Apparent poor extraction? LSF? from collimator dip variation

[AlexGKim]

coadd-9-80666-20210130.fits 39627824398603495 coadd-9-80666-20210116.fits

Two spectra of the same object look very different though maybe due to fiber misplacement since the target is an extended spiral galaxy.

If the emission line is real, it is poorly extracted as if the LSF is misestimated.

[AlexGKim]

Same problem in

coadd-1-80666-20210130.fits 39627836448836633

and others in this exposure and other exposures

80636 9 20210130 39628398565264831 80636 9 20210130 39628398565266270 80636 5 20210130 39628430932708590 80636 6 20210130 39628430945290012

[sbailey]

Thanks @AlexGKim . daily, cascades, or denali production? That first plot looks like an unmasked cosmic (or confirming: do you plot mask != 0 or ivar=0 wavelengths?), but I haven't tracked back to raw data to confirm.

[AlexGKim]

denali production. I confirm that the feature exists getting rid of mask != 0 and ivar=0 wavelengths.

[julienguy]

It's due to a combination of 3 effects: 1) The first night was during bright time with a very high sky continuum from the moonlight, (https://nightwatch.desi.lbl.gov/20210130/00073756/qa-summary-00073756.html) 2) There were no flat field exposures that day so the pipeline used a default flatfield from Dec 2020 (check keyword FIBERFLT in cframe file header). 3) There is the throughput dip at 4300A which varies in shape and amplitude with the fiber number (because the incidence angle of light on the collimator mirror varies with fiber number).

The variations of 3) from fiber to fiber is not well corrected because of 2) and this has a large impact on the target spectrum because of 1) (the sky subtraction works only if the flat field is correct).

So it's not something we can fix with the pipeline, and it's an example to demonstrate we still need daily calibrations. But I don't understand why there is such a variation of transmission between Dec and Jan ...

[julienguy]

Obtained with

from desispec.io import read_fiberflat
flat1=read_fiberflat("/global/cfs/cdirs/desi/spectro/redux/denali/calibnight/20201214/fiberflatnight-b9-20201214.fits")
flat2=read_fiberflat("/global/cfs/cdirs/desi/spectro/redux/denali/calibnight/20210131/fiberflatnight-b9-20210131.fits")
plt.plot(flat1.wave,flat1.fiberflat[480]/flat1.fiberflat[250],label="ratio of spectra from fiber 480 and 250 of b9 on 20201214")
plt.plot(flat2.wave,flat2.fiberflat[480]/flat2.fiberflat[250],label="ratio of spectra from fiber 480 and 250 of b9 on 20210131")

[AlexGKim]

There are may of these extraction issues for 1/30/2021

This may be related. denali coadd-9-80637-20210130.fits 39627773307786246

coadd-2-80637-20210130.fits 39627791385239498 39627797433426465

same coadd pedal 5 39627815573782717

coadd-9-80666-20210130.fits 39627842513800662

[sbailey]

@akremin has demonstrated that this is strongly correlated with humidity, but the correction is tricky because both the shape and wavelength is changing. It's work in progress but unlikely to converge before Fuji, so I'm removing it from the Fuji project required list.