---
title: Farmer COSMOS2020 Catalog Overview
url: https://www.emergentmind.com/topics/farmer-cosmos2020-catalog
type: topic
---

# Farmer COSMOS2020 Catalog Overview

to=exec code: {"cmd":"python - <<'PY'\nimport urllib.request, urllib.parse\nids=['2110.13923','2310.07757','2207.11740']\nfor aid in ids:\n    url=f'https://export.arxiv.org/api/query?id_list={aid}'\n    try:\n        with urllib.request.urlopen(url, timeout=10) as r:\n            txt=r.read(300).decode('utf-8','ignore')\n        print(aid, 'OK', txt[:120].replace('\\n',' '))\n    except Exception as e:\n        print(aid, 'ERR', e)\nPY"}񎓂assistant to=exec code  ปมถวายสัตย์ฯ 买天天中彩票{"exit_code":0,"stdout":"2110.13923 ERR <urlopen error [Errno -3] Temporary failure in name resolution>\n2310.07757 ERR <urlopen error [Errno -3] Temporary failure in name resolution>\n2207.11740 ERR <urlopen error [Errno -3] Temporary failure in name resolution>\n","stderr":""}
The Farmer COSMOS2020 Catalog is the profile-fitting photometric branch of the COSMOS2020 release, produced for the COSMOS field as one of two complementary catalogs alongside the aperture-based “Classic” catalog. In COSMOS2020, source detection and multi-wavelength photometry are performed for 1.7 million sources across the \(2\,\mathrm{deg}^{2}\) COSMOS field, and \(\sim 966{,}000\) of these are measured with all available broad-band data using both traditional aperture photometric methods and The Farmer. The Farmer itself is an automated, reproducible profile-fitting photometry package built around smooth parametric models from The Tractor, designed to address source confusion and blending in deep optical, near-infrared, and mid-infrared imaging while delivering native total fluxes without aperture corrections [2110.13923, 2310.07757].

## 1. Position within COSMOS2020

Within COSMOS2020, the Farmer catalog occupies the role of the model-fitting alternative to the Classic catalog. The full COSMOS2020 release was designed to provide a new reference photometric redshift catalog based on newly collected imaging and spectroscopy in the COSMOS field, and it explicitly distributes both photometric methodologies rather than replacing one with the other. The Farmer catalog focuses on the UltraVISTA footprint, using source detection on a weighted mean chi-squared coadd (“CHI_MEAN”) constructed from HSC \(i,z\) and UltraVISTA \(Y,J,H,K_s\) bands, and it excludes areas affected by bright star masks [2110.13923].

The source counts reported for the Farmer branch make its survey scale explicit. The catalog contains 964,506 sources in total, with 816,944 lying outside HSC bright star masks and within UltraVISTA. Its wavelength coverage extends from the ultraviolet to the mid-infrared, including CFHT/MegaCam \(u\), broad optical HSC \(g,r,i,z,y\), Subaru/Suprime-Cam medium and narrow bands, UltraVISTA \(Y,J,H,K_s\), and Spitzer/IRAC channels. This broad coverage is central to its use in photometric redshift estimation, stellar population inference, and high-redshift source selection [2110.13923].

A common misconception is that “COSMOS2020” and “the Farmer catalog” are interchangeable terms. They are not. COSMOS2020 is the survey release and analysis framework; the Farmer catalog is one of its two complementary photometric realizations, with different measurement assumptions, systematics, and strengths [2110.13923].

## 2. Profile-fitting methodology

The Farmer was developed to confront the central observational problem of deep ground-based surveys: source confusion at depths approaching those of space-based imaging but at lower spatial resolution. Its core method is profile fitting with a decision tree that selects among a library of smooth parametric models, fits neighboring sources jointly, and then propagates the resulting morphology into forced photometry on other bands while leaving brightness free to vary. Because the fitted models are total-light models, the resulting photometric measurements are naturally total, and no aperture corrections are required [2310.07757].

Detection and segmentation are performed with SEP, and the field is subdivided into overlapping “bricks” that are processed in parallel. Source groups are constructed from dilated segments so that potentially blended neighbors are modeled simultaneously rather than independently. The fit uses band- and position-dependent PSF information, including spatial grids where needed, and thus avoids the PSF homogenization required by fixed-aperture workflows [2110.13923, 2310.07757].

The model family used by the Farmer is discrete and hierarchical:

| Model | Description | Typical role |
|---|---|---|
| PointSource | PSF model | Unresolved objects |
| SimpleGalaxy | Circular exponential, fixed size | Marginally resolved sources |
| ExpGalaxy | Exponential profile | Resolved disks |
| DevGalaxy | de Vaucouleurs profile | Centrally concentrated galaxies |
| CompositeGalaxy | Sum of exponential and de Vaucouleurs profiles | Full galaxy model |

Model selection is governed by diagnostics such as per-source reduced chi-square, with progressively more complex models adopted only when statistically justified. Forced photometry then fixes the position and shape determined from higher-resolution data and re-optimizes only the flux in lower-resolution bands. This design is especially important for IRAC imaging and for crowded regions where aperture methods degrade rapidly [2310.07757].

## 3. Redshifts, physical parameters, and released products

COSMOS2020 computes photometric redshifts for all sources in each catalog using two independent codes, LePhare and EAZY. In the Farmer branch, the photometric redshift performance reaches sub-percent accuracy for \(i<21\), remains sub-percent for \(21<i<22.5\), is \(\sigma_{\rm NMAD}\approx 0.015\)–0.025 with outlier fractions of \(\sim 2\)–4\% for \(22.5<i<25\), and is \(\sim 0.04\)–0.05 with outlier fractions of \(\sim 14\)–20\% for \(25<i<27\). Compared to COSMOS2015, COSMOS2020 reaches the same photometric redshift precision at almost one magnitude deeper [2110.13923].

The Farmer catalog also propagates a large set of derived quantities. These include high-precision astrometry, total fluxes and errors in more than 30 bands, morphological parameters such as model profile, effective radius, axis ratio, and position angle for resolved sources, goodness-of-fit diagnostics, photometric redshifts from both LePhare and EAZY with confidence intervals and PDFs, and physical parameters from template SED fitting such as stellar mass, rest-frame magnitudes and colors, star-formation rate, stellar population parameters, and source classification flags. Mask flags and ancillary matches, including GALEX, Chandra, and HST/ACS where available, are پڻ included [2110.13923].

An important methodological nuance is that the Farmer’s native photometric errors are reported as smaller and smoother than those of the Classic catalog, but for SED fitting they were scaled by a factor of \(2\times\) to match spectroscopic performance. This does not negate the photometric model-fitting framework; it reflects the difference between formal model-based uncertainties and empirically calibrated redshift-confidence behavior [2110.13923].

Public release is through standard astronomical archive systems, including ESO Phase 3, IPAC IRSA, CDS, and the Institut d’Astrophysique de Paris COSMOS2020 site. The data are distributed as FITS tables with extensive documentation and additional files for photometric-redshift PDFs [2110.13923].

## 4. Validation and comparative performance

The Farmer was benchmarked on realistic COSMOS-like images with real galaxy morphologies from HST, injected point sources, matched depths, and matched PSFs. In these tests, the photometric bias is reported as \(<0.05\) mag across all bands to the survey limit, including Spitzer/IRAC. Number counts are smoothly recovered up to the nominal limiting magnitude without artificial plateaus or spikes, and shape parameters such as effective radius, axis ratio, and position angle are recovered to within 1% for resolved galaxies [2310.07757].

The validation results clarify where the Farmer improves on fixed-aperture methods and where its gains are conditional. Joint fitting of grouped sources is described as essential: simulations show that simultaneous modeling succeeds in deblending crowded sources, whereas independent or iterative subtraction approaches systematically fail. Aperture photometry begins to fail at moderate source densities, while the Farmer remains accurate deeper into the confusion-limited regime. At the same time, a hard limitation remains: if sources are not separated at the detection or segmentation stage, the Farmer cannot deblend them later [2310.07757].

Comparison with the Classic catalog is correspondingly nuanced rather than categorical. Farmer and Classic magnitudes and colors agree to \(<0.1\) mag everywhere. At brighter magnitudes, the Classic catalog is marginally better in photometric-redshift performance, while at the faintest limits the Farmer is superior, largely because of improved deblending and total-flux estimation. The Farmer therefore should not be understood as a universal replacement for aperture photometry; it is a complementary measurement regime whose principal advantages emerge for faint, blended, and crowded sources [2110.13923].

## 5. Scientific uses

The Farmer catalog has been used directly in several lines of COSMOS science because its deblending and total-flux estimates alter source recoverability in precisely the regimes where crowding and confusion dominate. In the search for \(z\geq 7.5\) galaxies, COSMOS2020 combined Classic and Farmer photometry, and four candidates were selected that would be rejected using fixed aperture photometry. That study identified 17 new \(7.5<z<10\) candidate sources and confirmed 15 previously published candidates, illustrating how the Farmer changes high-redshift sample construction rather than merely refining fluxes within an already fixed sample [2207.11740].

The Farmer version of COSMOS2020 has also been central to large-scale-structure work. A systematic search for protocluster candidates at \(z\geq 6\) used the Farmer catalog and found 15 significant (\(>4\sigma\)) candidate galaxy overdensities across \(6\leq z\leq 7.7\). For lower redshift large-scale structure, the density field was reconstructed from the Farmer catalog for a magnitude-limited \(K_s<24.5\) sample of \(\sim 210\,k\) galaxies at \(0.4<z<5\), using weighted kernel density estimation with a von Mises-Fisher kernel and explicit corrections for masked regions and edges [2210.17334, 2312.10222].

The Farmer data products have likewise supported galaxy-population inference beyond direct catalog use. A self-organizing-map approach to estimating stellar mass and star-formation rate was trained and calibrated on COSMOS2020 specifically because it provided the required high-quality panchromatic data set. More recently, a supervised CatBoostClassifier was trained for quiescent-versus-star-forming classification in the Farmer catalog using 28 color features from 8 mutual photometric bands, and when applied to COSMOS2020 it classified 365,877 galaxies within the adopted cuts [2206.06373, 2509.03039].

These applications suggest a general pattern: the Farmer catalog is most consequential where total-flux recovery, profile-based deblending, and UV-to-IR consistency materially affect object detection, redshift inference, or environment reconstruction.

## 6. Limitations, extensions, and continuing role

Several limitations recur across work that uses the Farmer catalog. First, the catalog’s advantages are strongest in crowded and faint-source regimes, not uniformly across all magnitudes. Second, photo-\(z\) uncertainties and mass completeness become progressively more restrictive at high redshift in environment studies, diluting environmental metrics and making quenching statistics noisier. Third, the deblending framework remains bounded by the initial detection image: unresolved blends at that stage are a fundamental failure mode rather than a post-processing nuisance [2110.13923, 2310.07757, 2411.09522].

Subsequent work has extended the Farmer catalog into domain-specific value-added products. The machine-learning quiescent classification for the Farmer sample reported an F1-score of 89% for quiescent galaxies on the mock test data, compared with 54% for SED fitting, and predicted a systematically higher fraction of quiescent galaxies across \(0.2<z<3.5\) than NUVrJ or SED-based selections. In a different direction, the “FARMER LP” subset of COSMOS2020 v4.1.1 over \(1.44\,\mathrm{deg}^{2}\) was used to build empirical CO and [CII] line-intensity cubes for masking forecasts, with the explicit caveat that predictions based purely on the empirical catalog are conservative lower limits because of incompleteness [2509.03039, 2606.17210].

The enduring importance of the Farmer COSMOS2020 Catalog lies in this combination of breadth and methodological specificity. It is not merely a source list derived from COSMOS imaging, but a profile-fitting measurement framework tied to a public multi-band release, a dual-photo-\(z\) inference pipeline, and a growing ecosystem of downstream catalogs and analyses. Its technical identity is defined by reproducible joint modeling, native total photometry, and explicit deblending; its scientific identity is defined by the classes of problems for which those properties are decisive [2110.13923, 2310.07757].

Source: https://www.emergentmind.com/topics/farmer-cosmos2020-catalog