O6.2 The Palomar Transient Factory

The Palomar Transient Factory or

Adventures in High Fidelity Rapid Turnaround Data Processing at IPAC Jason Surace Russ Laher, Frank Masci, Wei Mi (did the IPAC work) Branamir Sesar, Eran Ofek, David Levitan (students & post-docs) Vandana Desai, Carl Grillmair, Steve Groom, Eugean Hacopians, George Helou, Ed Jackson, Lisa Storrie-Lombardi, Lin Yan (IPAC Team) Eric Bellm (Project Scientist), Shri Kulkarni (PI)

What was/is PTF/iPTF? • PTF is a robotic synoptic sky survey system designed to study transient (time-domain) phenomena. • Surveys 1000-3000 square degrees a night, predominantly at R-band to a depth of 20.5. • Primarily aimed at supernova science. • But also can study variable stars, exoplanets, asteroids, etc. • And produces an imaging sky survey like SDSS over larger area. • PTF ran 4 years on-sky starting in 2009, now “iPTF” for another 3. Early foray into the next big theme in astronomy. • Total budget ~$3M. Surace 2014

Former CFHT 12k Camera -> PTF Camera

Eliminated nitrogen dewar; camera now mechanically cryo-cooled. New field flattener, etc. 7.8 square degree active area.

Surace 2011 Surace 2014

The Venerable 48-inch Telescope

Surace 2014

PTF camera installed in late 2008; Operations started 2009

Fully robotic operation. Automatically opens, takes calibrations, science data, and adapts to weather closures. Human intervention used to guide science programs. Surace 2014

Infrared Processing and Analysis Center

IPAC is NASA’s multi-mission science center and data archive center for IR/submm astronomy. Specifically, we handle processing, archiving, and/or control for numerous missions including: IRAS, ISO, Spitzer, GALEX, Herschel, Planck, and WISE, as well as 2MASS, KI, and PTI. Also the seat of the Spitzer Science Center, NExSci, NED, NStED, and IRSA. Approximately 150 employees in two buildings on the CIT campus. Surace 2014 2009

R-band Holdings

1292 nights, 3.1 million images 47 billion source apparitions (epochal detections) Surace 2014

g-band Holdings

241 nights, 500 thousand images Surace 2014

H-alpha Holdings

99 nights, 125 thousand images Surace 2014

P48 NERSC Image Subtraction and Transient Detection/RB Pipeline Caltech/C ahill

Ingest

Photometric Pipeline

IPAC

Realtime Image Subtraction Pipeline

Epochal Images and Catalogs

Transient Candidates

SSOs

Lightcurve Pipeline

Lightcurves

Reference Images

Reference Catalogs

Reference Pipeline Surace 2014

Moving Object Pipeline

IPAC Infrastructure • Data transmission from Palomar via microwave link to SDSC. • ~1TB of data every 4-5 days. • 24 drones with 240 cores. Mixed Sun and Dell blade units running RHE. • Roughly 0.5 PB spinning disk in Nexsan storage units. • Associated network equipment. • Database and file servers. • Archive servers. • Tape backup. IPAC Morrisroe Computer Center Surace 2014

Cluster/Parallelization Architecture • PTF data are observed on a fixed system of spatial tiles on the sky. Vastly simplifies data organization and processing. PTF fields and CCD combinations are the basic unit to parallelize processing over multiple cluster nodes. Each node processes a CCD at a time. • “Virtual Pipeline Operator” on a master control node oversees job coordination and staging. • Multi-tiered local scratch disk, “sandbox” (working area) and archive disk structure; inherited architecture from previous projects driven by issues with very large file counts and I/O heavy processes. • Disk system shared with archive for budget constraint issues. Surace 2014

Software Structure • Individual modules written predominantly in C, but also FORTRAN, PYTHON, MATLAB, and IDL. • Connected with PERL wrapper infrastructure into discrete pipelines. • Postgres database used for tracking dataflow, data quality, etc. Relational database not used in the operations system for catalog storage; not needed, and flat file access is more efficient. • Heavy use of community software: sextractor, swarp, scamp, astrometry.net, daophot, hotpants. Cheaper not to re-invent the wheel. • Software replaced as needed by new code development. • Highly agile development program: unknown and changing science requirements, small team, and no separate development system due to budget constraints! • Continuous refinement process. There’s a trap with big data development on a new instrument.

Surace 2014

Realtime Pipeline • Realtime – data is processed as received, turnaround in 20 minutes. Needed for same-night followup. • Astrometric and photometrically calibrated. • Image subtraction against a reference image library constructed from all the data to-date. In-house software. • “Streak detection” for fast-moving objects; moving object pipeline constructs solar system object tracklets. • Transient candidate detection and extraction via psf-fitting and aperture extraction. • Machine-learning “scores” candidates. • Image subtractions and candidate catalogs are pushed to an external gateway where they are picked up by the solar system, ToO, and extragalactic marshalls. Surace 2014

Realtime Image Subtraction and Transient Detection

Originally the community “HOTPANTS” package, now replaced with a more sophisticated in-house image subtraction algorithm.

Surace 2014

Photometric Pipeline • This pipeline processes data in the traditional manner. • Starts up at the end of the night, after all the data has been received. • Calibration is derived from the entire night’s worth of data. Specifically, the bias and flat-fields are derived from the data themselves. • Photometric calibration is derived from extracted photometry from all sources, fitting color, extinction, time and large-scale spatial variations vs. the SDSS. Typically reach an accuracy of a few %. • Astrometric calibration is done individually at the CCD level, against a combined SDSS and UCAC4 catalog. Typically good to 0.15”. • Output from this pipeline are calibrated single-CCD FITS images and single-CCD catalog FITS binary tables (both aperture and psf-fit). These are archived through IRSA. Available 1-3 days after observation.

Photometric Pipeline Output Single R-band thumbnail image of Arp 220, 8 arcminutes across. Aperture extractions catalog (sextractor-based) overlaid. All observations and detections of everything are saved in the archive. Products are a reduced image, bit-encoded data quality mask, and catalogs. All products are FITS.

Reference Image Pipeline • Once enough individual observations accumulate, the “reference image” pipeline is triggered. • This pipeline coadds the existing data, after selecting “best frames”, e.g. best seeing, photometric conditions, astrometry, etc. • Coaddition is done based on CCD id, PTF tile, and filter. • These images are the reference of the static sky, at a level deeper than the individual observations. • “Reference Catalogs” are extracted from these images. • This concept is important, because these are both the underlying basis of the image subtractions, and also the basis of the light-curve pipeline. • Like PTF coverage, the depth of these is variable, but is current 5