Maximum Likelihood Source Detection Breakout Session

The first I heard of maximum likelihood source detection was from @jbosch during the LSST DESC School in February 2015, but he tells me that the idea goes back well before that and was fleshed out to a considerable degree by Nick Kaiser as a part of the PanStars survey (a couple of technical reports were written by him in the early 00’s). Despite its promise I don’t think that maximum likelihood source detection was ever implemented, perhaps due to other unexpected project demands.

Some of the potential advantages of such an algorithm (e.g. easy streaming co-addition, optimal detection of extended low surface brightness objects, treatment of spatially variable and correlated noise, etc.) make it an attractive option for LSST. Such a method has implications for groups concerned with the detection of transient solar system objects, strong lenses, other extended objects with low surface brightness, and perhaps more broadly extended to general object detection.

I think now might be a good time to have a workshop session on the practical applications of a maximum likelihood based detection process with LSST data. @connolly and his student @jbkalmbach have apparently been using such a method to good success, as have @mdschneider and I on a separate project. So hopefully we can get together and have some detailed discussions on practical experiences and whether/how we might implement such a method with the LSST survey, as well as engage other LSST users to see if such a method would benefit their science. It would also be good to hear from DM whether such method is planned, and if so to what extent.

I believe those techniques were implemented in Nick’s imcat software, which was used to produce the input photometric catalogs for DEEP2. We had to do a lot of work to fix the deblending in those catalogs. . .

We’re definitely planning to do maximum likelihood detection in terms of coadding PSF-correlated images. We have not yet decided to what extent we’ll try to use optimal spatial filters for extended sources, or use anything beyond independent detection in multiple bands for the spectral dimension. But we’re considering those options.

I do think this is an area where there are relatively easy experiments that science collaborations could do that would inform DM choices. DM would ultimately do at least some of those experiments itself if science collaborations don’t, but science collaboration work in this area is a way to guarantee that some detailed requirements for specific science cases are represented, and it might get answers faster than if DM was left to its own devices. Seems like a good topic for a breakout.

FYI unfortunately I won’t be able to make it to the meeting. My family is currently moving to Livermore. Please keep me posted on any relevant discussions.