[Paper Review] Identification of coronal holes and filament channels in SDO/AIA 193 {A} images via geometrical classification methods

Martin Reiß, Manuela Temmer+3 more|arXiv (Cornell University)|Aug 12, 2014

Solar and Space Plasma Dynamics1 references3 citations

TL;DR

This study proposes and evaluates geometrical shape descriptors to distinguish coronal holes (CHs) from filament channels (FCs) in SDO/AIA 193Å EUV images, addressing misclassification in solar wind forecasting. Using a combination of established and novel shape measures—particularly direction-dependent neighbor analysis and symmetry analysis—it achieves a 17% overlap in classification, significantly reducing false positives and enabling real-time screening for improved solar wind speed predictions at 1 AU.

ABSTRACT

In this study, we describe and evaluate shape measures for distinguishing between coronal holes and filament channels as observed in Extreme Ultraviolet (EUV) images of the Sun. For a set of well-observed coronal hole and filament channel regions extracted from SDO/AIA 193 {A} images we analyze their intrinsic morphology during the period 2011 to 2013, by using well known shape measures from the literature and newly developed geometrical classification methods. The results suggest an asymmetry in the morphology of filament channels giving support for the sheared arcade or weakly twisted flux rope model for filaments. We find that the proposed shape descriptors have the potential to reduce coronal hole classification errors and are eligible for screening techniques in order to improve the forecasting of solar wind high-speed streams from coronal hole observations in solar EUV images.

Motivation & Objective

• To reduce misclassification of filament channels (FCs) as coronal holes (CHs) in automated EUV image analysis, which degrades solar wind speed forecasts.
• To investigate whether geometric properties—particularly asymmetry and elongation—can reliably distinguish FCs from CHs despite similar low intensity in EUV images.
• To evaluate the effectiveness of standard shape descriptors (e.g., compactness, elongation) and newly developed geometrical methods for CH/FC separation.
• To develop a fast, automated screening method applicable to real-time solar wind forecasting using SDO/AIA data.
• To validate that geometric features reflect underlying magnetic configurations, supporting models like the sheared arcade or weakly twisted flux rope for filaments.

Proposed method

• Applies histogram-shape-based thresholding to extract low-intensity regions from SDO/AIA 193Å images, identifying candidate CH and FC regions.
• Uses established shape measures: compactness (c = U²/4πA), roundness (r = 1/c), and elongation (e = dl/ds), where U is perimeter and A is area.
• Introduces two novel geometrical methods: (1) direction-dependent analysis of neighboring pixels per direction, normalized by maximum value, and (2) symmetry analysis based on pixel configuration symmetry.
• Employs an overlapping scanning technique to compute the standard deviation of the normalized neighbor count function as a shape descriptor.
• Tests all methods on a dataset of 2011–2013 CH and FC regions, comparing classification overlap and separation performance.
• Uses the overlap metric (percentage of misclassified regions) to evaluate method effectiveness, with lower overlap indicating better distinction between CHs and FCs.

Experimental results

Research questions

• RQ1 Can geometric shape descriptors reliably distinguish filament channels from coronal holes in SDO/AIA 193Å EUV images despite similar low intensity?
• RQ2 Do established shape measures like elongation, compactness, and roundness provide sufficient discrimination between CHs and FCs?
• RQ3 Do the newly proposed methods—symmetry analysis and direction-dependent neighbor analysis—improve classification accuracy compared to standard measures?
• RQ4 Can the proposed geometrical methods be applied in real-time to support automated solar wind forecasting at 1 AU?
• RQ5 Is the observed asymmetry in filament channel morphology consistent with the sheared arcade or weakly twisted flux rope model?

Key findings

• The direction-dependent neighbor analysis method achieved the lowest overlap of 17% in distinguishing FCs from CHs, outperforming standard shape measures.
• Symmetry analysis yielded an overlap of 20%, demonstrating strong potential for identifying asymmetric FCs versus more symmetric CHs.
• Standard shape measures (elongation, compactness, roundness) showed a clear trend but lacked sufficient distinctiveness for reliable separation.
• The proposed methods are invariant to translation and scaling, making them robust for automated analysis of solar EUV images.
• The algorithms are computationally efficient and suitable for real-time application in solar wind forecasting pipelines.
• The results support the sheared arcade or weakly twisted flux rope model for filaments, as FCs exhibit pronounced directional asymmetry and elongated morphology.