This work package will be devoted to Bayesian measurement of photo-z from multi-band images, extracting for each galaxy a p(z) distribution, directly from the blended images (see Jones and Heavens 2019). Compared to the state of the art, which uses template fitting or neural networks applied to colors (difference between the measured magnitudes in two bandpass) to derive a value of the redshift with a (Gaussian) uncertainty, and is prone to catastrophic errors where degeneracies yield to selecting a bad solution, we plan to work directly at the pixel level, on multiband image cubes, and output a p(z) probability distribution, using Bayesian networks.
Some work has already been done on that topic, within LSST-France (Pasquet et al. 2019), but using deep convolutional networks that output probability distribution functions, considering the problem as a classification problem, rather than Bayesian networks. We plan to use Bayesian techniques on multicolor image cubes, using both LSST 6-band images, and LSST+Euclid 10-band images.
This work will be led by APC with contributions from LORIA