The research, led by Oleg Stepanyuk of the Institute of Astronomy and National Astronomical Observatory at the Bulgarian Academy of Sciences, together with Werner Pötzi of the Kanzelhöhe Observatory for Solar and Environmental Research at the University of Graz, Momchil Dechev, Rositsa Miteva, and Kamen Kozarev, extends a hybrid algorithmic and data-driven approach the team had previously applied to data from the Atmospheric Imaging Assembly aboard NASA’s Solar Dynamics Observatory. In that earlier work, the method was used to segment extreme ultraviolet waves and shock fronts associated with solar eruptions. The new effort shifts the focus to hydrogen-alpha observations from Kanzelhöhe Observatory, an Austrian facility that has monitored the Sun with high temporal resolution for decades and whose data are now integrated into the European Space Agency’s Space Weather Service Network.

Why does ground-based hydrogen-alpha imaging matter for this task? Space telescopes offer uninterrupted coverage, but ground-based instruments provide a long historical archive and frequent sampling during daylight hours, capturing the slow morphological evolution of filaments that often precedes an eruption. Researchers have long noted that changes in a filament’s shape, height, and internal motion can signal that the magnetic flux rope anchoring it is approaching a critical state. Theoretical work on mechanisms such as the torus instability and kink instability, together with statistical studies linking prominence destabilization to coronal mass ejections, has established that filaments are not merely passengers on erupting structures but often visible tracers of the eruption process itself. Systematic, automated monitoring of these tracers across many years of data is therefore a natural foundation for forecasting.

The technical core of the new study is image segmentation: teaching a neural network to label, pixel by pixel, which parts of a full-disk solar image belong to a filament. The team employed compact U-Net-based architectures, a family of convolutional neural networks originally developed for biomedical image segmentation that have become a workhorse in solar physics as well. U-Net’s encoder-decoder design, with skip connections that carry fine spatial detail from the encoding path to the decoding path, is well suited to the elongated, low-contrast, and highly variable shapes of solar filaments. The models were pre-trained on synoptic and normalized image data, and the researchers systematically examined how performance depends on training-set engineering, preprocessing choices, and the selection of loss metrics used to guide optimization.

One of the study’s most distinctive contributions is a leakage-aware, sparse multi-year sampling strategy. In machine learning applied to images, data leakage occurs when samples that are too similar to one another end up split between training and test sets, inflating apparent performance. Solar images are especially vulnerable to this problem because filaments persist for days and evolve slowly: consecutive frames are nearly identical, and even images taken days apart can share large structures. By carefully spacing the sampled images across multiple years and accounting for temporal correlation, the team ensured that their reported accuracy reflects genuine generalization to unseen solar conditions rather than memorization of long-lived features. This kind of methodological rigor, borrowed in part from lessons learned in other imaging domains such as digital pathology, is essential if segmentation models are to be trusted in operational settings.

Perhaps the most conceptually interesting result concerns normalization, the practice of rescaling input data or internal network activations to stabilize training. Deep networks commonly rely on techniques such as batch normalization or weight normalization, and practitioners routinely apply per-image normalization schemes such as z-score or min-max rescaling to raw inputs. The authors performed a combined analytical and numerical study of how externally imposed, fixed normalization compares with letting the network work from native input values in compact U-Net models for hydrogen-alpha filament segmentation. Their conclusion is striking: synoptic photometric calibration already supplies a physics-aware normalization, internal normalization layers supply learned, per-channel scaling, and an additional external per-image normalization inserted between the two is at best redundant and at worst actively destructive for sparse, photometrically defined classes like hydrogen-alpha filaments.

The practical guidance that follows from this analysis is concrete. The authors recommend keeping synoptic photometric calibration intact rather than stripping it away with per-image rescaling, applying at most a fixed, dataset-global linear rescale to bring inputs into a range suitable for optimizer stability, and placing normalization blocks immediately after each skip-connection concatenation in the decoder so that encoder and bottleneck streams are equalized before subsequent convolutions. For training across multiple instruments, such as combining data from ChroTel, the Global Oscillation Network Group, and Kanzelhöhe, they suggest histogram matching to a canonical reference frame rather than per-image equalization, preserving relative photometry while removing instrument-to-instrument drift. They also recommend leaving the final one-by-one convolution un-normalized and using Dice or Tversky losses, which handle the severe class imbalance that arises when thin filaments occupy only a small fraction of each image.

This attention to compact, efficient models reflects a broader trend in solar research. Lightweight segmentation networks are increasingly being deployed on edge computing devices and field-programmable gate arrays, enabling real-time analysis at observatories without recourse to large data centers. Related efforts in the literature include ultralightweight U-Net variants designed specifically for full-disk hydrogen-alpha filament segmentation, semi-supervised deep learning methods for universal filament detection, and object-detection frameworks adapted for solar features. The Kanzelhöhe-based work contributes to this ecosystem not only with its models, which the team publishes and regularly updates in a public repository, but with hard-won engineering knowledge about what actually matters when training data are sparse, photometric conventions are meaningful, and the target class is geometrically delicate.

The scientific payoff of reliable automated filament segmentation extends well beyond cataloging. Once filaments can be consistently detected and tracked across years of observations, their morphological evolution can be correlated with eruption onset, providing quantitative precursors for forecasting models. Filament material itself has been directly observed within interplanetary coronal mass ejections, confirming the physical chain from quiescent structure to Earth-directed disturbance. Statistical studies of critical heights for prominence destabilization and of the relationship between erupting filaments and coronal mass ejection kinematics all depend on accurate, consistent measurements of filament position, area, and shape over time, measurements that manual cataloging cannot deliver at the scale of modern archives.

Funded by the Bulgarian National Science Foundation and Austria’s Agency for Education and Internationalisation, the work exemplifies how international collaboration between observatories and research institutes can convert routine monitoring data into forecasting capability. As solar activity cycles through its maxima and the volume of ground-based imagery continues to grow, tools like the ones described in this study will become indispensable sentinels, quietly scanning the Sun’s dark ribbons for the first signs that a storm is brewing. For a civilization increasingly dependent on satellites, aviation routes over the poles, and interconnected power infrastructure, teaching machines to read the Sun’s warnings may prove one of the most quietly consequential applications of artificial intelligence in the solar sciences.

Subject of Research: Deep learning segmentation of solar filaments from ground-based hydrogen-alpha observations for space weather forecasting

Article Title: Data-Driven Segmentation of Solar Filaments Based on Ground-Based Instrument Data

Article References: Data-Driven Segmentation of Solar Filaments Based on Ground-Based Instrument Data. (n.d.). https://doi.org/10.1007/s11207-026-02741-y

Image Credits: AI Generated

DOI: 10.1007/s11207-026-02741-y

Keywords: solar filaments, solar prominences, coronal mass ejections, space weather, deep learning, U-Net, image segmentation, normalization, Kanzelhöhe Observatory, hydrogen-alpha, Solar Physics, machine learning

The research, led by Ying Li of Kunming University of Science and Technology in Yunnan, China, together with colleagues at the same institution and at Yunnan Observatories of the Chinese Academy of Sciences, introduces a tracking architecture called BF-TrackFormer. The name reflects its core innovation: a bidirectional, autoregressive approach to tracking that works both forward and backward in time. The team built the system specifically to address a problem that has long frustrated automated filament monitoring, namely that filaments do not behave like tidy, well-behaved objects. They split, merge, twist, and fragment, and conventional tracking algorithms routinely lose track of them when their outlines break apart or when a detector momentarily fails to recognize them.

To train and evaluate the new method, the researchers assembled a dataset of 630 H-alpha full-disk solar images drawn from three distinct observational sources: the Big Bear Solar Observatory in California, the Solar Magnetic Activity Research Telescope operated by Kyoto University in Japan, and the Global Oscillation Network Group, a worldwide network of six stations operated by the U.S. National Solar Observatory that provides nearly continuous coverage of the Sun. The dataset was divided into one training set and five testing sets, a deliberate design choice meant to expose the algorithm to different data sources, time intervals, and observing conditions. This diversity matters because H-alpha imagery varies considerably from one instrument to another in resolution, calibration, and image quality, and a tracking system that only works on data resembling its training set would be of limited practical value for operational forecasting centers.

The technical foundation of BF-TrackFormer builds on TrackFormer, a transformer-based multi-object tracking framework originally developed for computer vision applications. Transformers, the same architectural family that powers modern language models, process images by attending to relationships between different regions of the frame, allowing a detector to identify objects and assign them consistent identities across successive frames in a single pass. The Chinese team adapted this paradigm to the peculiar challenges of solar imagery, where filaments appear as dark absorption features against the bright chromosphere and where their boundaries can be diffuse, elongated, and highly irregular compared with the compact objects typical of pedestrian or vehicle tracking benchmarks.

The genuinely novel component is the backward autoregressive branch, which consists of two modules the authors call the ID Embedding Head and the Recheck Network. In a standard forward-only tracker, each new frame is processed in sequence, and objects are matched to tracks established in previous frames. If a filament splits into two pieces, the detector may capture only one fragment, or if a filament temporarily fades below the detection threshold, the track is broken and the object reappears later as a brand-new entity with a new identity. The backward branch effectively gives the algorithm a second chance: it revisits frames where objects were missed or misclassified as background, re-embedding identity information so that fragments of a split filament can be reassociated with their parent track, and discontinuous segments of a single filament’s lifecycle can be stitched back together into one coherent trajectory.

This capability is scientifically significant because fragmented filament evolution is not merely a nuisance for software; it is a physical signal. Observational studies in recent years have documented that filament splitting, bifurcation, and reconstruction are often driven by magnetic reconnection and the emergence of new magnetic flux, processes that can precede or accompany eruption. A filament that breaks apart in H-alpha images may be undergoing the very magnetic restructuring that signals instability. If an automated system fragments a single physical filament into several spurious tracks, or loses it entirely at the critical moment, the observational clues that forecasters and researchers rely on are corrupted. By maintaining identity through splitting events, BF-TrackFormer preserves the integrity of the lifecycle record, which is precisely what statistical studies of filament behavior and eruption prediction models require as input.

The performance figures reported in the paper illustrate both the promise and the remaining difficulty of the task. Averaged across the five testing sets, the method achieved an IDF1 score of 70.1 percent, a measure of how consistently identities are maintained over time; a MOTA of 46.6 percent, the standard multi-object tracking accuracy metric that combines false positives, misses, and identity switches; precision of 75.1 percent; recall of 70.9 percent; and an identity survival rate of 10.1 percent. The authors note a clear trend: as the time interval between consecutive images decreases, all of these metrics improve. This makes intuitive sense, since filaments evolve slowly by human forecasting standards but can change enough between widely spaced frames to confuse any tracker, and denser temporal sampling gives the autoregressive machinery more reliable continuity to work with.

The relatively low identity survival rate underscores how hard the problem remains. Filaments can persist on the Sun for days or weeks, and tracking one across its full lifetime requires surviving changing illumination, projection effects as solar rotation carries features across the disk, overlapping structures, and the episodic fragmentation that motivated the study in the first place. Even so, the authors report that BF-TrackFormer performs particularly well on fragmented filaments, the case that conventional methods handle worst. In the ecosystem of automated solar monitoring, where earlier systems relied on classical image processing techniques such as morphological operations, region growing, and hand-tuned thresholds, the shift to learned, identity-aware tracking represents a generational change in how filament catalogs can be built.

The broader context is the growing recognition that space weather is an operational hazard demanding the same vigilance as terrestrial weather. Filament eruptions are closely associated with coronal mass ejections, the massive expulsions of plasma and magnetic field that are the primary drivers of major geomagnetic storms. Statistical studies spanning multiple solar cycles have established that a large fraction of filament disappearances coincide with coronal mass ejections, making filaments one of the most informative precursors available to forecasters. Automated detection and tracking systems, running continuously on ground-based H-alpha networks, can in principle flag filament activation, disappearance, and fragmentation in near real time, extending the reach of human observers who cannot watch every active region simultaneously.

The authors acknowledge the contributions of the observatories whose data made the work possible, including Big Bear Solar Observatory, the Global Oscillation Network Group, the Solar Magnetic Activity Research Telescope, and the CHASE satellite mission supported by the China National Space Administration. As solar maximum approaches its most active phases and humanity’s dependence on space-based infrastructure deepens, tools like BF-TrackFormer point toward a future in which the Sun’s most dangerous structures are watched continuously, their every split and twist logged by algorithms that never blink, and their eruptive potential assessed with the statistical rigor that only complete, unbroken lifecycle records can provide. The research is a reminder that forecasting the storm begins with faithfully following the calm before it.

Subject of Research: A bidirectional autoregressive deep learning method for detecting and tracking solar filaments across their lifecycles to improve space weather forecasting.

Article Title: Bidirectional Autoregressive Tracking Method for Solar Filaments

Article References: Li, Y., Yang, Y., Zhang, X., Feng, S., Dai, W., Liang, B., & Xiong, J. (2026). Bidirectional Autoregressive Tracking Method for Solar Filaments. Solar Physics, 301(9), Article 136. https://doi.org/10.1007/s11207-026-02737-8

Image Credits: AI Generated

DOI: 10.1007/s11207-026-02737-8

Keywords: solar filaments, space weather, deep learning, multi-object tracking, H-alpha imaging, coronal mass ejections, BF-TrackFormer, solar physics, transformers, Big Bear Solar Observatory, GONG, filament eruptions