LSST DESC Blending Task Force - Shared screen with speaker view
Unrecognized blends: Seems fairly straightforward to provide “hints” in the database — a few parameters that would indicate that it might be a blend. This would be a few parameters that capture clumpiness, asymmetry, and non-uniformity of colors. This lets you identify sources for post-processing or for eliminating from further analysis.
I have a question
I have a question for Peter: when you are fitting for the SED, can we also fit for the redshift at the same time that we perform deblending?
Looks like the algorithm imposes *hard* constraints that the objects have uniform color — i.e. wavelength-independent morphologies. This means that it will be impossible to have the de-blended objects have color gradients. Is that correct? Is there a way to soften this to allow modest color gradients?
Apologies…I have a conflicting meeting starting in 2 minutes, so have to sign off.
Not quite, Harry. Fred will talk about two-component models, which is the right way of accounting “color gradients”
After you find the most obvious objects do you go back with that model to refine the list of remaining detections?
@Chris: Not yet, but worthwhile discussing
have you tried this on a larger spiral and seen if you over-deblend when you have visible arms?
Presumably the monotonicity constraint would shred large spiral galaxies?
@Jon: Yes. But we could allow for azimuthal variation, as long as the radial profile is declining. I’m looking into that.
Does the algorithm produce uncertainties on fluxes or other quantities, e.g. by samples of some posterior?