Glaciology · dating · an instrument on real data

Most of the Time Is at the Bottom

The EPICA Dome C ice core is dated to 807,900 years at the bottom of its record, 3,192.75 metres down. The years are not spread evenly. The top half of the core reaches back only 121,300 years, and the older half of all the time is held in the last 468 metres. Move a cut along the published chronology, watch eight ice ages spread out when the same measurements are laid out by years instead of metres, and see the diffusion arithmetic that makes the deep years impossible to count.

In December 2004 a drill at Dome C, a summit of the East Antarctic plateau where very little snow falls, stopped at 3,259.72 metres, about fifteen metres above the rock, because the drillers did not want to reach the meltwater they expected there. The core's standard chronology, AICC2012, dates it down to 3,192.75 m, where the ice is 807,900 years old. That is the famous number. The less famous one is how unevenly those years are spread along the core.

Cut the dated core in half by length and the top half reaches back only 121,300 years: 15% of the time. Cut it in half by time and the cut falls at 2,724 m, so the older half of the whole record, four hundred thousand years, is held in the last 468 metres, 15% of the length. The top hundred metres hold 2,553 years; the bottom hundred hold 110,200. The oldest tenth of the time is in the last 65 metres. Every one of those figures is computed in your browser from the published chronology, and you can move the cut yourself.

Why the years get thinner

Snow that falls on Dome C is buried by the snow of later years, and the ice sheet beneath it is not still: it spreads, slowly, outward under its own weight. A layer that is stretched sideways must get thinner, and the deeper a layer has sunk, the longer it has been stretched. AICC2012 carries this as two columns beside every depth: the accumulation rate when that ice fell as snow (between 1.1 and 4.2 centimetres of ice a year, lowest in the cold glacial periods) and a thinning factor from a flow model. Their product is the thickness of one year. At the top it is 29 mm of ice; at the bottom of the dated core it is 0.56 mm, and it is thinnest, 0.40 mm, at 3,039 m, in a glacial period when little snow fell.

The simplest model of this is John Nye's, from 1963: assume no melting at the bed and a vertical squeeze that is the same all the way down, and a layer's thickness is proportional to its height above the bed. Dansgaard and Johnsen, setting out Nye's model in 1969, wrote down the age it implies. If an ice sheet of thickness H gets snow at a steady rate a, the age at depth z is (H/a) ln(H/(H−z)). The logarithm is the whole story of this page in one symbol: it grows without limit as z approaches the bed. Taken literally, with the ice about 3,275 m thick (the drill stopped about 15 m short of 3,275) and today's snowfall, the formula puts the ice at 3,192.75 m at 414,600 years; with the average snowfall over the whole record, 636,300. (It treats the depth as if it were all solid ice; the loose snow near the top is a small correction here.) The shape is right and the number is not, which is the reason real chronologies are built the hard way.

What a millimetre can remember

Thin years would still be years, if you could see them. At Dome C you cannot, and the reason is that the ice does not keep a sharp edge. The climate signal in the water itself, the ratio of heavy to light isotopes, is smeared out while the snow is still porous firn: "the amplitude of the original isotopic signal is smoothed by diffusion in the firn column, mainly through the vapour phase," as Simonsen and colleagues put it. The smearing is a Gaussian blur, and its width, the diffusion length, has been estimated for Dome C at about 7 cm of firn by the time the pores close (7.23 cm for deuterium: Gkinis and colleagues' 2021 conversion of an estimate by Holme and colleagues, 2018, who gave it as ice). As solid ice that is about 6.3 cm. A year of snow at Dome C is about three centimetres of ice. The blur is about twice as wide as the thing it would have to preserve.

Deeper down it gets worse. Ice that spends a long time warm keeps diffusing: Pol and colleagues suggest water veins between large crystals, in ice that has spent more than 200,000 years above −10 °C. And the estimates for the ice of Marine Isotope Stage 19, between 3,147 and 3,190 m, disagree with one another: at least 40 cm (Pol and colleagues, 2010), 31 ± 5 cm (Shaw and colleagues, 2024), 16 to 22 cm from models of the physics. All of them are measured in metres of ice at that depth, where one year is less than a millimetre.

The presets are the published estimates; the slider is there because they disagree. Try the first: at a hundred metres the one-year bar is not small, it is gone, and a ten-year cycle keeps about a third of its swing. This is the arithmetic behind a sentence in the AICC2012 paper: "Layer counting is not possible for deep Antarctic ice cores recovered in low accumulation areas." Pol and colleagues put the practical floor in central Antarctica at about twenty years. At the bottom, with a diffusion length of 31 cm, even a thousand-year cycle keeps less than a tenth of its swing.

So how is a year at the bottom dated?

Not by counting. Greenland can count: its snowfall is several times heavier, and the GICC05 chronology counts annual layers back 60,000 years, using the water isotopes only for the last 7,900 (beyond that, "isotopic diffusion obliterates the annual cycle", Andersen and colleagues wrote) and then electrical conductivity, chemistry measured in a continuous flow, and the visible bands in the ice. Its accumulated counting error at 60,000 years is 2,601 years, stated as two standard deviations. The North GRIP core itself, on AICC2012, ends at 3,082 m and 119,800 years, where one year is 14 mm of ice.

Dome C is dated the way you would date a clock you cannot read directly. A model of snowfall and ice flow gives a first guess at the age of every depth (those are the two columns this page multiplies). Then the guess is pinned wherever the ice carries a date of its own: a volcanic horizon, the peak in beryllium-10 from the Laschamp event, the Brunhes–Matuyama reversal of Earth's field (found between 3,161 and 3,170 m and dated independently to 776,000 ± 12,000 years), and, below the last interglacial, tie points from the oxygen in trapped air and the air content of the ice, which follow the slow cycles of Earth's orbit. AICC2012 combines all of this, for five cores at once, with a Bayesian tool called Datice. What it will not do is pretend: its age at the bottom of the dated core carries a standard deviation of 12,400 years, and the readout above prints the one for wherever you cut.

And the record stops before the ice does. Below about 3,200 m, Jouzel and colleagues wrote in 2007, "we have strong arguments that the core stratigraphy has been disturbed over its bottom 60 m"; a later study of that ice concluded that "a clear paleoclimatic signal can therefore not be inferred" from it. The bottom of the core is older still; it is just no longer in order.

The oldest ice

At its thinnest, 3,039 m, one metre of Dome C holds about 2,480 years. In January 2025 the Beyond EPICA project announced that it had drilled a 2,800 metre core to the bedrock at Little Dome C, and that from preliminary analyses "the uppermost 2,480 meters contain a climate record that goes back to 1.2 million years in a high-resolution record where up to 13,000 years are compressed into one meter of ice." That is the same pile-up, pushed further. The project's announced age is "1.2 million years, and probably beyond"; the 1.5 million years it aims for is a target, not yet a result.

Older ice exists, just not in order. In the Allan Hills of Antarctica, where wind strips the snow and old ice comes up to the surface, Yan and colleagues dated ice at 2.7 ± 0.3 million years in 2019, and Shackleton and colleagues reported a sample of 6.0 ± 0.7 million years in 2025, close to the bedrock. Those ages come from the argon in the trapped air, not from counting or from a flow model, and the authors call the ice stratigraphically discontinuous: snapshots, not a record.

curl -sO https://artwaste.land/checks/verify-how-old-is-the-bottom-of-an-ice-core.mjs
node verify-how-old-is-the-bottom-of-an-ice-core.mjs --mutate

Sources

1. Bazin, L. et al. (2013). An optimized multi-proxy, multi-site Antarctic ice and gas orbital chronology (AICC2012): 120–800 ka. Climate of the Past 9, 1715–1731. doi:10.5194/cp-9-1715-2013. Data: PANGAEA collection 824894 (CC-BY-3.0): EDC 824865, NGRIP 824867, EDC δD 824891.
2. Veres, D. et al. (2013). The Antarctic ice core chronology (AICC2012): an optimized multi-parameter and multi-site dating approach for the last 120 thousand years. Climate of the Past 9, 1733–1748. doi:10.5194/cp-9-1733-2013.
3. Parrenin, F. et al. (2007). The EDC3 chronology for the EPICA Dome C ice core. Climate of the Past 3, 485–497 (the Brunhes–Matuyama depth and age, the tie points).
4. Jouzel, J. et al. (2007). Orbital and millennial Antarctic climate variability over the past 800,000 years. Science 317, 793–796. doi:10.1126/science.1141038.
5. Tison, J.-L. et al. (2015). Retrieving the paleoclimatic signal from the deeper part of the EPICA Dome C ice core. The Cryosphere 9, 1633–1648. doi:10.5194/tc-9-1633-2015 (final depth 3,259.72 m, December 2004).
6. EPICA community members (2004). Eight glacial cycles from an Antarctic ice core. Nature 429, 623–628 (site; the 3,309 ± 22 m thickness estimate).
7. Nye, J. F. (1963). Correction factor for accumulation measured by the thickness of the annual layers in an ice sheet. Journal of Glaciology 4(36), 785–788. Dansgaard, W. & Johnsen, S. J. (1969). A flow model and a time scale for the ice core from Camp Century, Greenland. Journal of Glaciology 8(53), 215–223.
8. Grisart, A. et al. (2022). Sub-millennial climate variability from high-resolution water isotopes in the EPICA Dome C ice core. Climate of the Past 18, 2289–2301. doi:10.5194/cp-18-2289-2022 (the attenuation formula, their Eq. 2, after Johnsen et al. 2000; σ_firn = 0.07 m for EDC).
9. Gkinis, V. et al. (2021). Numerical experiments on firn isotope diffusion with the Community Firn Model. Journal of Glaciology 67(263), 450–472. doi:10.1017/jog.2021.1 (Dome C close-off diffusion lengths from Holme et al. 2018).
10. Simonsen, S. B. et al. (2011). Climate of the Past 7, 1327–1335. doi:10.5194/cp-7-1327-2011 (diffusion mainly through the vapour phase).
11. Pol, K. et al. (2010). New MIS 19 EPICA Dome C high resolution deuterium data. Earth and Planetary Science Letters 298, 95–103. doi:10.1016/j.epsl.2010.07.030 (abstract read; at least ~40 cm). Pol, K. et al. (2011). Climate of the Past 7, 437–450. doi:10.5194/cp-7-437-2011 (~8 cm at MIS 11; the ~20 year floor).
12. Shaw, F. et al. (2024). The Cryosphere 18, 3685–3698. doi:10.5194/tc-18-3685-2024 (31 ± 5 cm for MIS 19, 3,147 to 3,190 m; the 16–22 cm model range).
13. Svensson, A. et al. (2008). A 60 000 year Greenland stratigraphic ice core chronology. Climate of the Past 4, 47–57; GICC05 data file gicc05-60ka-20yr.txt (maximum counting error 2,601 years at 60,000, 2σ). Andersen, K. K. et al. (2006). Quaternary Science Reviews 25, 3246–3257.
14. Beyond EPICA – Oldest Ice, press release of 9 January 2025, as published by the British Antarctic Survey. Chung, A. et al. (2025). The Cryosphere 19, 4125–4140 (the 1.5 million year aim).
15. Yan, Y. et al. (2019). Two-million-year-old snapshots of atmospheric gases from Antarctic ice. Nature 574, 663–666. doi:10.1038/s41586-019-1692-3. Shackleton, S. et al. (2025). Miocene and Pliocene ice and air from the Allan Hills blue ice area, East Antarctica. PNAS 122(44), e2502681122. doi:10.1073/pnas.2502681122.