Organizations: School of Computer Science, Jiangsu University of Science and Technology, Zhenjiang 212100, People′s Republic of China · State Key Laboratory of Space Weather, Chinese Academy of Sciences, Beijing 100190, People′s Republic of China · School of software, Southeast University, Nanjing, People′s Republic of China · Yunnan Astronomical Observatory, Chinese Academy of Sciences, Kunming 650216, People′s Republic of China · Radio Cosmology Lab, Centre for Astronomy and Astrophysics Research, Department of Physics, Faculty of Science, Universiti Malaya, 50603 Kuala Lumpur, Malaysia · National Centre for Particle Physics, Universiti Malaya, 50603 Kuala Lumpur, Malaysia · Department of Electrical Engineering, Faculty of Engineering, Universiti Malaya, 50603 Kuala Lumpur, Malaysia
In this work, we develop, for the first time, a supervised classification framework with class-dependent rewards (CDR) to predict ≥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 & tables
Group name of knowledge-informed features
Keyword
Description
Reference
R_VALUE & AREA_ACR
R_VALUE
Sum of flux near polarity inversion line
Schrijver (2007)
AREA_ACR
Area of strong field pixels in the AR
Bobra & Couvidat (2015)
Gradient features
Gradient mean
To condense the gradient information into single descriptors (features) for each image, the gradient features are computed as the (1) mean, (2) standard deviation, (3) maximum, (4) minimum, (5) median, (6) skewness, and (7) kurtosis of the gradients in each image.
Gradient std
Gradient median
Gradient min
Table 1: 31 knowledge-informed features calculated from SHARP LOS magnetogram images.
Dataset
Training(No-flare/C/M/X)
Validation(No-flare/C/M/X)
Testing(No-flare/C/M/X)
No.1
223/168/46/8
70/37/11/1
58/33/16/4
No.2
223/168/46/8
64/38/12/1
63/37/16/3
No.3
223/168/46/8
67/37/10/1
59/30/14/4
No.4
223/168/46/8
66/37/8/2
53/29/17/5
No.5
223/168/46/8
62/37/10/2
56/30/13/4
No.6
223/168/46/8
64/33/9/1
50/30/17/3
Table 3: The distribution of the number of ARs from the 10 CV datasets used in this study, including magnetogram images and knowledge-informed features. The “No-flare /C/M/X” in the table represents the number of ARs corresponding to their respective categories among the training, validation, and testing sets.
Parameter name
CNN
CNN-BiLSTM
Transformer
Batch size
16
16
10
Epochs
120
120
150
Loss function
cross-entropy
cross-entropy
cross-entropy
Optimizer
SGD
SGD
Adam
Learning rate
0.01
0.01
0.0001
Table 4: The parameter configuration of deep learning models.
Feature
Model
Metric
TSS
BSS
The combination of LOS and vector magnetic field parameters
Transformer
0.814±0.029
0.452±0.095
8 vector magnetic field parameters
0.719±0.090
0.403±0.098
2 LOS magnetic field parameters
0.697±0.137
0.448±0.079
Table 7: The TSS and BSS scores for ≥ M-class flare prediction using the Transformer on the testing set with three distinct combinations of magnetic field parameters.
Models
p-value
TSS
BSS
CDR-Transformer-10 and Transformer-10
0.262
0.094
CDR-Transformer-2 and Transformer-2
0.366
0.431
Table 9: The paired t-test results between the CDR-Transformer and Transformer models.
TP reward
TN reward
FP reward
FN reward
TSS
BSS
5
4
-20
-15
0.803±0.048
0.461±0.172
6
4
-20
-15
0.811±0.062
0.467±0.160
7
4
-20
-15
0.800±0.064
0.475±0.148
8
4
-20
-15
0.785±0.076
0.481±0.136
9
4
-20
-15
0.801±0.082
0.482±0.136
10
4
-20
-15
0.829±0.059
0.489±0.120
Table 10: The effects of applying a perturbation (step size=1) to the TP reward on the TSS and BSS scores of the CDR-Transformer-10 model.
TP reward
TN reward
FP reward
FN reward
TSS
BSS
10
1
-20
-15
0.797±0.067
0.492±0.119
10
2
-20
-15
0.826±0.061
0.492±0.117
10
3
-20
-15
0.815±0.062
0.489±0.119
10
4
-20
-15
0.829±0.059
0.489±0.120
10
5
-20
-15
0.826±0.063
0.488±0.121
10
6
-20
-15
0.821±0.070
0.486±0.123
Table 11: The effects of applying a perturbation (step size=1) to the TN reward on the TSS and BSS scores of the CDR-Transformer-10 model.
TP reward
TN reward
FP reward
FN reward
TSS
BSS
10
4
-15
-15
0.807±0.060
0.497±0.106
10
4
-16
-15
0.813±0.061
0.492±0.101
10
4
-17
-15
0.820±0.065
0.495±0.112
10
4
-18
-15
0.825±0.066
0.494±0.110
10
4
-19
-15
0.819±0.072
0.491±0.114
10
4
-20
-15
0.829±0.059
0.489±0.120
Table 12: The effects of applying a perturbation (step size=1) to the FP reward on the TSS and BSS scores of the CDR-Transformer-10 model.
TP reward
TN reward
FP reward
FN reward
TSS
BSS
10
4
-20
-10
0.801±0.072
0.498±0.096
10
4
-20
-11
0.819±0.051
0.483±0.122
10
4
-20
-12
0.807±0.056
0.493±0.110
10
4
-20
-13
0.813±0.061
0.485±0.107
10
4
-20
-14
0.813±0.070
0.486±0.113
10
4
-20
-15
0.829±0.059
0.489±0.120
Table 13: The effects of applying a perturbation (step size=1) to the FN reward on the TSS and BSS scores of the CDR-Transformer-10 model.
Data
Model
TSS
Recall
FPR
Filtered testing dataset
NASA/CCMC
0.853
1
0.147
iTransformer
0.906 ± 0.043
1.000 ± 0.000
0.094 ± 0.043
Transformer
0.933 ± 0.057
0.980 ± 0.060
0.047 ± 0.020
CDR-Transformer
0.968 ± 0.016
1.000 ± 0.000
0.032 ± 0.016
Original testing dataset
NASA/CCMC
0.633
0.706
0.072
iTransformer
0.668 ± 0.060
1.000 ± 0.000
0.332 ± 0.060
Table 14: The performance results for CDR-Transformer-10, Transformer-10, iTransformer, and NASA/CCMC at the probability thresholds corresponding to their respective optimal TSS.
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.
Figure 6: The waterfall plot for the correct prediction of a positive class on Transformer-10 for AR12257. The AR12257 produced an M-class flare at 04:13 UTC on January 13, 2015.
Figure 7: The waterfall plot for the correct prediction of a negative class on Transformer-10 for AR12367. The AR12367 produced a C-class flare at 19:07 UTC on June 20, 2015.
Solar eruptions, including flares and coronal mass ejections (CMEs), have a significant impact on Earth. Some flares are associated with CMEs, and some flares are not. The association between flares and CMEs is not always obvious. In this study, we propose a new deep learning method, specifically a hybrid neural network (HNN) that combines a vision transformer with long short-term memory, to predict associations between flares and CMEs. HNN finds spatio-temporal patterns in the time series of line-of-sight magnetograms of solar active regions (ARs) collected by the Helioseismic and Magnetic Imager on board the Solar Dynamics Observatory and uses the patterns to predict whether a flare projected to occur within the next 24 hours will be eruptive (i.e., CME-associated) or confined (i.e., not CME-associated). Our experimental results demonstrate the good performance of the HNN method. Furthermore, the results show that magnetic flux cancellation in polarity inversion line regions may well play a role in triggering flare-associated CMEs, a finding consistent with literature.
Jialiang Li, Vasyl Yurchyshyn, Jason T. L. Wang +6
Institute for Space Weather Sciences, New Jersey Institute of Technology, University Heights, Newark, NJ 07102, USA · Department of Computer Science, New Jersey Institute of Technology, University Heights, Newark, NJ 07102, USA · Big Bear Solar Observatory, New Jersey Institute of Technology, 40386 North Shore Lane, Big Bear City, CA 92314, USA +4
Solar flares, particularly those of the M- and X-class, have a significant impact on human life because of their potential to disrupt critical infrastructure and communication systems on Earth. Accurate prediction of solar flares is crucial for mitigating these risks, but the black-box nature of conventional deep learning models used in flare prediction limits their trustworthiness and interpretability. In this paper, we propose a new approach to solar flare prediction using photospheric magnetic field parameters or features with deep learning. To improve model interpretability, we integrate explainable artificial intelligence (XAI) techniques, including SHapley Additive exPlanations (SHAP) and partial dependence plots (PDPs), into our prediction framework. XAI methods provide transparency by analyzing the importance and interactions of features used by our model. Specifically, SHAP values offer a global and local understanding of the features, while PDPs provide insights into feature-level trends. These techniques demonstrate the potential of XAI in deploying AI-driven solutions in high-impact applications such as solar flare prediction, paving the way for more informed decision-making in solar physics and space weather studies.
Yash Chaudhary, Jason T. L. Wang, Chunhui Xu +2
New Jersey Institute of Technology Newark, NJ 07102, USA · State University of New York Oneonta, NY 13820, USA
We present FlareEUV, a multimodal deep learning framework for predicting daily extreme ultraviolet (EUV) irradiance at 6.5 nm over three consecutive days during significant solar flares, using multi-instrument observations from NASA's Solar Dynamics Observatory (SDO). We consider 33 significant flares in the period between 2011 and 2014 in Solar Cycle 24. The SDO observations include 13 co-aligned full-disk images, comprising eight AIA EUV/UV and five HMI magnetic/continuum products. FlareEUV learns the relationship between magnetic structure and coronal emission from the raw imaging data using a lightweight attention-based architecture. Our experimental results demonstrate the good performance of FlareEUV in short-term EUV irradiance forecasting during the significant flares and its superiority over baseline methods.
Sathvik Soman, Jason T. L. Wang, Haimin Wang +1
New Jersey Institute of Technology, Newark, NJ, USA · Sam Houston State University, Huntsville, TX, USA