Photometric quality of alert photometry

Science Alerts & Brokers
#1

This question came up during the TVS TDE subgroup telecon. We noticed the photometric quality of the alerts, in particular at the fainter end, is not yet where it should be.

For example this SN candidate: ANTARES

Or low-level variability in this bright quasar: ANTARES

We see ~0.3 mag intra-night variability, while these sources should have a constant flux within one night. This lack of photometric accuracy presents a challenge for photometric classification. We also noticed some issues with the alert astrometry (eg, the example quasar above shows some large offsets between filters).

I suspect these issues are already on the radar of the project. However I couldnā€™t find information about this at https://prompt-products.lsst.io or elsewhere. Iā€™m raising this question here to find out how we can learn more about the current efforts to improve the alert photometry and to see if we can help.

3 Likes
#2

Hi @sjoert, thanks for your message! Pointers to data quality issues are always extremely helpful.

We are working on performance reports now to be included in the DP2 and AP papers, so weā€™ll be able to provide more quantification soon. As you note, the first alerts data provide great test cases for photometric repeatability.

Looking at specific data points for the two examples, both have relatively low reliability scores, indicating that the image difference sources may not be well-subtracted. This will bias the psfFlux measurement (particularly in the quasar case, where the stamps show dipole structure). Our experience is that the single-frame direct image will have good photometric calibration.

You may wish to consider applying filters on the reliability score when selecting sources for analysis. We also are planning to add reliability statistics to the DiaObject to allow easier selection of well-subtracted lightcurves.

For hostless point sources, you may also wish to consider the scienceFlux measurement in the DiaSource or DiaForcedSource recordsā€“these are forced measurements on the direct image and so are less affected by poor image subtraction (so long as the DIA centroid is reliable).