Bayesian Single-Epoch Photometric Classification of Supernovae

(abridged) Ongoing supernova (SN) surveys find hundreds of candidates, that require confirmation for their use. Traditional classification based on followup spectroscopy of all candidates is virtually impossible for these large samples. We present an automatic Bayesian classifying algorithm for supernovae, the SN-ABC. We rely solely on single-epoch multiband photometry and host-galaxy (photometric) redshift information to sort SN candidates into the two major types, Ia and core-collapse supernovae. We test the SN-ABC performance on published samples of SNe from the SNLS and GOODS projects that have both broad-band photometry and spectroscopic classification (so the true type is known). The SN-ABC correctly classifies up to 97% (85%) of the type Ia (II-P) SNe in SNLS, and similar fractions of the GOODS SNe, depending on photometric redshift quality. We further test our method on large artificial samples to explore possible biases, and find that, in deep surveys, SNe Ia are best classified at redshifts z >~ 0.6, or when near maximum, with success rates of the order of 95%. Core-collapse SNe are best recognized several weeks after maximum, or at z <~ 0.6, and reach success rates near 80%. The SN-ABC also allows the rejection of SN "impostors" such as active galactic nuclei (AGNs), with half of of the AGNs we simulate rejected by the algorithm. In a separate paper, we use our method to classify a new sample of high-z SNe that we have found in the Subaru Deep Field.