Dust Extinction Measures for ~8 Galaxies using Machine Learning on JWST Imaging - Abstract

Aug 4, 2026

Upload by Violet

Machine learning is used to measure dust content in JWST-observed galaxies at z > 6, leveraging trained neural networks built from high-resolution IllustrisTNG simulations. Dust is treated as a key unknown in early galaxy evolution and as a degeneracy source in spectral energy distribution (SED) fitting. A new SED-independent method predicts dust attenuation and sSFR via supervised regression and convolutional neural networks trained with dust parameterized by E(B-V) and A(V).

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Abstract

Introduction

Dust effects on early galaxies

Dusty star-forming galaxies and observational challenges

Motivation for accurate high-redshift dust measurements

arXiv :2403 . 18458v1 [ astro-ph .GA] 27 Mar 2024 Dust Extinction Measures for 􀁉 ∼ 8 Galaxies using Machine Learning on JWST Imaging Kwan Lin Kristy Fu, 1★ Christopher J. Conselice, 1 Leonardo Ferreira,2 Thomas Harvey, 1 Qiao Duan, 1 Nathan Adams, 1 Duncan Austin 1 1 Jodrell Bank Centre for Astrophysics, University of Manchester, Oxford Road, Manchester, UK 2 School of Physics and Astronomy, University of Victoria, Victoria, BC, Canada Accepted XXX. Received YYY; in original form ZZZ ABSTRACT We present the results of a machine learning study to measure the dust content of galaxies observed with JWST at z > 6 through the use of trained neural networks based on high-resolution IllustrisTNG simulations. Dust is an important unknown in the evolution and observability of distant galaxies and is degenerate with other stellar population features through spectral energy fitting. As such, we develop and test a new SED-independent machine learning method to predict dust attenuation and sSFR of high redshift (z > 6) galaxies. Simulated galaxies were constructed using the IllustrisTNG model, with a variety of dust contents parameterized by E(B-V) and A(V) values, then used to train Convolutional Neural Network (CNN) models using supervised learning through a regression model. We demonstrate that within the context of these simulations, our single and multi-band models are able to predict dust content of distant galaxies to within a 1􀁦 dispersion of A(V) ∼ 0. 1. Applied to spectroscopically confirmed z > 6 galaxies from the JADES and CEERS programs, our models predicted attenuation values of A(V) < 0.7 for all systems, with a low average (A(V) = 0.28) . Our CNN predictions show larger dust attenuation but lower amounts of star formation compared to SED fitted values. Both results show that distant galaxies with confirmed spectroscopy are not extremely dusty, although this sample is potentially significantly biased. We discuss these issues and present ideas on how to accurately measure dust features at the highest redshifts using a combination of machine learning and SED fitting. Key words: keyword1 – keyword2 – keyword3 1 INTRODUCTION Interstellar dust represents potentially a significant unknown in observing early galaxies in optical and near-infrared wavelengths (e.g., Casey et al. 2014; Fudamoto et al. 2020) . Due to dust effects, such as extinction and the scattering of light, images of early galaxies are often obscured, and properties such as dust composition (Calzetti et al. 1994; Min et al. 2006; Spoon et al. 2006; Dwek et al. 2014), AGN (Treister et al. 2010), and metallicity (Shivaei et al. 2020) can affect spectral energy distribution modelling accuracy (e.g., Juodžbaliset al. 2023) . On the other hand, solid measurements of the dust content or reddening can help build accurate models of galaxies, and are important in our understanding of galaxy evolution. We are now in a position, with ALMA and JWST, to learn more about dust in the early universe and how dust affects the output of light from early galaxies. However, there remains a significant amount we do not yet understand about dust in the early universe, in part due to the difficulties of observing this feature within faint high-redshift galaxies in the absence of rest-frame far-IR light. New approaches and ideas are needed to trace this aspect in the early universe. Observations starting from IRAS and COBE in 1980s and 90sand leading into JCMT/SCUBA, SMO, and ALMA, among others, ★ [E-mail: kwanlinkristy.fu@postgrad.manchester.ac.uk](E-mail: kwanlinkristy.fu@postgrad.manchester.ac.uk) (KTS) have identified many high-redshift galaxies which are faint or invisible in the optical but bright in the far-IR/sub mm (e.g., Casey et al. 2014; Hodge & da Cunha 2020; Dayal et al. 2022) . These galaxies are sometimes known as Dusty Star-Forming Galaxies (DSFGs) due to being enshrouded in dust, which obscures observations at optical and UV wavelengths. Observed

FAQs

How does the study estimate dust content for galaxies at z > 6?

It trains neural networks using high-resolution IllustrisTNG simulations and applies supervised regression to predict dust attenuation and sSFR from JWST imaging.

Why is dust extinction difficult to measure in early, high-redshift galaxies?

Dust causes extinction and scattering, and rest-frame far-IR light is often difficult to observe in faint high-redshift systems, leaving major uncertainty.

What is the relationship between the proposed machine learning results and SED fitting?

The CNN predictions indicate larger dust attenuation but lower star formation compared to values inferred from SED fitting, highlighting differences tied to degeneracies in spectral modeling.

Aug 4, 2026

Upload by Violet