Data-Mining for Astronomy

Submitted by Dr. Michael J. Way, NASA Goddard Institute for Space Studies.

Given the recent release of the Galaxy Zoo Data Release 1 researchers can begin to explore the myriad ways that one can use the most accurate and numerous database of galaxy morphology ever compiled. To that end we have used galaxy photometry and redshift information from the Sloan Digital Sky Survey in combination with precise knowledge of galaxy morphology via the Galaxy Zoo project to calculate photometric redshifts using Gaussian process regression.

We are primarily interested in obtaining accurate photometric redshifts for a subset of SDSS galaxies called the luminous red galaxies. These galaxies are normally found in denser regions of the local universe. They are interesting because they tend to be accurate tracers of the large scale structure in the universe and have been used for measuring the Baryonic Acoustic Oscillation signal thus putting better constraints on present day cosmological models.

The Galaxy Zoo database is used to segregate the elliptical galaxies from the spirals (we focus on the former). Then we obtain a variety of derived primary and secondary isophotal shape estimates from the Sloan Digital Sky Survey imaging catalog (e.g. the amount of light within the 50% Petrosian radius). Using these shape estimates in combination with the five bandpass photometry of elliptical galaxies with redshifts from the SDSS we using a non-linear regression training set method (Gaussian process regression) to estimate their photometric redshifts.  The root mean square error for luminous red galaxies classified as ellipticals is as low as 0.0118 which is nearly a factor of 2 lower than typical estimates for galaxies in the SDSS (See Figure).

One can see in the lower left panel that estimates of the photometric redshift are lowest for the luminous red galaxies classified as ellipticals. The best results are obtained when using their 5-band photometry and a variety of isophotal shape estimates denoted as B. See the paper on arXiv for more details.

The next step will be to use classification techniques from the Machine Learning literature to classify all of the elliptical galaxies in the ~350 million object database of the SDSS. This has already been attempted by one group using the ~900,000 Galaxy Zoo morphologies and isophotal shape estimates as training samples. One would expect to be  able to classify approximately 50-100 million luminous red galaxies as ellipticals. These in turn can be used as the most accurate probes thus far in estimating Baryonic Acoustic Oscillations at unprecedented depth.