CASA: Common Astronomy Software Applications

• CASA is a comprehensive software suite designed for radio astronomical data calibration, imaging, and analysis across various telescope systems.
• Its architecture leverages a modular Measurement Set data model with C++ casacore libraries and Python bindings, streamlining reproducible, high-throughput pipelines.
• CASA employs advanced algorithms, including RIME-based calibration and iterative deconvolution, to achieve scalable processing and precise spectral line modeling.

The Common Astronomy Software Applications (CASA) package is the de facto standard software environment for calibration, imaging, and analysis of radio astronomical data from single-dish, aperture-synthesis, and Very Long Baseline Interferometry (VLBI) telescopes. Originally developed by an international consortium led by NRAO, ESO, NAOJ, and later JIVE, CASA underpins the production and reduction pipelines for ALMA and the Karl G. Jansky Very Large Array, and has become a foundational component for high-throughput VLBI analysis, automated data pipelines, and spectral line modeling frameworks. CASA's architecture centers on the Measurement Set (MS) relational data model, an extensible Python interface for scripting and pipelines, and HPC-ready APIs for scalable processing (TEAM et al., 2022, Emonts et al., 2019, Janssen et al., 2019).

1. Architecture, Data Model, and Core Components

CASA is structured around a modular architecture built atop the C++ casacore libraries, which provide I/O primitives, numerical methods, and the Measurement Set data abstraction. The MS format consists of a MAIN table (rows ≡ time × baseline × spectral window) along with metadata subtables for ANTENNA, SPECTRAL_WINDOW, FIELD, POLARIZATION, and all derived calibration solutions. Unlike legacy packages, CASA requires no monolithic file conversion through multiple formats, and calibration and flagging tables augment the MS without modifying its data columns (TEAM et al., 2022, Petry et al., 2014).

The major functional layers are:

• casacore: Implements MS I/O, coordinates, transformations, and linear algebra.
• casatools: Pythonic bindings for data access, calibration, imaging, visualization.
• casatasks: High-level task scripts for standard workflows (e.g., gaincal, fringefit, tclean).
• casampi: Message Passing Interface (MPI) plugin for parallelizing imaging and calibration (Raba et al., 2019, Janssen et al., 2019).
• GUI tools: Standalone viewers (PlotMS, CARTA) for visualization.

CASA tasks are invoked either as Python functions or from within Jupyter, enabling native integration with standard scientific computing libraries (NumPy, AstroPy) and supporting both batch execution and interactive workflows (Raba et al., 2019, Emonts et al., 2019). With the release of CASA 6, installation is modular via PyPI wheels, decoupling the core processing from GUI and allowing direct embedding in Jupyter notebooks or cluster environments (Raba et al., 2019).

2. Mathematical Framework and Calibration Algorithms

At the algorithmic core, CASA’s calibration tasks are grounded in the Radio Interferometry Measurement Equation (RIME):

$$V_{ij}^{\rm obs}(\nu, t) = J_i(\nu, t) J_j^\dagger(\nu, t) V_{ij}^{\rm true}(\nu, t) + N_{ij}(\nu, t)$$

where $J_i$ includes antenna-based Jones matrices for gain, bandpass, delay, D-term, atmospheric, and geometric/faraday rotation terms (for VLBI), and $N_{ij}$ is thermal noise (TEAM et al., 2022, Emonts et al., 2019). Calibration proceeds via tasks including:

• gaincal: Solves for time-dependent per-antenna gains via least-squares minimization.
• bandpass: Fits frequency-dependent gains across spectral windows.
• polcal: Determines polarization leakage and cross-phase (D-term solutions).
• fringefit: Implements the Schwab–Cotton global fringe-fitting algorithm solving for per-antenna phase offsets, group delays, and fringe rates for VLBI (Bemmel et al., 2019, Bemmel et al., 2022).
• applycal: Applies an ordered list of calibration tables to the MS.

Amplitude calibration utilizes standard metadata (ANTAB) such as $T_\mathrm{sys}$ and DPFU, with tables generated by custom Python parsers and gencal calls. The corrected visibility amplitude is given by $$V_\mathrm{corr}(i, j) = V_\mathrm{raw}(i, j) \sqrt{G_i G_j / (T_\mathrm{sys,i} T_\mathrm{sys,j})}$$ (Janssen et al., 2019).

For single-dish reduction, similar formalism is applied—with calibration steps (sdcal, sdbaseline, tsdimaging) operating on the FLOAT_DATA column, and atmospheric tools (e.g., ATM-based ozone correction) directly accessible through casatools (Sawada et al., 2021).

3. Imaging and Deconvolution Processes

CASA’s flagship imaging engine, tclean, executes a sequence of weighting, gridding, FFT inversion, and deconvolution steps:

• Gridding: Resamples irregular (u, v) visibilities to a regular lattice, applying selected weighting (natural, uniform, Briggs robust).
• Fourier inversion: Computes the dirty image as the FFT of the weighted visibilities.
• Deconvolution: Minor cycles identify the peak residuals and subtract scaled PSFs; major cycles reconstruct the model and update visibilities.
• Deconvolver selection: Supports Hogbom, Clark, Multiscale, Multi-term Multi-Frequency Synthesis (MTMFS) for joint intensity and spectral index mapping (Petry et al., 2014, Emonts et al., 2019).
• Primary beam correction and mosaicking: Incorporated for widefield, mosaic, and subarray imaging modes.

Parallelization is enabled for tclean via mpicasa (or casampi in modular usage), yielding speed-up factors of 4–7× versus single-core execution on moderate-sized mosaics (Janssen et al., 2019, Emonts et al., 2019).

4. Automation, Reproducibility, and Workflow Scripting

CASA is designed for maximal automation. Complex calibration and imaging workflows, such as those required in ALMA QA2 or for VLBI pipelines, are captured in reproducible, parameterized Python scripts or externally in CSV/JSON configuration files. Notably:

• rPICARD wraps the end-to-end VLBI calibration chain (import, flag, accor, gencal, fringefit, bandpass, polcal, applycal, tclean), managing parallel execution using MPI and adjusting solution intervals dynamically on a per-scan basis (Janssen et al., 2019, Janssen et al., 2019).
• VASCO extends rPICARD with fully blind search for calibrators and reference antennas using FFT-based fringe detection, metadata heuristics, and orchestration via ALFRD for large-scale VLBA and archival datasets (Kumar et al., 19 Apr 2026).
• AMIsurvey and chimenea enable fully automated imaging, leveraging drive-casa to control the CASA task stack by emulating an interactive shell and incorporating iterative RMS estimation and source finding (Staley et al., 2015).

Python bindings enforce provenance and reproducibility; tasks are referentially transparent under some wrappers, supporting memoization and partial recomputation for exploratory analysis—substantially reducing turnaround during iterative data reduction (Nikolic et al., 2017).

5. VLBI and Advanced Heterogeneous-Array Data Reduction

CASA incorporates dedicated VLBI algorithms through new tasks:

• fringefit and accor: Mirror and extend AIPS FRING and ACCOR, supporting global least-squares across antennas, multi-band, non-dispersive and dispersive delay, and per-baseline FFT detection (Bemmel et al., 2022, Bemmel et al., 2019).
• importfitsidi, gencal: Support advanced FITS-IDI metadata, Tsys/gain curve ingestion, and DiFX quantization corrections.
• Polarization and phase referencing: Native support for D-term correction and phase transfer, with solutions chained via ordered gaintables in applycal.
• Automated pipeline execution: VASCO and rPICARD achieve >97% success rates in blind calibration of heterogeneous VLBA archive (1372 observations, 1000 sources) with per-source execution times ~30 min on 20-core clusters (Kumar et al., 19 Apr 2026, Janssen et al., 2019).

Parallelization via MPI enables scalable reduction of wide-band, multi-epoch, and multi-source VLBI datasets—approaching linear efficiency up to 16–20 cores in fringe-fit and calibration stages.

6. Extensibility: Spectral Line Analysis and Simulation

CASA’s ecosystem incorporates line modeling and simulation via:

• XCLASS: Embedded in CASA for 1D LTE radiative transfer modeling of multi-component, dust-attenuated spectra. XCLASS and MAGIX handle spectral fitting and parameter estimation using a bundled SQLite3 molecular database (fed by CDMS/JPL/VAMDC), with χ² minimization supporting parallel optimizers (e.g., Levenberg–Marquardt, genetic algorithms) (Möller et al., 2015).
• Simulations: simobserve, simanalyze, and simalma support full-array and site-realistic visibility and image simulation.
• Atmospheric correction: Single-dish atmospheric line removal (notably mesospheric O₃) by ATM-coupled tasks, achieving suppression factors of 10–20× using calibrated, optimized atmospheric profiles (Sawada et al., 2021).

7. Supported Observatories, Pipelines, and Future Directions

CASA supports a diverse array of telescopes and standard pipelines:

• ALMA, VLA, VLASS, Nobeyama 45m: End-to-end pipeline scripts tuned per facility and observing mode, with QA logic and auto-masking for high-throughput operations (TEAM et al., 2022, Petry et al., 2014).
• VLBI arrays: EHT, GMVA, VLBA, EVN—the rPICARD, VASCO, and custom pipelines provide robust, reproducible calibration, including for geodetic and spectral line modes (Janssen et al., 2019, Kumar et al., 19 Apr 2026).
• HPC-readiness and modularization: CASA 6 supports modular install and embedding in Jupyter and cloud workflows, with ongoing migration towards CNGI (CASA Next-Generation Infrastructure) based on Dask, Xarray/Zarr for scalable, distributed processing (Raba et al., 2019, TEAM et al., 2022).

CASA remains under active development, with quarterly releases, expanded Python 3 support, continuous integration testing, and community-driven feature enhancements to accommodate the demands of next-generation radio arrays and automated analysis at petascale volumes.

---

References:

• (Janssen et al., 2019) rPICARD: A CASA-based Calibration Pipeline for VLBI Data
• (Emonts et al., 2019) The CASA software for radio astronomy: status update from ADASS 2019
• (Bemmel et al., 2019) CASA on the fringe: VLBI data processing in the CASA software package
• (Bemmel et al., 2022) CASA on the fringe -- Development of VLBI processing capabilities for CASA
• (TEAM et al., 2022) CASA, the Common Astronomy Software Applications for Radio Astronomy
• (Raba et al., 2019) CASA 6: Modular Integration in Python
• (Möller et al., 2015) eXtended CASA Line Analysis Software Suite (XCLASS)
• (Petry et al., 2014) ALMA service data analysis and level 2 quality assurance with CASA
• (Janssen et al., 2019) rPICARD: A CASA-based Calibration Pipeline for VLBI Data. Calibration and imaging of 7 mm VLBA observations of the AGN jet in M87
• (Sawada et al., 2021) Offline Correction of Atmospheric Effects on Single-Dish Radio Spectroscopy
• (Staley et al., 2015) Chimenea and other tools: Automated imaging of multi-epoch radio-synthesis data with CASA
• (Nikolic et al., 2017) Minimal Re-computation for Exploratory Data Analysis in Astronomy
• (Kumar et al., 19 Apr 2026) VASCO: A fully automated CASA pipeline for large volume VLBI data calibration
• (Petry, 2012) Analysing ALMA data with CASA