Update BGS n(z) about desimodel

[dkirkby]

Copying a question from @moustakas to @amjsmith in desi-bgs:

Are these samples selected in the SDSS or DECaLS r-band? There's
about an 0.1-magnitude shift between the two filter curves, which will
affect the distribution and overall normalization (but perhaps not at
a level we care about).

Whatever the answer, the headers of these files should be updated to specify the filter used.

[amjsmith]

These samples are selected in the SDSS r-band. Below are attached new versions of these files where I've changed the file headers to specify that the filter is SDSS. I've also attached files where the samples are selected in the DECaLS r-band. Note that I had to change the file extensions from .dat to .txt to upload them here, and feel free to rename them.

For the DECaLS r-band selected samples, the integrals have increased to 866.9 and 699.4 per sq deg for the bgs_bright and bgs_faint samples, respectively.

nz_BGS_bright.txt
nz_BGS_bright_decals.txt
nz_BGS_faint.txt
nz_BGS_faint_decals.txt

[dkirkby]

Presumably DECam-r is what we want here, since that's what the imaging surveys use?

[amjsmith]

In desi-bgs, Shaun Cole said

The mocks on which these are based are nominally tuned to SDSS Petrosian r-band magnitude which differs both in pass band and magnitude definition from the DECaLS Tractor magnitudes.
I think we should stick to the stated (800 and 600) objects per sq deg.
Alex, could you remake the n(z) file cutting in magnitude so that you get these surface densities.
(I think when this catalogue is processed to DECaLS magnitudes we should again adjust the magnitude limits to hit these same surface densities. I know the BGS plan is written in terms of magnitude cuts but nowhere does it say what magnitude we are talking about. For now it is better to keep the surface density fixed as otherwise things will be hard to compare.)

I've attached new versions of the files, in the same format as before, but adjusted the magnitude limits so that the number densities are 800 and 600 per sq deg for the bright and faint samples. In the SDSS r-band, this corresponds to r < 19.5255 and 19.5255 < r < 19.9957.

nz_BGS_bright_800.txt
nz_BGS_faint_600.txt

[djschlegel]

Alex- The DESI imaging catalogs only have a single flux measure, so there's no choice to make there. SDSS Petrosian magnitudes were designed for the SDSS main sample (rPetro < 17.6), and won't be particularly well-behaved at much fainter magnitudes as they become an unstable parametric fit.

I find that r < 19.5 selects about 550 extended sources per sq deg in DECaLS DR5. But such a sample extends down to a fiber magnitude of 22, which will be impossible to get a redshift for. We've got to take that question up!

[moustakas]

This issue has remained open far too long and I've found myself with the need for a proper BGS dn/dz in desimodel. Unless there are any objections I'm going to check @amjsmith's most recent distributions into desimodel/data/targets as nz_bgs_bright.dat, nz_bgs_faint.dat and nz_bgs.dat (the sum of the two distributions), and remove the _BG.dat file as being obsolete.

This will likely break something in desisim but I'll fix that separately.

[moustakas]

OK, changes have been checked into the desimodel SVN, which probably means we need a new tag for 18.12. Do we want to update any of the other dn/dz files while we're at it?

from astropy.io.ascii import read
import matplotlib.pyplot as plt

faint = read('nz_bgs_faint.dat', names=('zmin', 'zmax', 'dndz'))
bright = read('nz_bgs_bright.dat', names=('zmin', 'zmax', 'dndz'))
sum = faint.copy()
sum['dndz'] = faint['dndz'] + bright['dndz']
sum.write('nz_bgs.dat', overwrite=True, format='ascii.basic')

fig, ax = plt.subplots()
zz = sum['zmin'] + (sum['zmax'] - sum['zmin']) / 2
ax.step(zz, sum['dndz'], label='BGS')
ax.step(zz, faint['dndz'], label='BGS Faint')
ax.step(zz, bright['dndz'], label='BGS Bright')
ax.legend(loc='upper right')
ax.set_xlabel('BGS Redshift')
ax.set_ylabel(r'BGS Target Density (deg$^{-2}$)')
ax.margins(0.02)

Note that I put in the following header comment by hand:
# Number of BGS galaxies (r_SDSS < 19.9957) per square degree from v0.0.4 of MXXL HOD mock.

[moustakas]

This will likely break something in desisim but I'll fix that separately.

Actually, it looks like desisim.targets.sample_nz doesn't read / use nz_BG.dat, so I think we're set.

@sbailey @dkirkby if all this looks OK please let's close this issue.

[sbailey]

Looks ok; closing. Re-open or re-post if new problems come up.

[sbailey]

With apologies for the spam, github isn't allowing me to close the issue. Trying again...