FAQ for astronomers

Most questions about running Hebog and reading its products are answered on other pages, linked in the first section. The rest of this page answers what those pages do not: what the input and settings cannot do, and how to judge what a catalogue supports, with pointers to the source-finding literature.

Answered elsewhere

Question	Answer
Can I use Hebog for a survey catalogue now?	Scientific status
Which images can I give it, and what can I supply when the header lacks a value?	Supported inputs and what the finder reads
Does an accepted beam or mosaic guarantee valid measurements?	Input limitations
How should I mark blank or zero-padded pixels?	Invalid pixels
What do the detection and island thresholds mean?	Set detection thresholds
Should I use the `continuum` or the `compact` profile?	Choose a profile
Do I need Dask?	Use several cores or a Dask cluster
Is Hebog faster than PyBDSF?	Performance
How does Hebog differ from PyBDSF, Aegean and other finders?	Hebog and other source finders
How are the background and RMS estimated, and why does the RMS vary?	Estimate background and noise and the RMS image
What does an empty catalogue mean?	If the catalogue is empty
Will Hebog separate blends and recover diffuse emission?	Deblend and fit Gaussians and the known limitations
Is a source the same as an island, a Gaussian or a galaxy?	Three populations, not one and catalogue relationships
Which source flux should I use?	What the two source fluxes measure
Where does a source's position lie, and when does it have a shape?	Source fields and beam deconvolution
How are the errors computed, and do they allow for correlated noise? Is the RA error in seconds of time?	Point estimator and covariance and uncertainty status
Is the source mask a photometric aperture or a CLEAN mask?	The source mask
Why does a detection have no catalogue row?	Measurement dispositions
Are background, model or residual images published?	Publish products
What should I check before using a result as evidence?	Evaluation checklist
How do I re-run or reproduce a result?	Re-run or retry and provenance
How do I run PyBDSF and Aegean on the same images?	Use the notebooks

The input and the settings

Should I use a primary-beam-corrected image?

Choose the image for the quantity you want to measure, and record the choice. Hebog measures the one image it is given: it applies no primary-beam correction and cannot detect on one image while measuring on another, so fluxes from an apparent-sky image stay apparent. A primary-beam-corrected image is accepted, but its noise rises towards the edge of the beam; check the RMS image there. Hebog does not combine a corrected image with a flat-noise one.

Are 5σ, 3σ and seven pixels defaults?

No. SourceFinderConfig has no default detection threshold, island threshold or minimum island size; the tutorial's values are an example, not a recommendation. Only the profile (continuum) and the absent maximum island size have defaults. The minimum island size counts pixels, so what it means in beams depends on how finely the image samples the beam: a beam covers 1.13 FWHM² pixels, so seven pixels is about a quarter of a beam whose FWHM spans five. Choose the settings for your data and measure their effect.

Can I supply my own RMS or background map?

No. The request takes no external RMS or background image, and the profile fixes the noise grids. You can compare the RMS image Hebog publishes with your own.

Judging the catalogue

Does a 5σ threshold guarantee a reliable catalogue?

No. A threshold is a rule applied to an estimated noise field, not the probability that an entry is real. Correlated noise, imaging artefacts, a misestimated background and the number of independent beams searched all add false detections, and raising the threshold trades faint-source recovery for reliability by an amount you have to measure on representative data. In a challenge of eleven finders, the noise estimate and deblending changed catalogues more than the threshold did (Hopkins et al. 2015); Huynh et al. (2012) compare fixed and false-discovery-rate thresholds by completeness and reliability. Hebog has no false-discovery-rate threshold and gives no per-row probability that a detection is real.

What are completeness and reliability, and how should I measure them?

For a stated population and matching rule, completeness is the fraction of true sources recovered and reliability the fraction of reported detections that are true sources. Source-finding challenges measure both, together with position and flux errors, for compact and extended emission separately (Hopkins et al. 2015).

To measure them for Hebog, inject sources spanning flux, size, morphology, crowding and distance from the edge into representative noise, including beam-correlated noise, and keep independent cases for the final evaluation. Report recovery and residuals as functions of those properties. Image-plane injections test the finder on that image; they do not measure emission lost in imaging. A deeper catalogue helps, but a row it lacks is not necessarily false, because the reference has its own selection and resolution (Boyce et al. 2023 evaluate several finders this way on simulated and real images).

Why does Hebog disagree with PyBDSF at the same thresholds?

Equal threshold numbers do not make equal noise maps, deblending, fit admission, source grouping or photometry, so the catalogues differ even on one image. Before attributing a difference to accuracy, match the input image, frame, units and beam, and compare like populations, as the evaluation checklist describes. Look at missed and extra detections, one-to-many matches and the flux and position residuals, not only the row counts: two catalogues can have the same number of rows and different objects. Agreement with PyBDSF does not establish truth, and neither does a longer catalogue; the SKA Science Data Challenge 1 showed that detection and characterisation must both be judged across source types (Bonaldi et al. 2021).

Do the errors include everything?

No. The published errors are the statistical errors of each fit. They do not include the flux-scale calibration, primary-beam or imaging systematics, or the uncertainty of the background and noise estimates, and their own calibration across morphologies and signal-to-noise ratios is not yet complete. Add the systematic terms your analysis needs.

Beyond one image

Can I measure spectral indices, polarisation or light curves?

Not with Hebog alone. It analyses one Stokes I plane at one reference frequency and has no forced photometry or association across epochs or frequencies. A source missing from a run on another epoch or band is not a measurement of zero flux: to measure flux at known positions, use a finder with forced fitting, such as Aegean's priorized fitting (Hancock et al. 2018); see which finder to use.

What should I report in a paper?

Keep the exact Hebog version, the four products and the returned result record; the diagnostics' provenance identifies the input, configuration, science profile, implementation and any supplied metadata. State which population (SOURCES or GAUSSIAN_COMPONENTS) and which flux you used, your selection and matching rules, and the evidence for completeness, reliability and the errors. Releases make no compatibility promise, so a result is reproducible only with the version that made it.