Sky Survey Schedule (SSS) Insights

• Sky Survey Schedule (SSS) is a systematic framework that organizes sky survey observations by integrating survey area, cadence, filter, and instrumental constraints.
• It employs a hierarchical and adaptive architecture, combining long-term strategies with short-term heuristic scheduling across multi-site and space-based surveys.
• SSS translates science requirements into actionable observation blocks, ensuring precise calibration, optimized dwell times, and robust simulation-driven validation.

Sky Survey Schedule (SSS) denotes the formal organization of sky-survey observations in angle, time, filter or frequency, and instrumental configuration. In the cited literature, the term is used both for the general scheduling problem that links survey area, depth, cadence, calibration, and operational constraints, and for a specific adaptive scheduling framework for multi-site optical networks (Spangelo et al., 2014, Yuan et al., 3 Sep 2025). Across radio, optical, and infrared surveys, an SSS is not merely a calendar: it is the mechanism by which science requirements are translated into tiles, blocks, epochs, visibility windows, and downstream data products (Lacy et al., 2019, Tereno et al., 2015).

1. Conceptual scope and architectural layers

A recurring pattern in the literature is hierarchical decomposition. For all-sky spacecraft surveys in low Earth orbit, one explicit three-layer structure consists of instrument and spacecraft configuration optimization, a high-level year-long strategy, and a short-horizon heuristic scheduler. In that formulation, the configuration layer guarantees geometric feasibility over the year, the annual layer allocates coverage goals across seasons, and the short-horizon layer fills gaps between preplanned pointings while enforcing Sun, Earth, and Moon constraints (Spangelo et al., 2014).

Distributed telescope arrays motivate a second hierarchy. A multilevel framework for time-domain large-area surveys separates a global scheduler, a site scheduler, and a telescope scheduler, with a distinct recording and visualization layer. The global scheduler allocates long-term observation opportunities across sites, the site scheduler refines them into per-telescope task queues under local conditions, and the lower layer executes and reports outcomes for subsequent replanning (Zhang et al., 2023).

A more recent named framework, explicitly called Sky Survey Schedule, generalizes this logic to coordinated multi-site operations for space-object cataloging and astronomical survey work. Its architecture comprises a Target Ephemeris Calculation subsystem, a Target Weight Adjustment subsystem, and a Mission Scheduler subsystem, tied together by a database-centric design. It supports two principal observation modes—fixed sky regions and target-centered tracking—and uses HEALPix-backed sky partitioning plus adaptive weighting to combine operator priorities, observation history, and environmental constraints into space-time priority matrices (Yuan et al., 3 Sep 2025).

These formulations converge on the same principle: scheduling is separated into geometry, long-baseline survey policy, and short-baseline execution. This suggests that SSS is best understood as a control stack rather than a single optimization pass.

2. Governing quantities: area, depth, cadence, and constraint sets

The dominant scheduling variables are survey footprint, per-visit depth, revisit cadence, available observing time, and the constraint set imposed by the platform. The Karl G. Jansky Very Large Array Sky Survey (VLASS) illustrates how tightly these are coupled. VLASS covers $33{,}885\ \mathrm{deg}^2$ at declination $\delta > -40^\circ$, is executed in three epochs, and ties its $\approx 32$-month cadence directly to the VLA’s 16-month configuration cycle because the required $\approx 2.5''$ resolution demands B and BnA configurations. Its schedule is therefore set jointly by sky area, array configuration, and on-the-fly mosaicking efficiency rather than by an arbitrary temporal spacing (Lacy et al., 2019).

Space-based wide surveys show the same coupling in different form. Euclid translates cosmological requirements into a six-year step-and-stare survey over $\sim 15{,}000\ \mathrm{deg}^2$ of extragalactic sky, with $\approx 20\%$ of observing time allocated to calibrations, a pre-tessellation into $\sim 50{,}000$ fields, and a fixed elementary observation pattern of four dithered frames per field. Its schedule is bounded by solar-aspect and tilt constraints, calibration blocks that may last up to about one week, and sky masks defined by zodiacal light, extinction, and bright stars (Tereno et al., 2015).

Roman’s proposed all-sky near-infrared program emphasizes a different balance. The white paper frames the first epoch as essentially single-visit per field, because Roman does not aim for LSST-style 100-visit light curves across tens of thousands of square degrees. Instead, it prioritizes an early first epoch so that later Roman, Gaia, LSST, and Euclid data create long astrometric baselines. The survey is partitioned into LSST high-latitude sky, non-LSST high latitude, and the low-latitude Galactic plane or bulge, with F158 used at $|b|>20^\circ$ and F146 at $|b|<20^\circ$ (Han et al., 24 Feb 2026).

Pointed survey missions introduce an additional temporal dimension: the distribution of observation times themselves becomes a science metric. For CSST, a 10-year mission with about 7 survey years and a main sky area of $\approx 17{,}500\ \mathrm{deg}^2$, the first year covers about $\delta > -40^\circ$0 in at least one band, with subsequent years filling missing bands and revisits. The analysis shows that heterogeneity and concentration in observation times directly affect astrometric performance, so the schedule must be judged not only by visit count but also by the temporal dispersion of visits (Fu et al., 2023).

3. Tiling, survey speed, and cadence formalisms

An SSS must convert high-level requirements into tiles, blocks, and dwell-time relations. VLASS partitions the sky into 899 tiles of roughly $\delta > -40^\circ$1, grouped into 4–8 hr scheduling blocks, and uses on-the-fly mosaicking so that antennas sweep continuously rather than stop at discrete pointings. The row separation is $\delta > -40^\circ$2, the nominal scan speed is $\delta > -40^\circ$3, the phase center updates every $\delta > -40^\circ$4, and the correlator dump time is $\delta > -40^\circ$5. Those values set the integration time per beam and therefore the achievable per-epoch sensitivity (Lacy et al., 2019).

Roman makes the time–area relation explicit. For the high-latitude F158 survey, it adopts 3 $\delta > -40^\circ$6 85 s exposures with an HLWAS-like 3-point dither, overheads of $\delta > -40^\circ$7, and a first-epoch survey speed of

$\delta > -40^\circ$8

so that

$\delta > -40^\circ$9

For the low-latitude F146 survey, $\approx 32$0 at $\approx 32$1 requires 67 days with a 2-point dither, implying an effective speed of $\approx 32$2 (Han et al., 24 Feb 2026).

All-sky LEO surveys introduce orbit-driven pointing limits. For the SPHEREx observing-scenario strategy, Sun avoidance fixes the cross-track tilt $\approx 32$3, Earth avoidance then limits the in-track nod $\approx 32$4, and the maximum contiguous pointing duration becomes

$\approx 32$5

with $\approx 32$6 the orbital period. For the stated SPHEREx parameters, $\approx 32$7 varies from about 9 to 19 minutes across the year as the solar $\approx 32$8 angle changes, and that seasonal variation determines how many pointings per orbit are feasible (Spangelo et al., 2014).

Ground-based wide-field surveys often solve the same problem through tiling geometry and dithering. WFST divides the sky into rectangular tiles tied to the focal-plane footprint, groups approximately 50 connected tiles into one observing block, and uses a greedy scheduler built from sky background, altitude, angular distance, historical observations, and recent-repeat penalties. To mitigate chip gaps and the unusable corners of the mosaic CCD layout, it optimizes an eight-visit dithering pattern and reports a uniformity metric of 0.974, with only 2.183% of sky area observed fewer than six times (Chen et al., 2023).

4. Optimization paradigms

The scheduling literature spans deterministic sequencing, greedy heuristics, integer programming, simulated annealing, and evolutionary optimization.

Euclid exemplifies a hybrid of rule-based and stochastic design. Its reference survey is produced deterministically by pre-scheduling calibration blocks, filling the remaining science windows with contiguous field sequences, progressing mostly up and down ecliptic meridians within horizontal bands, and minimizing large slews. The same problem is also cast as a combinatorial optimization problem in which simulated annealing perturbs field sequences or patches and accepts changes according to a temperature-controlled cost function (Tereno et al., 2015).

ZTF provides a fully explicit integer-linear formulation for time-domain imaging. The night is discretized into 30-minute blocks, exposures are assigned to blocks and filters via an ILP, and within-block ordering is then refined by a Traveling Salesman Problem solve to minimize slews. The science objective is volumetric: each exposure is weighted by

$\approx 32$9

where $\approx 2.5''$0 is the predicted limiting magnitude. The ILP maximizes the sum of $\approx 2.5''$1 over the night while penalizing filter changes, thereby enforcing strict per-field exposure counts and cadence patterns (Bellm et al., 2019).

ALTSched advances a top-down alternative to greedy next-best-pointing strategies. Instead of maximizing a scalar merit function at every step, it makes global nightly decisions about which half of the sky to observe and which filter sequence to use, then scans along the meridian to maximize signal-to-noise ratio. For LSST, that structure yields a sharply peaked $\approx 2.5''$2-day inter-night revisit distribution, rotating tilings for uniformity, and substantially improved time-domain metrics relative to the baseline greedy scheduler, while keeping effective survey time comparable (Rothchild et al., 2019).

Transient-targeted surveys have motivated explicitly probabilistic schedulers. For CHASE, transient occurrences are modeled as Poisson processes, and the schedule is scored by the expected number of detections younger than a specified age: $\approx 2.5''$3 where $\approx 2.5''$4 is the event rate, $\approx 2.5''$5 the revisit interval, $\approx 2.5''$6 the time after explosion when the event becomes detectable, and $\approx 2.5''$7 the duration of detectability. A genetic scheduler, including NSGA-II for multi-objective cases, then maximizes this expectation under exposure and slew constraints (Förster et al., 2010).

Simulation-centered benchmarking has become a parallel track. DeepSurveySim defines schedules as ordered sequences of $\approx 2.5''$8 and evaluates them through benchmark reward functions. Its core per-observation quality proxy is the effective exposure time

$\approx 2.5''$9

which supports optimization problems centered on observation quality, target-of-interest coverage, or uniformity of image quality across sites (Voetberg et al., 2023).

Paradigm	Representative implementation	Characteristic control variable
Deterministic plus stochastic refinement	Euclid	Bands, calibration blocks, simulated-annealing patch order
Global optimization with local routing	ZTF	ILP on 30-minute blocks, then TSP slew minimization
Top-down cadence design	ALTSched	Nightly hemisphere choice, meridian scans, rotating tilings
Detection-probability optimization	CHASE	Expected detections $\sim 15{,}000\ \mathrm{deg}^2$0
Benchmark simulation for algorithm comparison	DeepSurveySim	$\sim 15{,}000\ \mathrm{deg}^2$1 and task-specific reward functions

5. Simulation, validation, and data-product coupling

A mature SSS is inseparable from simulation and downstream products. The LSST DESC DC2 simulated sky survey demonstrates this explicitly: it takes the minion_1016 OpSim output, restricts it to a $\sim 15{,}000\ \mathrm{deg}^2$2 WFD region and a $\sim 15{,}000\ \mathrm{deg}^2$3 deep-drilling field, applies translational and rotational dithers, and simulates five years of the planned ten-year survey. The visit list then drives catalog generation, image simulation, and Rubin pipeline processing, making cadence a first-class input to end-to-end cosmology validation (Collaboration et al., 2020).

AstroSkyFlow pushes this schedule-driven logic further by ingesting a chronological observing schedule, simulating observatory control system behavior such as slews and readout dead time, and then generating multi-epoch images with time-dependent source variability, atmosphere, and sensor response. In the reported comparisons, AstroSkyFlow reproduces noise characteristics and point spread function properties more accurately than SkyMaker and recovers injected photometric and motion signals such as exoplanet transits and asteroid trails. This establishes the schedule itself as the principal variable in validation and machine-learning dataset generation (Li et al., 1 Jun 2026).

For astrometric surveys, schedule quality can be quantified directly from the time series. CSST defines $\sim 15{,}000\ \mathrm{deg}^2$4 as the standard deviation of observation times per source and finds a global mean $\sim 15{,}000\ \mathrm{deg}^2$5 days. At $\sim 15{,}000\ \mathrm{deg}^2$6, sources with $\sim 15{,}000\ \mathrm{deg}^2$7 achieve $\sim 15{,}000\ \mathrm{deg}^2$8 mas/yr, $\sim 15{,}000\ \mathrm{deg}^2$9 mas/yr, and $\approx 20\%$0 mas, whereas sources with $\approx 20\%$1 yield $\approx 20\%$2 mas/yr, $\approx 20\%$3 mas/yr, and $\approx 20\%$4 mas. The paper explicitly attributes these differences to time-distribution heterogeneity in the schedule (Fu et al., 2023).

Data release strategy is likewise schedule-coupled. VLASS makes raw visibilities available immediately after observation, targets initial calibration products within $\approx 20\%$5 week and final versions within $\approx 20\%$6 months, produces Quick Look mosaics within $\approx 20\%$7 weeks, and then releases single-epoch and cumulative images and catalogs on longer timescales. Its three-epoch schedule therefore determines not only transient sensitivity and coadded depth, but also the sequence of public data products (Lacy et al., 2019).

6. Representative implementations, misconceptions, and broader implications

Representative surveys show that “Sky Survey Schedule” is not a single algorithmic doctrine but a family of strategies tied to platform physics and science priorities.

Survey or framework	Scheduling signature	Primary implication
VLASS	Three epochs, B/BnA-only, OTFM raster scans	Cadence set by array configuration cycle (Lacy et al., 2019)
Roman all-sky path	Phased GAS-funded expansion, single-visit first epoch	Long-baseline astrometry favored over dense cadence (Han et al., 24 Feb 2026)
Euclid	Step-and-stare, calibration-first time skeleton	Calibrations define the mission timeline (Tereno et al., 2015)
SPHEREx strategy	$\approx 20\%$8-season planning with $\approx 20\%$9	Orbit geometry sets pointing duration (Spangelo et al., 2014)
Adaptive SSS framework	HEALPix regions, dynamic target weights, Focus/Attention modes	Closed-loop multi-site scheduling (Yuan et al., 3 Sep 2025)

Several misconceptions are corrected by these studies. First, an SSS is not simply an even cadence over a footprint. Roman explicitly argues for minimal cadence in the first epoch and instead exploits long baselines with Gaia, LSST, Euclid, and later Roman observations (Han et al., 24 Feb 2026). Second, uniform visit counts are not sufficient. CSST shows that clustered visits can severely degrade proper-motion and parallax accuracy even when the number of observations is adequate (Fu et al., 2023). Third, greedy next-observation scoring is not guaranteed to optimize survey-wide performance. ALTSched was introduced precisely because bottom-up greedy selection does not directly control global cadence structure, while ZTF demonstrates that blockwise integer programming can improve sequence completion and volumetric survey performance (Rothchild et al., 2019, Bellm et al., 2019).

At the same time, fully global optimization is not always necessary. The SPHEREx observing-scenario strategy emphasizes a constraint-based architecture that guarantees feasibility and uses a heuristic that always schedules the most under-covered feasible declination ring, while WFST shows that a comparatively simple greedy return metric can still produce acceptable annual coverage and dithering uniformity when the geometry is well structured (Spangelo et al., 2014, Chen et al., 2023).

Taken together, these works define SSS as the technical interface between science requirements and observing reality. Area, depth, cadence, overheads, orbit or mount geometry, calibration cadence, simulation infrastructure, and product-release policy are all schedule variables. A plausible implication is that future survey design will continue to move toward explicit co-design of scheduler, simulator, and downstream analysis pipeline, rather than treating observing cadence as an isolated operations problem.