Astro Lunch

True Bayesian redshifts in the Dark Energy Survey

Some of the most important current and future cosmological missions, aiming to make high precision measurements of the large-scale structure of galaxies and its weak gravitational lensing effects, will rely on photometry of galaxies for successfully accomplishing their goals. Therefore, the redshift of galaxies in such missions, arguably one of the most important observables, needs to be derived from photometry only, with the so-called photometric redshift technique. In this talk, I will describe the current plan for deriving the redshift distribution estimates and their uncertainties for the populations of galaxies used for weak lensing (WL) in the upcoming analyses of the Dark Energy Survey (DES). We propose a Bayesian scheme to use the effectively noiseless DES deep-field galaxies (from Supernova fields) as an intermediate step between spectroscopic or many-band training/validation samples, and the target sample (WL source galaxies). This scheme provides a comprehensible way to separate the measurement-noise part from the color-redshift relation in the problem, and allows us to use all the information at hand in both ends. A self-organizing map (SOM) using photometry from deep-field galaxies is used to characterize the DES multi-color space, and populated with galaxies with accurate redshift information to constrain the DES color-redshift relation. Then, for the measurement-noise half, we empirically estimate the connection between wide (noisy) and deep (noiseless) photometries, including effects of WL selection, using the Balrog software, which inserts fake objects in real imaging to accurately characterize measurement biases. The explicit character of the framework also enables an easy propagation of systematic uncertainties affecting the different ingredients, and allows for the addition of clustering information into the redshift inference problem in a natural way.