FITS headers reference¶

Everything on this page is written by argos.core.imaging.fits_writer. The data array is uint16, linear, and still in its Bayer mosaic — ARGOS never debayers, stretches or scales what it writes to disk.

Note

Many headers are written only when the acquisition supplied the value. A frame taken with no site configured has no SITELAT; a frame taken without a solved target has no TARGRA. The tables below mark those conditional.

Reading one for yourself

from astropy.io import fits

with fits.open("XX_Cyg_ir_light_e020000ms_g080_00024.fit") as hdul:
    print(repr(hdul[0].header))

The format is FITS 3.0 (Pence et al. 2010); keyword conventions follow the ASCOM/AAVSO usage common to acquisition software, so most of what follows will already be familiar from N.I.N.A. or MaxIm DL.

File naming¶

{OBJECT}_{FILTER}_{TYPE}_e{EXPOSURE_MS}ms_g{GAIN}_{INDEX}.fit

For example XX_Cyg_ir_light_e020000ms_g080_00024.fit: a 20 s exposure at gain 80 through the IR filter, frame 24 of the series.

The prefix is fixed for one homogeneous series; only the final index changes, which is what Siril’s sequence discovery requires.
Exposure is in milliseconds, zero-padded to six digits. The index is 1-based and padded to five.
_g{GAIN} is omitted when the gain is unknown.
The extension is .fit, not .fits.
There is deliberately no timestamp in the filename — the precise time belongs in DATE-OBS.
Object, filter and type are sanitised for the filesystem and lower-cased (except the object, which keeps its case).

Session folder¶

{sessions}/{RUN}_{OBJECT}/
├── lights/
├── darks/
├── flats/
├── biases/
├── session.json
├── photometry_quality.json
└── diagnostics/

{RUN} is the run identifier, or the first frame’s UTC timestamp formatted %Y%m%dT%H%M%SZ when there is none — for example 20260906T213734Z_XX_Cyg. Image-type folders are lower-case plurals, as Siril expects.

Structure¶

Keyword	Meaning
`SIMPLE`	file conforms to FITS standard
`BITPIX`	number of bits per data pixel — always 16
`NAXIS`, `NAXIS1`, `NAXIS2`	axis count and lengths (X/columns, Y/rows)
`EXTEND`	FITS dataset may contain extensions
`IMAGETYP`	`Light Frame`, `Dark Frame`, `Flat Frame` or `Bias Frame`

BZERO = 32768 and BSCALE = 1 are added by astropy for unsigned 16-bit data.

Timing¶

Keyword	Meaning
`DATE-OBS`	UTC date/time of exposure start
`DATE-LOC`	local date/time of exposure start
`DATE-AVG`	UTC date/time of exposure midpoint
`MJD-OBS`	MJD of exposure start (UTC)
`MJD-AVG`	MJD of exposure midpoint (UTC)
`EXPTIME`	[s] exposure time
`EXPOSURE`	[s] exposure time (alias)

The midpoint headers are the ones photometry uses: a light curve point is timed by DATE-AVG, not DATE-OBS.

Sensor and gain¶

Keyword	Meaning
`GAIN`	camera gain setting (0–600 on the Seestar), not a physical unit
`EGAIN`	[e⁻/ADU] electron gain, with its source named in the comment
`GAIN_E`	[e⁻/ADU] electron gain (AAVSO alias)
`EPERDN`	[e⁻/ADU] electron gain (alias)
`RDNOISE`	[e⁻] read noise, from the sensor reference
`FULLWELL`	[e⁻] full-well capacity, from the sensor reference
`XBINNING`, `YBINNING`	binning factors — always 1
`CCD-TEMP`	[°C] sensor temperature (conditional)
`OFFSET`	electronic offset / bias setting (conditional)
`READOUTM`	sensor readout mode (conditional)

Warning

GAIN and EGAIN are different quantities. GAIN is the dimensionless slider value; EGAIN is the electrons-per-ADU conversion the photometry noise model needs. Do not use one for the other.

Which ADU? An unresolved question — read before you trust EGAIN

The sensor is a 12-bit ADC whose output is scaled into a 16-bit file. That leaves a factor of 16 hanging over every electron conversion, and this documentation cannot tell you which side of it EGAIN sits on, because the code does not record it.

The consistency check that ought to hold is \(\text{EGAIN} \times (\text{saturation ADU}) \le \text{FULLWELL}\). With ARGOS’s own reference values it fails at every shipped gain — see Differential photometry — method and equations §3.5 for the table. Something in EGAIN / FULLWELL / camera.linearity_max_adu is referenced to a different ADU scale from the others.

Until it is resolved: EGAIN and FULLWELL are reference-table values, not measured calibrations of your unit. If your reduction depends on absolute electrons — a real SNR, a photon-noise model — measure the gain yourself from a pair of flats rather than reading it out of this header. Everything differential is unaffected, because a constant gain error cancels.

Instrument and optics¶

Keyword	Meaning
`TELESCOP`	telescope name, from the active profile
`INSTRUME`	sensor name — `IMX585` for the S30 Pro
`APTDIA`	[mm] aperture diameter — 30.0 for the S30 Pro
`FOCALLEN`	[mm] focal length — 160.0 for the S30 Pro
`FOCRATIO`	focal ratio — f/5.3 for the S30 Pro
`XPIXSZ`, `YPIXSZ`	[µm] unbinned pixel size — 2.9
`BAYERPAT`	Bayer pattern — `GRBG`
`XBAYROFF`, `YBAYROFF`	Bayer offsets
`EQUINOX`, `RADESYS`	equinox and celestial reference system

Frames written before 0.4.1

FOCRATIO was written as f/3.2 instead of f/5.3, and APTDIA was not written at all. Both are corrected in 0.4.1 — check any frame you already hold.

Target and pointing¶

Keyword	Meaning
`OBJECT`	target name, truncated to 68 characters
`FILTER`	the physical filter-wheel selection: `IR`, `LP` or `Dark`
`TARGRA`, `TARGDEC`	[h], [deg] requested target coordinates (J2000) (conditional)
`RA`, `DEC`	[h], [deg] pointing coordinates (J2000) (conditional)
`OBJCTRA`, `OBJCTDEC`	the same pointing in sexagesimal (conditional)
`ALTITUDE`, `AZIMUTH`	[deg] telescope altitude and azimuth (conditional)
`PIERSIDE`	side of pier (conditional)

Sexagesimal formats are HH MM SS.ss for OBJCTRA (two decimals) and ±DD MM SS.s for OBJCTDEC (one decimal).

Important

FILTER is the filter wheel’s position, never a photometric band. The band used for differential photometry is a separate setting (photometry.default_band, TG by default) and is not stamped into this header.

Sky conditions¶

Keyword	Meaning
`AIRMASS`	airmass at exposure start, Pickering (2002) (conditional)
`MOONSEP`	[deg] angular separation between target and Moon (conditional)
`MOONALT`	[deg] Moon altitude at the site (conditional)
`MOONPHAS`	Moon illuminated fraction, 0–1 (conditional)

Observer and site¶

Keyword	Meaning
`OBSERVER`	observer name (conditional)
`SITELAT`	[deg] site latitude (conditional)
`SITELONG`	[deg] site longitude, east positive (conditional)
`SITEELEV`	[m] site elevation (conditional)

Software¶

Keyword	Meaning
`SOFTWARE`	acquisition software and version
`SWCREATE`, `CREATOR`	the same, under their conventional aliases
`ANNOTATE`	session annotation (conditional)

Frame quality¶

Measured on the green plane and written per frame when the metric is available.

Keyword	Meaning
`HFD`	[px] half-flux diameter, at the subsampled scale
`FWHM`	[px] mean star FWHM on the green plane
`NSTARS`	detected star count
`SKYLEVEL`	[ADU] median sky background
`ECCENTR`	mean star eccentricity, 0 = round

Important

HFD and FWHM are both in green-plane pixels, which are twice the size of raw sensor pixels. To convert to arcseconds, multiply by 2 and then by the full-resolution plate scale — 3.74″/px on an S30 Pro, so one green pixel is 7.48″. Using the full-resolution scale directly gives an answer a factor of two too small.

These describe the frame, not the photometry: they are the trends Review plots to tell a good hour from a bad one.

What is not in the header¶

The frames carry no astrometric solution

ARGOS writes no WCS. There is no CRVAL, no CRPIX, no CD matrix, no CTYPE, and no plate-scale keyword of any kind among the 65 written above. RA/DEC and OBJCTRA/OBJCTDEC record where the mount was pointing — they are not a solved position and cannot be used to convert pixels to coordinates.

The practical consequence for your reduction: every frame must be re-solved downstream. Siril will do it, and ASTAP is already on your machine. Budget for it, and do not assume a stacking script that expects a WCS will work on these files unmodified.

The solution ARGOS computes during a session lives in the session record and in the interface, not in the pixels you keep.

The Bayer pattern is not negotiable¶

The array on disk is the raw CFA mosaic with BAYERPAT = 'GRBG'. Any tool that reads it must debayer with that pattern, or the colours — and the photometry — are wrong. ARGOS’s own measurements never debayer at all: they read the green plane described in Differential photometry — method and equations.

Warning

If you open one of these frames and it looks grey and grainy at 1:1, nothing is broken — you are looking at an undebayered mosaic. Siril debayers on import when BAYERPAT is present; check that its Debayer setting is reading the header rather than a forced pattern.