MultiscaleGradientCorrection (MGC)
MultiscaleGradientCorrection (MGC) compares the image with a reference map of the sky (MARS). Structures smaller than a selectable scale are taken from your own image, larger ones are matched to the reference.
Purpose
| Task | Setting |
|---|---|
| remove gradients in fields with real large-scale emission | gradient scale 1024 to start |
| weight your own data more than the reference | larger gradient scale |
WhyMGC rather than GradientCorrection?
Models large-scale brightness and colour gradients (light pollution, moon, residual vignetting) and subtracts them – structures are kept out of the model by protection masks. Open fundamentals → judges by smoothness only: everything large-scale counts as gradient. MGC compares with a reference map and can thus distinguish real large-scale signal – Milky Way emission, extended nebulae – from gradients.
Principle
- MARS is a reference of the sky made from broadband data. Image areas larger than the gradient scale are matched to it; smaller structures come from your own image.
- For this comparison MGC needs the image’s brightness in absolute units, which SPFC (SpectrophotometricFluxCalibration) provides beforehand.
- If the image differs spectrally from the broadband reference (e.g. with a line filter), a small scale also matches real differences to the reference.
Parameters
| Parameter | Effect | Typical choice |
|---|---|---|
| Gradient scale | boundary between your image and the reference | 1024 (default) |
| Gradient scale | Effect | Risk |
|---|---|---|
| smaller (e.g. 512) | more thorough gradient removal | differences from the reference are subtracted as gradient |
| 1024 | balanced | – |
| larger (1536–2048) | your data also determine larger structures | large-scale residual gradients possible |
Operation
Prerequisites: Gaia DR3/SP and MARS data installed; image cropped (DynamicCrop); astrometric solution present.
- ImageSolver, if the solution does not come from Chains calibration, cosmetic correction, weighting, registration, normalisation and integration into a reproducible pipeline – from individual subs to aligned master lights. Open fundamentals →.
- SPFC (SpectrophotometricFluxCalibration): choose the sensor’s QE curve and the filter curves used; Gray filter has no meaning for RGB images. SPFC only computes scaling factors for flux units – no colour calibration.
- Apply MGC with gradient scale 1024, preferably on a copy.
- Then colour calibration with Calibrates the colours of a broadband or OSC image by comparing thousands of stars with Gaia spectra – objective and reproducible, including background neutralisation. Open fundamentals →.
Checking the SPFC result:
| Criterion | Assessment |
|---|---|
| Total sources | several thousand per channel – a good basis |
| Dispersion / scale factor | broadband typically 1–3 %; with a line or multi-band filter higher (about 5–9 %) because SPFC does not know the filter – sufficient for gradient correction |
| Curves (sorted scale factors) | flat over most stars; steep ends (faint or saturated stars) are normal |
Typical applications
Multi-band or line filters on colour cameras
With a filter that emphasises emission lines (e.g. quad band-pass), the image differs spectrally from the broadband reference. Try a larger scale (e.g. 1536) so that less is matched to the deviating reference; compare two copies (1024 / 1536).
Comparing GC ↔ MGC
Preview on faint outer parts, drag it onto both results, 1:1, same STF. MGC often preserves faint structures slightly better; the background may stay slightly warmer, because extended emission is real signal. The background colour is set by colour calibration anyway.
What a good result looks like
Typical errors
| Observation | Cause | Remedy |
|---|---|---|
| Message “No reference data found” | region not covered by MARS | use Models large-scale brightness and colour gradients (light pollution, moon, residual vignetting) and subtracts them – structures are kept out of the model by protection masks. Open fundamentals → |
| MGC does not start or computes wrongly | SPFC missing | run SPFC first |
| Real structures weakened | scale too small, image differs spectrally from the reference | raise the gradient scale |
| Large-scale residual gradient | scale too large | lower the gradient scale |
When to use it – and when not
| Good for | Better another tool |
|---|---|
| broadband and OSC data with large-scale emission | pure narrowband masters → Models large-scale brightness and colour gradients (light pollution, moon, residual vignetting) and subtracts them – structures are kept out of the model by protection masks. Open fundamentals → (MARS is based on broadband data) |
| when GC pulls nebula into the model | region without MARS coverage → GradientCorrection or DBE |
GradientCorrection · HaLRGB workflow · OSC workflow with a quad band-pass filter · SpectrophotometricColorCalibration (SPCC) · WeightedBatchPreprocessing (WBPP)