Abstract

Poster - Splinter EScience

Thursday, 12 September 2024, 15:15   (S13)

Probabilistic Reconstruction of Spectra from Photometry

Johanna Riedel, Nikos Gianniotis, Kai Polsterer
HITS gGmbH

Spectroscopic data contains significantly more information than photometric data, however spectroscopy observations are more costly. Additionally, comparing photometry from different filter systems is not always a straightforward task. We adopt a probabilistic approach that allows to reconstruct the most probable full spectrum from photometry with various filter systems. Our approach is based on Probabilistic Principal Component Analysis (PPCA), but has the advantage of accounting for heterogeneous uncertainties of the wavelengths. This model produces a general representation of spectra that can be used to map a spectrum in a latent space with fewer dimensions. Additionally, we can reversely map a point from this latent space back to a spectrum, and obtain a reconstruction of the most likely full spectrum. By taking advantage of the linearity of PPCA and the folding with the photometric filters, our model can utilize any filter system as well as combine multiple filter systems.