Peering through the OH-forest: a new technique to remove residual sky features from SDSS spectra

Abstract

The Sloan Digital Sky Survey (SDSS) currently provides by far the largest homogeneous sample of intermediate signal-to-noise ratio (S/N) optical spectra of galaxies and quasars. The fully automated SDSS spectroscopic reduction pipeline has provided spectra of unprecedented quality that cover the wavelength range 3800-9200A. However, in common with spectra from virtually all multi-object surveys employing fibres, there remain significant systematic residuals in many of the spectra due to the incomplete subtraction of the strong OH sky emission lines longward of 6700A. The S/N over substantial wavelength regions in many spectra is reduced by more than a factor of 2 over that expected from counting statistics. We present a method to automatically remove the sky residual signal, using a principal component analysis (PCA) which takes advantage of the correlation in the form of the sky subtraction residuals present in each spectrum. Application of the method results in spectra with essentially no evidence for degradation due to the incomplete subtraction of OH emission features. A dramatic improvement in the quality of a substantial number of spectra, particularly those of faint objects such as the bulk of the high-redshift quasars, is achieved. We make available IDL code and documentation to implement the sky residual subtraction scheme on SDSS spectra included in the public data releases. We illustrate the power of the sky-residual subtraction scheme using samples of SDSS galaxy and quasar spectra, presenting tests involving the near-infrared CaII triplet absorption, metal absorption line features in damped Lyman-alpha systems and composite spectra of high-redshift quasars.