MSc Thesis Defense: Efe ÖZTABAN, MACHINE LEARNING DRIVEN DETERMINATION OF THREE-DIMENSIONAL STRUCTURES OF MOLECULAR CLOUDS IN HIGH-RESOLUTION MILLIMETER IMAGES, Date & Time: May 12, 2026 – 1:00 PM , Place: FENS L067
MACHINE LEARNING DRIVEN DETERMINATION OF
THREE-DIMENSIONAL STRUCTURES OF MOLECULAR CLOUDS
IN HIGH-RESOLUTION MILLIMETER IMAGES
Efe Öztaban
Physics, MSc Thesis, 2026
Thesis Jury
Prof. Emrah Kalemci (Thesis Advisor)
Asst. Prof. Arkadaş İnan Özakın
Prof. Ünal Ertan
Date & Time: May 12th, 2026 – 1:00 PM
Place: FENS L067
Keywords : Dust scattering halos, Molecular clouds, 4U 1630−47,
Three-dimensional reconstruction, Machine learning
Abstract
Dust scattering halos offer a powerful geometric probe of highly absorbed Galactic X-ray binaries by constraining distances important for BHXRB characterization, but their interpretation depends strongly on the structure of the intervening interstellar material. In this thesis, high-resolution APEX 12CO observations are used to study the molecular clouds along the line of sight to 4U 1630−47, whose distance has remained uncertain because of strong extinction and complex foreground gas. The narrow field of view and crowded velocity structure of the APEX data make standard cloud identification methods insufficient for this problem. Therefore, a dedicated reconstruction pipeline is developed to extract three-dimensional molecular cloud morphologies from the data by combining spectral peak identification, segmentation, thresholding, and soft clustering. The reconstruction separates the emission into 15 molecular cloud components and traces their three-dimensional morphologies in position-position-velocity space. When these clouds are incorporated into dust scattering halo modeling and compared with the observed Chandra halo image, the results favor a source distance of approximately 11.5 kpc over the alternative 4.85 kpc and 13.6 kpc solutions, as presented in Kalemci et al. 2025. This work shows that detailed molecular cloud morphology can play a central role in dust scattering halo modeling and in improving distance estimates for highly absorbed X-ray binaries.