Solar Physics最新文献

Sunspots are darkened regions on the Sun’s surface caused by intense magnetic disturbances and serve as crucial indicators of the solar cycle, which typically follows an approximately 11-year periodicity. These cycles have a profound influence on space weather, Earth’s climate, and the stability of modern technological systems, including satellite communications, GPS navigation, and power grid operations. The Sunspot Number (SSN) dataset, consisting of historical sunspot counts, represents a complex, nonlinear, and cyclic time series driven by solar magnetic activity. This study explores sunspot forecasting using advanced deep-learning architectures and large language model (LLM)-inspired approaches.

To improve prediction accuracy, three novel models are proposed: Hybrid H1 (CNN-BiLSTM-GRU), Hybrid H2 (CNN-GRU-RNN), and an Ensemble model that integrates both hybrid architectures with the Chronos framework. These models leverage convolutional layers for feature extraction, recurrent layers for temporal dependencies, attention mechanisms for enhanced focus on relevant patterns, and hyperparameter optimization to maximize performance across four SSN resolutions: daily, monthly mean, 13-month smoothed, and yearly data. Model performance is rigorously evaluated using RMSE, MAE, MSE, and (R^{2}) metrics, along with the Friedman ranking test and graphical analyses including line plots, scatter plots, box plots, and Taylor diagrams.

Experimental results demonstrate that all proposed models outperform traditional statistical methods and baseline deep learning and LLM-based models. For yearly SSN data, the Ensemble model shows the highest accuracy, with a 44% improvement over TimesFM and 27% over BiLSTM, while Hybrid H1 and H2 provide improvements of 41% and 42% over TimesFM, respectively. On the 13-month smoothed dataset, the Ensemble model improves forecasting by 47% over TimesFM and 15% over CNN, with Hybrid H1 and H2 yielding similar gains. For monthly mean data, the Ensemble model achieves an 18% improvement over TimeGPT and a remarkable 19% over LSTM. Daily data predictions show a 9.47% improvement for the Ensemble model over TimesFM and 21% over RNN, with Hybrid models following closely behind.

Using these best-performing models, long-term forecasts were extended on the yearly SSN dataset through the year 2110. The models predict peak sunspot activity in 2024, 2034, 2045, 2055, 2067, 2078, 2090, and 2102, corresponding to Solar Cycles 25 through 32. These predictions exhibit strong alignment with historical solar patterns, highlighting the reliability and advanced forecasting capacity of the proposed hybrid and ensemble deep-learning approaches, enriched by LLM-driven techniques, for modeling complex solar dynamics.

Magnetic-field parameters of solar active regions, such as the 16 Space-weather HMI Active Region Patches (SHARP) indices, are widely used to characterize active region magnetic complexity. Here we investigate whether deep-learning models can learn the physical information encoded in vector magnetograms well enough to recover these diagnostics directly from the images, thereby establishing a representation that can support subsequent downstream tasks. We propose a hybrid CNN-Swin Transformer regression framework, where the CNN extracts local magnetic features and the Swin Transformer captures large-scale spatial context through hierarchical, shifted-window self-attention. The model is trained on 750,430 SHARP vector magnetograms from 2011 – 2024 with a loss that combines mean squared error and Pearson correlation coefficient (PCC). On a held-out test set, the model achieves an average PCC of 0.934 across the 16 parameters. Performance is highest for geometry- and shear-related quantities (e.g., MEANGAM, MEANSHR, SHRGT45) and for non-potentiality proxies (e.g., MEANPOT), while current- and derivative-sensitive parameters remain more difficult (e.g., MEANJZD). These results indicate that the proposed model can recover physically meaningful SHARP diagnostics from vector magnetograms and can serve as a foundation for downstream active-region analysis and prediction tasks.

We present the Brightness–Location (BriLo) method, a novel single-spacecraft technique which exploits the Thomson scattering theory for localizing extended coronal features such as streamers using white-light (WL) imaging. Beyond determining the longitude and latitude of coronal features, the method also provides estimates of their geometrical properties, such as angular width (column depth). Validation is performed through geometrical triangulation with multi-viewpoint coronagraphs (the Solar TErrestrial RElations Observatory A COR2 and the Solar and Heliospheric Observatory C2–C3). The method is applied to ten coronal streamers observed by the Wide-Field Imager for Solar Probe (WISPR) on board the Parker Solar Probe (PSP) between encounter 1 – 17. We applied BriLo to two different data products, L3 and LX, which differ in K-corona treatment and absolute brightness levels. The L3 and LX results show good agreement in deriving streamer directionality, with differences of 2 – 30° in longitude and 1 – 6° in latitude. Both datasets provide longitude and latitude estimates that are broadly consistent with triangulation results. We further classified streamers and compared their locations with potential-field source surface (PFSS) extrapolations of the heliospheric current sheet (HCS). Helmet streamers are generally found close to the HCS, whereas pseudostreamers in proximity to active regions. In conclusion, the application of BriLo to LX data yields realistic streamer widths of several to ten degrees, while L3 data produce unrealistically narrow values below one degree. This discrepancy arises from the line of sight (LOS) integration of the observed signal and the dependence of F-corona removal on background estimation and coronal conditions. Overall, BriLo proves to be a robust tool not only for streamer localization but also for assessing and validating WL imaging techniques.

This study explores the dispersion properties of slow magnetoacoustic (MA) waves in solar coronal loops, emphasizing the combined effects of thermal conduction, thermal misbalance, and waveguide geometry to refine coronal seismological techniques. We analyzed the dispersion relations derived under infinite magnetic field and thin flux tube approximations, introducing characteristic timescales ((P_{mathrm{cond}}) for conduction-induced transition to isothermal sound speed, (P_{mathrm{mis}}) for misbalance effects, and (P_{mathrm{gI}}) for isothermal waveguide dispersion). The timescale (P_{mathrm{cond}}) is determined solely by plasma parameters and is unrelated to the waveguide size. These scales delineate bands where phase velocities approach adiabatic (c_{mathrm{S}}), isothermal (c_{mathrm{SI}}), misbalance-modified (c_{mathrm{SQ}}) sound speeds, or corresponding tube speeds (c_{mathrm{T}}), (c_{mathrm{TI}}), (c_{mathrm{TQ}}). Our analysis also reveals that, depending on the ratio (P_{mathrm{gI}} / P_{mathrm{cond}}), the behavior of dispersion curves can change qualitatively. The joint influence of waveguide dispersion and dispersion caused by thermal conduction expands the range of influence of the isothermal regime, which can in particular be expressed in the fact that the phase velocity even in the range of periods corresponding to the adiabatic regime may not reach the speed of sound (c_{mathrm{S}}), and the tube speed (c_{mathrm{T}}) for this band is a better estimate. Applied to observed hot loop oscillations, the model yields magnetic-field strengths of (sim 10~mathrm{G}), matching observed periods and damping times more accurately than adiabatic estimates alone.

Non-Linear Force-Free Fields (NLFFF) play a key role in our understanding of the nature and evolution of coronal magnetic fields. Two of the most common methods for their construction are the “extrapolation” and “evolution” approaches. The aim of the present paper is to compare results from these two approaches when they have the same vector magnetic field on the bottom boundary. To begin with a NLFFF evolution simulation of AR10977 is carried out to produce a time series of the vector field at the lower boundary covering the full life-span of the active region. Next at eight unique times in this time series, NLFFF extrapolations are constructed using the simulated vector boundary data. The resulting 3D coronal magnetic fields are then compared. During the early stages in the lifetime of AR10977, when the coronal magnetic field is composed of simply connected field lines, both NLFFF approaches produce a high level of agreement as long as the full vector field is injected into the extrapolation. When injection is limited to only strong field locations, a poorer agreement is found. In contrast, once a flux rope has formed during the later stages in the lifetime of the active region poor agreement is found between the two approaches, regardless of how the boundary information is injected in the extrapolations. This indicates that once a flux rope has formed through flux cancellation and risen into the corona the information held within the boundary vector field is insufficient to capture the complexity of the 3D coronal magnetic field. This result is also supported by the poor agreement that arises when comparing the relative magnetic helicity between the two modelling approaches. While the present study considers one extrapolation approach, it is important to repeat the study using alternative extrapolation methods that exist in the published literature.

Long-duration gamma-ray flare (LDGRF) events observed by the Fermi/ Large Area Telescope have presented a puzzle for modelers. One of the ideas to account for their long duration is to assume that particles accelerated at shocks driven by fast coronal mass ejections (CMEs) would be the able to be trapped in the space between the shock and the Sun and would be able to precipitate from this volume over an extended amount of time. We present a simple leaky-box type model for the precipitating > 500 MeV proton flux in a system, where a coronal shock feeds accelerated protons into the volume between the shock and the solar surface and a relatively small amount of scattering keeps the distribution isotropic and homogeneous inside the volume. We demonstrate that by choosing fully realistic shock parameters the total number of precipitating protons can be brought to an agreement with observations. We also demonstrate that durations of several hours for these events are fully within reach of the modeling without using unreasonable choices for parameters. Thus, CME-driven shocks in the coronal and inner solar wind plasma are a plausible candidate to account for the LDGRF events.

Large solar flares (GOES M-class or higher) are usually associated with eruptions of material. However, when considering flare irradiance enhancements and dynamics such as chromospheric evaporation, potential contributions from erupted material have historically been neglected. We analyse nine eruptive M- and X-class flares from 2024 to early 2025, quantifying the relative contributions of erupted material to irradiance enhancements during the events. Atmospheric Imaging Assembly (AIA) images from four different channels had ribbon and eruption irradiance contributions separated using a semi-automated masking method. The sample-averaged percentages of excess radiated energy by erupted material over the impulsive phase were (10^{+4}_{-4}%), (24^{+14}_{-14}%), (21^{+14}_{-10}%) and (13^{+6}_{-9}%) for the 131 Å, 171 Å, 304 Å and 1600 Å channels, respectively. For three events that were studied in further detail, hard X-ray (HXR) imaging showed little to no signatures of nonthermal heating within the eruptions. Our results suggest that erupted material can be a significant contributor to UV irradiance enhancements during flares, with possible heating mechanisms including nonthermal particle heating, Ohmic heating, or dissipation of MHD waves. Future work may clarify the heating mechanism and evaluate the impact of eruptions on spectral variability, particularly in Sun-as-a-star and stellar flare observations.

One of the many outcomes of the Solar Orbiter mission is the evidence for the solar atmosphere being filled by highly impulsive bursts, down to ≈ 200 (mathrm{km}) scale: the limit of the spatial resolution of EUV instruments. Small-scale events of this kind were already known, but their observation was occasional or with limited, lower resolution. Solar Orbiter has revealed that small-scale, highly impulsive events are everywhere on the quiet Sun, all the time, at even smaller scales. Their similarities with known larger features, are the witnesses that the physical processes causing them are independent of the spatial scales involved. Their highly dynamic property is the signature of energy transfer and/or local dissipation. Their investigation can thus elucidate on the dominant physical processes acting on the solar atmosphere and on the possible role in the origin of the hot solar corona.

In this review, we will present a summary of the observational and simulation results on this topic, led by the results from data taken by the Extreme Ultraviolet Imager (EUI)/High Resolution Imagers (HRI_EUV) instrument. Here, we will cover both statistical properties and analyses of individual events.

The solar chromosphere plays an important role in the transport of energy between the cool, dense photosphere and hot, tenuous corona. The determination of the thermodynamic and kinematics conditions of the chromosphere is crucial to understand the above coupling which is often done with the help of spectral inversion codes (ICs). We study and describe the performance of two ICs that are fundamentally unique in principle, implementation, and speed to infer the temperature of the solar chromosphere over a 3 arcmin field-of-view. We obtained spectral line scans of NOAA active region (AR) 12765 with the Ca ii 8542 Å line using the narrow-band imaging spectro-polarimeter on the 50-cm Multi-Application Solar Telescope at the Udaipur Solar Observatory, India. The temperature of the lower to mid-chromosphere was inferred from the following two ICs, namely, i) Non-LTE Inversion COde using the Lorien Engine (NICOLE) and ii) CAlcium Inversion based on a Spectral ARchive (CAISAR). A similar fit quality of the spectra is obtained with the resulting temperature stratification in good agreement with one another for a significantly large part of the field-of-view, despite the intrinsic differences in the two ICs. A few locations at the umbra-penumbra boundary and palge regions show a deviation in the two temperature stratifications which, to a first order, can be attributed to the generation of the archive spectra in CAISAR, where localised, Gaussian enhancements are provided to the temperature profile and the line-of-sight velocity is not considered/ignored. Once the spectral archive is created, CAISAR is several orders of magnitude faster than the conventional IC NICOLE, can be executed on a simple off-the-shelf workstation, and provide 3-dimensional temperature cubes with temporal information, that is ideally suited for processing observations from imaging spectrometers at most modern solar facilities.

In this study, we report on the processing and analysis of around 10,000 images of the solar disk, obtained from the SOHO and SDO satellites, available in the Helioviewer database. Using our own code, PyMoST (Python Morphological Sunspot Tracker), which employs a mathematical morphology technique, we were able to automatically track and analyze approximately 52,000 sunspots in these images, covering the period from January 1, 1998 to August 28, 2025. By using PyMoST we were able to obtain the distribution of the number of spots as a function of solar latitude and time, accurately reproducing Solar Cycles 23, 24 and the current phase of Solar Cycle 25, in agreement with the literature. In addition, our code was also able to determine the areas and number of sunspots, showing excellent correlation with the results reported by the Royal Greenwich Observatory (RGO), Kodaikanal Solar Observatory (KSO) and the International Sunspot Number - Solar Influences Data Analysis Center (SIDIC), thus consolidating the efficiency of our tool and its importance for the investigation of solar magnetic activity.