Photometric redshift determination is crucial to the success of dark energy missions such as LSST and DES. A KIPAC postdoc has developed an important tool for photometric redshift estimation and applied it to 60 million galaxies from the Sloan Digital Sky Survey.
Six randomly chosen galaxies' photometric redshift probability distributions from Cunha and colleagues' SDSS galaxy sample.
The major current and upcoming probes of dark energy are large astronomical sky surveys that observe billions of galaxies, hundreds of thousands of supernovae, and thousands of galaxy clusters. These include the currently deployed Dark Energy Survey (DES) and the future Large Synoptic Survey Telescope (LSST), both of which have major KIPAC participation. The dark energy constraints from any such survey depend critically on reliable determinations of the redshifts of the objects in question.
Redshifts are a measure of how much an object's light has shifted to longer wavelengths due to the expansion of the Universe. The traditional technique of redshift determination - taking a full spectrum of each object and observing the precise wavelength of emission and absorption lines - is impossible for sky surveys with billions of galaxies. Instead, the surveys must rely on photometric redshifts, known as "photo-zs", where the redshift of each object is estimated from its brightness in a few broadband colors, defined by filters on the telescope. Because photo-zs are estimated with a relatively small amount of information, and distant galaxies can differ markedly in their properties, errors in photo-z estimation are a major concern, and photo-z estimation techniques are a prominent topic of investigation in the astronomical community.
So-called 'empirical' methods for photo-z estimation involve comparing the colors of a galaxy for which the redshift is desired to those of galaxies with known redshifts. The most bias-free, and therefore most accurate estimations, can be achieved only when the relatively small set of galaxies with known redshifts - the training set - matches the properties of the large set of unknown galaxies in every significant way, or when an accurate correction is applied to account for the differences.
A method of doing exactly that was developed was developed by current KIPAC postdoc and Kavli Fellow Carlos Cunha along with a few colleagues from the University of Chicago and Fermilab several years ago. The technique involves 'weighting' - or assigning proportional importance - such that each galaxy in the training set's importance is proportional to how many galaxies in the unknown set are near to it in the multi-dimensional space of colors. The redshift probability distribution for any unknown galaxy is then a weighted average of the training set galaxies that have colors most like it, and the overall redshift distribution of the unknown galaxies matches the weighted redshift distribution of the training set.
The weighting technique of Cunha and colleagues naturally gives the full photo-z probability distribution for each galaxy, and not just the most likely redshift. This has the obvious advantage of carrying extra information that can be included in the determinations of cosmological quantities from surveys with large numbers of galaxies. Cunha and colleagues have now applied their technique to data from the Sloan Digital Sky Survey (SDSS) and provided individual redshift probability distributions for 60 million galaxies from SDSS's Data Release 8. While these photo-z distributions will be useful for those working with this round of SDSS data, the real importance of the photo-z technique demonstrated lies in its potential benefit for the science of dark energy constraining surveys such as DES and LSST.
This work is described in part in a paper submitted to Monthy Notices of the Royal Astronomical Society (MNRAS) and available from astro-ph at arXiv:1109.5192. Dr. Cunha's algorithms are publically available and are described in a paper published in MNRAS (2009, 396, 2379).