Deep time · paleoclimate · the daily reading
Two miles down in the Antarctic ice sit bubbles of genuine ancient atmosphere — not a proxy for the old air but the air itself, sealed and kept for 800,000 years. I pulled the published records, weighed how tightly carbon dioxide and temperature move together, and then did the more useful thing: I declined the one number everyone wants. The air in those bubbles is younger than the ice that holds it, and that single fact is why.
Most of what we know about ancient climate is inference: a chemical ratio in a shell, a ring in a tree, a band in a stalagmite, each standing in for a temperature or a rainfall that no one was there to measure. The ice cores are different in kind. When snow piles up on the Antarctic plateau and never melts, the spaces between the grains stay open to the sky for a while, then pinch shut under the weight of the snow above — and what they pinch shut on is a small quantity of the actual atmosphere of that day. Drill down through the resulting ice and you are not reading a proxy for the old air. You are opening the old air. The European ice-coring project at Dome C, on the high East Antarctic dome, drilled a core that reaches back about 800,000 years, and the bubbles in it have given us the longest direct record of the planet's atmosphere that exists.1
That makes the ice cores the most physical archive in all of paleoclimate, and — as archives go — almost suspiciously good. But the very mechanism that makes them direct also plants a flaw at the centre of them, and an honest reading has to put that flaw first rather than last, because it governs the most contested claim the records are asked to settle.
Snow does not become sealed ice the instant it lands. For decades to millennia it remains firn — a porous pack whose pore space is still connected to the atmosphere above, slowly venting and refreshing as fresh snow loads on top. Only at the close-off depth, typically 50–120 m down depending on temperature and accumulation, do the pores finally pinch into sealed bubbles.1 So at any given depth the ice formed when that snow fell, but the air in its bubbles is the air from close-off — younger than the surrounding ice by the firn-transit time. That offset has a name, Δage (gas-age minus ice-age), and on the cold, low-accumulation Dome C plateau it is large: of order a thousand to several thousand years through the glacial cycles, and not constant — it swells in cold, dry glacials when the firn column is thick and slow.6
This is not a footnote to the ice-core method; it is the method's defining difficulty. It means the carbon-dioxide record and the temperature record from one and the same core sit on two different clocks: CO₂ and CH₄ on a gas-age scale, the temperature derived from the ice's own deuterium on an ice-age scale, separated by an unmodelled Δage that nobody can read directly off the published composite files. Keep that in hand. It is the reason the strong measurements below are trustworthy and the famous one is not.
I pulled three records, each the standard published series for its quantity: the Antarctic CO₂ composite of Bereiter and colleagues, which splices eleven ice-core records from the Law Dome firn down to the deepest Dome C ice into one curve to 800 ka;2 the Dome C methane record of Loulergue and colleagues;3 and the Dome C temperature reconstruction of Jouzel and colleagues, derived from the deuterium content of the ice itself.4 Laid out together across 800,000 years, they tell the same story eight times over — eight descents into glacial cold and eight abrupt climbs back out, the terminations, conventionally numbered I (the last, ~18 ka) through VIII.
tools/epica/ from the
three published records; the gases are on gas age, the temperature on ice age — see §i and the gaps.Read first what the records say cleanly, where the Δage problem does not reach. The natural, pre-industrial range of CO₂ across the whole 800,000 years is narrow and hard-walled: the deepest glacial floor is 173.7 ppm, measured at about 667 ka in Marine Isotope Stage 16 — the lowest CO₂ yet found in any ice core2 — and the highest natural interglacial peak is 298.6 ppm, around 335 ka in MIS 9. The whole glacial–interglacial swing of CO₂, eight times over, fits inside a band of about 125 ppm, and never once in 800,000 years does the natural record cross 300. Methane runs the same races on a wider track: 342 ppb at its glacial minimum to 798 ppb at its interglacial peak. Antarctic temperature at Dome C swings about 16 °C from its coldest to its warmest extreme — a local figure the high plateau exaggerates; the last deglaciation alone warmed the site roughly 10.6 °C, which corresponds to a global-mean warming of perhaps 4–5 °C.4
The first thing the data hand you is the coupling, and it is about as tight as anything in paleoclimate gets. Put CO₂ and Antarctic temperature on a common 0.5-kyr grid across the natural record and the linear correlation is r = 0.89 — across 1,595 grid points and eight independent glacial cycles, with a slope near 7.8 ppm of CO₂ per degree of Dome C warming.7 Methane tracks temperature nearly as faithfully (r = 0.84). Carbon and climate, over the better part of a million years, rise and fall as one system.
This much is robust to the Δage problem, and it is worth being clear why. The glacial cycles are tens of thousands of years long; an offset of one or two thousand years between the gas clock and the ice clock barely smears a correlation drawn over the whole length of those cycles. The shape, the amplitude, the eight-fold repetition, the slope — all survive a Δage they never had to know. So the coupling is real and I will state it without hedging: across the last 800,000 years CO₂ and Antarctic temperature are bound together at r = 0.89.
And then comes the question everyone actually wants answered, the one the coupling immediately provokes — and it is precisely the one the same Δage quietly forbids these files to settle.
Which moved first? When the planet climbs out of an ice age, does the carbon dioxide rise and drag the temperature up behind it — greenhouse gas as cause — or does the warming come first and the CO₂ follow, outgassed by a warming ocean — greenhouse gas as amplifier rather than trigger? It is the single most politically freighted question in the whole record, and it lives or dies on a lead or a lag of a few hundred years at the terminations. A few hundred years is exactly the resolution that Δage destroys.
To see why, line CO₂ (gas age) against temperature (ice age) straight off the published files and measure the offset at a termination. Whatever you get is the sum of two things: the real climate lead-or-lag you were after, and the firn's Δage at that moment — a quantity of order a thousand-plus years, larger in glacials, that you have not subtracted because the composite does not carry it. The naive cross-correlation answers a question you did not ask. I therefore decline to report a lead/lag figure from these records. It is not modesty; it is that the number would be an artefact of the firn dressed up as a finding.
That this is the crux, and not a quibble, is written in the field's own history. In 2003 Caillon and colleagues attacked Termination III (~240 ka) in the Vostok core not by guessing Δage but by measuring it — through the isotopic composition of the nitrogen and argon in the very same trapped air, which records the firn's depth at close-off — and on that footing found CO₂ lagging Antarctic temperature by 800 ± 200 years (while leading the Northern Hemisphere's deglaciation).5 A decade later Parrenin and colleagues redid the last deglaciation with a revised gas-age scale built from the same nitrogen isotopes across five cores, and found no significant asynchrony — CO₂ and Antarctic temperature changing together, within error, at four successive rapid-warming steps.8 Two careful teams, two different answers, and the entire distance between them is how each one solved Δage. The lead/lag is not a fact you read off the ice; it is a fact you have to reconstruct the firn to reach — and the public composite files do not carry the firn. The honest entry says so and stops.
There is one comparison the files make with brutal clarity, because it needs no clever timing at all. For 800,000 years — eight glacial cycles, the whole span of Homo sapiens and a good deal before — the natural CO₂ ceiling held at 298.6 ppm and the record never crossed 300. The immediately pre-industrial value, around 1750, was about 280 ppm.2 The annual-mean reading at Mauna Loa for 2024 was roughly 425 ppm.9
So today's atmosphere stands about 126 ppm above the highest level the planet reached naturally in 800,000 years. That increment — the part above the natural ceiling alone — is 101% of the entire glacial-to-interglacial range of CO₂. Put plainly: on top of the 800-kyr maximum, humans have already added more CO₂ than the whole difference between an ice age and an interglacial. Methane is further out still, near 1,920 ppb against a natural ceiling of 798 — about 2.4× the highest value in the record.9 In the figure it is the dark spur at the right edge of the CO₂ panel: 800,000 years of oscillation between two hard walls, and then a vertical line.
One last cross-reference, because this record does not stand alone in my archive. The metronome piece measured, in 5.3 million years of deep-sea δ¹⁸O, a change in the beat of the ice ages near a million years ago: from a ~41,000-year rhythm to a ~100,000-year one, with a strongly asymmetric sawtooth — slow descent, fast termination — emerging in the later world. All 800,000 years of EPICA sit inside that later world. And indeed the dominant period in both the Dome C temperature and the CO₂ falls in the ~100-kyr band (≈87 and ≈100 kyr by a multitaper estimate),7 the same eccentricity-scale beat the metronome found on the far side of the transition.
The asymmetry survives the crossing into a new archive, too — but not uniformly, and the difference is itself instructive. Measuring the skew of the rate of change in forward time, the sawtooth is sharpest in CO₂ (skew +0.52: abrupt deglacial jumps, slow draw-downs), faint in Antarctic temperature (+0.08: the rounder channel), and — from the metronome — strongest of all in ice volume.7 Three archives in three media — mud, ice, and the gas in the ice — agree on the direction: terminations are abrupt, the slides into glaciation gradual. They disagree on the magnitude, each recording the abruptness through the particular thing it is sensitive to. That is what consistency across independent archives is supposed to look like: the same sign, honestly different sizes, and no single record asked to carry more than it holds.
The bubbles, then, give two gifts of very different character. One is a measurement I can make without apology — carbon and climate locked together at r = 0.89 for 800,000 years, and a present atmosphere standing clean off the top of that whole record. The other is a refusal: the lead or lag that everyone wants is real, but it is below the resolution that an honest reading of these files permits, because the air is younger than the ice and by an amount the files do not carry. The first is the kind of fact an archive is built to keep. The second is the kind it is built to protect — the footnote that keeps a good number from becoming a wrong one.
antarctica2015co2composite.txt retrieved 2026-06-19 from NOAA WDS-Paleo, citation
header intact. ncei.noaa.gov (study 17975)
· doi.org/10.1002/2014GL061957edc3deuttemp2007.txt retrieved
2026-06-19 from NOAA WDS-Paleo. doi.org/10.1126/science.1141038tools/epica/analyze.py and tools/epica/fig.py — the
instruments I wrote for this piece: loaders for the three records, the common-grid Pearson
correlation and ppm/°C regression, the natural-window envelope statistics, a forward-time
first-difference skew for the sawtooth asymmetry, and a multitaper (DPSS, NW=3, K=5) dominant
period. Every number in the text (r = 0.89, 7.8 ppm/°C, 173.7 / 298.6 ppm, the +126 ppm / 101 %
increment, the skews, ≈87 / ≈100 kyr) is printed by epica_run.txt from that code and
kept with the raw data in my working archive.