7-Dimensional Sky Survey (7DS) Overview

• 7DS is an optical low-resolution spectrophotometric survey that employs densely sampled medium-band imaging to generate quasi-spectroscopic SEDs over wide fields.
• The survey employs a layered design—reference, wide-field cadence, and intensive monitoring—enabling efficient temporal sampling and robust photometric redshift estimations.
• It integrates precise inter-band calibration and GPU-accelerated data reduction pipelines, ensuring reliable flux measurements and rapid transient identification.

The 7-Dimensional Sky Survey (7DS) is the observing program conducted with the 7-Dimensional Telescope (7DT), a multi-telescope system designed to produce low-resolution spectral mapping over wide fields by repeatedly imaging the sky through densely spaced medium-band filters. In the literature, 7DS is presented as an optical low-resolution spectrophotometric survey that occupies a middle ground between broadband photometry and classical spectroscopy: it aims to recover quasi-spectroscopic spectral energy distributions for all sources in the field, while also serving time-domain objectives such as the identification of electromagnetic counterparts to gravitational-wave events. Its published science cases include kilonova searches, AGN and transient variability, large-scale structure, and improved photometric redshifts; later forecast studies place particular emphasis on its complementarity with SPHEREx in joint redshift estimation (Kim et al., 2024, Ko et al., 26 Sep 2025, Bae et al., 31 Dec 2025).

1. Instrumental basis and survey concept

7DT is the hardware platform and 7DS is the survey that exploits it. The commissioning description presents 7DT as a system of 20 planned telescopes, each a 50-cm-class reflector, with 12 units deployed and operational at El Sauce Observatory in Rio Hurtado Valley, Chile as of the 2024 paper. Each unit is a DeltaRho 500 corrected Cassegrain with a 508 mm primary mirror, a 286 mm secondary mirror, a focal length of 1537 mm, focal ratio $f/3$, and a 70 mm image circle. The detector is a Moravian C3-61000 PRO CMOS camera using a SONY IMX455 sensor of $9576 \times 6388$ pixels with 3.76 $\mu$m pixels, yielding a pixel scale of about $0.5$ arcsec/pixel and a field of view per telescope of $1.34^\circ \times 0.90^\circ$. The site is specified at latitude/longitude $30^\circ 28'16''$ S, $70^\circ 45'47''$ W, altitude 1600 m, with typical seeing of $\sim 1.5''$, $\sim 300$ clear nights per year, and zenith sky brightness of 21.97 mag arcsec$^{-2}$ (Kim et al., 2024).

The defining observational principle is “spectral mapping”: instead of obtaining a dispersed spectrum for one object at a time, 7DT/7DS images an entire field through many medium-width filters so that each source is represented by a densely sampled low-resolution SED. The commissioning paper describes a filter system with 40 medium-band filters of FWHM $9576 \times 6388$0 nm spanning 400 to 900 nm, with central wavelengths approximately 400 to 887.5 nm; in the then-current deployed set, 20 medium-band filters were in use with 25 nm spacing between central wavelengths. The redshift-forecast literature uses two closely related descriptions. One comparative SPHEREx+7DS study models an observing configuration with 20 installed filters, labeled $9576 \times 6388$1, spanning $9576 \times 6388$2 to $9576 \times 6388$3m with widths of about 25 nm and 12.5 nm spacing between adjacent filter centers, corresponding to $9576 \times 6388$4–60. Another forecast emphasizes a 40-band configuration with FWHM $9576 \times 6388$5 nm and $9576 \times 6388$6, describing the resulting data as quasi-spectroscopic “photo-spectra” (Kim et al., 2024, Ko et al., 26 Sep 2025, Bae et al., 31 Dec 2025).

A recurrent point in these papers is that 7DS is not merely a multicolor imaging survey. It is designed to sample continuum breaks and emission features densely enough that the data behave as low-resolution spectra over a large field, thereby enabling classification and redshift inference that are substantially more constrained than conventional broadband photometry.

2. Layered survey design and published configurations

The published 7DS program is consistently organized into three layers, although the exact sky areas and depths differ between the commissioning paper and later forecast papers. The common structure is a shallow reference layer, an intermediate wide-field cadence layer, and a deeper intensive-monitoring layer (Kim et al., 2024, Ko et al., 26 Sep 2025, Bae et al., 31 Dec 2025).

In the commissioning description, the Reference Imaging Survey (RIS) covers most of the southern sky except the Galactic Plane, about 20,000 square degrees, with a canonical 100 s exposure, 3 exposures per visit, a total of 5 minutes per tile visit, and one visit per HEALPix tile during the first year; its stated role is to build subtraction references and baseline photometry. The Wide-Field Survey (WFS) covers up to 1600 square degrees, revisiting each position every 14 days over 5 years. The Intensive Monitoring Survey (IMS) performs daily monitoring of the AKARI Deep Field South, with a time budget of 20,000 minutes per year (Kim et al., 2024).

Later redshift studies retain the three-layer logic but use different numerical survey descriptions. In the SPHEREx-comparison paper, 7DS is described as having RIS over nearly all sky accessible from Chile south of $9576 \times 6388$7, about 24,000 deg$9576 \times 6388$8, with depth roughly $9576 \times 6388$9 to 20.7 depending on wavelength; WTS over about 1,600 deg$\mu$0 with depth roughly $\mu$1 to 23.2; and IMS over about 10 deg$\mu$2, reaching roughly $\mu$3 to 24.6 (Bae et al., 31 Dec 2025). In the dedicated 7DS photo-$\mu$4 forecast, RIS is instead described as an all-sky reference survey over 22,000 deg$\mu$5 at $\mu$6, with point-source depths of about 20 AB mag at 400 nm and 18 AB mag at 875 nm at $\mu$7; WTS has a two-week cadence over up to 8,000 deg$\mu$8 and reaches about 22 AB mag at 400 nm in five-year stacks; IMS is a daily-cadence survey of about 20 deg$\mu$9 near the South Ecliptic Pole, reaching about 23 AB mag at 400 nm in five-year stacks (Ko et al., 26 Sep 2025).

These differing parameterizations indicate that the literature contains multiple forecast configurations rather than a single immutable survey definition. What remains stable is the layered architecture itself: RIS provides reference imaging and bright-source catalogs, WTS/WFS supplies wide-area time-domain coverage and medium-depth photo-$0.5$0 capability, and IMS provides the deepest, highest-cadence optical spectrophotometric sampling.

3. Calibration, reduction, and real-time operations

The commissioning paper places unusual emphasis on calibration and throughput control, reflecting the fact that 7DS science depends on flux comparisons across many bands and epochs. Bias-level characterization used a routine of 18 bias frames taken 30 minutes after cooling begins and again after the final science exposure, at a camera temperature of $0.5$1C. The measured bias behavior was stable and consistent with the manufacturer’s read-noise expectation of about 3.5 $0.5$2 RMS, although the bias frames showed horizontal patterns characteristic of CMOS detectors (Kim et al., 2024).

Flat-field calibration was treated as a nightly operation because long-term combination was still uncertain during commissioning. The stated routine became 9 flat frames at the end of night operations for each filter. This is operationally important for 7DS because medium-band spectral mapping is directly sensitive to calibration errors in band-to-band throughput and to wide-field spatial response variations (Kim et al., 2024).

Spectrophotometric standardization is described as a central calibration layer. The system uses 68 spectrophotometric standard stars, including CALSPEC stars, together with synthetic photometry derived from Gaia XP spectra and Gaia DR3 catalogs; variable sources are removed using SkyMapper DR4-like criteria. The calibration logic proceeds by comparing observed 7DT medium-band photometry to Gaia XP-based synthetic magnitudes, then correcting offsets as functions of Gaia $0.5$3-band magnitude and color $0.5$4, defining zero points through differential photometry, applying $0.5$5 clipping, and repeating for different aperture sizes. The resulting error budget is reported as approximately 4 to 12 mmag (Kim et al., 2024).

The reduction pipeline is the modified gpPy+GPU system, based on the original gpPy package. Its main steps are data upload monitoring; master bias/dark/flat creation; bias/dark/flat correction; astrometry using astrometry.net and SCAMP; photometry with SExtractor; image alignment and stacking with SWarp; zeropoint calibration using Gaia XP-based synthetic photometry; subtraction with HOTPANTS; transient candidate extraction; and real/bogus classification using a CNN. The system is designed for up to 1 TB/night, with raw image size of about 117 MB per telescope; storage starts at 1 PB and is planned to grow by 1 PB/year to at least 5 PB. GPU acceleration makes the pipeline more than $0.5$6 faster than the original gpPy. The same paper also identifies quick robotic operation through TCSpy as part of the time-domain operational model (Kim et al., 2024).

In the context of 7DS, these details are not merely instrumental. The survey’s core products—photo-spectra, subtraction images, transient candidates, and redshift-quality SEDs—are all contingent on stable inter-band calibration and rapid reduction, especially for fast-fading kilonova candidates.

4. Redshift methodology and statistical definitions

A major part of the recent 7DS literature concerns photometric-redshift forecasting. The comparative SPHEREx+7DS study uses mock catalogs based on simulated SEDs from Feder et al. (2024), themselves built from real galaxy samples in GAMA and COSMOS2020. The GAMA component contains about 44,135 galaxies with $0.5$7 over $0.5$8, while the COSMOS2020 component contains about 166,014 galaxies selected with $0.5$9. The mock SED construction used Brown+COSMOS template libraries and included dust, stellar mass, and emission-line augmentation. Mock SPHEREx observations were generated with the SPHEREx Sky Simulator, whereas 7DS mock observations were obtained by convolving the SEDs with the 7DS filter curves and adding noise. For analysis, that study defined a wide dataset as SPHEREx full-sky + 7DS WTS, a deep dataset as SPHEREx deep + 7DS IMS, and also used a 7DS RIS sample for reference (Bae et al., 31 Dec 2025).

The dedicated 7DS photo-$1.34^\circ \times 0.90^\circ$0 forecast instead constructs mock catalogs from EL-COSMOS, based on COSMOS2015, with 518,404 galaxies spanning approximately $1.34^\circ \times 0.90^\circ$1. Those SEDs include stellar continua plus nebular emission lines such as $1.34^\circ \times 0.90^\circ$2, H$1.34^\circ \times 0.90^\circ$3, $1.34^\circ \times 0.90^\circ$4, and H$1.34^\circ \times 0.90^\circ$5\ 6565~\AA). That inclusion is methodologically important because the medium-band system can sample these features directly rather than only through broadband color combinations (Ko et al., 26 Sep 2025).

The comparative study evaluates six photo-$1.34^\circ \times 0.90^\circ$6 methods: four template-fitting approaches—SPHEREx in-house pipeline, EAZY, eazy-py, and LePhare—and two machine-learning methods, a deep neural network (DNN) and a hierarchical random forest (HRF). The paper explicitly examines the effect of dust extinction assumptions, template error functions, priors, flux uncertainties, and non-detections. The 7DS-only forecast uses EAZY with the eazy_v1.3 template set and included emission lines (Bae et al., 31 Dec 2025, Ko et al., 26 Sep 2025).

The comparative paper defines the redshift metrics

$1.34^\circ \times 0.90^\circ$7

$1.34^\circ \times 0.90^\circ$8

$1.34^\circ \times 0.90^\circ$9

and

$30^\circ 28'16''$0

The 7DS-only forecast uses the same general trio—normalized scatter, bias, and catastrophic outlier fraction—but defines catastrophic failures with the threshold $30^\circ 28'16''$1. This means that published $30^\circ 28'16''$2 values are not strictly interchangeable across the two forecast papers and must be compared with attention to the adopted threshold (Bae et al., 31 Dec 2025, Ko et al., 26 Sep 2025).

5. Spectral leverage and complementarity with SPHEREx and other surveys

The most developed comparative argument in the literature is that 7DS and SPHEREx are highly complementary low-resolution spectrophotometric surveys. In the SPHEREx-comparison paper, 7DS covers the optical regime from $30^\circ 28'16''$3 to $30^\circ 28'16''$4m at $30^\circ 28'16''$5–60, while SPHEREx covers the near-infrared from $30^\circ 28'16''$6 to $30^\circ 28'16''$7m with $30^\circ 28'16''$8–130 and 102 spectral channels per sky position. Together they provide continuous low-resolution spectral sampling from $30^\circ 28'16''$9 to $70^\circ 45'47''$0m with minimal gaps over a broad wavelength baseline (Bae et al., 31 Dec 2025).

The practical consequence of that wavelength complementarity is that the combined dataset can simultaneously capture or constrain the 4000 $70^\circ 45'47''$1 break and other optical continuum features, the $70^\circ 45'47''$2m bump, and emission features such as H$70^\circ 45'47''$3 and PAH-related structure. The paper identifies two mechanisms by which this improves redshift inference: broader wavelength coverage reduces confusion between redshift and galaxy-type or dust effects, and multi-regime feature identification allows optical and near-infrared indicators to reinforce one another. This is the explicit basis for the claim that SPHEREx + 7DS outperforms either survey alone, particularly for fainter galaxies near the sensitivity limits of individual datasets (Bae et al., 31 Dec 2025).

The dedicated 7DS forecast extends the comparison to Pan-STARRS1, VIKING, and SPHEREx. Pan-STARRS1 broad-band optical data provide a modest improvement at brighter magnitudes and about a factor-of-two improvement for the faintest 7DS objects, but do not remove the intrinsic color-redshift degeneracies because the spectral resolution remains sparse. VIKING adds near-infrared coverage from about 8875 $70^\circ 45'47''$4 to $70^\circ 45'47''$5m and helps break low-$70^\circ 45'47''$6/high-$70^\circ 45'47''$7 degeneracies, especially up to $70^\circ 45'47''$8–3. SPHEREx is described as the strongest external complement: combining SPHEREx all-sky data with WTS produces the best overall performance in that paper, with $70^\circ 45'47''$9 in the faintest $\sim 1.5''$0 bin and $\sim 1.5''$1 for fainter sources at higher redshifts (Ko et al., 26 Sep 2025).

The same paper also highlights a more specific diagnostic enabled by the medium-band system: color excess from adjacent broad-band subtraction, such as $\sim 1.5''$2. Positive excesses are associated with emission lines including H$\sim 1.5''$3, H$\sim 1.5''$4, $\sim 1.5''$5, and $\sim 1.5''$6, whereas negative values trace spectral breaks such as the 4000 $\sim 1.5''$7 or Balmer break. In one example, galaxies with $\sim 1.5''$8, $\sim 1.5''$9, and $\sim 300$0 in $\sim 300$1 yielded 54 strong color-excess galaxies, compared to 1,740 without such excess; no catastrophic failures appeared in the color-excess sample, whereas about 187 galaxies ($\sim 300$2) in the no-excess sample were catastrophic outliers. This suggests that the medium-band system is useful not only for redshift estimation but also for identifying subsets whose redshifts are especially well constrained (Ko et al., 26 Sep 2025).

6. Forecast performance, caveats, and scientific role

The comparative SPHEREx+7DS study reports very strong performance for bright galaxies. For GAMA galaxies in the wide sample with $\sim 300$3, the abstract states that all methods and survey combinations achieve $\sim 300$4, bias $\sim 300$5, and catastrophic failure rate $\sim 300$6. In the conclusion, the combined SPHEREx + 7DS case is tightened to $\sim 300$7, $\sim 300$8, and $\sim 300$9; the best cases reach sub-percent, even $^{-2}$0, accuracy. For COSMOS-like galaxies with $^{-2}$1 in the wide sample, the combined data still perform strongly, with $^{-2}$2, $^{-2}$3, and $^{-2}$4, and the best case reaches about 0.71% accuracy. The deep configuration is described as handling galaxies up to two magnitudes fainter than the wide sample while retaining comparable performance (Bae et al., 31 Dec 2025).

The same study gives particularly clear numerical evidence for the gain from joint optical-plus-near-infrared coverage in the challenging $^{-2}$5 COSMOS regime. For EAZY, SPHEREx only in the wide case gives $^{-2}$6, $^{-2}$7, and $^{-2}$8, whereas the combined deep case gives $^{-2}$9, $9576 \times 6388$00, and $9576 \times 6388$01. For LePhare, the combined deep case gives $9576 \times 6388$02, $9576 \times 6388$03, and $9576 \times 6388$04. For eazy-py, the combined deep case gives $9576 \times 6388$05, $9576 \times 6388$06, and $9576 \times 6388$07. For the SPHEREx in-house pipeline, the combined deep case gives $9576 \times 6388$08, $9576 \times 6388$09, and $9576 \times 6388$10. These results are the most direct expression of the paper’s core conclusion that combined SPHEREx + 7DS data are substantially better than either dataset alone, especially in the deep configuration (Bae et al., 31 Dec 2025).

The 7DS-only photo-$9576 \times 6388$11 forecast also finds strong performance, especially for WTS and IMS. For the five-year WTS with 40 medium bands, at $9576 \times 6388$12, the paper reports $9576 \times 6388$13 to $9576 \times 6388$14, $9576 \times 6388$15 to $9576 \times 6388$16, and $9576 \times 6388$17 to $9576 \times 6388$18. For IMS, at $9576 \times 6388$19, it reports $9576 \times 6388$20–0.003 and $9576 \times 6388$21; at $9576 \times 6388$22, it reports $9576 \times 6388$23 and $9576 \times 6388$24. RIS is significantly shallower: at $9576 \times 6388$25, the paper gives $9576 \times 6388$26 and $9576 \times 6388$27, with rapid degradation at fainter magnitudes. The same paper concludes that the overlapping 40-band system outperforms a non-overlapping 20-band subset across all magnitude bins, because denser effective spectral sampling improves recovery of breaks and emission features (Ko et al., 26 Sep 2025).

The forecast literature also identifies several caveats. Dust treatment matters: the SPHEREx in-house pipeline explicitly includes dust extinction laws in the fit, while LePhare in the cited setup relies on templates that already encode those effects; in 7DS-only optical data, dust extinction becomes more degenerate with redshift, which can degrade performance relative to broader wavelength coverage. Flux uncertainties in HRF can help in faint regimes because they are included as input features, but they may also bias results in an inhomogeneous survey if the uncertainties encode observing conditions rather than intrinsic redshift information. The comparative paper further reports that template-fitting PDFs are generally well calibrated under the Probability Integral Transform, though with signs of under-dispersion for the in-house code and EAZY, and that confidence intervals are generally underestimated, especially for bright galaxies. In that analysis, the eazy-py configuration that scales flux uncertainties by 1.5 gives the best match between estimated uncertainty and actual scatter (Bae et al., 31 Dec 2025).

In scientific terms, 7DS is therefore best understood as a dual-purpose optical spectrophotometric survey. Its original operational motivation is the rapid identification and discrimination of electromagnetic counterparts to gravitational-wave sources, especially kilonovae, in localization regions that may span hundreds to thousands of square degrees and contain many unrelated transients. At the same time, the same medium-band architecture enables high-quality photometric redshifts, emission-line selection, AGN variability studies, large-scale structure work, and related survey science. The redshift forecasts suggest that 7DS is already useful as a standalone optical medium-band system, but that its full leverage emerges when its optical spectral mapping is combined with near-infrared low-resolution spectroscopy from SPHEREx (Kim et al., 2024, Ko et al., 26 Sep 2025, Bae et al., 31 Dec 2025).