Self calibration of photometric redshift scatter in weak lensing surveys

TL;DR

This paper introduces a self-calibration method for photometric redshift errors in weak lensing surveys using shear-density correlations, achieving high accuracy without relying on priors, and discusses systematic biases affecting calibration.

Contribution

It proposes a novel self-calibration technique that uses survey data itself to accurately determine photo-z outlier rates without prior assumptions.

Findings

Photo-z outlier rates can be calibrated to 0.01-1% accuracy for z<2.

The method preserves all cosmological information from shear measurements.

Galaxy distribution bias is a significant systematic affecting calibration.

Abstract

Photo-z errors, especially catastrophic errors, are a major uncertainty for precision weak lensing cosmology. We find that the shear-(galaxy number) density and density-density cross correlation measurements between photo-z bins, available from the same lensing surveys, contain valuable information for self-calibration of the scattering probabilities between the true-z and photo-z bins. The self-calibration technique we propose does not rely on cosmological priors nor parameterization of the photo-z probability distribution function, and preserves all of the cosmological information available from shear-shear measurement. We estimate the calibration accuracy through the Fisher matrix formalism. We find that, for advanced lensing surveys such as the planned stage IV surveys, the rate of photo-z outliers can be determined with statistical uncertainties of 0.01-1% for $z<2$ galaxies. Among the several sources of calibration error that we identify and investigate, the {\it galaxy distribution bias} is likely the most dominant systematic error, whereby photo-z outliers have different redshift distributions and/or bias than non-outliers from the same bin. This bias affects all photo-z calibration techniques based on correlation measurements. Galaxy bias variations of $O(0.1)$ produce biases in photo-z outlier rates similar to the statistical errors of our method, so this galaxy distribution bias may bias the reconstructed scatters at several-$\sigma$ level, but is unlikely to completely invalidate the self-calibration technique.