Source detection and measurement#

Sources in the visit and deep coadd images are detected, deblended, and measured following (approximately) the process in Bosch et al. (2018), as described in Citation: Rubin Observatory Science Pipelines Developers (2025); The LSST Science Pipelines Software: Optical Survey Pipeline Reduction and Analysis Environment https://doi.org/10.71929/rubin/2570545 [BibTeX]. All measurements are stored in the catalog Data products.

Key terminology:

source: a detection or measurement in a single processed visit image
object: a detection or measurement in a deep coadd image

Detection#

Detection refers specifically to the process of finding regions with above-threshold flux in images, with one or more peaks within them. These regions are called “footprints”. The threshold used for detection is a signal-to-noise ratio > 5.

If the footprint contains one or more flagged pixels, e.g., for cosmic rays, detector edge, bad pixels, or flux saturation, the source is also flagged (see Catalog flag columns).

Deblending#

After detection, footprints with multiple peaks are deblended into “children”, and the original blended footprint called the “parent”. The Scarlet Lite algorithm performs the deblending, as described in “The current state of scarlet and looking toward the future” (dmtn-194.lsst.io). The catalogs only include the deblended children.

Measurement#

A large variety of centroid, size, shape, and flux measurements are made on the deblended children. It is important to note that child fluxes may be sub-threshold (signal-to-noise ratio < 5), if their parent was above-threshold.

In visit images, measurements of sources assume a PSF or Gaussian shape. The results are stored in the Source catalog.

In deep coadd images, measurements of objects include both PSF and extended shapes. A wide variety of flux measurements are pre-calculated, such as the composite model (cModel), Gaussian-aperture-and-PSF (GaaP), and Sersic models. The results are stored in the Object catalog.

Star/galaxy classification#

A new floating-point, per-band and griz model_extendedness classifier using Sersic fluxes and sizes generally performs better than the previous refExtendedness. However, griz_model_extendedness tends to classify everything as a galaxy fainter than approximately i = 24; adjusting the threshold can help, but optimal faint-star selection will require user input. A classification for bad or bogus detections is still lacking.

Forced photometry#

In general, “forced” photometry refers to a measurement made at a fixed coordinate in an image, regardless of whether an above-threshold region was detected in that particular image.

Most of the photometry measurements in the Object catalog, made on deep coadd images, are forced photometry measurements, made at the centroid from the “reference” band. The reference band is chosen based on detection significance using the priority order of i, r, z, y, g, then u band.

When a photometry algorithm also involves fitting an ellipse, three different approaches are used:

In most elliptical galaxy photometry (e.g. Kron), the aperture is a multiple of the second-moment shape measured in the reference band.
In cModel galaxy fitting, the models are fit independently to each band, and the fit from the reference image is used for additional forced photometry in all bands.

Forced PSF photometry measurements are also made on all visit images and all difference images at the locations of all objects. The results are stored in the ForcedSource catalog.

There are a total of four variants of forced photometry in Rubin data release processing, combining two reference catalogs (Object and DiaObject) with two measurement images (visit images and difference images). Each of the two forced photometry catalogs includes rows for both measurement images and a single reference catalog: ForcedSource uses Object positions, while DiaForcedSource uses DiaObject positions.

Forced photometry on difference images is expected to behave best in crowded regions, since image subtraction removes neighbors more effectively than the coadd deblender (there is no deblending when producing the DiaSource catalog either). There is no deblending whatsoever in forced photometry on visit images.