How this works

Every active satellite is tracked from the ground, and the resulting orbits are published as element sets. Left alone, an orbit changes only in ways physics dictates: atmospheric drag slowly lowers it, the Earth’s equatorial bulge rotates its plane, the Sun and Moon tug on it. All of that is smooth. A maneuver — a thruster firing — is none of those things. It appears as a sudden break in an otherwise smooth curve. This page watches the published orbits of active payloads and surfaces those breaks.

The measurement

For each satellite we turn its published mean motion into a semi-major axis — the size of the orbit — and read its inclination — the tilt of its plane:

where $\mu = 398600.4418\ \text{km}^3/\text{s}^2$ and $n$ is the mean motion in radians per second. We track both over a rolling 90-day window. A burn along the direction of travel changes the orbit’s size and energy; a burn out of plane changes its tilt.

Finding the break

We remove the smooth decay each orbit is already following; what remains is noise, plus the occasional step. We measure each satellite’s own noise level — robustly, so a real maneuver can’t inflate it — and flag a step only when it clears that noise by a set margin (three standard deviations by default). Independently, we propagate each published orbit forward to the moment of the next one and measure how far the prediction misses: between two honest, maneuver-free orbits the miss is small; a burn in between throws it off. When both tests agree, the detection is marked high-confidence.

Estimating the push

From the change in the orbit we estimate the velocity change — the $\Delta v$ — behind each maneuver. For a burn along track,

and for a change of orbital plane by angle $\Delta i$ at orbital speed $v$,

A geostationary satellite nudging back into its slot spends fractions of a metre per second; one changing orbits entirely spends tens or hundreds. The number you see is an estimate, not telemetry.

What this is not

This is inference from public tracking data, and it has honest limits. The published orbit takes time to settle after a burn, so the detected moment lags the real one, sometimes by days. A geomagnetic storm thickens the upper atmosphere and drags many satellites down at once, which can mimic a small maneuver — because it hits a whole population rather than one object, we treat shared dips as weather and keep only object-specific breaks. And the catalog itself carries noise. So read each entry as “this orbit changed in a way that looks deliberate,” not as a claim about intent, payload, or operator. The satellites that maneuver in the most interesting ways are often the ones that states would rather you not watch closely. This is one small, public way of watching anyway.