GSST/CBERS-5 International Symposium on Machine Learning in Heliophysics and Space Weather

Aug 17, 2026, 9:00 AM → Aug 21, 2026, 5:30 PM America/Sao_Paulo

Fernando de Mendonça - LIT (National Institute for Space Research, São José dos Campos, SP, Brazil)

Luis Eduardo Antunes Vieira (INPE), Jose Marchezi (National Institute for Space Research - INPE), Franciele Carlesso (INPE)

The GSST/CBERS-5 International Symposium on Machine Learning in Heliophysics and Space Weather will bring together researchers, engineers, and data scientists working at the interface of artificial intelligence and solar–terrestrial physics. 

This focused symposium aims to explore how Machine Learning (ML) and Computer Vision (CV) techniques can enhance our understanding of heliophysical processes and improve predictive capabilities in space weather and space climate. Discussions will emphasize scientific and technological opportunities within the frameworks of the Galileo Solar Space Telescope (GSST) and the China–Brazil Earth Resources Satellite-5 (CBERS-5) missions, fostering collaborative efforts across Brazilian and international scientific communities.

While the primary emphasis is on heliophysics and space weather, contributions addressing applications of Machine Learning in Astrophysics and Earth Sciences are also strongly encouraged. Particular attention will be given to enabling technologies that underpin next-generation observational capabilities, including advanced instrumentation, high-performance computing, data assimilation frameworks, photonics and optical systems, remote sensing platforms, and autonomous space-based systems. Contributions involving large-scale data analysis, modeling, predictive methodologies, and sensor development are especially welcome.

The event is organized by the National Institute for Space Research (INPE), the University of Vale do Paraíba (UNIVAP), and Mackenzie Presbyterian University, with the support of the Brazilian Society of Space Geophysics and Aeronomy (SBGEA).

The organization of this symposium builds directly on discussions held during the 11th Annual BRICS Astronomy Working Group (BAWG) Workshop*, where ongoing and planned solar and space-weather observation missions were examined, including the Galileo Solar Space Telescope (GSST, Brazil), Aditya-L1 (India), the Chinese Solar Polar Orbit Observatory (China), and CBERS-5 (China/Brazil). These discussions underscored the importance of coordinated international efforts and advanced data-driven approaches for the next generation of heliophysics and space-weather research. 

*The 11th Annual BRICS Astronomy Working Group (BAWG) Workshop was financed, in part, by the São Paulo Research Foundation (FAPESP), Brasil. Process Number #2025/17984-0.

Topics to be Covered

• Computer Vision and Machine Learning applications in heliophysics

• Solar magnetism and magnetic-field inference

• Solar activity: flares, CMEs, and energetic particles

• Solar wind dynamics and heliospheric structures

• Space weather monitoring and forecast

• Weather, Climate, and Atmospheric Sciences

• Physics-Informed Neural Networks

• Research-to-Operations (R2O) Processes Augmented by AI and Computer Vision

 

Format and Participation

The symposium will feature invited talks, contributed oral presentations, and discussion sessions. Participation from both academia and operational space-weather centers is encouraged to bridge research and applications.

Important Dates

• February 11 2026: Submission of abstract for oral and poster sessions opens/ Call for forecast application opens/ Submission of applications for financial support opens

• March 30 2026 June 17 2026: Deadline for abstract submission / Call for space weather forecast application closes

• April 30 2026 June 27 2026: Notification of abstract acceptance

• March 23 2026 July 05 2026: Deadline for financial support application

• May 29 2026 July 12 2026: Deadline for early registration (participants awarded  financial support)

• August 17–21 2026: Conference     

• December 31 2026: Paper submission deadline

 

Registration & Abstracts

• Registration is open at  https://indico.global/e/gsstAI

• Conference fees: US$ 250

• Submission of applications for financial support opens (The organization will provide limited financial support to cover the conference fees). 

 

Organizing Committee

        LOC

• Luis Eduardo Vieira (Chair;INPE)
• José Marchezi (Chair; INPE)
• Franciele Carlesso (Chair; INPE)
• Alisson Dal Lago (INPE)
• Daniele Medeiros (INPE)
• Hadassa Raquel Peixoto Jácome (INPE)
• Adriany Barbosa (INPE)
• Ivana Yoshie Sumida (INPE)
• Paulo Jauer (INPE)
• Marcos Vinicius Silveira (INPE)
• Letícia Rocha Silva Ferreira (INPE)

 

        SOC

• Luis Eduardo Vieira (Chair;INPE)
• José Marchezi (Chair; INPE)
• Alisson Dal Lago (INPE)
• Joaquim Costa (INPE)
• Ligia Alves da Silva (INPE)
• Laysa Resende (INPE)
• Livia Ribeiro Alves (INPE)
• Jean Carlo Santos (INPE)
• Marlos Rockenbach (INPE)
• Paulo Simões (Mackenzie)
• N.L. Vijaykumar (INPE)
• Alan Prestes (UNIVAP)
• Fernando Luis Guarnieri (INPE)
• Rafael Santos (INPE)

 

Venue and Dates

The event will take place at INPE – National Institute for Space Research, São José dos Campos, Brazil, during the second week of August 2026 (August 17–21, 2026).

Support

Interested in supporting? Contact the organization at solarAI2026@inpe.br 

 

Templates SolarAI 2026 Poster Template.pptx

Mon, August 17

Opening Session: Opening Talks

Invited Talks

1 Selected advances achieved by space- and balloon-borne missions

The Sun is observed from the ground, from the stratosphere and from space. Although there are some observations that can only be done from space, e.g. in wavelengths such as the extreme ultraviolet that are absorbed in the Earth's atmosphere, and others that are currently possible only from the ground, e.g. the highest resolution observations with 4m diameter telescopes. However, there are a range of observations that are of great importance for our understanding of solar activity and space weather that can be carried out from each of the three environments. Perhaps the most fundamental of these are measurements of the Sun’s magnetic field. This talk will highlight advantages of observing the Sun from space and from the stratosphere, and provide some example results. It will also point out some of the challenges to observing the Sun from these environments.

Speaker: Sami Solanki (MPAE)

Oral Contributions

2 Global Coronal Response to an Extreme Coronal Mass Ejection: Compressive Oscillations in the Outer Solar Corona

Coronal mass ejections are well-established drivers of large-scale wave phenomena in the low corona, yet the wave dynamics operating in the extended corona and inner heliosphere have remained almost entirely unexplored. We report here the first combined observational and numerical evidence of coherent, global compressive oscillations propagating through the outer corona and inner heliosphere. Our analysis is supported by both observational data from the SOHO/LASCO C3 coronagraph and a state-of-the-art magnetohydrodynamic (MHD) simulation that reproduces the large-scale coronal and heliospheric conditions of the event. To extract wave signatures from the complex coronal brightness data containing multiple concurrent phenomena, we employ Spectral Proper Orthogonal Decomposition (SPOD). This method can effectively decompose the perturbations in the original signal into modes that capture a physically distinct, energetically ranked oscillation at a single temporal frequency. Applying this framework to the extreme coronal mass ejection event of 2012 July 23, we isolate two physically distinct wave signatures. The first is a fast-mode shock-like compressive wave propagating preferentially along the CME propagation axis, with speeds exceeding 1000 km s⁻¹, that dissipates within approximately 2.5 hours, consistent with strong local damping or rapid energy transfer to the surrounding plasma. The second, more remarkable, is a large-scale global circular wavefront that persists for approximately 7 hours across the full LASCO C3 field of view. This behavior is fully consistent with fast-mode MHD wave propagation in a low-beta coronal plasma, where magnetic pressure dominates, and compressive disturbances travel coherently over vast distances with minimal dispersion. This second mode constitutes the first observational detection of a global-scale oscillation of this kind in the outer corona. Both wave signatures are independently recovered from SPOD applied to synthetic white-light data and directly from the MHD variables, confirming the physical interpretation. The compressive front spans approximately 60°-90° around the CME nose, while the global mode extends across nearly the full 360° angular extent of the C3 field of view. The analysis of synthetic white-light data further suggests that such wave signatures should be detectable with future missions such as PUNCH. These findings reveal a previously unrecognized component of CME-driven wave activity operating on spatial and temporal scales far beyond what has been characterized to date, providing new physical constraints on energy transport and plasma structuring in the extended corona and inner heliosphere.

Speaker: Suzana Silva (University of Sheffield)

10:40 AM Coffee Break

Oral Contributions

3 Studying the Response of the Lower Solar Atmosphere to X-class Flare: 30 THz Thermal Diagnostics and Multi-Instrument Data Integration

We investigate the post-flare amplification of chromospheric 3-minute oscillations within a sunspot umbra following the SOL2024-08-08 X1.3-class flare. Utilizing high-resolution data from the DKIST Visible Broadband Imager at 450 nm and the AR30THz telescope focused on AR13777, we demonstrate the global dominance of 5-minute oscillations and the strict localization of 3-minute modes within the deep umbral photosphere. By spatially cross-correlating the deep photospheric wave field with the mid-infrared chromosphere, we measure a near-zero representative phase lag of 3.2 +/- 1.88s, consistent with the sub-cutoff evanescent nature of the waves, whereby the entire squeezed umbral column oscillates in unison as a phase-locked coherent unit. We then reveal a temporal disconnect (~54 minutes) between the flare’s impulsive phase and the sustained 30 THz wave amplification. We conclude that two sequential processes were required: the impulsive flare provided the initial mechanical perturbation that excited the natural oscillation frequencies of the umbral column, while delayed, gradual thermodynamic density enhancement subsequently altered the local H- and free-free opacity. This opacity tuning physically shifted the 30 THz formation height upward by ~200–250 km into the oscillating umbral chromosphere, allowing the wave power to permanently cross the 95% significance threshold. These results establish the mid-infrared continuum as an essential, scatter-free diagnostic for tracking how mechanical wave energy is processed and dissipated within the dynamic lower solar atmosphere.

Speaker: soumya shree sadangaya

4 Physically Constrained Multi Task Reconstruction of the Martian Nightside Thermosphere and its Sensitivity to Solar and Heliospheric Drivers

Modeling the Martian nightside thermosphere is challenging due to the absence of direct solar illumination and highly irregular in-situ sampling. We present a multi task physics informed neural network (MT-PINN) trained on more than a decade of MAVEN/NGIMS observations (MY 32-38) to reconstruct the densities of O, CO₂, N₂, and Ar. To represent coupling with space weather drivers, the model incorporates solar wind parameters with hour time lags, implicitly accounting for possible delays in dayside-to-nightside energy transfer.

Unlike purely data driven regressors that often produce non physical density inversions in sparsely sampled regions, the architecture introduces a weak monotonicity prior via automatic differentiation. This constraint penalizes positive vertical gradients in logarithmic density, encouraging behavior consistent with hydrostatic expectations while preserving the model’s ability to capture small scale variability.

Validation using a strict orbit disjoint split shows that the physics informed regularization substantially reduces inversion rates (up to 90% for certain species) while maintaining high predictive accuracy (RMSE ≈ 0.22). The resulting differentiable surrogate enables rapid profile reconstruction along arbitrary trajectories and provides an efficient tool for exploring the sensitivity of the Martian upper atmosphere to solar and heliospheric forcing.

Speaker: Sergey Nikiforov (NYUAD)

5 Multi-Agent Reinforcement Learning for Adaptive Decentralized Space Weather Sensing

Satellite thermal management systems are critical for maintaining the operational integrity of spacecraft, yet they are typically designed as reactive systems. This study presents a novel methodology for predictive thermal health monitoring using Physics-Informed Neural Networks (PINNs) to safeguard satellites against extreme space weather events. The primary objective is to differentiate between nominal thermal drift and anomalies induced by high-energy particle flux and geomagnetic disturbances.

The methodology integrates real-time satellite telemetry—specifically temperature data sourced from on-board Phase Change Material (PCM) thermal control units—with external space weather indicators, including solar proton flux and planetary geomagnetic indices. By training a PINN model on both physical thermal equations and historical space weather datasets, the system learns to predict the satellite's internal thermal state under varying environmental stressors. This approach allows for the identification of thermal excursions before they exceed safety thresholds.

Main results demonstrate that the PINN framework achieves higher predictive accuracy than standard statistical models by adhering to the governing thermodynamic constraints of the spacecraft. This research significantly improves mission assurance, offering a proactive tool for autonomous satellite operators to mitigate the risks of hardware overheating during solar events. The significance of this work lies in its interdisciplinary application, demonstrating how integrating material-level thermal expertise with machine learning can enhance the longevity and autonomy of small-satellite constellations in increasingly dynamic space environments.

Speaker: Allen John Darin J (Mahendra Engineering College)

6 Analysis of astrophysical shockwaves using Physics informed neural networks

Dusty plasma comprises charged dust particles, ions, electrons, etc. They are prevalent outside of Earth as well. In this work, we aim to study the plasma environment outside of Jupiter. Specifically, the shock waves that form when solar wind interacts with Jupiter’s magnetosphere. We also draw comparisons from shock waves in Earth’s and Saturn’s magnetospheres. The methodology comprises Lie groups and self-similarity analysis. The research problem is defined using a collisionless multi-fluid plasma system, which is represented by partial differential equations (PDEs) and converted into ordinary differential equations (ODEs) using self-similarity transformations and physics-informed neural networks. The results highlight the dynamics of the flow variables, which help us understand the behavior of shocks in dusty plasma around Jupiter. The research concludes that due to the dominance of the thermal effect, the extraterrestrial shock follows an exponential path and decays with no steep fronts.

Speaker: Akshita Bhardwaj (Indian institute of technology Roorkee)

12:30 PM Lunch

Invited Talks

7 From Pixels to Physics: Reconstructing the Solar Atmosphere with Physics-Informed AI

Artificial Intelligence (AI) is rapidly transforming how we analyze and interpret solar observations, enabling new approaches to long-standing challenges in heliophysics. In particular, recent advances in machine learning provide a pathway from data-driven image analysis toward physically consistent models of the solar atmosphere. Rather than treating observations as isolated measurements, these methods enable the reconstruction of continuous, multi-dimensional representations constrained by both observations and underlying physical principles.
In this presentation, I will discuss how Physics-Informed AI can bridge the gap between observations and models of the solar atmosphere. I will highlight recent developments emerging from ongoing, collective efforts across the heliophysics and space weather communities to leverage machine learning for 3D reconstruction, inversion, magnetic field extrapolation, and multi-instrument data integration. Together, these approaches demonstrate how sparse and heterogeneous observations can be transformed into physically meaningful representations of the solar atmosphere. These methods illustrate a broader paradigm shift—from pixel-level analysis to learning physically meaningful representations of the Sun. Finally, I will outline key open challenges and opportunities for physics-informed AI to enable the next generation of data-driven, physically consistent models of the solar atmosphere.

Speaker: Dr Robert Jarolim (High Altitude Observatory, NSF NCAR, USA.)

Oral Contributions

8 High-Cadence Solar Flare Dynamics via Deep Learning: Soft X-ray Spectral Analysis for for Aditya-L1/SoLEXS

Solar Low-Energy X-ray Spectrometer (SoLEXS) is a Sun-as-a-star payload onboard the Aditya-L1 mission designed to monitor solar coronal emissions and flare energetics. It has been in continuous operation at the L1 Lagrangian point for almost two years, capturing solar soft X-ray (SXR) spectra at a 1-second cadence. The stability of the instrument and its observing conditions has produced a large, homogenous, and high-cadence dataset ideal for machine learning analysis. However, analyzing this 1-second data presents a significant computational challenge. The study of rapid plasma heating during solar flares, a key science goal, is fundamentally limited by the trade-off between temporal resolution and signal-to-noise ratio (SNR). While key dynamic processes occur on shorter timescales, the low SNR of these short-integration spectra makes traditional forward-modeling techniques infeasible, as they require long integration times (e.g., 30-180s) for reliable fitting. This temporal averaging obscures the fast dynamics targeted for investigation.

A novel analysis pipeline is presented, designed to overcome this limitation and extract plasma parameters (e.g., temperature, elemental abundances) at the full 1-second cadence from SoLEXS spectra. The method employs a hybrid ‘Teacher-Student’ deep learning approach. First, a ‘Teacher’ model, consisting of a fully connected neural network, is trained on reliable, high-SNR time-integrated spectra, learning to rapidly emulate the results of traditional fitting. A second, recurrent model (LSTM) is then trained to learn the temporal dynamics, using sequences of noisy 1-second spectra to predict the parameters of the integrated interval. This allows the pipeline to leverage both the reliability of high-SNR data and the temporal information in the 1-second-cadence sequences, effectively ‘de-noising’ the data by learning how plasma parameters evolve smoothly in time.

This pipeline produces a 1-second-cadence time series of plasma parameters, capturing rapid variations previously lost in time-averaged analysis. This high-cadence output reveals new phenomena. Notably, less-intense flare events are observed that are typically missed by traditional, long-integration analysis. Furthermore, the 1-second temperature model output reveals evidence of quasi-periodic pulsations (QPPs) in the thermal emission. This technique opens a new observational window, enabling the quantitative study of fast heating and cooling processes in the flare plasma at the instrument's native cadence.

Speaker: Abhilash Sarwade (U. R. Rao Satellite Centre, ISRO)

9 CMEAT: A Multimodal Dataset for Predicting CME Earth Impact and Transit Time

Geoeffective coronal mass ejections (CMEs) can disrupt satellites, power grids, and navigation systems, making accurate early warning critical for space weather operations. We present CMEAT, a curated multimodal dataset and fusion framework for predicting CME Earth impact and Sun–Earth transit time. CMEAT pairs CDAW LASCO observations (1996-2025) with ICME arrival labels and upstream L1 context, producing several thousand events, including hundreds of verified impacts. We extract a compact physical feature set and texture/geometric descriptors from LASCO C2/C3 running-difference imagery, and evaluate classical ML baselines alongside fusion strategies. On temporally held-out test sets, our fusion approaches achieve the best F1 score of 88.4% for impact classification and a mean absolute error of 19.04 hours for transit-time prediction, outperforming single-modality baselines.

Speaker: Besma Guesmi (Ubotica Technologies)

10 The new Solar Ultraviolet to Near-infrared Spectrometer (SUNS): searching for white-light flares

Since the first observation of a solar flare in 1859 by Carrington and Hodgson, explaining the origin of the excess visible continuum emission (white-light flares, WLFs) remains a challenge in the understanding of these events. Identifying the radiation mechanism involved is crucial for understanding the transport and deposition of energy in the solar atmosphere. However, spectral data for solar WLFs are relatively rare and insufficient to dispel the ambiguity of their origin: photospheric blackbody radiation or Paschen continuum from hydrogen recombination in the chromosphere. Due to the lack of solar observations with spectral resolution covering the entire visible range, we present a new telescope to meet this demand: the Solar UV-NIR Spectrometer (SUNS), developed and integrated at the Center for Radio Astronomy and Astrophysics Mackenzie (CRAAM), in partnership with the Steiner Institute, with funding from FAPESP and MackPesquisa. We will present its optical, mechanical, and spectral characteristics, observation conditions, future steps, and expected results. SUNS will provide resolved spectra in the visible range, which will allow us to test current visible continuum emission models and move towards a solution to the more than 165-year-old mystery of the origin of WLFs.

Speaker: Paulo Simões (Universidade Presbiteriana Mackenzie)

3:40 PM Coffee Break

Invited Talks

11 Machine learning techniques for remote sensing of the Sun and inner heliosphere

Remote sensing of the Sun and the inner heliosphere remains the primary observational approach for investigating the physical mechanisms underlying the Sun’s short-term variability and its associated space weather phenomena. Given the significant societal and economic implications of space weather, a growing number of current and planned space- and ground-based observatories are dedicated to monitoring the solar atmosphere and the inner heliosphere. Over the past decade, solar physics has increasingly embraced state-of-the-art machine learning (ML) methodologies to address the challenges posed by the increasing volume, complexity, and dimensionality of scientific and operational solar data. In this talk, I present a curated compilation —assisted by artificial intelligence —of established and emerging ML applications, highlighting key trends across four major application domains:

• Data calibration (e.g., image deconvolution, super-resolution, denoising, and spectropolarimetric calibration).
• Feature classification and detection (e.g., of coronal holes, active regions, coronal mass ejections, and flare ribbons).
• Measurement and reconstruction (e.g., Stokes inversions, plasma velocity inference, coronal magnetic field extrapolations, coronal electron density mapping, and three-dimensional reconstruction of CME morphology).
• Forecasting (e.g., solar flares, solar energetic particle events, CME arrival times, sunspot number, and solar irradiance variability).

Speaker: Dr Francisco Iglesias (Grupo de Estudios en Heliofísica de Mendoza (GEHMe), Universidad de Mendoza, Argentina)

Oral Contributions: Heliophysics & Space Weather

12 Forecasting Ionospheric Irregularities Using Interpretable Generalized Linear Models

The ionosphere poses challenges for accurate forecasting due to its complexity and variability. Irregularities in the lower ionosphere are influenced by local time, season, geographic location, solar activity and space weather, complicating precise predictions. However, understanding this region is crucial for radio communication, navigation and Global Navigation Satellite System (GNSS) accuracy. This study presents the application of Generalized Linear Models (GLMs) to forecast ionospheric conditions. The method is illustrated using 11-year solar cycle data from a GNSS station located in Brasília (geographic coordinates - 15°S and 47°W, 1106.02 m for altitude). Designed for simplicity and interpretability, these models use at most four independent variables to predict the ionosphere behaviour, all of them related to local time, seasonal variability, solar cycle and magnetosphere state. These selected variables were: Time Left to Sunrise (TLS), Maximum Elevation Angle of the Sun (EAS), F$_{10.7}$ and K$_p$, and combinations among themselves, during the 24th solar cycle. The third quartile of Rate of TEC index (ROTI) was used to classify the state of the ionosphere, defined as a binary response variable in the model: values below 0.5 indicated regular condition, while high values indicated irregularity. The models were trained (with 70% of random data) and tested (with 30% of random data) using ROTI data from GNSS receivers in Brasília (2010–2022). After training and calibration, the optimal model, featuring a probit link function, achieved near-perfect classification of the ionosphere. A logit link function also yielded good classification scores, having the advantage of providing easier interpretation of the model optimal parameter. In conclusion, the GLM model represents a practical and promising alternative for predicting ionospheric behavior, also providing valuable insights for space weather applications.

Speaker: Marco Antonio Ridenti (Instituto Tecnológico de Aeronáutica)

13 Detection of Rapid Coronal Hole Boundary Reconfigurations Using Correlation Dimension Mapping

This study investigates how magnetic reconnection reshapes coronal-hole (CH) boundaries during eruptive events. Using high-cadence EUV and magnetogram observations from the Solar Dynamics Observatory on 2015 June 4, we apply Correlation Dimension Mapping (CDM), a technique designed to quantify the geometric complexity of CH boundaries across multiple spatial scales.
We find that localized jet eruptions are systematically accompanied by rapid increases in the correlation dimension, indicating enhanced boundary irregularity during magnetic reconfiguration. Two of the three detected events correspond to coronal jets occurring within a multipolar magnetic configuration characterized by a polarity inversion line and a fan–spine topology favorable for reconnection.
In both cases, the peak in boundary complexity occurs shortly after the jet’s intensity maximum, suggesting a delayed response of the large-scale magnetic structure to localized reconnection.
These results demonstrate that CDM provides a sensitive diagnostic of dynamic restructuring at coronal-hole boundaries. The method offers a new way to track the geometric response of open–closed magnetic interfaces during eruptive activity and may help identify regions prone to reconnection-driven events in the solar corona.

Speaker: Dr Eduardo Flandez (Universidad de Chile)

Tue, August 18

Invited Talks

14 Solar magnetic field observation for HSOS: from SMFT to SPO

Solar magnetic field observations are mainly divided into two categories: high-resolution local magnetic field measurements and full-disk magnetic field observations. Over the past four decades, Huairou Solar Observing Station (HSOS) has conducted systematic research on the theory and technology of solar magnetic field observations based on these two categories. For high-resolution local magnetic field observations, the 35-cm Solar Magnetic Field Telescope (SFMT) has accumulated one of the longest homogeneous datasets of solar vector magnetic fields in the world, enabling key discoveries such as flare precursor motions and magnetic supergranule properties. As the world’s first mid-infrared solar magnetic field telescope, AIMS has revolutionized solar magnetic measurements by directly detecting Zeeman splitting at Mg I 12.3 μm and achieving a precision better than 10 Gauss. For full-disk magnetic field observations, the full-disk vector magnetograph SFMM has provided high-cadence full-disk solar vector magnetic field data with unprecedented spatial and temporal coverage, significantly improving the accuracy and reliability of solar activity forecasting and space weather research. The Solar Polar-orbit Observatory (SPO) has delivered unprecedented high-resolution observations of the solar polar regions, enabling breakthrough studies of polar magnetic fields, solar wind origin, and greatly advancing the global understanding of the solar dynamo and space weather drivers. This report presents the detailed characteristics and scientific contributions of SFMT, AIMS, SFMM, and SPO.

Speaker: Junfeng Hou

Oral Contributions

15 1-m off-axis mid -infrared solar magnetic field telescope (AIMS)

Magnetic field is the most important observational quantity in contemporary solar physics, as nearly all solar activities are closely associated with the solar magnetic field and its evolution. Currently, measurements of the solar magnetic field are primarily based on the Zeeman effect, where longer wavelengths offer higher detection precision and sensitivity. Therefore, conducting solar magnetic field observations in the mid-infrared band is the most effective approach to improving the accuracy of solar magnetic field measurements. The 1-m off-axis mid -infrared solar magnetic field telescope (AIMS) is the first solar telescope dedicated to mid-infrared solar magnetic field measurement in the world. The first-generation scientific instruments of AIMS are a 12.32 μm Fourier Transform Spectrometer and an 8–10 μm imager. After a decade of development, AIMS began scientific observations in October 2025. This report highlights the latest development progress and observational results of AIMS: AIMS has observed triply-split spectral lines in solar active regions and spectral evolution during solar flares at the 12.32 μm, and has captured multiple instances of radiation enhancement associated with solar flares using its 8–10 μm imager.

Speaker: Yuliang Shen (naoc)

16 Stacking Ensemble Learning with Geomagnetic and Meteorological Features for Atmospheric Electric Field Forecasting

Atmospheric electric field forecasting represents a significant challenge in space weather monitoring applications due to the complex interaction among geomagnetic disturbances, atmospheric dynamics, and local meteorological variability. This study proposes a hybrid machine learning model based on Stacking Ensemble Learning for atmospheric electric field forecasting using geomagnetic and meteorological variables acquired through high-temporal-resolution environmental sensors.
The research integrates an electric field mill, a magnetometer, and a meteorological station for environmental data acquisition. Independent variables were selected according to their atmospheric-physical relevance, geomagnetic interactions, statistical dependence, and predictive contribution to atmospheric electric field variability. Meteorological variables considered include temperature, humidity, dew point, atmospheric pressure, wind speed, precipitation, solar radiation, and ultraviolet index, owing to their influence on atmospheric conductivity, ionization processes, cloud electrification, and charge transport mechanisms. In addition, geomagnetic field measurements were incorporated because of their relationship with ionospheric disturbances and space weather variability. The final feature set was further supported through correlation analysis, multicollinearity assessment, and feature importance evaluation within the machine learning framework.
Prior to predictive model development, an Exploratory Data Analysis (EDA) will be conducted to identify statistical distributions, temporal patterns, outliers, missing values, and nonlinear relationships among geomagnetic and meteorological variables associated with atmospheric electric field variability. The study is conceptually grounded in Faraday’s law of electromagnetic induction, which establishes the relationship between temporal variations in magnetic fields and the generation of induced electric fields.
The proposed methodology employs a stacking ensemble architecture composed of both regularized linear regression and nonlinear learning models, including Ridge Regression, Elastic Net, Random Forest, XGBoost, and LightGBM. These models are integrated through a meta-model designed to optimize predictive performance and reduce generalization error. Furthermore, temporal feature engineering techniques, including lag variables and rolling statistics, are incorporated to capture dynamic patterns associated with atmospheric and geomagnetic disturbances.
Predictive performance will be evaluated using Time Series Split cross-validation and regression metrics such as the coefficient of determination (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Additionally, model interpretability will be assessed using feature importance analysis and SHAP values to identify the relative influence of geomagnetic and meteorological variables on atmospheric electric field behavior.
The expected result is to demonstrate that the integration of geomagnetic and meteorological variables through co-learning models significantly improves predictive accuracy compared to independent traditional approaches, contributing to the development of intelligent atmospheric monitoring and forecasting systems for space weather applications.

Speakers: Dr Juan Jesús Soria-Quijaite (Escuela Profesional de Ingeniería Ambiental, Universidad Peruana Unión, Lima, 150118, Perú), Dr Manuel A. Bravo (Centro de Instrumentación Científica, Universidad Adventista de Chile, Chillán, 3780000, Chile), Orlando Poma Porras (Escuela Profesional de Ingeniería Ambiental, Universidad Peruana Unión, Lima, 150118, Perú)

17 JW-ASTClaw: A Generalizable Multi-Agent Framework for Autonomous Solar Telescope Observation

Solar observation faces complex challenges that conventional automated observation systems struggle to address, including rapidly changing weather conditions, potentially anomalous data, and the need for prompt follow-up observations of eruptive phenomena. The rapid advancement of artificial intelligence technologies offers new possibilities for tackling these challenges. This paper presents the JW-ASTClaw framework, a multi-agent system built upon Large Language Models (LLMs) and the Model Context Protocol (MCP). The system comprises three perception agents (Seeing Analyzer, Cloud Analyzer, and Sunspot & Flare Analyzer) and one execution MCP (Observation Control Middleware), all coordinated by a central decision-making agent (Reasoning Engine). The framework has been successfully deployed on Solar Full-disk Multi-layer Magnetograph (SFMM), a key instrument of Phase II of China's Meridian Project, and its effectiveness has been validated through recent observations. The design features high scalability and generality, making it applicable to other telescopes and large-scale scientific instruments.

Speaker: Dr Liyue Tong (National Astronomical Observatory of the Chinese Academy of Sciences)

10:40 AM Coffee Break

Oral Contributions

18 Identifying Satellite Crossings of the Earth's Magnetosheath Using Unsupervised Learning Algorithms

In this study, we applied machine learning techniques to perform an unsupervised clustering of THEMIS satellite orbits to detect magnetosheath crossings. We used the DBSCAN algorithm to analyze crossings within a range of less than 40 Earth radii, focusing on data from the THEMIS-B (THB) and THEMIS-C (THC) spacecraft during 2008 and 2009. These spacecraft were selected due to their eccentric orbits, which facilitate multiple crossings through the magnetosheath. Using electron, ion, and magnetic field data, our algorithm effectively identified several magnetosheath crossings, demonstrating the robustness and applicability of the unsupervised approach. As a final result, we created a consolidated database compiling the magnetosheath crossings identified for the THEMIS mission, which constitutes a valuable resource for the detailed study of magnetospheric dynamics and has the potential to contribute to the development of more accurate models of the magnetosphere in the future.

Speaker: javier silva (Instituto Nacional de Pesquisas Espaciais)

19 Interpretable Machine Learning for Solar Active Region Detection via 2D Circular Kernel Time-Series Transformation and Entropy Features

Solar active region detection from high-cadence EUV imagery is important for data-driven space weather monitoring, yet robust pixel-level characterization remains challenging because active and non-active bright structures can overlap in intensity and morphology. We present an interpretable machine-learning framework for solar active region detection in 193 Å images from the Atmospheric Imaging Assembly onboard NASA’s Solar Dynamics Observatory. The method converts the circular neighborhood around each sampled pixel into a one-dimensional data series using a 2D circular kernel transformation, enabling compact feature extraction from local image structure.
We investigate two feature-selection strategies: (i) statistical and entropy descriptors of the transformed series—median, 95th percentile, distribution entropy, and fuzzy entropy—and (ii) direct use of the transformed series values as input features. A support vector classifier with an RBF kernel is trained to assign each local region to one of three classes: quiet/no active region, non-eruptive bright zone surrounding an active region, and flaring active region. Using repeated stratified 10-fold cross-validation, the proposed framework yields classification accuracies of 0.900 with entropy features and 0.914 with statistical features, while the combined feature set reaches 0.940. Among individual descriptors, fuzzy entropy provides the best performance (A_KF = 0.895), outperforming distribution entropy (0.738), the 95th percentile (0.873), and the median (0.840).
The resulting classifications preserve a realistic spatial distribution of solar activity and support the definition of a generalized solar activity indicator derived from the fraction of flaring-region pixels. The proposed approach is scalable, interpretable, and suitable for automated event detection pipelines in heliophysics and space weather applications.

Speaker: Dr OLUDEHINWA Irewola Aaron (Department of Physics, Federal University of Agriculture, Abeokuta, Nigeria.)

20 Explainable AI for Operational Space Weather Forecasting: An Interpretable Prototype for the EMBRACE-INPE Monitoring Network

Operational space weather monitoring requires not only accurate forecasts but also transparent and interpretable decision support tools. We present an Explainable Artificial Intelligence (XAI) prototype designed for the EMBRACE-INPE environment. The system integrates GOES X-ray flux, SYM-H index, and AE index, capturing the causal propagation from solar activity to geomagnetic and auroral responses. A sequence-based forecasting module produces six-hour predictions with hyperparameters optimized via Optuna. The XAI layer incorporates uncertainty bands, lag-aware propagation markers, and interpretable multivariate relationships. A three-dimensional regression representation exposes nonlinear dependencies between delayed solar drivers and auroral activity. The prototype is implemented as an interactive dashboard supporting human forecasters. Results illustrate how explainable AI can be embedded directly into operational interfaces, bridging machine learning forecasts and human-centered situational awareness.

Speaker: Reinaldo Rosa (National Institute for Space Research (INPE))

21 Machine Learning Approaches for Ionospheric Scintillation Analysis Using ScintPi Observations

Ionospheric scintillation refers to rapid phase and amplitude fluctuations of radio signals as they pass through ionospheric irregularities. While commercial scintillation monitors have been used extensively to study scintillation, their relatively high costs have limited scientific use. To address this, we have been developing low-cost scintillation monitors based on single-board computers and Global Navigation Satellite System (GNSS) receivers. These instruments, referred to as ScintPi, have been deployed at different locations over the globe, producing a very large and valuable dataset which has motivated the development of data-driven processing and analysis techniques.

This work presents a two-stage machine learning (ML) framework applied to scintillation measurements made by ScintPi monitors. More specifically, the framework consists of: (1) a supervised classification to automatically identify scintillation in GNSS data, and (2) a supervised regression to infer scintillation severity from Total Electron Content (TEC) measurements.

In the first stage, the goal is to distinguish true ionospheric scintillation from non-geophysical disturbances such as multipath and radio frequency interference (RFI). Supervised classification models were trained using a labeled dataset with known disturbance sources: multipath, RFI, or low-latitude scintillation. Features were derived from signal power spectra and TEC. Preliminary results show that tree-based models can distinguish scintillation from multipath and RFI with approximately 99% classification accuracy. These results were used to generate a cleaned dataset of approximately 1,100,000 low-latitude, quiet-time scintillation observations during 2024.

In the second stage, the goal is to use the cleaned dataset from the first stage to interpret the Rate of Change of TEC Index (ROTI) derived from lower-rate GNSS observations. While ROTI is expected to be correlated with the amplitude scintillation index S4, the relationship between these two variables depends on many factors including link geometry, irregularity drift, and irregularity spectrum. Determining a relationship between S4 and ROTI is important since it could allow estimates of scintillation severity from the wide network of geodetic receivers rather than specialized scintillation monitors alone. Supervised regression models were trained to infer S4 from ROTI. Features include ROTI from 1 Hz TEC, link geometry, local time, and solar activity. Model performance was evaluated using an independent low-latitude station outside the longitude sector used for training. Preliminary results show strong agreement between estimated and observed S4 (Pearson r ~ 0.92).

Overall, the results demonstrate the usefulness of ML techniques to aid processing of scintillation measurements and to relate scintillation parameters to different types of observations (e.g., ROTI).

Speaker: Isaac Wright (University of Texas at Dallas)

12:30 PM Lunch

Invited Talks

22 Computer Vision with small-scale magnetic elements in the photosphere

Magnetic Bright Points (MBPs) are small-scale magnetic features, ubiquitous across the solar surface. They are short-lived, highly dynamic features which can transfer energy between layers of the solar. Furthermore, as they are small intensity enhancements, they appear brighter than the quiescent environment and can contribute to Total Solar Irradiance. With the advent of new facilities such as the Daniel K. Inouye Solar Telescope, MBPs are sampled at unprecedented resolutions. Here we report on a novel deep learning computer vision techniques developed to isolate and track these features in high resolution data. The results of these techniques are compared to more traditional computer vision methods.

Speaker: Dr Peter Keys (Astrophysics Research Centre, Queen’s University Belfast (QUB), U.K.)

Oral Contributions

23 Multidirectional Muon Detector for Solar-Terrestrial Physics

One important topic within Heliophysics is the modulation of <100GeV cosmic rays due to solar activity. Since 2001, a prototype of a multidirectional muon detector is in operation at the Southern Space Observatory, in Sao Martinho da Serra (SMS), Brazil. It is part of the Global Muon Detector Network (GMDN), composed by detectors in Nagoya (Japan), Kingston (Australia), Sao Martinho da Serra (Brazil) and Kuwait City (Kuwait). The SMS detector went through an upgrade in 2025 and it now operates with a collecting area of 36 m2. Since then, the SMS instrument was able to observe a few Forbush Decreases (FD) resulting from solar Coronal Mass Ejections (CMEs) and their interplanetary counterparts (ICMEs). We will show some preliminary ground cosmic ray directional observations of these recent FDs/CMEs (atmospheric pressure and temperature corrected observations). The SMS detector was a donation from Shinshu University (Japan) to the National Institute for Space Research (Brazil), in 2001 and 2005. The recent upgrade was funded by the Japan Society for the Promotion of Science (JSPS-Japan), the National Institute for Space Research (Brazil), the Brazilian Space Agency, and the Brazilian CAPES funding agency. Additional funding for data analysis is being requested to FAPESP funding agency.

Speaker: Alisson Dal Lago (INPE - National Institute for Space Research)

24 Modeling Neurocognitive Damage from Cosmic and Solar Radiation

Deep space missions beyond Low Earth Orbit expose crews to significant doses of galactic cosmic radiation (GCR), composed of high-energy protons and high-atomic-number, high-energy (HZE) ions including ⁵⁶Fe, ²⁸Si, ⁴⁸Ti, ¹⁶O, and ⁴He. Unlike the protection offered by Earth's magnetosphere, GCR cannot be attenuated by available spacecraft materials, and its cumulative effects on the central nervous system (CNS) represent one of the most critical risks for crewed missions to the Moon and Mars. Data from the RAD instrument aboard the Curiosity rover indicate that a six-month transit to Mars exposes crews to approximately 60% of the recommended career limit. Solar minimum periods further intensify GCR exposure during long-duration missions.
Experimental rodent studies at NASA's Space Radiation Laboratory (NSRL) demonstrate that exposure to mission-relevant doses (≤15 cGy) produces persistent neurocognitive deficits, including impairment of spatial memory, executive function, and attentional shifting. These effects are ion-type- and dose-dependent: ⁵⁶Fe at 1–10 cGy produces impairment in the simple discrimination (SD) and compound discrimination (CD) stages of the Attentional Set-Shifting Test (ATSET), while ⁴He and ²⁸Si produce distinct cognitive profiles in different brain regions, suggesting cerebral circuits are differentially affected by radiation field composition.
This work synthesizes recent advances in applying supervised machine learning (ML), including support vector machines (SVM), Gaussian naïve Bayes (GNB), and artificial neural networks (ANN), for predicting individual-level cognitive impairment in rodents subjected to GCR. Using pre-irradiation behavioral scores as multidimensional input features, classifiers demonstrate better than chance predictive capability, surpassing conventional population level statistical analyses. ML models based on U-Net architectures calculate physical dose distributions of heavy ions with deviations under 2% from Monte Carlo simulations in milliseconds, highlighting ML potential for accurate modeling of HZE effects in physical planning and biological prediction. Neurobiological mechanisms, including microglial activation, reduced hippocampal neurogenesis, loss of dendritic spines, and disruption of synaptic plasticity, establish the biological plausibility of individual susceptibility to GCR. Genetic modifiers such as ATM heterozygosity and ApoE isoforms further modulate this sensitivity, reinforcing the need for personalized pre-mission risk stratification. ML tools trained with ground-based GCR data can screen astronaut candidates and predict cognitive impairment before departure, supporting permissible exposure limits and countermeasures integrated into space weather monitoring.

Speakers: Alexandre da Silva Santos (Federal University of Maranhão), Roger Pinheiro Presoti (Universidade Federal do Maranhão (UFMA))

25 Automatic Characterization and Three-dimensional Reconstruction of Coronal Mass Ejections Using Deep Learning Techniques

The study of space weather critically depends on the three-dimensional (3D) morphological and kinematic characterization of coronal mass ejections (CMEs). This process can be done via generic 3D point position estimation (using e.g., tie-pointing plus triangulation, differential emission measure tomography, polarization ratio or neural radiation fields techniques) or model-based 3D reconstructions. The former is typically limited by scarce spatial coverage and/or limited temporal resolution, critical for extended and dynamic events such as CMEs. Model-based approaches are usually done via a manual fit of a parametric geometric model to the CME outer shell observed in coronographic images, which is time-consuming and prone to subjective biases. To address these limitations, we present a deep learning framework structured into three fundamental stages. First, to mitigate the lack of pixel-level labeled observational data, we develop a methodology for generating large-scale synthetic training datasets; this is achieved by combining real coronal backgrounds with synthetic CME brightness images produced by applying ray-tracing techniques to a density distribution defined by the Graduated Cylindrical Shell (GCS) geometric model. Second, using these synthetic data, a Mask R-CNN convolutional network was trained in a supervised manner for the instance segmentation of the CME outer envelope in differential coronographic images. Its outcome demonstrates a high capacity to discriminate the CME from other radially moving structures without requiring explicit kinematic information. Finally the 3D reconstruction is addressed using a deep convolutional network architecture, designed to automatically infer the six parameters of the GCS model from input coronagraph images acquired from three different perspectives. Altogether, the implementation of these neural networks represents a significant advance toward the automated detection and 3D morphological characterization of CMEs, reducing human intervention and accelerating computational analysis.

Speaker: Mariano Sanchez Toledo (Grupo de Estudios en Heliofísica de Mendoza, Universidad de Mendoza, Mendoza, Argentina)

3:40 PM Coffee Break

Invited Talks

26 Individual and Collective Dynamics of Vortical Structures in the Solar Atmosphere

In this presentation, we discuss recent advances in understanding vortical plasma motions within the solar atmosphere. Using high-resolution magnetoconvection simulations from the Bifrost, MURaM, and R2D2 codes, together with observational data (SST and Sunrise), and state-of-the-art techniques for identifying plasma flows and transport barriers (machine learning, forward and backward finite-time Lyapunov exponents, FTLE), we show that vortex structures play a fundamental role in energy transport across different layers of the solar atmosphere and support wave propagation. Analogously to the application of FTLE in hydrodynamic flows, the ridges of forward and backward FTLE fields can indicate the locations of energy sources in the simulated solar atmosphere. In particular, we focus on three distinct classes of vortices, i.e. kinetic, magnetic, and energy vortices, highlighting their physical properties, formation mechanisms, and contributions to plasma dynamics. We also describe the network of Poynting flux inferred from the FTLE fields at different heights, connecting the photospheric and chromospheric layers. Beyond individual vortical structures, we explore their collective behaviour within (intra-connected) and between (inter-connected) communities of vortical motions, examining how these structures interact across a wide range of spatial and temporal scales. We demonstrate that their coupled dynamics can enhance energy transport to the higher layers of the solar atmosphere and beyond.

Speaker: Dr Viktor Fedun (School of Electrical and Electronic Engineering, The University of Sheffield, UK.)

Oral Contributions

27 Self-Organizing Maps: Identification of Magnetic Reconnection Sites at the Dayside Magnetopause

Magnetic reconnection is a fundamental physical process that occurs in magnetized plasmas and serves as an efficient mechanism for accelerating charged particles by converting magnetic energy into kinetic and thermal energy. In the interaction between the solar wind and Earth's magnetosphere, particularly during periods of southward interplanetary magnetic field orientation (Bz < 0), magnetic reconnection occurs at the dayside magnetopause, enabling the transfer of magnetic flux to the nightside magnetosphere (magnetotail) and directly influencing the development of geomagnetic storms and substorms. This project aims to develop machine learning techniques capable of identifying magnetic reconnection events in near real time, with the goal of integrating these methods into space weather monitoring systems and improving forecasting and response capabilities. The project will be carried out in collaboration with the Galileo Solar Space Telescope (GSST) mission team and the SolarAI initiative.

Speaker: Daniele da Silva Ferreira Medeiros (CBJLSW/INPE)

28 Ionospheric Response to the March 2015 Geomagnetic Storm: GNSS-TEC Analysis over Morocco

We investigated the ionospheric response to the geomagnetic storm of March 17, 2015, using GPS total electron content (TEC) measurements from the IGS station RABT in Rabat, Morocco (34°N, 7°W). This storm, with a minimum SYM-H of -233 nT, was the most severe of solar cycle 24. Vertical TEC was derived from dual-frequency pseudorange measurements using a single-layer mapping function (H=350 km). A physically consistent preprocessing strategy was applied, including satellite-specific and receiver Differential Code Bias (DCB) corrections, precise elevation-angle computation from broadcast ephemerides, and a minimum elevation cutoff of 30° to reduce multipath and mapping errors. Storm-induced TEC deviations (ΔVTEC) were computed relative to the quiet day of March 15.
Results show significant TEC enhancement during the storm main phase, with ΔVTEC reaching 45.78 TECU on March 17, representing a 158% increase above the quiet-day baseline of 28.92 ± 12.91 TECU. Correlation analysis revealed the strongest Kp-ΔVTEC relationship at a 3-hour lag, with r ≈ 0.53 (storm periods), consistent with a delayed ionospheric response to geomagnetic forcing. The recovery phase (March 18) exhibited mixed storm effects, likely associated with thermospheric composition changes and the propagation of travelling atmospheric disturbances.
These findings contribute to understanding mid-latitude ionospheric dynamics during extreme space weather events and have implications for GNSS-based positioning and navigation systems in the European-African sector.
Keywords: Geomagnetic storm, Total Electron Content, GNSS, Ionosphere, Kp

Speaker: Nouhaïla Bouhadi

Wed, August 19

Invited Talks

29 Wave theories and modelling in the lower solar atmosphere

In this talk I will review recent progress and the remaining challenges in both modelling and analysing observed magnetohydrodynamic (MHD) waves in the lower solar atmosphere. To model waves in sunspot umbrae, we have had to go beyond the standard cylindrical flux tube model due to the irregular shapes of the MHD waveguides. For sunspots the umbra/penumbra boundary shape is reasonably static over the time scales of oscillations of interest. This makes solving the eigenvalue problem for MHD wave modes in sunspots tractable because of the assumption of time independent boundary shape conditions. However, the same cannot be said about pores that have highly dynamic boundary shape changes. I will highlight recent progress that has been made in this very challenging area of research.

Speaker: Dr Gary Verth (School Mathematical and Physical Sciences, University of Sheffield, UK.)

Oral Contributions

30 Characterization of Solar Flare Precursors Using Machine Learning

Solar flares represent a major focus in space weather research due to their potential to disrupt satellite operations and critical terrestrial technologies. These phenomena are characterized by rapid variations in X-ray flux, resulting in intense energy releases within the solar atmosphere. To better understand their behavior, this study has utilized the sunpy library to access the X-ray flux time series from the Geostationary Operational Environmental Satellite (GOES) for eleven pre-selected events and to analyze the variations of soft X-ray flux and the temperature and emission measures of each event, aiming to identify precursors of solar flares. Furthermore, an image database is being constructed using four extreme ultraviolet channels (94 Å, 131 Å, 171 Å, and 193 Å) from the Atmospheric Imaging Assembly onboard the Solar Dynamics Observatory (SDO). The objective is to use these data to train machine learning models for the automated recognition and phase characterization of solar flares. Ultimately, the integration of GOES and SDO data aims to contribute to the identification of future events, thereby aiding in the monitoring and forecasting of space weather.

Speaker: Raul Toscano Faria

31 Deep Learning-Based Automatic Detection of M and X Class Solar Flares in the Solar Disk

Solar flares are transient energy release events in the solar atmosphere, typically associated with magnetic reconnection in active regions. While the Geostationary Operational Environmental Satellite X-ray Sensor (GOES/XRS) provides continuous monitoring of flare activity, its lack of spatial resolution limits the identification, localization, and characterization of flaring events. In this work, we present a deep learning-based method for the automatic detection of solar flares in the central region of the solar disk, where projection effects are reduced. The model uses multi-channel observations from the Atmospheric Imaging Assembly (AIA) onboard the Solar Dynamics Observatory (SDO), specifically the 304Å, 1600Å, 171Å and 131Å passbands, which sample different layers and temperature regimes of the solar atmosphere. We trained a convolutional neural network (CNN) using curated flare catalogs aligned with AIA imagery. The model is designed to detect flare events, estimate their location on the solar disk, and produce a structured description of each event. This approach provides a consistent framework for combining multi-wavelength observations with data-driven methods, with potential applications in automated monitoring and near-real-time analysis of solar activity.

Acknowledgements: We acknowledge support from ANID Chile through FONDECYT grant No. 11251905 (VAP).

Speaker: Nicolas Campos (Universidad de Santiago de Chile (USACH))

32 Adaptive Detection of Hot Onset Precursor Events in GOES Soft X-Ray Observation

Accurate and timely detection of solar flares is essential for advancing our understanding of solar activity and improving space weather forecasting capabilities. In this work, we evaluate the performance of an automated flare identification system by comparing its trigger outputs against two independent solar flare catalogs. The analysis highlights how detection performance depends critically on the characteristics of the reference dataset used for validation.

When evaluated against the operational NOAA flare catalog, which is based on fixed intensity thresholds and manual validation procedures, the system achieves a high flare level detection rate exceeding 96%. However, a large fraction of triggers remain unassociated with reported flares, suggesting a significant number of apparent false detections. This discrepancy reflects the inherent limitations of the operational catalog, which may exclude weaker, short duration, or low contrast events due to reporting thresholds and subjective validation.

To address this limitation, we repeat the analysis using a homogeneous archival catalog derived from systematic reprocessing of GOES soft X-ray observations. This catalog employs a consistent and automated flare detection methodology, including background correction and objective onset definitions. Under this framework, more than 93% of triggers are associated with flare intervals, and the fraction of unassociated triggers is significantly reduced. These results indicate that many detections classified as false positives under the operational catalog correspond to physically meaningful small scale activity captured by the archival dataset.

Detection performance is further analyzed as a function of flare magnitude and classification scheme. Detection rates increase with flare strength, approaching near-complete detection for the most energetic events. Differences between peak flux based and background corrected classifications highlight the role of background emission in flare identification and emphasize the importance of consistent event definitions.

Overall, this study demonstrates that the apparent performance of automated flare detection systems is strongly influenced by catalog completeness, threshold criteria, and timing definitions. The results underline the importance of using homogeneous and systematically processed datasets for validation and suggest that automated approaches, including those based on machine learning, can provide a more complete representation of solar activity by capturing events beyond traditional reporting limits.

Speaker: Solomon Perriyil (Centro de Rádio-Astronomia e Astrofísica Mackenzie - CRAAM)

10:40 AM Coffee Break

Oral Contributions

33 Physics-based Gaussian Processes for Solar and Stellar Variability

Context. The solar and stellar magnetic activity can cause spots and faculae on the photosphere that imprints variability signals on its brightness. Many different approaches have been proposed in the literature to reconstruct the signals of magnetic activity on the stellar surface from the brightness measurements, such as Doppler imaging, photometric surface mapping, and planetary transit mapping. Gaussian Process frameworks have been proposed as statistical models for solar and stellar variability, primarily as tools to aid in the mitigation of their contamination on Sun-as-a-star and stellar photometric and spectroscopic observations.
Aims. We aim to model the long-term modulation in brightness of the Sun and magnetic active stars, to derive quantitative physical properties for the surface features such as spots and faculae.
Methods. In this work, we applied a physics-based Gaussian Process framework called FENRIR to model solar and stellar activity as a stochastic process, approximated by a Gaussian distribution. We derive marginalized distributions for physical parameters, like: rotational period, latitude, and lifetime of the spots and faculae, and the solar magnetic cycle.
Results. We applied our method to total solar irradiance data from unresolved solar observations from the SORCE/TIM satellite and to stellar photometric observations from the CoRoT space mission as a proof of concept for “statistical photometric surface mapping”. This approach provides reasonable average values for the properties of solar and stellar starspots and faculae. In particular, by using only the brightness variation of the targets, we derived the rotational period at the average latitude of the spots, and managed to obtain periods in accordance with the literature for the Sun and the star CoRoT-2.

Speaker: Andre O. Kovacs (Center for Radio Astronomy and Astrophysics Mackenzie (CRAAM))

34 Impacts of Space Weather on the Orbital Decay of the Amazonia-1 Satellite During Solar Cycle 25

Low Earth Orbit (LEO) satellites, such as Brazil’s Amazonia-1, are subject to atmospheric drag resulting from
variations in thermospheric density, which intensify during periods of elevated solar activity. This study
investigates how space weather phenomena—specifically geomagnetic storms and solar flares—influence the
orbital decay of Amazonia-1 between 2021 and 2023. Real orbital data (TLEs), Dst geomagnetic indices, and
GOES flare records were analyzed. The dataset was segmented into distinct periods, and monthly average altitudes
were calculated to identify correlations between orbital decay and solar activity. Results indicate that both
geomagnetic storms and X/M-class flares significantly contribute to atmospheric densification and drag,
particularly when such events occur continuously. In some periods, monthly altitude loss exceeded 110 meters.
These findings underscore that during the active phases of Solar Cycle 25, increased drag compromises satellite
orbital stability. This highlights the necessity of integrating space weather considerations into satellite mission
design and planning. The study reinforces the value of continuous monitoring and predictive modeling of
thermospheric behavior to enhance the resilience and longevity of LEO missions.

Speaker: Leonardo Molliet

35 Change Ratios of Magnetic Helicity and Magnetic Free Energy During Major Solar Flares

Magnetic helicity is an important concept in solar physics, with a number of theoretical statements pointing out the important role of magnetic helicity in solar flares and coronal mass ejections (CMEs). Here we construct a sample of 47 solar flares, which contains 18 no-CME-associated confined flares and 29 CME-associated eruptive flares. We calculate the change ratios of magnetic helicity and magnetic free energy before and after these 47 flares. Our calculations show that the change ratios of magnetic helicity and magnetic free energy show distinct different distributions in confined flares and eruptive flares. The median value of the change ratios of magnetic helicity in confined flares is −0.8%, while this number is −14.5% for eruptive flares. For the magnetic free energy, the median value of the change ratios is −4.3% for confined flares, whereas this number is −14.6% for eruptive flares. This statistical result, using observational data, is well consistent with the theoretical understandings that magnetic helicity is approximately conserved in the magnetic reconnection, as shown by confined flares, and the CMEs take away magnetic helicity from the corona, as shown by eruptive flares.

Speaker: Quan Wang (National Astronomical Observatories of the Chinese Academy of Sciences)

36 Implementation of IMF Diffusion Processes for GCR Transport in the Inner Heliosphere using Geant4

The flux of Galactic Cosmic Rays (GCR) reaching the solar atmosphere is a key ingredient for studies of heliospheric modulation and secondary-particle production. The transport of charged particles in the interplanetary medium is commonly described by Parker’s transport equation (Parker, 1965), which provides the theoretical framework for diffusion, convection by the solar wind, gradient-curvature drifts, and adiabatic energy losses under the influence of the Interplanetary Magnetic Field (IMF). In this context, Seckel, Stanev, and Gaisser (1991) developed a seminal model for secondary-particle production by GCR interactions with the solar atmosphere, emphasizing the crucial role of magnetic-field effects in the transport of GCR toward the solar surface. Later, the model developed by Bobik et al. (2012) provided a refined treatment of GCR transport by solving Parker’s equation with rigidity-dependent diffusion coefficients and heliospheric conditions, allowing for a more accurate description of modulation across different energy regimes. In this work, we present an implementation of IMF diffusion processes using the Geant4 Monte Carlo simulation toolkit, aiming to model the transport of GCR with energies in the 1 GeV to 10 TeV range from Earth’s orbit to the solar surface. We use a Geant4 application developed by Tuneu et al. (2021) to implement the transport of charged particles in the presence of magnetic fields, based on the guiding-center approach, and extend it to include the effects of magnetic mirroring and gradient-curvature drifts, solar-wind modulation, stochastic magnetic turbulence through pitch-angle scattering, and adiabatic energy losses. The results of our simulations show that the inclusion of these transport effects suppresses the low-energy GCR flux compared to the purely ballistic case, while the heliospheric transparency increases progressively at higher energies. The resulting energy-dependent GCR flux at the solar atmosphere will be used as input for modeling gamma-ray emission from the quiescent Sun produced by GCR interactions.

Speaker: Raphael Malagoli Thereza (Universidade Presbiteriana Mackenzie - CRAAM)

12:30 PM Lunch

Invited Talks

37 The impact of machine learning, AI, and novel analysis tools to better understand the plasma dynamics of the Sun

Solar atmospheric plasmas host highly dynamical magnetohydrodynamic (MHD) processes, including reconnection, shocks, turbulence, and waves, that couple the photosphere, chromosphere, and corona and ultimately seed variability throughout the heliosphere. This talk connects these drivers to their signatures in the solar wind and at Earth’s ionosphere, emphasizing the need for consistent “Sun-to-geospace” diagnostics. I will highlight how modern collaborative efforts are tackling long-standing obstacles in identifying and quantifying MHD wave modes in structured magnetic features, including the role of radiative transfer, partial ionisation, and multi-height spectropolarimetry. I will then demonstrate how modern machine learning is accelerating plasma inference through use of neural classifiers for Stokes-profile morphologies, transformer-based stratified inversions, and the emerging promise of Physics-Informed Neural Networks (PINNs) to connect inversions directly to governing MHD physics. Remaining challenges include the domain shift between simulations and observations, and the robust coupling of wave energetics to atmospheric heating. Finally, I will outline how next-generation instrumentation, such as integral field spectropolarimeters and next-generation space missions such as the Galileo Solar Space Telescope, will deliver the cadence, sensitivity, and coverage needed to close these gaps.

Speaker: David B. Jess

Oral Contributions

38 Machine Learning-Based Forecasting of Geomagnetic Activity: A Benchmark of Models Applied to the Kp and Dst Indices

This study presents a benchmark of machine learning models for forecasting geomagnetic indices, specifically Kp and Dst, using NASA’s OMNI dataset with hourly resolution over the period from 2015 to 2024. The input variables include solar wind parameters and interplanetary magnetic field components, such as total magnetic field intensity, Bx, By, and Bz components, solar wind speed and density, and plasma temperature. After preprocessing, which involved removing missing values and inconsistent records, the final dataset was structured with 86,737 samples.

To capture the system’s temporal dynamics, lagged variables were constructed for each physical variable, considering windows of up to 24 previous hours at different temporal resolutions (0, 1, 2, 3, 6, 12, and 24h), resulting in 49 additional attributes used as model inputs. The problem was formulated as a supervised regression task with a three-hour forecasting horizon, with models trained separately for each target (Kp and Dst), enabling a direct comparison of performance across different geomagnetic variability regimes.

The experimental evaluation considered three main models: linear regression, random forest, and XGBoost. The data split was performed temporally, with training on the period from 2015 to 2022 (69,361 samples – 80.3%), validation in 2023 (8,544 samples – 9.9%), and testing in 2024 (8,562 samples – 9.9%), ensuring the absence of temporal leakage and simulating a realistic operational forecasting scenario.

The results show consistent performance across models, with progressive improvement as model complexity increases. For the Kp index, R² values ranged from 0.522 for linear regression to 0.556 for XGBoost, while for the Dst index they ranged from 0.572 to 0.639. It was observed that Dst benefited more from nonlinear models, suggesting a more complex relationship between input variables and its dynamics.

Finally, an analysis of feature importance was conducted, showing a predominance of solar wind speed and the Bz component of the interplanetary magnetic field, along with their temporal lags, indicating that both the intensity and persistence of these conditions are key determinants for geomagnetic activity forecasting. The results also suggest that, given the set of variables used, there is a moderate predictability limit, with maximum performance close to an R² of 0.64 for Dst and 0.56 for Kp.

Speaker: Marjori Klinczak (Unifatec)

39 Solarfall: A Multi-Stage Machine Learning Architecture for Solar Flare Forecasting

Forecasting solar flares is essential for mitigating space weather risks that threaten technological infrastructures. This research investigated the most relevant attributes for predicting solar flares by comparing the predictive performance of models based on two different predictive philosophies: an "Effect-to-Effect" approach based on X-ray inertia, and a "Cause-to-Effect" approach focused on magnetic topology of active regions. Additionally, an arbiter Meta Model was proposed to integrate both datasets. To address the high dimensionality and class imbalance characteristic of astrophysical data, an hierarchical cascade architecture named Solarfall was developed. This architecture employs four sequential XGBoost-based specialist classifiers to separate a solar flare signal from systemic noise.
Specifically, Solarfall divides the detection problem into four specialized stages. The initial stages prioritize maximizing recall by separating potential alerts from solar calm and filtering out low-impact noise. Subsequent stages act as mechanisms, progressively isolating medium flares from severe threats, and ultimately distinguishing Class X events from Class M storms. To prevent the propagation of cascading errors and overcome exposure bias, the intermediate and final classifiers underwent conditioned training. They were trained on the residual distribution—containing both true signals and false positives—that survived previous cuts, allowing the algorithm to learn the mathematical signatures of predecessor errors and protect the final decision's rigor.
The study used data spanning from 2010 to 2024, employing a chronological split to test the models on Solar Cycle 25. Results indicated that the multi-stage cascade has potential to reduce noise, filtering the initial 158,906 samples down to 26,428 (17%) at the third stage and to 4,440 (3%) at the fourth and last stage, marking a 97% reduction. However, the X-ray approach proved inadequate for forecasting extreme Class X events, yielding a negligible recall of 0.03 at the final stage (capturing 3% of extreme flares). Furthermore, evaluating the integration of x-ray flux and magnetic topology over the test set refuted the hypothesis of a superior combined model. The unification triggered an effect where X-ray flux acted as a confounding variable that degraded the system's accuracy and corrupted predictions.
Conversely, the architecture relying on magnetograms achieving a robust Precision-Recall Area Under the Curve of 0.462 and a precision of 0.48 for events Classes M and X. The research concludes that physical preconditions of an active region like current density and magnetic free energy, constitute the most relevant indicators for solar flare forecasting, outperforming generalized macroscopic approaches.

Speaker: Eduardo Ferraz de Campos (Federal Institute of São Paulo)

40 Automatic Extraction of Filaments in H-alpha and Radio Solar Images using Python

Solar filaments are dark, thread-like structures of cool, dense plasma seen on the Suns’s surface. They usually mark a boundary between two opposite magnetic regions and may last for multiple days, changing their form, but eventually they vanish. Their disappearance may end up with a CME associated with geomagnetic storms that affect the geospace. Therefore filament tracking is an important task of space weather. In this work I implemented an algorithm in Python for the automatic extraction and tracking of the evolution of filaments in solar H-alpha and radio images. I will describe the main steps in the algorithm that involve: preprocessing of the images, application of morphological closing operations with multi-directional linear structuring elements, noise removal and, finally, feature selection. The results of the application of the algorithm for two cases involving observations in H-alpha and Radio are presented, which demonstrate its efficiency.

Speaker: Dr Jean Carlo Santos (instituto nacional de pesquisas espaciais)

3:40 PM Coffee Break

Invited Talks

41 The Data-Driven Evolution of Heliophysics: Bridging Physical Mechanisms and Machine Learning in the Inner Magnetosphere

Heliophysics was born as a data-driven scientific field. From the discovery of the radiation belts and the confirmation of the existence of the solar wind, with the beginning of the space age, and until the latest discoveries in solar, magnetospheric, and ionospheric dynamics and coupling mechanisms, our field has always relied on the use of abundant in-situ and remote measurements to confirm physical processes and theoretical ideas, and to discover new phenomena. Thus, with every leap in technology, the field has greatly advanced. In recent years, the abundant data generated by ground and space-based and ground instruments, coupled with the increasing availability of such data in both long-term repositories and near-real time, has created both challenges and opportunities for new science and predictive modelling. In particular, the explosion of Data Science, Machine Learning Methods, and Artificial Intelligence algorithms to handle large datasets has been crucial to enable new scientific discoveries and develop low-cost data-driven models for Space Weather applications. While we are still in the beginning of the data-driven era of Space Weather, significant advancements have occurred in the past years, in part, due to the lower cost of running models in real time compared to the more traditional physics-based methods that usually require large data centers. This talk discusses the pivotal role of Machine Learning (ML) and Artificial Intelligence in navigating this data-rich era. I will present an overview of the latest scientific results in solar wind-magnetosphere coupling, with a focus on radiation belt processes. In particular, I will discuss recent results in MeV electron acceleration and losses during geomagnetically active times, as well as the effects of solar-wind driven perturbations in the inner-magnetosphere dynamics. The scientific results will then be put in the context of the recent efforts that have been made in using this physical knowledge and data availability for creating better predictive models in the Earth's inner magnetosphere.

Speaker: Victor A. Pinto

Oral Contributions

42 Application of Machine Learning Techniques for Instability Prediction and Risk Management Support in Mining Tailings Pile

The stability of tailings storage structures represents one of the major challenges in contemporary geotechnical engineering, particularly considering the environmental, social, and economic impacts associated with geotechnical failures. The increasing adoption of filtered tailings systems and dry stacking methods has expanded the use of tailings piles as an alternative to conventional disposal techniques. Despite the advantages associated with these structures, factors related to hydraulic behavior, infiltration processes, moisture variations, slope stability, and environmental conditions make their analysis a complex problem.

At the same time, advances associated with Industry 4.0 have enabled the application of Artificial Intelligence, Machine Learning, remote sensing, and real-time geotechnical monitoring techniques for identifying patterns related to structural stability. However, most existing studies remain focused on tailings dams, while only a limited number of investigations have specifically addressed tailings piles through the integrated use of multiple data sources.

In this context, the present study proposes an integrated approach based on Machine Learning techniques to identify instability precursor patterns in tailings piles using geotechnical, climatic, instrumental, and geospatial data. Models such as Random Forest, XGBoost, and Long Short-Term Memory (LSTM) networks will be evaluated to identify complex and temporal relationships within the datasets. The expected results aim to contribute to the development of methodologies for preventive monitoring, risk mitigation, and the advancement of scientific knowledge regarding the application of Artificial Intelligence in the mining ind

Speaker: Dr Matheus Alves de Barros (Topotech - Soluções em Topografia e Georreferenciamento)

43 A CNN Classifier for Hidden Solar Flares in Historical UV Spectroscopic Plates From Skylab SO82B

The NRL SO82B spectrograph on board of Skylab captured more than 6,000 far-ultraviolet photographic exposures of the Sun between 1973 and 1974. UV flare spectra are still rare with only a few reported during the mission. This work presents a supervised binary image classifier based on a ResNet-18 convolutional neural network (CNN) to identify uncatalogued flares in the SO82B data. The CNN was trained with transfer learning from ImageNet weights to distinguish flare from non-flare SO82B plate images. Training on 1,281 labelled frames (321 flares and 960 non-flares) and evaluating on a held-out test set of 641 independent examples (161 flares and 480 non-flares), the model achieved a ROC-AUC of 0.88 and an area under the precision-recall curve (AUPRC) of 0.75, confirming adequate discrimination capability. At the standard threshold of p > 0.50, the model flagged 400 candidate frames, achieving a held-out flare recall of 72.7% and a false-positive rate of 9.8% on held-out non-flares. Further analysis revealed a strip-level systematic: after plate-level suppression the 253 candidates (p > 0.70) reduce to 136 unique plates, and after strip-level suppression to all 3 film strips. This collapse suggests the model partially responds to strip-level properties, such as emulsion batch or scanner calibration artefacts, rather than exclusively to frame-level flare morphology, which it was expected. To isolate genuine events from this systematic baseline, a within-strip outlier analysis was applied, retaining only plates whose maximum frame score exceeds the strip median by more than 2σ. This conservative approach has led to identification of 68 plates of statistically significant, of which 20, it was from strip 1B (May 30 – Jun 17, 1973) and 48 from strip 2B (Aug 09 – Sep 12, 1973). Cross-referencing the 68 outlier candidates against the NOAA/NGDC flare catalogue for 1973 reveals that 5 candidate plates have capture times that fall within the duration of a catalogued optical flare event, and with 10 more within 30min of a catalogued flare; the strongest temporal match is on 1973 June 15, where four consecutive SO82B plates were acquired during a class 1F optical flare at active region AR 2382. Gradient-weighted class activation maps confirm that the network attends to the chromospheric emission core of the UV spectrum rather than plate artefacts or edges. The resulting candidate list provides a prioritised target list for spectral follow-up and for extending the SO82B flare database.

Speakers: Tiago Mendes Ferrer (@Mackenzie), Paulo Simões (Universidade Presbiteriana Mackenzie)

Thu, August 20

Invited Talks

44 Coupling Between Complex Solar Wind Structures and Outer Radiation Belt Dynamics

The solar photosphere produces various solar structures, such as Coronal Mass Ejections (CMEs) and High-Speed Solar Wind Streams (HSSs), which can propagate toward Earth and interact in the interplanetary medium, forming complex solar wind structures. In this study, we focus on events composed of two identical solar wind structures, specifically those formed by two CMEs or by two HSSs. While the coupling between a single ICME or HSS and the Earth’s magnetosphere, as well as its impact on electron flux variability in the outer radiation belt, is relatively well understood, the effects of complex solar wind structures remain insufficiently explored, particularly regarding outer radiation belt dynamics. This work uses the solar wind parameters measured at L1 Lagrangian point to identify the main characteristics of each complex solar wind structure, especially in terms of velocity components and the energy transport, which exhibit markedly different behaviors. In the inner magnetosphere, in situ measurements of high-energy electron flux and magnetospheric waves, such as ultra-low frequency (ULF) and whistler-mode chorus waves, together with calculations of radial and pitch angle diffusion coefficients, indicate that these processes respond differently during electron flux enhancements driven by each type of complex structure. We explicitly compare how the different solar wind drivers modulate energy transfer, wave activity, and diffusion processes, thereby governing the dynamic response of the outer radiation belt. These results demonstrate that the coupling between complex solar wind structures and outer radiation belt dynamics depends on the specific type of solar structures composing the event, as their intrinsic properties differ substantially. Furthermore, our findings suggest that empirical models used to calculate radial and pitch angle diffusion coefficient may require refinement to account for the distinct characteristics of different complex solar wind drivers.

Speaker: Dr Ligia Alves da Silva (Instituto Nacional de Pesquisas Espaciais - INPE)

Oral Contributions

45 GNSS-TEC Characterization of the 17 March 2015 Geomagnetic Storm over Morocco as a Benchmark for Data-Driven Space Weather Monitoring

We investigated the ionospheric response to the 17 March 2015 geomagnetic storm, the most severe event of solar cycle 24 (SYM-H = -233 nT), using GPS-derived total electron content (TEC) observations from the RABT station in Morocco (34°N, 7°W). Vertical TEC (VTEC) was derived from dual-frequency GPS measurements using differential code bias corrections and a 30° elevation cutoff. The results reveal a strong storm-time enhancement, with ΔVTEC reaching +45.78 TECU, corresponding to a 158% increase above the quiet-day baseline of 28.92 TECU. Lagged correlation analysis shows the strongest relationship with Kp at approximately 3 hours delay (r ≈ 0.53), indicating structured ionospheric variability linked to geomagnetic forcing. The temporal evolution is consistent with rapid electrodynamic effects during the main phase and composition-related changes during the recovery phase. This study highlights the usefulness of GNSS-TEC time series for regional space weather monitoring and provides a physically interpretable benchmark for future machine-learning approaches to ionospheric disturbance detection and prediction in the still understudied North African sector.

Speaker: Nouhaïla Bouhadi

46 Integrating Machine Learning for Space Weather Risk Mitigation in PSM-Validated Electronic Subsystems

The increasing reliance on Commercial Off-The-Shelf (COTS) components for small satellite propulsion and suborbital platforms, such as the Brazilian Suborbital Microgravity Platform (PSM), introduces significant vulnerabilities to space weather phenomena. This study investigates the intersection between heliophysics data and the operational reliability of electrospray propulsion systems and power processing units (PPU). While the PSM provides a high-quality microgravity environment for technology validation, its suborbital trajectory exposes payloads to variable ionospheric conditions and solar-induced radiation, which can lead to transient faults in unhardened COTS electronics.
We propose a framework where Machine Learning (ML) algorithms, trained on solar activity and magnetospheric data, are utilized to predict performance anomalies in propulsion telemetry collected during PSM flight campaigns. By correlating environmental heliophysics datasets with real-time housekeeping data—such as current stability in ion emitters—it is possible to distinguish between inherent hardware failures and externally induced environmental interference. Furthermore, this research aligns with international engineering protocols by applying European Cooperation for Space Standardization (ECSS) tailoring to define safety-critical margins for missions operating under high solar activity. The integration of predictive ML models into the systems engineering lifecycle enhances the sustainability of the suborbital economy by ensuring that recoverable payloads can withstand the dynamic challenges of the space environment.

Speaker: Felipe Portela Aguilar de Oliveira (Christianne Andrea Portela Aguilar)

47 Effect of solar wind plasma on Space weather

Today’s challenge for space weather research is to quantitatively predict the
dynamics of the magnetosphere from measured solar wind and interplanetary mag
netic field (IMF) conditions. Correlative studies between geomagnetic storms (GMSs)
and the various interplanetary (IP) field/plasma parameters have been performed to
search for the causes of geomagnetic activity and develop models for predicting the
occurrence of GMSs, which are important for space weather predictions. We find a
possible relation between GMSs and solar wind and IMF parameters in three different
situations and also derived the linear relation for all parameters in three situations.
On the basis of the present statistical study, we develop an empirical model. With the
help of this model, we can predict all categories of GMSs. This model is based on the
following fact: the total IMF Btotal can be used to trigger an alarm for GMSs, when
sudden changes in total magnetic field Btotal occur. This is the first alarm condition
for a storm’s arrival. It is observed in the present study that the southward Bz compo
nent of the IMF is an important factor for describing GMSs. A result of the paper is
that the magnitude of Bz is maximum neither during the initial phase (at the instant
of the IP shock) nor during the main phase (at the instant of Disturbance storm time
(Dst) minimum). It is seen in this study that there is a time delay between the maxi
mum value of southward Bz and the Dst minimum, and this time delay can be used
in the prediction of the intensity of a magnetic storm two-three hours before the main
phase of a GMS. A linear relation has been derived between the maximum value of
the southward component of Bz and the Dst, which is Dst = (−0.06)+(7.65)Bz+t.
Some auxiliary conditions should be fulfilled with this, for example the speed of the
solar wind should, on average, be 350 km s−1 to 750 km s−1, plasma β should be low
and, most importantly, plasma temperature should be low for intense storms. If the
plasma temperature is less than 0.5×106 K then the Dst value will be greater than the
predicted value of Dst or if temperature is greater than 0.5 ×106 K then the Dst value
will be less (some nT)

Speaker: Dr Balveer Singh Rathore (Government Holkar Science College, Indore)

10:40 AM Coffee Break

Oral Contributions

48 Modeling the Variability of Relativistic Electron Flux in the Outer Radiation Belt Using Deep Learning

The main factors governing the variability patterns of high-energy electron flux in the outer radiation belt are well established in the literature and are modulated by external physical mechanisms. These correspond to different solar wind structures, such as Interplanetary Coronal Mass Ejections (ICMEs) and High-Speed Solar Wind Streams (HSSs). These processes are associated with low-energy electron flux injections and also with wave-particle interactions, in which electrons are accelerated through L-shell layers. Thus, attempting to capture, predict, and explain the dynamics of hight-energy electron flux enhancement and dropout effects during perturbed periods using machine learning methods can yield new insights into the physical mechanisms controlling such variations in the outer radiation belt, as well as into the interplanetary medium, since one construct derives from the other. The models are trained using electron flux data from the Relativistic Electron Proton Telescope (REPT), an instrument aboard the Van Allen probes, solar wind data from ACE satellite, and global geomagnetic indices. The results obtained by the models, capable of reproducing the flux variability, are compared with the physical processes observed during quiet and disturbed periods of the previous solar cycle, between 2015 and 2018, a period whose onset of cadence is similar to the state of the current solar cycle. In addition to the results inferred by the models, methodologies capable of providing a comprehensive and coherent interpretability of the processes that occur during the breaking of the third invariant, an effect capable of contributing to the increase in the asymmetry of the radiation belts, will be investigated. Finally, this data science-based computational modeling work aims to improve the capabilities of forecasting space weather, contributing to the understanding of the dynamics of radiation belts under different space weather conditions, while also revealing relevant information about the main causative drivers.

Speaker: Thiago Sant Anna (National Observatory - ON)

49 Predicting CME speed at 20 using machine learning approaches

Coronal mass ejections (CMEs) are significant drivers of space weather, and accurately predicting their propagation speed is crucial for mitigating their impact on Earth’s environment. In this study, we leverage machine learning techniques to model and predict CME speed at 20R utilizing data from the Coordinated Data Analysis Workshop catalog. We considered data from Solar Cycles 23 and 24, divided into their rising, maxima, decline, and minima phases, to train multivariate linear regression, Random Forest, and XGBoost machine learning models aimed at predicting CME speeds at 20R. The machine learning models use linear speed, acceleration, width, and kinetic energy as input features to estimate CME speeds at 20R. Our results indicate that Random Forest and XGBoost models significantly outperform linear regression model across all datasets, achieving high R2 values (≈0.97) and low relative errors (6%) for most phases, especially during high solar activity. Feature importance analysis identifies CME linear speed and acceleration as the dominant predictors of CME speed at 20R. This result is consistent with physical models, which describe CME propagation as being influenced primarily by initial speed and the drag force acting through acceleration or deceleration in the interplanetary medium. The trained models were applied to available events from Solar Cycle 25, to predict CME speeds at 20R. The predicted values showed very good agreement with the actual speeds reported in the CDAW catalog. This successful application demonstrates the models’ generalizability and potential for forecasting future CME dynamics. Furthermore, such data-driven predictions can complement physics-based models—such as the Drag-Based Model—by providing reliable speed estimates at specific heliocentric distances, thereby enhancing the accuracy of space weather forecasts.

Speaker: Manjunath Hegde (Indian Institute of Astrophysics)

50 EoFTCNets: Efficient Solar Flare Nowcasting using 3D Temporal Convolutional Networks (3DTCN)

In the evolving landscape of 21st-century space science, forecasting space weather events such as solar flares and Coronal Mass Ejections (CMEs) are crucial yet challenging. Solar flares are intense bursts of radiation caused by the release of magnetic energy in active regions and are often accompanied by CMEs. These events can significantly impact Earth’s space environment, causing disruptions in radio communication, satellite operations, and power grids. Monitoring the temporal evolution of active regions and providing early warnings of solar flares is essential to mitigate these risks. Deep learning techniques have demonstrated significant success in detecting and predicting time-dependent events. By leveraging spatial data through convolution operations with temporal correlations, we introduce 3D Temporal Convolutional Networks(3DTCNs) to efficiently analyze active region patches over time, leveraging spatial and temporal correlations. Additionally, we introduce separate predictor modules based on flare classification to enhance the performance of our EoFTC-Nets (Eye-on-Flare Temporal Convolutional Networks) nowcasting system. Our results demonstrate that the proposed architecture matches or outperforms state-of-the-art approaches in the literature, achieving an accuracy exceeding 96% for a 24-hour forecasting window. Furthermore, the model is computationally efficient, consuming approximately 1.2 watts on Intel Movidius Myriad X, making it well-suited for onboard deployment and real-time space weather monitoring.

Speaker: Besma Guesmi (Ubotica Technologies)

51 F10.7 Solar Radio Flux Prediction using a Multi-Stage Self-Organizing Map-Autoencoder-LSTM Model

F10.7 is a solar radio flux measured at a wavelength of 10.7, which represents the vital proxy of solar activity. In this study, the impact of solar ultraviolet (UV) radiation on the upper atmosphere of Earth is considered by F10.7. This study aims to develop a predictive system for solar radio flux at F10.7 for short-term prediction from 1 to 3 days ahead. We implemented a multi-stage learning model to improve the short-term forecast of solar radio flux at F10.7. In the previous study, we explored a total solar irradiance forecast using a self-organizing map-autoencoder-LSTM model. In this study, we extend this approach to forecast solar radio flux at F10.7 using the same self-organizing map-autoencoder-LSTM model. We applied a self-organizing map (SOM) to cluster and segment the solar image features of continuum images, magnetograms. The μ maps represent the heliocentric angular parameter as an additional input channel alongside continuum images and magnetograms to account for center-to-limb variation and spatial projection effects across the solar disk. This model generates the feature map that serves as input to the autoencoder (AE) model. The encoder is used to reduce feature dimension and extract the compressed features. The compressed latent features are further integrated with the lagged solar radio flux at F10.7 to represent the input of a long short-term memory (LSTM) network to perform the prediction of solar radio flux at F10.7 for a few days ahead.
To the best of our knowledge, this study represents the first implementation of a SOM-AE-LSTM framework to forecast solar radio flux at F10.7. The proposed hybrid data-driven framework complements the existing F10.7 solar radio flux reconstruction and modeling approaches by integrating image-based solar information with lagged solar radio flux at F10.7.
We evaluated the performance of this model using the mean squared error (MSE), the root mean squared error (RMSE), the correlation coefficient (R), the determination coefficient (R2), and the mean absolute percentage error (MAPE) as performance metrics. We execute the forecast performances with the baseline and benchmark models, and conduct the ablation study to quantify the contribution of each component of the proposed architecture for this study.

Speaker: Idowu Raji (National Institute for Space Research (INPE))

12:30 PM Lunch

Invited Talks

52 The Hainan Ionogram Dataset: Supporting AI-Enabled Spread-F Monitoring, Prediction, and Space Weather Operations

Equatorial and low-latitude Spread-F (SF) can severely affect radio communication and navigation systems, making timely identification and prediction important for space weather monitoring and operations. Although artificial intelligence (AI) has shown strong potential for automating ionogram analysis, broader progress has been limited by the lack of large-scale, well-annotated, and publicly available datasets. Traditional manual ionogram scaling and interpretation are labor-intensive and subject to inconsistencies, especially for long-term observations and borderline cases.

Here we present a long-term ionogram dataset from the Hainan station (19.5°N, 109.1°E), covering 2002–2016 and spanning an entire solar cycle. The archive includes more than 517,000 expert-labeled raw ionograms. Building on this long-term observational foundation, we established a standardized human-in-the-loop active learning workflow to improve labeling consistency and scalability. This process led to a curated public dataset of 150,000 high-quality images, balanced across five classes—frequency SF, range SF, mixed SF, strong range SF, and non-SF—with 30,000 samples per class. The released dataset provides a practical benchmark resource for AI-based ionogram analysis and supports the development of scalable methods for future large-volume observations.

We further illustrate the value of this dataset through three published AI studies based on the Hainan observations: (1) real-time automated detection and subtype classification of SF using image-based deep learning; (2) nowcasting of ionogram sequence evolution using a spatio-temporal ConvGRU framework; and (3) physically constrained generative modeling of quiet and disturbed ionospheric features using IonoGAN. Together, these studies demonstrate that the Hainan dataset can support diverse downstream tasks, from automated event identification to short-term prediction, and can serve as a data foundation for AI-enabled space weather research and operations.

Speaker: Zheng Wang (National Space Science Center, CAS)

Oral Contributions

53 The Alfvén Critical Surface via Parker Solar Probe: Current Findings and Future Directions

This work aims to identify and characterize the critical Alfvén surface, which marks the boundary of the solar corona. Within this surface, the solar plasma is in a sub-Alfvénic regime, whereas beyond it, the solar wind becomes super-Alfvénic. The study presents theoretical foundations on the solar structure, plasma characteristics and Alfvén waves in the solar corona. The analysis was performed using data from the Parker Solar Probe from the eighth, ninth and tenth solar flybys. The parameters evaluated to determine the behavior of the plasma in the study region were: the magnetic field, its radial component, the particle density, the Alfvén velocity, the radial velocity, the plasma beta, the ratio between the kinetic energy, the magnetic energy and the Mach number. The results indicate that, during the eighth encounter, it was possible to identify the moment of the spacecraft’s entry into and exit from the solar corona, whereas during the ninth and tenth encounters only the entry was identified. Future research perspectives include investigating the energy balance (kinetic, magnetic and internal), along with performing solar wind simulations, incorporating the spacecraft trajectory during intervals in which it is within the sub-alfvénic plasma regime.

Speakers: Caroline Botelho (Instituto Nacional de Pesquisas Espaciais), PAULO RICARDO JAUER (Instituto Nacional de Pesquisas Espaciais INPE)

54 Benchmarking a synthetically-trained neural network for CME Segmentation in SolO/EUI and PROBA3/ASPIICS observations

Coronal Mass Ejections (CMEs) are critical drivers of space weather, requiring precise kinematic and morphological characterization to predict their geoeffectiveness. We previously demonstrated that fine-tuning the deep neural model Mask R-CNN on synthetic CME images, generated via Graduated Cylindrical Shell (GCS) shapes and raytracing, allows the automated segmentation of CME outer envelopes in coronograph images from SOHO/LASCO and STEREO/COR.

Following the validation of our core methodology, this work focuses on evaluating the model's performance when applied to solar observations from multiple instruments (not present in the training dataset). Specifically, we expand our validation framework by incorporating images from the Extreme Ultraviolet Imager (EUI/FSI) aboard Solar Orbiter and the ASPIICS coronagraph on Proba-3.

To quantify the network’s robustness, we assess standard segmentation metrics, including intersection over union, precision, and recall, and morphological parameters such as central position angle and angular width. All derived by comparing the predicted masks against a validation set of manually labeled images. This evaluation aims to establish the reliability of deep learning-based segmentation for future solar instruments and missions.

Speaker: Yasmin Machuca (CONICET - Grupo de Estudios en Heliofísica de Mendoza - Universidad de Mendoza)

55 Intelligent Atmospheric Density Forecasting for Autonomous Orbital Collision Avoidance

Increasing solar activity significantly alters the thermospheric density, causing unpredictable orbital decay for satellites in Low Earth Orbit (LEO). This study proposes an intelligent forecasting framework that utilizes Machine Learning to predict real-time changes in atmospheric drag and their subsequent impact on collision probabilities. The primary objective is to develop an autonomous system capable of adjusting satellite maneuver windows based on predicted space weather-induced atmospheric expansion.

The methodology leverages public datasets, including historical Two-Line Element (TLE) orbital data and solar indices (e.g., F10.7 flux and Kp indices). We employ a Long Short-Term Memory (LSTM) network to model the time-series relationship between solar activity and atmospheric drag at various orbital altitudes. By integrating this density forecast into a collision avoidance algorithm, the system calculates the "maneuver necessity" for active satellites. This approach eliminates the reliance on ground-based, high-latency atmospheric models.

The proposed framework is expected to improve the precision of orbit-decay predictions by 25% compared to standard models, enabling more efficient fuel usage for station-keeping. The significance of this work is its direct applicability to the current LEO satellite congestion crisis. By providing operators with an automated tool for anticipating drag-driven orbital changes, this research provides a scalable solution for maintaining safety in an increasingly crowded orbital environment, ultimately supporting the long-term sustainability of space operations.

Speaker: Tumma Anusha (Mahendra Engineering College)

3:40 PM Coffee Break

Poster Session

56 3D MHD Simulation of Solar Wind Interaction with the Lunar Environment: A Preliminary Study

We present a preliminary study of the interaction between the solar wind under supersonic and super-Alfvénic conditions in the lunar environment outside the Earth's magnetosphere using a three dimensional magnetohydrodynamic model that solves the ideal MHD-PRJ equations. The model incorporates observationally plasma parameters from upstream monitors (ACE, DSCVR, WIND) as input, allowing a physically consistent description of plasma conditions. The simulation captures the evolution of key parameters such as velocity, density and magnetic field, showing the formation of discontinuities, shock-like structures and a wake region downstream of the Moon. Preliminary results show that the model can reproduce plasma structures and can be extended to include additional physical processes.

Speaker: vinicius deggeroni (INPE)

57 Adaptive Hybrid Distribution Modeling of Radiation Belt Electrons

The Earth’s radiation belts host electron populations that frequently deviate from thermodynamic equilibrium, exhibiting pronounced suprathermal tails and anisotropies driven by wave–particle interactions. Accurately representing these non-Maxwellian features is essential for understanding acceleration processes and improving predictive capabilities. In this work, we introduce an adaptive hybrid distribution modeling framework for radiation belt electrons. The method automatically identifies the optimal functional representation of the observed electron flux among Maxwellian, kappa, bi-Maxwellian, bi-kappa, and hybrid Maxwellian–kappa distributions. Model selection is performed algorithmically based on goodness-of-fit metrics, enabling a data-driven characterization of the underlying plasma regime. The selected distribution parameters are then used to reconstruct the electron flux density and to provide physically interpretable quantities such as density, characteristic temperature, and suprathermal index. Building upon this representation, we implement a neural network forecasting model that leverages the fitted distribution parameters to predict electron flux variability. This adaptive framework provides a unified, physically grounded approach to modeling and forecasting radiation belt electron populations, capturing both near-equilibrium and strongly non-thermal regimes.

Speaker: Jose Marchezi (National Institute for Space Research - INPE)

58 Application of a wavelet technique to identify planetary magnetosphere boundaries

The study of planetary magnetospheres boundaries (bow shocks and magnetopauses or induced magnetopauses) is important to study solar wind magnetosphere physical processes. Thus their identification from in situ spacecraft data is of great importance. In this work we present a method based on the Haar wavelet decomposition technique applied to magnetic field measurements from spacecraft in different planetary magnetospheres. Cases of application to Saturn (Cassini mission) and Mercury (MESSENGER mission) are presented. The methodology based on variance obtained by scale and edge identifications was shown to be a simple tool to perform this task. The results confirm that the Haar transform can efficiently identify the planetary magnetosphere boundaries characterized
by the abrupt magnetic field changes. This technique could be applied elsewhere in the heliosphere to identify abrupt variations in plasma and magnetic field parameters (interpanetary shocks, discontinuities).

Speaker: Ezequiel Echer (Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brasil)

59 Machine learning algorithm to forecast VTEC over the Brazilian low-latitude region

R. de Jesus1,2,3, G. Yang1, A.A. Pimenta3, A.J. de Abreu3,4, L.F.R. Vital1,2,3, O.M. Adebayo3, V.F. Andrioli3, A.M. Santos3, P.P. Batista3, M.P.P. Martins3, L.C.A. Resende3, C.S. Carmo1,2,3, D. Medeiros1,2,3, I.S. Batista3, K. Venkatesh5, P.R. Fagundes6, C. Wang1, H. Li1, Z. Liu1

1State Key Laboratory of Space Weather, National Space Science Center/Chinese Academy of Sciences, Beijing, China, 2China-Brazil Joint Laboratory for Space Weather, National Space Science Center, São José dos Campos, Brazil, 3National Institute for Space Research (INPE), São José dos Campos, Brazil, 4Instituto Tecnológico de Aeronáutica (ITA), Divisão de Ciências Fundamentais, São José dos Campos, SP, Brazil, 5Physical Research Laboratory, Ahmedabad, India, 6Universidade do Vale do Paraíba /IP & D, São José dos Campos, SP, Brazil

Abstract
The scientific community has long been interested in developing models for ionospheric forecasting. Several studies have applied machine learning to develop more sophisticated models. In this work, machine learning models have been developed to predict vertical total electron content (VTEC) at Cachoeira Paulista (CHPI; 23ºS, 45ºW), Brazil. For this, the Random Forest (RF) and Support Vector Regression (SVR) from machine learning algorithms will be used. The GNSS and meteor radar observations from CHPI during 2019-2020 are used in this investigation. The performance of the machine learning models is compared with that of Seasonal Autoregressive Integrated Moving Average (SARIMA). Models performance is validated using Root Mean Square Error (RMSE) and Mean Square Error (MSE). Initially, the input parameters considered in the models were Dst index and F10.7 solar flux. Preliminary results showed that the SARIMA model provides slightly better VTEC forecasts than the RF and SVR models. This result suggests that additional input parameters may be necessary to improve machine learning models. Therefore, wind data derived from the meteor radar, as well as days of the year (DOY) and local time will be incorporated as additional predictors.

Speaker: Rodolfo de Jesus (China-Brazil Joint Laboratory for Space Weather)

60 Using Machine Learning to Understand the Impact of Space Weather on Exoplanets

Space weather, caused by stellar activity such as flares and stellar winds, can strongly affect the atmospheres of exoplanets. These effects are important for understanding whether a planet can keep its atmosphere and potentially support life.
In this work, we present a simple theoretical overview of how machine learning can help study these effects. We explain how variations in stellar activity can influence exoplanet atmospheres and how machine learning can be used to find patterns between them.
The goal is not to build a model, but to show how artificial intelligence could be used as a tool to better understand complex interactions between stars and planets.
This study provides an accessible introduction to the connection between space weather, exoplanets, and machine learning, and highlights the importance of combining physics with modern data analysis methods.

Speaker: Salma Makhtich (Faculte of Science Semlalia Marrakech Morocco)

61 Prediction of relativistic electron fluxes (>2 MeV) in geostationary orbit using deep learning techniques.

Relativistic electrons (>2 MeV) in geostationary orbit (GEO) are of great scientific interest because, during disturbed geomagnetic conditions, their fluxes can increase by several orders of magnitude on timescales of just hours. This phenomenon causes internal charging in satellites, damaging their electrical circuits and affecting their operation. Various studies have shown that the flux of relativistic electrons in GEO presents a strong correlation with different geomagnetic indices, as well as with plasma parameters and the magnetic field of the solar wind.
In this work, a predictive model based on Deep Learning techniques is developed, using Convolutional Neural Networks (CNN) to forecast relativistic electron fluxes of >2 MeV in GEO, comparing single-output and multi-output configurations. The model was trained with historical electron flux data measured by GOES satellites through the EPEAD instrument, along with solar wind parameters from the OMNIWeb database. The model performance was evaluated for prediction horizons of 1, 12, and 24 hours. Likewise, models trained exclusively with data corresponding to periods of geomagnetic storms were implemented, analyzing predictive capability under disturbed conditions.
The results show that CNN-based models, when compared to each other, present consistent performance across different prediction horizons, maintaining similar RMSE values. Likewise, no significant difference is observed between single-output and multi-output configurations. On the other hand, when training exclusively with geomagnetic storm data, a deterioration in the overall performance of the models is observed, reflecting the difficulty of generalization under disturbed conditions. In this scenario, for a 24-hour horizon, the multi-output approach shows a better ability to capture the system dynamics compared to the single-output approach.

Speaker: Camila Peña Cifuentes (Universidad de Santiago de Chile)

62 The Magnetopause Standoff Position Variation During the Extreme Geomagnetic Storm of May 2024: MHD Simulation

The solar wind is an ionized gas built mostly of hydrogen nuclei and energized electrons. These particles are constantly ejected by the Sun. During the period of high solar activity, the plasma ejections become more intense, and extreme events can occur, such as intense coronal mass ejections (CMEs) and solar flares, which can result in severe magnetic storms. The interaction between a super magnetosonic solar wind plasma and the Earth’s magnetosphere creates a bow shock that deflects and decelerates the plasma. During fast CMEs or high-speed streams, the bow shock is compressed and, in consequence, compresses the magnetopause. The study of the magnetopause location is of utmost importance since it protects the Earth and satellite equipment orbiting near space. With the magnetopause compression, the possibility of energized particles, mass, and momentum transfer into the inner magnetosphere may impact the Earth's plasma convection. This study has the goal to analyze the solar wind parameters and the response of the magnetopause during the severe geomagnetic storm of May 2024, by using a magnetohydrodynamic computational simulation (MHD) available on the CCMC-NASA website. In addition, we calculate the magnetopause location obtained by Shue et al., (1998) empirical model and compare those results with the magnetopause nose position given by the simulation. The Dst and Sym-H indices were analyzed to classify the storm phases, and the behavior of the magnetopause as a function of the storm phases was analyzed. Our work demonstrates that during the sudden impulse commencement and main phase of the geomagnetic storm, the most intense compression of the magnetopause occurs. Due to the arrival of the solar structures, the magnetopause gets closer to the Earth. A high degree of similarity was observed between the magnetopause position calculated by the simulation and the Shue empirical model, except for some discrepancies arising from the distinct calculation methods: namely, the Shue model relies on the interplanetary magnetic field (Bz) and the dynamic pressure (Pd), whereas the simulation is based on the peak of current density (J). The results obtained allowed the understanding of the scientific fundamentals of space weather, with an emphasis on learning about the processes resulting from the interaction between the solar wind and the Earth's magnetosphere during extreme events.

Speaker: Nathan Mazzon Marquezin (USP)

63 A PINN-Based NLFFF Framework for Deciphering Coronal Magnetic Topology and Loop Parameters in NOAA AR 3663

We present a data-driven and artificial intelligence-based method for the three-dimensional characterization of magnetic field lines in the solar atmosphere, applied to NOAA AR 3663, one of the most complex and eruptive active regions of Solar Cycle 25. The method is based on the extrapolation of the nonlinear force-free coronal magnetic field (NLFFF) via Physically Informed Neural Networks (PINNs), using vector magnetograms from the SDO/HMI as a boundary condition, following the formulation of Jarolim et al. (2024). From the 3D volume generated by the PINN, the field lines are reconstructed by bidirectional numerical integration with the 4th-order Runge-Kutta algorithm (RK4). The morphological validation of the magnetic structures is performed by comparing them with coronal loops observed in extreme ultraviolet (EUV) images from the SDO/AIA. To complement the magnetic modeling ($\beta \approx 0$) with independent thermodynamic constraints, we performed Differential Emission Measurement (DEM) analysis using the CHIANTI atomic database, determining the electron density and temperature distribution along the reconstructed loops. Preliminary results reveal the geometric distribution (length and height) of the loops and the thermal structure of AR 3663, providing insights into free energy storage and the magnetic conditions that precede eruptive activity. The method demonstrates the potential of PINN-based NLFFF solvers as physically consistent and computationally efficient tools for coronal diagnostics and modeling of space weather originating in active regions.

Speaker: Bruno Fernandes Garcia

64 Adapting Terrestrial Radial Diffusion Theory to Saturn’s Relativistic Electron Belt

Radial diffusion is a key process controlling the dynamics of relativistic electrons in planetary radiation belts. In Earth’s outer belt, the radial diffusion coefficient (Dll) is typically derived from well-established analytical formulations that relate electromagnetic field fluctuations, particularly ultra-low frequency (ULF) waves, to stochastic radial transport. For Saturn, by contrast, available Dll descriptions are often reduced to simple power-law dependences on radial distance, without explicit consideration of the magnetospheric conditions that may control particle transport. However, the direct application of these formulations to Saturn remains uncertain due to the planet’s distinct magnetospheric environment, including rapid rotation, the presence of moons and rings, and the lack of continuous in-situ or upstream measurements. This work investigates how classical terrestrial formulations of radial diffusion can be adapted to Saturn’s relativistic electron belt. The study revisits the analytical derivations used to construct Dll from the Fokker-Planck equation, with emphasis on the assumptions that relate the radial diffusion coefficient to wave power and other associated parameters. It then assesses how this framework must be adapted for Saturn, identifying which components can be retained and which require reformulation to account for the planet’s magnetospheric environment and the processes governing its electron radiation belt dynamics. The objective is to establish a physically consistent pathway for extending the terrestrial radial diffusion theory to Saturn. This provides a theoretical foundation for future efforts aimed at quantifying radial diffusion and developing parameterizations for Saturn’s radiation belts.

Speaker: Edu Rockenbach (National Institute for Space Research)

65 Automated Detection of Solar Flare-induced disturbances on VLF signals using Hidden Markov Models

Space weather, particularly solar flares, poses a significant risk to global technological infrastructure, affecting satellites, power grids, communications, and navigation. Very Low Frequency (VLF) signals provide a terrestrial alternative for monitoring space weather: their propagation through the Earth‑ionosphere waveguide is sensitive to D‑region disturbances caused by solar X‑rays, enabling indirect, low‑cost, and real-time detection.

In this preliminary study, we develop and evaluate a Gaussian Hidden Markov Model (HMM) with three hidden states representing the signal's physical states during a flare perturbation (normal, onset, and decay). The model was trained on VLF signals from the NAA transmitter (24 kHz) recorded in Piura, Peru, during 2025. Preprocessing included FIR filtering, downsampling, and manual labeling. The signal was decomposed into background‑removed amplitude and its first derivative (velocity). Critically, the transition matrix was derived from labeled events, imposing physical constraints consistent with solar flare evolution.

Results show that the HMM achieves 71% precision and recall for detecting the onset state, successfully identifying solar events under noisy conditions while maintaining temporal logic aligned with flare physics. The model also demonstrates adequate computational efficiency for real‑time operation.
As future work we consider implementing Gaussian Mixture Model HMMs and implementing an autoregressive HMM.

Speaker: Aldo Arriola Cordova (Comisión Nacional de Investigación y Desarrollo Aeroespacial, Universidad Nacional de Ingeniería)

66 Characterization of Interplanetary Shocks Across Distinct Interplanetary Medium Environments Using Multi-Spacecraft Observation

Interplanetary shocks are fundamental agents of energy dissipation and particle acceleration in the heliosphere, playing a critical role in space weather phenomena that affect Earth's technological infrastructure. This study presents a comprehensive investigation of complex interplanetary shock events using multi-spacecraft observations to characterize their physical properties, associated structures, and geoeffective impacts across distinct interplanetary medium environments. The research utilizes in-situ plasma and magnetic field measurements from multiple missions including ACE, STEREO, Parker Solar Probe, and Solar Orbiter, enabling robust identification of shock events through comparative analysis across different heliospheric positions. Shock events are detected by examining abrupt discontinuities in solar wind parameters—magnetic field strength, plasma density, temperature, and velocity—with multi-spacecraft timing analysis employed to determine propagation directions and spatial extents. The Rankine-Hugoniot conservation relations are systematically applied to calculate key shock parameters including shock normal vectors, compression ratios, Alfvén Mach numbers, and the critical angle θBn between the upstream magnetic field and shock normal, allowing classification into quasi-parallel and quasi-perpendicular geometries. Beyond shock characterization, the investigation extends to the interplanetary structures driving these disturbances—interplanetary coronal mass ejections (ICMEs), sheath regions, and magnetic clouds—establishing causal linkages between solar eruptions and heliospheric shock formation. Geoeffectiveness assessment is performed by correlating shock arrival times at near-Earth spacecraft with geomagnetic indices (Dst and Kp), quantifying how specific shock characteristics influence magnetospheric disturbances. Statistical analysis of shock events across varying longitudinal separations, building upon previous work demonstrating a 50% probability cutoff at 90° angular separation, provides insights into shock front expansion and propagation evolution through the inner heliosphere. By integrating multi-point observations, theoretical shock physics, and space weather impact assessment, this research contributes to improved understanding of solar-terrestrial coupling and enhances predictive capabilities for geomagnetic storm forecasting, addressing the growing societal need for reliable space weather preparedness as technological system dependencies continue to increase.

Keywords: Interplanetary shocks, multi-spacecraft observations, Rankine-Hugoniot conditions, geoeffectiveness, interplanetary coronal mass ejections, space weather, magnetic clouds, shock parameters, heliospheric propagation

Speaker: ALIU KARIM (NATIONAL INSTITUTE FOR SPACE RESEARCH (INPE))

67 Electron Flux Variations Associated with Plasmaspheric Plumes and Hiss Waves under Different Solar Wind Conditions

The variation of electron flux in the energy range from 0.015 to 51.768 keV is analyzed in association with the occurrence of plasmaspheric plumes under different solar wind conditions. During the formation of these structures, regions of enhanced plasma density extend to higher L-shell values, allowing the propagation of whistler-plume waves. The observed flux variations may be associated with wave–particle interaction processes occurring in these regions, suggesting interaction between hiss waves and the low-energy electron population of the plasmasphere. In this work, two distinct periods are analyzed: a geomagnetically quiet interval on 2015-08-04/05 and a disturbed interval during the arrival of an ICME on 2015-03-01. During the quiet period, in the presence of a plasmaspheric plume and whistler-plume waves, a decrease in electron flux is observed in the energy range from 0.043 to 0.140 keV, followed by a rapid enhancement after the plume passage. In the higher-energy channels, between 27.153 and 41.749 keV, an increase in electron flux is observed during the plume occurrence; however, this enhancement does not begin at the exact onset of the plume and decreases before the plume ends, with flux levels increasing again after the plume passage. During the disturbed interval associated with the ICME, electrons in the energy range from 0.526 to 1.047 keV exhibit an enhancement during the occurrence of the plume and the presence of whistler-plume waves, reaching a maximum peak followed by a gradual decrease. In contrast, lower-energy electrons around 0.067 keV and higher-energy electrons above 4.147 keV remain relatively stable throughout the event. These results indicate that plasmaspheric plumes associated with whistler-plume waves may contribute to distinct electron flux responses under different solar wind conditions, particularly within low-energy electron populations in the plasmasphere.

Speaker: Pedro Henrique Ferreira Fister

68 Estimating the Recovery Time of Forbush Decrease Events through Cosmic Ray Diffusion

The solar magnetic field is complex and, due to the motion of the solar plasma, it continually twists and accumulates magnetic energy in regions of strong instability. During periods of maximum solar activity, Coronal Mass Ejections (CMEs) occur frequently; these are phenomena in which large quantities of plasma, together with part of the solar magnetic field, are expelled into interplanetary space. When these plasma structures reach Earth, they partially shield Primary Cosmic Rays (PCRs). When PCRs collide with the Earth's atmosphere, they produce the so-called Secondary Cosmic Rays (SCRs). For PCRs with energies up to 100 GeV, pion production occurs. Charged pions decay into positively and negatively charged muons, accompanied by muon neutrinos and antineutrinos, respectively. We detect these muons using a detector at the Southern Space Observatory (OES/INPE), located in São Martinho da Serra, RS, Brazil. This detector is part of the Global Muon Detector Network (GMDN), whose other detectors are installed in Nagoya, Japan; Hobart, Australia; and Kuwait City, Kuwait. Since muons are generated from the interaction of PCRs with the Earth's atmosphere, the reduction in the arrival of these primary cosmic rays causes a decrease in muon counts at ground level. This phenomenon is known as a Forbush Decrease. This work aims to study the recovery time of the PCR flux after the passage of a CME.

Speaker: João Gabriel Sant' Ana Rodrigues (INPE, UFSM)

69 Magnetosphere response to extreme events during solar cycle 25

Understanding and predicting magnetospheric response to variations in interplanetary conditions is a central goal of space physics. Prominent interplanetary structures, such as Coronal Mass Ejections (CMEs) and Corotating Interaction Regions (CIRs), are the main drivers of the geomagnetic storms, which are classified according to the magnitude of Earth’s magnetic field perturbations. These disturbances are quantified by the Dst index, and the storms are classified as weak (Dst between -30 and -50 nT), moderate (Dst between -50 and -100 nT), intense (Dst between -100 and -250 nT) and super-intense (Dst lower than -250 nT). During intense and super-intense storms, the Earth’s magnetopause undergoes significant compression, enhancing energy transfer into the magnetosphere . Several theoretical models have been developed to describe magnetopause dynamics, for example the model proposed by Shue et al., (1997, 1998), which estimates the magnetopause location as function of solar wind parameters. . Although the model is widely used in the space physics field, it still has limitations that can be evidenced especially during extreme events. The goal of this work is to analyse the magnetopause variation during intense and super-intense geomagnetic storms. The analysis uses in situ satellite data from solar wind, magnetosheath, and inner magnetosphere. At this point the extreme events were identified and categorized. OMNI dataset are used for an initial characterization of the events and identification of the driving interplanetary structures. Magnetopause crossings are identified using data from MMS, THEMIS, and GOES and compared with the empirical model predictions to investigate its limitations. The data are obtained using the Python version of the SPEDAS software (PySPEDAS), which was developed for the analysis and visualization of data from various space missions.

Speaker: Bruna Parisi Hilsdorf (Escola de Engenharia de Lorena - Universidade de São Paulo)

70 Particle In Cell Modeling of Whistler-Driven Electron Heating in the Solar Wind Plasmas

Electron heating in the solar wind and the evolution of velocity distribution functions (VDF) remain fundamental questions in space physics. This work presents a numerical study of wave-particle interactions, focusing on electron heating driven by whistler waves. To capture the detailed kinetic effects governing this dynamics, we employ the KEMPO (Kyoto University ElectroMagnetic Particle Code) for one-dimensional (1D) Particle-In-Cell (PIC) simulations. As input parameters, the model is initialized with the plasma beta ($\beta_e$), temperature anisotropy ($T_{\perp}/T_{\parallel}$), and the ratio between plasma and cyclotron frequencies ($\omega_{pe}/\Omega_e$). Furthermore, Kappa velocity distribution functions are utilized to represent suprathermal populations, specifically the electron strahl. Based on the temporal evolution of these conditions, the resulting parameters extracted from the simulation include electromagnetic field fluctuations ($\delta B$, $\delta E$), the evolution of pitch-angle scattering, and variations in the electron heat flux. The 1D approach allows for isolating energy and momentum exchanges along the wave propagation direction, providing high-resolution analysis of particle energization. The expected results aim to deepen the theoretical understanding of kinetic signatures observed in situ by space missions, connecting the microscale physics of plasma instabilities with the macroscopic thermodynamics of solar wind expansion.

Speaker: Laíne Soares (Instituto Nacional de Pesquisas Espaciais)

71 Study of the Earth's bow shock position through in situ observations and empirical models

The interaction between the supersonic solar wind and the Earth’s magnetic field forms the bow shock, a collisionless shock wave whose shape and position vary according to solar wind conditions and the interplanetary magnetic field. Empirical models have been developed to describe this shock surface, such as Chao et al. (2002) who proposed an axi-symmetric formulation dependent on solar wind parameters, and Lu et al. (2019), who introduced a threedimensional asymmetric model that, in addition to solar wind parameters, incorporates the effect of the terrestrial dipole tilt. In this study, both models were evaluated using 2703 bow shock crossings observed by the MMS (Magnetospheric Multiscale) mission, in order to assess their ability to reproduce the shock shape and location. Although the Chao et al. (2002) model provides an excellent average representation of the bow shock, particularly in the dayside region, its axi-symmetric nature limits its ability to describe lateral and nightside asymmetries. In contrast, the Lu et al. (2019) model more accurately captures the observed three-dimensional asymmetries, exhibiting better physical consistency and improved global agreement with in situ observations.

Speaker: Ana Clara Garcia Ilha (Universidade de São Paulo)

72 The Role of Hiss Waves in Electron Precipitation over the South Atlantic Magnetic Anomaly

Energetic electron precipitation over the South Atlantic Magnetic Anomaly (SAMA) is strongly influenced by wave–particle interactions occurring in the inner magnetosphere, especially during disturbed geomagnetic conditions. These precipitation processes can modify the ionization of the upper atmosphere and affect atmospheric chemistry. In this work, a case study of the geomagnetic storm on 22 June 2015, associated with the arrival of an interplanetary coronal mass ejection (ICME), is presented to investigate the role of hiss waves in driving electron precipitation over the Brazilian sector. Measurements from the Van Allen Probes are used to examine wave activity and energetic electron dynamics in the inner radiation belt, while particle precipitation signatures are identified using observations from the Proba-V satellite during its passages across the SAMA region. The analysis focuses on electrons in the energy range of 500–600 keV. Results indicate enhanced precipitation during the storm main phase, accompanied by a significant expansion of the precipitation region to latitudes outside the typical SAMA boundaries. Enhanced plasmaspheric wave activity, particularly magnetospheric hiss and magnetosonic emissions, is observed simultaneously with the precipitation signatures. Preliminary analysis suggests that hiss waves play a major role in pitch-angle scattering of inner radiation belt electrons, contributing to the observed precipitation patterns. Ongoing investigations include resonance analyses between waves and particles, as well as the identification of the magnetospheric conditions favorable for hiss wave generation during ICME-driven storms.

Speaker: Karen Julia Coldebella Ferreira (National Institute for Space Research (INPE) / NASA Goddard Space Flight Center)

73 Towards Transparent Deep Learning Solar Image Processing

Introduction: In this study, we applied Explainable Artificial Intelligence (XAI) techniques (LIME and SHAP) to a Convolutional Neural Network (CNN) trained to classify Helioseismic and Magnetic Imager (HMI) magnetograms according to the presence/absence of an Hsx sunspot group. Our focus is on Solar Satellite Images (SSIs), where the data constitute a multimodal mixture of regular image collections and associated semi-structured Solar Flare textual information that provides critical context and labels.
Methodology: The dataset employed in this study was obtained from the Multiple Data-source Solar Extraction, Transformation, and Load (MS-ETL) pipeline, which aggregates raw HMI image collected by the Solar Dynamics Observatory (SDO), incorporating metadata such as timestamp, solar coordinates, active-region identifiers, and associated NOAA classifications. Each image corresponds to a line-of-sight HMI magnetogram with a mapped resolution of 224x224 pixels. MS-ETL performs several preprocessing steps (disk detection, limb darkening correction, coordinate referencing, and cropping) to produce standardized solar patches centered on active regions. These cleaned and uniformly formatted images serve as an ideal basis for DL workflows. In this work, we extracted all samples labeled under the presence or absence of the Hsx sunspot classification (canonical category of the McIntosh - Mount Wilson). The dataset is therefore divided into two classes: (a) no-Hsx: magnetograms without an Hsx-type sunspot group, and (b) with-Hsx: magnetograms containing an Hsx-class sunspot, characterized by a compact, well-formed magnetic core. To interpret the CNN’s predictions and assess whether it identifies physically meaningful solar features, two complementary XAI techniques were applied: LIME and SHAP. These methods provide local explanations for individual predictions, but differ significantly in methodology and interpretive power.
Results: We found that LIME identifies influential regions in close proximity to the localized physical structures of the sunspots, effectively isolating the umbra and penumbrae as the primary drivers of the classification. In contrast, SHAP attributions display a more global distribution, indicating that the neural network's decision-making process is also influenced by background distortions and the geometric boundaries of the solar limb.
Although both methods confirm the model's sensitivity to magnetic complexity, the disparity in their output suggests that LIME is more effective at capturing local morphological features, whereas SHAP uncovers the model's reliance on broader contextual and structural artifacts within the satellite imagery. These findings demonstrate that utilizing a multi-method XAI approach is relevant for identifying unintended model biases and for ensuring the transparency to DL methods.

Speaker: Marcela Xavier Ribeiro (UFSCar)

74 ULF wave coupling during geomagnetic storms in the South American sector: a multimission analysis

Ultra-Low Frequency (ULF) waves are fundamental drivers of energy transport within the coupled Magnetosphere-Ionosphere system. While the generation mechanisms for Pc5 (1.7-6.7 mHz) and Pc3 (22-100 mHz) pulsations are global, driven respectively by Kelvin-Helmholtz instabilities or internal resonances, and upstream ion-cyclotron waves, their transmission to the ground at low latitudes is significantly modified by regional electrodynamics. The South American sector acts as a unique natural laboratory for this coupling due to the South American Magnetic Anomaly (SAMA) and the Equatorial Electrojet (EEJ). This study presents an initial investigation into the Space-Ground coupling efficiency of ULF waves, testing the hypothesis that the South American ionosphere functions as an active electrodynamic filter. We employ a multi-point observation strategy combining in-situ magnetic field data from the Swarm constellation and Van Allen Probes with ground-based magnetograms from the Embrace/INPE network. Focusing on selected geomagnetic storm events in 2015, we analyze Swarm passes traversing low-latitude L-shells (L < 2) over the South American sector. This configuration allows for a direct comparison between wave signatures observed at specific L-shells in space and the response at their corresponding magnetic footprints on the ground. Preliminary results reveal a distinct North-South asymmetry, with wave packets in the SAMA region exhibiting significantly higher amplitudes compared to their conjugate footprints in the Northern hemisphere. Concurrently, ground-based analysis indicates a distinct amplification of Pc5 wave power within the SAMA region during the recovery phase. Specifically, we assess whether particle precipitation in the SAMA enhances local conductivity and Field Line Resonance (FLR) efficiency, and whether the EEJ modulates these modes. This work aims to establish a framework for quantifying the active role of South American electrodynamics in modulating global magnetospheric energy transfer.

Speaker: Jayne Alencar de Melo (Instituto Nacional de Geofísica Espacial)