Prediction of Magnetic Flux Evolution During Solar Active Region Emergence using Long Short-Term Memory Networks
E.Eren Dogan \orcid 0009-0002-9903-9134, S.Spiridon Kasapis \orcid 0000-0002-0972-8642, S.Sarang Patil \orcid 0009-0005-2657-1156, J.Jonas Tirona \orcid 0009-0007-3333-6685, J.John Stefan \orcid 0000-0002-5519-8291, I.Irina Kitiashvili \orcid 0000-0003-4144-2270, M.Mengjia Xu \orcid 0000-0002-3627-8261, A.Alexander Kosovichev \orcid 0000-0003-0364-4883 · 2026
Read Prediction of Magnetic Flux Evolution During Solar Active Region Emergence using Long Short-Term Memory Networks by E.Eren Dogan \orcid 0009-0002-9903-9134, S.Spiridon Kasapis \orcid 0000-0002-0972-8642, S.Sarang Patil \orcid 0009-0005-2657-1156, J.Jonas Tirona \orcid 0009-0007-3333-6685, J.John Stefan \orcid 0000-0002-5519-8291, I.Irina Kitiashvili \orcid 0000-0003-4144-2270, M.Mengjia Xu \orcid 0000-0002-3627-8261, A.Alexander Kosovichev \orcid 0000-0003-0364-4883 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-03507.txt.
Introduction
Dogan and colleagues build machine-learning models that predict magnetic flux changes during solar active region emergence from brightness and solar wave power maps. Their simpler MagFluxLSTM beats the more complex MagFluxEnc-Dec encoder-decoder: among the top 100 tuning settings, 98 percent use MagFluxLSTM. The simpler model predicts flux rise 3 to 10 hours ahead within a 12-hour window for five test regions.
Catalog record
| CREATOR | E.Eren Dogan \orcid 0009-0002-9903-9134, S.Spiridon Kasapis \orcid 0000-0002-0972-8642, S.Sarang Patil \orcid 0009-0005-2657-1156, J.Jonas Tirona \orcid 0009-0007-3333-6685, J.John Stefan \orcid 0000-0002-5519-8291, I.Irina Kitiashvili \orcid 0000-0003-4144-2270, M.Mengjia Xu \orcid 0000-0002-3627-8261, A.Alexander Kosovichev \orcid 0000-0003-0364-4883 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44924 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.03507
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Prediction of Magnetic Flux Evolution During Solar Active Region Emergence using Long Short-Term Memory Networks E.Eren Dogan \orcid 0009-0002-9903-9134, S.Spiridon Kasapis \orcid 0000-0002-0972-8642, S.Sarang Patil \orcid 0009-0005-2657-1156, J.Jonas Tirona \orcid 0009-0007-3333-6685, J.John Stefan \orcid 0000-0002-5519-8291, I.Irina Kitiashvili \orcid 0000-0003-4144-2270, M.Mengjia Xu \orcid 0000-0002-3627-8261, A.Alexander Kosovichev \orcid 0000-0003-0364-4883 Abstract Solar active regions (ARs) are the primary drivers of space weather events, making their early prediction crucial for operational forecasting systems. We develop machine learning models capable of predicting the evolution of magnetic flux during AR emergence using 1D time series of the continuum intensity and solar oscillation power maps for 53 active regions and their surrounding quiet-Sun areas. Each observable is sampled over a fixed 30.66 ∘ × 30.66 ∘ 30.66^{\circ}\times 30.66^{\circ} field of view. These observations capture the temporal evolution of each active region and serve as inputs for training and validation of our MagFluxLSTM and MagFluxEnc-Dec models. The MagFluxLSTM architecture implements a single-stage standard Long-Short Term Memory (LSTM) network. MagFluxEnc-Dec represents an LSTM encoder-decoder with teacher forcing. To test and evaluate the models’ performance, we use the continuum intensity and oscillation power maps (calculated for several frequency bands from Doppler velocity) as input to predict the magnetic flux. Among the top 100 hyperparameter configurations ranked by validation derivative RMSE, 98% correspond to MagFluxLSTM , compared to only 2% for MagFluxEnc-Dec . Thus, although the MagFluxEnc-Dec architecture has higher model complexity, it leads to poorer generalization to ARs outside the training set and less stable training than the simpler MagFluxLSTM , which can predict magnetic flux emergence 3–10 hours in advance within a 12-hour prediction window in both experimental and operational-type settings for the 5 testing active regions. keywords Solar Active Regions, Space Weather Forecasting, Magnetic Flux Emergence, Acoustic Power, Continuum Intensity, Quiet-Sun, LSTM Encoder-Decoder, MagFluxLSTM, Hyperparameter Optimization, Teacher Forcing 1 Introduction Solar active regions (ARs) are areas of intense and complex magnetic field concentration on the solar photosphere and are the primary source of space weather phenomena, including solar flares, coronal mass ejections, and solar energetic particle events ( 33 , e.g.,) . Therefore, early detection and prediction of AR emergence are critical for space weather forecasting, as they can potentially enhance warning capabilities for geomagnetic disturbances that affect Earth’s magnetosphere and impact technological infrastructure. Understanding subsurface dynamics and changes in turbulent motions that cause the stochastic excitation of solar oscillations at the photosphere, the power spectrum of which can be sensitive to subsurface perturbations caused by emerging AR ( 8 ; 36 ; 39 ; 19 ) . Such pre-emergence signatures can also be detected through H \alpha imaging ( 27 ) , surface flow analysis ( 7 ; 34 ) , and helioseismic travel-time variations ( 21 ; 22 ; 15 ; 37 ) . A number of helioseismic studies have confirmed that variations of photospheric disturbances could signal the upcoming emergence of magnetic flux before it can be identified visually ( 15 ; 14 ; 3 ; 23 ; 1 ; 37 , e.g.,) . Machine learning (ML) models for forecasting space weather events have recently emerged as a powerful approach for analyzing the growing volume of heliophysics data ( 5 ) . These methods have been successfully applied to a wide range of predictive tasks, including the prediction of solar flare onset ( 16 ; 6 , e.g,) , solar energetic particles ( 18 ; 13 , SEPs; e.g.,) , and coronal mass ejections ( 4 ; 9 , CMEs; e.g,) . One of the most recent advances in this domain is the SuryaFM model ( 30 ; 29 ) , a 366-million-parameter foundation model for heliophysics designed to learn general-purpose solar representations from multi-instrument SDO observations. In this light, deep learning architectures designed for sequential data offer promising avenues for capturing the temporal dependencies inherent in the emergence of active regions. Long Short-Term Memory (LSTM) networks–an advanced architecture of recurrent neural networks (RNNs)–have demonstrated good performance in time-series prediction tasks across diverse domains ( 10 ; 12 ; 35 ) , making them particularly suitable for modeling the complex temporal evolution of solar magnetic fields. 19 ; 20 employed LSTM architectures to predict the decrease in continuum intensity associated with the emergence of ARs before they were observed on the solar surface. The LSTM models were trained on Solar Dynamics Observatory/Helioseismic and Magnetic Imager ( 28 ; 32 , SDO/HMI;) observations processed into the 1D ML-ready dataset, the Solar Active Region Emergence Dataset ( 17 , SolARED;) . The LSTM models were designed to predict the onset of continuum intensity decrease associated with AR emergence within a 12-hour window. The results demonstrate that the LSTM-based analysis of oscillation power maps can successfully forecast AR emergence 5–29 hours before the associated continuum intensity decrease, at a stage when the magnetic flux has reached only 4%–9.6% of its eventual maximum. This study aims to test whether ML models can capture disturbances in solar acoustic power and continuum intensity to predict the magnetic flux emergence before it is visible on the solar photosphere. The presented deep learning framework for predicting the evolution of magnetic flux builds on the methodology of 20 and addresses three key challenges: a) capturing long-term temporal dependencies and pre-emergence signatures in acoustic power and continuum intensity observations, b) demonstrating the ability to predict magnetic flux given (a), and c) achieving operational magnetic flux forecast capabilities with sufficient lead time for space weather applications. The paper is organized as follows: Section 2 presents the methodology followed, such as the data preprocessing steps, the LSTM architectures used, the loss optimization, and the LSTM training procedures. Section 3 describes the prediction results on five testing ARs, including quantitative performance metrics and qualitative analysis of prediction capabilities. Finally, Section 4 provides an overview of the key findings, discusses the models’ limitations, and offers some concluding remarks. 2 Data and Methodology We provide detailed information about the AR emergence dataset (Section 2.1 ) the different LSTM architectures (Section 2.2 ), and the model training (Section 2.3 ) and implementation details (Section 2.4 ) in the following subsections. 2.1 Dataset In this study, we train and test the developed ML models on the compressed ML-ready data sets derived from preprocessed solar surface observations of 53 ARs, observed from March 2010 to June 2023 using the SDO/HMI ( 31 ; 11 ) instrument. The dataset’s 1D time-series represent the evolution of spatially averaged in 9\times 9 bins: a) continuum intensity I_{c} , b) unsigned magnetic flux \Phi_{m} , and c) M=4 frequency ranges of acoustic power P_{a} (i.e., 2–3 mHz, 3–4 mHz, 4–5 mHz, and 5–6 mHz), all at 1-hour cadence, covering the full emergence and evolution lifecycle of each AR. The acoustic power maps are derived from an 8-hour series of Dopplergrams with one-hour shifts, capturing the temporal evolution of subsurface dynamics throughout each AR’s lifecycle. A detailed description of the Solar Active Region Emergence Dataset (SolARED) is presented in 17 . The dataset is available through the SolARED interactive website at https://sun.njit.edu/sarportal/ . The SolARED dataset utilized in this study is already pre-normalized to account for the solar geometric effect. In the original dataset, the evolution of these observables was tracked and remapped onto the heliographic coordinates using