GUIDES / PREDICTIONS
How TLE data and SGP4 predictions work
By Jeroen Wouda · 2 November 2026 · Predictions
Every pass prediction on this site, and on every app and website that does the same job, comes from two things: a text file describing an orbit, and a mathematical model that pushes that description forward in time. Neither is mysterious, and knowing roughly how they work explains most of the oddities you will meet, from times that shift between visits to two sources that disagree.
The element set
A two-line element set, or TLE, is a compact snapshot of an orbit at a stated moment, called the epoch. It is literally two lines of fixed-width text, with a name line above them. Encoded in those characters are the orbital inclination, the eccentricity, the orientation of the orbit in space, the position of the satellite along it at the epoch, and the rate at which the satellite goes round.
There is also a drag term, a small number describing how quickly the orbit is decaying, and a checksum digit at the end of each line so that transmission errors can be caught.
The numbers are produced from tracking: radar and optical observations of the object, fitted to an orbit model. For the station and most catalogued satellites the tracking comes from the US Space Force's space surveillance network, and the resulting element sets are distributed publicly. CelesTrak is the most widely used source, and the one behind most consumer predictions.
The propagator
An element set on its own tells you where the satellite was at the epoch. To know where it will be at 21:14 on Thursday you need a model, and the standard one is SGP4: Simplified General Perturbations, version 4.
SGP4 is an analytic model. Rather than integrating the equations of motion step by step, it applies a set of formulas that account for the main perturbations on a satellite in low orbit: the fact that the Earth is not a sphere, the drag of the thin atmosphere at 400 kilometres, and the slow drifts those cause in the orbit's shape and orientation. Run it for a given time and it returns a position and velocity in an Earth-centred frame.
The word simplified is honest. SGP4 is a compromise between accuracy and speed, chosen because it can produce thousands of predictions in the time a numerical integrator would take to do one. For looking up at a point of light it is more than good enough.
From orbit to your sky
The rest is geometry. The prediction code takes the satellite's position, subtracts your position on the rotating Earth, and converts the result into a direction and a distance: azimuth, elevation, range. It then searches forward in time for the moments when elevation crosses zero, which gives rise and set, and finds the highest point in between.
Brightness comes from the same calculation. Given the Sun's direction, the satellite's position and yours, the code can estimate the phase angle and the range, and from those a magnitude. That estimate is rough by nature, because it assumes a standard shape and reflectivity, but it is good enough to separate a brilliant pass from a marginal one.
Why predictions drift
An element set is a fitted snapshot, not a measurement of the future. Its accuracy decays with time from the epoch, and the rate of decay depends on things that are not in the file.
The biggest is atmospheric drag. The thermosphere expands and contracts with solar activity, and a satellite at 400 kilometres runs into noticeably more air when it does. A drag term fitted last week may be wrong this week.
The second is manoeuvring. The station regularly fires its engines, often using the thrusters of a docked cargo vehicle, to raise its orbit and compensate for drag. After such a burn the old element set describes an orbit the station is no longer in, and a prediction made from it can be minutes out, or can show a pass that does not happen at all until fresh elements appear.
Why sources disagree
Two sites predicting the same pass can differ by a few seconds. Usually they are using element sets of different ages, or slightly different assumptions about the horizon and atmospheric refraction. Neither is broken. If the difference is larger than that, one of them is working from stale data.
The practical rule
Element sets for the station are refreshed frequently, but no prediction is a promise. Check the day before to plan, and check again on the day you intend to go out. That is not a flaw in the method. It is how the method works.
Where to go next
- See the result of all this on the ISS passes page, where each row is a propagated prediction.
- Pass tables or an app: which should you use covers how to read those numbers without over-trusting them.
- What the ISS actually looks like from the ground explains the brightness estimate in the last column.
More guides
- Pass tables or an app — which should you use — Both use the same orbital data. The real difference is how you plan, how you find the direction, and what happens to your night vision.
- Best time of year to watch the ISS — There is no single best month. The good weeks depend on the angle between the orbit and the Sun, and here is how that cycle works.
- What the ISS actually looks like from the ground — Why the space station is a point of light rather than a shape, and why its brightness changes from one pass to the next.