Photometric Covariance in Multi-Band Surveys: Understanding the Photometric Error in the SDSS

TL;DR

This paper analyzes the photometric uncertainties in SDSS, revealing under-estimated errors and strong inter-band correlations, and provides methods to correct these issues for improved multi-band survey accuracy.

Contribution

It offers a detailed correction method for SDSS photometric errors and covariance, emphasizing the importance of empirical calibration for future surveys.

Findings

SDSS pipeline under-estimates photometric errors by 20-100%.

Photometric errors are highly correlated across bands, especially for variable objects.

Correcting for covariance reduces color error over-estimation by 2-3 times.

Abstract

In this paper we describe a detailed analysis of the photometric uncertainties present within the Sloan Digital Sky Survey (SDSS) imaging survey based on repeat observations of approximately 200 square degrees of the sky. We show that, for the standard SDSS aperture systems (petrocounts, counts_model, psfcounts and cmodel_counts), the errors generated by the SDSS photometric pipeline under-estimate the observed scatter in the individual bands. The degree of disagreement is a strong function of aperture and magnitude (ranging from 20% to more than a factor of 2). We also find that the photometry in the five optical bands can be highly correlated for both point sources and galaxies, although the correlation for point sources is almost entirely due to variable objects. Without correcting for this covariance the SDSS color errors could be in over-estimated by a factor of two to three. Combining these opposing effects, the SDSS errors on the colors differ from the observed color variation by approximately 10-20% for most apertures and magnitudes. We provide a prescription to correct the errors derived from the SDSS photometric pipeline as a function of magnitude and a semi-analytic method for generating the appropriate covariance between the different photometric passbands. Given the intrinsic nature of these correlations, we expect that all current and future multi-band surveys will also observe strongly covariant magnitudes. The ability of these surveys to complete their science goals is largely dependent on color-based target selection and photometric redshifts; these results show the importance of spending a significant fraction of early survey operations on re-imaging to empirically determine the photometric covariance of any observing/reduction pipeline.