Source-linked AI summary
SoFiA: a flexible source finder for 3D spectral line data
Paolo Serra, Tobias Westmeier, Nadine Giese, Russell Jurek, Lars Flöer, Attila Popping, Benjamin Winkel, Thijs van der Hulst, Martin Meyer, Bärbel S. Koribalski, Lister Staveley-Smith, Hélène Courtois
TL;DR
3D spectral-line cubes contain sources with diverse structures, creating a need for flexible detection and parameterization tools. SoFiA combines selectable multi-scale source-finding and processing algorithms in one application, with outputs supporting downstream analysis. Its performance comparisons are still being extended, and processing very large cubes remains constrained by system memory.
Problem
Diverse 3D source structures make complete and reliable detection and parameterization difficult, while growing data cubes require general source-finding tools.
Method
SoFiA combines modular filtering, multi-scale detection, source identification, reliability processing, parameter measurement, and output generation in one selectable workflow.
Results
SoFiA detects multi-scale emission, including faint extended structure and low-level emission in a resolved edge-on galaxy, while generating source parameters and output products.
Takeaways & Limitations
SoFiA provides a publicly available, flexible package whose algorithms and outputs can be selected for different data cubes and scientific questions.
Takeaways & Limitations
Performance comparisons are not yet definitive because several algorithms have been improved since the prior study and further testing is planned.
Abstract
from arXiv · showhide
We introduce SoFiA, a flexible software application for the detection and parameterization of sources in 3D spectral-line datasets. SoFiA combines for the first time in a single piece of software a set of new source-finding and parameterization algorithms developed on the way to future HI surveys with ASKAP (WALLABY, DINGO) and APERTIF. It is designed to enable the general use of these new algorithms by the community on a broad range of datasets. The key advantages of SoFiA are the ability to: search for line emission on multiple scales to detect 3D sources in a complete and reliable way, taking into account noise level variations and the presence of artefacts in a data cube; estimate the reliability of individual detections; look for signal in arbitrarily large data cubes using a catalogue of 3D coordinates as a prior; provide a wide range of source parameters and output products which facilitate further analysis by the user. We highlight the modularity of SoFiA, which makes it a flexible package allowing users to select and apply only the algorithms useful for their data and science questions. This modularity makes it also possible to easily expand SoFiA in order to include additional methods as they become available. The full SoFiA distribution, including a dedicated graphical user interface, is publicly available for download.
1 INTRODUCTION
3D spectral-line data cubes contain sources with diverse structures, making complete and reliable detection and parameterization difficult. SoFiA addresses the need for a general, flexible source-finding tool as data cubes grow in number and size.
- 3D data cubes map spectral-line flux across sky position and line-of-sight velocity, but source-finding methods vary substantially between projects.
- Sources can combine bright emission, complex extensions, multiple velocity peaks, and faint emission that is difficult to detect.
- An ideal 3D source finder should detect and parameterize diverse sources completely and reliably.
- Growing numbers and sizes of data cubes create a need for standard, sufficiently general source-finding tools.
- SoFiA combines detection algorithms and techniques from several source finders in a flexible Source Finding Application for data cubes.
- SoFiA is designed for data cubes independent of telescope or observed spectral line and was developed alongside preparations for WALLABY, DINGO, and APERTIF surveys.
2 DESCRIPTION OF SoFiA
SoFiA is a modular application that lets users assemble source-finding and parameterization workflows for data cubes. It supports command-line and graphical use, configurable complexity, and varied output products.
- SoFiA provides selectable modules for modifying cubes, detecting signal, identifying sources, rejecting false detections, optimizing masks, measuring parameters, and producing outputs.
- SoFiA produces catalogues, cubes, moment maps, position-velocity diagrams, and integrated spectra.
- The software is written in Python and C++ and can run on Unix or Linux systems when its required libraries are available.
- Users can choose module combinations and algorithms through a Qt GUI or plain-text parameter file.
- The workflow can be configured from simple moment-image creation to filtering, false-detection rejection, and spectral model fitting.
- SoFiA's modular principle is intended to remain stable while new algorithms may be introduced in future releases.
2.1 Data cube, weights cube, mask cubes and filters
SoFiA uses data, weights, binary-mask, and object-mask cubes at different processing stages, with filtering and noise normalization supporting source detection. Users can flag problematic regions or provide masks for subsequent analysis.
- SoFiA uses data, weights, binary-mask, and object-mask cubes at different processing stages.
- A weights cube accounts for noise variations and imaging artefacts by weighting voxel values across the input cube.
- Because source-finding algorithms assume uniform noise, SoFiA can apply inverse-noise weights before detection and undo them before parameter measurement.
- Users can derive local noise levels along any or all cube axes, or supply a weights cube or analytic weight description.
- Filtering options include user-selected 3D-kernel convolution and 2D-1D wavelet de-noising.
- Users can flag regions affected by bright continuum sources, Galactic emission, or radio-frequency interference before source finding.
- Existing or newly generated masks can support additional source-finding runs or parameterization-only workflows.
2.2 Detection of spectral line signal
SoFiA offers multiple selectable detection algorithms, generally targeting emission across multiple spatial and velocity scales. These methods produce binary masks that can be combined with filtering strategies, while comparative performance evidence remains subject to ongoing testing.
- SoFiA offers selectable detection algorithms, with multiple-scale searches central to detecting diverse 3D sources.
- The simple threshold method detects voxels above a user-specified flux or noise-relative threshold and does not operate on multiple scales by itself.
- S+C smooths the cube with user-specified 3D kernels at multiple angular and velocity resolutions, then unions thresholded masks.
- CNHI extracts individual or bundled spectra and uses the Kuiper test to identify regions inconsistent with noise.
- Users can combine source finders with filtering methods to design detection strategies suited to their data and target sources.
- Comparative work reported higher completeness and reliability for wavelet-denoised thresholding on some unresolved narrow-line sources, while S+C was favored for extended objects.
- The algorithms have since been improved, and further testing within SoFiA is planned to assess their performance.
- All implemented finders return binary masks, which can include noise peaks when thresholds are low.
2.3 Merging detected voxels into sources
SoFiA converts a binary mask of detected voxels into individual sources using a configurable 3D merging algorithm, then derives basic source parameters. This approach can recover extended and kinematically complex objects across multiple scales.
- 2.3 Merging detected voxels into sources: SoFiA identifies individual sources from the binary mask using a C++ one-pass algorithm with a user-specified elliptic-cylinder linking element.The implementation is equivalent to friends-of-friends grouping with an elliptic cylinder.
- 2.3 Merging detected voxels into sources: The merging step returns basic parameters including total flux, peak flux, and source size, normalized by the noise level.
2.4 Reliability and rejection of false detections
SoFiA addresses false detections by combining threshold choices with size filtering and reliability estimation. Users can trade catalogue completeness against reliability and obtain reliability values for positive detections.
- 2.4 Reliability and rejection of false detections: Lower detection thresholds include more noise peaks, which can be mistaken for separate objects when sufficiently distant from real sources.
- 2.4 Reliability and rejection of false detections: A size filter removes detections below user-defined minimum dimensions, but may also remove bright unresolved sources.
- 2.4 Reliability and rejection of false detections: SoFiA estimates object reliability by comparing positive and negative source distributions in parameter space, assuming noise peaks have matching distributions.
- 2.4 Reliability and rejection of false detections: Reliability calculation requires both positive and negative noise peaks and a sufficient number of negative peaks for meaningful distribution estimates.
- 2.4 Reliability and rejection of false detections: A high threshold yields a reliable but potentially incomplete catalogue, whereas a lower threshold followed by false-detection rejection can provide reliability values for positive detections.
2.5 Mask optimization
SoFiA optimizes object masks because thresholding can omit faint outer emission and bias measured parameters. It provides geometry-based and dilation-based mask-growth methods for different source morphologies.
- 2.5 Mask optimization: Mask optimization compensates for faint outer emission missed by detection masks, which can otherwise underestimate total flux and bias source parameters.
- 2.5 Mask optimization: The ellipse-based method fits a moment-0 ellipse, applies it across occupied velocity channels, and enlarges it until total flux reaches a maximum.
- 2.5 Mask optimization: The ellipse-based method is mainly suited to unresolved, face-on, or symmetric sources; for complex 3D structures it can reduce integrated signal-to-noise.
- 2.5 Mask optimization: SoFiA also offers binary dilation along the spatial axes and optionally by a user-selected number of frequency channels.
- 2.5 Mask optimization: After mask optimization, SoFiA recomputes basic parameters and measures additional quantities such as positions, flux extrema, sizes, and line widths.
2.7 Output products
SoFiA produces catalogues, masks, moment maps, cut-out cubes, spectra, and position-velocity diagrams for downstream analysis. Its current implementation processes cubes in memory on a single core, with runtime scaling approximately linearly with cube size.
- 2.7 Output products: SoFiA outputs object catalogues, final masks, moment-0 and moment-1 images, and object-specific data products.
- 2.7 Output products: Object cut-outs include masks, moment-0, moment-1, and moment-2 images, integrated spectra, and position-velocity diagrams along the morphological major axis.
- 2.7 Output products: These products support higher-level analyses, including measuring additional source parameters and producing Gauss-Hermite velocity fields for kinematical studies.
- 2.8 Performance of SoFiA: The current implementation loads the input cube into memory and processes it on a single core.
- 2.8 Performance of SoFiA: Execution time scales approximately linearly with cube size in the tested range, while memory usage is typically 2–3 times the cube size for large cubes.Occasional memory peaks reach 3–4 times the cube size.
- 2.8 Performance of SoFiA: The S+C finder takes most of the runtime in the example, with memory peaks from smoothing and plateaus from noise-level calculation.
- 2.8 Performance of SoFiA: Merging voxels, reliability processing, and moment-image creation account for later memory peaks, while other post-S+C algorithms are typically faster.
2.9 Source-finding based on a catalogue of 3D coordinates
SoFiA can search arbitrarily large cubes at catalogued 3D positions by processing fixed-size sub-cubes sequentially, while current whole-cube processing remains memory-limited.
- Current boundary: Whole-cube processing is currently limited by available system memory, although users can divide the cube into smaller regions and combine their products afterward.Integrated automatic partitioning of a large cube is described as future development.
- Catalogue-based search: SoFiA searches for emission around any number of catalogue-provided 3D coordinates within an arbitrarily large data cube.This mode is intended, for example, to search H I data at galaxies listed in an optical spectroscopic catalogue.
- Catalogue-based search: The software loads only the sub-cube surrounding each position, allowing large-cube searches without loading the entire cube into memory.Users set a common 3D sub-cube size, and positions are processed sequentially.
- Catalogue-based search: SoFiA can merge the catalogues from individual positions into one output source catalogue.
2.10 Comparison to other source finders
Compared with general-purpose packages and Duchamp/Selavy, SoFiA integrates a broader set of source-finding, reliability, parameterization, and output capabilities, while not providing full morphology analysis.
- Comparison: SoFiA integrates ready-to-use source-finding algorithms instead of requiring users to develop custom code around general-purpose interferometric software.
- Comparison: Duchamp uses threshold detection followed by source growth, whereas SoFiA provides different and more numerous algorithms for source finding and parameterization.The two packages are described as potentially complementary, with Duchamp performing best for unresolved sources in the cited comparison.
- Distinct capabilities: SoFiA uniquely combines methods including S+C and CNHI finding, 2D-1D wavelet denoising, reliability estimation, mask dilation, catalogue-based searches, and busy-function fitting.Cubelets and position-velocity diagrams are also available in SoFiA but not Duchamp, while Selavy provides them through distributed processing.
- Boundary: SoFiA does not provide full source-morphology analysis because nearby detected voxels are merged according to the selected merging element rather than decomposed into distinct components.Its object-mask cube can nevertheless serve as a starting point for further morphological analysis.
3 SUMMARY
SoFiA is a modular, publicly available package that combines flexible 3D spectral-line source-finding and parameterization methods for broad data-cube use. It supports multiscale and artefact-aware detection, reliability estimation, catalogue-guided searches, and varied analysis products, while remaining extensible.
- Scope: SoFiA combines source-finding and parameterization algorithms developed for ASKAP and APERTIF H I surveys into one package usable on any data cube.Its intended use is independent of the telescope or emission line observed.
- Detection: SoFiA searches for spectral-line signal across multiple spatial and frequency scales while accounting for noise variations, errors, and artefacts.These capabilities are presented as supporting complete and reliable detection and parameterization of 3D sources.
- Outputs: SoFiA estimates individual-detection reliability, supports catalogue-guided searches in arbitrarily large cubes, and produces catalogues, moment images, cut-outs, masks, position-velocity diagrams, and integrated spectra.
- Outputs: The output products are designed to describe detected sources and facilitate subsequent analysis.
- Modularity: SoFiA’s modularity lets users select algorithms for their data and sources of interest and allows future expansion with additional methods.The software is publicly available with technical information and regular updates.