Earth

The day is not 24 hours

—

Every day the IERS has measured since 1962 — 23,614 of them — drawn as how much longer or shorter than exactly 24 hours the planet actually took to turn. Press play. The fine wave is the weather: the atmosphere shifts enough mass around to swing the spin by 1.26 ms over each year, longest around 28 March and shortest around 10 July. The slow arch it rides on is thought to be the liquid core trading momentum with the mantle, and it crests in 1972 — the year of the longest day on record, 4.36 ms over. Then it comes down: every year from 2020 to 2025 averaged below the line, the first run of its kind here, and the shortest day ever measured is 5 July 2024 — the seasonal low landing on the modern one. The strip underneath is what all of this does to the clocks, and every rung in it is a second someone had to add — 27 leap seconds since 1972, and none for nine years, already the longest wait there has been. Tap or hover any day.

shorter longer

a leap second — one whole second added

1963–72, when UTC was still adjusted by fractions

Data & source

Data: IERS Earth Orientation Parameters, C04 combined solution (1962–2026), extended past 2026-01-06 with the rapid service file finals2000A.all · International Earth Rotation and Reference Systems Service (IERS), Earth Orientation Centre, Paris Observatory · freely available for scientific use with attribution · fetched —.

Scope and gaps, plainly. Two quantities, both measured, and they are not independent. Excess length of day is how much longer than 86,400 SI seconds the Earth actually took to complete one turn — a couple of milliseconds either way, which is why nobody notices. UT1−UTC, the strip below, is how far the Earth's own time has drifted from atomic time; it is the running sum of the excess, reset each time a step is inserted into UTC. Steps here are derived from the data rather than read from a table: any day-to-day jump above 40 ms is a human decision, because the planet cannot change its spin that fast. That finds both regimes — the 9 fractional steps of 1963–72, when UTC was still "rubber band" time running at a deliberately offset rate, and the 27 whole leap seconds since. The last was 1 January 2017, and the wait since is already the longest there has ever been. Precision is not constant across this record and the table says so year by year: the formal uncertainty on a day's length was about 1.4 ms in 1962, when the measurement was optical astronomy, against about 0.007 ms now, when it is VLBI and satellite laser ranging. The shape of the curve is trustworthy throughout; the absolute level in the 1960s is not, to the same degree, and the early series is smoothed rather than daily-independent. The two source files overlap from 1973 on, and over the last five years — the era the join actually falls in — they agree to about 0.02 ms in both quantities, far below anything this page can draw. In the 1970s they diverge by up to 3 ms, which is why the early record is taken from the combined solution alone and never from the rapid file. The join is marked by a tick on the date axis, and values after it are IERS "final" rows only — the Bulletin A predictions that run a year past today are not measurements and are not drawn.