CASA, the Common Astronomy Software Applications for Radio Astronomy

1. INTRODUCTION

• CASA supports calibration, imaging, and analysis of radio-telescope data and is the primary processing software for ALMA and the VLA.It is also commonly used for other aperture-synthesis and single-dish telescopes.Source: §1 · pp. 1–2
• Built on Casacore, CASA combines calibration, imaging, image analysis, visualization, and telescope-specific support in C++ applications accessible through Python.The AIPS++ code was reorganized and migrated to CASA in 2004, when its scripting language shifted from Glish to Python bindings.Source: §1 · pp. 1–2
• The paper outlines CASA’s structure, data model, API, documentation, processing stages, observatory pipelines, development process, and tentative next-generation design.It briefly discusses calibration and imaging pipeline support for ALMA, the VLA, and VLASS.Source: §1 · p. 2

2.1. CASA structure and Python

• CASA combines a C++ library with Python scripting, offering monolithic and pip-installable modular distributions and a layered tools-and-tasks interface over C++ algorithms and Casacore infrastructure.The modular distribution, available from version 6, lets users work in their own Python environment.Source: §2.1 · pp. 2–3

2.2. The MeasurementSet and Data Model

• The MeasurementSet stores interferometric or single-dish data in a relational Main table and metadata subtables, with columns for original, calibrated, modeled, flagged, and weighted data.Main-table rows index time, phase-reference direction, spectral window, and baseline; the two-dimensional data array represents polarization and spectral channels.Source: §2.2 · p. 3

2.3. Application Programming Interface (API)

• CASA’s API exposes functionality through functional casatasks and object-oriented casatools, supports installation in a user’s Python environment or a monolithic distribution, and is tested for each release.The API also includes visualization packages and allows users to configure and extend CASA.Source: §2.3 · p. 3

2.3.1. CASA configuration and shell

• Users can configure CASA before launch with config.py and customize its interactive shell through startup.py, including setting paths, importing packages, and running preparation code.Configuration options include reference data paths, pipeline installation, and GUI display settings.Source: §2.3.1 · pp. 3–4

2.3.2. CASA tasks, tools, and GUIs

• CASA tools provide low-level operations, while higher-level Python tasks package defined processing steps for scripting and pipelines; GUIs support visibility diagnostics and image viewing.The plotms GUI plots visibility and calibration-table data, and the CASA viewer visualizes image products.Source: §2.3.2 · p. 4

2.3.3. External data repository

• CASA’s external data repository includes Measures tables and other runtime data, which may need updating between software releases for operations on recently observed data to work correctly.The Measures tables include Earth Orientation Parameters and solar-system ephemerides.Source: §2.3.3 · p. 4

2.4. Parallel processing

• CASA uses MPI-based parallel processing across cores and computing nodes, with Python-level parallelization for calibration, data manipulation, and continuum imaging and C++-level parallelization for spectral-cube imaging.Imaging partitions continuum data by time and spectral cubes by frequency; starting a multi-process session requires mpicasa.Source: §2.4 · pp. 4–5

2.5. CASA Documentation

• CASA Docs provides task descriptions and parameters, installation information, examples, release details, and known issues; CASA Guides offer step-by-step processing recipes for multiple telescope teams.Task documentation can be opened from within CASA using doc(‘taskname’).Source: §2.5 · p. 5

3. DATA PROCESSING

• Figure 2 presents a typical end-to-end CASA processing workflow, encompassing import and export, examination and flagging, calibration, manipulation, imaging, visualization and analysis, and simulations.The section describes how CASA handles these stages.Source: §3 · p. 5, Figure 2 · p. 5

3.1. Data processing I: Import and export

• ALMA and VLA raw data are imported from ASDM into CASA MeasurementSet format with importasdm; CASA also accepts other telescope data and UVFITS or FITS-IDI files.ASDM contains observatory-specific metadata and uses binary objects for bulk data and large metadata subtables, with XML for smaller metadata subtables.Source: §3.1 · p. 5

3.2. Data processing II: Examination and Flagging

• CASA provides tasks and graphical tools to inspect and plot measurement data; flagging excludes compromised observations, including data affected by shadowing, noisy band edges, or radio-frequency interference.The flagdata task supports manual selection and automated flagging, including clip, tfcrop for strong narrow-band interference, and sliding-window rms outlier detection with rflag.Source: §3.2 · pp. 5–6
• Flagging marks MeasurementSet entries without deleting data, and flagmanager supports saving and restoring named flag versions.A copy of the FLAG column can be stored in a separate flagversions directory.Source: §3.2 · p. 6

3.3.1. Interferometric calibration

• Interferometric calibration derives complex corrections for instrumental and environmental effects, using a multiplicative Measurement Equation whose terms are selected according to scientific goals.Solutions typically use characterized calibrators; baseline-based corrections require care because they can absorb antenna-based information and undermine later antenna-based calibration.Source: §3.3.1 · pp. 6–8

3.3.2. Calibration methodology in CASA

• CASA calibration tasks solve for gain, bandpass, polarization, and delay corrections, while applycal applies specified solutions and writes CORRECTED_DATA for imaging.The calibration solutions are stored in separate calibration tables.Source: §3.3.2 · p. 8
• CASA calibration proceeds by bootstrapping each correction from the best available information, then revising solutions and source models through self-calibration and reimaging.The largest practical benefit comes from iterative revision of time-dependent gains, mainly due to the troposphere, and the source visibility model.Source: §3.3.2 · pp. 8–9

3.3.3. Single-dish calibration

• Single-dish calibration uses system-temperature and blank-sky corrections in the spectral domain, followed by accurate fitting and subtraction of spectral baselines.CASA provides sdcal, sdgaincal, and sdbaseline for calibration, time-dependent gain removal, and baseline subtraction, respectively.Source: §3.3.3 · p. 9

3.4. Data processing IV: Data manipulation

• mstransform combines most visibility-data transformations, separately or together in an in-memory pipeline, avoiding unnecessary input/output steps.Historically, individual tasks handled each manipulation step.Source: §3.4 · p. 9
• CASA tasks concatenate, average, split, weight, and regrid visibility data, and support smoothing, subtraction, position correction, and phase-center shifting.Examples include concat, split, statwt, cvel, hanningsmooth, uvcontsub, uvsub, fixplanets, and phaseshift.Source: §3.4 · p. 9
• Most visibility-data manipulation occurs after calibration and before imaging.Source: §3.4 · p. 9

3.5.1. Interferometric imaging

• Because finite sampling leaves the interferometric measurement operator noninvertible, CASA iteratively reconstructs sky brightness from calibrated visibilities using dirty images, deconvolution, and PSF-based restoration.The dirty image is shaped by the instrument’s PSF and noise; deconvolution estimates the sky model, which is convolved with a Gaussian representing the PSF main lobe to form the clean image.Source: §3.5.1 · p. 10

3.5.2. Weighting schemes

• Natural, uniform, and Briggs weighting trade off image noise, resolution, and sensitivity, while an optional Gaussian uv taper downweights long baselines.Natural weighting typically yields the lowest noise but lower resolution; uniform weighting raises resolution and noise, while Briggs weighting adjusts between the two via a robustness parameter.Source: §3.5.2 · pp. 10–11

3.5.3. Imaging mode and Gridding

• CASA gridding resamples weighted visibilities onto a regular spatial-frequency grid before Fourier inversion, with imaging modes for continuum, cubes, polarization, and mosaics.Advanced A- and W-Projection options correct direction-dependent wideband and widefield effects, including primary-beam variation and non-coplanar baselines.Source: §3.5.3 · pp. 11–12

3.5.4. Deconvolution and image reconstruction

• CASA’s tclean alternates major visibility-domain cycles with minor image-domain deconvolution cycles, iteratively building a sky model and subtracting its predicted visibilities.Minor-cycle algorithms include point-source, maximum-entropy, multiscale, and wideband methods; users can set stopping thresholds and masks, and the final model is restored with the PSF’s main-lobe resolution, optionally corrected for primary-beam response.Source: §3.5.4 · pp. 12–13

3.5.5. Single-dish imaging

• CASA’s tsdimaging task forms single-dish images or cubes by populating an image grid with calibrated data and configurable convolution kernels.It selects defaults such as pixel size, image dimensions, and spectral resolution, which users can adjust.Source: §3.5.5 · p. 13

3.5.6. Image combination

• CASA’s sdintimaging combines single-dish and interferometer data through feathering and joint deconvolution to recover emission across spatial scales.It supports cubes, wideband multi-term imaging, mosaics, frequency-dependent primary-beam corrections, and adjustable relative weighting of the two data sets.Source: §3.5.6 · pp. 13–14

3.6. Data processing VI: Analysis & Visualization

• CASA GUIs support inspection of raw telescope data and metadata, including browsetable for tables, plotants for antenna positions, and plotms for visibility data.The section also refers readers to Sect. 3.2 for plotms.Source: §3.6 · p. 14
• CARTA is external image-visualization and analysis software designed for ALMA, the VLA, and SKA pathfinder telescopes.It is developed by a consortium involving ASIAA, IDIA, NRAO, and the University of Alberta’s Department of Physics.Source: §3.6 · p. 14
• CARTA is being developed with next-generation radio telescopes in mind and is intended to become a standard visualization tool for CASA-processed data.Source: §3.6 · p. 14

3.7. Data processing VII: Simulations

• CASA can simulate observations and datasets for VLA, ALMA, and other existing or future observatories, supporting telescope-capability assessment and benchmarks of CASA performance, optimization, and reproducibility.It can create simulated MeasurementSets for any interferometric array, with configuration files available for many arrays.Source: §3.7 · p. 14
• simobserve creates simulated MeasurementSets for interferometric or total-power observations with a specified telescope.Source: §3.7 · p. 14
• simanalyze images and analyzes simulated MeasurementSet data, producing diagnostic images and plots.Source: §3.7 · p. 14
• simalma streamlines simobserve and simanalyze into one-step ALMA simulations, combining 12-m array configurations, the 7-m ACA, and total-power measurements.Source: §3.7 · p. 14
• CASA simulator tools provide additional flexibility, especially for non-ALMA use cases, including calculating and applying calibration tables for atmospheric and instrumental corruption.Source: §3.7 · pp. 14–15

3.8. Very Long Baseline Interferometry

• CASA processes VLBI data, including from the VLBA and EVN (Zensus & Ros 2015), using dedicated and standard interferometric tasks.The software can also process data from other VLBI telescopes.Source: §3.8 · p. 15
• fringefit solves for phase, delay, delay-rate, and optionally dispersive delay by time and spectral window, enabling correction of visibility-phase errors.The errors can arise from the atmosphere, instrument signal paths, or other pre-calibration factors.Source: §3.8 · p. 15
• accor derives amplitude corrections from mean autocorrelation spectra to correct sampler-threshold errors, including those caused by the VLBA’s DiFX correlator.Source: §3.8 · p. 15
• Janssen et al. (2019) built rPICARD, the first generic CASA-based VLBI calibration and imaging pipeline, used to process Event Horizon Telescope data that yielded the first black-hole image (Event Horizon Telescope Collaboration et al. 2019a,b; Janssen et al. 2019).Source: §3.8 · p. 15
• JIVE leads ongoing CASA VLBI development with NRAO; van Bemmel et al. (2022) describe its VLBI tools, tasks, and operation in a companion paper.Source: §3.8 · p. 15

3.9. Algorithm Research & Development

• CASA’s R&D continually improves algorithms as needs evolve, in collaboration with the NRAO Algorithm R&D Group, telescope and pipeline scientific staff, and the CASA development team.Source: §3.9 · p. 15
• Planned algorithms include wideband Adaptive Scale Pixel (ASP) deconvolution and Full Mueller imaging for wide bandwidths (Jagannathan 2018).Source: §3.9 · p. 15
• Other planned algorithms include automated flagging based on binning (u, v)-data and a GPU implementation of widefield A-Projection.Source: §3.9 · p. 15
• The team is developing GUIs for examining, processing, and simulating data, with better Python integration and usability in Notebook environments.Source: §3.9 · p. 15

4. ALMA AND VLA PIPELINES

• CASA supports ALMA calibration and imaging pipelines and maintains versions for VLA, VLASS, and Science Ready Data Products processing.These pipelines support automated and customized ALMA data processing.Source: §4 · p. 15
• The pipelines combine standard CASA tasks and tool methods with custom pipeline tasks and analysis utilities, and are bundled only with select CASA versions.Source: §4 · p. 15
• CASA also supports Nobeyama 45-m single-dish processing in On-The-Fly mode and allows custom tasks and external pipelines, including VLBI pipelines.Source: §4 · p. 15

• Stakeholder representatives inform CASA’s development plans, whose priorities balance user needs with software maintenance and evolution on a half-yearly timescale.VLBI and Single-Dish input is coordinated through partner developer teams; CASA releases software every few months.Source: §5 · p. 15
• Each feature proceeds from requirements and specifications through development to internal verification and external validation, with functional and unit tests targeting code and interface coverage.Source: §5 · p. 15
• Stakeholder-verification tests target dominant usage modes such as operational pipelines, while performance and benchmark test suites are in development.External science staff typically validate applicability to stakeholder use cases; selected numerical or algorithmic features receive additional characterization to assess effectiveness for intended use cases.Source: §5 · pp. 15–16

6. NEXT-GENERATION CASA

• The CNGI prototype uses Zarr storage, the Xarray API, and the Dask parallel processing framework.CNGI is a prototype implementation of the next-generation CASA infrastructure.Source: §6 · p. 16

7. CONCLUSIONS

• CASA supports calibration, imaging, and analysis for ALMA, the VLA, and other radio telescopes, including aperture-synthesis arrays and single-dish instruments.Comprehensive documentation and stakeholder-based development help CASA serve as a leading radio-data processing software package.Source: §7 · p. 16
• CASA’s core pipeline support and modernized software infrastructure target the increasing data rates and scalability challenges of current and next-generation radio telescopes.The paper concludes that these efforts will help CASA handle increasingly complex data.Source: §7 · p. 16