Foreword: I had originally planned to post a story on this, but Steven Goddard of the UK Register sends word that he has already done a comparison. It mirrors much of what I would have written. There is a clear discrepancy between the two data sources. What is unclear is the cause. Is it differing measurement and tabulation methods? Or, is it some post measurement adjustment being applied. With a 30 percent difference, it would seem that the public would have difficulty determining which dataset is the truly representative one.
UPDATE: The questions have been answered, see correction below – Anthony
Arctic ice refuses to melt as ordered
Published Friday 15th August 2008 10:02 GMT – source story is here
Just a few weeks ago, predictions of Arctic ice collapse were buzzing all over the internet. Some scientists were predicting that the “North Pole may be ice-free for first time this summer”. Others predicted that the entire “polar ice cap would disappear this summer”.
The Arctic melt season is nearly done for this year. The sun is now very low above the horizon and will set for the winter at the North Pole in five weeks. And none of these dire predictions have come to pass. Yet there is, however, something odd going on with the ice data.
The National Snow and Ice Data Center (NSIDC) in Boulder, Colorado released an alarming graph on August 11, showing that Arctic ice was rapidly disappearing, back towards last year’s record minimum. Their data shows Arctic sea ice extent only 10 per cent greater than this date in 2007, and the second lowest on record. Here’s a smaller version of the graph:
The National Snow and Ice Data Center (NSIDC)’s troublesome ice graphThe problem is that this graph does not appear to be correct. Other data sources show Arctic ice having made a nice recovery this summer. NASA Marshall Space Flight Center data shows 2008 ice nearly identical to 2002, 2005 and 2006. Maps of Arctic ice extent are readily available from several sources, including the University of Illinois, which keeps a daily archive for the last 30 years. A comparison of these maps (derived from NSIDC data) below shows that Arctic ice extent was 30 per cent greater on August 11, 2008 than it was on the August 12, 2007. (2008 is a leap year, so the dates are offset by one.)
Ice at the Arctic: 2007 and 2008 snapshotsThe video below highlights the differences between those two dates. As you can see, ice has grown in nearly every direction since last summer – with a large increase in the area north of Siberia. Also note that the area around the Northwest Passage (west of Greenland) has seen a significant increase in ice. Some of the islands in the Canadian Archipelago are surrounded by more ice than they were during the summer of 1980.
The 30 per cent increase was calculated by counting pixels which contain colors representing ice. This is a conservative calculation, because of the map projection used. As the ice expands away from the pole, each new pixel represents a larger area – so the net effect is that the calculated 30 per cent increase is actually on the low side.
So how did NSIDC calculate a 10 per cent increase over 2007? Their graph appears to disagree with the maps by a factor of three (10 per cent vs. 30 per cent) – hardly a trivial discrepancy.
What melts the Arctic?
The Arctic did not experience the meltdowns forecast by NSIDC and the Norwegian Polar Year Secretariat. It didn’t even come close. Additionally, some current graphs and press releases from NSIDC seem less than conservative. There appears to be a consistent pattern of overstatement related to Arctic ice loss.
We know that Arctic summer ice extent is largely determined by variable oceanic and atmospheric currents such as the Arctic Oscillation. NASA claimed last summer that “not all the large changes seen in Arctic climate in recent years are a result of long-term trends associated with global warming”. The media tendency to knee-jerkingly blame everything on “global warming” makes for an easy story – but it is not based on solid science. ®
Bootnote
And what of the Antarctic? Down south, ice extent is well ahead of the recent average. Why isn’t NSIDC making similarly high-profile press releases about the increase in Antarctic ice over the last 30 years?
The author, Steven Goddard, is not affiliated directly or indirectly with any energy industry, nor does he have any current affiliation with any university.
NOTE OF CORRECTION FROM STEVEN GODDARD:
The senior editor at the Register has added a footnote to the article with
excerpts from Dr. Meier’s letter, and a short explanation of why my analysis
was incorrect.
To expound further – after a lot of examination of UIUC maps, I discovered
that while their 2008 maps appear golden, their 2007 maps do not agree well
with either NSIDC maps or NASA satellite imagery. NSIDC does not archive
their maps, but I found one map from August 19, 2007. I overlaid the NSIDC
map on top of the UIUC map from the same date. As you can see below, the
NSIDC ice map (white) shows considerably greater extent than the UIUC maps
(colors.) The UIUC ice sits back much further from the Canadian coast than
does the NSIDC ice. The land lines up perfectly between the maps, so it
appears possible that the UIUC ice is mapped using a different projection
than their land projection.
Click for larger image
Because the 2007 UIUC maps show less area, the increase in 2008 appears
greater. This is the crux of the problem. I am convinced that the NSIDC
data is correct and that my analysis is flawed. The technique is
theoretically correct, but the output is never better than the raw data.
Prior to writing the article, I had done quite a bit of comparison of UIUC
vs. NSIDC vs. NASA for this year. The hole in my methodology was not
performing the same analysis for last year. (The fact that NSIDC doesn’t
archive their maps of course contributed to the difficulty of that
exercise.)
My apologies to Dr. Meiers and Dr. Serreze, and NSIDC. Their analysis,
graphs and conclusions were all absolutely correct. Arctic ice is indeed
melting nearly as fast as last year, and this is indeed troubling.
– Steven Goddard
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Jeff (22:04:53) says:
Otherwise, your results don’t agree with his.
But they do on the two images that were used in his article. Hopefully he will be back at some stage to clarify things.
Meanwhile perhaps we will get an independent verification or refutation from RealClimate.
I should mention that there is a lot of behind the scenes communications going on between Dr. Meier, myself, and Mr. Goddard.
Dr. Meier has been most gracious with his interactivity. I hope to have more to report soon.
The basic problem here is that the UIUC/CT images are a very poor representation of the sea ice data. Between projection issues (the data have been re-projected at least twice, and the final the projection is undefined), JPEG compression artifacts (which make it impossible to do a quantitative analysis), anti-aliasing/smoothing of the texture map (look along the ice margins), and an imprecise color palette (Mr. Goddard, could you give me the rgb value you use for 30-15% ice concentration on the UIUC maps? I can’t find any pixels below 30% on the images)–it’s not appropriate to use them for data analysis.
It’s now been 6 days since this article first appeared. Not a single person has stated what the Earth areas of the sea ice coverages on Aug. 12, 2007 and Aug. 11, 2007 in the data set that UIUC/CT used is. Without the actual areas, how does anyone know what the difference is?
Two days ago, I asked where the missing ice in NSIDC’s data is. All you have to do is compare the maps (and I’ve posted where NSIDC’s maps are) and show the area(s) where there’s ice in the UIUC/CT maps but not in the NSIDC maps. No one has done this. Why?
Looking at the UIUC/CT and NSIDC ice area graphs, it looks like the ice area typically decreases this time of year by a little over 1 percent per day this time of year, with occasional spikes of a few percent. Yet according to Steven Goddard’s results, the difference between corresponding days in 2007 and 2008 changed by more than 5 percent on half the days that he measured. How can the difference change by significantly more than the ice itself changed?
Simmon (05:33:44) says:
The basic problem here is that the UIUC/CT images are a very poor representation of the sea ice data…
While I can’t disagree with any of that, I do believe they are good enough that the discrepancies cannot be accounted for just by the factors you mention.
I can’t find any pixels below 30% on the images
Not many for sure. I found 110 pixels in 15-30% range on 20080811, in a few scattered pockets. Is this typical of sea ice? If the maps are missing a big chunk of low concentration ice that could certainly screw up extent values.
Jeff (06:56:20) says:
All you have to do is compare the maps…
Sorry, your links have scrolled into the misty past on this very long thread while I was counting pixels. The main difficulty in comparing maps from different sources is change in projection.
dipole
“Sorry, your links have scrolled into the misty past”
ftp://sidads.colorado.edu/pub/DATASETS/seaice/polar-stereo/nasateam/final-gsfc/browse/north/daily/2007/nt_20070812_f13_v01_n.png
ftp://sidads.colorado.edu/pub/DATASETS/seaice/polar-stereo/nasateam/near-real-time/browse/north/nt_20080811_f13_nrt_n.png
A fine correction — about what one would expect from a science professional without an agenda to push. I’ve noticed, in comparing the NSIDC maps, that ice areas come and go; not so much the melting but from the movement of ice floes due to wind and currents. I’m still curious to see how the refreeze progresses this winter and how it’ll compare to the refreeze from the past winter.
Anthony Watts: Thank you for publishing the correction, and thank you for pursuing this, and not using a hit and run approach to this story.
Regarding some posted comments: The Northeast Passage seems to have opened today;
http://www.iup.uni-bremen.de:8084/amsr/arctic_AMSRE_nic.png
The Northwest Passage still shows some ice in the Viscount Melville Sound, and in the last week ice floes (50% ice) has been pushed down to block the end of the McClure Strait. Last year the NW Passage cleared on August 21; this year, the inland portion should be clear in the next few days. There is a ship in the NW Passage:
http://www.sailwx.info/shiptrack/shipposition.phtml?call=DBLK
Steven Goddard,
As one of those who has rather pursued you over this matter on this thread, may I offer my respect to you for the correction you have published.
With regards,
S.
I second Steven Talbot’s sentiment.
The UIUC data was misleading? Then maybe I was wrong about the method not working?
Yes – a good resolution of this issue.
Jeff (19:23:21):
The UIUC data was misleading? Then maybe I was wrong about the method not working?
Given the absence of detail it was not unreasonable to be sceptical ;^). The 30% extent difference was also verified independently by another post on RealClimate, as well as further up this thread. Add that to the visual evidence from masking the matching pixels and it seems to me that pixel counting can be accurate and reproducible on the UIUC images.
So this does seem to put the spotlight on the pre-2008 UIUC images. Did they discover an error in their mapping process which they silently fixed without updating the old images? Since the images are made to look like false-colour photographs it is natural for the unwary to interpret them as photographic evidence.
I posted to this effect on the corresponding RealClimate thread, but was stuck in moderation for several hours and now the site is down.
dipole, you missed the point. Mr. Goddard’s calclulated 30% increase was a sure sign that the calculation was wrong, as the sea ice extent was only higher by 10-14% than last year on August 11. A lot of the commenters here realized Steven Goddard’s result was wrong, right off the bat, because the result was contradicted by other sources.
We still don’t know if Mr. Goddard used the correct method. A key piece of input data, the UIUC map from a year ago is different because it was apparently generated by a different method than this year. But we still don’t know if his pixel count method would have worked, using the correct input. We still don’t know the projection and method UIUC uses to generate their image, unless I missed it in all the detail above.
But if we go to the data used to generate the maps and images, the sea ice extent has only increased about 10-14%, NOT 30%. This is the key.
More importantly, all the general conclusions in the article above, relating to sea ice this year, have been pretty negated, including the title “Arctic ice refuses to melt as ordered”.
Arctic sea ice area has continued to drop since this article was published on the net, and the sea ice area (a better measure than extent), is within 600,000 to 700,000 sq km of the minimum set in mid-September last year.
Not likely to set a new low, without some unusual weather, but clearly should set the 2nd lowest NH sea ice area in the record.
Paul K,
The pixel counting technique is done properly and works fine. I have found a few NSIDC maps from 2007, and when I apply the same methodology I get an exact match vs. their graphs.
BTW – those of you concerned about the day to day variations in measurements should look at this. There is a lot of imprecision in the raw ice measurements.
http://nsidc.org/data/seaice_index/images/daily_images/N_unsmoothed_alldata_timeseries.png
Paul K (22:49:45) says:
dipole, you missed the point. Mr. Goddard’s calclulated 30% increase was a sure sign that the calculation was wrong,
‘Scuse me? Or one of his data sources was wrong. As he says:
while their (UIUC) 2008 maps appear golden, their 2007 maps do not agree well with either NSIDC maps or NASA satellite imagery.
The CT site invites you to compare images from 2 different dates. It’s not unreasonable to assume that the images are in fact generated similarly and directly comparable. It appears this is not the case, and this was the undoing of his argument, not the pixel counting.
UIUC maps seem to compare very well with the high resolution images obtained with the AMSR-E imager, see below for a comparison on 8/11/07:
http://iup.physik.uni-bremen.de:8084/amsredata/asi_daygrid_swath/l1a/n6250/2007/aug/asi-n6250-20070811-v5_nic.png
http://arctic.atmos.uiuc.edu/cryosphere/deetest/deetmp.7104.png
The NSIDC image looks rather similar too to be honest when one considers its lower resolution (for day earlier):
http://nsidc.org/news/press/2007_seaiceminimum/images/20070810_extent.png
Hi Folks,
There is no difference between the data or the way the 2008 and 2007 images were produced in the comparison images on the Cryosphere Today. The apparent differences Mr. Goddard observed between the NSIDC values and those produced comparing images from the CT are almost entirely due to the mistake of using pixel counting to compute area on severely distorted satellite projections.
Minor differences may come from:
(1) NSIDC uses a longer temporal averaging (around 10 days, I think) compared to the CT single-day plots.
(2) NSIDC uses a 15% threshold data cutoff; the CT cuts off concentrations below 30% in the comparison images.
Still, the above only account for percentage change differences of ~3%. The majority of the apparent difference comes from projection fun.
The problem with the satellite projection is that the center of the images are distorted when compared to the periphery of the images. The distortion is such that a ‘feature’ in the middle of the image (higher latitudes) has more pixels representing it than the same feature would at the periphery of the image (lower latitudes). The ‘feature’ in this case can be thought of as the ice extent difference. I am referencing a schematic I drew that helped me understand the issue:
http://arctic.atmos.uiuc.edu/cryosphere/equal.area.projection.png
The satellite projected images should not be used to compute area/extent (or differences between area/extent from one year to the next). When I use the raw equal area grids as they come from NASA, before reprojecting onto my satellite projection, I get a 2007/2008 difference of around 10% in ice area around August 11 – consistent with what NSIDC is reporting. This has reduced to around 5% at present.
If my understanding of the issue is complete, a similar difference in winter, when the ice edge was nearer to the periphery of the CT satellite images, would be misrepresented in the opposite sign (smaller) than that computed on a standard equal-area grid (ala NSIDC).
As with any map projection comparison, there is nothing ‘wrong’ with any of the different projections, they all have their place. Problems do sometimes arise when comparisons of summary statistics computed on one are made with similar numbers from the other, as is the case here.
REPLY: Thank you for your posting Mr. Chapman. Here are questions for you –
1) Why don’t you answer your email contact link on the CT web page? I and other have sent queries this year, with no response. For example, I sent an email two days ago.
2) Why do you have a quote from a politician (Al Gore) on a web page presenting science? This is a question many people have raised.
3) What prompted the color scheme change in recent days
Thank you for your consideration – Anthony Watts
so now i am wondering if the incorrect UIUC 2007 maps will be corrected…. suspect they have already caused quite a few drama’s….recently saw a local doco (ABC 4 corners) which seemed to use the incorrect data as the basis of its presentation. Total sea ice since 1979 doesn’t seem to have changed much …if this data is correct??
so its all a bit of a storm in a tea cup…but would expect a small loss of ice due to our current position in regard to total sunspot activity trend.
William Chapman,
Thanks you very much for your fine explanation. The problem I (and others) had was that the side by side comparison link from your web site showed visually a large difference between August, 2007 and August, 2008. August 15, 2007 vs. August 14, 2008 is a good example.
Pixel counting is simply a way to quantify what is processed by the brain when viewing an image. In this case, the 2008 image has 41% more ice pixels than the 2007 image. NSIDC extent graphs and UIUC area graphs both correctly show about 10% difference for that date.
This is a substantial difference (41% vs. 10%) and misled me and numerous other people. Perhaps you should put a disclaimer on the page that the images are not proportionate and can not be directly compared? Having a side by side comparison is quite misleading in the absence of such information.
1) Why don’t you answer your email contact link on the CT web page? I and other have sent queries this year, with no response. For example, I sent an email two days ago.
limited time – sorry.
2) Why do you have a quote from a politician (Al Gore) on a web page presenting science? This is a question many people have raised.
didn’t realize it was a concern for many people. All references to Al Gore have been removed.
3) What prompted the color scheme change in recent days
I added three new color schemes about 40 days ago (July 11; is that ‘recent’?). I was hoping for more detail in the images “from the satellite perspective” in the images shown on the main page. The AMSR-E data provide more spatial resolution so I switched data sources and color schemes for those home page images. IMPORTANT: The data used for all other timeseries and comparison graphics have stayed the same (SSMI) obviously, to avoid any issues with data inhomogeneity in time. The AMSR-E data source is only used for the high resolution Northern Hemisphere graphics on the main page. I hope to convert the Southern Hemisphere as well over the next month. The AMSR-E is a relatively new platform, so maybe after it has been around for 10-15 years or so, and has a proven track-record, we can switch the timeseries and other data over entirely to that platform. I have included links to the old SSMI images on the main page for those who prefer them or want to compare current conditions to historic conditions (prior to the AMSR-E launch).
William Chapman (07:27:26) says:
Hi Folks…
Thanks for your input, and especially the confirmation of the consistency of your image sequence. Sorry to have questioned this. I have an amateur interest in map projections and had tried to allow for the distortion as described elsewhere on this thread.
Anyway you have given me an incentive to have another look – the exact quantification of the error in the original article remains an interesting puzzle and it would be nice to tie up the loose ends. Is there a pointer somewhere to the exact projection used?
Uhoh. Escaped italics.
Anthony, Mr. Albert Gore Jr. hasn’t held a political office for almost eight years. He has been active in advancing scientific efforts for over 30 years. Why isn’t his view on scientific developments appropriate, especially in view of his Nobel prize? None of the (sometimes dubious) sources of information you post here have a Nobel, that we know of. In fact, we don’t know anything about the author of the current article under discussion.
Are you saying that science information sites shouldn’t have anything related to various policies discussed? This would be a serious limitation on analyzing alternatives.
Or are you saying that former politicians who held government positions, shouldn’t be involved in activities involving government entities? If this is the case, would you object to former presidents like George H.W. Bush or Bill Clinton addressing a high school class? This also cuts off the rights of such persons from participating in many activities after their period of public service is completed.
I don’t understand your point, nor how it is relevant to the current discussion.
REPLY: And I don’t understand yours. The less politics involved in science the better. Gore has been, and still is a politician. See this:
Story here: http://www.iht.com/articles/2008/08/19/america/gore.php
So shall you now try to tell me that his “coveted slot” (to quote the article) at a political convention next to the potential next president of the United States is not political work on his part but is science?
[…] Kudos to Mr. Chapman for his willingness to consider the issue, and for acting quickly when it was pointed out. You can read the original comment here […]