AI-FLARES: Artificial Intelligence for the Analysis of Solar Flares Data

TL;DR

AI-FLARES addresses the challenge of forecasting solar flares and interpreting their underlying physics by integrating data-driven AI with multi-modal solar data. The authors develop feature-based predictors from HMI magnetograms and a video-based Deep Learning pipeline (LRCN) for flare forecasting, introduce EUV desaturation via SE-DESAT to recover saturated AIA images, and build imaging-spectroscopy reconstructions for hard X-ray visibilities using maximum-entropy, PSO-based forward fits, and CLEAN-like methods. A regularized electron-map framework links particle acceleration to observable photon signatures, enabling quantitative insights into energy release during flares. The work demonstrates improved forecasting capabilities and physically interpretable reconstructions, contributing to space weather prediction and solar-physics diagnostics, supported by public code and ongoing physics-aware methodological developments.

Abstract

AI-FLARES (Artificial Intelligence for the Analysis of Solar Flares Data) is a research project funded by the Agenzia Spaziale Italiana and by the Istituto Nazionale di Astrofisica within the framework of the ``Attività di Studio per la Comunità Scientifica Nazionale Sole, Sistema Solare ed Esopianeti'' program. The topic addressed by this project was the development and use of computational methods for the analysis of remote sensing space data associated to solar flare emission. This paper overviews the main results obtained by the project, with specific focus on solar flare forecasting, reconstruction of morphologies of the flaring sources, and interpretation of acceleration mechanisms triggered by solar flares.

Figures (5)

• Figure 1: Feature-based machine learning for flare forecasting. Top left panel: very few descriptors (x-axis) are sufficient to achieve high values of the True Skill Statistic (TSS) score (y-axis) with two machine learning methods (hybrid LASSO, HLA and Random Forest, RF). Bottom panel: feature-ranking applied to machine learning outcomes show that the Ising energy significantly increases its rank when the AR producing the flaring event is included in the training set. Top right panel: the TSS notably increases when the topological feature introduced in 2021ApJ...915...38C is added as first descriptor (red solid line), with respect to the case wehm the feature is not used in the training set (blue dashed line).
• Figure 2: Deep learing for flare forecasting. Left panel: the AI-FLARES neural network is made of a Long Short-Term Memory (LSTM) network fed by the outcomes of several Convolutional Neural Networks (CNNs). Right panels: the rates of true positives and true negatives are significantly high thanks to the use of video-based deep learning.
• Figure 3: Desaturation of EUV maps. Left panel: an intense solar flare saturates an extended region of an image recorded by SDO/AIA. Right panel: AI-FLARES desaturation algorithm is able to restore the signal in the core of the flaring source.
• Figure 4: Image reconstruction from hard X-ray visibilities. A constrained maximum entropy method, a forward-fit algorithm based on Particle Swarm Optimization (PSO), and an automated version of CLEAN deconvolution metod provided the reconstructions in the left, middle, and right panels, respectively.
• Figure 5: Regularized imaging spectroscopy from hard X-ray visibilities. These panels represent electron flux maps at different electron energies (regularization introduced a smoothing constraint across the energy direction.)