What Is a TLE? How Satellite Tracking Actually Works

A reasonable assumption about satellite tracking is that satellites report where they are. They mostly do not, and even when they do, that information is rarely public.

What is public is a description of each object's orbit, published by the organisations that track them with radar and optical sensors. Feed that description into the right algorithm and you can compute a position for any moment — past or future. That description is the two-line element set.

What a TLE describes

A TLE does not contain a position. It contains a set of parameters that define an orbit's shape and orientation, plus a timestamp — the epoch — saying when those parameters were measured.

An orbit needs six numbers to be fully specified, which is why a TLE carries six orbital elements plus supporting information. Between them they answer: how big is this orbit, how elliptical, how is it tilted, which way does the tilt face, where in the ellipse is the closest approach, and where along the path was the satellite at the epoch.

• Inclination — the tilt of the orbital plane relative to the equator, which also sets the maximum latitude reached
• Right ascension of the ascending node — the orientation of that tilt in space
• Eccentricity — how far from circular the orbit is; Starlink orbits are very nearly circular
• Argument of perigee — where in the orbit the lowest point sits
• Mean anomaly — where along the orbit the satellite was at the epoch
• Mean motion — how many orbits per day, which implies the altitude

Why the format looks like that

A TLE is two lines of exactly 69 characters, with every value at a fixed column position and no delimiters. Decimal points are sometimes implied rather than written, and exponents use a compressed notation that omits the letter E.

This is not a design choice anyone would make today. The format dates from the 1960s and was built for 80-column punched cards, where every character was precious. It survived because an enormous amount of software reads it and because the tracking catalogue never stopped publishing it.

The consequence is that TLEs are unforgiving. A single misplaced space breaks parsing, and the checksum digit at the end of each line exists to catch exactly that kind of transcription error.

TLEs only work with SGP4

This is the part most often missed, and it matters.

The values in a TLE are mean elements, not instantaneous ones. They have been fitted so that a specific propagation model — SGP4, the fourth Simplified General Perturbations model — reproduces the observed track. They do not describe the satellite's actual orbit at the epoch in a textbook sense.

Feed TLE values into a simple two-body orbital calculation and you will get an answer that looks plausible and is wrong. SGP4 is not optional; it is half of the data format. The model includes specific corrections for the Earth's equatorial bulge and for atmospheric drag, and the TLE's numbers are calibrated to those corrections.

A TLE is not orbital data plus a choice of algorithm. It is orbital data that only means anything when read by SGP4.

Why accuracy decays

A TLE is a snapshot, and reality drifts away from it for reasons the model cannot know in advance.

Atmospheric drag is the main one. At Starlink altitudes the air is thin but present, and its density varies with solar activity in ways that cannot be predicted precisely. The TLE's drag term captures an average, not the actual conditions over the coming week.

Then there are manoeuvres. A satellite that fires its thruster changes its orbit, and no TLE issued before that burn reflects it. Starlink satellites manoeuvre routinely — station-keeping, collision avoidance, orbit raising — which makes their elements go stale faster than a passive object's would.

In practice, a Starlink TLE a few hours old gives positions good to within a kilometre or so; several days old, the error grows to tens of kilometres; a few weeks old, it can be useless. For naked-eye observing this maps to: fresh data gives you timings good to seconds, week-old data gives you timings good to minutes, and month-old data may point you at empty sky.

• Hours old — sub-kilometre accuracy, timings good to seconds
• A few days old — kilometres of error, timings good to tens of seconds
• Weeks old — tens of kilometres or worse, and any manoeuvre invalidates it entirely

Where TLEs come from

The underlying tracking is done by a network of radar and optical sensors operated by the United States Space Force, which maintains the public satellite catalogue. Element sets are generated from those observations and published through Space-Track, with CelesTrak providing widely used curated feeds organised by constellation.

For large constellations, operators also publish their own ephemeris data, which is generally more accurate than externally derived TLEs — the operator knows about a manoeuvre before it happens. TLEs remain the common currency because everything reads them.

This site refreshes its Starlink elements from the public feed several times a day, which is what keeps the globe and the pass predictions trustworthy. Cached data would be cheaper and quietly wrong.

What SGP4 does with them

Given a TLE and a target time, SGP4 propagates the orbit forward or backward and returns a position and velocity in an Earth-centred inertial frame.

Converting that into something useful for an observer takes two more steps: rotating from the inertial frame into an Earth-fixed frame, which requires knowing exactly how far the planet has turned, and then converting to a local horizon frame — the elevation, azimuth and range that tell you where to look.

All of this runs comfortably in a browser. The 3D globe on this site propagates the entire active constellation continuously in JavaScript, which is a reasonable illustration of how much cheaper computing has become relative to the era the format was designed in.