Quantitative Image-Based Validation Framework for Assessing Global Coronal Magnetic Field Models

TL;DR

This work introduces a quantitative framework to validate image-based coronal magnetic-field tracing against global MHD models, uniting COR-1 white-light observations, QRaFT feature tracing, MAS/MHD outputs, and FORWARD synthetic-pB modeling. By computing pixel-wise angular differences between traced quasi-radial structures and model-driven magnetic-field orientations, and by analyzing the resulting distributions with KDE, JSD, KLD, and Tukey HSD metrics, the authors demonstrate that large-scale coronal magnetic geometry can be recovered to within about $\pm 10^\circ$ of the MAS ground truth. They also decompose errors into graphical and physical sources, including observational artifacts, model boundary conditions, and tracing limitations, enabling targeted improvements. The framework provides a transferable, interpretable set of metrics for validating coronal-segmentation methods and constraining boundary conditions in space-weather models, with broader potential to integrate ground-based data and optimize coronal heating and magnetic boundary inputs. Overall, the study establishes a robust data-model cross-validation approach that can improve both segmentation algorithms and coronal simulations, advancing our ability to forecast space-weather impacts.

Abstract

Coronagraph observations provide key information about the orientation of the Sun's magnetic field. Previous studies used various algorithms to segment quasi-radial features in coronagraph images and approximate their local plane-of-sky geometry and orientation which can be used as input for optimizing and constraining coronal magnetic field models. We present a new framework that allows for further quantitative evaluations of image-based coronal segmentation methods against magnetic field models, and vice-versa. We compare quasi-radial features identified from QRaFT, a global coronal feature tracing algorithm, in white-light coronagraph images to outputs of MAS, an advanced magnetohydrodynamic model. We use the FORWARD toolset to produce synthetic polarized brightness images co-aligned to real coronagraph observations, segment features in these images, and quantify the difference between the inferred and model magnetic field. This approach allows us to geometrically compare features segmented in artificial images to those segmented in white-light coronagraph observations against the plane-of-sky projected MAS coronal magnetic field. We quantify QRaFT's performance in the artificial images and observational data, and perform statistical analyses that measure the accuracy and uncertainty of the model output to the observational data. The results demonstrate that a coronal segmentation method identifies the global large-scale orientation of the coronal magnetic field within $\sim\pm10^\circ$ of the plane-of-sky projected MAS magnetic field.

Figures (11)

• Figure 1: Flowchart style overview of defined framework for quantitatively assessing image-based coronal segmentations against magnetic field models, and vice-versa. Each arrow represents an output from a component of the framework that is used as an input into the component the arrow is connected to.
• Figure 2: (Left) Synthetic forward pB image co-aligned to the equivalent observational perspective as the cor-1 observation. (Right)cor-1 pB observation on 2017-08-29.
• Figure 3: qraft processing steps shown for (left) a synthetic pB image produced by mas and the forward code and (right) an observed pB image obtained from cor-1 on 2017-08-29. The qraft image is aligned with the cor-1 image as explained in the text. In each example the upper left panel (a) is the original cor-1 pB image after radial detrending and smoothing, the bottom left panel (b) is the same image in plane polar coordinates, the top right panel (c) is the signed second-order azimuthal difference, the bottom right panel (d) is the unsigned second order difference after azimuthal detrending revealing a fine quasi-radial structure in the example observed image, and (e) are the resulting segmented features created by tracing the azimuthal gradients and automatically removing unrealistic features.
• Figure 4: Polar plot showing the Carrington longitudes of each cor-1 observation chosen in this study. These six observations correspond to cr 2194 and were chosen to provide roughly $60^\circ$ difference of Carrington longitude between each observation.
• Figure 5: (Left) maspos B orientation with QRaFT filtered features from ne central model image for 2017-08-29. (Right) Model pos B orientation with QRaFT filtered features from cor-1 representative median image for 2017-08-29. The red lines in each plot represent the segmented features from qraft. The green arrows represent the directions of the mas pos B for the 2017-08-29 model image, while the black lines represent the corresponding magnetic field lines.