exclude W2 from the set of bands fit with fastphot

[moustakas]

In the fits to the broadband photometry (using fastphot) in Denali, the chi-squared distribution for LRGs had a very large tail. By comparison, the chi-squared distribution for ELGs and other target classes were roughly symmetric and centered on ~1 (see below).

I suspect the issue is hot dust emission from AGN in a fraction of the LRG sample, which boosts the (observed-frame) W2 flux in a way that's not capture by the "stars-only" population synthesis models used in fastphot. This hypothesis is a little hard to test because we don't write out the W2 photometry synthesized from the best-fitting model, although I did visually inspect a few dozen model fits with large chi-squared values.

So in Everest I plan to exclude W2 from the list of bands fitted and we can revisit in Fuji.

[moustakas]

Or maybe it's just QSO contaminants in the target selection? Here's one example fit to an LRG target (in SV1) where W2 is included-- https://data.desi.lbl.gov/desi/users/ioannis/fastspecfit/everest/sv1/tiles/cumulative/80605/lrg/sv1-lrg-80605-39627658597767397-cumulative.html

And here's the proposed changed with W2 excluded from the fit (on-the-fly, by temporarily setting the inverse variance to zero)--

fastphot /global/cfs/cdirs/desi/spectro/redux/everest/tiles/cumulative/80605/20210205/redrock-1-80605-thru20210205.fits -o ./fastphot.fits --targetids 39627658597767397 --specprod everest
fastspecfit-qa --fastphotfile ./fastphot.fits -o /global/cfs/cdirs/desi/users/ioannis/tmp2

Note that #22 is a related ticket.

[moustakas]

Done in https://github.com/desihub/fastspecfit/commit/47e4f76cad8e679461a2d882f03bfe92c5dfab60.