# What is the accuracy / uncertainty of Two Line Elements (TLEs)?

I use TLE sets supplied on this website to find the future position of satellites (well usually just the ISS) using the SGP4 propagation method implemented in C++.

My question is:
What is the accuracy of the satellite position in the supplied TLEs in terms of kilometers or error % ? And also if it is possible to estimate the error of the satellite's position after propagation with using SGP4 for a given time interval ?

Any sources or references supporting the answers would also be very much appreciated.

Edit:
It is really not important for the error estimation to be precise and result of a rigorous analysis. But it should be based on some valid assumptions.

It depends on a lot of factors, including:

• Time since epoch
• Altitude
• Atmospheric density variations
• Spacecraft maneuvers
• Accuracy, number and distribution of observations used to fit the TLE
• Fit span used for differential corrections

Typical errors for a TLE for a non-maneuvering spacecraft at an altitude higher than 400km and with good observation data are approx 1 km error at epoch, growing at about 1 to 2 km per day.

Many TLEs issued by JSpOC (which is the source of the data on Kelso's Celestrak website) fit those criteria, but many do not, and you can't generally tell just by looking at the TLE itself. If there aren't enough observations or they are "cross-tagged" (identified as the belonging to an incorrect object) then the quality can be much worse, tens or hundreds of km in error even at epoch.

Planet Labs operates a number of 3U cubesats, determines their orbits independently and publishes daily the RMS error of the JSpOC TLEs: http://ephemerides.planet-labs.com/jspoc_matches.txt

• Thank you for your answer. Do you have a source for the sentence telling about typical numbers for a TLE ? And is there any "official" resource that also gives such approximation (and is by chance valid for the ISS too) ? Thank you. – James C Dec 3 '14 at 19:32
• It's by personal observation over a couple of years of operating a fleet of satellites, and from talking with experts in the field such as Kelso and Vallado. For the ISS, accuracy is likely to be better. See also spaceflight.nasa.gov/realdata/sightings/SSapplications/Post/… – pericynthion Dec 3 '14 at 19:47
• I would like to ask something, just to make sure I understood your answer correctly. The error of satellite's position is about 1 km at epoch of the obtained TLE. Then by using the SGP4 propagator, the computed positions degrade in accuracy by about 1-2 km per day of propagation ? Thank you. – James C Dec 4 '14 at 0:44
• Yes, that's the rule of thumb that's traditionally used, assuming good initial conditions. – pericynthion Dec 4 '14 at 1:05

Here is a link to a paper discussing a probabilistic error model for TLEs for MEO and GEO satellites. https://arc.aiaa.org/doi/abs/10.2514/6.2018-5241