Skip to content


August 29, 2019

By Perry Spector


This is the first guest post on this blog, and the point is to describe a data-model comparison project called DMC:ANTARCTICA that I have been working on for the past year. The overall objective of the project is to provide synoptic evaluation of Antarctic ice sheet models that simulate the evolution of the ice sheet over long time periods using the entire dataset of cosmogenic-nuclide measurements from Antarctica. This project is still under development, but there now exists a publicly available DMC:ANTARCTICA web interface that serves data-model comparisons for every site in Antarctica where data exist and for 15 different ice-sheet model simulations.


We rely on numerical ice-sheet models for establishing the climate sensitivity of the Antarctic ice sheet and for forecasting its future behavior and contribution to sea level. Although these models provide glaciologically-feasible reconstructions ice-sheet evolution, they are of limited use unless they can be shown to agree with observations of past ice-sheet change.

Considerably progress actually has been made on this front. Recent simulations by David Pollard, Pippa Whitehouse, Rob Briggs, and others that are constrained by geologic observations (e.g. relative sea-level curves from Antarctica, exposure ages, etc.) do, in fact, agree with those observations remarkably well in many cases.

Existing data-constrained models, however, are primarily restricted to the period between the LGM and the present. Longer simulations, spanning important time periods when the climate was as warm or warmer than present, have not been thoroughly tested. One limitation to doing this is that many of the constraint datasets, such as RSL curves from Antarctica, do not extend beyond the LGM. On Pleistocene-Pliocene timescales, cosmogenic-nuclide measurements from Antarctica are some of the only data available to test hypotheses of past ice-sheet behavior.

The existing set of Antarctic cosmogenic-nuclide measurements

All known published (and many unpublished) cosmogenic nuclide data from Antarctica have been compiled, mostly by Greg Balco, in the ICE-D:ANTARCTICA database. The database currently comprises 4361 measurements on 2551 samples. For this to be a useful set of ice-sheet model constraints, it should span as much of the continent as possible and also be distributed in time. Let’s look at the breakdown. The figure below shows the distribution for samples of glacial erratics and for bedrock surfaces.

Minimum and maximum circle sizes represent 1 and 88 samples, respectively.

Both populations do a pretty good job highlighting where outcropping mountains exist in Antarctica. There are large areas, however, that are entirely unrepresented, such as the East Antarctic interior and the coast between northern Victoria Land and the Lambert-Amery Basin. Fortunately these are sectors where most ice-sheet models predict relatively small changes over glacial-interglacial cycles. For glacial periods, models predict large ice thickness changes in the Ross, Weddell, and Amundsen Sea sectors, areas where the coverage is relatively good.

Interestingly, around 80% of the samples are of glacial erratics while the remaining 20% are of bedrock surfaces. This distribution is probably, in part, due to the combination of two factors. First, outcropping bedrock surfaces in Antarctica that were ice-covered during the LGM commonly have exposure ages much older than the LGM because, in many cases, past ice cover was frozen to the bed and non-erosive. Second, the objective of many exposure-dating projects has specifically been to establish the ice sheet’s configuration during or following the LGM, and so these projects likely opted to sample glacial deposits rather than bedrock. However, as has been discussed by some authors (e.g. here) and as will be discussed below, well-preserved bedrock surfaces can provide important information about exposure and ice cover over multiple glacial-interglacial cycles.

The next set of maps show the distribution by nuclide.

Most samples in the database only have a single analysis of a single nuclide, and, unsurprisingly, that nuclide is usually Be-10. Multiple nuclides (e.g. Be-10 and Al-26) have been measured on some samples, as shown below. These samples are particularly important because paired nuclide data can provide information not just about past exposure, but also about past ice-cover and erosion.

The next figure shows histograms of exposure ages (note logarithmic axis) and gives a sense of the temporal distribution of the database.

Although a large fraction of the samples only provide information about the last ice age and the subsequent deglaciation (i.e. the focus of many exposure dating studies), a substantial population provides constraints on timescales extending as far back as the Miocene.

Therefore, to summarize, the set of Antarctic cosmogenic-nuclide measurements is reasonably well distributed in space and time, which suggests that it should be of some use for evaluating numerical ice-sheet models that run over a range of time periods. It would be useful if the database contained more samples from slowly-eroding bedrock surfaces, but the coverage isn’t terrible at least.

A Data-model comparison framework

For most samples, there are an infinite number of different histories of exposure, ice-cover, and erosion that could produce the observed nuclide concentrations. Therefore, these data cannot be simply inverted to generate unique ice-thickness chronologies. However, they can be used to distinguish incorrect chronologies predicted from ice-sheet models from plausible ones. The basic idea is that, for a given rock in Antarctica, the history of exposure and ice-cover simulated by a model predicts certain concentrations of cosmogenic nuclides at the present-day that can be directly compared to concentrations that have actually been measured in the rock.

This isn’t a new idea. In a 2012 paper, Sujoy Mukhopadhyay and others reported that Be-10 and Ne-21 concentrations measured in bedrock samples from the Ohio Range were consistent with concentrations predicted by a 5 million year ice-sheet model run. Since then, the only other example that I am aware of is a paper I recently wrote with John Stone and Brent Goehring, which is under review in The Cryosphere, that evaluates several models using results from the Pirrit Hills and the Whitmore Mountains.

However, rather than doing one-off data-model comparisons for one or two sites, the objective of the DMC:ANTARCTICA project is to perform synoptic evaluation of model performance using all available cosmogenic-nuclide measurements from Antarctica. At the moment, that capability is not yet available, but it is coming. What is available is the DMC:ANTARCTICA web interface which serves comparisons between every site in Antarctica where data exist and a number of published ice-sheet model simulations. The rest of this post will give a basic description of how the web interface works, and highlight a couple interesting results.

For a given site in Antarctica and ice-sheet model of interest, the modeled ice-thickness history at the site is extracted via bi-linear interpolation. Here is an example from the model of Pollard & DeConto (2009) for the Pirrit Hills.

The next steps are to (i) generate synthetic samples of both bedrock surfaces and glacial erratics spanning a range of altitudes at the site, (ii) determine the history of exposure and ice cover experienced by each sample, (iii) use those histories to calculate the present-day concentrations of various nuclides, and then (iv) compare those predictions to concentrations that have actually been measured in samples from the site.

For each DMC request, the calculations are performed in real time on a Google Cloud virtual machine. Concentrations, as well as exposure ages, are calculated using the Matlab/Octave code behind the version 3 online exposure age calculator, which was described by Balco et al. (2008) and has since been updated. Additional details about data-model comparison calculations are available here.

All samples are assumed to have zero cosmogenic nuclides at the beginning of the model simulation. Thus running the web interface with a site where there are 12 million year Ne-21 exposure ages, paired with a LGM-to-present ice-sheet model, will obviously produce large data-model misfits that are unrelated to the actual performance of the model. Similar misfits can be obtained by doing the reverse, that is, pairing a long-term model with a site that only has, say, Holocene exposure ages. This latter scenario can be dealt with by employing the option to change the start time of the model so you can look at, for example, only the last 20 kyr of a 800 kyr run.

The web interface can also serve model predictions for any latitude and longitude in Antarctica, even if no samples have been collected at that location. This feature could potentially be useful for planning future field projects.

Some interesting DMC results: (i) elevation transects of bedrock surfaces and glacial erratics

DMC simulates periodic deposition of erratics at a site whenever the ice sheet there is thinning. This is because most glacial erratics in Antarctica are ablation deposits, meaning that they were transported englacially from upstream erosion areas to ablation areas on the ice surface (where they were initially exposed to the cosmic-ray flux), and subsequently deposited on mountain flanks by ice thinning. Periodic deposition of erratics is simulated for all sites, regardless of whether erratics are actually known to be present.

Ice-sheet models that simulate many glacial-interglacial cycles can therefore produce erratics at similar elevations but at different times. The effect of this is shown in the upper-right panel below for the 5 Myr simulation of Pollard & DeConto (2009) at Mt. Turcotte in the Pirrit Hills (see ice-thickness time series above).

Different colors represent different nuclide-mineral pairs as follows: orange = C-14 (quartz); red = Be-10 (quartz); blue = Al-26 (quartz); neon = Ne-21 (quartz); gray = He-3 (pyroxene/olivine).

Concentrations for a given nuclide in erratics are scattered and have a relatively weak relationship with elevation. The exception is for C-14 (shown in orange), which decays rapidly with a 5730 year half life and therefore has no memory of the exposure and ice-cover history prior to around 30 kyr ago. In contrast, the bedrock surfaces (upper-left panel), which are predicted to have been intermittently ice covered, show concentrations that always increase monotonically with altitude. In part, this is due to the fact that production rates increase with elevation, but high elevation samples also must have spent a lesser proportion of their existence ice-covered than low elevation samples from the same site.

Note 1: Both bedrock and erratics are assumed to not experience erosion nor get covered by till or snow. For erratics, once they are deposited they are assumed to not change elevation, roll over, etc.

Note 2: The four panels above have different x and y axis ranges. Sorry. Working on that.

The predictions for Mt. Turcotte bedrock and erratics (upper panels) are interesting because they are qualitatively pretty similar to what is actually observed there (lower panels), as well as at many other Antarctic sites. Observed Be-10, Al-26, and Ne-21 in bedrock samples monotonically increase with elevation. There are only minor exceptions to this, which are probably related to small amounts of erosion experienced by some samples or, in the case of Ne-21 (shown in neon), radiogenic production. In contrast to the bedrock samples, Be-10 measurements on glacial erratics (lower-right panel) are scattered and have no obvious concentration-elevation relationship. Thus, at least qualitatively, the model predictions for both bedrock and erratics are consistent with what we actually observe.

One take-home point from this comparison is that, in Antarctica, where past ice cover was commonly cold-based and non-erosive, weathered bedrock surfaces can, in some cases, be much better indicators of long-term exposure and ice cover than glacial erratics. Next time you’re in the field, collect an elevation transect of weathered bedrock surfaces.

Some interesting DMC results: (ii) LGM-to-present ice-sheet models consistently simulate premature deglaciation in the western Weddell Sea

Okay, now let’s look at an example in which we actually evaluate the performance of ice-sheet models.

Our knowledge of the post-LGM deglaciation history of the two largest embayments in Antarctica, the Ross and the Weddell, has remained very lopsided. Decades of geological (marine and terrestrial) and glaciological research in and around the Ross Embayment has generated a large network of constraints on the thickness and extent of grounded ice during the LGM and during the subsequent retreat. In contrast, relatively few field studies in the Weddell Embayment have produced commensurately few constraints there.

Recent papers by Keir Nichols, Jo Johnson, Brent Goehring, and others (published and in review here and here) are increasing our knowledge of the deglacial history of the Weddell Sea. These papers describe C-14 exposure ages from an elevation transect of 8 bedrock samples from the Lassiter Coast on the east side of the Peninsula. All samples have mid-Holocene ages and indicate abrupt thinning there at this time.

Let’s see how these results compare to predictions from state-of-the-art ice sheet models. In the figures below, orange circles show measured C-14 exposure ages plotted against sample elevation. Gray circles show model predictions. Note that the 8 bedrock samples on which C-14 concentrations were measured were collected from a few different but nearby peaks, which have been classified as different sites in ICE-D:ANTARCTICA. Thus, only 5 of the 8 samples are shown below, which does not affect the data-model comparison.

The first figure shows the predictions from the reference simulation of the model by Jonny Kingslake and others (2018).

The second figure shows predictions from the best-scoring run of the large ensemble by David Pollard and others (2016). Other recent models by David Pollard and others predict relatively similar transects.

The third figure shows predictions from the last 20 kyr of the model by Tigchelaar et al. (2018).

These three state-of-the-art ice-sheet models consistently simulate premature thinning on the Lassiter Coast by several millennia, which likely means that the grounding line in the western Weddell Sea retreats too early in the models. Note that the observation of premature thinning is not limited to this site but is also seen at the Heritage Range and at the Pirrit Hills, sites which are presumably sensitive to ice-sheet behavior in the western Weddell Sea.

The model that performs best at this site appears to be that of Pollard et al. (2016), which is perhaps not surprising because, unlike the models by Kingslake et al. (2018) and Tigchelaar et al. (2018), it was calibrated using a large network of observational constrains from various other sites in Antarctica.


The overall goal of the DMC:ANTARCTICA project is to systematically evaluate ice-sheet model simulations using the entire set of cosmogenic-nuclide measurements. This isn’t available yet, but it’s coming.

The DMC:ANTARCTICA web interface is publicly available for dynamically comparing observations and model predictions anywhere in Antarctica. Check it out.

The web interface helps address the issue of ice-sheet model opacity. Typically the complex behavior of an ice-sheet model is conveyed only in a few small figures in a paper. Now it’s possible for anyone to explore models and see what they predict anywhere in the continent.

The web interface still has some issues to be ironed out. Some of these are discussed here. Let me know if you find bugs or if there is a specific feature you’d like to see implemented.

Don’t see your data? The problem is probably either that (i) it’s not in ICE-D:ANTARCTICA yet, or (ii) certain information needed to calculate exposure ages (e.g. sample thickness) is missing from ICE-D. Contact me or Greg to fix that.

Have a model you would like included in this project? Get in touch.

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: