COSMOS-Web: estimating physical parameters of galaxies using Self-Organizing Maps
Authors:
- Fatemeh AbediniFAFatemeh Abedini
- Ghassem GozaliaslGGGhassem Gozaliasl
- Akram Hasani ZonooziAZAkram Hasani Zonoozi
- Atousa KalantariAKAtousa Kalantari
- Olivier IlbertOIOlivier Ilbert
- Hollis B AkinsHAHollis B Akins
- Natalie AllenNANatalie Allen
- Rafael C Arango-ToroRARafael C Arango-Toro
- Caitlin M CaseyCCCaitlin M Casey
- Nicole E DrakosNDNicole E Drakos
- Andreas L FaisstAFAndreas L Faisst
Open Access
- 23 February 2026
- journal article
- Published by Oxford University Press (OUP) in Monthly Notices of the Royal Astronomical Society
- Vol. 548 (1)
Abstract
The COSMOS (Cosmic Evolution Survey)-Web survey, with its unparalleled combination of multiband data, notably, near-infrared imaging from James Webb Space Telescope (JWST)’s Near-InfraRed Camera (NIRCam) (F115W, F150W, F277W, and F444W), provides a transformative data set down to $\sim 28$ mag (F444W) for studying galaxy evolution. In this work, we employ Self-Organizing Maps (SOMs), an unsupervised machine learning method, to estimate key physical parameters of galaxies – redshift, stellar mass, star formation rate (SFR), specific SFR (sSFR), and age – directly from photometric data out to $z=3.5$. SOMs efficiently project high-dimensional galaxy colour information onto 2D maps, showing how physical properties vary among galaxies with similar spectral energy distributions. We first validate our approach using mock galaxy catalogues from the HORIZON-AGN (HORIZON simulation including Active Galactic Nuclei feedback) simulation, where the SOM accurately recovers the true parameters, demonstrating its robustness. Applying the method to COSMOS-Web observations, we find that the SOM delivers robust estimates despite the increased complexity of real galaxy populations. Performance metrics ($\sigma _{\mathrm{NMAD}}$ typically between 0.1–0.3, and Pearson correlation between 0.7 and 0.9) confirm the precision of the method, with $\sim$ 70 per cent of predictions within 1$\sigma$ dex of reference values. Although redshift estimation in COSMOS-Web remains challenging (median $\sigma _{\mathrm{NMAD}} = 0.04$), the overall success of the highlights its potential as a powerful and interpretable tool for galaxy parameter estimation. A key advance of this work is the use of JWST/NIRCam photometry, particularly the F444W band, which enhances SOM training and allows more accurate estimation of stellar mass, SFR, and age compared to previous studies using Infrared Array Camera (IRAC)/Spitzer filters.
Keywords
All Related Versions
Funding Information
- Centre National d’Etudes Spatiales (ANR-22-CE31-0007)
This publication has 115 references indexed in Scilit:
- A review of unsupervised learning in astronomyAstronomy and Computing, 2024
- SciPy 1.0: fundamental algorithms for scientific computing in PythonNature Methods, 2020
- The many flavours of photometric redshiftsNature Astronomy, 2018
- Cosmology and fundamental physics with the Euclid satelliteLiving Reviews in Relativity, 2018
- Fast automated analysis of strong gravitational lenses with convolutional neural networksNature, 2017
- Fitting the integrated spectral energy distributions of galaxiesAstrophysics and Space Science, 2010
- Stellar population synthesis at the resolution of 2003Monthly Notices of the Royal Astronomical Society, 2003
- Measuring the redshift evolution of clustering: the Hubble Deep Field SouthMonthly Notices of the Royal Astronomical Society, 2002
- Self-Organizing Maps2001
- Introduction to Multivariate Analysis1980