Design for issue #50 (PyAutoMind prompt
ground_based_instruments_optional_noise_psf.md). Ground-based survey
imaging for colour context on lenses whose modeling data has no
optical counterpart (especially ALMA targets) — explicitly not
modeling inputs, which is why the RMS noise map and PSF are optional by
design.
DES, SDSS, HSC and their peers deliver pre-reduced coadds through
public cutout services. Reducing raw DECam/SDSS frames would be a large
new pipeline for products the survey collaborations already made better;
so this is a third adapter domain (cutout) beside imaging and
visibility: fetch + package, never reduce. The branch mirrors the
visibility pattern — its own small pipeline behind the shared registry
and the reduce_target domain dispatch.
| Adapter | Service | Bands | Scale | Noise | PSF |
|---|---|---|---|---|---|
legacy_surveys |
legacysurvey.org fits-cutout (DR10) |
griz (default grz) | 0.262" | yes — &invvar appends an inverse-variance HDU to the same request |
no |
sdss |
astroquery.sdss frames + WCS cutout |
ugriz (default gri) | 0.396" | no (phase 1) | no |
panstarrs |
STScI ps1filenames.py + fitscut.cgi |
grizy (default gri) | 0.25" | no (phase 1) | no |
DES coverage note: the DESI Legacy Imaging Surveys DR10 include the
DECam/DES footprint (and far more sky), so legacy_surveys is the DES
door — no separate DESDM service integration is needed.
Products per band: <out>/<survey>/<band>/data.fits (native scale,
survey WCS), noise_map.fits only where the service ships variance.
reduction.json carries a products_optional block stating what was
NOT produced and why — a survey cutout must never masquerade as a
modeling-ready al.Imaging dataset (no PSF, uncharacterised coadd
correlations).
- HSC PDR — cutout service is credential-gated (STARs account); deferred until a real need justifies the auth plumbing.
- unWISE + GALEX — served by the same Legacy viewer endpoint via
the
layer=parameter (unwise-neo7,galex); the cheapest multi-wavelength extension on the books — IR + UV colour for one smalllayerdial on the existing fetcher. - SDSS/PS1 variance — SDSS frames carry gain/dark-variance/sky
metadata and PS1 serves
.wtstack files; both are recorded follow-ups if anyone actually needs ground-based noise maps. - PSF — Legacy catalogs carry per-brick
psfsize_<band>; a Gaussian-FWHM approximate kernel (clearly flagged approximate in provenance) is the recorded follow-up if colour context ever needs even a rough PSF.
- The branch never imports the imaging stages; the only shared surfaces
are the adapter registry,
TargetSpec(survey_bandsdial, reusingcutout_shapefor size), andreduction.jsonprovenance. - Unit tests are numpy/astropy-only with fetchers monkeypatched; the
real-network demonstration lives in
prototypes/survey_cutouts_spike.py.