Description
A satellite or aircraft trail is the streak left by a moving object crossing the field during an exposure.
The sensor integrates the object's light signal along its entire path, producing a thin line (1 to 3 pixels wide for a satellite, wider and stroboscopic for an aircraft with flashing navigation lights).
The defect affects only one frame at a time and is very effectively removed by statistical rejection during stacking (sigma clipping, Winsorized Sigma Clipping, ESD), provided enough frames are available so that the algorithm has a clean majority for each pixel.
The issue has become significantly more prevalent since 2020 with the mass deployment of Starlink, OneWeb, and other constellations: it is now rare to complete a full night without several passes per session, particularly during twilight hours when low-Earth-orbit satellites are still sunlit.
Visual signature
On a single frame: a very thin straight line, typically 1 to 3 pixels wide, crossing all or part of the field.
For a satellite, the trail is continuous and of fairly uniform brightness (sometimes variable if the satellite is tumbling, producing periodic flashes).
For an aircraft, the signature is typically double or triple: a continuous line from the steady navigation light plus a series of equidistant dots from the strobe lights.
On a stacked master without proper rejection: a residual trace visible as a faint, diffuse streak, sometimes discontinuous if sigma clipping partially filtered it.
With Winsorized Sigma Clipping properly configured on 30+ frames, the trail disappears entirely.
Differential diagnosis
Distinguish from a cosmic ray (a point impact or very short track spanning only a few pixels, never crossing the entire field).
Do not confuse with a guiding glitch or oscillation burst (stars deformed everywhere, not just a single line on a dark background).
Different from a geostationary satellite which, in sidereal tracking, leaves only a short, regular streak (it drifts slowly relative to the tracked stars).
Do not mix up with a dead sensor column (always in the same position, perfectly vertical, present on every frame) or a meteor trail (often brighter, sometimes fragmented or showing luminous persistence).
If several parallel trails appear on the same frame, it is almost certainly a freshly deployed Starlink train. If the trail curves, that is field rotation during a long exposure, not a satellite.
Probable causes
- Starlink constellation (more than 5,000 active satellites in 2026), particularly dense at the start and end of night
- Other LEO constellations: OneWeb, Kuiper, Qianfan, Iridium NEXT
- Legacy satellites (ISS, Hubble, debris), with total orbital traffic continuously rising
- Commercial aircraft, ultralight aircraft, drones in controlled airspace
- Sessions conducted during astronomical twilight, when the ratio of sunlit to unlit satellites is at its highest
- Targets low on the horizon (below 30 degrees altitude), where the apparent density of illuminated satellites is higher
- Targets near the ecliptic during high satellite traffic windows
- Long individual exposures (over 3 minutes) that increase the probability of a crossing
Course of action
- Stack with statistical rejection: Winsorized Sigma Clipping (PixInsight) or Generalized Extreme Studentized Deviate on 20+ frames
- Set kappa low/high between 2.5 and 3.0 depending on the number of frames
- For datasets under 15 frames, manually identify and mask trails with CloneStamp or DefectMap before integration
- In Siril, use Linear Fit Clipping or Winsorized Sigma as the stacking rejection method
- Enable inter-frame dithering to prevent a trail in the same sky region from landing on exactly the same pixels across multiple subs
- Dedicated tools: satellitedetector (PixInsight script), Astrometry.net annotation to identify satellite passes
- Schedule acquisitions outside astronomical twilight for high-priority targets
- Keep frames with trails: sigma clipping does its job, manual exclusion is unnecessary
- For a single short exposure (planetary, lunar), reshoot the sequence if a trail ruins the result
The Doc's advice
Twenty years ago, a satellite trail was the event of the evening -- you proudly posted it on the forums. Today it has become background noise in astrophotography, and the absolute weapon for eliminating it comes down to two words: Winsorized Sigma. If your trails survive the stack, it is almost always because you do not have enough frames or your kappa is too permissive. The solution is rarely more aggressive filtering -- it is more integration.
Think you can see this defect in your image?
Run a diagnosisFrequently asked questions
Should I discard frames that contain satellite trails?
No, absolutely not. Modern statistical rejection algorithms (Winsorized Sigma Clipping, ESD) are specifically designed to eliminate these outlier values pixel by pixel without throwing out the entire frame. You would lose far more signal by discarding 10 out of 60 frames than by letting the algorithm filter the trails. The only exception: if you have only 5 to 10 frames in total, statistical rejection lacks enough data and manual masking becomes preferable.
Why do I see Starlink trails appearing as a train, dozens at a time?
During an initial deployment, SpaceX releases satellites in a cluster before they disperse to their final orbits. For the first few days after launch they remain bunched together and pass one after another, producing those characteristic trains that are very visible to the naked eye and devastating for an astrophoto session. Sites such as Heavens-Above or CelesTrak allow you to predict passes and avoid pointing toward the expected trajectory.
Does dithering really help eliminate satellite trails?
Indirectly, yes. Dithering does not remove an existing trail but ensures that a trail in the same region of sky does not fall on exactly the same pixels from one frame to the next. This helps rejection algorithms better discriminate legitimate signal from trails, especially along the edges of the track where rejection is more delicate. Dithering remains primarily an anti-fixed-pattern noise strategy (FPN, walking noise), but its benefit for satellite trail handling is a welcome bonus.
Are there tools to predict satellite passes during a session?
Yes. Heavens-Above, CelesTrak, and the Stellarium app all use up-to-date TLE (Two-Line Elements) databases that let you predict visible passes. For a serious session, cross-referencing your target against Starlink/OneWeb predictions for a given time window allows you to avoid the worst moments. NINA also offers a satellite visualization plugin in the framing assistant. Note that prediction accuracy remains limited for satellites actively performing orbital maneuvers.