Guest Post by Willis Eschenbach
There’s a new study entitled “Global warming preceded by increasing carbon dioxide concentrations during the last deglaciation”, Shakun et al. (paywalled, hereinafter Shakun2012). The paper claims to show that in the warming since the last ice age, CO2 leads temperature. Anthony wrote about it in his post “A new paper in Nature suggests CO2 leads temperature, but has some serious problems“. The press release says (emphasis mine):
A new study, funded by the National Science Foundation and published in the journal Nature, identifies this relationship and provides compelling evidence that rising CO2 caused much of the global warming.
Lead author Jeremy Shakun, who conducted much of the research as a doctoral student at Oregon State University, said the key to understanding the role of CO2 is to reconstruct globally averaged temperature changes during the end of the last Ice Age, which contrasts with previous efforts that only compared local temperatures in Antarctica to carbon dioxide levels.
“Carbon dioxide has been suspected as an important factor in ending the last Ice Age, but its exact role has always been unclear because rising temperatures reflected in Antarctic ice cores came before rising levels of CO2,” said Shakun, who is a National Oceanic and Atmospheric Administration (NOAA) Post-doctoral Fellow at Harvard University and Columbia University.
“But if you reconstruct temperatures on a global scale – and not just examine Antarctic temperatures – it becomes apparent that the CO2 change slightly preceded much of the global warming, and this means the global greenhouse effect had an important role in driving up global temperatures and bringing the planet out of the last Ice Age,” Shakun added.
The good news about the paper is that they have provided the temperature records (Excel spreadsheet) for the 80 proxies used in the study. My compliments to them.
Me being a suspicious fellow, however, I figured “trust but verify”, so I plotted up the temperature records that they used. I always begin with the original data, without any additions or distractions. Figure 1 shows the data that they used.
Figure 1. Records and types of proxies used in the Shakun2012 study
As you can see, some of the ice core records are down where we’d expect them to be, well below zero. Those are the GRIP and NGRIP records from Greenland. But there are some oddities about these proxies.
One problem that is immediately obvious is the timing. The peaks for the previous interglacial period (the Eemian, about 130,000 BC) don’t line up. That may not be much of a problem, though, because the paper is about the warming from the most recent ice age.
One oddity is that there are ice core records that are right around freezing (0°C). In addition, there are pollen records around freezing as well. This shows that we actually have a mix of anomaly records and actual temperature records. This is not a problem, just an oddity.
Next, let’s take a look at the location of the proxies. Figure 2 is from their paper:
Figure 2. Location of the proxies used in the Shakun2012 study.
This looks good, it looks like there may be passable coverage. So let’s look at the last glacial transition, we’ll look at the time since 26,000 BC.
Figure 3. Same data as in Figure 1, but showing the warming from the last ice age.
Here, you can see the Antarctic ice core records (yellow and green lines near 0°C) mentioned above that are shown as variations, with the modern value taken to be 0°C.
Some other observations. Greenland (yellow temperatures at bottom) seems to be an outlier in terms of change in temperature. The Antarctic ice cores and all of the rest of the records show much less warming since the ice age.
In order to compare these eighty proxies to each other, what we need to do is to “standardize” them. This means to first subtract the mean (average) of each proxy from the individual values. Then each of the individual values is divided by the standard deviation of the entire record for that proxy. The result will vary between about -3 and 3. Standardizing preserves the shape and timing of the data, it just makes all the proxies have a mean of 0 and a standard deviation of 1.
Next comes the part that the authors of these multi-proxy studies seem to have generally ignored. This is to look at each and every one of these proxy records and think about what they seem to mean. I’ll look at them sixteen at a time. Figure 3 shows the first sixteen of the Shakun2012 proxies.
Figure 4. Proxies from the Shakur2012 study. All of these cover the period from 26,000 BC to 1980 AD. Vertical dashed lines show the minimum (light blue) and maximum (dark red) values for the each proxy. Minimum and maximum times rounded to nearest 100 years. Colors as shown in Figure 1. Click for larger version.
NOTES BY NUMBER
1, 2: These are the Greenland ice cores. They show a warming of 32 and 27 degrees respectively, which is much more than any other proxy. Warming begins earlier than 20,000 BC.
4: The warmest date is at 1200 AD.
6: Warmest date is 1000 AD. Warming doesn’t start until 12,600 BC.
9: Maximum warmth is at 14,600 BC.
15: Very unusual shape, 11° warming.
Figure 5. Same as Figure 4, proxies from the Shakur2012 study. All of these cover the period from 26,000 BC to 1980 AD. Vertical dashed lines show the minimum (light blue) and maximum (dark red) values for the each proxy. Minimum and maximum times rounded to nearest 100 years. Click for larger version.
19: Warming doesn’t start until 10,800 BC
21: Maximum warmth precedes maximum cold.
28. Maximum doesn’t occur until 400 BC.
30. Maximum doesn’t occur until 1400 AD.
31. Maximum doesn’t occur until 2400 BC.
32. Maximum doesn’t occur until 1500 AD.
Figure 6. Same as Figure 4. Click for larger version.
34: Maximum at 1600 AD
35: Maximum at 14,000 BC
36: Strange shape, constant warming until the present.
42. Maximum not until 400 AD.
44: Warming until the end of the record in 8200 BC.
Figure 7. Same as Figure 4. Click for larger version.
50: Maximum not until 1100 AD.
51: Constant rise beginning to end.
52: Large drop and rise after maximum warmth.
53: Rises beginning to end.
54: Rises beginning to end.
58: Maximum not until 1300 AD.
59: Maximum not until 1600 AD.
60: Large rise in 1100-1200
Figure 8. Same as Figure 4. Click for larger version.
67: Warming starts at 25,900 BC.
68: Warming only one tenth of a degree
76: Warming occurs almost instantaneously
Discussion
The variety in the shapes of these graphs is quite surprising. Yes, they’re all vaguely alike … but that’s about all.
The main curiosity about these, other than the wide variety of amounts of warming, is the different timing of the warming. In some proxies it starts in 25,000 BC, in others it starts in 15,000 BC. Sometimes the warming peaks as early as 14,000 BC, and sometimes around 5,000 BC or later. Sometimes the warming continues right up to the present.
The problem becomes evident when we plot all of these 80 standardized proxies together. Figure 9 shows all of the standardized temperature traces.
Figure 9. All 80 temperature proxies from Shakun2012. Colors as shown in Figure 1.
Now, there’s plenty of things of interest in there. It’s clear that there is warming since the last ice age. The median value for the warming is 4.3°C, although the range is quite wide.
But if you want to make the claim that CO2 precedes the warming?
I fear that this set of proxies is perfectly useless for that. How on earth could you claim anything about the timing of the warming from this group of proxies? It’s all over the map.
Final Conclusion
The reviewers should have taken the time to plot the proxies … but then, the authors should have taken the time to plot the proxies.
w.
[UPDATE] A hat tip to Jostein, who pointed in the comments to the Shakun Nature paper being available here.
[UPDATE] Some folks wanted to see the CO2 data they used on the same timescale. Other folks said the colors in Figure 9 were misleading, since ice cores were printed on top, obscuring others below. We’re a full-service website, so here’s both in one:
Figure 10. All proxies, along with CO2 record used in Shakun2012.
My best to all,
w.
I decided to take a look at the various proxies by proxy type. There are ten different kinds of proxies.
Figure 11. Proxies averaged by type.
A few notes, in no particular order. The ice core records are similar, but the timing is different.
Foram assemblages seem to be useless. The same is true of the Tex86 proxies.
Pollen has a consistent signal, but the warming doesn’t start until about 10,000 BC.
MBT/CBT perfectly exemplifies the problems with this approach. Which one are we supposed to believe? Which one is it that is lagging the CO2?
Finally, the Mg/Ca and the UK’37 proxies kinda sorta have the same shape, but no uniformity at all regarding the timing of the rise.
Let me close with a black-and-white version of the above chart. This allows you to see where the denser areas are located.
Figure 12. Proxies by type. Blue line shows CO2 data as used in the study.
Note the difference in the underlying shapes of the different types of proxies, and the differences in their timing with respect to the rise of CO2.
Next, note that the CO2 record they are using is from Antarctica. That is the reason for the good fit with the single “ice core ∂18O and dD” proxy (left graph, second row) and the “ice core dD” (center graph, second row). Both of those are Antarctic records as well.
Also, as you can see, even within each proxy type there is no unanimity regarding the timing of either the onset or the end of the warming from the last ice age.
CO2 is the blue line … so was the warming before or after the blue line?
w.
[UPDATE]—The discussion continues at Shakun Redux: Master tricksed us! I told you he was tricksy!
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Jeff Alberts says:
April 8, 2012 at 5:54 pm
“major9985 says:
April 7, 2012 at 4:37 am
” Is it not obvious Willis has taken all the temperature records and not shown them as a global temperature? Even if at different times in different areas of the planet temperatures did not match up, you simply take an average over all.. ”
And you end up with something completely meaningless.”
There have been a gazillion papers written by Climate Scientists on the average global temperature. They certainly believe the average global temperature is an important and meaningful quantity, even if you do not. It is your opinion that is meaningless.
I’m confused about this post.
Shakun et al claim that the proxies differ: that the single antarctic proxy previously used is not representative of the global picture. You plotted the proxies all on one plot, and they end up all over the place, which is perfectly consistent with their claim.
What’s the issue?
Of course, Shakun et al then take this complete mess and summarize it intelligibly by doing an area-weighted average. That gives us a global temperature map to look at, from which more interesting conclusions can be drawn — for example, about the timing of the temperature increase.
Just because they want it to have meaning doesn’t mean it does. Look up “intensive variables”.
Shakun et al claim that the proxies differ: that the single antarctic proxy previously used is not representative of the global picture. You plotted the proxies all on one plot, and they end up all over the place, which is perfectly consistent with their claim.
The reason they can make the assertion that they do is that the SH warming (apparently) precedes the NH warming by a statistically resolvable time lag. As the South Pacific warmed and the Pacific in general followed — being the world’s largest ocean, covering roughly half of the Earth — the CO_2 level went up. It had to go up, because there is a huge amount of CO_2 dissolved in sea water and raising the temperature alters the equilibrium concentration, although we don’t really know what the temperature of the bulk of the ocean was at that time. When Willis plotted the temperatures by latitudinal band in the “Tricksie” post, a potential problem with both the original paper and Willis’ method is revealed (that I just noticed).
Obviously the warming observed at any given site is completely different. This is apparent in the non-renormalized data. However, the conclusion of Shakun is highly dependent on the SH warming first. The way Willis plotted it in Tricksie, it does — the polar region in particular appears to lead the CO_2 by 1000 years or more. The two localized proxies from the intermediate southern latitudes appear to perfectly bracket the CO_2, and indeed all of the equatorial data from both the NH and SH do as well. It is the NH data from the upper latitudes and the poles that appears to lag the CO_2 (although it then must dominate Shakun’s estimate of mean global temperature.
However, there is an implicit assumption built into this that is not really justifiable. The ranges of these curves are not really comparable — in particular the anomaly presented on the left is not a meaningful quantity. How, and in particular when, can one normalize it? If one shifts it up it makes warming precede CO_2. If one shifts it down, it works the other way. Since one has no absolute number from which it begins or to which it goes at the end, and since the tropical anomaly is not close to the polar anomaly in size, one is comparing apples to oranges to grapes to bananas, and ignoring all of the inland guavas and kumquats and kiwis and pears and…
This makes everything arguable — in both directions. Looking at the NH polar data in Tricksie, it looks like the NH polar temperatures bumped some first, before either CO_2 or southern polar temperatures. But then it they stubbornly refused to bump and actually went down until there was a sudden surge right before the YD, even as CO_2 was steadily increasing. Shift this curve down a bit and it looks like CO_2 leads NH polar temperatures (and since there is no absolute scale, who knows what “zero” really is)? Shift it up a bit, and CO_2 leads the temperature.
Willis, one thing you might want to do is plot not the curves themselves, but their slopes or even their second derivatives against CO_2 concentration. What one is looking for is force, or acceleration of temperature trends, inflection points where something changes. The “lag” in the climate behaves something like a “mass” in physics — changing the driving may not even change the slope of the temperature curve any more than pressing on the brake instantly makes your car move in reverse, but the causal effect of a brake is a negative acceleration that is instantly observable.
To put it yet another way, it isn’t the absolute magnitude of the temperature that matters — that is determined by past history in a complicated non-Markovian differential form. It isn’t the slope of the temperature that matters — that too appears to have an “inertia” that causes it to lag or lead driving forces by as much as 1000 years (for understandable reasons, e.g. thermalization times of the oceans). It is the rate that the slope changes — when one changes the “forcing” of the system, one really should fairly rapidly see the slope itself start to change in close correspondence with the important drivers.
This works both ways — it applies to the CO_2 curve as well.
In the data for 60N to 90N, the slope of the temperature goes from flat to increasing around 21 kya, when the CO_2 curve is flat (especially flat if one includes the rest of the CO_2 data omitted by Shakun) and boring. Around 17 kya it actually goes flat to negative — there was a significant thermal deceleration around 19 kya and it remains flat with no forcing correlation with the CO_2 right up to 14 kya. It then spikes up — in response to some enormous “force”, peaks, and then falls almost as rapidly with no possible way to interpret CO_2 as the active agency responsible for either one. Only then, around 12 kya, around the end of the Younger Dryas, does a strong positive slope emerge that appears almost perfectly correlated with the CO_2. However, the way it is plotted one can argue that the CO_2 is dragging up the temperature. It is only when one looks at the inflection points, not the non-normalizable curves themselves, that it is clear that this is not what is happening.
rgb
major9985 says:
April 7, 2012 at 4:37 am
Major, you and Jeff Alberts seem to think that any average is meaningful … here’s a couple examples showing why that’s not true.
You can go out and measure people’s income. If you measure the income of every man and woman in America, you get a meaningful number for the income of the average American.
But that doesn’t mean all averages are meaningful. You and Jeff seem to be the kind of guys who would do the measurements and come back to proudly announce that you have researched the question extensively, and that the average American has exactly one breast and one testicle … sure, you can take the average of anything, but the answer may not mean anything …
w.
rgbatduke says:
April 9, 2012 at 8:38 am
Take a look at the graph I presented of the different latitude zones, here, I’ll show it again:
As you point out, the issue is when the inflection point occurs, the turning upwards of temperature or CO2.
Starting from the South Pole, the region 60-90S starts to increase at the same time as CO2.
The next region, from 30-60S, turns upwards slightly before the CO2, so their hypothesis is already in trouble. It should turn later than the region further south.
The next three regions? I’d say they are too diffuse to even begin to say when they turned. However, all three appear to have turned at roughly the time the CO2 did, with one perhaps turning a bit before and another a bit after the CO2.
Then we have the northern data. It turns upwards slightly later than the others.
Net result? Well, if we take the southernmost region as our starting point, three of the other five regions don’t show us anything. Of the remaining two, one is going the right direction and one is going in the wrong direction …
As a result, I’d have to say that their “south to north” hypothesis, while not falsified, is certainly not supported by the data that they have presented. And the difficulties don’t stop there. Look at e.g. proxy 42, and see if you can tell me the timing of the transition …
w.
numerobis says:
April 8, 2012 at 9:33 pm
Area-weighting is completely inappropriate for this dataset, it is far too sparse. All it has done in their case is to hide the effect of the one completely unusual area represented by the two Greenland proxies.
w.
Do you have any explanation for why it’s invalid to do area-weighted averaging here? It certainly seems like a better approach than the prior approach, which was to take a single ice core as being globally representative.
I’m also failing to understand the analogy between area-weighted averaging of temperature proxies and summing exotic fruits. These aren’t smoothies we’re making, it’s converting various proxies into temperature anomalies and averaging those to get a global record of prehistorical temperatures. Proxies have generally bigger errors than historical instrumental records, and we’d prefer to have more of them than just 80, but that just means you have bigger error bars.
Willis, please re-read my posts. I’m arguing the exact opposite. Some averages are meaningful. A global average temperature is not.
Jeff Alberts says:
April 9, 2012 at 6:31 pm
Indeed, my apologies for the misunderstanding.
w.
numerobis says:
April 9, 2012 at 12:20 pm
Indeed I do, and it’s the subject of my new post here.
Thanks,
w.
“Indeed, my apologies for the misunderstanding.”
😉
Then we have the northern data. It turns upwards slightly later than the others.
Are we looking at the same figure? The northern data turns up well before the inflection point in CO_2, then goes down as CO_2 initially rises before turning around to catch back up. In fact, the northern exhibits a relatively weak warming trend some 3000 years before anything else, at a time the southern pole is if anything cooling and CO_2 is basically flat.
I agree that the major upward inflection appears to be delayed, but the data still leave that initial warming to be explained, as it occurred very early and was if anything correlated with flat to dropping CO_2. If anything, this suggests that the entire CO_2 rise was directly associated with warming oceans. Warming at the inland greenland sites was obviously not well-correlated with oceanic warming, and there isn’t a lot of northern polar open ocean, and what there is was very likely at a very uniform -4 to 4K temperature. In the south and central latitudes, however, when the warming started it was associated with warming the ocean, releasing trapped CO_2. This may well have provided some positive feedback to the warming, but it wasn’t a very strong one or the cold phase ice age wouldn’t be stable to the many southern warming events that happen across the 90,000 years out of every 100,000 that fail to trigger a reversion to warm phase globally.
Until one fully understands why the cold phase is cold and what destabilizes it so that warm fluctuations can grow (where before they did not) work like Shakun’s cannot reliably support any CO_2 linked hypothesis, only the observation that yeah, warming oceans release CO_2 so that the two things go hand in hand globally. Positive feedback should have predicted a continued rise in temperatures following CO_2 across the YD. Instead we have the opposite — a levelling of CO_2 in response to levelled or dropping temperatures.
rgb
Willis:
Thankyou for yet another of your superb critiques of a paper.
I write to support two comments you have made in the subsequent thread.
At April 6, 2012 at 6:01 pm you say;
“What I’m seeing from these proxies is that we don’t know as much as I thought we did about the shape and timing of the emergence from the last ice age. The other proxies tell a very different story from the ice cores …”
And at April 6, 2012 at 7:28 pm you say;
“Whatever ‘something’ these proxies are measuring, it’s clear that they are not measuring the same something. For example, the GRIP greenland ice cores show warming, starting 27,000 years ago. They warm slowly for about 10,000 years, then they warm rapidly to a peak about 10,000 years ago, and after that, they gradually cool down.”
Yes! Oh, yes!
I have repeatedly pointed out (in several places including WUWT) that
(1) ice core data are useful because they indicate CO2 concentration and isotope-derived temperature data from the same trapped gas bubbles
but
(2) ice core data are NOT a direct indication of anything because
(2a) different ice cores provide different indications
and
(2b) other proxies (e.g. stomata data) provide different indications to those of the ice cores and to each other.
The entire AGW edifice is built on dubious data. Therefore, much that is asserted as being “known” (e.g. temperatures and their changes, atmospheric CO2 concentrations and their changes, etc.) is very, very debateable. Please note that this applies to both proxy-derived data of the distant past and to measurement-derived data averaged to provide global and hemispheric information of the recent past.
With the possible exception of the satellite-derived data, all of the basic climate parameters are of unknown accuracy, reliability and precision. And this problem becomes obvious whenever different data sets for the same parameter are compared.
Richard
Hi Willis,
your work looks impressive, but as a non-expert I am wondering how you produced your graphs. One of the things about the hoaxers in the climate debate is the lack of transparency. So the more transparency you can provide the more credible your case will be. Can I ask you to explain what tools you have used to produce these graphs ? I presume you loaded the data from the excel sheet provided by Shakun into a tool and then ran some scripts. It would be great if you could document these steps and publish them. This is all the more important asthere are already allegations out there about manipulation. It would also allow your supporters to redo the plots and confirm or correct what you are saying.
These comments are made in the spirit of open, honest science, reading what people like you have to say without prejudice.
Wishing you all the best
Kevin Mannerings
Kevin Mannerings says:
April 17, 2012 at 1:29 am
Thanks, Kevin. I do my work in a combination of Excel and the computer language “R”. R is free and it’s wicked powerful. One of the best things about R is that at any time you can run one line or part of a larger program. This gives great freedom, and leads to execrable code, at least when I’m writing it.
I’ve zipped the CSV data files, along with the R file, and put them online here (127 kb zip file).
However, I must warn you that my code is very poorly commented, and in no particular order, because I just pick the lines that I want to run and run them.
All the best,
w.
Hi Willis,
Don’t worry about the comments in your scripts. As a tester I am used to that :-), I would like to have a look at it, as it might be useful for the tests I am doing on Piers Corbyn’s eathquake forecasts. Unfortunately, the link you gave is not working. You can mail me on kevinDOTmanneringsATgooglemailDOTcom.
Rgds, Kevin
Kevin Mannerings says:
April 17, 2012 at 11:07 pm (Edit)
Sorry, Kevin, bad link. I fixed it now, give it another try.
All the best,
w.