Starkiller: Removing stars and satellites from IFU data

Starkiller removes stellar contamination from integral field unit (IFU) spectroscopic datacubes. It works by:

1. Fetching a source catalogue (Gaia DR3 by default) for the observed field
2. Extracting PSF-photometry spectra for each star in the cube
3. Matching each spectrum to a stellar model (CK or ESO library) accounting for interstellar extinction
4. Normalising the best-matching model to catalogue photometry (Gaia G by default)
5. Injecting each modelled star into a synthetic scene cube using the measured PSF
6. Applying a secondary flux correction to calibration sources
7. Subtracting the synthetic scene from the input cube to produce a differenced FITS file

PSFs can be trailed (for non-sidereal tracking), Moffat or Gaussian, and can be constructed from the data itself. Satellite streaks can also be detected and removed.

---

Installation

pip install git+https://github.com/CheerfulUser/starkiller.git

Requirements: Python ≥ 3.8, NumPy < 2.0 (the pysynphot and extinction dependencies are compiled against NumPy 1.x and will fail to import under NumPy 2.x).

---

Usage

Non-siderially tracked data (e.g. comets, asteroids)

from starkiller import starkiller

ifu = starkiller(
    file='path/to/cube.fits',
    cal_maglim=20,    # brightest mag limit for PSF calibration stars
    run=True,
)

Siderially tracked data

from starkiller import starkiller

ifu = starkiller(
    file='path/to/cube.fits',
    trail=False,           # no trailing — sidereal tracking
    numcores=7,
    savepath='output/',
    spec_catalog='ck+',    # extended CK model grid (recommended for faint stars)
    psf_profile='gaussian',
    cal_maglim=18,
    model_maglim=25,
    psf_preference='data', # build PSF from bright calibration stars in the cube
)

Siderially tracked data with an extended source (e.g. nebula, galaxy)

Set fuzzy=True to detect and mask any extended emission before PSF fitting, preventing it from contaminating the stellar models.

from starkiller import starkiller

ifu = starkiller(
    file='path/to/cube.fits',
    trail=False,
    fuzzy=True,            # mask extended emission (nebula, galaxy) before PSF fitting
    psf_preference='data',
    spec_catalog='ck+',
)

Satellite streak removal

from starkiller import starkiller

ifu = starkiller(
    file='path/to/cube.fits',
    spec_catalog='ck',
    savepath='output/',
    satellite=True,        # detect and model satellite streaks
)

The satellite spectrum is modelled as a highly-extincted solar spectrum and subtracted, recovering the underlying source.

---

Key parameters

Parameter	Default	Description
trail	True	Estimate and model trailing angle. Set False for sidereal data.
cal_maglim	18	Faint magnitude limit for PSF calibration stars.
model_maglim	25	Faint magnitude limit for stars included in the scene model.
spec_catalog	'ck'	Stellar spectral library: 'ck' (Castelli-Kurucz), 'ck+' (extended), 'eso'.
psf_profile	'gaussian'	PSF profile shape: 'gaussian' or 'moffat'.
psf_preference	'data'	Build PSF from 'data' (calibration stars in cube) or 'model' (analytic).
fuzzy	False	Mask extended sources (nebulae, galaxies) before PSF fitting.
satellite	False	Detect and subtract satellite streaks.
numcores	5	Number of CPU cores for parallel spectral fitting.
wavelength_sol	'air'	Wavelength frame: 'air' or 'vacuum'.
background	False	If True, subtract a 2D sky background before modelling.

---

Accessing results

After a run the starkiller object exposes the key products as attributes:

ifu = starkiller('cube.fits', run=True)

ifu.diff_cube    # differenced cube (stars removed)
ifu.scene        # synthetic star scene that was subtracted
ifu.ebvs         # E(B-V) extinction values per star
ifu.cat          # source catalogue used (pandas DataFrame)
ifu.psf          # fitted PSF object

Extinction values can be used directly for further analysis:

---

Output FITS file

The differenced cube is saved as a FITS file with four extensions:

Extension	Contents
0	Primary HDU (copy of input header)
1	Differenced cube (input − scene)
2	Error cube
3	Binary table of fitted PSF parameters

The WCS from the input cube is preserved in all image extensions.

---

2D image reduction

For single broad-band images (rather than IFU cubes), use starkiller_image:

from starkiller import starkiller_image

img = starkiller_image(
    image=data_array,
    wcs=wcs_object,
    ref_filter='r',
    cal_maglim=[16, 20],
    model_maglim=25,
    run=True,
)

---

Papers

Starkiller was developed for the MUSE 2I/Borisov dataset to recover observations that would otherwise have been unusable due to stellar contamination. See Deam et al. 2026 for details.

The starkiller method is described in Ridden-Harper et al. 2025 (The Astronomical Journal, 169, 148).

If you use starkiller, please cite the Zenodo release using the DOI badge at the top of this page.

---

Current limitations

• Simultaneous PSF fitting for overlapping/crowded sources is not yet implemented.
• Very long trails that extend beyond the detector edges may cause the WCS matching and PSF fitting to struggle. Setting fuzzy=True can help in these cases.