Skip to content

YD vs. ACR: Who Would Win?

February 8, 2019

Any readers with kids — of the six people in the world who read this blog, statistics suggest that at least one or two will have children — may be familiar with the “Who Would Win” series of books by prolific children’s book author Jerry Pallotta, including such titles as “Who Would Win: Komodo Dragon vs. King Cobra,” “Who Would Win: Jaguar vs. Skunk,” and many, many, many others. This posting is about an equally epic matchup that does not appear in that series, but is a good way to highlight yet another database project and some ideas about what we should be doing with it.

The database project is the ‘ICE-D:ALPINE‘ database, which is basically a copy of the now-reasonably-well-established ‘ICE-D:ANTARCTICA‘ database of all known cosmogenic-nuclide data from Antarctica. The difference, as indicated by the name, is that the ICE-D:ALPINE database is intended to contain all known cosmogenic-nuclide exposure-age data from alpine glacier systems worldwide (except for Antarctica of course). This is a much bigger project, because alpine glacial landforms in populated mountain ranges of the world are a lot easier to get to than glacial deposits in Antarctica — a substantial fraction of them are directly accessible by train from the Zurich airport — so the density of geologists and therefore exposure-dated boulders is correspondingly higher. Thus, the quantity of data is a lot bigger. If we count by measurements, the alpine database is about 4 times larger than the Antarctica database at the moment; if we count by published journal articles, alpine glaciers are winning by about a factor of 3. The alpine glacier database is also a much bigger project, and there are more people working to make this happen. The initial data content was primarily based on the global compilation of exposure ages on boulders put together by Jakob Heyman, and this has now been augmented by an international group of contributors whose names appear on the database front page. It’s also less complete than the Antarctica database. We guess that it will have to grow by 25-50% to include all published data, and of course the rate of publication of new exposure-age data for alpine glacial deposits worldwide is a lot higher. So right now it’s a pretty good, but not totally comprehensive, representation of most of the data that are out there. The present distribution of data looks like this, where each dot signifies a glacial landform, typically an alpine glacier moraine, that’s been exposure-dated at least once. There are 2163 dots.

The matchup concerns the two biggest climate events — or at least the ones with the largest media presence — from the period of global deglaciation following the Last Glacial Maximum. The Younger Dryas (“YD”) event is a 1200-year cold period between 12,900 and 11,700 years ago that is extremely prominent in climate records from Greenland and the North Atlantic region. The Antarctic Cold Reversal (“ACR”) is sort of like the opposite of the YD: it is a 1700-year cold period between 14,700 and 13,000 years ago that is prominent in Antarctic and Southern Hemisphere climate records. The idea is that during the ACR, it was warm in Europe and cold in Antarctica; during the YD it was the opposite.

Most paleoclimatologists and glacial geologists would probably expect that alpine glaciers worldwide paid attention to these climate events. Glaciers expand during cold periods, and contract during warm ones. An extraordinary amount of work in the glacial-geological and paleoclimate literature, which will not be summarized or enumerated here because this is a blog entry and not a peer-reviewed journal article, has been devoted to associating particular glacier advances in particular places with the YD or the ACR. Nearly all this work consists of single studies of single glacier systems, or a handful of glacier systems in the same general area. But if we were to do a global synthesis of all dated alpine glacier deposits, a decent initial hypothesis would be that in the Northern Hemisphere and particularly around the North Atlantic, it was cold during the YD and warm during the ACR, so we should see a lot of YD moraines and not a lot of ACR moraines. In the Southern Hemisphere, it was the opposite, so we should see a lot of ACR moraines and not a lot of YD moraines.

So let’s see if this is really true, using the entire data set of all dated moraines and not just a few individual hand-picked ones that happen to be part of our own Ph.D. dissertation. To do this, I’ll use what is generally thought to be more or less the present state of the art for production rate scaling and calibration, specifically (i) the time-dependent ‘LSDn’ scaling method developed by Nat Lifton and others in 2014-16, as implemented in version 3 of the online exposure age calculators described by Balco et al. (2008) and subsequently updated, and (ii) the CRONUS-Earth “primary” production rate calibration data sets for Be-10, Al-26, and He-3. The majority of the exposure ages in the database are derived from Be-10 measurements, but there are significant numbers of Al-26 and He-3 data in there too, and for purposes of this I’ll treat them all equally. Using these tools, calculate exposure ages for all the samples on all 2163 moraines in the database. For each landform, apply the outlier-detection and averaging scheme used in the online exposure age calculators — which basically attempts to discard the minimum number of outliers necessary to yield a single population — to generate a summary age and uncertainty for each moraine. Assume that these define a Gaussian uncertainty distribution and compute what fraction thereof lies within the YD (11700-12900 years BP) and ACR (14700-13000 BP). Basically, this is an estimate of the probability that a particular landform was emplaced during each of these climate events.

As an aside, a measure of how useful it is to have all these data in a modern online database is that this entire analysis takes slightly less than 5 minutes to run in MATLAB, even with a slow household network connection.

Now plot the results on a map.

Obviously, lots of moraines have ages that are nowhere near either the YD or the ACR; these have zero probability of belonging to either event and are denoted by black dots. But a decent number of moraines have a significant probability of belonging to one event or the other. Note that because the typical overall uncertainty in a moraine age is similar in magnitude to the length of these climate events, this algorithm will never yield a 100% probability; given this, a high probability of belonging to a millenial-scale climate event will be anything greater than approximately 50%. So the colored circles on both maps denote moraines that have more than half a chance of belonging to each climate event. Out of 2163 moraines, this scheme identifies 44 likely YD moraines (the green circles) and 156 likely ACR moraines (the blue circles).

What is interesting about these results is that they are not quite what we expect. It does seem evident that YD moraines are rare, and ACR moraines are common, across all latitudes in the Pacific Cordillera. I am not sure whether we expect this: the ACR is typically explained as something having to do mostly with Atlantic heat transport, and although ACR-age glacier advances in much of South America have been explained as a response to cooling in the Southern Ocean, there is significant disagreement about whether this should be true in the tropics, and it is definitely not obvious why it should also happen in the north Pacific. What is not what we expected, though, is that YD moraines are not dominant, or even very common at all, anywhere, even in Europe, where there are certainly lots of late-glacial moraines, but most of them seem to be ACR age, not YD-age. It is important to remember that these data only pertain to alpine glaciers, not ice sheets, so the general lack of evidence for YD glacier advances may be in agreement with the idea that the seasonality of YD climate change resulted in different responses from alpine glaciers and ice sheets. But the main point is that this analysis does not at all show that the YD and ACR each have their own geographic sphere of influence on glacier change. In fact, nothing like it. ACR wins and YD loses globally.

We also get this result if we just look at the derived ages of all 2163 moraines as a histogram.

The green and light blue boxes indicate the YD and ACR, respectively. Viewed in this way, it is actually evident that there are fewer moraines that belong to the YD than there are for the rest of the late glacial: there are a lot of moraines older than 13-ish ka and a lot of moraines younger than 12-ish ka, but not a whole lot in the middle. Weird.

So, what is going on here? One thing that is interesting about this analysis is that it has reached the level of complexity where it’s hard to know exactly what’s going on in detail. This is potentially bad because it’s hard to know what’s going on in detail. For example, although I think that the outlier-removal scheme mostly behaves rationally, if there are some pathological age distributions in here that caused it to do the wrong thing, I’ll never know it. On the other hand, it’s good because it makes the analysis unbiased: even if I have some preconception of what the results ought to be, it is so complicated that it would be really hard to even figure out how to consciously or unconsciously fudge the algorithm or the data to get the desired result. But, still, there are some reasons why these results could be systematically wrong. A couple are fairly obvious. One, summary moraine ages could skew young because of postdepositional disturbance (just because of how the outlier-rejection scheme works in this example, they are less likely to be skewed old due to inheritance). But this would be mitigated by the fact that production rate calibration data are largely from the same types of landforms, which would skew production rates to the young side and neutralize this effect. And, in any case, if this were happening it would push moraines out of the YD to the young ages and also push other moraines into it from the old side, so it would not be expected to produce a systematic lack of YD moraines, except by accident. Two, production rate estimates could be inaccurate. If we globally lowered production rates by 5%, that would push the big group of ages near 11-11.5 ka right into the end of the YD near 12 ka. But we would have to ignore a lot of presumably correct production rate calibration data to do that, so that is not very compelling either.

Many people would argue that this entire analysis is oversimplified, misleading, and probably wrong exactly because the data set is not highly curated and subject to detailed examination and quality control at the level of each individual landform. Many specialists in exposure-dating of glacial deposits, if you get them talking privately, will reveal that data collected by other research groups are sometimes not good. Scattered. Hopelessly contaminated by postdepositional disturbance. Not to be trusted. Garbage. Certainly not to be used in an analysis of this sort for fear of leading the results astray. Of course, this data set is so large that I have no idea who even collected most of the data, so even if I had some of these opinions myself it would be nearly impossible to act on them to influence the results. But we can make a bit of an effort to de-garbage the data set by applying some objective criteria. For example, we can only accept exposure-age distributions where we can’t reject the probability that the ages belong to a single population at the 95% confidence level, which basically has the effect of further screening out landforms whose ages are scattered more than can be dealt with by discarding a couple of outliers. The effect on the global distribution of moraine ages:

is to make the relative absence of moraines of YD age in relation to older and younger time periods even more striking. Applying this filter geographically produces this:

Again, being more strict about discarding scattered age distributions results in rejecting a whole lot of moraines that might have appeared YD-age just because of excess scatter, and makes ACR-age moraines (and “Preboreal,” i.e. post-YD, moraines) even more dominant. We are down to only a handful of YD moraines in the Alps and only one in the western US. What do we conclude? We conclude that YD-age alpine glacier moraines are actually kind of rare, which is at least not exactly what I expected, and possibly not what you did either. Perhaps the YD was cold, but also dry, so alpine glaciers shrank everywhere. Perhaps this is a manifestation of the seasonal distribution of YD cooling, something like Meredith Kelly suggested in this paper. Or perhaps Gerard Roe is right and glacier change is just an autoregressive process driven by stochastic variability in a steady climate.

But in any case, ACR wins! Game over.

So, to summarize, that was kind of fun, but what is the point? It could be argued that this analysis is oversimplified, misleading, based on crappy data, and largely BS. Maybe that’s why this is an unreviewed blog posting instead of a Nature article. But what is really important here is that this exercise highlights that we can move away from making large-scale inferences about glacier response to climate change based on small sets of highly curated data, and move towards doing it with much larger and more inclusive data sets, even if they are noisy, and, hopefully, with globally applied and less biased algorithms. We have, or almost have, a modern database infrastructure with dynamic, internally consistent, and mostly transparent, exposure-age recalculation, that we can use to do this. We should do it. And if, in fact, my shot at this really is oversimplified, misleading, and largely BS, then go do it better. All the data are here.



No comments yet

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: