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


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

Paul Klemencic

Wow, Ned, you really went off there. Do you really think a correlation of 0.5 is that good? What was the correlation for Day 0 of the most recent run? Was it even 0.5?

Anyway, I had spent the last few hours going over the NSIDC daily data, which I didn't get until just this last week, and did compare MASIE and NSIDC daily numbers, with a preliminary first pass look, then I wrote a response offline, and I came here to post it, and there is your comment…

I am going to post it the way it read, not changing a word:

OK, Ned, I am throwing in the towel. But not in the direction you think, and not for the reason you want.

I looked at the NSIDC numbers for July, August, and September last year, and the July and August numbers this year. I was going to check the correlation you calculated, but before I did that I simply wanted to see the daily "error", using NSIDC as the benchmark and the MASIE as the variable. I got a shock.
These numbers are preliminary, and I need to look day by day to check for gaps, but the results so far are pretty compelling. Its too bad we didn't have daily NSIDC SIE data earlier in the season.

Last year the daily errors came in as I would expect:
July 2011 NSIDC was higher by an average of 220k with the root mean square error (RMSE) = 224k, since there were two days with negative errors.
July 2011 RMSE= 224k
August 2011 RMSE= 247k
September 2011 RMSE= 200k (NSIDC and MASIE paths crossed)

I was able to reduce the errors, and match observable events (storms, date of minimum, unusual pack movements) by introducing a 5-day lag for the MASIE reported data.
5d lag July 2011 RMSE= 214k versus 224k
5d lag August 2011 RMSE= 121k versus 247k
5d lag September 2011 RMSE= 101k versus 200k

I expected July 2012 to more or less match July 2011 since the other SIE measures tracked very close to July 2011 over the month. And if there were bigger errors they would be due to the longer lag.

But alas, no. MASIE showed a much bigger RMSE in July of this year, than last year, with RMSE in July hitting 363k and then in August skyrocketing to 715k so far. Even before the storm, the RMSE hit 436k on the one first few days in August.

These differences are much higher than last year. Since MASIE has only been around since 2010, we simply don't have the track record for this report, to compare and spot discrepancies. Introducing a lag won't be able to fix this large an error.

Here are the numbers, and note that using a 12d lag to correct August 2012 worked well, but threw off July:
12d lag July 2012 RMSE= 540k versus 362k
12d lag August 2012 RMSE= 112k versus 715k so far…

I still see a 12d lag in response to timed events like the GAC-2012, but clearly MASIE has bigger problems than a simple reporting lag. And we won't be able to reverse-engineer these problems like you said.

MASIE should still see a delayed minimum, but I can't begin to guess how that minimum will compare with Bremen or NSIDC, let alone IJIS.

I'm finished until after the minimum hits.

Paul Klemencic

Clarification to last comment upon re-reading it: I am agreeing with your earlier observation Ned, that we won't be able to reverse-engineer the problems in MASIE.

Peter Ellis

Relevant here, from The Other Place:


"A quick clarification on the different sources: the quote above refers to concentration, which will affect \”area\” estimates. NSIDC uses \”extent\”, which counts any ice above 15%, so a concentration bias won\’t affect the detection of ice extent except at low concentrations. There is still some ice missed by passive microwave sensors using the extent threshold, but it\’s generally much less than for area. NSIDC and the IMS/MASIE extent estimates generally agree well, with NSIDC being a bit lower. This year is unusual in that there is a lot of very low concentration ice in the Chukchi Sea that passive microwave is not seeing, but IMS/MASIE analysis has been counting as ice. That ice will likely melt completely in the next couple of weeks and the area will open up in IMS/MASIE.

A key point is that IMS/MASIE and other operational sources, such as from NIC, use a variety of data sources that are inconsistent in quantity and quality, as well as subjective human analysis to create maps of ice. A primary purpose of these maps is to support navigation in ice-infested waters. So they tend to be conservative and count even areas sparsely covered with ice as \”ice-covered\”. The passive microwave data is produced by completely automated processing that is consistent over the entire record dating back to 1979. Thus, while absolute estimates of ice cover may be biased, the trends and variability (e.g., comparing records, determining a record low extent) is more accurate than from using operational sources.

Walt Meier

Paul, will you argue with the head of the ice service at NSIDC, or will you give over?

Paul Klemencic

Peter Ellis: This is exactly what I didn't want to happen. I didn't want the problem to become this public, which is why I used this old thread to discuss it.

I especially don't want any discussion with Steve Goddard, who believes that water exists at its triple point in the Arctic (he can't read a phase diagram), doesn't understand either Henry's law, Raoult's law, or Dalton's law; and doesn't know how to estimate the thermal energy released by air and water vapor as they cool… and doesn't want to seem to want to learn. He brings to mind Mark Twain's famous quote: "Never argue with an idiot, because the onlookers may not be able to tell the difference."

I have spent an inordinate time studying this problem, and have talked to Dr. Meier's people, and there is nothing I want to say publicly. I am going dark on this issue.


Paul, it's really unimportant, but just to acknowledge your earlier comment - I know you thought you answered my question, but you didn't - you avoided it and asked a different question in return (it's easy to get distracted by minutiae). Subsequent comments by folks here, and most importantly by Walt Meier on the ASI update 10 post, do not support your case, and does support the idea that there is a fair bit of really low concentration ice out there, read by IMS and MASIE, and not by Bremen or IJIS. But lets leave this until the dust has settled on a terrible summer for Arctic ice.

Peter Ellis

Paul, I suggest you get over yourself with the "didn't want to to become public" angle. It's BEEN public for years - upthread I pointed you to some of the published, peer-reviewed research on it. Journal articles are hardly private. Moreover, scientists are generally more than happy to discuss this sort of thing.

Case in point, here's some further clarification from Walt Meier, this time at WUWT.

"I’ll make a few points for clarification on the post above. First, MASIE and IMS are the same product. MASIE is simply a repackaging of the IMS data in easier to use formats. IMS is produced by the National Ice Center (NIC), using similar sources and methods as they use for their daily interactive maps. So all three of the examples provided are closely related and not independent measurements.

The passive microwave estimates all show a record low for the Arctic. These aren’t completely independent either – they all measure microwave emission, but there are difference sensors (SSMIS, WindSat, and for the first time AMSR2: http://www.eorc.jaxa.jp/en/imgdata/topics/2012/tp120825.html, which is pretty exciting), and there are different processing methods as well. There can always be potential errors in data, especially in near-real-time, so having multiple sensors showing consistency provides confidence that one sensor doesn’t have an error, which has happened from time to time. When it does, we go back and reprocess and correct the errors.

I worked at the National Ice Center for a couple years and have collaborated with them many times since, so I’m familiar with their methods and their focus. Their mandate is to map as much as ice as possible as accurately as possible each day and week in support of ships (particularly DoD ships) operating in and near ice-covered waters. They work hard on getting today’s data analyzed and then tomorrow they start over. They are not concerned with the past. If they can detect more ice today than yesterday, then they map it. If they lose a sensor, they do the best they can with what they have left. If they make an error, they don’t go back and correct it -it’s on to the next day. NIC doesn’t discuss climate or climate change because that is not their purpose and from my experience working there, they just don’t have the time – they’re focused on the here and now.

The charts are produced manually, so there is subjectivity in the analysis that we don’t have in our fully automated processing. This means that there can even be inconsistencies in adjacent regions if they were analyzed by different people.

They have created an archive of their weekly ice charts, which is archived at NSIDC: http://nsidc.org/data/g02172.html. There was some attempt to homogenize the charts (at least remove regional discrepancies) during the production, but they do not produce a consistent timeseries. MASIE, though it only goes back to 2006, has similar issues of consistency.

The folks at NIC do a great job at what they’re focused – navigational support. MASIE is an excellent data set and we at NSIDC find it very useful looking at specific details about the ice (e.g., is the Northwest Passage open or not), but the NIC products are not applicable to studying climate-scale changes.

Walt Meier

Peter Ellis

Watching the IMS charts over the last week or so, it's really clear how only parts of the ice boundary get updated each day. Much of it stays pixel-for-pixel identical from day to day.

Presumably the updates are applied to areas where new data becomes available - perhaps an overflight of a radar satellite in a cloudy area, or some such.

It's clearly designed for operational / precautionary purposes.

Rob Dekker

Thank you for correcting Anthony at WUWT, where he desperately tried to find 'multiple' products that sustained his belief that records were not broken.

And so typical for Anthony, rather than admitting a false assertion, he has to have NSIDC's Walt Meier himself come over there to explain it to him personally.

Then, rather than thanking Meier for correcting the mistake in Anthony's already heavy cherry-picked selection of products, mistake, Meier, Stroeve and Serreze are being rediculed and insulted by the various posters there, including the sock-puppet of a WUWT moderator...

Artful Dodger

Hi Rob

Denial is like digging a basement in your houseboat: steady work, with a chance of drowning.

Now back to that river in Egypt...


Ned Ward

Yeah, that thread over at WUWT was appalling.

My impression was that in the past, Anthony made at least a half-decent effort to be polite to Meier and Stroeve -- presumably in an effort to keep them around, since they're pretty much the only actual scientists in any field willing to post on Anthony's site.

But this year's sea ice fiasco seems to have driven Anthony crazy. His comments are even more absurd and erratic than usual, and he was downright rude to Meier and Stroeve.

Anthony has a huge emotional investment in "it's not happening!" When reality so blatantly contradicts that line, he will (understandably) have difficulty coping with it. That's just human nature.

Ned Ward

Peter, thanks for posting that comment from Walt Meier. This is exactly the point you and I and others have been trying to make in this thread. The sea ice maps derived from fully-automated processing of microwave radiometer imagery are fundamentally different from the sea ice maps derived from manual interpretation of visible, infrared, SAR, and microwave radiometer imagery.

Both types of products are interesting and worthwhile. It's not that one is clearly "right" and one is clearly "wrong". The end-user just needs to understand the characteristics, strengths, and limitations of each data set ... and not try to make inappropriate comparisons between apples and oranges.

Artful Dodger

Uni Bremen has updated their SIE chart as of 19 Sep 2012:


Looks like it came in just a smidgen under 3.0 M km² SIE.

2007/11 were a virtual tie at around 4.2 M km² SIE.

Note that this chart is based on SSMIS data rather than AMSR2, so inter-annual comparisons are fair.

Peter Ellis

No, the time series graph splices together the data from multiple satellites, see the table at the bottom of this page:

In fact, they specifically say it's not comparable and that this year will be (relatively) underestimated.

L. Hamilton

Although UB cautions against comparing their current values with past years, the current ones can fairly be compared with each other. Their time series hit its lowest point this year (so far) on 9/16, and two days later stands 80k above that mark.

With the new AMSR-2 data there should be opportunities to retrospectively adjust the past year's values so the UB time series has better continuity going forward.

The comments to this entry are closed.