feat: add DES Y band (u band deferred pending sourcing decision)

[aboucaud]

Summary

• Adds the Y filter to surveycodex/data/DES.yaml, completing DES's grizY wide-survey filter set.
• Does not add a u filter — see "Why u was not added" below.

Values added (Y band)

Field	Value	Source
sky_brightness	17.96 mag/arcsec²	arXiv:1801.03181 (DES DR1), Table 1
psf_fwhm	0.90 arcsec	arXiv:1801.03181, Table 1
full_exposure_time	450 s (10 × 45 s)	arXiv:1801.03181, Section 2.2 ("cumulative exposure time of ... 450 s in Y"), footnote 9
zeropoint	25.56 mag	computed with speclite (decam2014-Y), matches scripts/check_zeropoints.py exactly
effective_wavelength	989.297 nm	computed with speclite (decam2014-Y), matches scripts/check_effective_wavelengths.py exactly

The full_exposure_time reference comment was updated to clarify that the existing 900 s value (10x90s) is specific to g/r/i/z and does not apply to Y.

Key finding: Y's exposure time is NOT 900s like g/r/i/z

This is the crux of the task. The DES wide-field survey used 90 s single exposures for g/r/i/z but only 45 s for Y, giving cumulative coadd exposure times of 900 s (griz) vs. 450 s (Y) during Y1-Y3 (the data covered by arXiv:1801.03181). A footnote in that paper notes Y's single-exposure time was only raised to 90s starting in Y4, while keeping the same 450s cumulative depth target — so even DES's own later data keeps the same Y cumulative value, just reached with fewer, longer exposures. Copying 900s from g/r/i/z would have been wrong.

Why u was not added

I dug into whether DES has a u-band survey configuration comparable to g/r/i/z/Y and found it does not:

• arXiv:1801.03181 (DES DR1) and arXiv:2101.05765 (DES DR2) — the papers describing the actual wide-field survey — have no u-band data at all. DES's wide survey is grizY only.
• The only DES-associated u-band imaging is described in arXiv:2012.12824 (Hartley et al. 2022, "DES Y3: Deep Field Optical + Near-Infrared Images and Catalogue"). That u-band data:
  • comes from ~15 separate community-time programs (different PIs, different programs — Table 2), not a homogeneous DES-run survey;
  • is restricted to the 10 SN deep pencil-beam fields + COSMOS, ~30 sq deg total vs. ~5000 sq deg for the wide survey;
  • has wildly heterogeneous per-field cumulative exposure times (~3,300 s-45,380 s depending on field/program — Table A1);
  • has no standardized single-epoch cadence: Table A1's "SE_TRUTH" (single-epoch-equivalent) row is blank (—) for u in every single field, unlike g/r/i/z which all have SE_TRUTH entries.

In short, there's no legitimate single value that would represent "DES u" the way full_exposure_time/sky_brightness/psf_fwhm represent g/r/i/z/Y — any such value would be either cherry-picked from one arbitrary community program/field, or fabricated. Per this repo's own rule ("Values without a source are not accepted"), and per this task's explicit goal of not repeating an earlier unverified-assumption mistake (copying a sibling filter's exposure time across a genuinely different observing configuration), I'm flagging this rather than guessing. Happy to add u with speclite-only zeropoint/effective_wavelength plus a clearly-caveated field-specific choice if maintainers want that tradeoff — I've left it out of this PR pending that decision.

Test plan

• pytest --cov — 36 passed
• pre-commit run --all-files — all hooks pass
• python -m surveycodex — DES now shows ['g', 'r', 'i', 'z', 'Y'], Y filter prints correctly
• surveycodex --refs — references table renders correctly, includes updated full_exposure_time comment
• python scripts/check_zeropoints.py — DES section shows 5 filters (g/r/i/z/Y), all matching speclite exactly (Y: 25.56 mag vs 25.56 mag)
• python scripts/check_effective_wavelengths.py — DES section shows 5 filters, all matching speclite exactly (Y: 989.30 nm vs 989.30 nm)

🤖 Generated with Claude Code