The Hyper Suprime-Cam Software Pipeline

TL;DR

The paper presents the Hyper Suprime-Cam Software Pipeline, a deep optical-imaging processing system built as a customized extension of the LSST Data Management stack to reduce HSC-SSP data and general HSC observations. It details a four-stage processing flow—CCD Processing, Joint Calibration, Image Coaddition, and Coadd Processing—plus a comprehensive suite of algorithms for instrumental signature removal, PSF modeling, detection, deblending, and multifaceted photometry. Key contributions include a robust joint astrometric/photometric calibration framework, a PSF-coaddition approach that preserves PSF information in coadds, a

Abstract

In this paper, we describe the optical imaging data processing pipeline developed for the Subaru Telescope's Hyper Suprime-Cam (HSC) instrument. The HSC Pipeline builds on the prototype pipeline being developed by the Large Synoptic Survey Telescope's Data Management system, adding customizations for HSC, large-scale processing capabilities, and novel algorithms that have since been reincorporated into the LSST codebase. While designed primarily to reduce HSC Subaru Strategic Program (SSP) data, it is also the recommended pipeline for reducing general-observer HSC data. The HSC pipeline includes high level processing steps that generate coadded images and science-ready catalogs as well as low-level detrending and image characterizations.

Figures (21)

• Figure 1: Conceptual flow of processing in the HSC Pipeline. Filled rectangles are the four high-level stages of the pipeline discussed in Section \ref{['sec:pipelines']}. Unfilled rectangles show the most important data products and their granularity. Dataset names are those used by the data butler (see Section \ref{['sec:io-and-provenance']} and Table \ref{['tbl:data-products']}).
• Figure 2: Differences in CCD position ($x$: upper, $y$: middle) and rotations (bottom) in the focal plane from joint astrometric fitting in different tracts in $i$.
• Figure 3: Examples of coaddition algorithms in the presence of image artifacts: a satellite trail (top) and a pair of optical ghosts (bottom). The images in the left column were computed with a direct mean, in which the PSF is preserved but single-image artifacts clearly print through to the coadd. The middle column images were created with an iterative 3$\sigma$ clip, which invalidates the PSF and still fails to remove the wings of the artifact. The right column images were produced with our "safe clipping" algorithm, which does a better job of removing the artifacts and preserves the PSF in the rest of the image. Note that none of the methods fully remove the rightmost ghost in the lower panel.
• Figure 4: Diagram of the coadd processing flow described in Section \ref{['sec:coadd-processing']}. As we detail in the text, we start by detecting sources in each band independently, then merge these detections to build a consistent cross-band object list. We then detect and measure in each band separately. Finally, we consider all bands together to select a reference band for each object and then perform forced photometry in each band. This algorithm is naturally extended to include narrow-band filters (in the order described in the text) in the Deep and UltraDeep layers.
• Figure 5: Residuals of the PSF models for size and ellipticity for the S16a $i$-band data as a function of PSF magnitude, both before (blue) and after (purple) brighter-fatter correction. The lower right shows the stacked residuals of the corrected image and the PSF model as a function of PSF magnitude, where the brightness decreases from top left to bottom right.