Climbing the Cliffs: Classifying YSOs in the Cosmic Cliffs JWST Data using a Probabilistic Random Forest

TL;DR

This study uses a Probabilistic Random Forest machine learning approach to identify and analyze young stellar objects in the JWST Cosmic Cliffs data, revealing a larger and more detailed population than previous surveys.

Contribution

It introduces a novel application of probabilistic machine learning to classify YSOs in JWST data, uncovering many new candidates and sub-stellar objects.

Findings

Identified 450 candidate YSOs, 413 of which are new detections.

YSO surface density correlates with Herschel-derived column densities.

Detected a significant population of sub-stellar YSOs, implying higher star formation efficiencies.

Abstract

Among the first observations released to the public from the James Webb Space Telescope (JWST) was a section of the star-forming region NGC 3324 known colloquially as the "Cosmic Cliffs." We build a photometric catalog of the region and test the ability of using the Probabilistic Random Forest machine learning method to identify its Young Stellar Objects (YSOs). We find 450 candidate YSOs (cYSOs) out of 19~497 total objects within the field, 413 of which are cYSOs not found in previous works. These classifications are verified with several different metrics, including recall and precision. Using the obtained probabilities of objects being YSOs, we employ a Monte Carlo approach to determine the surface density of cYSOs in the Cosmic Cliffs, which we find to be largely coincident with column densities derived from Herschel data, up to a column density of 1.37 $\times$ 10$^{22}$ cm$^{-2}$. The newly determined number and spatial distribution of YSOs in the Cosmic Cliffs demonstrate that JWST is far more capable of detecting YSOs in dusty regions than Spitzer. Comparisons of the observed colors and brightness of faint cYSOs with those of pre-main-sequence models suggest JWST has detected a significant population of sub-stellar YSOs in the Cosmic Cliffs. The size of this population further suggests previous estimates of star formation efficiencies in molecular clouds have been systematically low.