feat: add SDSS survey

[aboucaud]

Summary

Adds the Sloan Digital Sky Survey (SDSS) as surveycodex/data/SDSS.yaml: the 2.5m SDSS telescope at Apache Point Observatory with the SDSS imaging camera (Gunn et al. 1998), filters u, g, r, i, z (speclite's sdss2010 filter group).

Sourced values

Field	Value	Source
pixel_scale	0.396 arcsec/pixel	Xin et al. 2018 (arXiv:1805.02845), Sec 2: "0.396 arcsec per pixel". Matches Gunn et al. 1998's design value (rounded to 0.4″/24µm pixel, image scale 16.5″/mm).
gain	6.0 e-/ADU	Gunn et al. 1998 (arXiv:astro-ph/9809085), Sec 5: nominal conversion gain is 2 e-/ADU (u'), 5 e-/ADU (g'), 6 e-/ADU (r', i', z'). Used 6, the value shared by 3/5 bands incl. the r-band reference filter, as the single survey-level representative value.
mirror_diameter	2.5 m	Gunn et al. 1998, Sec 1.
obscuration	0.27	Gunn et al. 1998, Sec 3 (Eq. 2): "a 27% obscuration of incoming flux by the secondary mirror and the light baffles".
zeropoint_airmass	1.3	speclite docs: sdss2010 filters use a reference APO atmosphere at airmass 1.3.
sky_brightness (u,g,r,i,z)	22.01, 21.84, 20.84, 20.16, 18.96 mag/arcsec²	SDSS DR17 imaging data-quality page, median (50th pct) per-filter sky brightness for moon-down observations.
full_exposure_time	54.1 s (all bands)	Xin et al. 2018, Sec 2: single-pass TDI drift-scan exposure time. Consistent with Gunn et al. 1998's ~54s figure.
psf_fwhm (u,g,r,i,z)	1.47, 1.39, 1.32, 1.27, 1.23 arcsec	Derived from the official SDSS median r-band seeing of 1.32″ (SDSS DR17 page above), scaled per band via the Kolmogorov relation FWHM ∝ λ^-0.2 (Xin et al. 2018, Eq. 7) using speclite's effective wavelengths — that paper shows this approximates SDSS's measured wavelength dependence of seeing.
zeropoint, effective_wavelength	computed	Computed with speclite's sdss2010-{u,g,r,i,z} filters (see scripts/check_zeropoints.py / check_effective_wavelengths.py).

Judgment calls called out:

• gain and psf_fwhm vary per band in the primary sources but the schema stores one scalar survey-level gain; I picked the value shared by the majority of bands (r/i/z = 6 e-/ADU) rather than averaging.
• No single published table gives per-band SDSS seeing (psf_fwhm); I derived it from an official median r-band anchor plus a peer-reviewed wavelength-scaling relation rather than guessing, and documented the derivation explicitly in the YAML references comment (same pattern already used for LSST.yaml's psf_fwhm).

Validation methodology

Before trusting the speclite-based zeropoint/effective-wavelength computation for SDSS, reproduced DES's already-shipped g-band values using the exact formula in scripts/check_zeropoints.py (compute_zeropoint_mag with exposure_time=1s, effective_area from DES's own mirror_diameter/obscuration) and confirmed an exact match (zeropoint=26.7239≈26.72, effective_wavelength=489.0037≈489.004) before applying the same method to SDSS.

Test plan

• pytest --cov — 36 passed
• pre-commit run --all-files — all hooks pass
• python -m surveycodex — SDSS prints correctly
• surveycodex --refs — SDSS references print correctly
• python scripts/check_zeropoints.py — speclite vs surveycodex match exactly for all 5 SDSS filters
• python scripts/check_effective_wavelengths.py — speclite vs surveycodex match exactly for all 5 SDSS filters
• CI checks green (see PR checks)

🤖 Generated with Claude Code

[aboucaud]

Addressed code review findings:

• psf_fwhm was derived using the theoretical Kolmogorov exponent (α=-0.2) with a comment claiming Xin et al. 2018 supports it — the paper actually finds the data "clearly deviate" from α=-0.2 and favors α≈-0.3 (von Kármán model), and separately flags the z-band as not following the power law as well as u/g/r/i. Recomputed with α=-0.3 (u: 1.47→1.55, g: 1.39→1.43, i: 1.27→1.25, z: 1.23→1.18; r unchanged as the empirical anchor) and rewrote the comment to accurately reflect the paper's finding, with an explicit z-band uncertainty caveat.
• Fixed two mis-cited section numbers (values were already correct, only the locators were wrong): gain's "Sec 5" → "Sec 7" (Gunn et al. 1998), pixel_scale/full_exposure_time's "Sec 2" → "Sec 1" (Xin et al. 2018).

All validation re-run and clean (pytest, pre-commit, speclite agreement).