the Air Vent

Because the world needs another opinion

A Little Bit of Magic

Posted by Jeff Id on February 28, 2009

Sorry for the delay, I haven’t actually been lazy at all. I have been working on a couple of what I find very interesting posts this one today and possibly another tomorrow. You aren’t going to believe this…

Those who have followed the Steig reconstruction know that 3 curves were used with different weightings to recreate the entire antarctic. The 3 curves look like this.

three-pcs

Linear combinations of these 3 curves make the temperature trend for the entire antarctic. RegEM was used to create the data prior to 1982 for the satellites by basically copying surface station data onto the ends of the sat data. RegEM is supposed to follow the data that exists so the PC’s should look the same in the pre 1982 period as the post. But they don’t! The paper makes the statement that 3 pc’s are used in RegEM however in the result PC3 is near zero and therefore there are really only 2. In this post I made a video of the Steig temperature anomalies for the antarctic. None of the data in this video is real, ALL of it is highly processed. The data post 1982 is derrived from PCA analysis of real data. The pre 1982 data is pasted on ‘imputed’ data.

Here are some plots of the distribution of he 3 pc’s or eigenvectors as calculated by Steve M at climate audit. These graphs are not temperature but rather weightings (positive and negative multipliers) times the 3 pc’s to create the final temperature anomaly.

eig-1

eig-2

eig3

Here’s the video of raw temp anomalies I created from R. If you consider the pc3 in the very first graph is basically zero until 1982 when it suddenly steps we see what we expect in the video. An oscillation blue and red of the whole continent (PC1) combinded with a top to bottom oscillation which corresponds to PC2 (you can see the boundary). These are the only PC’s of consequence in the pre-1982 RegEM reconstruction portion. After 1982 PC3 kicks in and the data begins fluctuating left and right on the graph as well. If you’re wondering what that means let me put it this way, temperature patterns did NOT change in Antarctica in 1982 coincidentally with the launch of a polar satellite! (that was fun.)

video-cover-1

The video is a little difficult to interpret when you don’t know what you’re looking for so I did a second one where removed the mean of the plot from each month. So this video is the same as above except that it shows an equal amount of positive and negative anomaly for each frame (month). The idea was to bring out more of the features in the data…. IT DID!

video-cover-2

r1-454

Note the pink ring of reversed trend around the mid Antarctic. This is not a minor point at all. It shows quite clearly in the second video the ring around the antarctic is OPPOSITE in temperature from the land immediately adjacent. When the center is blue, the edge is red and they change positions constantly. I don’t know how this happened because we don’t have the REAL data only this PC nonsense, but I sure as heck don’t see an explanation in the paper anywhere. If you take a close look at Steve McIntyre’s plot of eigenvalues it’s in there too. PC1 has yellow creeping all around the rim PC2 has some as well from the bottom edge. Let’s see where the actual data comes from.

surface-stations-location

The real data is all around the edge. Somehow these guys managed to invent some math which reversed the trend of the surface stations within a couple hundred kilometers from the actual stations themselves and apply the reversed trend to the entire area circumscribed by the actual data where nobody knows what really happened.

———————-

Update — here are some of the still frames from the second movie. Keep in mind this is the data as presented on Steig’s own website from the dataset representing the final conclusion of the paper.

r1-545

r1-4491r1-43

r1-389

r1-350

So now we know how they got a warming trend when many of the eastern stations showed cooling. If it’s related to ocean air, I would expect it to follow inner temps at some time but it does not. If we had the real data we would know but it doesn’t look good for Steig 09 to me.

20 Responses to “A Little Bit of Magic”

  1. Layman Lurker said

    I can’t see this holding up. To me it looks like a mess, a complete meat grinder. Trying to deconstruct is like trying to make steak after it was turned into hamburger.

    BTW, I reserve the right to revisit my comments on the nature of the spatial correlation of the surface stations. Maybe this is just my own ignorance but there are things about RegEm which seem contradictory to me if it is just normal distance correlation. On one hand we know from Steig’s hints and from your work that RegEm as used can’t distinguish geography of the individual stations inputed. If it can’t relate the location of the stations into matricies and calculations and stuff, then how can it pick up the distance correlation? It doesn’t know if the distance between two given stations is 2 miles or 2000 miles apart.

    For the peninsula, the stations are all clustered into a small geographic area and the distances don’t matter that much – therefore strong correlations between any two given points. But what is the explanation for the greater continent? I think that there may be strong correlations between most of these stations regardless of distance between points. Think about it…almost all on the coast (ocean as disk boundary), similar distance from geographic center which happens to be near the pole therefore latitudinal correlation emanating from the pole/geographic center.

  2. Gary P said

    I have never understood how all of these coastal stations could be used to determine temperature trends in the interior of the continent. Is the Hadley cell over the Antarctic so strong that the wind is always outward?

    But if the wind is always outward, then the temperature trend off the coast would have to follow the coastal stations. They should not be out of phase.

  3. David Jay said

    The phase difference is clearly unconnected to any physical explanation. It must be an artifact of RegEM.

  4. Jeff C. said

    Good Lord, it gets worse every day. It took watching it a few times to sink in what I was seeing. Up to 1981, since only two PCs are effectively at work, the peninsula/West Antarctica behave in a monolithic fashion opposite of East Antarctica. It is fascinating to watch the temp anomaly alternate back and forth between the two sides. Does anyone really believe that the temperatures of a diverse continent behave this way? It should be easy to disprove based on station data alone. When the third PC kicks in 1982 things diversify somewhat, but not that much.

    “If you’re wondering what that means let me put it this way, temperature patterns did NOT change in Antarctica in 1982 coincidentally with the launch of a polar satellite!”

    I’m not so sure of that. Didn’t Al Gore say that AGW risked altering the energy balance between Earth and the rest of the universe? If little old Earth can do that, who knows what a 1000 pound satellite can do.

  5. Jeff Id said

    Great comments, at this point I’m not sure this paper is any better than M08. I’m feeling a little embarrassed for stating that I believed it was probably right at the beginning. Why wouldn’t the Antarctic warm after all, everything else did.

    Whether this paper is right or…uh…. not. IMHO, we just don’t have the data to make a conclusion. Hell we don’t even have the data to know if the ring around the Antarctic is reasonable.

    What if temps on the coast were dampened by sea temps in the satellite data. In this case they used 50km gridded data to prove a trend. Suppose that the temp stations were in close enough proximity to the coast that the 50km block placed the sat temps in many cases, right in the coastal pixel. Then we correlate, we know the costal pixel is dampened for trend compared to the center by water temps so RegEM then applies an increased trend to the center. We see ring around the Antarctic and the only way to counter it would be with better spatial resolution of the sat data or this 3 pc nonsense.

    I don’t know but the Sat data is pretty critical. Hell they could still be right after all and our economy could recover next month too right.

  6. page48 said

    “Today’s scientists have substituted mathematics for experiments, and they wander off through equation after equation, and eventually build a structure which has no relation to reality.”

    Nikola Tesla

  7. page48 said

    I am sure my question will sound very dumb. I’m just now learning the basics of PCA and can’t contribute a lot to the current discussions. But, I have to ask about this one thing.

    What do the satellite measurements of temperature, taken alone and since 1982, report for the interior of Antarctica.

    Why was this whole study necessary?

    It seems that the satellite measurements of the interior, if they have been taken since the currant warming began, would be important in their own right.

    What do they say?

    Sorry if I sound stupid.

  8. Tim L said

    It looks like what I said before…” some kind of inversion going on ”
    ((x-y)-(y-x))
    ya this works but is extremely simple…lol

    Nice work Jeff!!!!
    are your eyes crossed yet?

  9. Tim L said

    I needed a place for this… this is proof of old corrup data being left up on the web. links below :)
    OLD

    http://nsidc.com/arcticseaicenews/2009/020309.html

    NEW

    http://nsidc.com/arcticseaicenews/

    This is one thing DR. Leif has been fighting at WUWT.

  10. Layman Lurker said

    Since it would seem that the patterns are artifacts of autocorrelation, then the inclusion of more factors might show more comlexity in your videos but they would still be artifacts and thus would be moot. Since the spatial warming patterns generated cannot be trusted, one could argue that this therefore limits us to just one PC to represent the overall trend and eliminate the possibility of “Bias introduced by finite eigenvector choice” as suggested by David Stockwell at CA.

    You would still be left with the issue of RegEM not recognizing that the total peninsula data should be constrained to approx. 5% weighting.

    Jeff, I suggest you post a PC1 trend on your spatial grid weighted data from Jeff C. as a best guess at the actual Antarctic trend.

  11. Phillip Bratby said

    Jeff,

    Just seen your article with Jeff C at WUWT. Stunning. Thanks. It really says something about the standard of peer-review employed by the major journals when Steig et al can get published in Nature with such fundamental flaws. If only we could find out who the peer-reviewers were?

  12. Jeff Id said

    #10 I think you’re right that an area weighted analysis with pc1 would be a good way as long as we don’t infill with RegEM in that process. Anthony’s cleaning up a post Jeff C and I did which has another method. We’ve been waiting for him to finish it I found it very interesting.

    #7 We don’t know what satellites say about the interior because they won’t release the data yet. The post I just described will give you some idea though.

    I would say this is an important paper simply because they need the doom and gloom ice/ flooding polarbear’s drowning, penguin’s can’t migrate to the left because they always walk right, scenario. If the darn ice won’t melt how can we talk about drowning the poor people and tax/steal their money??

    Sorry, I’m getting even more cynical.

    Really this paper is about using the satellite data from surface temperatures of the snow and ice as daily air temps so they can paste on the old surface station records and turn cooling into warming.

    —–

    Here’s what I see now. There is a ring of temperatures around the antarctic which appears moderated presumably due to ocean temperatures. This sounds reasonable on the face of it. But how wide is that band and what stations are affected. I showed in a previous post that there is correlation blending of temps over thousands of kilomters, SteveM’s plots show that the 3 PC’s will blend trends across half a continent.

    This is important I think. Now we have a situation where seaside trends, perhaps only a hundred kilometers from a surface station are blending in quite far as well. Look at the last 5 graphs.

    If we take the seaside trends and use them to “calibrate” the satellite data knowing that the satellite data will be reduced by 3 pc’s at the edges the trends at the center will respond with an amplification. A scientist looking for warming could say, ocean temps by the shore are cooler so we expect the inner continent to have greater variability and it will sound perfectly reasonable.

    IMO, there are some very serious problems with it though.

    1 The temperature gradient at the edge of the continent becomes a factor in how much amplification the center of the continent receives.
    2 The PC’s are guaranteed to blend the temperatures over such short distances from the surface stations to the ocean, increasing the effect of ocean cells on the edge located stations.
    3 The paper mentions specifically the difficulties with taking data in the winter. If the winter data is incomplete or inaccurate, the surface station data will be infilled with something. Leaving the summer months where the sea ice is almost gone, as the determination factor for the temp gradient. What I mean is, there’s not as much gradient in the winter.
    4 It sounds simple but in a high gradient area, an accurate recon relies on the ‘very’ precise location of the surface stations relative to the water.
    5 Finally, the size of the satellite gridcells can cause additional blending of trend.

    Anyway, if that makes sense to anyone but me there is potentially a difficult to rectify problem in the math which we may or may not be able to discover with the sat data.

    This land edge gradient may be the crux of the intent, think about it. The trends on the interior get amplified according to the difference in the sat data from center to edge. What happens to the center if the edge get’s smeared inland?

    I may be off base here and am just putting my thoughts down but it would create a paper with an incorrect result that we may even know is incorrect yet find it difficult to disprove the satellite recon portion.

  13. Harold Ambler said

    Impressive work, Jeff.

    A point and a question:

    First the point: Svensmark’s theory calls for Antarctic cooling during periods of strong solar activity (which many solar physicists consider the last 75 years to have been): strong solar wind, diminished galactic cosmic rays, fewer clouds, higher albedo over Antarctica … cooling. If Svensmark is correct and if we enter a deep prolonged solar minimum, GCRs should increase, clouds should increase, and Antarctica (unlike the rest of the world) should begin to warm somewhat due to the lowered albedo (Svensmark clouds have higher albedo than everything other than pristine ice). That’s his theory, anyway, which the growth of sea ice from 1979-2007 would seem to bear out.

    And now the question: I’ve done my best to follow the Steig controversy on wuwt and CA (and elsewhere) ever since the story broke with the absurd Science cover. And yet one thing I have never seen (that would be very useful for my writing projects) is an average continent-wide temperature at the beginning and end points of data collection. I have seen Steig’s alleged trend, but not the raw data. For instance, is the continent-wide temperature average in 1957 approximately -50 degrees Celsius and now it’s -49.5? (I think I’ve seen figures in that ballpark…) Would it be possible to give me your best sense of these numbers?

  14. Joe Black said

    The videos are great.

    To my eye there seems to be rotation of the patterns. Is it possible to make either 4 seasonal videos or pick 4 months from the year and make a video of the data pattern for just each given month?

    Climate is so much more than an annual temperature or annual temperature trend.

  15. Layman Lurker said

    Here is a map of the current distribution of temp on Antarctica (this pattern may change with time):

    http://www.wunderground.com/global/Region/AN/Temperature.html

    These are actual temps, not anomalies. I don’t know how this distribution relates to the anomalies. In terms of physical forces influencing climate however, the map no doubt shows the effect of oceans, and also latitude. The pole is inland, the oceans are on the coast. Coldest air is closest to the pole and gradually warms as you move out. Moderation of the oceans is greatest on the coast, and decreases as you move inland. A layman’s logic would say that changes in oceanic forces which could produce anomalies would be felt greatest on the coast, and would dampen as you move inland. Vice Versa with latitude – forces which tend to concentrate cold air to converge in the center would dampen as you move away from the center toward the coast. Two symetrical physical forces working in different directions but the patterns are synergysitc.

    Now look at eigenvector 1:

    The spatial pattern of the vector is very similar to the current temp map of Antarctica. Yet this is a map of spatial trend weighting, not temperatures. Am I missing something or is this vector inverted from what one would think is actually a reflection of the physical forces?

    Would this not explain why the trend in PC1 is negative rather than positive? The two synergistic forces which would produce warming at the coast but less warming in the interior would be spuriously “flipped”.

    I could easily be missing something here, so I would welcome comments and insights.

  16. Layman Lurker said

    Now I see the area of -50 to -60 temps in the center of Antarctica is completely gone from the map at my link.

  17. TCO said

    Layman, my impression is the sign of the PC is irrelevant as the follow-on parts of the algorithm will weight it with positive or negative numbers (so that it gets “flipped” to the orientation which makes most sense in correlation analysis).

    REPLY: Fished from the bucket again. I don’t know why. I’ll be around more this weekend and next week so it won’t take so long.

  18. Layman Lurker said

    #15

    Further thoughts: One could make the argument that if two forces I discuss above actually exist, then there should be two symetrical eigenvectors with opposite spatial weightings.

  19. Layman Lurker said

    Hope this works. Here is a link to Antarctic temp snapshots over the last 18 hours or so: http://www.wunderground.com/global/Region/AN/pxTemperature.html

  20. Layman Lurker said

    #17 TCO

    Thanks for the comment TCO. I am not sure if I have connected all the dots or not. My question is not so much whether the sign of PC1 matters as much as what it might tell us.

    There is defintely something not right in the areas of the reconstruction covered by the surface stations. Our first clue is PC3. When the Jeff’s imposed spatial weighting of the RegEM output, the discontinuity disappears. What does that tell us? I think it tells us that the reconstructed areas of the coast and the peninsula are an artifact of the pre-1982 portion of the reconstruction. There is little doubt that this PC is wonky but why? I submit that RegEM “sees” that there is variability between specific geographical areas of Antarctica pre and post 1982. Because RegEM sees this and hones in on these areas, it uncovers the negligeable spatial weighting that these areas recieve – hence the tiny variability/amplitude around zero pre 1982 of PC3. The difference between these pre-1982 areas of essentially zero trend and the post 1982 sat data with proper spatial representation is a discontinuity which RegEM sees as the third ranked source of variability. When the Jeff’s did the grid weighting on the output, it gave these areas weight in proportion to its relative area and “overuled” the eigenvector weighting. Presto, more amplitude pre-1982 for PC3. If these areas were weighted prior to being fed into RegEM, I think it is likely that RegEM would find an entirely different 3rd rank PC as there would be no artifact to detect. This is why I suggested to Jeff in another thread that this should be done (he likely figured that out already). Jeff C. said yesterday at CA that he has done this and will post.

    Are is there any other evidence that something is wrong with the temp trends in areas of the reconstruction covered by the surface stations? Ryan O posted on this at Lucia’s: http://rankexploits.com/musings/2009/steigs-antarctica-part-three-creative-mathemagic/
    He observed the differences between the actual data and the same areas in the reconstruction. He asks two questions and then hints at a third. The first question is “Why”. The second is “Is the difference statistically significant?”. He then goes on to show that it is indeed significant. I don’t know if he has posted his third question yet but here is my guess at what the question is: What does this tell us?

    The case I am presenting here is that the trends in the areas of the reconstruction represented by the surface stations are not correct. By extension, because these areas have almost no spatial weighting, this means that the spatial weightings
    for the PC’s are not correct and must be underweighted on the coast and peninsula. By further extension, because the sum of the weightings must equal 1, this means that weightings for the interior are not correct and must be overweighted (also recognized by Ryan O).

    Because almost all of these stations fall at the perimeter of the continent this presents a curious relationship with vector #1. One could speculate that if the data was corrected for autocorrelation and the spatial weighting issue removed, that a new eigenvector might emerge that would resemble an inverted version of the current eigenvector #1. This would impart more weight at the perimeter and less in the interior. Instinctively, this would be consistent with a physical phenomenon connected to the oceans with it’s effects more prominent on the coast and dampened inland (pure speculation on my part).

    Ryan O’s first picture at Lucia’s (above link) shows the spatially distributed warming of Antarctica with decidedly less warming trend at the perimeter of the continent. The relationship even holds in West Antarctica, where coastal warming is not as strong as inland areas of the west. If the spatial weightings used to derive the PC’s are more or less inverted, then a revised map would show a relative warming at the perimeter and a relative cooling of the interior when compared to current. The revision would be dependant correction for autocorrelation yielding revised spatial weighting and revised PC’s.

    My query after all this is: Is the sign of PC1 an artifact of the above issues? Is it unrelated? I also understand that I have not considered the impact of post 1982 on PC1 which has a different combination of issues (and a lot of unknowns).

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: