Guide
How to remove a light pollution gradient from a Seestar image
You stretch a Seestar stack and one side of the picture is brighter than the other, in orange, brown or grey. That is a gradient. It is the most common reason a first image looks muddy, and it is the first thing to fix.
What a gradient is
A gradient is a slow ramp of brightness across the frame that is not part of the object. Light from towns and streetlights, the moon, and the glow near the horizon all add light to the sky, and they add more to one side of the frame than the other.
In a linear stack you do not see it, because the whole image is dark. A stretch lifts the faint signal, and the gradient with it. That is why it appears as soon as you stretch, and why you remove it before you stretch.
It also changes during a night. The target moves across the sky, so the part of the frame that points at the town changes, and a long session can carry a more complex gradient than a short one.
How to remove it
The idea is the same in every program: estimate the smooth ramp, then subtract it. Do it on the linear image, after stacking and before the colour balance and the stretch.
In GraXpert
- Open the stacked FITS file.
- Run background extraction. You can use the AI model, which needs no sample points, or the interpolation method, where you place points on empty sky.
- Look at the background model it shows you. It should be smooth, a ramp and nothing more. If it has bright blobs where your object is, the model has taken part of the object for sky.
- Save the result and carry on with the colour and the stretch.
In Siril
- Open Background Extraction on the linear image.
- Place sample points on empty sky only: away from the object, away from bright stars and away from the black borders stacking leaves. Use few points and a low polynomial degree first. A model with many points follows the object, not the sky.
- Apply it and look at the result at a strong stretch. The sky should be even.
Siril can also call GraXpert’s background extraction from inside the program. Which to use is covered in GraXpert or Siril for Seestar images.
When it goes wrong
- The object is large. When a nebula or a galaxy fills most of the frame, there is little empty sky to measure, and the model can take the faint outer part of the object for gradient and remove it. The Orion Nebula, M42, is a case where this is a risk. Use fewer points, keep them on clear sky, and compare before and after at the edges of the object.
- You removed it twice. Run it once if you can. A second pass has less gradient to find, and it may start to remove real signal, so compare before and after if you try one.
- Dark patches near the object. A sample point on a faint cloud made a dark hole. Move the point.
Fewer gradients to start with
- Point away from the moon, and wait for it to set if you can.
- On emission nebulae under a light-polluted sky, the Seestar’s dual-band filter keeps the contrast, because it lets through only a few narrow colours of light. Use it for those targets if you have it.
- Keep the horizon out of the frame when you plan the target.
Software removes a smooth gradient well. It cannot bring back detail that the pollution hid in the noise.
What Stelora does about it
The first processing stage in Stelora’s pipeline is gradient removal, with GraXpert’s background extraction. It is one of the stages that make the difference between the before and after on the gallery pages. It has the same limit as above: a background model cannot always tell the faint edge of a large object from gradient, so that edge can lose a little. If GraXpert fails on a frame, the image goes on unchanged and your job does not fail. How it works has the other stages.