Today’s post in my series about Michael Mann is going to be a bit different in that it “doesn’t matter.” I’m not going to discuss anything central to Mann’s results or arguments. Instead, I’m just going to discuss Mann being incompetent in his work.
You should remember Mann’s reconstruction used 415 different data series. Approximately 300 were combined via a method he called Principal Component Analysis (PCA), reducing them to 31 proxies. Another 81 series were used directly as proxies. That gave a total of 112 proxies.
As you should also remember, only 22 of those proxies extended back to 1400 AD. We’ve already discussed how two of those 22 proxies, the most important ones, were inappropriate for Mann’s purposes. I won’t revisit them. Instead, I’ll look at some of the other 20.
We can find a list of the proxies which extended back to 1400 AD here. The first four are quelc1-o18.dat, quelc1-accum.dat, quelc2-o18.dat and quelc2-accum.dat When we look at the file describing the various series used in the paper, we see those series were described:
Quelccaya Ice Core summit Ice O-18 (air temp) 14S 71W 470 Thompson 1982
" " Ice accumulation precip " " 488 "
" Core 2 Ice O-18 (air temp) " " 744 "
" " Ice accumulation precip " " 744 "
There are a couple of obvious questions here. First, why are four of the 22 proxies that extend back to 1400 AD all from the same location? One of the primary reasons to use something like PCA is to prevent over-sampling areas. That purpose is defeated if many proxies you don’t use PCA on are from a single location.
Second, why are two of them listed as precipitation proxies? Are we to believe precipitation proxies should be used to reconstruct temperatures? I have to assume so. That’s the only explanation I can see for three more proxies from that period: seprecip-nc.dat, seprecip-sc.dat and seprecip-ga.dat. If the “precip” in there names isn’t obvious enough, here are their descriptions:
Southeast U.S-N. Carolina Dendro ring widths precip 36N 80W 1005 Stahle et al 1988 Southeast U.S-S. Carolina Dendro ring widths precip 34N 81W 1005 " Southeast U.S-Georgia Dendro ring widths precip 33N 83W 1005 "
That’s five proxies explicitly stated to be precipitation proxies. ~25% of the proxies that covered Mann’s entire period don’t measure temperatures at all. Combined with the two we’ve previously discussed, some ~30% of his 1400 data was inappropriate. That value goes up to ~35% if we don’t oversample Quelccaya (located in the Andes mountains).
It gets weirder if we look at the full, 112 proxy roster. In addition to the proxies mentioned above, it lists floresc.dat, redsea-o18.dat, chiriq-o18-ann.dat, and java.dat. These are described:
Burdekin River Coral-fluorescence Precip/Runoff 20S 147E 1746 Lough 1991 Red Sea Coral-O-18 SST/Precip 29.5N 35E 1788 Heiss, 1994 Gulf of Chiriqui, Panama Coral-O-18 Precip 7.5N 81W 1708 Linsley et al, 1994 Java Dendro ring widths precip 8S 113E 1746 Jacoby & D'Arrigo 1990
Three of the four are listed as precipitation proxies, and a fourth is listed as both a sea surface temperature and precipitation proxy. We won’t count that fourth one. That puts us up to eight different precipitation proxies used to reconstruct temperatures.
Three more are found when we look at the output of PCA applied to the “Stahle Oklahoma” network. It’s listed as giving three PCs, pc01.out, pc02.out and pc03.out. These three PCs were derived from 14 individual series. Here are the descriptions of the first three:
Neosho R. OK Dendro ring widths precip 37N 94W 1681 Stahle & Cleaveland 1993 Keystone Res. OK " " 36N 96W 1613 " Canadian Res OK " " 35N 98W 1682 "
That pattern continues. That means those three PCs were precipitation proxies. That puts us up to 11. Eleven out of 112 proxies used by Michael Mann were stated to be precipitation, not temperature, proxies.
But there’s more. Another 11 proxies listed prec-1820-01.dat, prec-1820-03.dat, prec-1820-04.dat… prec-1820-12.dat. One would naturally think those are precipitation proxies, putting our count up to 22. However, that misses out on an important point. Those 11 are described under the section, “LONG INSTRUMENTAL DATA”:
Station Precipitation " precipitation 12.5N 82.5E 1813 "
" " " 17.5N 72.5E 1817 "
" " " 37.5N 77.5W 1809 "
The eight I don’t show are in the same form. What this shows is those 11 proxies aren’t just measures of precipitation, they’re instrumental records. It’s difficult to understand how one does a paleoclimatic reconstruction while using instrumental data. As though that’s not weird enough, if you check the latitude/longitude listings of these series, 10 of the 11 values aren’t even close to the correct locations.*
In any event, that puts us up to 22 precipitation proxies, 11 of which are instrumental. The next set of proxies to examine are temp-1820-01.dat-temp… 1820-05.dat, temp-1820-07.dat, temp-1820-09.dat… temp-1820-13.dat. These eleven proxies are listed under the same “LONG INSTRUMENTAL DATA” header (I’m showing only the first three):
Station temperature gridded average surf air temp 42.5N 92.5W 1820 Jones & Bradley 1992
" " " 47.5N 2.5E 1757 "
" " " 47.5N 7.5E 1753 "
These are all temperature measurements done by man. That is, Mann used 11 instrumental temperature series to help reconstruct temperatures. There’s also another series (england.dat) which is mostly instrumental data, but we won’t consider it here. Instead, we’ll just leave the count as 22 precipitation proxies and 22 instrumental series, 11 of which overlap.
If you believe a temperature reconstruction should be made with temperature data, you must conclude 22 of Mann’s 112 (20%) proxies were inappropriate. If you believe a paleoclimate reconstruction should be made with data taken from nature, not from manmade measurements, you must conclude 22 of Mann’s 112 (20%) proxies were inappropriate. If you believe both, feeling paleoclimatic temperature reconstructions should be made with temperature data taken from nature, you must conclude 33 of Mann’s 112 (29%) proxies were inappropriate.
And that’s a minimum, gotten if we don’t even look at any of the data or sources.
*A description of the error and how it happened can be found in this post. That post also shows the same error was repeated in a 2007 paper Mann authored. That’s remarkable as the error was pointed out to him multiple times between 1998 and 2007.
Izuru 
It might not be appropriate for -this- post, but I’d find it useful to expand on what the proxies are being used as proxies -for-.
That is: Pretend all 112 proxies -were- straight up high-quality thermometers (instead of the various things chosen). They’re pruned down with weighting to just the few key proxies and then given the task of proxying NH temperature. Or global temperature.
1) Even at the peak of thousands of actual HCN surface stations in the US alone the error bars for the -instruments- don’t approach Mann’s confidence in his reconstruction.
2) It’s easier to see how odd it is to say “this one!” is best. Yes, Seattle might be the perfect proxy -this- year for ‘GMST’, but it might be Dubai next year, and … weather changes shift the perfect proxy some, and -climate- changes should ‘shuffle the deck’ entirely.
Al, my plan has always been to write an overview after I finish this series. I don’t know what exactly it’ll cover though. It’s hard to know how much detail to cover at any given point.
One thing I want to be clear on though is Michael Mann didn’t prune the 112 proxies. Mann started with 415 series, and he reduced ~300 of them to 31. Even then, he didn’t prune any. He just combined a bunch. It’s sort of like if you averaged series ~300 series in sets of 10. You’d wind up with ~30 series, and all the original series would have some influence (though Mann’s methodology made it so some series had more influence than others).
That said, your point is right. In 1820 Michael Mann had 112 proxies with which to reconstruct temperatures for the northern hemisphere. In 1400, he had 22. I haven’t covered this yet, but in another paper he extended his 1998 reconstruction back to 1000 AD. He had only 14 proxies for that.
There’s no way to justify the certainty Mann stated his results with, even if none of the problems I have pointed out in this series were true.
When you do your wrapup, you may want to refer to “The role of statisticians in public policy debates over climate change” by Richard L. Smith
Department of Statistics ans Operations Research, University of North Carolina. Published in the American Statistical Association newsletter, Spring 2007 http://www.pnl.gov/statenvi Summarizes a talk on this subject by Wegman at a statisticians conference; interesting reading.
And of course the long report by two statisticians showing that the error bars on the temp proxies were so wide as to make Mann and others temp reconstructions worthless. Can’t recall the paper’s author/title, but extensively discussed at CA & elsewhere at the time.
I’m enjoying the series! Cheers — Pete Tillman
“Pruning” was a poor word choice on my part.
I was talking more about the -effective- elimination inherent to the methodology. The worst of the 112 (or however many, in whichever time period) has -far- less impact than the #1 best. IIRC (Which I may well not), Steve ran Mann’s method with -large- chunks of the “worst” proxies removed.
This is a separate issue from the “robustness” issue, where the plan is to “remove #1 best, and show good agreement (or not)” then “remove #2 best, show good agreement (or not)”.
This is: remove #20 through #112 (after sorting) and still get a near identical result. It indicates a such a low impact that you’re -effectively- down to an untenable number of actually used key proxies.
“some series had more influence than others” is a massive understatement IMNSHO.
Ah, I see what you mean Al. You’re right about some proxies having far more impact than others. In effect, Mann’s hockey stick comes from NOAMER PC1 and the artificially extended Gaspe. You can replace the rest of the proxies with random noise, and you’d still get a hockey stick. 110 of the proxies basically do nothing but shape the structure of the noise in the reconstruction.
I think you misunderstood my comment you quote at the end of your comment though. I said that in reference to Mann’s faulty implementation of PCA, back when there were still 415 series. The proxies you discuss in your post are after his PCA methodology was used when there were only 112 proxies. It doesn’t change your conclusion though. What you say is true for both parts. Mann’s PCA massively overweighted ~20 NOAMER series, creating the PC1 with a huge hockey stick. That PC1 (plus Gaspe) then determined the final reconstruction’s shape (effectively being given more weight than the other 20 series).
Pete Tillman, I don’t know that I’d want to refer to Richard L. Smith’s piece. I appreciate the piece, but it’s mostly just a summary of two reports. I’d probably just reference the reports directly. As for the other reference you mention, I believe you’re referring to McShane and Wyner (2010). That paper was an interesting examination of Mann’s work, and a series of papers (in favor and in opposition) stemmed from it. I don’t think those papers received the attention they deserved. I’m sure I’ll write at least a bit about them. I don’t know if I’ll discuss them in the wrapup though. The wrapup should be a fairly short piece which attempts to summarize things. It’d be hard to squeeze McShane and Wyner into that.
Of course, that’s a fair ways off. I’ve written a dozen posts thus far, and I expect to write at least a dozen more. I’ll know more about what I’ll include as time passes. Right now I’m still working out what all I want to discuss. The next couple posts will obviously be about Michael Mann’s 1999 extension of the reconstruction I’ve discussed thus far, but what should I discuss after that? Mann has a lot of papers, and there are even some unpublished ones I could discuss. Plus there’s his behind the scenes activities and various RealClimate posts I could discuss. And tons more.
In any event, I’m glad to hear you’re enjoying the series! I have topics for the next four or so posts mapped out, and I’m hoping to have one posted every other day.
Don’t forget that the locations of Mann’s instrumental precipitation series are wrong. “The rain in Maine falls mainly in the Seine”. Ironically he did the same thing in Mann et al 2008 with an instrumental precipitation series measuring the rain in Spain itself.
I didn’t. I even linked to a post discussing it in the note at the end of the post.
By the way, if teleconnections don’t care where on the planet data comes from, could they also explain why Mann’s supposed CO2 adjustment in MBH99 only affects the 1000-1400 period? If spatial locations don’t matter, who’s to say temporal ones do?
If teleconnections make any sense at all, it should be possible to come up with a concrete error estimate using existing instrumentation.
That is: Pick a calibration period (1978-1988), find -the-best- surface stations (that have records 1978-now) when compared to a satellite temperature (use Mann’s method if you like to rank and weight). Allow for ‘offset’. Now find the -error- bars “If Mann is correct, then this thermometer now reads northern hemisphere temperature … ± XX degrees C … and it should do so well outside the training period.” Now test if that makes the slightest bit of sense by testing 1988-1998, 1998-2008, and 2008-now.
I’ll predict that:
1) The error bars are -large-, at least 5C and possibly 10C or more.
2) The error bars are not the same in the different decades. (Meaning this is -not- a true calibration failing to turn a ‘proxy’ into a ‘ruler’)
3) If checked, you don’t get the same “best instruments” in the different decades.
Hi
May I vote for your proposal to look into Mann’s “… behind the scenes activities “?
I have the impression and have formed the opinion that many of the studies and publications Mann has claimed “independently” verify his hockey stick are not independent at all, but instead reflect his stage management. The withdrawn Marcott effort seems to me to be such a Mann-made production.
Pouncer, I don’t think there has been as much management by Michael Mann as you seem to think. Regardless, the documentation I’ve seen shows him using pressure to try to silence critics and otherwise bolster his case, but I’m not aware of him being responsible for new work being created. If it’s there, I’m not aware of it. I’m also not aware of what you mean by “[t]he withdrawn Marcott effort.” As far as I know, Marcott et al worked without Mann’s input and have never withdrawn anything. The only temperature reconstruction I can recall being withdrawn was by Gergis et al.
Given the information I have, I don’t think I could write what you seem to have in mind. Fortunately, the issue of non-independence isn’t tied to any of that. Non-independence stems primarily from the massive overlap in paper authors and data. Off the top of my head, I’d estimate there are maybe 20 authors responsible for 15 notable papers. More troubling, there are fewer than a dozen proxies which show any sort of a hockey stick. You don’t need any behind the scenes manipulation for non-independence in that. It’s pretty much guaranteed.
That’s actually an issue I’ve wanted to write about for some time. The problem I see is it isn’t tied to Mann’s lawsuit so I don’t think it’d fit this series. I’m toying with the idea of covering it after the Mann series instead.
That is, if I can actually finish this series. I got a library card last Friday, and since then I’ve read six books. I love it, but I lose track of time when I’m reading. Two days is the same as two weeks if I have enough material to read. I can still keep focus on something if I’m constantly involved (such as participating in a discussion), but remembering to write/post new posts? I think that’s the reason people have editors.
Now I’m going to go back to my latest book. Halfway through book seven!
As a moderation note, I’ve removed two comments from Kevin O’Neill because he is under moderation. He is fully aware of this but chooses to ignore my instructions regarding his moderation. You can find information about this action here.