Playing hockey – blowing the whistle

by Rud Istvan

This instantly ‘famous’ 2013 Science hockey stick paper derived from Marcott’s 2011 Ph.D thesis at Oregon State University, available here. His thesis doesn’t show a hockey stick ‘blade’ projecting above its anomaly baseline NCDC 1961-1990. H/T to Jean S, posted at Climate Audit. Something changed after the thesis was published to produce the new ‘blade’ in Science. That something was significant, since the Science paper’s Supplementary Information discussion said it did not enable discriminating such a temperature variation (i.e. a ‘blade’) on such a short a time scale.

See my previous post here, Lets Play Hockey Again, which was right about the Medieval Warm Period (MWP) but wrong about the ‘blade origins. Mea Culpa. Here we deduce whence the ‘blade’ came using ‘crowd-sourced’ science.

First, SI table S1 and thesis figure C1 confirm both use exactly the same 73 proxy records. The only difference is that thesis series MD98-2181 was renamed KY07-04-01 in the SI.

Second, thesis figure C7 graphed each proxy series individually using the mean and 2 sigma error band from its Monte Carlo statistical error estimation technique. A visual check confirms that there is no near T0 (1950) ‘blade’ visible in any of 73 dithered proxies, although five have distinct upticks near ‘present’ (the past 100-200 years or so visually). These include Dome C (Antarctic ice core δD with ±30% uncertainty), MD98-2165 (equatorial Mg/Ca), TN057.17 (temperate diatom mat with ±0.75°C uncertainty), Agassiz Renland (Arctic borehole δ¹⁸O with ±30% uncertainty), and GeoB3910, an equatorial alkenone. The thesis itself shows (below) that these are insufficient to uptick the ensemble.

Importantly for what follows, figure C7 also shows start and end dates for each proxy series as given by their references, along with their proxy age/date [radioisotope] controls. 20 unequivocally carried through about 1850, and 9 extended beyond T0 (1950). All the other proxies ended earlier in time. For example, proxy MD95-2043 (Cacho 2001, alkenone, Western Mediterranean) ended about 942 AD per its reference, with a notable ending uptick from the Medieval Warm Period. Remember this proxy for a couple of pages more.

Third, the outstanding Ph.D thesis on which this Science paper was based does not contain a hockey stick ‘blade’ anywhere. We can compare five important correlated pieces of information from its chapter 4 and supplemental data to deduce where and how the Science ‘blade’ originated. To start, both 2000-year reconstructions were compared to Mann 2008. The purple lines are the ‘standard’ simple area weighted mean 5×5 grid reconstructions.

The Science ‘blade’ is obvious. It does not exist in the thesis. Something changed.            

The ‘blade’ (above anomaly baseline) is visibly present in all Science reconstruction methods, yet in none in the thesis. It is therefore not an artifact of the various reconstruction methodologies. Something changed in either the data or in the data processing algorithms to produce the blade in Science.

The blade is obvious in the mean of all 1000 dithered Monte Carlo proxy simulations in Science, yet is non-existent in the equivalent 10000-simulation thesis mean. The thesis error band widens symmetrically and significantly as T0 is approached. It must, given that Science reported the mean proxy resolution is 180 years, and the median is 120 years. This validates the thesis’ basic statistical algorithm as directionally correct and obviously useful.

The schematic algorithms used to generate the above error information are identical. Without comparing the underlying code, it is impossible to ascertain with complete certainty that there was no algorithm change in Science. But beyond any reasonable doubt, the ‘blade’ arose from some data change rather than from some algorithm change

The ‘blade’ is only present in data for figure S12a’s 20-year sampled reconstruction after 1900.  It is not present in the 100-year version that goes to 1900, and which closely tracks the 20-year version to that time. Although lower frequency sampling will mask any earlier changes, significant data changes obviously occurred in the years after 1900 where the blade arose.

The latitude coverage in Science and the thesis is identical (horizontal grey bars). It has to be, since using the same proxies. The number of Mann2008 proxy records increases identically as the present is approached (vertical tan bars). They have to, since the same study. In both Science and the thesis, the number of Holocene proxies available at ‘2000’ is identically about 70. It should be. (And actually is exactly 70 based on scrutiny of thesis figure C7). These figures appear to be accurate visual representations of the underlying analytical facts.

But the number of Holocene proxies is not the same after about 1550 (500 before T0). Oddly, Science includes more proxies until after 1900. Even more oddly, in the thesis 9 ‘survived’ 1950, yet in Science none did.

Zero. And that is the proverbial smoking gun. The data were changed by re-dating, and Science figure 1G proves it without digging into proxy details buried deep within the SI proxy data spreadsheets.

The authors of this paper re-dated selected proxies after the thesis to take advantage of ‘mean dropout’ to fabricate a ‘blade’. As proxies end, only those that remain contribute subsequent information.  Three ‘proxies’ over five time periods illustrate how this works. Each hypothetical ‘proxy’ individually signals no change over time, although proxy 3 oscillates neutrally around a 0 anomaly.

Time period       4      3      2      1      T0

Proxy 1                -1     -1     -1     -1      na

Proxy 2               +1   +1    +1    +1     +1

Proxy 3                +1    -1     +1     -1     +1

Recon. mean   +1/3  -1/3 +1/3 -1/3 +1

To generate the blade in Science, it sufficed to ‘pull back’ in time some ending downtick proxies, and pull forward to T0 some ending upticks. It is evident from 1G compared to 4.3C that 9 were pulled back at least a few decades. We shall look at one below. It is also evident that about 10 others were pulled forward, since in Science at 1850 (100 before T0) there were still about 30 proxies in the area weighted mean, while in the thesis there were only about 20. The Science ‘blade’ was manufactured by ±time shifting proxy start/end dates to take advantage of mean dropout. Unfortunately, time shifting from reference dates with associated multiple age controls is nowhere noted or justified in the Science paper or the accompanying SI prose. It should have been.

The bigger question is whether it was justified even though not disclosed?

Comparing SI proxy data to the original reference data shows exactly what was done, and discloses whether re-dating might have been justified. The inestimable Climate Audit[or] Steve McIntyre has done so for the alkenone proxies (31 of 73 total), the largest subgroup. Steve plotted the re-dated Science SI alkenones using the paper’s ‘standard’ weighted area mean 5×5 reconstruction (SI data spreadsheet page 2, column AJ) to produce remarkable ‘McIntyre 1’:

The rise from 1920 to 1940 is over 1°C (more than IPCC AR4 over a century), ending far above the anomaly baseline NCDC 1961-1990, and even above NCDC 1990-2010 (red)! Alkenones alone essentially explain Science figure 1a.

Steve emailed Dr. Marcott on or about 3/13, inquiring how this ginormous sudden rise could be. The response was that the Science paper said uncertainty increased toward T0 as fewer proxies were available, so this result was probably ‘not robust’. He never said that to his many MSM interviewers about the new hockey stick, saying instead “It’s really the rates of change here that’s amazing and atypical”.  Not robust but still amazing? Steve’s email gave Dr. Marcott an opportunity to mention re-dating. He chose not to do so, perhaps unaware that published figures 1G versus 4.3C provided conclusive ‘smoking gun’ evidence without studying the gory SI data details in the SI Excel spreadsheet.

Marcott was probably also unaware that Steve had already said publicly that proxy re-dating was problematic, and that he would comment on it. The day after publishing Marcott’s nonresponse, Steve published his re-dating comment, with McIntyre 2 worth more than a thousand words. Black is Science with Marcott’s re-dating. Red is Marcott’s thesis before re-dating.

(Steve’s time axis is reverse labeled from Science, counting up not down.)

To get this subset result Marcott pulled back two negative alkenone proxies so they did not contribute to the most recent 20-year points.  OCE326-GGC30 with a steady decline was pulled back 191 years despite three radiocarbon age controls in the past two millennia. Here it is from figure C7 with its radiocarbon age controls (+).

The second ‘re-dated’ negative alkenone proxy (SI#23, Isono 2009, off central Japan) seems particularly indefensible. This proxy record is an unusually valuable composite of a piston core plus a box core from the same location. Piston cores penetrate deeply to sample long ago sediments, but the coring process makes the top unreadable (typically for many hundreds of years). Box cores are shallow in time, but preserve resolution to the core top. Isono dated his core top to 1991 (Marcott’s date notation -41), and carefully spliced the cores together with a large overlap period to create a long continuous record with about 32-year resolution. The reference core top date must be approximately correct, since 3 cm below was the classic isotope ‘bomb spike’ caused by atmospheric nuclear weapons testing in the 1950’s. Since it is not possible to ‘re-date’ the incontrovertible beginning of the atomic era, the most recent three (negative) proxy data points of Isono’s proxy just ‘disappeared’, conveniently removed from the 1940 Science ‘blade’ calculation. (Excel NaN [Not a Number] is triggered by division by 0.)

5 other alkenone proxies with ending upticks were pulled forward. MD95-2043 ends at ‘942’ according to its reference and 7 age controls. It of course shows a distinct MWP peak at its end. Marcott pulled it forward exactly 1008 years to T0 1950, so the MWP onset contributes to the 20^th century blade. (This wonderfully precise re-dating had the additional ‘virtue’ of diminishing the MWP signal.) Here are the proofs from Science SI Excel data comparison

Here are MD95 2011 (pulled forward an ‘unexplained’ 510 years) and 2015 (pulled forward an ‘unexplained’ 690 years) from thesis figure C7:

No justification is given for such extreme re-dating of proxies containing numerous radioisotope (+) age controls. If the original reference papers had large date errors, these should have been corrected in the thesis. In the absence of any kind of explanation, this has the appearance of either gross incompetence then or fabrication now.

The curious minor re-dating of GeoB 6518-1, oddly pulled back just 32 years, shows beyond any reasonable doubt it was not incompetence. Quite the opposite; the re-dating appears to have been carefully and skillfully done. This post 1950 uptick was pulled back just enough so that all of it got included in the pre-1950 ‘blade’ reconstruction. Here is that proxy from thesis figure C7:

Note the post-1950 age control (+) that Science re-dating overrode.

Hockey is a fast contact sport. The whistle has blown on an obvious high stick foul just 11 days after Science first published Marcott’s hockey stick.  This post is with sincere thanks to the formidable Steven McIntyre of Climate Audit, to whom I ‘passed the puck’ in my previous on-line posting ‘Let’s Play Hockey Again’ at Dr. Judith Curry’s invaluable Climate Etc.