.

I will follow the site you linked.

Ned Ward

Paul writes: Day 247 data will show an extent of somewhere between 4.60 and 4.66 million sq km consistent with the Bremen data and NSIDC data for the day six days earlier.

That's a pretty weak prediction. You're suggesting MASIE will decrease by 0.15 million from day 245 to 247. Large drops like that occurred twice in the preceding seven days (0.13 and 0.15 million km2). In addition, MASIE has just had two small increases, so perhaps it's primed for a larger than normal decrease that would have nothing to do with a "six day lag".

In fact, simple linear regression on recent MASIE data would project numbers that are basically identical to your "prediction" for day 247. If you look at the trend from the past 14 days, it would project 4.57. Based on the past 7 days, it would project 4.65.

So you can expect me to be completely unimpressed if MASIE shows a value of 4.60 to 4.66 on Day 247.

The second prediction you make (a 100K loss from Day 247 to 248) is slightly more out of the ordinary. Maybe it would warrant a raised eyebrow if that happened to occur. Would that be convincing evidence of a six-day lag? Not even remotely.


I can think of at least three ways you could convince me of the existence of a six-day lag:

(1) You could find something in writing documenting such a lag, or call up someone at NSIDC and get them to confirm that the data are lagged.

(2) You could look at a very long period of time, during which the extent both increased and decreased, and show that the correlation between MASIE and IJIS is much higher with a six day lag than without. Feel free to use crandles's spreadsheet if you want to try this.

(3) When each day's IJIS data were released, you could make a prediction for MASIE six days later. If you were successful at many such predictions, including a substantial number that were outside the range of what would normally be expected (i.e., not like predicting 4.60-4.66 for Day 247) I might find that convincing.

While we're at it, let's address this:

Paul writes: Ned, your correlation review of the MASIE data is not correct. I didn't base my conclusion that the MASIE data is date stamped incorrectly on the total extent data. I based it on observable events in the regions that matched the regional breakdown on the MASIE spreadsheet.

Why does it matter what you based your hypothesis on? We can test the hypothesis other ways.

If your hypothesis were true, the correlation between MASIE and IJIS should be higher for a six-day lag than for no lag. It isn't.

If your hypothesis were true, MASIE's numbers would be higher than IJIS when the ice is shrinking (March-Sept) but they would be lower when the ice is increasing (Sept-March). From crandles's spreadsheet, we can see that they are higher during both periods, not just the former.

Paul Klemencic

Enough on IJIS; now lets tackle MASIE. Lets start by looking at Ned Ward's statistical correlation between the two, with a little more detail than in my reply at 2:08 today.

Based on the content of my comments already, we can see Ned made quite a few errors because he doesn't understand the measurement and reporting systems for the different SIE products, and ignored the most compelling evidence for the MASIE data being dated incorrectly. Lets list a few errors:

1. A smaller grid (MASIE is 4 km) versus a larger grid (IJIS is 12.5 km) definitely should pick up more open water within the pack, and measure the edge of the pack more accurately. This is why the various groups are moving to smaller grids. The difference is huge; a 12.5 km grid is almost ten times (9.77) as large in terms of area. If there is any significant amount of 90% ice in that large grid, it is going to count the entire grid area as ice.

There is no reason to expect MASIE reported extents to be significantly larger than IJIS extents... especially in the summer when ice extent is falling. That the MASIE data appear to do so, should set off alarm bells.

2. The most compelling evidence is the regional breakdown, particularly on unusual days, like the big Flash Melt day of August 22nd. This day was so unusual because it appeared that over 50k sq km disappeared in the Chukchi region, out of what appeared to around 250k.

But the MASIE report for the day of the melt showed 328k sq km ahead of the melt, and 309k after (these numbers seemed to match satellite observations six days earlier), and this is only a 6.6% decline. We expected at least a 20% decline. And the Beaufort decline also fell far short of observed.

Then six days later, on August 28th, MASIE data for day 240 showed a decline from 247k to 180k, a decline of 67k, and a 27% decline, which is an easily observable huge drop, not seen on the map for August 28. The data posted on that day and the day before, matched the extent before and after the Flash Melt, and the size and percentage of decline expected for August 22. And the data posted that date matched losses in the Beaufort and E. Siberian for August 22 as well, for a total loss of 157k in those three regions. The net of the remaining regions showed a gain of 3k leaving a total extent loss of 154k. These massive losses in these three regions did not match the satellite map imagery for August 28. ('Houston, we have a problem.')

A few days later on day 243, a big 134k drop was reported by MASIE, centered primarily in the Canadian Archipelago, Beaufort, and Greenland sea, with a gain in the Chukchi. Didn't match the satellite observations at all, but did match the satellite observations for day 237. ('Houston, we have another problem.')

This is extremely strong evidence for dating problem with MASIE, which Ned ignored.

3. Ned ran a correlation matching IJIS reported extents with the same date MASIE reports, ignoring the fact that one is 24 hour data, and the other is 48 hour data. The IJIS date on their report is 20 hours after the midpoint of the data collection period being temporally averaged. The closest IJIS dated data to the normal MASIE dated data, would be the date one day before the MASIE date. His correlation is off by a day... the lag day 5 correlation listed should be lag day 6, etc.

4. Ned didn't look for to see if the winter peaks matched. The winter peak for IJIS is dated March 7, with the midpoint of the 48-hour collection period closer to March 6. The MASIE peak should hit six days later, and lo and behold, MASIE peaks on March 13.

5. Ned could have at least looked at my evidence for regional data supporting the 6 day mistake in dating the data, by following the data over the last 4-5 days, but didn't consider my evidence worth considering. That's arrogant. I looked at his correlation carefully. I considered his arguments ("you aren't certain MASIE should measure lower extents than IJIS"). He either didn't review my data and observations, or decided not to respond to the arguments supporting my hypothesis. This is a big mistake; in fact the biggest error he made.

More on MASIE to come...

Paul Klemencic

Wow, from Ned's latest response, he still hasn't bothered to examine my primary evidence.... Arrogance! When will they ever learn?

And Ned, at least 120k loss in two days in September, with roughly a 100k loss the next day... that is unusual. Especially since I am even predicting the regions where the melt will be centered.

My evidence blows your evidence away.

Paul Klemencic

Ned, one more thing, Who made IJIS the gold standard for extent measurement?

Just because IJIS publishes daily extent data that you can put into your statistics package, doesn't make their data the comparative standard for everyone else. Their data appears to be much different than the other SIE products, but I would be more wary of considering them the standard to compare against.

Your correlation of IJIS with MASIE is riddled with problems.

Kevin O'Neill

Paul, you convinced me long ago - but I think your last post was ill-considered. That we don't understand the IJIS algorithm makes it neither wrong nor right. If anything your analysis points to problems with MASIE - not IJIS.

I'm especially bothered that MASIE is supposed to be a quick release product when - as you've pointed out - it appears to have a serious lag.

Paul Klemencic

Kevin, you are probably right. If I understood the IJIS extent temporal averaging better, I would have much more confidence in the results. Right now my data seem to indicate it overestimates SIE the final report this morning by about 130-150k compared to Bremen and NSIDC, when they finally show an estimate for September 4.

As for MASIE, I posted on August 31, that as soon as I saw the MASIE reports for September 5 and September 6, that I would consider my case strong enough to send note to NSIDC. Which I intend to do. I was kind of hoping they would 'hear' about it, and fix the problem. I already sent them this note:


MASIE Site - I love it; but update it.
Date: September 1, 2011 10:36:13 AM PDT
To: nsidc@nsidc.org

I visit the MASIE site every day the last two months, and find the regional ice extent data very informative.

There does seem to be a problem updating the data, and improving documentation of the region sea areas, etc.
I saw this message there:

NSIDC has received support to develop MASIE but not to maintain MASIE. We are actively seeking support to maintain the Web site and products over the long term.

If you find MASIE helpful, please let us know with a quick message to NSIDC User Services.

I find it useful, so please see that it receives support. Understanding regional ice extent declines could be key to accurately forecasting Arctic summer ice. It should receive more attention and publicity... (hint: in the monthly NSIDC report, Arctic Sea Ice News and Analysis).

thank you,

Paul Klemencic

Kevin O'Neill

In case I wasn't clear; we've seen the UB map show massive amounts of ice disappear - only to reappear hours or days later. We know that NSIDC uses a 5-day average, but we have to interpolate the results on a low resolution graph. MASIE uses a lot of different data sources and their analysis must take several days because their output appears misdated.

So what's our problem with IJIS? We don't understand the details of their algorithm? Umm, we don't actually know the algorithms for ANY of the products. At least IJIS does provide daily FINAL numbers. NSIDC, MASIE, UB, or IJIS: who is the most transparent? IJIS.

Kevin O'Neill

Cross posted with ya Paul :)

Kevin O'Neill

Paul, the offset compared to NSIDC doesn't bother me. As long as I know it exists I can take it into account.

The daily data is invaluable. If NSIDC or UB released their datasets I'd be thrilled. If IJIS released regional data that would be great.

Or I suppose we could learn to process the raw data ourselves and build our own algorithms :)

Ned Ward

I disagree strongly with many things in Paul's replies to my recent comments.

For now, though, I'd just like to highlight this:

Paul writes: Ned, one more thing, Who made IJIS the gold standard for extent measurement?

I specifically said that I'm agnostic on the question of which data sources are "better" or "more accurate" than others.

The only reason I tend to cite IJIS more often is because as far as I know they're the only ones that provide a handy tabular listing of daily extents going back more than 28 days (and up to the present).

I suspect that's the reason why a lot of other people use IJIS data, too. There is no judgment about "gold standards" being made.

Aside from that, I'd just encourage people to read back through the past day's comments in this thread, and form their own opinions.

Paul Klemencic

Kevin, the offset doesn't bother me either, but needs to be corrected.

One of the reasons I haven't contacted NSIDC more specifically pointing out the problem yet, is because I was afraid they would shut down the site for awhile to assess and correct the data. I didn't want to miss last week and this week's data.

William Crump

Paul K:

I was able to find some old Cryosphere Today Charts by searching through prior year posts on other blogs. It was tedious, but I will try to find the 2007 chart for the Arctic Basin again. I had some posts where Frank D and I went back and forth at Patrick;s web site and earlier in 2011 on this web site that may have contained the prior year data.

I agree it is tough to find this information and I wish Cryosphere Today maintained a set of historical files on their site - if they do, I have not found it.

The best link I have is for the central Arctic basin that has historical info is the Tivy chart below:

http://www.arcus.org/files/page/images/639/figure8.png

Note the "Big Dip" in 2007 for September for the Central Arctic Basin.

William Crump

Paul K:

Here is the Cryosphere today chart for 2008:

http://climatesanity.files.wordpress.com/2008/12/arctic-basin-sea-ice-area2.jpg

compared to the current chart:

http://arctic.atmos.uiuc.edu/cryosphere/IMAGES/recent365.anom.region.1.html

I do not see a big difference between 2008 and 2011 for the point in time in which the graphs fall below 3.5 million km2 - beginning of July, 3.0 million km2 - beginning of August, and when they settle out at 2.5 million km2 at the beginning of September.

What differences do you detect between 2008 and 2011?

William Crump

Paul K:

Below is a link to a December 2008 article that discusses 2007 and the 2008 September minimum in the Arctic Basin.

https://climatesanity.wordpress.com/tag/cryosphere-today/

Those charting the demise of the Arctic ice at the September minimum by 2016 using a data set for the Arctic as a whole need to explain why the charts for area and extent for the Arctic Basin, which constitutes more than 80% of the ice that remains at the September minimum, do not show a rate of decline sufficient for this to happen.

IMO, the Arctic wide data set includes factors that distort the rate of decline and the extrapolations with the Arctic wide data should not be used to predict the demise of Arctic ice.

I believe that a more accurate extrapolation occurs if the data for the 4 plus million km2 region known as the Arctic Basin alone is used.


Peter Ellis

"Doing fine so far," he says as he plummets past the third floor window...

We've been over this multiple times, and I think we have to agree to disagree on this one.

Paul Klemencic

William, the article you linked to is at a site that essentially is doing "soup to nuts" disinformation for deniers.

Regarding your idea, I agree that the central Arctic Basin is going to be tougher than the lower latitude ice. As you may have noted in some of my other comments here, I really, really don't like extrapolating existing trends w.r.t. Arctic ice. We know the system is changing rapidly, and often unpredictably, when compared to prior history. Given that, extrapolating any trend must be thought through thoroughly.

The key problem with extrapolating central Arctic Basin extent, is the situations where the central ice becomes exposed are intermittent. In 2007, we saw the open seas get to 85N coming from the E. Siberian region ( I like using the regions defined on the MASIE home page; it makes talking about the ice pack easier.)

This year, we have been chewing on the pack on the east side since July. Although it took a beating, it held up. But with thinner ice year after year in the Central Arctic, and earlier loss of protective ice in the seas around the NP, the central pack is more exposed each year. This means the pack will be more mobile.

Given the right wind and weather this August, we could have had open seas within 300km of the NP. Next year the right conditions could do it, but more likely by 2016, and very likely before 2020.

I don't think extrapolating the central Arctic ice extent trends will allow you to predict anything. We need to look at the process data; melt rates, sea currents, chances of wind and weather conditions, and especially surrounding seas warming up due to Arctic amplification. If we study these things, I think we can predict "ice-free" central Arctic.

My bet is these factors substantially increase the chances that the central Arctic ice will simply weaken to the point where a weather system blows the remaining ice out of the central Arctic above 85N. And this will likely happen before 2020, and almost certainly happen before 2030.

Jon Torrance

William,

This is no doubt somewhat unfair (after all, as far as I know no one anticipated the 2007 melt season) but if it were December 2006 and you had the 1981-2006 portion of the Tivy chart of Central Arctic Ocean September sea ice area you linked to as your only basis for predicting when, if ever, the September area would drop to ~2.1M square kms, you would probably have concluded that it would take at least a few decades and might well never happen. You certainly wouldn't have predicted that it would happen the very next year. As I said, no one that I know of looking at all available Arctic data did (although props to Maslowski for having the courage to make his initial observation years before then about what would happen if trends continued) but I do think it is fair to say that also considering the trends in area and extent up to then for the Arctic as a whole would have given you greater awareness that there was a significant decrease in Arctic sea ice going on and made you more cautious about predicting that ~2.1M couldn't happen anytime soon, i.e. for that specific imagined test of predictive ability, considering data covering all the northern hemisphere sea ice would have outperformed considering only data covering the ice in the Central Arctic Basin.

crandles

William, if the April maximum ice volume become less than the volume that is melted in a typical season, what do you expect to happen?

PIOMAS show a steep decline in volume but we are less sure of the actual volume. If PIOMAS figures are about right that could happen by about 2016. I think it is possible that PIOMAS could be understated by upto 25% and in that case there may be time for ice thickness feedback to kick in and make the date a long way off. If the PIOMAS numbers are right, then there won't be much time for that negative feedback to kick in and volume will keep declining.

So there is a big if about PIOMAS being right, but if we assume that for the moment, what happens between now and 2016?

The ice between Greenland, Archipelago and pole gets thinner. It thickens up in the winter but the a substantial thickness is first year ice. This ice gets pushed into Beaford Sea. This higher salinity ice reduces the melting point so the first year ice part can melt for a greater proportion of the melting season. This means a faster rate of ice edge retreat particularly from Beauford direction. Also the difficult to melt region between Greenland, Archipelago and pole becomes thinner and thinner until it is able to melt out.

I think Neven once discribed your concentration on central basis as like measuring last 3cm of a pencil and concluding it would never run out. Perhaps that is a little exaggerated as I do not expect much reduction in thickness of first year ice and this could leave the ice edge too far from the difficult to melt region to be able to melt it out.

However, I keep coming back to the view that if the volume at max is less than the volume that melts in the season then it will more or less melt out and the increased speed of retreat due mentioned above together with more first year and less multiyear ice will need less energy to melt it.

That is my explanation. What is your explanation as to why we don't get down to a meltable volume at the maximum?

Bob Wallace

Here's what PIOMAS has to say about their model...

"Our comparisons with data and alternate model runs indicate that this new trend (new model graph) is a conservative estimate of the actual trend."

Seems like they are saying that the model error is likely to be predicting volumes too high than actual volumes.

--

Seems like this Arctic Basin stuff is implying that there's something magic about that part of the ice which makes it different from the other ice.

IMO, the Basin is simply the center of the ice cube. Until the outsides are melted, the center is protected. And the outsides are melting quicker, freezing back less as the years pass.

crandles

Re: " "Our comparisons with data and alternate model runs indicate that this new trend (new model graph) is a conservative estimate of the actual trend."

Seems like they are saying that the model error is likely to be predicting volumes too high than actual volumes."

They are saying the trend is conservative.

It is possible that trend is conservative and current volume understated while past volumes were understated by a larger amount.
.

Lots of people responding to William Crump's suggestion that we need to explain. However the explanations given seem to me to vary a lot. Does that make the explanations given rather questionable? Or is it just that there are lots of different explanations?


Daniel Bailey

Re: Bob Wallace

    "IMO, the Basin is simply the center of the ice cube. Until the outsides are melted, the center is protected."

If you are implying that there is no bottom melt in the Arctic Basin sea ice then you are incorrect. Thicknesses have gone from about 2 meters in 2001 to less than 1 meter now.

Reduced thickness mean reduced internal strengths of the slabs, making them more vulnerable to wind and tidal forces. This increased churn also means a thinning of the mixing layer under the ice, bringing ever-warmer subsurface waters to the under-ice margins thereby hastening the bottom melt, etc.

Everyone should remember to evaluate the ice in that most critical of all dimensions, the thickness. As goes the thickness, so goes the volume. One day, one year, we may wake up very surprised at the "flash melt from hell..."

The death-spiral continues...

Jon Torrance

How about "Until the outsides are melted, the center is more protected than it would be if no ice at all formed outside the center during the winter."? Can we all agree on that, with the possible exception of William Crump, who appears to believe that a mystical barrier isolates the Arctic Basin from the surrounding seas?

Also, crandles, I think there are simply a variety of ways, none of them yet effective, of trying to persuade Mr. Crump that his idea is foolish.

William Crump

Jon:

Your point is well taken that using the pre 2007 data you would not have predicted the 2007 drop. I doubt you can use this graph to predict single year levels, but I do think it can be used to predict average levels over 5 to 10 year periods.

If the only data points you had were 2006 and 2007, then perhaps you could draw a trend line showing an "ice free" Arctic within the next 10 years, but this is not a valid method and this is not how the ice behaves.

In looking at the graph it is clear some years report increases from the prior year and some years report decreases, so using it to predict a single year level is not valid. This is also why the pencil sharpening example does not work. The Arctic does not decline the same amount each year

The data set clearly shows a long term decline and based on this decline I would expect the average ice levels for the period 2016 to 2020 to be below the 2007 to 2011 average, but they would not be "ice free".

While I believe that single year drops like 2006 v. 2007 are likely to occur more frequently than in the past due to the loss of multi-year ice in the Arctic Basin, this will not generate a consistently "ice free Arctic. After a big drop year, subsequent years tend to show a rebound, just like the period 2008 through 2010 did compared to 2007.

The MASIE chart for September 10 shows 2011 as being about 125,000 km2 above 2007. The ranking of years for September 10 has 2009 with the largest area followed by 2010 which is higher than 2008. While 2011 is 100,000 km2 below 2008,it is 120,000 km2 above 2007. None of these differences are all that significant as even 2007 had 2.8 million km2 of ice at September 10th.

ftp://sidads.colorado.edu/DATASETS/NOAA/G02186/plots/r11_Central_Arctic_ts.png

The point I am making is that drawing trend lines using Arctic wide data will give a false trend line as the central Arctic Basin is not declining as fast as the rest of the Arctic. While there is no barrier which separates the Arctic Basin from other regions it is clearly different. The depth of water and lack of surrounding land mass, and the increased distance from Atlantic and Pacific ocean currents make it different. The central Arctic Basin is colder than regions at lower latitudes. Unlike these other regions, the Arctic Basin receives a larger influx of ice from surrounding regions. It is the combination of these factors that makes the Arctic Basin different from the rest of the Arctic.

So while there is no barrier, I believe it is incorrect to treat this region as if it were the same as other regions.

The most effective way to show me a trend that will result in an "ice free" Arctic in the next ten years is to provide a data base concerning the rate of decline in the thickness of first year ice in the Arctic Basin. Until then, I will foolishly use the extent data, which show more than 2.9 million km2 of ice remaining and the area data which show 2.5 million km2 of ice remaining in the Arctic Basin.

In spite of the steep drop that occurred in 2007, there is a greater area and extent of ice in the Arctic Basin four years later. This is why I am not persuaded that the Arctic will be "ice free" in the next 10 years.

Seke Rob

Did you miss Larry's todays apropos of the CT setting a new record low of 2.9047396 million for Area? Bremen bulletined a extent record as well. How is that larger?

Here's my take [See Chart] on the first 253 days average. Can you see it, none of last 5 years within 1 sigma. 2007+2011. 2007 is in fact right on 2 sigma, and 2011 is further below that 2 Std.Dev.

Ice free... if one can canoe to the NP, then that's ice free to me. Last year or the year before, several guys had to be helicoptered to the NP because else they'd have had to swim the last miles.

crandles

Re "In looking at the graph it is clear some years report increases from the prior year and some years report decreases, so using it to predict a single year level is not valid. This is also why the pencil sharpening example does not work. The Arctic does not decline the same amount each year"

Seem to me this is dismissing the idea
"Until the outsides are melted, the center is more protected than it would be if no ice at all formed outside the center during the winter."

Also why can you apply
"The data set clearly shows a long term decline and based on this decline I would expect the average ice levels for the period 2016 to 2020 to be below the 2007 to 2011 average, but they would not be "ice free".
when you like but not when it makes pencil sharpening look a little more valid.

Yes pencil sharpening is monotonic decline which it shouldn't be; the pencil model needs some noise adding.

I do accept there is more to your thoughts than can be dismissed by a pencil sharpening example even with some noise added. In particular, you mention ice movement into that region.

Do you have a reference for that?
My uneducated notion of what should happen is that during the freeze season, yes more ice area moves into the Arctic basin than moves out as the ice gets squeezed up against Greenland and Archipelago. To partially compensate the ice moving out into Beauford Sea is thicker multi year ice than the first year ice moving in.

However, during melt season I would suggest that these crunched up slabs will tend to break off and take up more area so there may be a net movement outwards.

Assuming that you are right that there is net movement into Arctic basin, have you managed to quantify it to show that the net movement in is going to remain sufficient to show the Basin won't melt out in the short term?

Clearly we currently have a net loss of ice volume as multiyear ice is getting thinner and less extensive using a trend over at least several years. Do you have data showing this trend is going to end? That is other than a near negligable Arctic Basin area trend which IMHO is not very relevant when the outside protection is declining at an increasing rate.

Bob Wallace

"If you are implying that there is no bottom melt in the Arctic Basin sea ice then you are incorrect."

No, Daniel. I'm very aware that the ice is thinning, even in the least melted areas.

That statement that the Russians used to put their "ice house" research stations on 20 foot thick ice back in the 1950s and now have to look around for some two meter thick ice says it all.

I was just trying to give William a concept around which he could reorganize his thinking about why there's still a lot of two-dimensional ice in the Arctic Basin.

(I'm a major advocate of watching volume. And also paying a lot of attention to April volume as we discuss September extent. IMO, it's where we start that determines where we finish. Weather makes the bottom number noisy.)

Bob Wallace

"After a big drop year, subsequent years tend to show a rebound, just like the period 2008 through 2010 did compared to 2007."

If you take the ten largest 'big drop' years and look at recovery you find that only one of the ten rebounded back to or above that year's loss. Nine of the ten years there was less ice frozen in the following cold season than had been melted.

Average recovery of big melt volumes was 94%. 2007, a very large plunge year, was the only one to refreeze more than was lost.

For the ten 'little drop' years the average volume rebound was 101%. The middle ten melt seasons showed recoveries of 98%.

The big drop years seem to be melting away ice which is not recovered.

Bob Wallace

William, might I offer another conceptual way to look at what is happening (I think) to the Arctic Basin?

Think of Arctic sea ice as a big angel-food cake pushed up against a wall. Vermin are able to get to the sides, the top and even crawl underneath and chew away.

As time goes on that cake shrinks in outside dimension. It gets eaten away on top and bottom.

The last to go will be the part up against the wall. The critters will have to eat their way there....

The comments to this entry are closed.