Simultaneous Estimation of Photometric Redshifts and SED Parameters: Improved Techniques and a Realistic Error Budget

TL;DR

This paper introduces improved techniques for joint galaxy photometric redshift estimation and SED fitting, emphasizing accurate uncertainty estimation, validation methods, and the benefits of flexible models for better parameter recovery.

Contribution

It presents a new annealing method for re-calibrating joint uncertainties and demonstrates the importance of flexible models and realistic star formation histories for improved parameter estimation.

Findings

Uncertainties on redshifts and SED parameters are correctly correlated.

Over- or under-estimated redshift uncertainties lead to similar biases in SED parameters.

Flexible models significantly improve the recovery of stellar ages.

Abstract

We seek to improve the accuracy of joint galaxy photometric redshift estimation and spectral energy distribution (SED) fitting. By simulating different sources of uncorrected systematic errors, we demonstrate that if the uncertainties on the photometric redshifts are estimated correctly, so are those on the other SED fitting parameters, such as stellar mass, stellar age, and dust reddening. Furthermore, we find that if the redshift uncertainties are over(under)-estimated, the uncertainties in SED parameters tend to be over(under)-estimated by similar amounts. These results hold even in the presence of severe systematics and provide, for the first time, a mechanism to validate the uncertainties on these parameters via comparison with spectroscopic redshifts. We propose a new technique (annealing) to re-calibrate the joint uncertainties in the photo-z and SED fitting parameters without compromising the performance of the SED fitting + photo-z estimation. This procedure provides a consistent estimation of the multidimensional probability distribution function in SED fitting + z parameter space, including all correlations. While the performance of joint SED fitting and photo-z estimation might be hindered by template incompleteness, we demonstrate that the latter is "flagged" by a large fraction of outliers in redshift, and that significant improvements can be achieved by using flexible stellar populations synthesis models and more realistic star formation histories. In all cases, we find that the median stellar age is better recovered than the time elapsed from the onset of star formation [abridged].