Improved photometric redshifts with surface luminosity priors

We apply Bayesian statistics with prior probabilities of galaxy surface luminosity (SL) to improve photometric redshifts. We apply the method to a sample of 1266 galaxies with spectroscopic redshifts in the GOODS North and South fields at 0.1 ≲ z ≲ 2.0. We start with spectrophotometric redshifts (SPZs) based on Probing Evolution and Reionization Spectroscopically grism spectra, which cover a wavelength range of 6000-9000 , combined with (U)BViz(JHK) broadband photometry in the GOODS fields. The accuracy of SPZ redshifts is estimated to be σ(Δ(z)) = 0.035 with an systematic offset of -0.026, where Δ(z) = Δz/(1 + z), for galaxies in redshift range of 0.5 ≲ z ≲ 1.25. The addition of the SL prior probability helps break the degeneracy of SPZ redshifts between low redshift 4000 break galaxies and high-redshift Lyman break galaxies which are mostly catastrophic outliers. For the 1138 galaxies at z ≲ 1.6, the fraction of galaxies with redshift deviation Δ(z)>0.2 is reduced from 15.0% to 10.4%, while the rms scatter of the fractional redshift error does not change much.