Testing KiDS cross-correlation redshifts with simulations

TL;DR

This study evaluates the accuracy of clustering-redshift techniques using simulated KiDS data, demonstrating their potential for precise redshift calibration in cosmic shear surveys, with some biases influenced by model assumptions.

Contribution

It provides a detailed assessment of clustering redshift performance with mock KiDS data, highlighting the importance of sophisticated models for future high-precision surveys.

Findings

Clustering redshifts are unbiased within |Δz|<0.006.

Redshift evolution of galaxy bias can be mitigated with auto-correlation.

Biases increase to |Δz|~0.04 when using simple colour-based models.

Abstract

Measuring cosmic shear in wide-field imaging surveys requires accurate knowledge of the redshift distribution of all sources. The clustering-redshift technique exploits the angular cross-correlation of a target galaxy sample with unknown redshifts and a reference sample with known redshifts, and is an attractive alternative to colour-based methods of redshift calibration. We test the performance of such clustering redshift measurements using mock catalogues that resemble the Kilo-Degree Survey (KiDS). These mocks are created from the MICE simulation and closely mimic the properties of the KiDS source sample and the overlapping spectroscopic reference samples. We quantify the performance of the clustering redshifts by comparing the cross-correlation results with the true redshift distributions in each of the five KiDS photometric redshift bins. Such a comparison to an informative model is necessary due to the incompleteness of the reference samples at high redshifts. Clustering mean redshifts are unbiased at $|\Delta z|<0.006$ under these conditions. The redshift evolution of the galaxy bias can be reliably mitigated at this level of precision using auto-correlation measurements and self-consistency relations, and will not become a dominant source of systematic error until the arrival of Stage-IV cosmic shear surveys. Using redshift distributions from a direct colour-based estimate instead of the true redshift distributions as a model for comparison with the clustering redshifts increases the biases in the mean to up to $|\Delta z|\sim0.04$. This indicates that the interpretation of clustering redshifts in real-world applications will require more sophisticated (parameterised) models of the redshift distribution in the future. If such better models are available, the clustering-redshift technique promises to be a highly complementary alternative to other methods of redshift calibration.