Campus, Varian 355
A key challenge for upcoming “Stage IV" weak lensing cosmology experiments (LSST, Euclid,WFIRST) will be measuring accurate redshifts for billions of faint galaxies using only broad-band photometric observations. Specifically, dark energy constraints from weak lensing cosmology depend on excellent characterization of the N(z) distributions of galaxies in ~10-20 cosmic shear bins. I will describe our development of an unsupervised machine learning-based technique (using the “self-organizing map”) to map the empirical high-dimensional galaxy color space (u-g, g-r, …, J-H) expected for Euclid and WFIRST. Importantly, this technique allows us to determine where (in galaxy color space) we currently have existing spectroscopy, and where it is lacking. The incomplete/heterogeneous sampling of the galaxy color space is an important weakness for traditional machine learning photo-z techniques. This analysis inspired the Complete Calibration of the Color-Redshift Relation (C3R2) survey - a multi-institution, multi-instrument survey with the Keck telescopes aimed at mapping out the empirical galaxy color-redshift relation in preparation for Euclid and WFIRST. C3R2 is a joint effort involving all of the Keck partners, with 44.5 nights allocated thus far. I will describe the status of the C3R2 survey, application of the technique and some its challenges, and prospects for the future.