Arguments
Real Skepticism About the New Marcott 'Hockey Stick'
Posted on 10 April 2013 by dana1981
A new global temperature reconstruction over the past 11,300 years by Marcott et al. (2013) has been described as 'the new hockey stick,' and adopted into 'the wheelchair' by Jos Hagelaars by including temperatures further in the past and projected for the future (Figure 1).
Figure 1: The temperature reconstruction of Shakun et al (green – shifted manually by 0.25 degrees), of Marcott et al (blue), combined with the instrumental period data from HadCRUT4 (red) and the model average of IPCC projections for the A1B scenario up to 2100 (orange).
The Marcott paper has been subjected to an immense amount of scrutiny, particularly in the climate contrarian blogosphere, with criticisms about everything from the wording of its press release to the timing of its Frequently Asked Questions (FAQ) publication. Unfortunately climate contrarians have been so noisy in their generally invalid criticisms that the media has begun to echo them, for example in this Washington Post blog.
With all the hubub, it's easy to lose sight of the important conclusions of this paper. The bottom line is that the rate of warming over the past century is very rapid and probably unprecedented for the past 11,000 years. That's actually both good and bad news.
Why Climate Contrarians Should Love the Hockey Sticks
The last, best hope for climate contrarians is for climate sensitivity (the total global surface warming in response to the increased greenhouse effect from doubled atmospheric CO2) to be low. We know the planet is warming due to humans increasing the greenhouse effect, and the only remaining plausible argument against taking action to do something about it is the hope that future climate change will be relatively minimal.
This is where climate contrarians lose the plot. It's understandable to look at 'hockey stick' graphs and be alarmed at the unnaturally fast rate of current global warming. But in reality, the more unnatural it is, the better. If wild temperature swings were the norm, it would mean the climate is very sensitive to changes in factors like the increased greenhouse effect, whereas the 'hockey stick' graphs suggest the Earth's climate is normally quite stable.
On the one hand, these graphs do suggest that current climate change is unnatural – but we already knew that. We know that humans are causing global warming by rapidly burning large quantities of fossil fuels. On the other hand, the past climate stability suggests that climate sensitivity is probably not terribly high, which would mean we're not yet doomed to catastrophic climate change. See, good news!
In their efforts to deny that the current warming is unprecedented and human-caused, climate contrarians are actually scoring a hockey stick own-goal because they're also arguing that the climate is more sensitive than the IPCC believes. For those who oppose taking major steps to reduce greenhouse gas emissions, that's the worst possible argument to make.
The good news for climate contrarians is that the current rate of global warming appears to be unprecedented over the past 11,000 years. During that timeframe, the difference between the hottest and coldest average global surface temperature is around 0.7°C, with the cooling between those temperatures happening slowly, over about 5,000 years.
Over the past 100 years, we've seen about 0.8°C global surface warming (Figure 2). While the time resolution in the Marcott reconstruction is relatively low, there is simply no evidence of a similarly rapid or large natural climate change in the past 11,000 years. As Tamino at the Open Mind blog shows, any similar rapid and large warming event to the current one would likely have shown up in the Marcott analysis, despite its low resolution. Tamino concludes,
"the Marcott et al. reconstruction is powerful evidence that the warming we’ve witnessed in the last 100 years is unlike anything that happened in the previous 11,300 years."
While it may seem counter-intuitive, that's a good thing, because it means the climate is not highly unstable.
Figure 2: Regularized expectation maximization (RegEM) Marcott reconstructions (black), plus the HadCRUT4 series in 20-year averages centered on the times of the Marcott reconstruction (blue). Created by Tamino.
The Hockey Stick 'Blade' is Real
Much of the manufactured controversy about the Marcott paper is in regards to the 'blade' or 'uptick' – the rapid warming at the end of the graph over the past century. While their reconstruction does identify an approximately 0.6°C warming between 1890 and 1950, the authors note in the paper that this result is probably not "robust." Tamino notes that this uptick appears to largely be a result of proxies dropping out (although a smaller uptick seems to be a real feature), as many individual proxies do not extend all the way to the year 1950. If proxies with colder temperatures drop out, the remaining reconstruction can show an artificial warming toward the end.
In the paper, where they talk about temperatures over the past decade, the authors reference the instrumental temperature record rather than the proxies. As Tamino notes in another excellent analysis of the paper,
"for the Marcott et al. reconstruction data coverage shrinks as one gets closer to the present. But that’s not such a problem because we already know how temperature changed in the 20th century."
Certain parties have complained that the press release and subsequent media coverage of the paper have not made it sufficiently clear that the 'blade' of this hockey stick comes from the instrumental temperature data. This is the focus of the Washington Post article, for example. However, the authors were clear on this point in the paper, and in several interviews and subsequent discussions, like the FAQ.
In reality, the 'blade' of the 'hockey stick' – the instrumental temperature record – is our most accurate temperature data set. As noted in the FAQ on RealClimate, the instrumental temperature record is also consistent with proxies from other studies. For example Anderson et al. (2012) compares their study's natural proxy temperature reconstruction (Paleo; solid line in Figure 3) to the instrumental surface temperature record (MLOST; dashed line in Figure 3) and finds a strong correlation (of 0.76) between the two. Reanalysis data, as in Compo et al. (2013), has also independently confirmed the instrumental global surface temperature record accuracy (correlations between 0.84 and 0.92), as of course did the Koch-funded Berkeley Earth study.
Figure 3: Paleo Index (solid) and the merged land-ocean surface temperature anomalies (MLOST, dashed) relative to 1901-2000. The range of the paleo trends index values is coincidentally nearly the same as the instrumental temperature record although the quantities are different (index values versus temperature anomalies °C). From Anderson et al. (2012).
There may be some valid criticism that the press release and some media discussions were not clear that the comments about recent unprecedented warming are based on comparing the instrumental temperature record to the Marcott reconstruction, but that is a very minor criticism that has no bearing on the scientific validity of the discussion or the Marcott paper. Unfortunately the media has begun amplifying this minor and scientifically irrelevant point.
Real Skepticism
It's worth taking a moment here to reflect on real skepticism. Spending literally dozens of blog posts attacking a study because its results seem inconvenient is not real skepticism. Comparing climate scientists to the mafia is not real skepticism. Nitpicking minor details in press releases and media articles while ignoring the discussion in the paper itself is certainly not real skepticism.
If you want an example of real skepticism, look no further than Tamino's Open Mind blog. Tamino read the Marcott paper, noted they had expressed doubt about the robustness of the final uptick in their proxies, looked at the data, identified the proxy dropout issue, tried some new analyses, and found that the proxy uptick is probably real but probably smaller than it appears in the paper. Also see similar efforts by Nick Stokes. These are the approaches of real skeptics. At least the manufactured controversy over the Marcott paper has served to show who the real skeptics and "honest brokers" are.
The irony is that the climate contrarians are being their own worst enemies here. A 'hockey stick' shape means less past natural variability in the climate system, which suggests that climate is relatively stable. It's revealing that in their zealotry to deny that the current global warming could possibly be unnatural and unprecedented, the contrarians are actively trying to undermine their only potentially valid remaining argument against serious climate mitigation.
Nevertheless, all signs indicate that the current rate of warming is very rapid, probably unprecedented in the past 11,000 years; that if we're not at the highest temperatures during that timeframe, we will be soon; and that despite the contrarians' best efforts to argue otherwise, we're not yet doomed to catastrophic climate change.
Also see this good post on the subject by John Timmer at Ars Technica, and this one by Climate Science Watch.
Joe T @100, you have picked up on one of the most interesting and innovative features of Marcott's reconstruction. Instead of just making a central estimate of the temperature, and showing error bars, they varied the data based on the error margins of the original proxies. By doing this, they get statistical data on all the ways that the reconstruction could be wrong. From among all those variations, only 18% of ten year intervals are warmer than 2000-2009. Allowing for some inherent smoothing in the method, that becomes 28%. That does not mean that there were any decades warmer than 2000-2009 in the Holocene. The actual temperature record will approximate to one of their statistical variations of the data, their "realizations". It is, however, as likely to be a cold realization as a warm one. Because reconstructions can be wrong by being to warm, or to cold, with equal probability, the mean does not vary as much as the realizations can, and does not show the potential warmest years.
This explanation will be easier to understand if you actually see the realizations plotted with the mean:
The idea is that current temperatures, while much higher than the mean (black line), are not higher than about the warmest of the realizations in for any given decade about 18% of the time.
The net effect from this can be seen in Marcott's Fig 3:
The 2000-2009 temperatures, with an anomaly of about 0.4 C lies in the upper end of the distribution of realizaitons (solid black curve). These can then be compared with the expected temperatures from various IPCC AR4 scenarios. (The coloured curves represent alternative means of reconstruction, and can be ignored for this discussion.)
Hey Tom, thanks very much for a very cogent explanation. I appreciate it.
Angusmac criticizes the Marcott reconstruction here, saying (in part):
Point (1) should be commonplace, although due to a common misinterpretation of Marcott et al, it is not. The misinterpretation is that the bold line in the center of the 1 sigma uncertainty interval in Marcott et al is the "reconstruction", whereas it is just the mean of an ensemble of reconstructions. Marcott et al state no preference among that ensemble as to which is the "real" value, so therefore all members of the ensemble must be given equal weight. All, or nearly all members of the ensemble frequently lie higher than the ensemble mean, and often much higher. It follows that temperatures greater than even 2015 are far more likely than not to have occured in the Holocene Climactic Optimum, and indeed may have occurred close to 1 in every 20 years over the period 5500 to 9500 BP.
Point (2), on the other hand is at best ill informed nitpicking. Ill informed because it assumes the differential rate of heating between land and ocean which leads to cooler oceans in a warming climate (and warmer oceans in a cooler climate) will be a significant factor over multi-centenial periods. For the Marcott mean value, the time resolution is 300 years, at which resolution the difference between land and ocean would be effectively nil. (This contrasts with the case for the much smaller resolution of reconstructions of the MWP.)
Nitpicking because while sea temperature proxies are 80% all proxies, the sea surface is 70% of global surface area. That is, sea surface temperatures are only overrepresented by 14.3%. This evidently creates a problem for somebody prepared to overlook that sea surface temperatures were underrepresented by 45.7% in the Lundqvist analysis that he prefers. Further, nitpicking because again in Marcott et al, NH proxies are again overrepresented (as is typically the case in reconstructions). Marcott el al write:
Despite Marcott's phrasing, the SH extratropics is well represented, with 15.1% of proxies for 18.3% of the surface area (17.5% under representation), but the NH extratropics has 39.7% of the proxies for 18.5% of the area (a massive 114.6% over representiation). Meanwhile tropical regions, with 45.2% of proxies for 63.7% of the area, are also under represented (29.1% under representation).
As can be seen below, NH temperatures rose much higher in the Holocene Climatic Optimum relative to tropical or SH temperatures:
As it turns out, the most under represented area relative to the proportion of the Earth's surface shows lower temperatures over the Holocene Climactic Optimum, while the area with the greatest early holocene warmth is massively over represented.
These biases are far more likely to impact the final reconstruction than are the relatively minor land/sea bias pointed to by Angusmac. Further, they bias in the opposite direction to his presumption regarding the land/sea bias, and so would surely cancel the effect (at least).
Finally, as it happens Marcott et al checked for the effect of biases against a simple model:
Clearly their methodology was sufficiently robust to eliminate all but a small part of the warm bias we would expect from looking at proxie locations alone.
I am currently short of time, and will discuss point three and further criticisms later.
Continuing my discussion of Angusmac's comment here:
Point (3) clearly misunderstands the nature of the Marcott reconstructions. (The misunderstanding is quite common among people who discuss climate science on the internet.) Specifically, while the mean of the stack of reconstructions has a resolution of approximately 300 years, the reconstructions themselves (as shown in figure S3, see Comment 101 above) have a resolution of 20 years. They are therefore quite appropriately compared to decadal temperatures. Further, in generating Figure 3 (see comment 101 above), which is the crux of Marcott et al, Marcott et al added noise to the reconstructions so that their variability matched the annually resolved Mann 2008 reconstruction. The statistical distribution is, therefore, quite appropriately compared to annual temperatures.
Given this, it is reasonable to criticize graphs that show only the stack mean vs modern temperatures. The falsely give the impression that we are experiencing temperatures never experienced before since the invention of agriculture. Rather, modern twenty year averages should be compared to the individual reconstructions in the stack, like this:
As you can see, even the 1996-2015 average was probably experience many times (if briefly) in the Holocene, but we are pushing towards the upper temperature limit of the Holocene. More significantly, the rate of increase of temperature over the last century is certainly in the highest 5% of holocene temperature trends, and may well be unprecedented. It is also ongoing.
More later.
This is my final response on this page to angusmac's comment here. In that comment, in addition to the three points addressed above he draws attention to the fact that the 19th and 20th century spike in the Marcott standard mean of reconstructions is not robust (something pointed out by Marcott et al in the original paper). He reinforces the argument by comparing the final stages of Figure S3 to the equivalent Figure C8 from Marcott's thesis:
He writes:
In the elided section, he adds the further, hypocritical criticism that Marcott's response to blog criticisms on a blog rather than formally through an corrigendum or explanandum published in the journal "...casts doubt on the robustness of the last 500 years of the reconstruction and perhaps even the entire paper". The double standard in not assuming that the failure of critics to formally publish their criticisms "casts doubt on the robustness" of those criticisms is breath taking. I will treat that "argument" with the contempt it disserves, and give it no further dicussion.
With regard to robustness, Tamino explained it first and best. It is entirely an artifact of the method of taking a simple average combined with the drop out of proxies towards the end of the record. Using other methods such as the method of difference (see Tamino's post), or the RegEM method used as an alternative method in Marcott et al, results in a much smaller, but robuts uptick occuring in the 20th century only:
Comparison of simple averages to the method of difference (or the RegEm method) shows the problem has little effect anywhere else in the reconstruction, and a scarcely discernible effect in the crucial years of the Holocene Climactic Optimum:
Given that angusmac is familiar with Marcott's Q&A at Realclimate, and even with Marcott's thesis, he must be familiar with this explanation of the problem, and ergo that it makes no relevant difference to the reconstruction in the HCO. Given that, I am surprised (to say the least) that he should raise this issue.
This, of course, does not explain the difference in the terminal section of the stacks generated for the paper (S3) and the thesis (C8). What angusmac does not show is the difference between Figure C8 and Figure C9 of the thesis:
As can be seen, calibration issues in just one proxy were able to make a significant difference to the robustness of the reconstructions through the HCO. Those issues were resolved in the paper, and the proxy consequently is used in the paper without problems. While that proxy (ODP 984) terminates around 1400 AD, and therefore does not contribute to the lack of robustness of the terminal period, it is likely that similar improvements in calibration and/or terminal dates explains the difference between C9 of the thesis and S3 of the paper with regard to the final two centuries.
Comparison of C9 and S3 shows the problem to only relate to the final two centuries which are not the point of the reconstruction, and which are not used to calibrate reconstruction temperatures to modern temperatures (which is done indirectly via the Mann 2008 reconstruction of temperatures over the last 2000 years). From this it follows that there is no basis in this data to doubt the HCO reconstruction from Marcott et al.
Nor is any substantive reason advanced to show the changes in data handling with regard to calibration and possibly terminal dates between the two does not represent an improvement. As it makes no difference to the substance of the reconstruction that is sufficient answer IMO. If it does not satisfy angusmac, he can do the necessary leg work by enquiring of Marcott re all the precise differences between thesis and paper among which the full explanation must be found.