This article is offered by Jonathan Drake. It discusses the unusual changes in trend noted in sea ice at various points of the historic record paying particular attention to the satellite data. I think it is worth posting, but I have varied opinions and questions regarding the different sea ice datasets displayed here. To me it could be characterized as ‘unusual trends’, as something is certainly up in the longer term combination of data.
Rather than reformat the PDF, I have uploaded it here: Could Instrumentation Drift Account for Arctic Sea Ice Decline:
the Air Vent 
Adjustments, like Einstein’s constant or parameter fitting, make me uneasy. Always, without exception, when I have adjusted data it has always been a mistake, the “dirty” data always has the clues that I need to solve the problem.
Is there anyway to do your sea ice simulations with completely unadjusted data?
I think the short answer is “no”, but I don’t have time to generate a longer version. It’s still an interesting question.
Data should not be treated as if they have been delivered to us on stone tablets by some smelly bearded shepherd.
The idea that trends in sea ice are an artifact of the fact that they are measured with satellites is interesting, but it is mostly idle speculation. More traditional measurements didn’t stop when the satellite era began. A random google search for example leads to this, just as one example.
Click to access Handbook%20for%20Sea%20Ice%20Analysis%20and%20Forecasting.07.pdf
Until someone presents some evidence that traditional observations are *in serious conflict* with the satellites (and Drake offers no such evidence), skeptics would be wise to be skeptical of this sort of left-field idea.
Doc it is good to hear from you. There is a paper on NSIDC site which claims brightness temperatures of the EMSR in the Antarctic matches well to observaiton yet there was a 2million sq km loss in the early portion of that sat record 1972 – 1978. I know this sat data is very questionable, however, I don’t really know if it matches observation well. Is there also a human observatoinal source which matches this huge sea ice loss in the Antarctic?
The paper is here:http://www.atmos.umd.edu/~kostya/Pdf/Seaice.30yrs.GRL.pdf
Figure 1.
I am curious if this serious loss in ice area actually happened?
I came to similar, but opposite, conclusions to Dr. Steig when I very briefly (!) researched and analysed the sea ice data for a few weeks, last year.
Basically, I found the Walsh dataset to be rather qualitative in nature. From what I could tell, a good deal of effort went into it, and Walsh et al. deserve credit for attempting the job. But, Walsh’s estimates don’t seem to be directly comparable to the satellite measurements.
Indeed, the Walsh dataset is itself a composite dataset (as is the satellite record of course), and I’m not convinced that the earlier parts of the record can be meaningfully compared to later parts. As far as I remember, NSIDC reached a similar conclusion, and strongly cautioned using the pre-1953 section.
It’s been a while since I looked at it, but I think the Hadley Centre did a bit of picking & choosing from the Walsh data set too. I think they replaced the 1871-1900 and 1940-1952 periods with “climatologies” (i.e., they calculated the average values for those periods and then used those values for every year in that period). Then for the 1973-Oct 1978 period they switched to early satellite estimates.
It’s been a while now, but as far as I remember, much of the apparent trends in the Hadley Centre’s dataset UP TO the post-1978 satellite record seemed to be associated with changes between those periods. That made me very wary of making direct comparisons between different periods.
I am cautious of relying on composite datasets in general, particularly those constructed using periods of minimal overlap.
So, I am reluctant to treat either the satellite record OR the Walsh dataset as continuous records. But, combining the two together seems pretty reckless, and would need a LOT more justification and consideration than the current “well, we want to make as long as record as we can, so why not but all the apples & oranges together in a row”…
There’s a description of the Hadley Centre’s analysis in Rayner et al., 2003: http://badc.nerc.ac.uk/data/hadisst/HadISST_paper.pdf
I was not aware of that paper — I will take a look.
Note that for Antarctica indeed matching observations with satellites is much more difficult than in the Arctic. There just aren’t that many people there most of the time, and mostly all we have is satellites. However it is quite unambiguous what is sea ice and what isn’t in satellites.
I can fully accept that explanation. It is very difficult for those of us not immersed in the field to know if the data doesn’t exist or if we just don’t know about it.
Thank you.
Data with cyclic behavior can create spurious trends depending on the points in teh cycle that the region being analyzed starts and stops. The trend used complete years, starting and ending at the same date. But the natural cycle being analyzed is biased high to start the year and low to end the year. I think the natural cycle and short number of cycles being evaluated has caused a spurious trend. The lower low values of each cycle suggest some decay, but the trendline presented is effected by the natural cycle of each year. For example, if you drew a linear trend using any single year’s data set, the line would suggest a dramatic decay; but that is only due to the cycle itself.
I’d guess that a more acurate trend would be to average all the points for a given year, i.e. each full cycle, and then calculate the trend based on the annual decay rate indicated by those values.
The same satellites in decaying orbits are reporting relatively constant Antarctic sea ice and shrinking Arctic sea ice. Any correction based on orbital decay is likely to show Antarctic sea ice rising. Although the Antarctic Peninsula represents only a small fraction of the area where seasonal sea ice forms in the Antarctic, there is strong evidence for surface warming and loss of permanent floating ice in this region.
The transitions between satellites occur in different seasons, when difficulties in distinguishing ice and water are quite different.
The satellite transitions are calibrated separately over northern and southern hemispheres:
http://www.remss.com/papers/tech_reports/Wentz_SSMI_Version6_Calibration.pdf
see figure 5: Annual sea-ice extent (area within the ice–ocean margin) derived from a new
‘Zakharov’ sea-ice data set (red), Northern Hemisphere sea-ice extent from the widely
used ‘Walsh’ data set (green) and zonal (70–90◦N) mean annual SAT (black) since
1900. The time series shown are 5-yr running means.
http://www.tellusa.net/coaction/index.php/tellusa/article/viewFile/14418/16180
forgot:
http://nwpi.krc.karelia.ru/e/climas/Ice/Ice_no_sat/Hice-no-sat.htm
John Bills, there’s some interesting data in that paper, thanks! I’m also interested by the ice thickness section (3.2 & Fig 6).
However, I really don’t get how they reached their conclusion: “Secondly, we believe there are strong indications that neither the warming trend nor the decrease of ice extent and volume over the last two decades can be explained by natural processes alone.”
Their data seems to suggest to me that much (if not all!) of the variability can be explained by natural processes.
Indeed, the lack of agreement between Fig 1a and Fig 1b should have been enough to make the authors question the reliability of their ECHAM4 GCM.
Can anyone explain to me what I’m apparently missing?
Any idea on the overlap in coverage between the Zakharov & Walsh datasets over time?
I see they say the Zakharov dataset covers ~77% of the Arctic & they blame the lack of dramatic melting during the satellite period on the fact that it doesn’t cover the Beaufort Sea or Chukchi Seas. But, the measurements used seem to have varied over time (e.g., mostly shipping lanes pre-1930s, mostly airplanes post-1940s). So, I suspect even that coverage underwent significant changes over the entire record. I had similar concerns over the Walsh dataset.
As Eric Steig & Jonathon Drake both implied above (possibly with different expectations), I also wonder what the post-1973 Walsh dataset would look like if they hadn’t started using satellites…
P.S. That http://nwpi.krc.karelia.ru link doesn’t seem to work. What does it lead to? The Zakharov dataset?
I meant JonathAn
I see Karelia is down for the moment
This one is very interesting (but down right now):
http://nwpi.krc.karelia.ru/e/climas/Ice/Ice_no_sat/Region.htm
Look at figure 5 and see the overlap: http://www.tellusa.net/coaction/index.php/tellusa/article/viewFile/14418/16180
It’s still down, but I’ll check again later.
I see the overlap/lack of overlap in Figure 5.
I’m curious what the overlap in the DATA used to construct the estimates is.
Clearly, the Walsh dataset covers a larger area (~12 million sq. km of ice extent vs. ~8.5 million sq. km of ice extent). But, the paper says the Zakharov estimate also includes new Russian data. So, what is the overlap in data sources between the two. Also, their data sources vary over time. How much of the apparent variability is due to fluctuations in data sources?
Recall that for terrestrial/aerial measurements, the estimates involve a lot of interpolation – the ships/planes/submarines aren’t sailing along the entire circumference of the ice extent! As far as I recall, they’re just noting ice that they observe on their voyages.
From what I gathered, Walsh et al. & presumably Zakharov et al. then extrapolate the observations they have for a given month to estimate the approximate ice extent for that month. BUT, their data sources vary substantially over time!
If a particular month only had a few observations, then the interpolation could dramatically under/overestimate the ice extent compared to other months.
The satellite measurements are completely different from the others in at least two ways:
1. They are theoretically estimated from microwave readings
2. They are constructed from a large number of orbital passes
The first point is a disadvantage – they don’t actually observe the ice, they infer it. This means the accuracy of the estimates depends on the reliability of the calculations (which Jonathan Drake is questioning). The second point is an advantage – they have an almost complete coverage (bar “the hole” at the poles), meaning they have a very high precision.
However, if you are comparing two datasets, you must AT BEST use the precision of the least precise and the accuracy of the least accurate.
The difference in BOTH precision AND accuracy means that the satellite estimates can’t be simply treated as an extension to Walsh’s terrestrial measurements. That was a mistake on Walsh et al.’s behalf.
I wonder how the precision and accuracy of the terrestrial estimates vary over the record, and in particular, what it was like in the 1920s-1930s warm period.
Found Zakharov:
Click to access wmo_td_782.pdf
At TallBloke’s Talkshop, Roger Andrews (April 18, 2012 at 6:13 pm) has posted a composite of datasets: http://tallbloke.wordpress.com/2012/04/17/jonathan-drake-could-instrumentation-drift-account-for-arctic-sea-ice-decline/#comment-23569 “It compares three observationally-based Arctic sea ice reconstructions – which match each other quite well – against the combined Chapman-Walsh satellite record”
Reports of underestimations from passive microwave are quite common. This one from marginal ice regions (Canadian) advises caution: http://www.tandfonline.com/doi/pdf/10.3137/ao.410405
Cheers, John! Some bedtime reading for tonight… 😉
It’s a long document, but a quick glance at p27-33 answers some of my questions and gives some interesting insight into the relationship between the 1920s-1940s warming, 1950s-1970s cooling and 1980s-2000s warming.
Jeff, if you’re still paying attention to this thread, you might be interested in having a look at the WMO Technical Document #782 (1997) that John posted above. It’s long, and it’s mostly related to terrestrial measurements, but you might find it complementary to your satellite analysis…
This paper is important: http://www.remss.com/papers/tech_reports/hilburn_SSMI_EIA_TB_RSS_Tech_Rpt_092811.pdf
For a good background, this is worth reading: http://www.ncdc.noaa.gov/oa/rsad/ssmi/swath/fnoc-ssmi-manual.pdf
Hi Jonathon,
(I’m posting this reply as a new comment, as the above inline commenting was getting quite deep)
Sorry, been pretty busy the last week. Thanks for pointing out Roger Andrews’ comment on Tallbloke’s post. I guess the same questions I asked above also apply to the Vinje and Miles datasets. Have you found out how they’re constructed yet?
A five second Google search (“vinje ice extent”) yields Vinje, 2001: http://journals.ametsoc.org/doi/pdf/10.1175/1520-0442%282001%29014%3C0255%3AAATOSI%3E2.0.CO%3B2 , but “miles ice extent” requires more than five seconds 😉 Have you found any direct links? I’ll have a more careful look for the Miles dataset and at Vinje’s paper later.
But, in my opinion, there are several essential questions that need to be resolved before ANY realistic assessment of the relative pros/cons of the various datasets can be made:
1. What data sources are used to construct each of the datasets?
2. In what way do these data sources vary over the course of the records?
The above questions both comprise at least two distinct and critical aspects:
a) What types of measurements were provided by the data sources. So far I’ve seen several methods of direct ice extent estimation – i) coastal observation, ii) ship observation, iii) submarine observation, iv) aerial observation and finally v) conversion of satellite measurements.
Anyone know of others? I’m ignoring (for now) the paleoclimate studies which attempt to use various proxies, e.g., Gerard Bond’s “ice-rafted debris”: http://en.wikipedia.org/wiki/Bond_event which I have several problems with, but was an interesting concept. But, I reckon, we should ignore the proxies until we have more confidence in the direct measurements. Do you agree?
EACH of those methods has its own distinct properties and could potentially lead to different results. Your focus in your article was that you suspect method (v) is unreliable. But, I’m dismayed at the manner in which all five methods are routinely lumped together as equivalent.
The Walsh dataset seems the worst in that, as the methods used for any given decade seem to be completely different. The Walsh dataset isn’t just comparing apples to oranges – it brings in grapefruits, melons and grapes too!!! 😉
b) How does the geographical coverage of each of the data sources vary over time?
e.g.,
Vinje seems to be confined to April ice in the Nordic Seas.
Zakharov used Russian measurements.
Walsh includes the Beaufort and Chukchi Seas.
I’ve also seen some discussion of the Koch Index, a VERY interesting estimate which is based on the amount of ice reaching Iceland: http://www.vedur.is/media/vedurstofan/utgafa/greinargerdir/1998/98035.pdf
(see also Lassen & Thejll, 2005: http://www.dmi.dk/dmi/sr05-02.pdf).
So, each of the datasets comprises a different geographical area, meaning direct comparisons require a lot of caution.
But, even within each dataset, the measurements, calculations and geographical area vary substantially over time. That should set off alarm bells with any good scientist.
If you look at the papers I linked on the Koch index, you’ll see that Koch didn’t give enough details in his 1945 study (which went from the settlement of Iceland to 1939), so people have had to guess what he did, in order to extend it up to the modern era.
—
In fractal mathematics, there is a well-known problem called the “coastline paradox”: http://en.wikipedia.org/wiki/How_Long_Is_the_Coast_of_Britain%3F_Statistical_Self-Similarity_and_Fractional_Dimension Essentially, the smaller the ruler you use for measuring the “length” of a coastline, the larger your answer!
Ice extent is definitely not the same as a “coastline length”. But, it does highlight the scientific problem of consistency of measurements. If your “ruler” (measurement/calculation techniques) changes OR the actual area you are measuring changes, you have major problems. But, here we seem to have both! 😦
—
So in summary, we have at least 4 questions that need to be resolved (1a, 1b, 2a and 2b). I genuinely think it COULD be done if the individual components of the data sources could be obtained and separately analysed.
Unfortunately, the current approach seems to be “throw everything into a blender and homogenize it until it matches the computer models” 😉
—
I suspect that the true Arctic sea ice extent variability since the 1880s should be closely related to the temperature variability in the region. Climate scientists (& intuition!) tell us that as the Arctic warms, the ice extent decreases. That’s apparently what the panic over the satellite-based estimates of the ice extent is about. Similarly, whenever the Arctic cools, the ice extent should increase.
But, most Arctic temperature studies report a very warm period in the 1930s in the Arctic, which was at the very least comparable to the recent warm period. Look at Polyakov et al., 2003 for instance: http://journals.ametsoc.org/doi/full/10.1175/1520-0442%282003%29016%3C2067:VATOAT%3E2.0.CO;2
However, NONE of the ice extent datasets discussed so far (possibly with the exception of the qualitative Koch index) show the dramatic 1920s-1930s decrease and subsequent recovery. That makes me sceptical of their reliability of any of them…
E-book with data.
Climate Change in Eurasian Arctic Shelf Seas
Centennial Ice Cover Observations
Ivan E. Frolov, Zalman M. Gudkovich, Valery P. Karklin, Evgeny G. Kovalev and Vasily M. Smolyanitsky
http://www.springerlink.com/content/978-3-540-85875-1#section=649439&page=1&locus=41
(be carefull with the data)
Blog Lurker (April 26, 2012 at 6:42 am)
I agree with your summary. There is a great deal to be checked and reassessed. The accuracy of data from sparse measurements is a huge problem, particularly in the earlier records and will likely remain rather subjective.
It is clear that satellite passive microwave is inadequate for shipping, hence the need for “more on visible imagery” http://nsidc.org/data/masie/about_masie.html. Since the advent of the satellite PM, most, possibly all, major sea ice datasets have included PM and are therefore not independent. Reconstructions using temperature records also need to be treated carefully.
There are some interesting parts to this thread at WUWT: http://wattsupwiththat.com/2012/05/02/cache-of-historical-arctic-sea-ice-maps-discovered/
Partington et al 2003: http://digitalcommons.unl.edu/cgi/viewcontent.cgi?article=1058&context=usdeptcommercepub
“Both chart data and passive microwave data show a negative trend in integrated arctic-wide concentration over the period 1979-1994. The difference between the passive microwave and chart trends is statistically significant only in the summer, where it is about 2 percent per decade steeper in passive microwave data.”
“Differences between the NIC ice chart sea ice record and the passive microwave sea ice record are highly significant despite the fact that the NIC charts are semi-dependent on the passive microwave data, and it is worth noting these differences.”
AQUA platform with AMSR-E was launched 2002 so that was not available to Partington et al.
John Bills & Jonathan Drake,
Sorry, things have been pretty hectic, so I didn’t get a chance to reply until now. But, I just wanted to say that this discussion has been informative, and prompted me to look at the various (Northern Hemisphere) sea ice records in more detail than I had before. Unfortunately, I don’t have the time to look into it more, right now. But, maybe in a few months, I might return to it.
I’m realising there is a considerable overlap between all of the non-satellite datasets, and the problems we were discussing above do not appear to have been adequately addressed.
@John, thanks for the link to Frolov et al., 2009. By the way, do you have the pdf of Chapter 2 (& 3), or are you just reading the free samples like me? I can access Springer Journals, but for Springer Books I’ll need to order it…
There are some unusual discrepencies at various points in the dataset (Appendix A), which are puzzling.
The Karelia links were also interesting, when I finally was able to access the server.
@Jonathan, I too am sceptical of the reliability of the satellite-based composite record. But, I suspect the general decline of Arctic sea ice extent since 1979 is probably correct. The exact magnitude (& continuity) of that decline is less clear.
I totally agree with you on “Reconstructions using temperature records also need to be treated carefully.”. I agree – there are a number of serious problems with the temperature records, which are difficult to exactly quantify. But, from my analysis, I am quite confident that the Arctic warm period in the 1930s was genuine, and is at least comparable to (if not greater than) the recent warm period. The fact that the Arctic sea ice extent estimates fail to show anything close to that should raise a large warning flag on their reliability.
Good luck with your own analysis…