DESI Legacy Imaging Survey

• DESI Legacy Imaging Survey is a coordinated multi-telescope initiative providing deep, uniform photometry critical for DESI target selection and precision cosmology.
• It employs a dynamic observing strategy and a forward-modeling pipeline to achieve consistent image quality and robust source measurements despite variable conditions.
• The survey’s open data releases and innovative photometric techniques enable a wide range of applications, from galaxy evolution studies to Milky Way mapping.

The DESI Legacy Imaging Survey (LS) is a coordinated set of wide-field optical imaging surveys designed to provide deep, uniform photometric coverage for target selection and ancillary science in the Dark Energy Spectroscopic Instrument (DESI) experiment. Combining multiple telescopes and instruments, LS’s architecture, observing strategy, and data processing pipeline are engineered to enable both the reliability and flexibility required for precision cosmological surveys and a broad range of extragalactic and Galactic applications.

1. Survey Architecture and Scope

The DESI Legacy Imaging Surveys comprise three public partners:

• The Dark Energy Camera Legacy Survey (DECaLS), utilizing the DECam imager on the 4-m Blanco telescope at Cerro Tololo Inter-American Observatory,
• The Beijing-Arizona Sky Survey (BASS) with the 90Prime imager on the Bok 90-inch at Kitt Peak National Observatory,
• The Mayall z-band Legacy Survey (MzLS) with the Mosaic-3 camera on the 4-m Mayall telescope, also at Kitt Peak (Dey et al., 2018).

The resulting combined footprint is approximately 14,000 deg², targeting the extragalactic sky observable from the northern hemisphere. This area is divided typologically by the Galactic plane into the Northern and Southern Galactic Cap regions, ensuring avoidance of high extinction and star-crowded fields. Imaging is acquired in three optical bands: $g$, $r$, and $z$, with 5σ depth goals of $g=24.0$, $r=23.4$, and $z=22.5$ (AB magnitudes), referenced to an exponential-profile galaxy with $r_{1/2} = 0.45"$ (Dey et al., 2018).

2. Dynamic Observing Strategy and Depth Uniformity

To achieve nearly uniform depth over such a wide area—despite night-to-night variability in atmospheric seeing, transparency, airmass, and ambient sky brightness—the survey employs a dynamic observing strategy (Dey et al., 2018). Key components include:

• On-site, near-real-time analysis of each exposure to measure delivered image quality, transparency ($T$), pointing errors, and sky brightness (${m_\mathrm{sky}}$).
• The exposure time for each field is dynamically scaled on-the-fly:

$$f = T^{-2} \cdot 10^{0.8 k_i (X - 1)} \cdot 10^{0.8 A_i E_{B-V}} \cdot 10^{-0.4(\Delta m_\mathrm{sky})} \cdot \left(\frac{N_\mathrm{eff}}{N_{\mathrm{eff, fid}}}\right)$$

where $k_i$ and $A_i$ are extinction coefficients, $X$ is airmass, and $E_{B-V}$ is the Galactic reddening (Dey et al., 2018).

This algorithm “closes the loop” during nightly operations, autonomously updating scheduling and exposure times for all remaining fields to compensate for variable conditions. The end result is a map with minimized depth variations, offering advantages for target uniformity and statistical power in clustering analyses.

3. Data Products and Probabilistic Cataloging

LS delivers a suite of data products:

• Calibrated single-exposure images and coadded images (inverse-variance-weighted), divided into manageable "bricks".
• Ancillary data: bad pixel masks, inverse-variance maps, weight maps, and 5σ depth maps that quantitatively reflect survey uniformity.
• Source catalogs generated via an inference-based, forward-modeling pipeline dubbed “The Tractor” (Dey et al., 2018). For each candidate source, The Tractor performs a consistent parametric model fit across all bands, using morphological options (point source, exponential disk, de Vaucouleurs, composite) and convolving each with the appropriate PSF for every contributing exposure.

Crucially, The Tractor applies identical parametric models to all bands, yielding high-fidelity, joint constraints on position, flux, and shape. This methodology is critical for:

• Accurate source deblending in crowded or low-latitude fields,
• Deriving reliable photometry for both faint and blended sources,
• Enabling "forced photometry" in the lower-resolution, four-band WISE mid-IR imaging, extending cataloged fluxes to $W1$, $W2$, $W3$, and $W4$ bands for every LS optical source.

4. Public Data Releases, Software, and Community Engagement

LS commits to a rigorous open data policy, with two scheduled releases each year. Raw and pipeline-processed data are distributed promptly via the NOAO Science Archive (Dey et al., 2018). The entire data reduction and cataloging software stack (including “legacypipe” for full data processing and “The Tractor” for probabilistic source inference) is open-sourced on GitHub. This ensures reproducibility, fosters community scrutiny, and enables external reprocessing or custom analyses (Dey et al., 2018). Releases encompass all calibrated images, value-added products, and source catalogs, with comprehensive documentation.

5. Target Selection for DESI and Scientific Utility

The uniform, deep, and morphologically consistent photometry of the LS is fundamental for DESI target selection, affecting both galaxy and quasar programs (Dey et al., 2018). Reliable selection criteria depend crucially on both depth uniformity and robust source characterizations. DESI’s spectroscopic program will utilize source positions and photometry derived from LS to assemble samples for redshift measurement across $>14,000$ deg², enabling mapping of cosmic structure to precision levels better than 1%.

Secondary science applications of the LS imaging include:

• Study of galaxy evolution and large-scale structure,
• Milky Way stellar population and substructure mapping,
• Joint optical/mid-IR photometric redshifts, improved stellar mass estimates, and detailed source deblending (critical near bright stars or complex backgrounds) (Dey et al., 2018).

6. Methodological and Technological Innovations

The use of dynamic real-time feedback for observing represents a methodological advance in large-sky surveys. The forward-modeling, probabilistic catalog methodology (The Tractor) exemplifies the shift from conventional aperture/segmentation photometry to rigorous, PSF-convolved, multi-epoch fitting, yielding deeply uniform and physically self-consistent measurements. The strategy enables robust detection and measurement of both diffuse and compact sources under varying observing conditions.

The forced-photometry framework for transferring LS-derived optical models to WISE imaging allows the construction of contiguous, cross-band flux catalogs even for faint or blended sources, leveraging the high spatial resolution of the LS and the sensitivity of mid-IR imaging (Dey et al., 2018).

7. Broader Impact and Legacy

The LS dataset and methodology underpin both immediate cosmological applications and a broad class of astrophysical research. Its design principles—dynamically managed uniform depth, rigorous model-based photometry, and open code/data releases—anticipate and inform future large-scale survey operations. LS methodology is now a reference for current and next-generation wide-area imaging efforts.

The project’s open-access paradigm and flexible pipeline have enabled a spectrum of community-led investigations spanning precision cosmology, galaxy evolution, stellar characterization, and multi-wavelength cross-surveys. The LS thus constitutes a foundational modern imaging resource, with enduring scientific and methodological influence (Dey et al., 2018).