Alert plans for fields with multiple visits in a night

A random thought that occurred to me at the AAS meeting: LSST will often observe a field multiple times in a night, either as part of an asteroid pair or a deep drilling sequence. Is there any plan to co-add images taken on the same night and run them through the image difference pipeline? Seems like one could detect things like supernova earlier using co-adds. Or is this something that the community could build as a level 3 product?

I don’t think that there are any plans. You might want to coadd the difference images.

Zeljko also pointed out that one could do it at the catalog level and if there are multiple 2-sigma detections at the same spot in a night it could be promoted to higher significance detection.

Presumably every object that differs sufficiently from the template will be picked up and sent out as an alert, with the relevant postage stamps, so for this subset at least it could be analyzed in a broker, giving a brightness vs time for the night. An interesting addition to the broker burden.

Right, but the problem is in the “differs sufficiently”. There will be transients below the detection threshold in individual visits, but could be detected if all the images in a night were co-added.

Indeed. And that indirectly brings up a divergence of opinion. Some believe that it is necessary for LSST to define its performance with respect to the faintest detectable targets - hence, as you point out, multiple visits get deeper! Others feel that brighter and easily detected targets will be the most productive and it’s OK to benchmark that. For me the attraction of punting to the broker is that the broker problem is still in definition and we can casually toss more requirements on, whereas LSST DM is now on a fixed budget and is obliged to resist growth in scope : )

I guess this does make an ideal case for a level 3 data product then. If someone wants to detect very faint transients, like supernova very early in their light curves, then they can go write the code to make nightly co-adds and find them and alert on them. But there’s no reason for DM to go do it, especially when there’s no guarantee anyone would use alerts on very faint objects.

Zeljko also pointed out that one could do it at the catalog level and if there are multiple 2-sigma detections at the same spot in

And that statement is true without doing nightly coadds right? Also, this should be possible over a larger time window (not just the same night) if you know what kind of object you are interested in, and in principle should increase detection rates but at higher computational costs.

Right, you wouldn’t need to make a coadd. But it would mean we have to save a large number of low-SNR detections from each difference image, or throw a ton of marginal detections onto the broker. It might be confusing if the majority of alerts are things we expect to be noise.

We are definitely not storing all low-SNR detections in the database. The DPDD says:

Any additional low-SNR measurements or coadditions to generate higher-SNR measurements would be additions to scope, as far as I understand things.

I agree with @ktl that this is not in scope as I’ve understood it. If we did want to add it, I think some form of coaddition would be more efficient than saving low-threshold detections and associating them in catalog space.