the Air Vent

Because the world needs another opinion

Antarctic Sat. Reconstruction Without Peninsula – Still no Sat. Data

Posted by Jeff Id on February 23, 2009

Just a short post this morning to show one of the different calculations I’ve done. I shouldn’t keep everything to myself after all. The graph below represents a regridding of the Antarctic as presented by Jeff C who was kind enough to send me his data.

shaded-cells


I used RegEM to reconstruct the trends without grids A,B,C or D which represent the peninsula of the surface station data along with the 3 pc’s of the original satellite data. This isn’t a perfect reconstruction of the no peninsula data because the 3 pc curves were allegedly derived from the missing satellite data for the complete antarctic. These pc’s would of course have information from the peninsula incorporated in them but since the peninsula is basically an ocean air temperature, this should take us a step closer to the actual arctic surface trend.

three-pcs-original-vs-recon-no-peninsula

PC3 on the left was the original PC from Steig09 with a large number of peninsula surface stations, the one on the right is the no peninsula version. While the right side certainly looks more reasonable, the pre1982 variance went appears larger than the instrument data.

The total trend changed from the original paper of 0.012C/yr to the regridded version of 0.007 to the regridded no peninsula of 0.005 C/yr but is still positive. I couldn’t get the Antarctic area plot to cut the peninsula cleanly this morning before work but the trend across the Antarctic is largely positive and what I would call ‘relatively uniform’ as shown in the original reconstructions. However, this is the reconstructed trend.

regridded-sat-trend-no-abcd

Again, because we don’t have the original satellite data or code we cannot complete a true analysis of the original result or the no peninsula data. Simply clipping the gridded satellite data (which is a combination of 3 pcs) will have no effect on the above reconstruction so without the satellite data (and code) we cannot determine the effect of the peninsula on the Steig09 reconstructions. This simply represents a step closer to the no peninsula result.

The AWS reconstruction is IMO nearly meaningless due to lack of data. The 3rd pc of the original paper is completely unnatural, so we are left right where we started from — not knowing if the claimed Antarctic warming is real or not.

The witholding of the single piece of data and code which could resolve this issue is not the best PR move on their part.

14 Responses to “Antarctic Sat. Reconstruction Without Peninsula – Still no Sat. Data”

  1. Layman Lurker said

    “…so we are left right where we started from — not knowing if the claimed Antarctic warming is real or not.”

    Jeff, correct me if I’m wrong, but you have have come very close to re-creating the 3 actual PC’s used in the sat reconstruction. You have shown that the original sat reconstruction depends on the “unatural” step in PC3. The “actual” warming may or may not be real due to lack of data, but can the “claimed” warming be legitimate?

  2. Jeff Id said

    When I see PC3 like that, I don’t see how the reconstruction can be trusted. I understand what they tried to do now and it doesn’t look like a good technique to me because they do very little to control the spatial weighting of the trend. A trend from one side of the continent can be applied to the opposite.

    One other thing to consider is that the paper found a uncertainty ratio of +/- 0.007C/yr based on variations in the data, by removing this tiny piece of land in the far edge of the continent and coming slightly closer to a reasonable area weighting we now have 0.005 +/- 0.007 so a slope of 0 falls within their own confidence interval. I also assume that if we had the data, by removing peninsula data from the sat pc’s we would get better and different spatial weighting which would result in a further reduction of trend.

    BTW: I still don’t agree with how the confidence interval was obtained.

  3. Layman Lurker said

    And it is likely that removal of peninsula data would likely increase the CI as well. No? So the conclusion of .14C warming trend depends on inclusion of the heavily weighted peninsula data, to obtain a combination of trend and CI produced from a reconstruction using a flawed set of PC’s. Doesn’t seem like a strong case to me.

    The only way I can see that this can be legit is if the use of the wierd PC3 does not significantly affect the reconstruction. However you have shown that it does. By weighting the peninsula data and re-doing the PC’s, you not only showed a reduced warming trend, but also that the Steig trend DEPENDS on the un-natural, pre-1982 section of PC3.

    You may not have access to sat data and code, but you know what it produced and I think you also know what it cannot produce – and that is a legit .14C warming without a flawed PC3.

    That is my take. Am I missing something?

  4. Jeff Id said

    In the paper I was surprisd to see 3 pc’s for the whole antarctic used in RegEM. Steve M didn’t mind the use of so few but IMO the fact that spatial weighting is carried out through correlation to these curves ONLY, there has to be more. It was telling that the paper describes physical processes for the first two PC’s but nothing is said about physical processes for the third.

    I believe your post is correct regarding PC3, I wonder what would happen if I simply deleted the 3rd and kept the same weighting for the others.

  5. Layman Lurker said

    “…nothing is said about physical processes for the third.”

    But it appears to be a proper inference that this PC reflects the discontinuity between improperly weighted peninsula temp trends and that of the greater continent prior to 1982. A hypothesis supported by your (and Jeff C.’s) gridded weighting.

  6. Steve M didn’t mind the use of so few

    I can picture an argument for using only the PC1.

    This debate over how many PCs to retain is a familiar issue as there was lots of ink spilled over whether to retain 2 or 5 PCs in MBH98. In order to salvage the bristlecones, Mann argued that a PC4 was “right” but the rule getting a PC4 in the NOAMER network didn’t yield the retained PCs in other networks. Of course, it’s climate science so no one other than me cared.

  7. Jeff Id said

    “so no one other than me cared.”

    I suspect no one other than you figured it out ;)

  8. Layman Lurker said

    OT… Steve, what is happening with your website? I hope you suffered no losses.

  9. Jeff C. said

    Jeff – this is somewhat OT, but I thought I’d put it out here to see if anyone had any comments or might be able to tell me where I might be making a mistake.

    I’ve been sifting through the data from numerous angles trying to understand the relationship between the surface station measured data and the satellite and AWS reconstructions. One of the things I calculated was the correlation of surface station data from various regions versus the reconstruction average. By recon average I mean the monthly mean of all 5509 points for the satellite recon or all 63 sites for the AWS recon.

    I used the surface station data from CA (data.tab) and then calculated the anomalies for each station. This is actual measured data, not the RegEM infilled data. I then grouped the stations into regional categories and calculated an average for each region. Here are the regions:

    Island (5 stations, those off the continent)
    Peninsula (15 stations exclusively on the peninsula)
    West Antarctica (5 stations including those bordering the Ross Ice Shelf)
    East Antarctica Interior (2 stations, Vostok and Amundsen-Scott)
    East Antarctic Coast (15 stations, coast of East Antarctica)

    I then calculation the correlation coeficients for each region vs. the satellite reconstruction average.

    Islands -0.194
    Peninsula -0.037
    West Antarctica +0.522
    East Antarctica Interior +0.655
    East Antarctic Coast +0.768

    I then did the same for the AWS recon.

    Islands -0.246
    Peninsula -0.029
    West Antarctica +0.850
    East Antarctica Interior +0.512
    East Antarctic Coast +0.559

    I haven’t fully digested this, but it seems to me that since both recons show a negative correlation to the islands and no correlation to the peninsula, RegEM is avoiding wide-scale spurious correlations to those regions over the continent. My averaging might be masking something important so I’m going to look at individual stations and gridcells.

  10. Jeff Id said

    Jeff

    This is good stuff again.

    Since all the sat data post-1982 is real, there should be good correlation in this section, even with 3 pc’s.

    I have two questions,

    – Does the pre-1982 show the same distribution as the post 1982, (this would make a really interesting post, it doesn’t matter which way it comes out).

    – Most of the variance in these correlations is likely due to high frequency effects. What do the low frequency i.e. trend correlations show? RegEM with this little data may be very insensitive to longer term signals in this much noise.

  11. FrancisT said

    Since this is the latest Steig post I’m putting it here, but it applies as a general comment. I’ve seen a bunch of people asking – so what does this mean. Here’s my response (please please correct me in any errors!)

    A layman’s guide to Steig and the Jeff, Jeff, Roman, Steve et al, issues with it

    Steig et al. was a paper published in Nature which purported to demonstrate, via a number of different analyses, that the Antarctic is in fact warming as fast (or possibly faster) as the rest of the world. This is contrary to previous research which indicated that the Antarctic was cooling.

    The issues boil down to a couple of points that are frequent bones of contention between the “team” of Mann, Hansen & co and the “auditors” – Steve McIntyre, Jeff, Jeff, Roman etc.

    The first issue is ease of replication. Steig has provided a certain amount of source code and raw data but it is patchy meaning that some of the analyses are easier to replicate than others. There was also a brief flurry of data accuracy issues in the initial data provided. Some of these were Steig misattributing data from other sources such as the British Antarctic Survey (BAS) others seme to have been errors in the BAS data itself which were identified when the “auditors” started to dig in to what Steig had done. The data issues appear so far to be comparatively minor in overall effect though the fact that the “auditors” found them in a few days but the original researchers and any fact checkers they may have employed did not is a trifle worrying.

    The second issue is the statistical methods used to reconstruct the historical temperatures in places where data is absent. The current posts here and at CA are all attempts to reconstruct Steig’s analyses and then see if there are flaws in how they are created. So far there seem to be one basic problem, which is that a very large amount of data is being reconstructed, particularly pre 1982. This means the final trends reported are based heavily on the reconstruction methods and assumptions. By changing the weighting of various parts of the data it seems possible to get very different results.

    The raw data used seems to rely heavily on stations in the “Antactic Peninsular” which is the most heavily populated part of Antactica and hence has more surface stations than other parts. At least one of the different weighting methods (grouping surface stations in the same area) seems to significantly reduce the trend because it groups the Antactic Peninsular stations together and thereby reduces their overall weight in the final reconstructions.

    Secondly they seem to be using a very limited number of Principle Components (3). This could be OK but it looks suspicious especially when using (almost?) any different number seems to reduce the size of the trends in the final result. There are also some questions about the third principle component which seems to have a significant step in 1982 – presumably to do with the integration of satellite data.

  12. Layman Lurker said

    #9 & #10

    What would these correlations look like with your grid weighted reconstruction?

  13. Layman Lurker said

    re: Jeff C. #9

    By looking at your post and the “Chladni” post at CA, it looks like the peninsula warming has been “exported” or “traded” to the greater continent.

    Perhaps a hockey analogy is in order. In 1967 The Chicago Blackhawks traded Phil Esposito, Ken Hodge, and Fred Stanfield to Boston, in exchange for Pit Martin, Jack Norris, and Gilles Marotte. After that the Bruins became anomalously (is that a word?) “hot” and Chicago very “cool”.

  14. I’m impressed, I must say. Seldom do I come across a blog that’s equally educative and entertaining, and let me
    tell you, you have hit the nail on the head. The issue is something which not enough folks are speaking intelligently about.
    Now i’m very happy that I found this in my hunt for something relating to this.

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

 
Follow

Get every new post delivered to your Inbox.

Join 140 other followers

%d bloggers like this: