FLOAT
EN FR

Block

Astrometry

FLOAT 0.2.0 · beta of 15 November 2026

What it does
It recognises the patch of sky in the stack from its stars, so that every pixel gets its position on the sky, and the later blocks can use the star catalogue.
What it shows
In Inspect, the image with the sky grid and the catalogue stars set on its stars, with the image scale, the field of view, the position residual and the focal length found; in the Report, the target, the field and the scale.
What can be set
Usually nothing, when the files hold the pointing and the focal length; otherwise give the Target name, or the Focal length.

The Astrometry block finds where the stack points on the sky. A first solution comes from Asterion, shipped with the installer; without it, or when it finds nothing, FLOAT’s own search takes over, blind when the target is unknown. The solution is then refined on the Gaia DR3 stars with a tangent-plane projection and a distortion polynomial. Colour calibration, the Gaia catalogues and the H-alpha reference read this solution. A run downloads nothing: the star catalogue is installed once.

  1. Selection
  2. Stacking
  3. Astrometry
  4. Gradient
  5. Deconvolution

1What goes in, what comes out

The block reads the stack and finds its stars (a background mesh of 64 pixels, a threshold of 15 times the noise, 6 pixels at least). It also reads what the project knows of the field: the target, the telescope pointing written in the files, the focal length and the pixel size.

It writes the world coordinate system of the stack: a tangent-plane projection with a distortion polynomial (the SIP convention). It also reports the centre of the field, its size, the scale in arcseconds per pixel, the angle of north, whether the image is mirrored, the residual of the fit and, when the pixel size is known, the focal length the solution implies. The Report also gives the residual in arcseconds.

2The first solution

Asterion first. The first solution comes from Asterion,[1] shipped with the installer. It is given at most 10 seconds. When the target’s position is known, it is searched near it; otherwise on the whole sky, if Blind solving is on. When the scale comes from the focal length and Asterion finds nothing at that scale, it is asked again at any scale, in case the focal length was entered wrongly. The fine fit of section 3 starts from this solution, and the search below is skipped.

FLOAT’s own search. When Asterion is not installed, cannot run or finds nothing, FLOAT’s own search finds the field. The log says which case it was. The search runs in four steps.

  • The scale. It is 206.265 × pixel size (µm) / focal length (mm), in arcseconds per pixel, divided by the drizzle scale. When it is unknown, 22 values from 0.4 to 40 arcseconds per pixel are tried, the focal lengths this camera was solved at before first.
  • The match. Triangle matching of the 150 brightest detections with the 150 brightest Gaia stars projected at the trial scale, both mirror images tried. Fifteen matched pairs are taken as a match.
  • The centre. The given centre first, then centres shifted by the Search range, a fraction of the field.
  • The catalogue depth. The faintest Gaia magnitude is set by the field’s area, from G 10.5 for a wide field to G 16 for a small one.

The random draws of the match are seeded: the same stack gives the same solution at every run.

Blind solving

When the centre is unknown, or the guided search fails, the field is identified on the whole sky from its brightest stars (up to 80). The principle is the one of Astrometry.net:[2] each group of four stars is coded by the positions of its two inner stars in the frame where its two most distant stars stand at (0, 0) and (1, 1), a code that does not change with translation, rotation or scale. The codes of the image are looked up in an index of the sky built from the bright stars of the Hipparcos and Tycho-2 catalogues.[3] Each candidate is verified by counting how many other stars of the image fall on catalogue stars, and is kept only when that count stands well above what chance gives. The index holds the stars of Hipparcos to V 9.5 and of Tycho-2 to VT 11, so the blind search is made for fields wider than about 1 degree.

Fields of few stars

Under 50 stars in the stack, the block uses another decision. Every triple of the 30 brightest stars is compared with every triple of the catalogue stars. A solution stands only when its Bayes factor reaches 109, by the decision of Lang et al.:[2] each star either has no counterpart (a quarter of them are allowed none) or lies within a Gaussian of 1 pixel around a catalogue star. No random draw is used. With fewer than 6 stars, no field is identified. No distortion polynomial is fitted, as it would follow the noise of so few stars.

3The fine fit on Gaia DR3

The first solution gives the stars’ approximate place on the sky. The fine fit then runs in passes. A tangent-plane projection with a distortion polynomial of the chosen order is fitted on up to 4000 detections. The Gaia stars are projected through it and matched to the detections within 3 pixels, then the fit is made again on up to 8000 detections, and twice more with matches within 1.5 and 1 pixel. Stars within 5 pixels of the edge are not used. The median residual of the last pass is the one reported.

The solution is written once the fit is done. The solution of a sparse field (section 2) must also give a median residual under 1 pixel, or no solution is given rather than a wrong one.

4What the solution is used for

Every later step that uses the sky relies on this solution.

  • The Gaia catalogues. The Catalogs step places the Gaia stars of the field on the image, their positions carried to the date of the frames with their proper motions, and selects the stars whose spectra the Color block uses.
  • The H-alpha reference. A survey of the sky in H-alpha is reprojected onto the stack. It shows where the faint nebulosity is, so that the Gradient block takes no sky sample there.
  • Colour. The Color block measures the stars the solution places and compares them with their Gaia spectra.

When no field is found. The setting If the field is not identified decides. With Continue without solution, the run goes on, the block is drawn in orange and its tooltip gives the reason. Color then takes its fallback balance, and Gradient chooses its samples without the survey. With Stop, the block fails.

A target far from the pointing recorded in the files is handled in two tries. When the two differ by more than the field’s diagonal plus 2 degrees, the typed target is tried first, without the blind solver. If that fails, the pointing of the files is tried, and the log asks to check the project’s target.

5The offline star catalogue

The stars FLOAT reads come from an offline Gaia DR3 catalogue installed once, by the installer or from Settings › Catalogs & astrometry. It derives from published extracts of Gaia DR3.[4][5] The star index of the blind solver is installed in the same place.

A run downloads nothing. When a catalogue is missing, the block says so and names the setting that installs it. A target name is looked up in the names already resolved and in the Messier, NGC, IC, Sharpless and Lynds catalogues that FLOAT ships. An unknown name needs its RA and Dec.

The extract keeps the 127 brightest stars of every cell of 14 arcminutes. In the Milky Way it can stop short of the magnitude asked for; it is enough for a solution. Without the extract, the block reads the sky tiles of the cache, and fails with the reason when they are not all there.

6Settings

Table 1.The settings of the block. The Advanced ones are marked.
SettingValueRangeWhat it does
Target namefrom the project or the filestextName of the main object, resolved to the approximate centre of the field. Not needed when RA and Dec are given or when the files hold the pointing.
Centre RAfrom the project or the files0 – 360 °Approximate right ascension of the centre. Within about 20 % of the field width is enough.
Centre Decfrom the project or the files−90 – 90 °Approximate declination of the centre.
Focal lengthfrom the files or the projectmm, above 0With the pixel size, it gives the image scale. Empty, the scale is searched (a few minutes without Asterion).
Pixel size (Advanced)from the camera or the filesµm, above 0Size of a sensor pixel.
Image scale (Advanced)computed, else searchedarcsec/px, above 0Size of a pixel of the stack on the sky.
Blind solving (Advanced)onon, offIdentifies the field on the whole sky when the centre is unknown or the guided search fails.
Search range (Advanced)0.2 × field0 – 1, step 0.05How far from the given centre the search looks for the field.
Distortion order (Advanced)30 – 5Order of the SIP polynomial: 3 for refractors with reducers and wide fields, 2 for small fields, 0 for a pure tangent plane.
If the field is not identifiedContinue without solutionContinue without solution, StopWhat the block does when no field is identified (section 4).

7Limits, and what was tried

  • A wrong focal length does not hide the field. When the field is not found at the scale of the entered focal length, it is searched again without it. A warning says the field is at another scale, and to check the focal length and the pixel size.
  • Blind solving needs its index. When the star index is not installed, the all-sky search cannot try, and the block asks for the target or its coordinates.
  • Narrow fields are hard to solve blind. The blind solver is built for wide fields, with enough bright stars in them.
  • Unknown scale and far centre can take minutes. Without Asterion, an unknown scale means a search over 22 scales. A match at another scale than the trial one is followed at once, to avoid walking the whole grid.
  • Sparse fields. Under 50 stars, a solution needs strong proof and has no distortion polynomial. A field that does not give it has no solution.

References

  1. Asterion Solver authors, Asterion, Apache License 2.0. github.com/asterion-solver/asterion
  2. Lang, D. et al. 2010, Astrometry.net: Blind astrometric calibration of arbitrary astronomical images, AJ 139, 1782. arXiv:0910.2233
  3. ESA 1997, The Hipparcos and Tycho Catalogues. cosmos.esa.int/web/hipparcos; Høg, E. et al. 2000, The Tycho-2 catalogue of the 2.5 million brightest stars, A&A 355, L27. ADS 2000A&A...355L..27H
  4. Cass, I., Knagg-Baugh, A., Melis, C. & Richard, C. 2025, Siril Astrometry Catalogue extracted from Gaia DR3, v1.0.0. doi:10.5281/zenodo.14692304
  5. Gaia Collaboration, Vallenari, A. et al. 2023, Gaia Data Release 3: Summary of the content and survey properties, A&A 674, A1. arXiv:2208.00211