Estimating Photometric Redshifts with Artificial Neural Networks and Multi-Parameters

We calculate photometricredshifts from the Sloan Digital Sky SurveyData Release 2 (SDSS DR2) Galaxy Sample using artificial neural networks (ANNs). Different input sets based on various parameters (e.g. magnitude, color index, flux information) are explored. Mainly, parameters from broadband photometry are utilized and their performances in red- shift prediction are compared. While any parameter may be easily incorporated in the input, our results indicate that using the dereddened magnitudes often produces more accurate pho- tometric redshifts than using the Petrosian magnitudes or model magnitudes as input, but the model magnitudes are superior to the Petrosian magnitudes. Also, better performance re- sults when more effective parameters are used in the training set. The method is tested on a sample of 79346 galaxies from the SDSS DR2. When using 19 parameters based on the dereddened magnitudes, the rms error in redshift estimation is σz = 0.020184. The ANN is highly competitive tool compared to the traditional template-fitting methods when a large and representative training set is available.

Estimating Photometric Redshifts with Artificial Neural Networks and Multi-Parameters