A Clipping Method to Mitigate the Impact of Catastrophic Photometric Redshift Errors on Weak Lensing Tomography

TL;DR

This paper introduces a clipping method to identify and discard galaxies with unreliable photometric redshifts, improving weak lensing tomography accuracy by reducing biases and overlaps between redshift bins.

Contribution

A novel clipping technique based on posterior likelihood widths effectively isolates galaxies with accurate photo-z, enhancing tomographic bin purity and reducing systematic biases in dark energy measurements.

Findings

Discarding over 70% of galaxies reduces photo-z bias to statistical error levels.

Adding NIR data and restricting magnitude/redshift ranges improves bin purity.

Higher S/N ratios are observed in cross-correlations involving photo-z outliers.

Abstract

We use the mock catalog of galaxies, constructed based on the COSMOS galaxy catalog including information on photometric redshifts (photo-z) and SED types of galaxies, in order to study how to define a galaxy subsample suitable for weak lensing tomography feasible with optical (and NIR) multi-band data. Since most of useful cosmological information arises from the sample variance limited regime for upcoming lensing surveys, a suitable subsample can be obtained by discarding a large fraction of galaxies that have less reliable photo-z estimations. We develop a method to efficiently identify photo-z outliers by monitoring the width of posterior likelihood function of redshift estimation for each galaxies. This clipping method may allow to obtain clean tomographic redshift bins (here three bins considered) that have almost no overlaps between different bins, by discarding more than 70% galaxies of ill-defined photo-z's corresponding to the number densities of remaining galaxies less than 20 per square arcminutes for a Subaru-type deep survey. Restricting the ranges of magnitudes and redshifts and/or adding near infrared data help obtain a cleaner redshift binning. By using the Fisher information matrix formalism, we propagate photo-z errors into biases in the dark energy equation of state parameter w. We found that, by discarding most of ill-defined photo-z galaxies, the bias in w can be reduced to the level comparable to the marginalized statistical error, however, the residual, small systematic bias remains due to asymmetric scatters around the relation between photometric and true redshifts. We also use the mock catalog to estimate the cumulative signal-to-noise (S/N) ratios for measuring the angular cross-correlations of galaxies between finner photo-z bins, finding the higher S/N values for photo-z bins including photo-z outliers.