This is something that needs wider circulation, hence its appearance here. I’ve been mulling over how to best present this, and decided there’s nothing I could do in the way of excerpts that still told the story effectively, so I decided to present it in full. I’m pretty sure Jeff won’t mind. Readers may recall Steig et al 2009 and its cover picture on Nature saying the whole of Antarctica is warming, and the skeptic response paper O’Donnel et al 2010 response demonstrating conclusively that the peninsula is warming, but the whole of the continent is not. The reason for this is the same flawed PCA flawed statistical methodology, similar to what was used by Mann to create the Hockey Stick. Jeff Condon, co-author of O’Donnell et al 2010 tells the story at his Air Vent blog. Andrew Montford also has a post on it where he opines about the Steig et al 2009 paper possibly being included in IPCC AR5. – Anthony
Posted by Jeff Condon
I know you guys missed me, Real Climate sure did. Eric Steig has written a letter to “The Guardian” (booming voice) in response to a Nic Lewis letter patiently explaining problems in an article written by yet-another-know-nothing with a keyboard. Unfortunately for us, the article itself has been updated in response so we can’t read the original. What is interesting about the exchange is Dr. Steig’s wild reply.
My bold.
Nicholas Lewis (Letters, 28 August) complained that your report (Arctic ice melt likely to break record, 24 August) gave the impression that typical temperatures in Antarctica have risen as much as on the Antarctic peninsula.
While he is correct about this, his letter also refers to an outdated study of his, which argued that previous estimates of overall Antarctic warming were too high. In fact, the work of Lewis and co-authors has been proven wrong.
The relevant paper here is Orsi et al, Geophysical Research Letters, vol. 39, 2012, which shows that the rate of warming in west Antarctica is as great, or greater, than what we showed in our original work (Steig et al, Nature, vol. 457, 2009). Moreover, Lewis’s own paper shows there has been
significant warming in west Antarctica and that the average trend over Antarctica is of warming, not cooling as is often stated.
The reality is that the Antarctic is warming up and is contributing significantly to sea level rise; and that there is strong potential for a greater contribution to sea level rise from Antarctica in the future.
Professor Eric Steig
University of Washington, Seattle, USA
Our 2010 study is now outdated???
Seriously!! This absolutely is the doctor who never learned about matlab.
For those who have not read the history of the Antarctic wars, here is a pictorial summary.
Steig said this on the cover of Nature magazine:
We said, no you screwed up the math so using the same data it is more like this: (Ryan O’Donnel)
Because without the satellites the temperature stations alone say this:
So it absolutely cannot be the image on the right:
Steig 09 smeared the peninsula warming across the continent (see how it is missing from the peninsula on the left), but now he says O10 has been “proven” wrong. This tells me that he has apparently never understood that the result we produced is nothing more than thermometer data. That is all it is.
Temperatures as reported by thermometers. It is a skeptic plot I tell you!!
Perilously, Steig 09 was precociously printed on the previously prestigious primary page of Nature publication. Carelessly comprised of contaminated and crappy satellite data with thermometers taking a tertiary role in tolling temperature. Sorrily, Steig’s seminal segment was further stuffed by sloppy math. (alitteral too far?)
So Jeff , what did he base his conclusion that the PCA distributed thermometers of O10 are now “outdated” and “proven wrong” on?
A single borehole temperature reconstruction at a single point……
One spot
—-> o <—-
Genius!!
It was workmanship like that which got me labeled as a skeptic in the first place.
Notes to Real Climate and Orsi:
The PCA method S09 attempted to employ, is about redistribution of thermometer information according to covariance of AVHRR satellite data. By nature, every temperature station affects every point in the reconstruction. Kriging the temp stations, is a far more controlled and far more verifiable solution for the same thing and it would produce the same result as O10. The 3 pc’s of the Steig 09 method “smeared” the thermometer data everywhere, so no matter what is published, S09 methods will NEVER be verified. S09 can never, and will never, be correct….because it isn’t! The fact that it is to be cited in AR5 is yet another wart on the last few grains of credibility the IPCC holds.
Sorry for that.
Does the error of S09 that mean that O10 is right? No, of course not. But O10 is very close to actual thermometer results. This is because in a “skeptic” plot, we cleverly used actual thermometers. Bunch of morons I say. This is in direct contrast to S09 which preferred 3 pc’s of highly noisy Satellite AVHRR data WHEREVER available. That was not a smart plan ….. Um, if you want good results.
Even with enough pc’s as O10 used, there are points in this sort of reconstruction with noisy data, where modes of the PC are a dominant factor in creating the local trends of the plot. These methods mean that station information can be and IS copied across an entire continent. This was proven by the S09 cover of nature(Figure 1 above). The trick is to minimize the information bleeding. What this means to me is that I am quite comfortable that the O10 reconstruction will never be proven wrong, not because something as massive and complex as o10 doesn’t contain a boo-boo, but rather because it is an approximation of a field. The best anyone will ever do — is improve on it.
Apparently, this is something that Steig has never figured out. He might not ever work it out, but science is a cold sport and my guess is that those who are smarter than him ….. will.
===
Other notes of surprise:
What normal thinking person would take a temperature from a lousy borehole and hold that out as superior to an actual thermometer?
???????
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Rob Dekker says:
September 10, 2012 at 2:43 am
Jeff Condon said :
Les, I haven’t tried it with the actual data but you may be right.
Maybe you should use Orsi et al data, Jeff. That borehole is right in the area where O’Donnell et al had only a single station to work with within a 1000 km radius : Byrd.
Rob,
I’m sorry that I haven’t been back to look at this thread for a bit. I’m afraid that I suffer from the view that a borehole holds zero value with respect to temperature. Due to its numerous errors, I hold the same opinion of Steig’s result.
As to your claim that Byrd is wrong so all of the west Antarctic is, you are missing a couple of points. First that Byrd station itself, from the BAS site, has a lot of missing data. Second, that PCA creates false resonances in the spatial temperature information. These show up as Chadni patterns in the result which tend to create localized mismatches in the data. Even though all PC’s were used in the reconstruction, it is my contention that noise in the data causes certain modes to remain prevalent in localized regions. Note the darker blue near Byrd. It is quite possible to get a region like Byrd to be slightly different than actual yet have it fully compensated by an adjacent region.
An example of the regional mismatch in the west Antarctic can bee seen by swinging your eyes to the temperature by closest station plot provided in this post as the second figure from the bottom. It looks different in the Byrd region – slightly. There isn’t as much localized cooling in closest station method at the bottom edge of the Antarctic. What I can say is that the average of the West by O10 is almost exactly the same as the method-independent closest-station reconstruction even though there are minor regional differences.
What is an artifact of the method vs actual weather pattern is impossible to determine. Therefore, even our corrected Steig method is not my preferred result but if you believe there is value in the improved spatial form of the satellite data because it incorporates natural weather patterns over boring closest station methods, the corrected Steig method might be your preferred result.
My short version answer is that you are incorrect that if we get a different Byrd answer the West Antarctic is wrong. As evidence, the PCA West result is the same as the averaged temperature stations for the region.
I’m sorry that I don’t have time to look at the data for you more on this. The data and full results are online at Climate Audit so you can find exactly the trend you want to locate.
If time permits, I do hope to apply Roman’s anomaly methods and Kriging to the problem. That would create a far easier to understand temperature field. The noisy AVHRR satellite data could be used for determining appropriate distance weighting in the matrix.
Kriging sounds like a boring publication but all of this noise from “climate scientists” about a single point in a reconstruction means that people are generally confused about PCA methods. While various principle component methods are the sexy thing to use these days in the field (on whatever random data can be found), these kinds of discussions by the pro’s tell me that they may not afford sufficient added value.
Rob Dekker wrote:
“How did you conclude that ? As far as I can see, the infilled trend at Byrd in O’Donnell et al is +0.08 C/decade. If I add 0.13 to that, I get +0.21 C/decade, which is still below Steig’s trend at Byrd of +0.23 C/decade, and certainly below the +0.32 C/decade trend at Byrd from Monaghan et al.”
Ah, you are confusing Steig’s reconstruction trend for the grid cell containing Byrd with Steig’s actual trend for the infilled Byrd station record – which he kept quiet about!
Steig’s RegEM code infills the missing temperature data for the 42 manned weather stations that he uses for his reconstruction, plus for the 3 AVHRR satellite principal components (PCs). He then produces a spatial (grid cell based) reconstruction from just those 3 PCs, a very unsatisfactory procedure that has the effect of smearing the warming from the many peninsula stations over much of the continent. His spatial reconstruction is accordingly pretty much worthless, irrespective of the merits of his infilling procedure. That was a key point made in the OLMC paper: Steig’s reconstruction is invalid, since its method is mathematically unsatisfactory.
The infilling method that Steig used was less sophisticated, and less stable, than that used by OLMC, and used fewer stations but included the 3 AVHRR PCs. Reflecting these differences, Steig’s 1957-2006 trend for the infilled Byrd station record was considerably higher than OLMC’s trend.
However, Steig’s 1957-2006 trend for the infilled Byrd station record was only 0.135 C/decade, NOT 0.23 C/decade.
I invite you to check this for yourself by running the relevant computer code on the data that Steig used. Or you can ask Eric Steig to check it, and thereby find out if he is as willing to provide information as the OLMC team have been.
The large discrepancy between the trend per the infilled Byrd station record and the trend per the spatial reconstruction at the same location illustrates how poor Steig’s reconstruction was. The two trends are close to each other in OLMC’s case.
Since Steig’s spatial reconstruction used only 3 PCs and thus had only 3 time series to define the temperature records for 5509 grid cells, and for various other reasons that you will be familiar with if you have read the OLMC paper, the Steig reconstruction trend in any particular grid cell is not very meaningful. That being so, in Steig’s case the trend for Byrd station itself, post infilling, is a much better measure of the trend at Byrd than the spatial reconstruction trend there.
Jeff Condon said
We don’t disagree on that, Jeff, but if the issue at Byrd was only that it had a lot of missing data, then O’Donnell et al would have found a trend of +0.25 C/decade at Byrd (if I understand Ryan correctly). That would have resulted in a West Antarctica trend of +0.16 C/decade, getting pretty close to Steig’s reconstructed trend in West Antarctica, and a continental trend of +0.07 +/- 0.08 C, barely insignificant.
But that did not happen, since O’Donnell modeled Byrd as 2 stations. That is reasonable, because there was a re-location at Byrd (from the old manned station to Byrd AGW, an automated station). But the problem then becomes on how you will calculate the “offset” between the two Byrd stations. O’Donnell et al used the nearest stations that cover the data void time period between the old and the new Byrd station, but these are more than a 1000 km away, across the Ross sea and beyond. Wouldn’t that cause data “smearing” at Byrd, and over West Antarctica, no matter which statistical method you use ?
It’s simply very difficult to obtain data where there is none, right ?
Now, when a lousy borehole comes along in exactly the right place, confirming the original trend (at +0.23 C/decade) would you not want to use that data to re-calibrate your offset at the two Byrd stations, and see what will come out for the O’Donnell et al conclusions ?
Thanks for bringing up the “darker blue near Byrd” in the O’Donnell reconstruction. It seems to be not just near Byrd, but smearing out all across the South part of West Antarctica, extending to the Ross Sea shore. I was wondering where these came from, but your explanation of PCA creates false resonances in the spatial temperature information. These show up as Chadni patterns in the result which tend to create localized mismatches in the data is way over my head.
Could you please explain in plain English which stations around the area are responsible for the “darker blue” ? Or if it was not station data, which data caused that part of West Antarctica to show up as cooling in the O’Donnell reconstruction ?