Oral Presentation
SCUBA-2 Ultra Deep Imaging EAO Survey (STUDIES): Confusion-limited 450 μm Galaxy Number Counts and Machine Learning Identifications
Presenter: Zhen-Kai Gao (ASIAA/NCU)
The SCUBA-2 Ultra Deep Imaging EAO Survey (STUDIES) is a JCMT large program designed to study the dust-obscured portion of cosmic star formation. To achieve this goal, we conduct observations at 450 μm in the COSMOS and SXDS fields, with the goal of reaching the confusion limit. To date, we have completed 330 hours of observations in the COSMOS field and have acquired approximately 37% (119 hours) of the allocated time in the SXDS field.
Using our observations and archival data, we have created a confusion-limited 450 μm map of the COSMOS field covering an area of approximately 450 arcmin2 with a noise level of around 0.6 mJy at the center. From this map, we have identified 477 sources at 3.5σ with flux densities ranging from 2 to 45 mJy, corresponding to a star formation rate range of ∼100 to ∼2000 solar masses per year at z=2, forming the deepest 450 μm galaxy samples to date. The samples allow us to construct deep source counts at 450 μm. We used Monte Carlo simulations to derive the intrinsic number counts and found that the trend of the faint-end counts shows no drop-off or flattening out. We also compared our counts to model predictions and found reasonable consistency from the faint end to around 20 mJy, but a steeper turn at the bright end in observations. In addition, we compared the integrated surface brightness down to 2.1 mJy of our counts with that measured by COBE and found that our confusion-limited SCUBA-2 image can directly resolve 41% of the 450 μm COBE extragalactic background light (EBL). Based on the results of the trend of the faint-end counts and the unresolved EBL, we conclude that there are still many faint (< 2 mJy) 450 μm sources that are unresolved by current SCUBA-2 imaging.
Finally, we present our preliminary results on using machine learning to identify 450 μm-selected submillimeter galaxies (SMGs) from the COSMOS2020 catalog. By using counterparts of our 450 μm sources identified with ALMA and radio observations, we trained a model with a high level (> 0.9) of precision, recall, and f1 scores. This model allows us to identify a large sample of SMG candidates, which will be useful for studying the spatial clustering, physical properties, and morphologies of this population, as well as for source de-blending in low-resolution images.
asiaa.sinica.edu.tw