The ALMA Interferometric Pipeline Heuristics

TL;DR

This paper details the development and refinement of the ALMA interferometric data processing pipeline heuristics, emphasizing automation, quality assurance, and reproducibility across multiple observation cycles.

Contribution

It introduces a comprehensive, Python-based framework with refined heuristics for calibration and imaging, supporting diverse data types and improving automation in ALMA data processing.

Findings

97% of datasets calibrated and imaged automatically

Heuristics refined through annual cycles since 2014

Pipeline supports ALMA, VLA, and single dish data

Abstract

We describe the calibration and imaging heuristics developed and deployed in the ALMA interferometric data processing pipeline, as of ALMA Cycle 9. The pipeline software framework is written in Python, with each data reduction stage layered on top of tasks and toolkit functions provided by the Common Astronomy Software Applications package. This framework supports a variety of tasks for observatory operations, including science data quality assurance, observing mode commissioning, and user reprocessing. It supports ALMA and VLA interferometric data along with ALMA and NRO45m single dish data, via different stages and heuristics. In addition to producing calibration tables, calibrated measurement sets, and cleaned images, the pipeline creates a WebLog which serves as the primary interface for verifying the data quality assurance by the observatory and for examining the contents of the data by the user. Following the adoption of the pipeline by ALMA Operations in 2014, the heuristics have been refined through annual development cycles, culminating in a new pipeline release aligned with the start of each ALMA Cycle of observations. Initial development focused on basic calibration and flagging heuristics (Cycles 2-3), followed by imaging heuristics (Cycles 4-5), refinement of the flagging and imaging heuristics with parallel processing (Cycles 6-7), addition of the moment difference analysis to improve continuum channel identification (2020 release), addition of a spectral renormalization stage (Cycle 8), and improvement in low SNR calibration heuristics (Cycle 9). In the two most recent Cycles, 97% of ALMA datasets were calibrated and imaged with the pipeline, ensuring long-term automated reproducibility. We conclude with a brief description of plans for future additions, including self-calibration, multi-configuration imaging, and calibration and imaging of full polarization data.