How Valuable is the Marcott 2013 Holocene Temperature Reconstruction?

This is part one of a two part series on Marcott 2013. You can read the second part here.

In 2013, Shaun Marcott and his collaborators published a global temperature reconstruction covering basically the entire Holocene - the last 11,300 years.[1] It was a significant achievement, and so it seems it must be vehemently attacked. The paper is nearly 10 years old at this point, and it's been both replicated and improved upon by later studies.[2][3] Perhaps therefore it's not really necessary for me to defend a paper that has both already been successfully defended[4][5][6][7] and has been replicated by further studies which use both more extensive proxies[2] and reanalyses.[3] However, the aspects of climate science that I find most fascinating are paleoclimate studies, and I would like this blog to develop a more or less "complete" treatment of the major paleoclimate studies that impact our understanding of climate science today.

Seasonal Biases

Long before Marcott's paper was published, it was well understood that existing proxy data used to reconstruct the warmth of the Holocene Thermal Maximum (HTM) was affected by seasonal biases. Proxy data was generally from the extratropical Northern Hemisphere, and they reflected summer temperatures more than annual temperatures, creating a warming bias in the proxy record.[8] Due to  variations in the earth's orbit (Milankovitch Cycles), Arctic summers and Antarctic winters were likely warmer than during the HTM while the tropics were significantly cooler. For this reason, it's likely that the HTM in Marcott is too warm due to these seasonal biases. The difference between expectations from orbital cycles and proxy evidence can be seen by comparing paleoclimate models to Marcott 2013.[9]

Comparison of Models and Proxies

However, in 2021, Samantha Bova and her colleagues developed a way to correct for these known biases in the proxy data. The a section of the graph below shows the Marcott reconstruction with the Pages 2K and Temperature 12K reconstructions. The b and c sections of the graph show the effects of correcting these biases in mean annual sea surface temperatures (MASST). Section d shows that the bias correction buts temperatures in stronger agreement with expected temperatures from changes in greenhouse gas concentrations throughout the Holocene.

Bova's Corrections to Seasonable Biases

So we can be pretty confident here that Marcott's reconstruction overestimates the warmth of the HTM relative to today, and so does the Temperature 12K reconstruction to a lesser extent. Continued use of Marcott, therefore, should include these qualifications. At the same time, though, when comparing recent warming in global temperatures to the rest of the Holocene, showing that current temperatures are even warmer than Marcott's too-warm estimate of the HTM should show just how significant current warming is compared to variability in the earlier parts of the Holocene.

Osman's Reconstruction of Global Temperatures

The newest reconstruction of global temperatures encompassing the entire Holocene (and more) shows the effect of correcting for these biases.[3] The long term trend over the Holocene is much flatter than Marcott's reconstruction, with the HTM not as much warmer than the LIA. So the last 100 years has witnessed what seems like extraordinary warming rates compared to Marcott, but these rates are even more extraordinary when compared to reconstructions that correct for seasonal biases. To be clear, these are GMST estimates, so variability in regional temperatures is not included here. There is still evidence that the Arctic has been warmer during the HTM than it is today due to the Earth's orbital geometry - the Arctic received more summer sunlight. But the Arctic is also currently warming 3x as rapidly as the rest of the global average, so this is not likely to last for long.

Robustness

Shortly after publication of Marcott's paper, it received extensive attacks, most notably from Steve McIntyre[11][12] and Roger Pielke, Jr.[13]. The most damaging of these attacks is the claim that the most recent years of the Marcott reconstruction (1890-1950) lacks robustness. On social media, this is frequently generalized to a claim the entire reconstruction is not robust. The initial criticism was generated by comparing Marcott's published paper to the doctoral dissertation version of the paper.[11] This observation leads McIntyre to wonder, "I wonder what accounts for the difference." Reading the paper might have answered this question, but more on that later. In an interview on RealClimate, Marcott clarified a point about the last 60 years of his dataset that was already in the paper.[4] He said,

Our global paleotemperature reconstruction includes a so-called “uptick” in temperatures during the 20th-century. However, in the paper we make the point that this particular feature is of shorter duration than the inherent smoothing in our statistical averaging procedure, and that it is based on only a few available paleo-reconstructions of the type we used. Thus, the 20th century portion of our paleotemperature stack is not statistically robust, cannot be considered representative of global temperature changes, and therefore is not the basis of any of our conclusions.

This statement led McIntyre to make what seems to be a pretty outlandish claim. "Without providing any links to or citation of Climate Audit, they now concede: '20th century portion of our paleotemperature stack is not statistically robust, cannot be considered representative of global temperature changes, and therefore is not the basis of any of our conclusions.'" So McIntyre is complaining that he didn't get credit for discovering what Marcott later had to "concede." On the same day, Pielke Jr's blogpost comes just short of accusing Marcott and Science of misconduct.[13] There's just one problem with all this. As we've already seen in the interview, Marcott says that this concession is "in the paper." And sure enough it is:

Without filling data gaps, our Standard5×5 reconstruction (Fig. 1A) exhibits 0.6°C greater warming over the past ~60 yr B.P. (1890 to 1950 CE) than our equivalent infilled 5° × 5° area-weighted mean stack (Fig. 1, C and D). However, considering the temporal resolution of our data set and the small number of records that cover this interval (Fig. 1G), this difference is probably not robust. Before this interval, the gap-filled and unfilled methods of calculating the stacks are nearly identical (Fig. 1D).

The paper concedes that the last 60 years is not robust due the "temporal resolution" of the data and the "small number of records" for the period between 1890 and 1950. Marcott was not admitting something because of some analysis by McIntyre. He was clarifying what was already in the paper. McIntyre then goes on to complain about the fact that he says the "difference" between the Standard and RegEM methods is not robust, not that the dataset between 1890 and 1950 not statistically robust. But what Marcott is saying is that the Standard and RegEM  methods are "nearly identical" before the 20th century. They are not identical for 1890-1950 because of the temporal resolution of the data and small number of records. The size of the uptick is too sensitive to the methodology used to make the reconstruction. In fact, the Standard method shows about ~0.1 C/decade more warming than RegEM between 1890 and 1950. That's obviously too much of a difference, especially when you compare this to the instrumental record. If that difference is not statistically robust, 20th century warming in Marcott's paper isn't robust either.

What's happening here is pretty clearly described in a Tamino post.[5] The "Standard 5x5" unfilled reconstruction shows more warming than the RegEM infilled reconstruction, and also more warming than the instrumental record, mostly because of a single data point in 1940. The reason for this uptick is precisely what Marcott indicated - temporal resolution and lack of proxy records. There are 73 total proxies, all of which overlap 5500-4500 BP, but only 18 overlap 1940. That means proxies dropped out of the stack with time. This drop out can easily affect the average of the "Standard 5x5" reconstruction, and it's not hard to see why. Take an example of a class with 25 students, and you want to track their average test scores. There are several ways you can increase the average test scores of the class. You can (1) give them the tools they need to improve their scores, (2) add 5 smart students to the class or (3) remove the 5 poorest students from the class. Each of these improve the average test scores, but neither (2) nor (3) should be confused with (1). In Marcott's Standard reconstruction, if more cool proxies than warm proxies drop out (are removed from the class) then the average temperature will increase due to the shortage of proxies, not because the globe is actually warming (students improving their grades). This problem does not effect the RegEM reconstruction as much as the Standard reconstruction, so the difference between them indicates that the 20th century warming is not robust. You can't say that there has been a robust warming because it's too sensitive to the method used. This is the point that Marcott clarified in the RealClimate post, and it's the sympathetic reading of the admittedly awkward statement in the paper.

The fact remains that the McIntyre and Pielke Jr criticisms of the paper don't actually affect the paper's integrity or value at all. This can be seen for two reasons: (1) Marcott did not base any conclusions on the 20th century data (he used the instrumental record instead) and (2) the 20th century portion of the data is not even the most interesting or valuable portion of the dataset. What's interesting is the proxy data before the beginning of the instrumental record, which is statistically robust. But the 1850-1950 portion of the dataset can be used to establish an alignment of the dataset with the instrumental record, and that is why it's included in the paper. McIntyre's criticism of the paper on the basis of the lack of statistical robustness doesn't hold water.

But what should be clear is that Marcott's reconstruction is robust over almost the entire time frame of the dataset, and the only portion that isn't robust is where it doesn't need to be robust. We have the instrumental record to show the large uptick in temperatures since 1850. In the graph at the top of this post, I showed both Marcott 2013 with the 1σ uncertainty envelope and HadCRUT5. I set both to the 1850-1900 mean and added 20-year smoothing to the instrumental record to match the resolution of Marcott. The results show (1) that GMST during the 21st century is likely warmer than the warmest 20-year means of Holocene, that (2) the rate of increase in global temperatures is unprecedented in the Holocene, and (3) given uncorrected seasonal biases, this is a conservative look at how rapidly we are leaving anything we've seen during the Holocene.

References:

[1] Marcott, Shaun et al. “A Reconstruction of Regional and Global Temperature for the Past 11,300 Years.” Science 339 (2013): 1198-1201. http://shpud.com/Science-2013-Marcott-1198-201.pdf

[2] Kaufman, D., McKay, N., Routson, C. et al. Holocene global mean surface temperature, a multi-method reconstruction approach. Sci Data 7, 201 (2020). https://doi.org/10.1038/s41597-020-0530-7

[3] Osman, M.B., Tierney, J.E., Zhu, J. et al. Globally resolved surface temperatures since the Last Glacial Maximum. Nature 599, 239–244 (2021). https://doi.org/10.1038/s41586-021-03984-4

[4] "Response by Marcott et al." Realclimate. https://www.realclimate.org/index.php/archives/2013/03/response-by-marcott-et-al/

[5] Tamino, "The Tick," Open Mind. 
https://tamino.wordpress.com/2013/03/22/the-tick/

[6] Tamino, "Global Temperature Change - The Big Picture," Open Mind. 
https://tamino.wordpress.com/2013/03/22/global-temperature-change-the-big-picture/

[7] Tamino, "Regional Marcott," Open Mind. 
https://tamino.wordpress.com/2013/03/23/regional-marcott/

[8] Michael Mann, “OK, But we do know it was warmer than present 6000 years ago, don’t we?” https://www.realclimate.org/index.php/archives/2004/12/ok-perhaps-recent-20th-century-warmth-is-anomalous-over-the-past-millennium-or-two-but-wasnt-it-warmer-during-the-holocene-optimum-some-6000-years-ago/

[9] Richard Telford, “Seasonally biased proxies and the Holocene temperature conundrum”

https://quantpalaeo.wordpress.com/2014/08/14/seasonally-biased-proxies-and-the-holocene-temperature-conundrum/

[10] Bova, S., Rosenthal, Y., Liu, Z. et al. Seasonal origin of the thermal maxima at the Holocene and the last interglacial. Nature 589, 548–553 (2021). 

https://doi.org/10.1038/s41586-020-03155-x

[11] Stephen McIntyre, "No Uptick in Marcott Thesis." Climate Audit. 
https://climateaudit.org/2013/03/14/no-uptick-in-marcott-thesis/

[12] Stephen McIntyre, "The Marcott Filibuster. Climate Audit. 
https://climateaudit.org/2013/03/31/the-marcott-filibuster/

[13] Roger Pielke Jr, "Fixing the Marcott Mess in Climate Science"
http://rogerpielkejr.blogspot.com/2013/03/fixing-marcott-mess-in-climate-science.html