![School of Natural Sciences Event](/sites/default/files/styles/two_column_medium/public/2019-09/sns_default.jpg?itok=IEu1CLXj)
Princeton University Thunch Talk
Probabilistic Component Separation: Deconstructing Photometric and Spectroscopic Pipelines
A ubiquitous problem in astronomy is correctly assigning absorption/emission in an image/spectrum to the multiple processes occurring along the line of sight within the field of view. We introduce a method to decompose the images/spectra, with full posteriors on the joint distribution of the components.
The decomposition is obtained by modeling each component as a draw from a high-dimensional Gaussian distribution in the data-space (the observed image/spectrum)---a method we call “Marginalized Analytic Data-space Gaussian Inference for Component Separation” (MADGICS). This technique provides statistically rigorous uncertainties and detection thresholds, which allows better leveraging of low signal-to-noise data.
I will discuss the application of these component separation techniques at scale to several of the largest photometric and spectroscopic surveys: the Dark Energy Camera Plane Survey (DECaPS2), Gaia Radial Velocity (RVS) spectra, and SDSS APOGEE spectra. We focus on applying these techniques in near-infrared wavelengths so as to penetrate through dust in the interstellar medium and map its spatial and chemical complexity at Galactic scales.