Prediction of Major Solar Flares Using Interpretable Class-dependent Reward Framework with Active Region Magnetograms and Domain Knowledge

TL;DR

The paper tackles the challenge of predicting $\geq M$-class solar flares within $24\ \mathrm{h}$ under severe class imbalance by introducing a first-of-its-kind class-dependent reward (CDR) framework that integrates domain-informed features with reinforcement-learning-inspired training. It systematically compares DL models trained on SHARP LOS magnetograms and 39 knowledge-informed features, including three CDR variants, and demonstrates that knowledge-informed features (particularly when combining LOS and vector data) outperform magnetogram-based DL approaches; among models, the CDR-Transformer-10 achieves the best overall performance, with robustness to reward perturbations and clear interpretability via SHAP analyses. The study also benchmarks against NASA/CCMC models, showing superior performance on a filtered testing set, and discusses practical implications for operational space weather forecasting. Overall, the work contributes a novel, interpretable, and robust framework for solar flare prediction that leverages physical insights to address class imbalance and improve predictive accuracy in real-time settings.

Abstract

In this work, we develop, for the first time, a supervised classification framework with class-dependent rewards (CDR) to predict $\geq$MM flares within 24 hr. We construct multiple datasets, covering knowledge-informed features and line-of sight (LOS) magnetograms. We also apply three deep learning models (CNN, CNN-BiLSTM, and Transformer) and three CDR counterparts (CDR-CNN, CDR-CNN-BiLSTM, and CDR-Transformer). First, we analyze the importance of LOS magnetic field parameters with the Transformer, then compare its performance using LOS-only, vector-only, and combined magnetic field parameters. Second, we compare flare prediction performance based on CDR models versus deep learning counterparts. Third, we perform sensitivity analysis on reward engineering for CDR models. Fourth, we use the SHAP method for model interpretability. Finally, we conduct performance comparison between our models and NASA/CCMC. The main findings are: (1)Among LOS feature combinations, R_VALUE and AREA_ACR consistently yield the best results. (2)Transformer achieves better performance with combined LOS and vector magnetic field data than with either alone. (3)Models using knowledge-informed features outperform those using magnetograms. (4)While CNN and CNN-BiLSTM outperform their CDR counterparts on magnetograms, CDR-Transformer is slightly superior to its deep learning counterpart when using knowledge-informed features. Among all models, CDR-Transformer achieves the best performance. (5)The predictive performance of the CDR models is not overly sensitive to the reward choices.(6)Through SHAP analysis, the CDR model tends to regard TOTUSJH as more important, while the Transformer tends to prioritize R_VALUE more.(7)Under identical prediction time and active region (AR) number, the CDR-Transformer shows superior predictive capabilities compared to NASA/CCMC.

Figures (12)

• Figure 1: Training process of the CDR framework.
• Figure 2: The architecture of flare prediction models.
• Figure 3: Boxplots of the effect of reward variation on TSS and BSS of the CDR-Transformer-10 model. In the boxplots, the red horizontal line in each box represents the median value of TSS or BSS.
• Figure 4: A bar chart depicting the global significance of the 10 features for Transformer-10 on one testing set from 10 CV datasets.
• Figure 5: Beeswarm plot showing the effect of the 10 features on Transformer-10 for each AR.