COSMOS-Web: JWST Extragalactic Imaging Survey

• COSMOS-Web Survey is a large-scale JWST imaging project that maps extragalactic fields with high-resolution near- and mid-infrared data, spanning cosmic time from the epoch of reionization to today.
• It employs innovative observational strategies with 152 visits and custom data reduction pipelines to achieve uniform 5σ depths (26.7–28.3 AB mag) and sub-arcsecond astrometric precision.
• The survey provides a benchmark dataset for studies in galaxy evolution, cosmic structure formation, and weak lensing, enhanced by extensive multi-wavelength ancillary data.

The COSMOS-Web Survey is the largest contiguous extragalactic program undertaken during JWST Cycle 1, delivering deep, high-resolution, near- and mid-infrared imaging across 0.54 deg² of the COSMOS field. With a comprehensive combination of four broad NIRCam filters (F115W, F150W, F277W, F444W) and parallel MIRI F770W imaging, COSMOS-Web is designed for statistical studies of galaxy assembly, structure, and environment over cosmic time, extending from the epoch of reionization (z ≫ 6) to the present. The survey’s unique scale, depth, and rich ancillary data make it a benchmark dataset for studies in cosmology, galaxy evolution, and high-redshift astrophysics.

1. Survey Architecture and Observational Design

COSMOS-Web executes JWST NIRCam imaging over a contiguous 0.54 deg² rectangular mosaic (41.5′ × 46.6′, Position Angle 20°), distributed in a regular grid of 19 columns × 8 rows (152 visits) with two 4-TIGHT dither sets per visit and four integrations (~257 s each). The sequence incorporates 12,880 total exposures, observed in three epochs: initial coverage in January 2023 (6 visits; ~77 arcmin²), main coverage in April 2023 (~half area, 77 visits), and completion/repeat coverage in December 2023–January 2024 (69 visits plus repeats). The planned on-sky time was 255 hours, with overheads bringing the total to approximately 270 hours (Franco et al., 3 Jun 2025, Casey et al., 2022). Parallel MIRI F770W imaging covers ~0.20 deg².

The four NIRCam bands (F115W, F150W – SW; F277W, F444W – LW) reach 5σ point-source depths of 26.7–28.3 AB mag (0.15″ apertures), with typical achieved FWHM in PSF of ~0.04″ in SW and ~0.08″ in LW. Contiguity of the field—rare among JWST extragalactic projects—suppresses cosmic variance and enables environmental and structure studies on scales up to ~120 Mpc at z ~ 8.

Ancillary data include HST/ACS F814W, VISTA/UltraVISTA, Subaru HSC, Spitzer IRAC, X-ray (Chandra, XMM), and extensive ALMA coverage, supporting robust photometric redshifts and multi-wavelength science (Shuntov et al., 3 Jun 2025, Casey et al., 2022).

2. Data Reduction Methodology

All raw exposures are processed using the official STScI JWST Calibration Pipeline (v1.14.0 with CRDS 0285), integrated with custom preprocessing steps aimed at maximizing data fidelity at the field scale, as detailed in Franco et al. (Franco et al., 3 Jun 2025). Major custom innovations include:

• Stage 1 (“detector1”): Standard steps (DQ init, saturation, reference-pixel, linearity, jump (CR) detection, slope fitting) are augmented by targeted strategies for artifact removal. These include “snowball” cosmic-ray showers (with tuned jump_detection and manual ShowerMasking), “wisp” scattered-light templates (built per filter/detector), 1/f noise correction (amplifier-wise row/column medians), persistence flagging (snowblind's PersistenceFlagStep), and the removal of “claws” optical artifacts via Gaussian-convolved template subtraction.
• Stage 2 (“image2”): Creation of *_cal.fits files with WCS registration, flat-fielding, and photometric calibration. A Gaussian fit to the masked pixel-value histogram is subtracted to homogenize detector backgrounds.
• Stage 3 (“image3”): Mosaics produced via inverse-variance weighted drizzle-coaddition in 20 independent overlapping tiles (~20.2′×26.2′ each), enabling manageable data processing and parallelism. Global sky background is refined through iterative sigma-clipping and Astropy Background2D statistical smoothing. Astrometric alignment uses HST/ACS F814W catalogs tied to Gaia DR3/COSMOS2020, with SW filters aligned through an LW intermediary and JHAT solution, achieving <3 mas (RA) and <4 mas (Dec) median offsets; MAD ≤14 mas.
• Product structure: Per-tile “i2d.fits” files with extensions for SCI, ERR, CONF, WHT, VAR_POISSON, VAR_RNOISE, VAR_FLAT, and supplied mosaics at 20, 30, and 60 mas pixel scales (Franco et al., 3 Jun 2025).

Data products are released to the public at COSMOS-Web DR1.

3. Photometric Depth, Sensitivity, and Calibration

The survey attains uniform depth, reaching 5σ limits (0.15″ apertures, no correction) as follows:

# Exposures	F115W (mag)	F150W (mag)	F277W (mag)	F444W (mag)
1	26.69	26.95	27.74	27.60
2	27.05	27.30	28.01	27.88
3	27.24	27.49	28.19	28.07
4	27.41	27.65	28.34	28.22

PSF FWHM ranges from 0.04″ (SW) to 0.08″ (LW), consistent with pre-flight and commissioning benchmarks. MIRI F770W reaches m_AB ∼ 25.5 in regions with four exposures, exceeding pre-launch expectations (Harish et al., 3 Jun 2025).

Calibration quality is monitored via stacked star profiles, PSF matching across bands, and full error propagation (including correlated noise) in all science mosaics. Astrometric registration is accurate to <5 mas with robust alignment to auxiliary catalogs (Franco et al., 3 Jun 2025).

4. Catalog Products and Derivation of Physical Parameters

High-level catalogs (COSMOS2025; (Shuntov et al., 3 Jun 2025)) combine JWST photometry (NIRCam+MIRI) with 30–40 ancillary bands, using a dual detection strategy:

• Source detection: positive-truncated χ² co-add of NIRCam bands, dual “hot/cold” SExtractor runs for completeness/deblending.
• Photometry: profile-fit photometry and multi-band model-based fits (SourceXtractor++) across all ground- and space-based imaging.
• Morphology: SE++ Sérsic and bulge+disk fits, Galight/Lenstronomy decompositions, and ML-based classification yielding spheroid, bulge-dominated, disk-dominated, and irregular likelihoods.
• Photometric redshifts and SED fitting: LePHARE/Bruzual & Charlot (2003) templates, emission lines, multi-extinction laws, IGM absorption; validation with >12 000 secure spec-z’s gives σMAD = 0.012 at m_F444W < 28 and factor-two improvement over COSMOS2020. Stellar mass completeness: ~80% at log(M*/M⊙) ∼ 9 at z ∼ 10 (Shuntov et al., 3 Jun 2025).

All products are publicly distributed with complete photometric error vectors, morphological parameters, photometric redshift PDFs, and value-added tables for scientific exploitation.

5. Scientific Goals and Legacy Science

Primary science goals (Casey et al., 2022, Shuntov et al., 3 Jun 2025, Franco et al., 3 Jun 2025):

1. Epoch of Reionization (EoR): Identification of ~4000 z ∼ 6–7, ~1500 z ∼ 7–8, and ~680 z ∼ 8–9 candidates, minimizing cosmic variance (σ_cv ~16% at z ∼ 9); mapping large-scale structure via kernel density in 12 independent redshift slices.
2. Quiescent Galaxy Populations at z > 4: Robust selection of sSFR < 10^–11 yr⁻¹ and M⋆ > 10¹⁰ M_⊙ systems, with legacy constraints on the quiescent mass function to log n ~ –7 Mpc⁻³ at 4 < z < 6.
3. Stellar-Mass–Halo-Mass Relation via Weak Lensing: Tangential shear measurements at 1 Mpc scales and extension of SMHR constraints to z ~ 2.5 with >110,000 background sources at z>2.5.

Additional legacy science:

• Detection of z>10 pair-instability supernovae; direct-collapse black holes; sub-dwarf and brown-dwarf Galactic populations.
• Strong lensing: Forecasts predict ~107 small-θ_E systems (88% with θ_E<1″), probing reionization-era sources to z_s ∼ 11 (Hogg et al., 11 Mar 2025).
• Discovery and analysis of rare populations (e.g., “little red dots,” extended massive dusty galaxies at z>5, compact star-formers, and AGN; (Akins et al., 2024, Gentile et al., 2024)).

6. Data Reduction Pipeline: Innovations and Challenges

The combination of JWST standard pipeline and targeted corrections was essential for high-fidelity science mosaics at COSMOS-Web's scale:

• Noise sources (1/f read-noise, “snowballs,” “wisps,” persistence, “claws”) required specialized masking, empirical templates, or amplifier/row-level corrections in custom pre-processing modules.
• Mosaicking in 20 overlapping tiles facilitated parallelism and computational tractability.
• Astrometric alignment to ACS+Gaia catalogs reached <5 mas precision, crucial for source extraction and cross-matching.
• Outlier and artifact detection used both pixel-level and object-level masks, with statistical post-processing for bad-pixel, cosmic-ray, and detector artifact suppression (Franco et al., 3 Jun 2025).

These methods enabled uniformity and reproducibility, supporting public releases and third-party analyses.

7. Public Release and Broader Impact

COSMOS-Web delivers an unprecedented dataset—spanning the widest contiguous JWST coverage at maximal extragalactic sensitivity—with all reduced mosaics, high-level science catalogs, masks, and documentation publicly available (e.g., via https://cosmos2025.iap.fr/ and DR1 web portal). Integration with legacy COSMOS field data allows immediate exploitation across fields ranging from cosmic structure formation to rare object discovery. The pipeline and reduction methodologies serve as a reference for future large-area JWST surveys.

COSMOS-Web's approach—combining official JWST pipeline outputs with specialized artifact correction, rigorous cross-band astrometry, tile-based mosaicking, and profile/model-based photometry—establishes new standards for wide-field space-based imaging surveys (Franco et al., 3 Jun 2025, Shuntov et al., 3 Jun 2025). The dataset underpins key advances in studies of galaxy evolution, large-scale structure, lensing, and the earliest phases of galaxy and SMBH assembly.