Survey Operations for the Dark Energy Spectroscopic Instrument

TL;DR

The paper documents the daily-operational framework of DESI, detailing how planning, field selection, and real-time observing control are coordinated to achieve its depth-first survey strategy over $14{,}000$ deg$^2$ and $0 < z < 3.5$. It presents the instrument architecture (5000 fibers across ten spectrographs on the Mayall telescope), the tile-based survey fields with a7-dark/4-bright tiling, and the nuanced strategies for airmass and slew optimization that drive survey speed. The analysis shows DESI achieved substantial progress in its first $1.1$ years—observing over $14$ million galaxies and $4$ million stars—and aligns well with simulations and expectations, validating the operational model and the effectiveness of the real-time feedback loop via the merged target list (MTL). The work demonstrates that near-real-time reductions, QA, and adaptive planning can support a large-scale, multi-program spectroscopic survey with rigorous constraints on field overlap and depth while maintaining homogeneous data quality. The approach has practical impact for planning and executing future wide-field spectroscopic surveys with similar scale and complexity.

Abstract

The Dark Energy Spectroscopic Instrument (DESI) survey is a spectroscopic survey of tens of millions of galaxies at $0 < z < 3.5$ covering 14,000 square degrees of the sky. In its first 1.1 years of survey operations, it has observed more than 14 million galaxies and 4 million stars. We describe the processes that govern DESI's observations of the 15,000 fields composing the survey. This includes the planning of each night's observations in the afternoon; automatic selection of fields to observe during the night; real-time assessment of field completeness on the basis of observing conditions during each exposure; reduction, redshifting, and quality assurance of each field of targets in the morning following observation; and updates to the list of future targets to observe on the basis of these results. We also compare the performance of the survey with historical expectations and find good agreement. Simulations of the weather and of DESI observations using the real field-selection algorithm show good agreement with the actual observations. After accounting for major unplanned shutdowns, the dark time survey is progressing about 7% faster than forecast, which is good agreement given approximations made in the simulations.

Figures (5)

• Figure 1: Survey completeness on 2022--06--14, in the dark (top) and bright (bottom) programs. Green areas are completely finished, while white areas are unfinished. Areas not included in the footprint are in gray. Regions with $E(B-V) > 0.3$ are outlined by the solid contours. The dotted and dashed lines show the ecliptic and Galactic planes. The survey aims to start observations near $\delta = 0^\circ$ and build out. Notable deviations from that pattern are areas just above $\delta=30^\circ$, which are driven by needing to avoid strong winds from the south, and a region $50^\circ$ from the ecliptic in the bright program in the north, driven by moon avoidance.
• Figure 2: The fraction of the sky that is covered by a given number of tiles in the seven-pass dark tiling and the four-pass bright tiling. On average, a given part of the sky is covered by 5.2 dark tiles and 3.2 bright tiles.
• Figure 3: The number of exposures that can reach any particular point of the sky, for the seven-pass dark program, were no areas excluded (e.g., due to low Galactic latitude or low declination). The twelve star-like regions with with slightly lower coverage corresponds to the points of the underlying icosahedral tiling of Hardin:2000.
• Figure 4: The footprint of the DESI survey resulting from the constraints of §\ref{['sec:footprint']}. Tiles are colored by the amount of time it would take to reach a fixed intrinsic galaxy depth, relative to observing at zenith in the absence of Galactic extinction. This is $f_\mathrm{dust} f_\mathrm{airmass}$, from Equations \ref{['eq:dust']} and \ref{['eq:airmass']}. Airmasses are computed using the design airmasses resulting from the optimization of §\ref{['subsec:airmass']}. The Galactic plane is shown as a dotted gray line, and the gray contour shows $E(B-V) = 0.3$ mag. Tiles in extinguished regions and at the declination bounds of the survey are most expensive, owing to both atmospheric and Galactic extinction.
• Figure 5: Schematic flow chart of DESI operations steps, running from planning for the night, through each night's observations, through their reduction and updates to the MTL. Steps in the dashed box are optional and may be skipped temporarily if systems are not available. See §\ref{['subsec:dailyops']} for details.