Does Marcott 2013 Show that Current Warming is Exceptional?

In a previous post, I considered whether the global temperature reconstruction continues to be valuable after 1) newer studies have been published with fewer seasonal biases and 2) criticisms that portions of the reconstruction are not robust. I concluded that the reconstruction continues to be valuable both because it remains a "conservative" look at temperature variability during the Holocene and because criticisms about the robustness of the reconstruction turn out to be entirely unfounded. What I'd like to do now is consider what the reconstruction tells us about natural climate variability during the Holocene and how this should cause us to evaluate current warming.

A casual look at the Marcott 2013[1] with the instrumental record is certainly alarming. Marcott's Holocene reconstruction shows relatively slow warming up until about 7000 years ago, then slow cooling into the the Little Ice Age. There's a "bump" in the graph associated with the Medieval Warm Period, but in this graph, the MWP wasn't all that warm and the LIA contained the "coldest" temperatures of the Holocene, though not anything like an actual "ice age."  At about the time human activity began significantly increasing GHG concentrations in the atmosphere, temperatures rapidly warmed - so rapidly that you can hardly detect the slope in the line. It's a dramatic change that has no precedent anywhere in the last 11,300 years. According to this reconstruction, current warming is more than just exceptional; it's completely unheard of. Global temperatures have increased from the coldest temperatures to beyond anything we've seen during the Holocene in about 150 yea

But is this accurate? After all, HadCRUT5 is a dataset built on the readings of real thermometers strategically placed all over the globe, whereas Marcott's is just proxies. And Marcott's reconstruction has been smoothed. On smaller time scales, there has certainly been more variability than this reconstruction shows. The question then becomes, how much variability in global temperatures is hidden by the resolution of the proxies and the smoothing used by Marcott? We need to consider the possibility that global temperatures are highly variable, but that variability has been smoothed out by the resolution of these ancient thermometers and Marcott's statistical methodology.

Smoothing

Let's deal first with the statistical methodology and Marott's application of smoothing. Much of this has already been dealt with pretty effectively in a Tamino post.[2] Any reconstruction like Marcott's will have both temporal uncertainties and temperature uncertainties. That is, a proxy record will have an error of some sort both in terms of it's temperature reading and the date of the proxy. Variability in Marcott's reconstruction, therefore, may not reflect the actual variability in global temperatures. It could be caused by the uncertainties in both temperature and time. So Marcott's reconstruction smoothed these out by averaging 1000 simulations with "perturbed" estimates of the ages of the proxies. The amount of perturbation was based on the uncertainty of the proxy. Doing this smooths out variability that is artificial - indicative of proxy uncertainties rather than climate variability. But it will also smooth out natural variability. That is, we can be pretty certain that actual climate variability is greater than what is shown in Marcott's graph (more on this later). 

Much of the contrarian confusion comes from this quote from Marcott's paper. "The results suggest that at longer periods, more variability is preserved, with essentially no variability preserved at periods shorter than 300 years." This is taken to mean, that any variability, regardless of the magnitude would be undetectable at periods shorter than 300 years. But that's not actually correct. What he's saying is that natural variability in the proxy data is small enough that it is not preserved at periods shorter than 300 years. If natural variability was large enough, there's no reason to think that Marcott's reconstruction wouldn't pick it up on smaller time scales. So how large would natural variability have to be for Marcott's reconstruction to preserve it at say, a time scale of 100 years? It turns out this is something you can test.

Tamino created an artificial temperature signal and put three warming spikes where global temperatures rose by 0.9 C in 100 years and then cooled by the same amount in the same amount of time. He then added these to Marcott's proxy data. This is what he came up with. Then he perturbed the ages 100 times and plotted the average. Each of the warming spikes are clearly detectable, even though smaller than the unperturbed signal.

Tamino's Comparison of Unperturbed and 100 Averaged Perturbed

He did it again using 1000 perturbed records, and the change in the magnitude of the spikes was largely unchanged. What this shows is that that Marcott's smoothing method is not sufficient to mask a 0.9 C/century warming spike. So Marcott's reconstruction does not preserve the natural variability that exists during the Holocene at time scales less than 300 years, but it would preserve a 1 C warming event that occurred in 100 years. If warming events like what we are experiencing now had happened during the Holocene, it would be picked up in Marcott's reconstruction.

Proxies

If warming events like what we're currently experiencing happened during the Holocene, Marcott's method wouldn't smooth it out, but what about the likelihood that Marcott's proxies are simply insufficient to pick up the warming signal? Even Tamino observed that when he introduced the above artificial spikes into the data, it didn't automatically raise global temperatures by 0.9 C, because some of the proxies didn't have any observations during the time frames of the introduced spikes. These proxies would be unaffected by Tamino's artificial signal, making the warming spike in the reconstruction smaller than the GMST signal Tamino created.

This is in part what confidence intervals are for. For most of Marcott's reconstruction, the 1σ CI appears to be about ± 0.2 C. Marcott's results are going to be off by less then 0.2 C about 68% of the time, but about 32% of the time, his results may be off by more than this. However, confidence intervals generally cannot account for systematic errors that introduce bias into the data. They account for random errors only. So biases in the data would cause errors that might push Marcott's reconstruction's confidence envelope outside what was actual global temperatures. The most obvious source of bias here would be coverage bias. Lack of sufficient coverage could introduce bias into Marcott's dataset. And we've already seen that proxy dropout did cause an artificial warming trend in the last 60 years of the Standard 5x5 reconstruction. But prior to this, during the time frame that matters, the unfilled Standard 5x5 reconstruction and the infilled RegEM reconstruction are virtually identical. This indicates that proxy dropout is not introducing coverage bias into the dataset before 1890.

We also saw in the last post that seasonal biases have affected Marcott's reconstruction, but this bias appears to be introducing too much variability in to Marcott's reconstruction, not removing it. Proxies from mid-latitudes tend to be summer proxies that are warmer than the annual mean, which makes temperatures warmer than they actually were. Correcting these seasonal biases actually diminishes the variability of global temperatures during the Holocene by making the HTM cooler, relative to the LIA. If anything Marcott's reconstruction contains more variability than actually occurred, so comparing the instrumental record to Marcott's reconstruction gives a more "conservative" evaluation of just how unique current warming is.

Plausibility

Underlying this entire discussion is, what kinds of climate variability can we plausibly expect to see by natural means? It's one thing to evaluate statistically to what extent Marcott's method or proxy data could miss a warming signal comparable to the last 100 years. But even if such a thing is statistically possible, that doesn't mean it's geologically plausible. Are there natural forcings that can cause 1 C warming to GMST in a century?

The causes for current warming are readily apparent. A geologist looking at earth millions of years from now would easily be able to detect the causes of current warming, even if humanity and all historical records were destroyed. He'd find evidence of the mining of fossil fuels, the concrete, the remains of power plants, etc. It would be no mystery to infer the increase in global temperatures from GHGs moderated by aerosols from the geologic evidence left behind from our activities. Other events in geologic history that have caused rapid warming or cooling are similarly detectable. The causes of warming during the eruptions of the Siberian Traps during the end-Permian extinction are readily detectable. The causes of the extinction event are far more complicated than rapid warming, but the warming itself is readily detectable in geologic evidence. The same can be said for the PETM. Rapid warming coming out of glacial cycles during the Pleistocene are also readily apparent. They can be seen in CO2 proxies as well as extent of glaciation, etc. None of these warming events averaged warming as rapid as the last 100 years, but all of them were far more rapid than anything detected in Marcott's reconstruction. The impact at the K-T boundary caused immediate cooling, and evidence for this can be seen in rock layers from the boundary all over the world as well. Rapid climate changes disrupt stability and are generally preserved in the geologic record.

Statistically, we've seen that Marcott's methodology could pick up rapid warming events, but it doesn't pick up any warming events comparable to any of the kinds of "rapid" warming events that have been detected elsewhere in geologic history. And none of the geologic evidence that would coincide with a rapid natural climate-changing event have been found either. We don't have evidence of a massive impact; we don't have evidence of massive GHG-spewing volcanic eruptions. We don't have evidence of a large increase in solar activity. Is it plausible that temperatures rose by 1 C in 100 years at various times throughout the Holocene and those climate changes have been completely undetectable by proxy or other geologic evidence? I think not. But perhaps it's the exception that proves the rule.

In 2015, David Kemp and his colleagues published a paper[3] considering the maximum rates of climate change, and they argued that geologists have systematically underestimated these maximum rates in the geologic record. Until this time, geologists have been unable to detect warming occurring at rates comparable to current warming, and in reality, they still haven't. But this paper argued that rates of climate change detected from proxies exhibit scaling that can alias natural variability. Even during some of the most rapid warming events detected in geologic history (like the end-Permian event and the PETM), the maximum rate of warming from natural variability may exceed what has been detected by geologists. Let's use Kemp's descriptions of these events as an example:

The Permian–Triassic boundary represents the most significant global warming of the Phanerozoic, when Earth’s largest mass extinction was accompanied by an increase in tropical sea surface temperature (SST) of ∼15 °C over an interval of nearly 1 Myr. Recent high precision U–Pb dating suggests that ∼10 °C of surface ocean warming occurred over a timespan of ∼60 kyr, that is, a rate of ∼1.7 × 10−4 °C per year. In a modern day context, this local rate is ∼42 times lower than the global surface ocean warming of 0.35 °C over the past 50 years. Similarly, through the PETM, a global surface ocean warming of ∼6 °C over 5–20 kyr has been linked to a pronounced perturbation of the global carbon cycle, and suggests a warming rate at least six times slower than modern.

These are the most significant warming events geologists have detected, and they warmed at average rates of ~42x slower and ~6x slower than current warming. In terms of average warming rates, this is what counts as extremely rapid geologically. But in terms of maximum warming rates over short periods of time, things may be different.

Taking into account timespan-dependent scaling, warming rates through intervals such as the Permian–Triassic boundary and the PETM likely exceeded current rates on decadal timescales, at least intermittently. Warming across the Permian–Triassic boundary stands out as the most significant temperature change of the past ∼0.5 billion years.

In other words, there are two events in geologic history where warming happened rapidly enough that, on decadal time scales undetectable by proxy evidence, warming occurred more rapidly than current warming. So it's true that proxy evidence can't preserve evidence of the maximum rates of climate change, and we can infer that warming on decadal time scales occurred more rapidly than proxies detect. However, the two candidates to potentially exceed current warming are found in exceedingly extreme events like the end-Permian extinction and PETM. The relatively stable climate of the Holocene through the LIA cannot be considered another plausible candidate - it's not plausible to think that natural warming at rates comparable to the last 100 years occurred 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] Tamino, "Smearing Climate Data," Open Mind. https://tamino.wordpress.com/2013/04/03/smearing-climate-data/

[3] Kemp, D., Eichenseer, K. & Kiessling, W. Maximum rates of climate change are systematically underestimated in the geological record. Nat Commun 6, 8890 (2015). https://doi.org/10.1038/ncomms9890