I’ve had a longstanding interest in high-resolution ocean proxies (with posts as early as 2005 – see Ocean Sediment tag) and had already written detailed reviews of many of the individual high-resolution series used in Ocean2K (e.g. here here here here here here). In these prior discussions, the divergence between the 20th century proxy data and 20th century instrumental data had been a major issue. The non-bladedness of Ocean2K data was therefore unsurprising to me.
Although, for their main figures, the Ocean2K authors made the questionable decisions to voluntarily degrade their data both into 200-year bins and from deg C to SD units, in their Supplementary Information, they identified a network of 21 series with high-resolution extending into the 20th century, showing results in 25-year bins but only for the short period 1850-2000, once again re-scaling, this time using only six values (25-year bins) for each series.
In my first post, I had undertaken to examine their data in higher resolution and will do so today using their high-resolution network – without the needless re-scaling and over the entire 0-2000 interval. The results clearly confirm the absence of a 20th century blade. The Ocean2K authors were singularly uninformative about this obvious result; I’ll show how they worked around this “problem”. I’ll also discuss untrue claims by Ken Rice (ATTP) and other ClimateBallers that the Ocean2K data “finishes in 1900” or is otherwise too low resolution to permit identification of a concealed blade.
Background: the Ocean Proxy “Divergence Problem”
I had initially become interested in high-resolution ocean data (especially alkenone and Mg/Ca, rather than dO18) because, as opposed to tree rings, they were directly calibrated in deg C according to standard equations (not ex post correlations).
Alkenone series are based on the ratio of C37:2 and C37:3 in coccolithopores, while Mg/Ca series are based on ratios in foraminifera, with surface dwelling foraminifera (especially G. ruber) being of interest. During the past 20 years and especially the past 10 years, alkenone samples have been widely collected throughout world oceans and coretop and sediment trap calibrations yield sensible maps of ocean temperature without jiggling. In deep time, they also yield “sensible” results. Alkenone series constitute 15 of 21 high-resolution of the Ocean2K dataset (26 of 57 overall) and also the majority of the Marcott ocean data (31 of 60), with foraminifera Mg/Ca being the second-largest fraction.
Alkenone and Mg/Ca series had originally been collected to shed light on “deep time”, but there were occasional box cores which both preserved the most recent sediments (a sampling problem with piston cores) and which had been sampled at sufficiently high-resolution to shed light on the past two millennia. I’ve made a practice of regularly examining the NOAA and Pangaea datasets for potentially relevant new data and, over the past 10 years, had already noticed and separately discussed many of the series in the high-resolution Ocean2K dataset (e.g. here here here here here here).
Here, for example, is a figure from Leduc et al 2010, previously shown at CA here, showing dramatic decreases in alkenone SST at two sites: Morocco, Benguela. (Both these sites are included in the Ocean2K high-resolution dataset, both with more than thirty 20th century values.) Numerous other CA posts on the topic are in the following tags: Ocean sediment; Alkenone.
Figure 1. From Leduc et al 2010. Both locations are in the Ocean2K high-resolution network.
In a number of CA posts, I had questioned the “alkenone divergence problem”, the term alluding to the notorious divergence between instrumental temperatures and tree ring density proxies that had given rise to various “tricks” to “hide the decline” in Mann’s section of IPCC TAR and other articles in order not to “dilute the message”. In important ways, the alkenone divergence problem is even more troubling as (1) there is a physical calibration of alkenone proxies, whereas tree ring densities are merely correlated after the fact; and (2) alkenone proxies have “sensible” properties in deep time.
The “problem” arising from divergence between a proxy reconstruction and instrumental temperature is that such divergence makes it impossible to have confidence in the proxy reconstructions in earlier periods without reconciling the divergence. Mann, for example, has always insisted that his reconstructions have statistical “skill” in calibration and verification periods, though the validity of such claims has obviously been at issue.
A Reconstruction from the Ocean2K “High-Resolution” Dataset
The Ocean2K data consisted of 57 series of wildly differing resolution: nine series had fewer than 20 values, while twelve series had more than 100 values. In geophysics, specialists always use high-resolution data where available and use low-resolution data only where better data is unavailable. In contrast, in their main figures, the Ocean2K authors degraded all their data into 200-year bins and made composites of the data only after blurring.
In their Supplementary Information, the Ocean2K authors identified a subset of 21 high-resolution series: twelve of the 21 series had more than 20 values in the 20th century, seven series had more than forty 20th century values and all but one had more than eight values. In Figure S10, they showed a high-resolution composite in 25-year bins, but only for the 1850-2000 period and only in SD Units (1850-2000 and only after binning).
Because the underlying proxy data is already in deg C, it is trivially easy (easier in fact) to do the Ocean2K calculations in deg C rather than SD Units and it’s hard to believe that the Ocean2K authors hadn’t already done so. Figure 2 below shows the composite of high-resolution ocean cores in 25-year bins over the 0-2000 period (rather than 1850-2000) and in deg C (rather than SD units.) For comparison, I’ve also shown instrumental HadSST (black) and the composite from the full network in Ocean2K technique using 200-year bins (but retaining deg C). Expressed in deg C, there is a major divergence in the 20th century between instrumental temperature and the proxy reconstruction. Even late 20th century proxy values are clearly below medieval values.
Figure 2. Red – 25-year bin composite of Ocean2K high-resolution ocean cores (not using one incongruous singleton coral series) retaining deg C. Magenta- composite for full network, calculated as in Ocean2K composite, but retaining deg C throughout. Black – HadSST global (since ERSST global series only begins in 1880.)
McGregor et al made no mention of this dramatic divergence in their main text, instead asserting that “the composite of reconstructions from tropical regions are in qualitative agreement with historical SST warming at the same locations”:
Although assessment of significance is limited by the number and resolution of the reconstructions, and by the small amount of overlap with historical SST estimates, we find that the composite of reconstructions from tropical regions are in qualitative agreement with historical SST warming at the same locations (Supplementary Fig. S10). Upwelling processes recorded at a number of the sites may also influence the twentieth-century composite (Supplementary Sections 1 and 8).
Even if the tropical composite was in “qualitative” agreement (a point that I will examine in a future article), this implies that the extratropical divergence has to be that much worse in order to yield the actual overall divergence. It is very misleading for the authors to claim “qualitative agreement” in the tropics without disclosing the overall divergence.
Deep in their Supplementary Information (page 44), they quietly conceded that the high-resolution composite did not yield the warming trend of the instrumental data, but there is no hint of this important result in the text of the article:
The 21_O2k and 21_Kaplan composites are non-significantly correlated (r2 = 0.17, df = 4, p = 0.42), with the warming trend in the 21_Kaplan not reproduced in the 21_O2k composite (Supplementary Fig. S10).
They illustrated this with the following graphic (in 1850-2000 SD units after binning). While the use of SD Units degrades the data, even this figure ought to have been sufficient to dispel the speculation of some ClimateBallers that the 1800-2000 bin might combine low 19th century values and high 20th century values, thereby concealing a blade.
Figure 3. Excerpt from Ocean 2K SI Figure S10, showing Kaplan SST (top panel) and Ocean2K high-resolution composite, both expressed in 1850-2000 SD Units (after binning).
While the SD units of the 200-year bin and 25-year bin figures are not the same, I think that it is still instructive to show the two panels with consistent centering. In the figure below, I’ve centered the panel showing the 25-year bins so that it matches the reference level of the final (1800-2000) bin of the 200-year reconstruction, further illustrating that its final bin does not contain a concealed blade.
Figure 4. Left panel – Ocean2K in 200-year bins (PAGES2K FAQ version from here); right – bottom panel of SI Figure S10a, with its zero value aligned to value of 1800-2000 bin in left panel. Both panels in SD Units (not deg C). I’ve been able to closely emulate results in the left panel, but not as closely in the right panel.
The Supplementary Information carries out corresponding analyses on subsets of the high-resolution data: tropical vs extratropical, upwelling vs non-upwelling, alkenone vs Mg/Ca. Trying to analyse the divergence through such stratification is entirely justified, though the actual statistical analysis carried out by the Ocean2K authors is far from professional standard. I’ll discuss these analyses in a separate post. For now, I’ll note that similar concerns have been raised about alkenone data in a Holocene context, even by Ocean2K authors. Lorenz et al 2006 (discussed at CA in early 2007 here) had contrasted trends in tropical vs extratropical alkenone data over the Holocene; in my commentary, I had pointed out the prevalence of upwelling locations in the tropical data.
Postscript: False ClimateBaller Claims that the Data “Finishes in 1900”
In reaction to my first post, Ken Rice (ATTP) and other ClimateBallers argued that there was no reason to expect the Ocean2K data to have a blade, since the data supposedly ended in 1900 or was otherwise too low resolution. Such claims were made at David Appell’s here, at Rice’s blog here and on Twitter.
As I observed in my post, the Ocean2K data archive is excellent and the measurement counts easily calculated. A barplot of measurements (grey) and cores (red) is shown below. Not only does the data not end in 1900, the number of individual measurements from the 20th century is larger than any previous century. Nor is this sample too small to permit analysis. 21 series are considerably more than the number of medieval proxies in many canonical multiproxy studies that are uncontested by IPCC or ClimateBallers. While it would be nice to have more data (especially in the Southern Ocean), there’s easily enough 20th century data to be worthwhile discussing.
Figure 5. Number of measurements in the Ocean2K dataset by 20-year period (grey); number of contributing series by 20-year period (red).
Now consider various assertions about the data made by Rice and others. Shortly after my original article, Rice stated (here and here) that the data ended in 1900 and thus there was no reason to expect a blade.
“As far as I’m aware, it finsishes in 1900 and the paper has “pre-industrial” in the title. So why would we expect it to have a blade?”
Rice even accused me of “misread[ing]” the x-axis:
Can’t quite work out how you’ve managed to misread the x-axis so badly?
I informed Rice in a comment at Appell’s that his belief that ended “in 1900” was incorrect as follows:
Ken says: “As far as I’m aware, it finsishes in 1900 and the paper has “pre-industrial” in the title. So why would we expect it to have a blade?” The data doesn’t end in 1900. There are more measurements in the 20th century than in any previous century. The 20th century data doesn’t have a Hockey Stick either, as you can see in their Figure S10a.
I had also posted a Twitter comment highlighting that, even in the Ocean 2K step graph, the final (1800-2000) bin of the step-graph extended to 2000. Rather than defend his false claims, Rice made a Gavinesque exit, but not before making an unsupported allegation that I was spreading “misinformation” about the Ocean2K study:
Nonetheless, a few days later, Rice returned to the topic in a blog article on Sept 13, re-iterating his untrue claim that the Ocean2K data ended “in 1900”:
Steve McIntyre (who was involved in the discussion on David Appell’s blog) seems to be highlighting that the recent Ocean2K reconstruction does not have a blade. Well, the data appears to end in 1900 and the paper title is Robust global ocean cooling trend for the pre-industrial Common Era, so why would we expect there to be a blade.
This time, one of his readers (improbably, Sou) pointed out to Rice that the 1800-2000 bin must include 20th century data. Sou speculated that 200-year bin could contain a concealed blade in the bin through a combination of cold 19th century values and warm late 20th century values – apparently unaware that this possibility had already been foreclosed by Supplementary Figure S10):
I don’t know that the recent ocean2k paper ended in 1900. I think what it did was end in the 1801 to 2000 “bin”, which would have included the coldest years of the past 2,000 years, as well as whatever proxy records were included up to 2000. The boxes in Figure 2 showed a lot of things, including the median for each 200 year bin, the latest of which was centred on 1900 – but went from 1801 to 2000.
Rice amended his post to say that his prior assertion (that the data ended in 1900) wasn’t “strictly correct”:
What I say here isn’t strictly correct.
However, the issue is not that his original assertion wasn’t “strictly correct”; the issue was that it was unambiguously wrong.