I am wandering if any discussion on how to estimate very faint objects’ magnitude error has been done. I have performed an exercise to retrieve a light curve of faint injected afterglow. However when using the forced photometry algorithm to estimate, let’s say, i-band, I have a very high error bars (in the detectable magnitude region) even if the detection is slightly over the detection limit for i-band.
To me this seems an issue with the concept of magnitudes more so than the LSST pipelines per se. LSST pipelines (forced) photometry measurements, for instance as listed in the DP0.2 schema, are commonly presented in columns providing fluxes rather than magnitudes. For cases where sigflux/flux < 1, one could imagine a straightforward procedure that translates the quoted flux and its quoted flux uncertainty from the pipeline into asymmetric magnitude uncertainties. Based on the size of the yellow errorbar (and assuming it represents +/- 1 sigma and not something else), the i-band case your plot shows is in a sigflux/flux >> 1 scenario, in which case to me it’d make more sense to plot the data point as some sort of limit, and I’m not sure I’d personally refer to it as a “detection”. Thanks…
