Inversion of CHASE H α \alpha Spectral Line during Solar Flares Based on RADYN Dataset via Deep Learning Volume: 3 Issue: 3
W. Xu \ORCID 0000-0002-2919-6989, Q. Hao \correspondingAuthor \ORCID 0000-0002-9264-6698, Z. Zheng \ORCID 0009-0006-5180-8052, J. Hong \ORCID 0000-0002-8002-7785, J. Hu \ORCID 0009-0000-5660-0572, Y. Qiu \ORCID 0000-0002-1190-0173, C. Li \ORCID 0000-0001-7693-4908, M.D. Ding \ORCID 0000-0002-4978-4972, C. Fang · 2026
Read Inversion of CHASE H α \alpha Spectral Line during Solar Flares Based on RADYN Dataset via Deep Learning Volume: 3 Issue: 3 by W. Xu \ORCID 0000-0002-2919-6989, Q. Hao \correspondingAuthor \ORCID 0000-0002-9264-6698, Z. Zheng \ORCID 0009-0006-5180-8052, J. Hong \ORCID 0000-0002-8002-7785, J. Hu \ORCID 0009-0000-5660-0572, Y. Qiu \ORCID 0000-0002-1190-0173, C. Li \ORCID 0000-0001-7693-4908, M.D. Ding \ORCID 0000-0002-4978-4972, C. Fang (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-2605-26511.txt.
Introduction
W. Xu and eight colleagues train a fully connected neural network to read solar-flare physics from hydrogen-alpha spectra. Direct measures of electron density, temperature, and plasma velocity during flares stay hard, so the network learns rapid inversion from synthetic RADYN simulation data. Spectra were matched to CHASE satellite resolution for direct use on real observations. Errors stay low across varied flare conditions, and the model matches RADYN closely. Applied to an X7.1 flare on October 1, 2024, it traces sensible spatial and temporal evolution of key parameters through flare phases.
Catalog record
| CREATOR | W. Xu \ORCID 0000-0002-2919-6989, Q. Hao \correspondingAuthor \ORCID 0000-0002-9264-6698, Z. Zheng \ORCID 0009-0006-5180-8052, J. Hong \ORCID 0000-0002-8002-7785, J. Hu \ORCID 0009-0000-5660-0572, Y. Qiu \ORCID 0000-0002-1190-0173, C. Li \ORCID 0000-0001-7693-4908, M.D. Ding \ORCID 0000-0002-4978-4972, C. Fang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32671 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.26511
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Inversion of CHASE H α \alpha Spectral Line during Solar Flares Based on RADYN Dataset via Deep Learning Volume: 3 Issue: 3 W. Xu \ORCID 0000-0002-2919-6989, Q. Hao \correspondingAuthor \ORCID 0000-0002-9264-6698, Z. Zheng \ORCID 0009-0006-5180-8052, J. Hong \ORCID 0000-0002-8002-7785, J. Hu \ORCID 0009-0000-5660-0572, Y. Qiu \ORCID 0000-0002-1190-0173, C. Li \ORCID 0000-0001-7693-4908, M.D. Ding \ORCID 0000-0002-4978-4972, C. Fang Abstract Solar flares represent one of the most intense forms of solar activity. Understanding the evolution of physical parameters in the solar atmosphere during flares is key to studying flare mechanisms and improving prediction capabilities. However, directly measuring quantities such as electron number density, temperature, and plasma velocity remains difficult. Here, we introduce a novel fully connected neural network, trained on synthetic data from the Radiative Hydrodynamics Code (RADYN) simulations, to perform rapid inversion of physical parameters from H α \alpha spectral profiles. The spectral data were processed to align with the observational resolution of the CHASE satellite, enabling seamless application of the model to real-world observations. Results demonstrate a high degree of consistency with RADYN simulations, achieving low errors under diverse flare conditions. Furthermore, we applied the developed model to analyze CHASE observations of a class X7.1 solar flare on October 1, 2024. The results reveal reasonable spatial and temporal evolution of key parameters throughout different flare phases. This work demonstrates the potential of deep learning techniques for fast and reliable spectral inversion, providing new tools for solar flare diagnostics based on H α \alpha data. \correspondent Qi Hao \correspondentEmail [email protected] \citeinfo Xu, W. et al. \pages 1–11 \StartPage 1 \MonthIssue May \copyrights 2026 keywords Solar Flare — Spectral Inversion — Deep Learning 1 Introduction Solar flares are one of the most intense forms of solar activity [ 21 ] . They are characterized by sudden, large-scale energy releases in localized regions of the solar atmosphere [ 10 ] . This process causes rapid heating and the emission of a broad spectrum of electromagnetic radiation. The wavelength of this radiation spans the entire electromagnetic spectrum. During solar flares, energy is deposited in the photosphere and chromosphere through mechanisms such as ion beam bombardment, thermal conduction, and magnetic reconnection. These processes induce significant changes in the physical parameters, including electron number density, temperature, and plasma velocity. The evolution of these parameters is closely linked to the onset and development of the flare. For instance, rapid changes in plasma velocity fields can trigger magnetic reconnection, which can then initiate solar flares. Accurately measuring and inverting these physical parameters is therefore essential for understanding the mechanisms behind solar flare eruptions and improving flare prediction accuracy. However, directly measuring these parameters is technically challenging. Instead, their variations are often reflected in the observed spectral features. By analyzing strong spectral lines in the visible, near-infrared, and ultraviolet spectra, we can infer the physical conditions. Spectral lines such as H, Ca II and Mg II reveal that the solar chromosphere, the atmospheric layer between the photosphere and the corona, is highly dynamic, finely structured, and complex [ 5 ] . Complex inversion methods based on physical models often rely on intricate physical assumptions and approximations, resulting in high computational costs and limitations in practical application [ 18 , 9 ] . In this context, machine learning methods offer a promising alternative, enabling rapid and efficient inversion based on large-scale statistical data, independent of physical models. The application of deep learning techniques in solar flare research is emerging as a powerful approach, driven by advances in computational capabilities [ 19 , 3 , 12 , 4 ] . 18 trained an invertible neural network (INN) based on the results of simulations with the one-dimensional radiative hydrodynamics code RADYN. The INN analyses the input spectral H \alpha and Ca II 8542 lines and outputs physical parameters, such as the temperature and electron number density of the solar atmosphere. In validating the analysis of a class M1.1 solar flare, the expected results were obtained. 9 used the results of a large number of 3D solar atmosphere simulations as a training set for a convolutional neural network (CNN) called SunnyNet. SunnyNet computes 3D nonlocal thermodynamic equilibrium radiative transfer in optically thick stellar atmospheres and can predict atomic populations in nonlocal thermodynamic equilibrium with reasonable accuracy. It can also predict the temperature and electron number densities in solar flares with reasonable accuracy. SunnyNet can make reasonable predictions for atomic populations in non-local thermodynamic equilibrium and is nearly 100,000 times faster than conventional codes. 13 inverted H \alpha and Ca II spectra using deep neural network (DNN) and avoided the gradient vanishing problem of DNNs by using a jump-joining method. The final inversion results were more accurate, and it took only 0.3-0.4 ms for each spectral line, which is 250 times faster than the traditional method. The final inversion results are more accurate, taking only 0.3-0.4 ms for each spectral line, which is 250 times faster than the traditional analysis. The Chinese H \alpha Solar Explorer (CHASE) is China’s first solar exploration satellite. It is equipped with a solar space telescope that can make continuous 24-hour observations in a sun-synchronous orbit at an altitude of 517 kilometers [ 14 , 15 , 16 ] . The successful launch of CHASE has provided a wealth of high-resolution H \alpha spectral data, enabling more detailed studies of solar flares. The H \alpha line during flares provides crucial insights into the underlying physical processes. In this study, we apply deep learning techniques innovatively to solar flare research. We aim to construct an appropriate fully connected neural network model that can swiftly and accurately invert solar chromospheric parameters, including electron number density, temperature, and plasma velocity, using H \alpha spectral data. We further validate the practical application of our model by analyzing real H \alpha spectral data obtained by the CHASE satellite during solar flares. Our work is expected to lay a solid scientific foundation for studying the physical mechanisms of solar flares and improving the accuracy of flare eruption predictions. Figure 1 shows the schematic of the workflow. The RADYN data source and data preprocessing are described in Section 2 , while the model selection and prediction are described in Section 3 . Section 4 presents the model application on CHASE data. Finally, discussion and conclusion are given in Section 5 . Figure 1: Schematic diagram of the workflow. 2 Dataset 2.1 RADYN Flare Dataset RADYN is a radiative hydrodynamics code originally developed by 7 , 8 to simulate acoustic shocks in the chromosphere. It has since been adapted to simulate flare responses in the chromosphere [ 1 , 2 ] . RADYN uses an adaptive grid method to solve the radiative hydrodynamics equations for one-dimensional plane-parallel atmospheres. It optimizes data resolution by taking denser samples in regions with larger parameter gradients. This approach ensures that maximum information is retained within the same storage capacity, making it particularly effective for modeling the complex dynamics of solar flares. The dataset used in our study was from the RADYN F-CHROMA database 1 1 1 https://star.pst.qub.ac.uk/wiki/public/solarmodels/start.html [ 6 ] . The dataset comprises 79 flare models, characterized by different spectral indices, cutoff energies, and total energy of the beam electrons that are injected at the flare loop top as an