the Air Vent

Because the world needs another opinion

Bias In Arctic Satellite Sea Ice Trend

Posted by Jeff Id on February 8, 2012

Updated below.

—-

I wrote recently that I wanted to check local regions of inland lakes to look for bias in the sea ice satellite record.   Satellites are not long lived creatures.  They are built of lightweight materials very carefully constructed to survive in incredibly harsh environments.  The result is design lifetimes far shorter than a decade.   Why is that important?   Because that means that the finest climate trend records, which are satellite based, are comprised of instruments continually re-calibrated to absolute detail but are knitted together wherever any switch of instrument occurs.

All kinds of things can affect measurements, time of day, altitude decay, instrument degradation, signal loss, on and on.   Scientists work very hard to correct for these changes but sometimes they are unable to achieve a perfect result.  Other times problems are missed.

One of the biggest critiques of satellite temperature data is the known offsets caused by orbital decay of the individual instruments and how they are knitted together during transitions to newer satellites.  The details of the corrections are impressive and when things in science are not simple, that often means not certain.  In the UAH and RSS series, that means that the scientists use additional data to re-knit the satellites.  Less detail is available in knitting of multi-satellite sea ice data.

What is more important is that sea ice data changed sensor types at about 1987.  I knew this before but had forgotten in which year this transition occurred.

For this experiment, I chose a section of Canada which included lakes that always will freeze over 100% in the winter and melt 100% in the summer.  You would expect that sea ice area for these lakes would produce a sine wave with clipped peaks as the signal reached 100% and 0%.

I can really improve these graphics but this will get the point across:

Total Sea Ice Area - no mask

Purple mask incorporating Great Bear, Great Slave, and Athabaska River Lake.

By running the sea ice code presented previously here and incorporating this purple highlighted mask, we get the sea ice area plot below:

Click to expand. One hundred day filtered three lake sea ice area.

Several things stand out.  First, there is little, yet a non-zero negative trend in the maximum ice from 3 lakes .   Then we see that the minimum ice level from these 3 lakes never drops below 10000 sq km average.   The daily unfiltered data approaches closer to zero but there is enough noise that zero isn’t ever reached in lakes which have zero ice.  The very last cycle is “preliminary data” which has not been fully processed so it’s similar depth to the first cycles may be representative of the processing rather than the data itself.  It makes one wonder what exactly was done to the different types of data.

Steve Fitzpatrick happened to be writing about the crazy thread on backradiation so I sent him this last plot with this comment:

Sea ice area from Great Bear, Great Slave, and Athabaska River.

Wouldn’t you expect that satellite data would show an even melt in the summer months?

And later:

Sea ice at these 3 lakes should have the same max/min.   There is a satellite switch right at 87-88.

I don’t know if there is an instrument type change at the same point.
More reading.

I had forgotten when the instrument type switch occurred but a bit of research turned up this:

This data set contains gridded brightness temperatures (TB) and sea ice concentrations, spanning October 1978 through August 1987, when the SMMR scanner was turned off.

In the end I realized that this is the point where the data turned from SMMR to SMM/I type sensors.  I’m not familiar with the details of the sensors but it seems this offset likely biases the critical Northern Hemisphere sea ice trend to some extent.

——-

Unfiltered area data per Carrick’s request:

————-

As an update to the above post, I’ve created a latitude ring around the NH which is comprised of sea ice known to melt in the summer.  I have verified the flat bottom of the melt cycle which lasts about 100 days for this entire range of data.

This is the ice area plot for the band.

The anomaly has that same step in the middle we found for the sea ice south of the Arctic circle.

Magnifying the lower edge of the sea ice graph reveals similar patterns to what we have discovered.  Remember, all of the sea ice in this band melts every year.  Again, this isn’t a huge difference by any means, but it isn’t zero either.

30 Responses to “Bias In Arctic Satellite Sea Ice Trend”

  1. Jeff Condon said

    Hmm, hundreds of views and nobody is impressed. I like this one.

    We can expect that a proper correction for the effect would result in more positive trends in both hemispheres. I don’t know by how much but it won’t be zero.

  2. Carrick said

    It’s an interesting step, but I have a feeling there are a number of other ones you have to take to complete this journey.

    What does the unfiltered data look like?

  3. Jeff Condon said

    “What does the unfiltered data look like?”

    I don’t have that on this machine but I can post it tonight. The valleys are really noisy with averages similar to the filtered version.

  4. glacierman said

    Well, I think you are on to something. This approach has the potential to document a trend introduced by the various calibrations/sensor replacements. What are the odds the trend is in one direction only?

  5. Layman Lurker said

    Great post Jeff. I don’t have time right now, but if you go to KNMI and download bands of sea ice time series data by each latitude and look at the anomalies, you will see the 1987 shift and the circa 1996 shift which you pointed out in the earlier post. These biases look different depending on the latitude band examined.

  6. Robert Austin said

    Very interesting! It would be interesting to mask other areas with total coverage in winter and ice free in summer such as Hudsons Bay to see if there is similar “weirdness” to the ice coverage as you have shown here.

  7. Nic L said

    Interesting work, Jeff. Satellite changes and corrections are a real problem, and you have found a neat test here. I wish governments would redirect some of the funds that go into supercomputers for GCMs to getting more satellites in operation so that there is always an overlap when one fails.

  8. stan said

    Jeff,

    OT (but sort of related) — could you point me toward a reliable explanation of the differences between GISS and DMI on Arctic temps? This is one of those debate points that seems to be potentially really interesting, but one side claims it proves Hansen is a fraud and the other side says it’s nothing. Is there anyone out there who has a reputation as an honest broker who has discussed the differences?

  9. Matthew W said

    Is all of your hard work going to be moot?:

    http://www.usnews.com/news/articles/2012/02/08/earths-polar-ice-melting-less-than-thought

    “While vast quantities of ice melting into the ocean is not exactly good news, Wahr says, according to his team’s estimates, about 30 percent less ice is melting than previously thought”

    Interesting !!

  10. Brian H said

    Hudson’s Bay would seem to be an even better benchmark, not least because it is salt water.

  11. […] the Air Vent: Bias In Arctic Satellite Sea […]

  12. TimTheToolMan said

    Nice work Jeff. I assume this is purely extent and there is no infuence of timing of the freezing/melting involved? Its too much of a coincidence there is a big jump at change of instrument though.

    What would be very interesting is for you to do a few more masks so you can be sure of the effect and then for you to make your own correction on the entire dataset🙂

  13. Jeff Condon said

    Thanks everyone. I’ve updated the post to include the unfiltered plot Carrick requested.

  14. Jeff Condon said

    TTTM,
    More are coming.

  15. curious said

    Nice – how does your trend line look if you do it as two split at the 87 mark?..Also how does the instrument respond to snow cover on land – is it relevant? 12, 14 sound good.

  16. John M said

    #9

    OMG, it’s not worse than we thought!

    Of course, you know what this means…

    Since less melt water is going into the ocean than previously thought, even more missing heat has to be accounted for to explain the rise in sea level.

    http://sciences.blogs.liberation.fr/home/files/Cazenave_et_al_GPC_2008.pdf

    OMG, it is worse that we thought!

  17. Anonymous said

    Jeff,
    From what I read at Wikipedia, the microwave radiation model for ice is pretty complicated, including the influence of salts dissolved in the water and a bunch of other factors. Maybe the differences between the lakes and the open ocean lead to the kind of nutty results for the lakes (they never show zero in summer!). I wonder also about the influence of the surrounding land… I mean, perhaps emissions from the adjacent land influences the calculated quantity of ice.

  18. Steve Fitzpatrick said

    Sorry, I wasn’t trying to hide my identity above.

  19. Jeff Condon said

    Steve,

    It is very interesting because there ain’t no personalities involved and the result on trend seems poorly explored.

  20. Carrick said

    Jeff thanks for the unfiltered data. Coupla’ things. First the 100-day filter seriously distorts the “approach to maximum” of the data.

    Secondly, it looks to me the summer time levels in both cases are just instrumentation noise floors. I suspect the drop after 1987 was simply an improvement in equipment and hence lower noise floor (but still noise).

    However, if the noise is “speckle-like”, you might be able to remove it by applying a median filter to the data before summing.

  21. Jeff Condon said

    “Secondly, it looks to me the summer time levels in both cases are just instrumentation noise floors. I suspect the drop after 1987 was simply an improvement in equipment and hence lower noise floor (but still noise).”

    Absolutely. The noise floor is offset from zero – a bit. I’ve looked at some Cryosphere area plots tonight (between that other thread) and they seem to remove it effectively though.

    I used a square filter window so I don’t believe that will change much. Also, I’ve plotted the area for every year and they all have flat bottoms in this latitude band.

  22. Jeff Condon said

    Sorry, Just read up on ‘median’ filters and you may be right that a median would give a slightly different result. I doubt it would be much but don’t know.

    Do you want some data to check out?

  23. Carrick said

    Jeff:

    Do you want some data to check out?

    Sure. Certainly beats Doug Cotton explaining to us how Claes Johnson has elegantly proven the Earth is shaped like a banana and other similar nonsense.

  24. Coldish said

    Jeff, Just read this post. I’m no expert, but I’m impressed. I like the method of calibrating from areas that oscillate between 100% and zero every year. The non-zero summer readings and the step down at c 1987 have automatically created a spurious downward trend for fresh-water areas that are 100% ice-free in the summer. Looks like something that should be followed up. Thanks for showing us.

  25. Frank said

    Jeff: There are likely to be areas of “permanent” summer snow and ice in the area of your purple mask. Do you know if the satellite data includes “sea ice” that could be on land as well as lakes?

    Reading the technical information suggests plenty of possible biases. Different satellites have been observing at different times of the day. The period of overlap was less than one month, so between satellite correction factors are good for a particular time of the year. The corrections are different for the Antarctic and Arctic.

    You could try consider finding the best fit with linear segments for each satellite and a variable step function adjustment when the satellites change. This will obviously fit better than a single linear trend for the whole period. A reasonable control would be to repeat the analysis assuming the satellites changed in different years than they actually changed. If the fit improves and the size of the step function change is greater when the correct changeover date is used, that would suggest that the bias introduced at changeover is real.

    “The weather filter used for the SMMR (Gloersen and Cavalieri 1986) was found to be inadequate for the SSM/I due to the SSM/I’s use of the 19.3 GHz channel (which is further up on the shoulder of the water vapor line at 22.2 GHz) rather than the 18.0 GHz channel. A different weather filter is used to reduce spurious sea ice concentrations from SSM/I that result from the presence of atmospheric water vapor, non-precipitating cloud liquid water, rain, and sea surface roughening by surface winds. This filter is a combination of the SSM/I 37.0 and 19.3 GHz channels, which effectively eliminates most of the spurious sea ice concentration measurements resulting from wind-roughening of the ocean surface, cloud liquid water, and rainfall. Another filter that is based on the 19.3 and 22.2 GHz channels is also used. The rationale behind combining the 19.3 and 22.2 GHz channels is based on the sensitivity of the 22.2 GHz to water vapor and on the need to minimize the effect of ice temperature variations at the ice edge.”

  26. Jeff Condon said

    Frank,

    “Do you know if the satellite data includes “sea ice” that could be on land as well as lakes? ”

    There is an unchanging ground mask – I have checked. However unmasked edges may incorporate some land.

    “Reading the technical information suggests plenty of possible biases.”

    For sure. And the fact that I’ve found a bias in the gridded data, doesn’t mean that the pros haven’t already fixed it. As Carrick aptly wrote above, there is a lot more to the story.

  27. unakite said

    John M said
    February 8, 2012 at 9:57 pm

    OMG, it is worse that we thought!

    Heh, maybe, finally, we can say, “it is worse than THEY thought!”

  28. Bruce said

    Is Steven Goddard correct? Is there a censored IPCC graph showing low sea ice in the 1970s?

    http://www.real-science.com/ipcc-early-1970s-arctic-sea-ice-persistently

    As I was looking for it I noticed Greenland temperatures were 2C higher than 1980 in the 30s … http://www.ipcc.ch/ipccreports/far/wg_I/ipcc_far_wg_I_chapter_09.pdf pg 271

  29. Dan Hughes said

    While the subject is, somewhat, latitudinal bands.

    The following are extremely rough thoughts regarding the Global Average Surface Temperature: WAGs to the max. I’ve been reluctant to throw them out for consideration due to (1) my lack of expertise in this arena, and (2) lack of experience and expertise relative to analysis of the temperature data.

    Temperature is the end result of matter under-going thermodynamic processes. Averaging the temperature should then account for this fact and include in the average only those that have experienced the same, or, in some un-specified way, nearly the same, processes. Developing a rule of thumb for estimating the average temperature for the working fluid experiencing a change in state in a process is a valid approach for only a limited set of conditions.

    For the Earth’s atmosphere, the processes that the air is subjected to vary mostly with time-of-day and the latitude and altitude of the location. The transport of energy from equatorial latitudes to polar latitudes is a dominant aspect of the Earth’s energy engine. We directly experience daily, seasonal, and yearly cycles. The cycles are continuous and we could identify any point in them as the starting condition. A natural initial condition might be the initial heating supplied as the Sun rises in the morning following the relatively quiescent conditions during the lack of energy input at night. These quiescent conditions allow development of significant temperature gradients within the hydrodynamically stable atmosphere and the consequent differences in initial temperature with altitude.

    Does anyone know if the temperature data have been averaged within latitude bands and then compared with the “usual” averages? Variations might include accounting for the altitude of the stations in the latitude bands. And maybe coastal stations are also special cases due to the strong influence of the nearby massive bodies of liquid water.

    There is probably not enough data to check the temporal correlation between two or more stations on a time-of-day basis. Although with minute-by-minute data coming online, this might could be checked for a couple of stations.

    I appreciate any thoughts about the concept, including its degree of WAG-ness and lack of usefulness.

  30. Jeff Condon said

    Dan,

    I thought I knew of something along the lines of what you suggested but can’t recall even if it was a blog or paper.

    Sorry.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

 
%d bloggers like this: