gap in flux vs. flux_err distribution from segstats

At a slight guess, those bright objects are likely to be stars, and they tend to have a much higher mean S/N. To be sure try comparing to the surface brightness or size of the objects.

A

Just wondering- could this be something to do with variable stacking depth in the data?

A

I did some quick tests with made-up data, see code and plots below. All stars are the same here except for their magnitudes. Star parameters, skyRMS, gain and magnitude range are "inspired" by KiDS r-band data (although obviously the magnitude distribution of stars etc. isn't very realistic). The point is that these plots show the same strange behaviour (even more pronounced since less noisy) as the real data, so I'm now suspecting there's something odd going on with the flux_err estimation (rather than with my star selection). flux_err is dominated by flux_err_skyRMS; flux_err_sky is a few %-level effect while flux_err_shot is negligible. The same thing happens if I don't pass a sky, skyRMS, or gain to ProFound.



library(ProFit)
library(ProFound)

skyc=2
Nstar=100
loopno=30
imagedim=500

segstats=data.frame()
for(i in 1:loopno){
print(i)

starmodel=list(
moffat=list(
xcen=runif(Nstar, 0, imagedim),
ycen=runif(Nstar, 0, imagedim),
mag=seq(19, 23, length.out=Nstar),
fwhm=rep(3, Nstar),
con=rep(2.1, Nstar),
axrat=rep(1, Nstar),
ang=rep(0, Nstar),
box=rep(0, Nstar)
)
)

starim=profitMakeModel(modellist=starmodel, dim=rep(imagedim,2))$z
testim=starim+rnorm(length(starim), mean=0, sd=4e-12)
testseg=profoundProFound(testim, skycut=skyc, sky=0, skyRMS=4e-12, gain=1.6e13)

segstats=rbind(segstats, testseg$segstats)
}

magplot(segstats$mag, segstats$flux/segstats$flux_err, xlab='magnitude', ylab='S/N (flux/flux_err)', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc)
magplot(segstats$flux, segstats$flux_err, xlab='flux', ylab='flux_err', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc)

Thanks to those plots I think I've sussed out what you are seeing here. In short: there's nothing really wrong, rather it is a natural feature of the dilation process for PSFs.

The plots below tell the full story. As you say the skyRMS mostly dominates the error (as we would expect for all but the brightest sources). This is well predicted by N100, and N100 for stars (PSFs) is sensitive to the number of dilations before convergence is reached. For fainter PSF sources more dilations are required on average, meaning larger apertures and larger errors (naturally enough). Since the dilations are a discrete process this produces the visual artefacts you pointed out. I think that solves the mystery. They are to be expected basically, and I don't think there is reason to be too concerned.

magplot(segstats$flux, segstats$flux_err_skyRMS, xlab='flux', ylab='flux_err_skyRMS', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc, log='xy')
magplot(segstats$iter, segstats$flux_err_skyRMS, xlab='iter', ylab='flux_err_skyRMS', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc, log='xy')
magplot(segstats$N100, segstats$flux_err_skyRMS, xlab='N100', ylab='flux_err_skyRMS', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc, log='xy')

magplot(segstats$iter, segstats$N100, xlab='iter', ylab='N100', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc, log='xy')
magplot(segstats$iter, segstats$R50, xlab='iter', ylab='R50', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc, log='xy')
magplot(segstats$iter, segstats$R100, xlab='iter', ylab='R100', col=rgb(0,0,1,0.5), pch=19, cex=0.2, main=skyc, log='xy')

Is there anything else to be worried about on your side, or was it more about understanding the origin?

Cheers,

A

Ah OK, that's good to know, thanks. It was more about understanding the origin really, essentially I was worried there might be some weird bias in my star selection, so it's good to know that it comes from somewhere else and is understood. Since I'm mainly using R50 then I guess it won't matter too much to me.
It could also be an explanation for why I see a range of concentrations for stars of a given R50 (which is something else I've been wondering about): Basically, R90 depends on the iteration number while R50 doesn't, so for a given R50 there's going to be a range of con corresponding to the range of iteration numbers. R90 tends to go down with iteration number, so con goes up; so higher concentration means fainter stars on average (with higher iter), which is also what I observed. Does that sound about right?