An Improved HDBSCAN-based Detection and Tracking Method for Solar Active Regions in Magnetograms
C. X. Shi, Q. Hao, P. F. Chen, Y. Guo · 2026
Read An Improved HDBSCAN-based Detection and Tracking Method for Solar Active Regions in Magnetograms by C. X. Shi, Q. Hao, P. F. Chen, Y. Guo (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23150.txt.
Introduction
Shi and three coauthors present HARDAT, an HDBSCAN based method for detecting and tracking active regions in magnetograms. Active regions drive storms and space weather, and follow-up matters for forecasts. Their DBSCAN based method had cutoffs, while HDBSCAN handles multi-density structures on the fly, finding diffuse regions well while keeping shape. A spin follow-up method with real motion models and Hamming distance likeness achieves multi-object follow-up. Tested on SOHO and SDO maps against NOAA and earlier lists, HARDAT wins on accuracy and stability.
Catalog record
| CREATOR | C. X. Shi, Q. Hao, P. F. Chen, Y. Guo |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49002 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.23150
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
An Improved HDBSCAN-based Detection and Tracking Method for Solar Active Regions in Magnetograms C. X. Shi, Q. Hao, P. F. Chen, Y. Guo Abstract Solar active regions (ARs) are the primary source of solar eruptions and space weather. Accurate detection and tracking of ARs is crucial for understanding their evolution and predicting solar activities. In the previous work, based on the density-based spatial clustering of applications with noise (DBSCAN) approach, we proposed the DBSCAN-based solar active region detection (DSARD) framework. To overtake its limitations, in this paper we applied the hierarchical density-based spatial clustering of applications with noise (HDBSCAN) approach to the detection of solar active regions, which is called the HDBSCAN-based solar active region detection and tracking (HARDAT) method. This enables the algorithm to handle multi-density magnetic structures dynamically, eliminating the need for fixed thresholds. Consequently, the algorithm can detect diffuse and small ARs more effectively while preserving morphological integrity. We have also developed a solar differential rotation based tracking algorithm that integrates physical motion models and Hamming distance similarity metrics to achieve robust multi-object tracking. Additionally, we propose a novel polarity inversion line extraction method that uses support vector classification, which offers superior generalization for complex AR boundaries. Processing line-of-sight magnetograms from SOHO/MDI (1996–2011) and SDO/HMI (2010–2024) and evaluating them against the National Oceanic and Atmospheric Administration (NOAA) and DSARD catalogues demonstrates that HARDAT is superior in terms of sensitivity, accuracy, and stability of detection and tracking. This is particularly evident when resolving clustered ARs and maintaining identity continuity. HARDAT therefore offers a comprehensive solution for the long-term analysis of AR evolution and space weather prediction. show][email protected] I Introduction Solar active regions (ARs) are localized areas of strong magnetic fields on the Sun. They are often associated with sunspots, flares and coronal mass ejections (CMEs; 30 ). Research on ARs involves analysing magnetic field structures, their evolution and the mechanisms by which energy accumulates within these regions, to assess the potential risk of eruptive events, such as solar flares and CMEs ( 5 ) . Studying them is essential for understanding solar dynamics and forecasting space weather ( 28 ) . High-resolution and high-cadence magnetographs, such as the Michelson Doppler Imager (MDI; 20 ) on board the Solar and Heliospheric Observatory (SOHO; 8 ) and the Helioseismic and Magnetic Imager (HMI; 22 ) on board the Solar Dynamics Observatory (SDO; 15 ), have rendered traditional processing methods inadequate for handling their massive datasets. These methods rely on manual inspection or threshold-based image processing. Consequently, researchers have developed a series of automated detection methods for ARs in magnetograms ( 33 ; 29 ; 3 ; 16 ; 10 ; 32 ; 6 ) . These methods provide crucial data support for the studies of solar ARs, solar activities, long-term solar dynamo modeling. In particular, we recently applied the unsupervised desity-based spatial clustering of applications with noise (DBSCAN; 24 ) algorithm to develop the DBSCAN-based solar active region detection (DSARD; 6 ) method. DSARD eliminates the need for the vast high-quality annotated data that are inherent in supervised learning, thereby reducing subjective bias. Its density-based nature enables the adaptive detection of ARs across varying intensities, including small-scale transient magnetic features that are often overlooked by conventional thresholding methods. This significantly enhances the accuracy and robustness of detection. Employing staged clustering in a two-stage DBSCAN density clustering process effectively avoids the tendency of the traditional methods (such as region growing and morphological algorithms) to excessively merge adjacent ARs or perform inappropriate segmentation within high-resolution magnetograms. This significantly improves the capability to analyze fine-scale magnetic structures. Compared with the National Oceanic and Atmospheric Administration (NOAA) and Spaceweather HMI Active Region Patch (SHARP; 3 ), the DSARD method achieved accuracy rates of 89.7\% and 78.2\% , respectively, whilst identifying small-scale active regions that are overlooked by conventional methods. We conducted detailed statistical analyses of characteristics such as the number, area, magnetic flux, tilt angle, and drift velocity of active regions during the solar cycle 24 and the rising phases of the solar cycle 25 using the active region database generated by the DSARD method. The majority of the statistical results were consistent with those of previous studies, thereby validating the reliability of the DSARD method. Furthermore, our research yielded several novel key conclusions such as the finding that large ARs adhere to Joy’s Law, whereas small ARs exhibit random distribution patterns. The proportion of ARs that violate Hale’s Law is markedly higher than in previous studies. However, we also found that the DSARD method had some limitations. This is because the DBSCAN algorithm uses fixed density threshold and distance parameter, which restricts its ability to detect diffuse regions and multi-density magnetic field structures. Furthermore, DSARD does not track the detected ARs. To avoid duplicate detections and minimise the projection effects, a margin of \pm 6^{\circ} from the center meridian of the solar disk is selected when evaluating the detection results. Consequently, the true positive rate of the results is therefore limited. Hierarchical density-based spatial clustering of applications with noise (HDBSCAN; 13 ; 4 ) is a powerful extension of DBSCAN that can handle clusters with varying densities by leveraging the concepts of mutual reachability distance and cluster stability. In this paper, we propose an HDBSCAN-based solar active region detection and tracking (HARDAT) method. The adoption of HDBSCAN makes it particularly suitable for magnetograms, where ARs exhibit a wide range of sizes and magnetic density distributions. Our method improves upon DSARD by incorporating dynamic density clustering and a tracking module informed by physics that accounts for solar differential rotation ( 23 ; 18 ) . We also propose a new polarity inversion line (PIL) extraction method based on support vector classification (SVC; 9 ). This method has a stronger generalization ability than the traditional mask-based method and can simultaneously extract magnetic neutral lines in dense and diffuse magnetic fields. The paper is organized as follows: Section II describes the data used in this study. Section III details the improved detection methodology, including HDBSCAN clustering and integration strategies. Section IV presents the tracking algorithm based on solar differential rotation. Section V introduces the SVC-based PIL extraction method. Section VI provides a discussion and conclusion, highlighting the advantages of HARDAT over existing methods. II Data Acquisition This study uses the solar full disk line-of-sight (LOS) magnetograms, which were provided by SOHO/MDI from the year 1996 to 2011 and SDO/HMI from 2010 to 2024. The SOHO/MDI magnetogram dataset provides solar photospheric magnetic distribution with a spatial resolution of 4″ (an image size of 1024×1024 pixels). SDO/HMI, the successor to SOHO/MDI, has greater temporal and spatial accuracy. It provides solar magnetograms with a spatial resolution of 1″ (an image size of 4096×4096 pixels). And we use mdi.fd\_M\_96m\_lev182 and hmi.M\_720s magnetorgrams data series for this study. As the cadences of these two datasets are differ, the sampling cadence of the data is set to 192 minutes for SOHO/MDI magnetograms and 3 hours for SDO/HMI magnetograms, ensuring the consistency in subsequent tracking accuracy. This yields solar magnetogram data spanning a total of approximately 28 years, from 1996 to 2024, including the whole solar cycles 23 and 24, and the rising phase of solar cycle 25. All data for this study