Space weather describes the dynamic conditions in near-Earth space, mostly driven by the variable interaction between the continuous flow of the solar wind and the Earth’s magnetic field.  Extreme space weather has the potential to disrupt or damage key infrastructure on which we rely, for example through the generation of large, anomalous Geomagnetically Induced Currents (GICs) in power networks and transformers.  Accurately forecasting a risk of large GICs would enable key actions to be taken to mitigate their impact.

Given the sparsity of direct GIC measurements, and their inherent specificity to the contemporaneous network properties and configuration, we turn to forecasting the driving factor: the changing ground magnetic field (R).  In this talk we discuss a recent model developed to forecast whether the rate of change of the ground magnetic field (R) will exceed specific, high thresholds in the United Kingdom.  The model uses a common space weather forecasting framework: an interval of data from the upstream solar wind is used to make a prediction as to future conditions at the Earth.  We will use this model as an example to discuss forecasting performance, particularly with respect to different magnetospheric driving and processes.  We demonstrate the use of techniques such as SHAP (Shapley Additive exPlanations) to investigate how and why the model is making the predictions that it does.  What physical processes can this model set up capture?  Where do we need to go in the future?

How to cite: Smith, A., Rae, J., Forsyth, C., Coxon, J., Walach, M.-T., Lao, C., Bloomfield, S., Reddy, S., Coughlan, M., Keesee, A., and Bentley, S.: Space Weather Forecasts of Ground Level Space Weather with Machine Learning: Performance, Limitations and Challenges, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9149, https://doi.org/10.5194/egusphere-egu25-9149, 2025.

The necessity to limit budget, size, weight and power consumption of the instruments placed on board space mission satellites results in several drawbacks, including the exclusion of dedicated instrumentation for the monitoring of the spacecraft environment. Understanding the environmental conditions of space missions is essential to correctly analyse their observations. Seldom the necessary interplanetary parameters, not measured in situ, can be gathered from nearby dedicated missions, however this is not always feasible. Other solutions envisage the application of machine learning models to estimate the missing parameters on the basis of those that are available on board the satellites. Despite the high performance of machine learning predictors, they come along with issues related to the model selection and training, the data pre-processing and the opaqueness of the outcomes returned to end-users. The application of tools developed in the explainable artificial intelligence (XAI) field can be considered to encode through symbolic knowledge the functional relationship between parameters observed in situ and correlated parameters for which measurements are lacking but useful. In this context, XAI methods in general, and symbolic knowledge extraction in particular, constitute a promising alternative to traditional machine learning models, enabling users to avoid the model selection and training phases and to obtain completely interpretable results. This presentation provides an overview on the application of symbolic knowledge-extraction techniques to perform rule induction from available in-situ data, aimed at carrying out a human-interpretable estimation and forecasting of missing platform parameters. Potentialities, drawbacks and challenges of this approach are discussed to highlight the direction from current results to future applications.

How to cite: Sabbatini, F. and Grimani, C.: Missing Interplanetary Data Estimation for Space Missions via Symbolic Rule Induction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7486, https://doi.org/10.5194/egusphere-egu25-7486, 2025.

Many machine learning models have provided significant results in predicting the geomagnetic activity quantified by Earth-measured geomagnetic indices. For instance, one such model is the SERENADE model that provides probabilistic forecasts of the Kp index, days ahead solely from solar imaging. It consists of three modules combining convolutional, recurrent, and linear neural network layers that first extract the important information contained in the input solar imagery and transform them into an intelligible forecast. To improve the performance of this model, we evaluate solar-imaging-adapted dimensionality reduction techniques that extract the features from the images and can therefore be used as the first layer of the forecast model. We use a solar imagery dataset formatted specifically for machine-learning research (SDOML). We applied the Principal Component Analysis method and trained AutoEncoders and Variational AutoEncoders (VAE) targeting several reduced dimensions. We consider the convolutional GoogLeNet method, which was pre-trained on the ImageNet dataset, as a baseline for our comparison. We analyze the information retained by the extracted features in terms of solar activity physical parameters and find high correlations between the latter and the the reduced representations of the images, with the VAE results standing out. In addition, we re-train the SERENADE model to predict the daily maximum of the Kp index two days in advance using the extracted features by the new dimensionality reduction methods as input to the model. We first use the same hyperparameters that were optimized for the GoogLeNet model and obtain more stable predictions using the dedicated solar imaging feature extractors than when using the baseline model, specifically in the VAE case. Furthermore, when fine-tuning SERENADE's hyperparameters to the VAE model, the predictive performance of the model was enhanced, notably during geomagnetic storms, which indicates that the use of adapted feature extractors could improve the geomagnetic activity forecasting.

How to cite: Tahtouh, M., Bernoux, G., and Brunet, A.: Evaluating Solar Imaging Feature Extraction Techniques for Enhancing Space Weather Prediction with Deep Learning Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8698, https://doi.org/10.5194/egusphere-egu25-8698, 2025.

Solar flares, sudden bursts of electromagnetic energy originating from magnetically active regions on the solar surface, pose significant risks to satellite infrastructure, communication systems, and power grids. Accurate forecasting of these events is crucial for advancing space weather prediction and safeguarding technological infrastructure. The interconnected nature of the Sun's atmospheric layers—from the corona to the lower photosphere—highlights the need for comprehensive data analysis techniques that leverage modern advancements in machine learning (ML) and physically informed models.

Traditional approaches have relied on features extracted from line-of-sight (LoS) magnetograms of solar active regions, historically linked to increased flare activity. However, recent studies employing LoS magnetogram time series have shown limited improvements, prompting the need for novel methodologies that integrate learning-based and physics-based insights.

To address this challenge, we present a deep learning-based framework for solar flare forecasting, leveraging the Solar Dynamics Observatory’s Helioseismic and Magnetic Imager (SDO/HMI) LoS magnetograms. Our model frames flare forecasting as a binary time series classification problem, aiming to distinguish active regions likely to produce M- or X-class flares within a 24-hour window. The approach integrates a Convolutional Neural Network (CNN) autoencoder for feature extraction and a Long Short-Term Memory (LSTM) binary classifier for flare activity prediction, achieving a 90% test accuracy.

By leveraging advanced ML techniques, this methodology demonstrates the potential of data-driven models in heliophysics. Our results highlight the transformative role of AI-powered science in advancing solar flare prediction and contributing to the development of reliable early warning systems for space weather forecasting.

How to cite: Doria Rosales, E., Carbone, P. V., Falanga, P. M., Ciaramella, P. A., and Di Nardo, PhD. E.: Machine Learning for Space Weather: Solar Flare Forecasting Using SDO/HMI Magnetogram Time Series, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9116, https://doi.org/10.5194/egusphere-egu25-9116, 2025.

How to cite: Stepanyuk, O., Pötzi, W., Kozarev, K., Dechev, M., and Miteva, R.: Hybrid AI Approaches for Solar Feature Recognition Using Ground-Based Instrument Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12654, https://doi.org/10.5194/egusphere-egu25-12654, 2025.

The rapid growth of high-dimensional data in solar physics presents significant challenges for analysis and interpretation, making it an excellent domain for the application of machine learning (ML) algorithms. Synoptic full-disk observations with the Solar Dynamics Observatory (SDO)  provide continuous observations of the solar magnetic activity over more than one solar cycle, facilitating the study of solar variability and space weather impacts. The Space-weather HMI Active Region Patches (SHARP) vector magnetic field (VMF) maps and parameters, based on Helioseismic and Magnetic Imager (HMI) full-disk observations, are developed to study the magnetic evolution of individual active regions and flare triggering mechanisms. We present a method for active region parametrization by combining empirical parameters and ML-extracted features. Time series of SHARP VMF maps are used as input for the Disentangled Variational Autoencoder (VAE), a Disentangled Representation Learning (DRL) algorithm that facilitates the extraction of a low-dimensional feature representation. The VAE model is used to encode generalized information about nonlinear dynamical systems, i.e., a solar active region, aiming to isolate distinct factors of variation in the data, allowing a clearer interpretation of physical processes. We demonstrate how the ML features can be used to identify and study the stages of the magnetic patches evolution. These are benchmarked with SHARP parameters, relating empirical and learned features. Furthermore, the empirical dataset enhanced with ML features can be used to analyze the development of individual active regions and searching for eruption precursors.

How to cite: Dineva, E., Miloshevich, G., Lapenta, G., Magdalenic Zhukov, J., and Poedts, S.: Parametrization of SHARP Vector Magnetic Field Using Disentangled Representation Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11565, https://doi.org/10.5194/egusphere-egu25-11565, 2025.

The ionosphere has important impacts on many different systems, such as communications, thus modeling it is a crucial task. The influence of the ionosphere is closely linked to its electron density, but this is difficult to measure adequately. Because of this, modeling requires the use of additional correlated values, such as solar activity metrics. These measures do not capture enough to reproduce small-scale changes in electron density, so we have developed a technique to expand our input space to include sparse measurements of Total Electron Content (TEC), or the integral of electron density.

TEC data is measured more densely than electron density, although it is still not consistent spatially, with many gaps in measurement coverage. Despite this, it is collected very consistently throughout time so it presents itself as a good candidate for an input to an ionospheric model. Even so, TEC has not been used as an input to such models, especially Machine Learning (ML) models, as the irregular coverage of the measurements makes it difficult to deal with.

We have developed a technique to use transformer-like architectures to move from an irregular domain to a fixed size embedded domain to facilitate further usage of the TEC data. This approach has enabled us to use TEC as a direct input to electron density models, noticeably improving performance. Our technique also enables the use of a variety of irregular inputs all at once, enabling a wider range of possible model inputs. Lastly, as a byproduct of the process, we can use the inverse of our embedding technique (which is also how we train the model) to perform TEC map completion, where we can predict TEC values even where no measurements have been taken.

How to cite: Smith, L. and Cohen, M.: Using Transformers to Integrate Irregular Data for Improved Ionospheric Modeling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14036, https://doi.org/10.5194/egusphere-egu25-14036, 2025.

Magnetic reconnection on the dayside magnetopause is considered the primary mechanism for transporting mass, momentum, and energy from the solar wind into the terrestrial magnetosphere. Several studies have demonstrated that the spatiotemporal dynamics of the dayside magnetic reconnection can be inferred remotely from the analysis of the time-energy dispersion of ions in the Earth’s cusps. Despite the immense number of in-situ cusp measurements acquired by numerous space-based instruments, it is still challenging to determine the overall cusp behavior owing to the intermittency of the measurement acquisition. To overcome this issue, this work implements a regression model of the three-dimensional (3-D) ion flux in the Earth’s Northern cusp based on deep learning techniques and numerous measurements of the cusp under varying solar wind conditions. For the training process, we have used solar wind parameters obtained from NASA's OMNI database as input and in-situ ion flux measurements acquired by the CIS/HIA instruments on board ESA’s multi-spacecraft Cluster mission during the period from 2001 to 2010 for supervised output. The model allows the reconstruction of the time-dependent, 3-D ion flux distribution within the cusp region, which serves to determine the boundaries of the high-altitude cusp, analyze its structural response to time-dependent solar wind conditions, and investigate the relationship between the cusp and dayside magnetic reconnection. The experiments under controlled input parameters show that our model is capable of reproducing  expected ion dispersion signatures as a response to variable solar wind conditions.

How to cite: Cucho-Padin, G., Sibeck, D., Da Silva, D., and Wang, X.: A Deep-learning-based Model of the Three-dimensional Ion Flux in the Earth’s Northern Cusp, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20652, https://doi.org/10.5194/egusphere-egu25-20652, 2025.

In recent years, the field of space situational awareness (SSA) has gained increasing attention, driven by the rapid rise in both active satellites and orbital debris. Therefore, being able to predict the orbit of a resident space object (RSO) as accurately as possible is more critical than ever in order to reduce collision risks and to preserve the orbital environment. However, incomplete knowledge of debris geometry, uncertain object characteristics, or simplified force models can cause prediction errors which exceed orders of several kilometers within just a few days, making it useless for reliable collision avoidance operations. Using modern Machine Learning (ML) algorithms can enhance prediction accuracy by addressing these challenges as recent studies have shown. In this context we present Artificial Intelligence for Precise Orbit Determination (AI4POD), a Python package that is designed to simplify the integration of ML-algorithms within the orbit prediction and determination process. AI4POD is structured as a comprehensive toolbox that includes a high-fidelity force model, various measurement functions, and classical orbit determination (OD) algorithms such as the batch least-squares estimation method. This integrated approach allows users to combine traditional orbit simulations with data-driven approaches to improve accuracy and to extend the predictability horizon. Based on this catalog, several approaches from artificial intelligence (AI) shall be tested in the future. Inspired by already proposed methodologies we are generating a training set of historical tracking data along with their corresponding orbit determinations using the AI4POD toolbox. Several machine learning algorithms will be explored to learn the nonlinear prediction errors, aiming to compensate for unmodeled or uncertain factors such as incomplete knowledge of satellite geometry or environmental conditions.

How to cite: Aigner, B., Dallinger, F., Andert, T., Haser, B., Pätzold, M., and Hahn, M.: AI-Enhanced Orbit Determination: The AI4POD Framework, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11790, https://doi.org/10.5194/egusphere-egu25-11790, 2025.

Artificial intelligence (AI), particularly machine learning (ML), is widely applied in fields such as medicine, autonomous driving, and manufacturing. Over time, ML has also seen increasing use in space and geosciences, where its algorithms hold the potential to enhance orbit prediction and orbit determination (OD) by utilizing measurement data. However, ML models like Artificial Neural Networks (ANNs) are limited to problems with abundant data and are often considered "black boxes", as their predictions lack interpretability in a scientifically meaningful way. To address these challenges, Raissi et al. 2018 introduced Physics Informed Neural Networks (PINNs), a specialized type of ANN. PINNs integrate the governing differential equations of a system into the learning process, imposing a physical constraint on the network's training and predictions. This approach allows effective training with small datasets, removing the reliance on large amounts of measurements. Additionally, PINNs can estimate unknown or poorly defined parameters within the differential equations, making them conceptually similar to classical OD algorithms like the Weighted Least Squares method. Building on this, Scorsoglio et al. 2023 successfully applied a variant of PINNs, called Physics Informed Extreme Learning Machines (PIELMs), for OD. In this study, a similar approach is employed for OD within the AI4POD (Artificial Intelligence for Precise Orbit Determination) software tool, focusing on resident space objects (RSOs) in low Earth orbit. Following this, we explore various methods, such as output perturbation, to determine the covariance matrix for the PINN-based OD approach. The covariance matrix provides an assessment of uncertainty in the predicted orbit and therefore being an essential tool in real space missions and collision avoidance. These methods are compared for their realism and effectiveness, both against each other and against the covariance matrix results from classical approaches. This study aims to evaluate whether the proposed methods can replicate and potentially improve upon traditional covariance estimation techniques.

How to cite: Dallinger, F., Aigner, B., Andert, T., Haser, B., Pätzold, M., and Hahn, M.: On Covariance Estimation in Physics Informed Neural Networks for Orbit Determination, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11680, https://doi.org/10.5194/egusphere-egu25-11680, 2025.

The solar wind, a stream of charged particles originating from the Sun, poses significant risks to technology and astronauts. It is driven by large structures on the solar surface like coronal holes and active regions, which can be identified in extreme ultra-violet (EUV) solar images several days before they become geoeffective. In this work, we propose to use a distributional regression algorithm to forecast the solar wind speed at the Lagrange 1 point from solar images. Instead of predicting a single value, this method models the entire conditional distribution as a function of input features. It allows computing the uncertainty of predictions and specifying the probability of the solar wind speed exceeding certain thresholds, which is especially useful for extreme event predictions like coronal mass ejections and high-speed solar wind streams. We employ a convolutional neural network to encode solar images from multiple wavelength channels into unstructured low-dimensional representations. Using a semi-structured distributional regression approach, we couple the deep learning encoder with structured physical input parameters, such as past solar wind properties and solar cycle information. Thereby, we incorporate physical knowledge into the model and enhance explainability. We predict the solar wind speed distributions with a one-hour cadence four days in advance. We train and evaluate our method using cross-validation on 15 years of data and compare it to current state-of-the-art models. We find that it provides an accurate forecast and especially models the heavy-tailed solar wind speed distribution well. We further show the advantages over standard regression approaches and how to use the predicted conditional quantiles to improve extreme event predictions, highlighting the potential for operational space weather forecasts.

How to cite: Collin, D., Shprits, Y., Hofmeister, S., Bianco, S., Klein, N., and Gallego, G.: Solar Wind Speed Forecasting From Solar Images Using Distributional Regression , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11587, https://doi.org/10.5194/egusphere-egu25-11587, 2025.

Solar activity significantly influences the near-Earth environment, leading to magnetic storms and magnetospheric substorms that can impact both technological and human systems. Understanding the physical processes that govern the Sun-Earth relationship and developing models to forecast magnetic disturbances on Earth are therefore of critical importance. In this context, we present a preliminary work to model and forecast the dynamics of magnetic storms, as measured by the SYM-H geomagnetic index, using Physics-Informed Neural Networks (PINNs). This approach is applied to models based on deterministic ordinary differential equations (ODEs), such as those described by Burton et al. (1975) and others, which were proposed to describe the evolution of geomagnetic indices during magnetic storms. The findings and significance of this approach are discussed in the context of Earth's magnetospheric dynamics and the relevance of PINN techniques in space weather research.

How to cite: Lacal, M., Camporeale, E., Consolini, G., and Piersanti, M.: Modeling Magnetic Storms' Dynamics with Physics-Informed Neural Networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18050, https://doi.org/10.5194/egusphere-egu25-18050, 2025.

The identification and characterization of the coronal mass ejections (CMEs) and fast solar wind flows in the in situ data are important for understanding dynamics of these phenomena and consequently for space weather forecasting. In this study, we apply Self-Organizing Maps (SOMs) and clustering techniques to analyze in situ solar wind observations. SOMs (Kohonen, T, 1982) [1] an unsupervised learning technique, is employed to project high-dimensional interplanetary plasma parameters such as velocity, density, temperature, and magnetic field onto a lower-dimensional representation, preserving the topological structure of the data. Clustering algorithms, such as k-means, are then applied to the SOM output to distinguish between ICME events, fast and slow solar wind flows.
Our approach is validated using a few months long interval of the ACE and Wind in situ observations, with labeled CME intervals from Richardson and Cane [2] as a benchmark. This combination of SOMs and clustering provides a framework for automated identification of interplanetary plasma structures, important for space weather studies but also for operational services. 

[1] T. Kohonen, ‘Self-organized formation of topologically correct feature maps’, Biol. Cybern., vol. 43, no. 1, pp. 59–69, Jan. 1982, doi: 10.1007/BF00337288
[2] Richardson, Ian; Cane, Hilary, 2024, "Near-Earth Interplanetary Coronal Mass Ejections Since January 1996"https://doi.org/10.7910/DVN/C2MHTH

How to cite: Carella, F., Magdalenić, J., and Bemporad, A.: Identification of fast solar wind flows and CMEs in the in situ data using Self-Organizing Maps and clustering techniques, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13602, https://doi.org/10.5194/egusphere-egu25-13602, 2025.

Segmentation and characterization of solar coronal structures are essential for advancing our understanding of solar atmosphere and accurately identifying key regions, such as active regions and coronal holes, which are precursors to phenomena like solar flares and coronal mass ejections (CMEs). In this study, we investigate two complementary approaches to automate this process. First, we employ a previously presented deep learning-based U-Net architecture tailored for segmenting and characterizing solar coronal structures. Second, we develop a lightweight algorithm aimed at optimizing resource efficiency, consisting of classical computer vision techniques, which include thresholding and morphological filtering. The approach that best balances segmentation performance and computational efficiency will be selected for integration into a prototype designed to support future space exploration missions.

To characterize the segmented regions, we propose a set of carefully designed hand-crafted features to represent and characterize the resulting segmentations. These representations are analyzed using unsupervised clustering techniques, such as K-means and t-SNE, to distinguish solar coronal structures, including active regions, coronal holes and bright points.

Our dataset spans multiple layers of the solar atmosphere, incorporating HMI magnetograms (photosphere) and AIA wavelengths—94 Å (flaring regions), 171 Å (quiet Sun), 193 Å (coronal structures), and 304 Å (chromosphere). The performance of both segmentation approaches is thoroughly evaluated using metrics such as Dice score and Intersection over Union (IoU), with comparisons made against state-of-the-art methods.

Future work will focus on developing feature encoding techniques to better understand and predict solar phenomena, such as solar flare emissions, while investigating the impact of different feature extraction strategies on model performance.

References:

• Galvez, Richard, et al. "A machine-learning data set prepared from the NASA solar dynamics observatory mission." The Astrophysical Journal Supplement Series 242.1 (2019): 7.
• Šimon Mackovjak et al. “SCSS-Net: solar corona structures segmentation by deep learning”, Monthly Notices of the Royal Astronomical Society, Volume 508, Issue 3, December 2021, Pages 3111–3124, https://doi.org/10.1093/mnras/stab2536
• Gonidakis, Panagiotis & Sóñora-Mengana, Alexander & Jansen, Bart & Vandemeulebroucke, Jef. (2023). Handcrafted Features Can Boost Performance and Data-Efficiency for Deep Detection of Lung Nodules From CT Imaging. IEEE Access. PP. 1-1. 10.1109/ACCESS.2023.3331315. 

How to cite: Gonidakis, P., Carella, F., Miloshevich, G., and Poedts, S.: Efficient Segmentation and Clustering of Solar Coronal Structures: A Comparison of U-Net and Classical Computer Vision Techniques Using SDO Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9849, https://doi.org/10.5194/egusphere-egu25-9849, 2025.

In this study, we develop an artificial intelligence (AI)-based solar surface flux transport (SFT) model. We predict global magnetic field distributions on the solar surface up to the next solar rotation (27.3 days) using deep-learning. Here we train and evaluate our deep-learning model, based on the Pix2PixCC architecture, using data sets of SDO/HMI, SOHO/MDI, and NSO/GONG synoptic maps with a resolution of 360 by 180 (longitude and sine-latitude) from 1996 to 2023. We present results of our model and compare them with those from the persistence model and the conventional SFT model, including the effects of differential rotation, meridional flow, and diffusion on the solar surface. Our AI-based SFT model generates magnetic field distributions for the next solar rotation, better than the conventional SFT model and the persistence model in the quantitative metrics such as RMSE, FSIM, and pixel-to-pixel CC. Our model successfully generates magnetic features, such as the diffusion of solar active regions and the motions of supergranules. Our model also generates small-scale magnetic features better than the conventional SFT models. Using synthetic input data with bipolar structures, we confirm that our model successfully reproduces differential rotation and meridional flow. Finally, we discuss the advantages and limitations of our model in view of magnetic field evolution and its potential applications.

How to cite: Jeong, H.-J., Jeon, M., Kim, D., Kim, Y., Baek, J.-H., Moon, Y.-J., and choi, S.: Prediction of Solar Surface Magnetic Fields Using an AI-based Surface Flux Transport Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18475, https://doi.org/10.5194/egusphere-egu25-18475, 2025.

Remote sensing observations, whether astronomical or within the solar system, are constrained by instrumental limitations, such as the point spread function in imaging. Ensuring the reliability of scientific analysis from such data requires robust deconvolution techniques. We present a spatio-temporal deconvolution method, to minimise the effect of an extended or complex-shaped point spread function, applicable to dynamic systems with various timescales. This approach enhances observational data by improving image contrast and resolving small-scale dynamic features.

Our method employs a deep neural network trained on state-of-the-art numerical simulations, enabling it to identify dynamic patterns in both spatial and temporal dimensions and to estimate and correct the degradation of intensity contrast. The resulting improvements in intensity representation and resolution facilitate more accurate analyses of small-scale features.

We apply this methodology to solar observations in the millimeter wavelength regime, recovering fine-scale structures critical for understanding the complex behaviour of the solar atmosphere, predict the generation of potentially harmful events, solar flares and the solar wind. By incorporating the temporal domain, our approach surpasses traditional 2D deconvolution techniques.

While initially developed for solar imaging, the method is versatile and can be adapted to various observational contexts across different wavelength regimes. This makes it a valuable tool for advancing future observational studies and expanding research capabilities.

How to cite: Eklund, H.: Spatio-temporal deconvolution method for enhanced image analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17531, https://doi.org/10.5194/egusphere-egu25-17531, 2025.