What causes ice ages?

Earth’s orbit stretches and relaxes over about 100,000 years, its axial tilt varies over about 41,000, and the direction of that tilt precesses over about 23,000. These change where and when sunlight falls, without changing the total much.

In 1976 a study of deep-sea cores from the Southern Hemisphere found those periods in three separate records of past climate, and concluded that changes in Earth’s orbit are the fundamental cause of the succession of ice ages. The record has been extended and refined many times since and the periods are still there.

A stack of 57 deep-sea records published in 2005 gives a continuous curve of benthic oxygen isotopes over the last five million years, sampled finely enough to see individual glacial cycles. Those isotopes carry global ice volume and deep-ocean temperature together, and a stack cannot separate the two.

It shows the character of the cycles changing about a million years ago, from a 41,000-year rhythm to a longer, larger, more sawtooth-shaped one.

Eccentricity moves the global average sunlight by about half a watt per square metre, against the fifty and more watts precession swings across a high northern summer. Yet for the last million years the ice ages have followed it, with enormous amplitude.

Something in the system is amplifying a small push. Ice sheet dynamics, carbon dioxide feedbacks, dust over the Southern Ocean, and the slow response of the crust under the weight of ice are all candidates, and combinations of them can be made to work.

The timing, at least, looks more orderly than it did. A 2025 analysis finds that since 900,000 years ago each deglaciation has begun with the first precession peak after a minimum in eccentricity, counting only the peaks that begin while the tilt is increasing. Why the cycles last that long, and why they keep to eccentricity at all, has no agreed answer, and it is the central unsolved problem of the subject rather than a detail at its edge.

The transition from 41,000-year cycles to longer ones happened with no significant change in the pacing or amplitude of the orbital forcing. The cause has to be internal.

Proposals include a long decline in atmospheric carbon dioxide and the gradual removal of loose sediment from under the northern ice sheets, which would change how they slide. In one Earth system model either can produce the switch, but only the two together give both its timing and the long cooling around it. The test is hard: continuous ice-core records of carbon dioxide reach back only 800,000 years, and older Antarctic ice gives scattered snapshots, which find the glacial lows falling across the transition while the average barely moves.

Deglaciations are fast compared with the slow slide into a glacial period, and the carbon dioxide rise that accompanies them is thought to come from old carbon stored in the deep ocean and released mainly through the Southern Ocean, a route still being characterised.

The ordering of events across a termination, ice, ocean circulation, carbon dioxide, temperature, is still being pinned down. For the last one, a reconstruction from 80 records finds global temperature rising with carbon dioxide and generally behind it. The ordering is what any causal account needs.