Source-linked AI summary
The First Release COSMOS Optical and Near-IR Data and Catalog
P. Capak, H. Aussel, M. Ajiki, H. J. McCracken, B. Mobasher, N. Scoville, P. Shopbell, Y. Taniguchi, D. Thompson, S. Tribiano, S. Sasaki, A. W. Blain, M. Brusa, C. Carilli, A. Comastri, C. M. Carollo, P. Cassata, J. Colbert, R. S. Ellis, M. Elvis, M. Giavalisco, W. Green, L. Guzzo, G. Hasinger, O. Ilbert, C. Impey, K. Jahnke, J. Kartaltepe, J-P. Kneib, J. Koda, A. Koekemoer, Y. Komiyama, A. Leauthaud, O. Lefevre, S. Lilly, R. Massey, S. Miyazaki, T. Murayama, T. Nagao, J. A. Peacock, A. Pickles, C. Porciani, A. Renzini, J. Rhodes, M. Rich, M. Salvato, D. B. Sanders, C. Scarlata, D. Schiminovich, E. Schinnerer, M. Scodeggio, K. Sheth, Y. Shioya, L. A. M. Tasca, J. E. Taylor, L. Yan, G. Zamorani
TL;DR
COSMOS photometry required improved calibration because scattered light makes standard flat-fielding inaccurate and standard-star coverage limited. The paper presents tractable multi-band data products and a combined detection catalog, achieving better-than-1% field consistency and corrected zero points accurate to better than 2%.
Problem
Standard flat-fielding is inaccurate at the 3-5% level, while limited standard-star observations yield zero-point calibrations accurate only to ±0.05 magnitudes.
Method
The study constructs common-grid data products and combines CFHT i* with Subaru i+ imaging to maximize detection dynamic range and support multi-band photometry.
Results
Better than 1% photometric consistency is achieved across the COSMOS field, while galaxy-based corrections produce zero points accurate to better than 2%.
Takeaways & Limitations
The resulting catalog and photometry support large-scale-structure studies and photometric-redshift science goals.
Takeaways & Limitations
Standard-star calibration remains limited, and future internal spectrophotometric standards are needed to verify the corrected zero points.
Abstract
from arXiv · showhide
We present imaging data and photometry for the COSMOS survey in 15 photometric bands between 0.3um and 2.4um. These include data taken on the Subaru 8.3m telescope, the KPNO and CTIO 4m telescopes, and the CFHT 3.6m telescope. Special techniques are used to ensure that the relative photometric calibration is better than 1% across the field of view. The absolute photometric accuracy from standard star measurements is found to be 6%. The absolute calibration is corrected using galaxy spectra, providing colors accurate to 2% or better. Stellar and galaxy colors and counts agree well with the expected values. Finally, as the first step in the scientific analysis of these data we construct panchromatic number counts which confirm that both the geometry of the universe and the galaxy population are evolving.
2.1. Data Products
COSMOS data products are organized on a common tiled grid with matched image, PSF, and noise-map products to support consistent analysis across bands.
- Data Products: The 4-square-degree field is divided into 144 overlapping 10′ × 10′ sections, each represented by a 4096×4096 image at 0.15′′ per pixel.Adjacent tiles overlap by 14.4′′, so most objects can be analyzed on a single image.
- Data Products: Subaru and SDSS bands provide original-PSF and field-homogenized images, with additional best-seeing images for Subaru BJ, r+, and i+.CFHT provides original-PSF images, while CTIO and KPNO provide PSF-homogenized images because of larger PSF variation.
- Data Products: RMS noise maps accompany every filter on the same tiling and flux scale, incorporating detector and processing noise but excluding photon noise from object flux.Included terms cover photon noise, background subtraction, flat fielding, masking, saturation, and cosmic-ray removal.
2.2. Subaru Suprime-Cam
The Subaru Suprime-Cam reduction combines extensive calibration, masking, sky subtraction, and a solved scattered-light correction to achieve stable photometry across the COSMOS field.
- 2.2. Subaru Suprime-Cam: Nine Suprime-Cam pointings cover COSMOS in BJ, VJ, g+, r+, i+, z+, and NB816, with additional intermediate-band observations.The camera has a 34′×27′ field of view and ten 2k × 4k CCD detectors.
- 2.2. Subaru Suprime-Cam: Suprime-Cam processing includes overscan and bias correction, flat-fielding, object detection, astrometric fitting, sky subtraction, defect masking, and final catalog generation.Astrometric solutions use fourth-order two-dimensional polynomials, with final scatter below 0.2′′ at 1σ.
- 2.2.2. Scattered Light Correction: The standard flat-field method is inaccurate at the 3-5% level because scattered light acts like an additive dark current rather than a multiplicative sensitivity variation.Suprime-Cam scattered-light patterns vary with lighting conditions and telescope position, reaching ±5% at field edges.
- 2.2.2. Scattered Light Correction: A true flat is solved from objects observed at multiple focal-plane positions by estimating regional correction factors and exposure-dependent terms.For photometric, airmass-corrected data, the exposure term Pe is zero.
- 2.2.2. Scattered Light Correction: The corrected flat remains stable across telescope position, time, and photometric quality, allowing one flat to serve multiple telescope runs.The flats are publicly available in the COSMOS archive.
2.3. CFHT Megaprime
CFHT Megaprime provides deep and shallow optical imaging with queue-based observing and standardized reduction procedures designed to maintain uniform image quality and photometry.
- 2.3. CFHT Megaprime: Megaprime has a 1 square-degree field of view with 36 CCD detectors and was used for deep u∗ and shallow i∗ COSMOS imaging.Five overlapping pointing centers cover the field, producing data four times deeper in the center than at the edges.
- 2.3. CFHT Megaprime: Queue observing and the Elixir pipeline provide consistent image quality and photometric calibration better than 1% across the field of view.Elixir corrects bias, dark current, flat fielding, and scattered light.
- 2.3. CFHT Megaprime: TERAPIX performs further astrometric and photometric calibration, sky subtraction, image combination, quality inspection, and rejection of images with excessive seeing.Images exceeding 1.3′′ in i∗ or 1.4′′ in u∗ are rejected.
2.4. CTIO ISPI and KPNO FLAMINGOS
CTIO ISPI and KPNO FLAMINGOS data are combined into a COSMOS-wide Ks image through repeated dithered coverage, two-pass infrared reduction, calibration, and resampling.
- 2.4. CTIO ISPI and KPNO FLAMINGOS: The single-detector ISPI and Flamingos cameras cover slightly more than 10′ × 10′ and provide 0.9µm–2.4µm infrared sensitivity for the combined Ks image.Depth varies spatially because of weather and instrument field-of-view constraints.
- 2.4. CTIO ISPI and KPNO FLAMINGOS: 81 KPNO or CTIO pointings cover COSMOS, with every position observed at least four times by KPNO and three times by CTIO.A second 64-pointing CTIO grid offset by half a pointing improves photometric consistency.
- 2.4. CTIO ISPI and KPNO FLAMINGOS: The two-pass reduction builds object-masked sky flats, subtracts temporally adjacent sky frames, removes residual background, and masks sources, satellites, and bad regions.The procedure uses IRAF tasks together with SExtractor-derived object masks.
- 2.4. CTIO ISPI and KPNO FLAMINGOS: Astrometric registration uses a fourth-order two-dimensional polynomial, achieving position-independent scatter below 0.3′′.The larger scatter reflects the 0.3′′ detector pixels and poor seeing.
- 2.4. CTIO ISPI and KPNO FLAMINGOS: Images are calibrated to 2MASS, shifted by 1.852 magnitudes to the AB system, smoothed to 1.5′′ FWHM, and combined on the COSMOS 0.15′′-per-pixel grid.An original-PSF Ks image is not produced because PSF variations are too large.
2.5. Sloan Digital Sky Survey
The SDSS data were incorporated into the COSMOS catalog through calibrated mosaics, PSF homogenization, and comparison with Subaru photometry. Residual strip-dependent systematics and seeing-related offsets constrain the reliability of some measurements.
- SDSS DR2 photometry supplements Subaru and CFHT data for objects saturated in those observations, extending reliable measurements to 10th magnitude.
- SDSS images were mosaicked on the COSMOS grid, calibrated, sky-subtracted, and matched to the COSMOS astrometric catalog.
- 5% systematic effects remain between SDSS stripes after Gaussian FWHM homogenization because of non-Gaussian PSF wings.
- 0.06 magnitudes of declination offset appears between SDSS passes from imperfect PSF matching, but total magnitudes remove it and colors are not significantly impacted.
- 3′′ apertures were used for PSF-matched multi-band photometry because the method accounts for non-Gaussian PSF structure; other apertures retain systematic effects.
4.1. Catalog Contents
The catalog combines multi-band measurements with standardized missing-data codes, aperture corrections, masks, and qualitative star and de-blending flags. These metadata support photometric analysis while identifying measurements requiring caution.
- The catalog contains PSF-matched 3′′ aperture photometry and errors for Subaru, CFHT, KPNO+CTIO Ks, and HST F814W data.
- Non-detections receive magnitude 99 with a 1σ limiting-magnitude error, while unavailable or defective measurements receive magnitude and error -99.
- Applying the catalogued aperture correction to any band provides that band’s total magnitude.
- The star/galaxy flag is qualitative, so studies requiring accurate separation should perform a more detailed analysis.
- Masked-region flags record the area of the photometric aperture falling inside a masked region in square arc-seconds.
- Heavily de-blended objects are flagged because aggressive de-blending near bright objects produces numerous false detections.
4.2. Catalog Usage Guide
Catalog flags must be interpreted according to the analysis: strict flag cuts improve photometric reliability but remove area, while de-blending and masks have analysis-specific effects.
- Objects with all mask and de-blending flags clear have the most reliable photometry, but this selection removes a significant fraction of survey area.
- De-blending flags may be ignored for multi-wavelength cross-correlation but are important for optical clustering, especially at faint magnitudes.
- All photometry masks must be applied for a clean photometric-redshift sample because masked photometry affects redshifts non-linearly.
4.3. Completeness and Confusion
Completeness and confusion are quantified through simulated-object recovery, with detection settings chosen to maximize depth while accepting spurious sources near the limit.
- Completeness and confusion can dominate photon noise in deep surveys and depend on both data quality and software.
- A factor-of-two seeing difference reduces measurement sensitivity by 0.3 magnitudes, peak surface brightness by 0.75 magnitudes, and increases blending fourfold.
- Detection settings maximize depth but produce spurious detections near the limit, which can be removed with masks and magnitude or FWHM cuts.
- Simulated objects spanning representative morphologies and magnitudes are recovered with SExtractor, including confusion effects because existing objects are not avoided.
- The i+ detected catalog is 91% complete at i+ = 25.0 and 87% complete at i+ = 26.0.
5.1. Consistency of Photometry
COSMOS photometry was designed to support photometric-redshift and large-scale-structure studies through stringent spatial calibration and cross-survey consistency checks. Comparisons show stable photometry across the field, with small scatter against independent optical and near-IR measurements.
- Less than 1% spatial photometric variation is required to keep systematic photometric-redshift errors below 1% for large-scale-structure studies.A 1% photometric error typically produces a 1% photometric-redshift error.
- COSMOS, CFHT-LS, SDSS, ACS, and 2MASS photometry were compared after filter-system conversion, magnitude harmonization, and positional catalog matching.The comparisons used survey-specific total-magnitude estimates and conversions into the COSMOS photometric system.
- The SDSS u and CFHT u* comparison was not performed because their filters lack a simple linear relationship.
- No systematic positional effects were measurable in the optical comparisons, and COSMOS photometry remained constant across the field to better than 1%.The ACS and Subaru i+ comparison indicates that observed variations involving CFHT-LS and SDSS largely arise from errors in those surveys.
- 0.02 magnitudes is the rms variation between COSMOS and 2MASS Ks photometry for unsaturated objects measured directly in the Ks image.Objects with i+ > 16 and at least 5σ Ks detections show no positional trend; brighter objects suffer from saturated detection images and incorrect aperture corrections.
5.2. Galactic Extinction Correction
The COSMOS catalog accounts for Galactic extinction using field-level reddening estimates and filter-dependent corrections derived from the extinction curve. The median extinction is small but non-negligible, especially in the u* band.
- The SDSS u and CFHT u* filters cannot be compared using a simple linear relation.
- Each catalog object receives a band-specific Galactic-extinction correction based on its estimated extinction and a filter-dependent factor.The factors are calculated by integrating each filter response against the Galactic extinction curve.
5.3. Absolute Photometric Zero-Point Corrections
The paper recalibrates absolute photometric zero points by combining external-survey comparisons with galaxy spectra and stellar-color checks. Spectroscopic-redshift corrections substantially improve agreement, while the Ks band remains constrained by limited intermediate-wavelength data.
- ±0.05 magnitudes is the standard-star calibration accuracy, and these offsets are larger than desired for accurate photometric redshifts.Only three to five standards per band were available because standard-star observations on Suprime-Cam require substantial overhead.
- ±0.06 magnitudes is the RMS amplitude of zero-point corrections independently estimated from CFHT-LS and SDSS comparisons.These corrections are slightly larger than the expected error and disagree with ACS F814W by -0.118 magnitudes.
- Galaxy-template fitting at known spectroscopic redshifts estimates offsets by combining rest-frame 100 Å wavelength-bin comparisons across templates.Weighted averaging reduces effects from calibration errors in individual galaxy templates.
- The Ks band cannot be corrected with the galaxy-spectrum method because COSMOS lacks photometric data between 0.9 µm and 2.2 µm.The catalog instead relies on the 2MASS zero point, despite a measured -0.097 magnitude offset.
- Offsets are within the expected ±0.05 magnitude error except for the Suprime-Cam BJ band, which has known calibration problems.
- Spectroscopic-redshift offsets remove systematic trends in photometric versus spectroscopic redshifts for 842 objects spanning 0 < z < 1.2.After correction, i+ and z+ agree with F814W to 0.007 magnitudes, and tests indicate zero points within 0.01 magnitude of the true AB zero points.
- The recommended photometric-redshift offsets were not applied to the released catalog because external calibration could not yet verify them.
5.4. Star Colors
Stellar color-color diagrams provide a sensitive test of COSMOS color accuracy and filter performance. Observed stellar colors agree closely with metallicity-matched theoretical and empirical libraries, while metallicity differences explain expected library offsets.
- Stars diagnose small color-calibration and filter-transmission differences because they form tight loci and their colors are sensitive to instrumental response.
- The BzK diagram selects stars for cleaner color-color comparisons, using only objects with greater than 10σ detections to limit faint-blue-star bias.The shallow Ks data make the selection less complete for faint blue stars, but the effect is described as minimal for the plotted sample.
- Solar-metallicity stellar libraries predict redder colors than the predominantly sub-solar-metallicity COSMOS thick-disk population, especially in the ultraviolet.Higher metallicity increases metal-line absorption, producing redder ultraviolet colors.
- Stellar colors agree extremely well with reference libraries, with median expected-to-observed offsets below 0.02 magnitudes in most bands.The comparison uses four color-color plots and theoretical libraries evaluated at [Fe/H]=-0.4 alongside the Pickles library.
5.5. Number Counts
COSMOS number counts agree with previous studies and reveal wavelength-dependent behavior linked to galaxy evolution and an expanding universe. The i+-selected catalog also has quantified completeness limits supporting these counts.
- The normalized number-count slopes steepen with increasing wavelength, indicating evolution in both the universe and the galaxy population.The u∗ slope remains close to 0.6, while near-IR Ks counts lie far below 0.6.
- The COSMOS data comprise 15 bands spanning 0.3µm–2.2µm with expected zero-point variations below 1% across the field.Flat fields constructed directly from object fluxes achieved this photometric consistency.
- 91% completeness at i+ = 25.0, 87% at i+ = 26.0, and 50% at i+ = 27.4 characterize the resulting catalog.
- Corrected photometric zero-points appear accurate to better than 2% based on star colors.Standard-star zero-points were initially accurate to 5%.
- Number counts from the i+-selected catalog agree well with previous studies.