Novel Methods for Predicting Photometric Redshifts from Broadband Photometry Using Virtual Sensors

We calculate photometric redshifts from the Sloan Digital Sky Survey MainGalaxy Sample, The Galaxy Evolution Explorer All Sky Survey, and The Two MicronAll Sky Survey using two new training-set methods. We utilize the broad-bandphotometry from the three surveys alongside Sloan Digital Sky Survey measuresof photometric quality and galaxy morphology. Our first training-set methoddraws from the theory of ensemble learning while the second employs Gaussianprocess regression both of which allow for the estimation of redshift alongwith a measure of uncertainty in the estimation. The Gaussian process modelsthe data very effectively with small training samples of approximately 1000points or less. These two methods are compared to a well known ArtificialNeural Network training-set method and to simple linear and quadraticregression. Our results show that robust photometric redshift errors as low as0.02 RMS can regularly be obtained. We also demonstrate the need to provideconfidence bands on the error estimation made by both classes of models. Ourresults indicate that variations due to the optimization procedure used foralmost all neural networks, combined with the variations due to the datasample, can produce models with variations in accuracy that span an order ofmagnitude. A key contribution of this paper is to quantify the variability inthe quality of results as a function of model and training sample. We show howsimply choosing the "best" model given a data set and model class can producemisleading results.Comment: 36 pages, 12 figures, ApJ in Press, modified to reflect published version and color figure