Handling deblended objects in catalogs

With @nchotard we discovered that when a source is deblended, the parent object remains in the catalog with apparently meaningful quantities attached to it. It means that when iterating over sources, one should be careful to reject objects with deblend_nChild != 0 otherwise there is some sort of double counting.
For such parent objects, we are wondering what is the meaning of quantities like modelfit_CModel_flux which seems to be defined for correctly shaped galaxy only ?
Have we missed something ?

Quantities such as centroids, shapes and model-fit fluxes are probably not very useful for parent objects, but it’s important to keep track of them in the catalog for the sake of deblending.

Yes, I agree that it is important to keep track of them, but I would expect meaningless values to be flagged (i.e. “modelfit_CModel_flag”)

We have a separate flag detect_isPrimary in the coadd processing that should be used in addition to any per-measurement flags. This will reject deblended parents and redundant objects in tract/patch overlap regions.

I recently filed an epic to improve the various astrometry-related flags. Would it be worth clarifying this as part of that epic?

On the other hand, are these sorts of details written down anywhere? The interaction of our various flags seems impossible to understand with out expert guidance.

This statement is rather broader than I would be willing to agree with. We have regular scientists doing science with HSC using essentially the same set of flags. Maybe we’ve got a bit of extra documentation (e.g., our pipeline description document has a section on flags) that LSST could steal, but I don’t think regular scientists really look at that document. It seems to me that what happens is that people with knowledge of the pipeline suggest some flags that people should care about, and then a scientist willing to experiment uses that information and suggests a “canonical set” of flags, and then everyone else follow suit. I think LSST isn’t processing science data in earnest, and so hasn’t gone through all of that. Maybe that kind of operation isn’t real “understanding”, but certainly regular scientists are able to use the flags with a bit of guidance.

The SDSS page is at http://www.sdss.org/dr12/tutorials/flags.

A couple of links down you’ll find my detailed document: http://www.astro.princeton.edu/~rhl/flags.html

Is there a one-to-one correspondence between LSST flags and SDSS flags, including their meaning?

We’ll have an equivalent page listing all flags along with tutorials looking at how filtering against different flags influences an analysis given real data.

DM-6655 is the ticket for this that @parejkoj originally created.

No, but some are cognate.