Automated Modeling with AAP-Imfit: Astrometry and Photometry via CASA

TL;DR

VLBI jet studies require precise flux decomposition of highly resolved but edge-poor intensity maps. The authors introduce AAP-Imfit, an automated pipeline that uses CASA imfit to fit 2-D Gaussian components, with a detection limit and artifact-removal workflow to extract accurate component fluxes and positions. Validation on 3C 279 and 3C 454.3 BEAM-ME/MOJAVE datasets shows close agreement with published fits in RMS metrics and model-to-map ratios, confirming the method's accuracy while highlighting the need for visual checks for complex features. This automation enables large-scale, reproducible VLBI jet analyses, dramatically reducing manual fitting time and facilitating statistical studies of jet dynamics and Doppler boosting.

Abstract

Very Long Baseline Interferometry (VLBI) provides the highest-resolution radio intensity maps, crucial for detailed studies of compact sources like active galactic nuclei (AGN) and their relativistic jets. Analyzing jet components in these maps traditionally involves manual Gaussian fitting, a time-consuming bottleneck for large datasets. To address this, we present an automated batch-processing tool, based on the Gaussian fitting capabilities of CASA, designed to streamline VLBI jet component characterization (AAP-Imfit). Our algorithm sets a detection limit, performs automatic 2D Gaussian fitting, and removes model artifacts, efficiently extracting component flux densities and positions. This method enables systematic and reproducible analysis, significantly reducing the time required for fitting extensive VLBI datasets. We validated AAP-Imfit by using VLBI observations of the blazars 3C 279 and 3C 454.3, comparing our results with published fits. The close agreement in residual root mean square (RMS) values and model/residual-to-map RMS ratios confirms the accuracy of our automated approach in reproducing original flux distributions. While visual inspection remains important for complex or faint features, this routine significantly accelerates VLBI component fitting, paving the way for large-scale statistical studies of jet dynamics.

Figures (10)

• Figure 1: Schematic of the AAP-Imfit routine. This flowchart is applied to each map to be modeled retrieving the astrometry and photometry of each Gaussian fitted. The algorithm consists of four main steps: First, all the parameters required by the code are extracted from the header of the FITS/IMAP file. Next, a detection limit is calculated, which separates real source emission from the background level. The final step involves removing Gaussian components that do not accurately represent the flux distribution. Finally, the map, model, and residuals are plotted for visualization, accompanied by a summary of the component properties and a general overview of all the fitted maps. Inputs and outputs are denoted in green and purple, respectively. Sections that use CASA tasks are denoted in yellow.
• Figure 2: Examples of different detection limits calculated for the VLBI intensity map of the blazar source 3C 279 under the BEAM-ME monitoring program. Top row: Observed map (Left) taken at 43 GHz on 2015 August 1, convolved with a beam of $0.15\times0.36$ mas$^{2}$ and a position angle of $-10^{\circ}$, background map (Middle), and RMS map (Right) returned by the detection limit function. The grey contours correspond to contour levels of 0.2, 0.4, 0.7, 1.4, 2.9, 5.8, 11.5, 23.0, 46.0, and $92.1\%$ of the peak total intensity. Bottom row: Masked pixels from the observed map with the detection limit calculated to be 6 (Left), 9 (Middle), and 12 (Right) times the median RMS. Pixels with values greater than the detection limit are masked with a value of 1, while pixels with values lower than the detection limit are set to be 0.
• Figure 3: Left panel: VLBI intensity map from the 3C 279 observation at 43 GHz in 2015 September 22, convolved with a beam of $0.15\times0.36$ mas$^{2}$ and a position angle of $-10^{\circ}$. Middle panel: Model of the observed flux distribution generated by the algorithm. Right panel: Residual image. The grey contours correspond to contour levels of 0.37, 0.75, 1.50, 3.00, 6.00, 12.00, 24.00, 47.93, and $95.90\%$ of the peak total intensity. The first contour level is at the detection limit.
• Figure 4: Upper row: Left panel: VLBI intensity map from the 3C 454.3 observation at 43 GHZ in 2016 December 23, convolved with a beam of $0.14\times0.33$ mas$^{2}$ and a position angle of $-10^{\circ}$. Middle panel: Model of the observed flux distribution generated by the algorithm. Right panel: Residual image. The grey contours correspond to contour levels of 0.20, 0.40, 0.79, 1.60, 3.17, 6.40, 12.68, 25.36, 50.72, and $99.00\%$ of the peak total intensity. Bottom row: Similar as above Left panel: VLBI intensity map from the 3C 454.3 observation at 15 GHz in 2016 August 09, convolved with a beam of $0.44\times1.00$ mas$^{2}$ and PA$=-3.75^{\circ}$. Middle panel: Model of the observed flux distribution delivered by the algorithm. Right panel: Residual image. The grey contours correspond to contour levels of 0.02, 0.05, 0.10, 0.21, 0.41, 0.83, 1.66, 3.32, 6.63, 13.27, and $36.73\%$ of the peak total intensity. The first contour level at both observations is at the detection limit.
• Figure 5: Comparison between RMS values for the source 3C 279 using the observations from the BEAM-ME survey. (a) Upper: Rate between the model and map RMS, close values to one are expected if the model accurately retrieves the original flux distribution. Bottom: Percentage differences between both rates. (b) Upper: Model RMS, similar values between samples are expected if the routine retrieves similar models to the control sample. Bottom: Percentage differences between both model RMS values with respect to the observed map RMS. (c) Upper: Rate between the residual and map RMS, values close to zero are expected if the residual does not show source information and only background info. Bottom: Percentage differences between both rates. (d) Upper: Residual RMS, similar values between samples are expected if the routine retrieves a similar residual to the control sample. Bottom: Percentage differences between both residual RMS values with respect to the map RMS. Green dashed lines represent the linear one-to-one relation. The Red dashed line denotes zero difference.