Here we greatly improve artificial intelligence (AI)–generated solar farside magnetograms using data sets from the Solar Terrestrial Relations Observatory (STEREO) and Solar Dynamics Observatory (SDO). We modify our previous deep-learning model and configuration of input data sets to generate more realistic magnetograms than before. First, our model, which is called Pix2PixCC, uses updated objective functions, which include correlation coefficients (CCs) between the real and generated data. Second, we construct input data sets of our model: solar farside STEREO extreme-ultraviolet (EUV) observations together with nearest frontside SDO data pairs of EUV observations and magnetograms. We expect that the frontside data pairs provide historic information on magnetic field polarity distributions. We demonstrate that magnetic field distributions generated by our model are more consistent with the real ones than previously, in consideration of several metrics. The averaged pixel-to-pixel CC for full disk, active regions, and quiet regions between real and AI-generated magnetograms with 8 × 8 binning are 0.88, 0.91, and 0.70, respectively. Total unsigned magnetic flux and net magnetic flux of the AI-generated magnetograms are consistent with those of real ones for the test data sets. It is interesting to note that our farside magnetograms produce polar field strengths and magnetic field polarities consistent with those of nearby frontside magnetograms for solar cycles 24 and 25. Now we can monitor the temporal evolution of active regions using solar farside magnetograms by the model together with the frontside ones. Our AI-generated solar farside magnetograms are now publicly available at the Korean Data Center for SDO (http://sdo.kasi.re.kr). 

 Solar magnetic fields play a key role in understanding the nature of the coronal phenomena. Global coronal magnetic fields are usually extrapolated from photospheric fields, for which farside data is taken when it was at the frontside, about two weeks earlier. For the first time we have constructed the extrapolations of global magnetic fields using frontside and artificial intelligence (AI)-generated farside magnetic fields at a near-real time basis. We generate the farside magnetograms from three channel farside observations of Solar Terrestrial Relations Observatory (STEREO) Ahead (A) and Behind (B) by our deep learning model trained with frontside Solar Dynamics Observatory extreme ultraviolet images and magnetograms. For frontside testing data sets, we demonstrate that the generated magnetic field distributions are consistent with the real ones; not only active regions (ARs), but also quiet regions of the Sun. We make global magnetic field synchronic maps in which conventional farside data are replaced by farside ones generated by our model. The synchronic maps show much better not only the appearance of ARs but also the disappearance of others on the solar surface than before. We use these synchronized magnetic data to extrapolate the global coronal fields using Potential Field Source Surface (PFSS) model. We show that our results are much more consistent with coronal observations than those of the conventional method in view of solar active regions and coronal holes. We present several positive prospects of our new methodology for the study of solar corona, heliosphere, and space weather.

The Sun-Earth Lagrange point L4, which is called a parking space of space, is considered one of the unique places where solar activity and the heliospheric environment can be observed continuously and comprehensively. The L4 mission affords a clear and wide-angle view of the Sun-Earth line for the study of Sun-Earth connections from remote-sensing observations. The L4 mission will significantly contribute to advancing heliophysics science, improving space weather forecasting capability, extending space weather studies far beyond near-Earth space, and reducing risk from solar radiation hazards on human missions to the Moon and Mars. Our paper outlines the importance of L4 observations by using remote-sensing instruments and advocates comprehensive and coordinated observations of the heliosphere at multi-points including other planned L1 and L5 missions. We mainly discuss scientific perspectives on three topics in view of remote sensing observations: (1) solar magnetic field structure and evolution, (2) source regions of geoeffective solar energetic particles (SEPs), and (3) stereoscopic views of solar corona and coronal mass ejections (CMEs).

For the first time, we generate solar coronal parameters (density, magnetic field, radial velocity, and temperature) on a near-real-time basis by deep learning. For this, we apply the Pix2PixCC deep-learning model to threedimensional (3D) distributions of these parameters: synoptic maps of the photospheric magnetic field as an input and the magnetohydrodynamic algorithm outside a sphere (MAS) results as an output. To generate the 3D structure of the solar coronal parameters from 1 to 30 solar radii, we train and evaluate 152 distinct deep-learning models. For each parameter, we consider the data of 169 Carrington rotations from 2010 June to 2023 February: 132 for training and 37 for testing. The key findings of our study are as follows: First, our deep-learning models successfully reconstruct the 3D distributions of coronal parameters from 1 to 30 solar radii with an average correlation coefficient of 0.98. Second, during the solar active and quiet periods, the AI-generated data exhibits consistency with the target MAS simulation data. Third, our deep-learning models for each parameter took a remarkably short time (about 16 s for each parameter) to generate the results with an NVIDIA Titan XP GPU. As the MAS simulation is a regularization model, we may significantly reduce the simulation time by using our results as an initial configuration to obtain an equilibrium condition. We hope that the generated 3D solar coronal parameters can be used for the near-real-time forecasting of heliospheric propagation of solar eruptions.

We study the source regions of six solar energetic particle (SEP) events accelerated near or behind the limbs of the Sun. We use AI-generated farside magnetograms at a near real-time basis developed by Jeong et al. and AIHMI-PFSS extrapolations up to 2.5Rsun computed using the input of the synchronic data combining AI-generated farside and HMI magnetograms. By comparing the AIHMI, HMI, Global Oscillations Network Group (GONG) synoptic magnetograms, and Air force Data Assimilative Photospheric flux Transport synchronic magnetograms, as well as the PFSS extrapolations, we find interesting differences between them in view of SEP source regions and magnetic field configurations. First, the structures and sizes of the source active regions (ARs) are changed. The total unsigned magnetic field fluxes of the ARs are mostly stronger in the AIHMI than in the HMI and GONG magnetograms. Second, newly emerging ARs are observed in the SEP source regions in the AIHMI magnetograms for two events. Third, the alterations in the magnetic flux, the emergence, and the dissipation of ARs lead to modifications in the locations of the global polarity inversion lines (PILs). The EUV wave propagation is typically observed to be oriented nearly perpendicular with respect to the local PIL, suggesting that the AIHMI-PFSS extrapolations around the source region are more realistic. This study shows that the continuous farside evolution of AR magnetic fields, which is accomplished by our AI synchronic magnetograms, can lead to an improved understanding of SEP source ARs.

In this study, we forecast solar wind speed for the next 3 days with a 6 hr cadence using a deep-learning model. For this we use Solar Dynamics Observatory/Atmospheric Imaging Assembly 211 and 193 Å images together with solar wind speeds for the last 5 days as input data. The total period of the data is from 2010 May to 2020 December. We divide them into a training set (January–August), validation set (September), and test set (October–December), to consider the solar cycle effect. The deep-learning model consists of two networks: a convolutional layer–based network for images and a dense layer–based network for solar wind speeds. Our main results are as follows. First, our model successfully predicts the solar wind speed for the next 3 days. The rms error (RMSE) of our model is from 37.4 km/s (for the 6 hr prediction) to 68.2 km/s (for the 72 hr prediction), and the correlation coefficient is from 0.92 to 0.67. These results are much better than those of previous studies. Second, the model can predict sudden increase of solar wind speeds caused by large equatorial coronal holes. Third, solar wind speeds predicted by our model are more consistent with observations than those by the Wang–Sheely–Arge–ENLIL model, especially in high-speed-stream regions. It is also noted that our model cannot predict solar wind speed enhancement by coronal mass ejections. Our study demonstrates the effectiveness of deep learning for solar wind speed prediction, with potential applications in space weather forecasting.

Accurate and reliable disaster forecasting is vital for saving lives and property. Hence, effective disaster management is necessary to reduce the impact of natural disasters and to accelerate recovery and reconstruction. Typhoons are one of the major disasters related to heavy rainfall in Korea. As a typhoon develops in the far ocean, satellite observations are the only means to monitor them. Our study uses satellite observations to propose a deep-learning-based disaster monitoring model for short-term typhoon rainfall forecasting. For this, we consider two deep learning models: a video frame prediction model, Warp and Refine Network (WR-Net), to predict future satellite observations and an image-to-image translation model, geostationary rainfall product (GeorAIn) (based on the Pix2PixCC model), to generate rainfall maps from predicted satellite images. Typhoon Hinnamnor, the worst typhoon case in 2022 in Korea, is selected as a target case for model verification. The results show that the predicted satellite images can capture the structures and patterns of the typhoon. The rainfall maps generated from the GeorAIn model using predicted satellite images show a correlation coefficient of 0.81 for 3-hr and 0.56 for 7-hr predictions. The proposed disaster monitoring model can provide us with practical implications for disaster alerting systems and can be extended to flood-monitoring systems.

This study is the first attempt to generate a three-dimensional (3D) coronal electron density distribution based on the pix2pixHD model, whose computing time is much shorter than that of the magnetohydrodynamic (MHD) simulation. For this, we consider photospheric solar magnetic fields as input, and electron density distribution simulated with the MHD Algorithm outside a Sphere (MAS) at a given solar radius is taken as output. We consider 155 pairs of Carrington rotations as inputs and outputs from 2010 June to 2022 April for training and testing. We train 152 deep-learning models for 152 solar radii, which are taken up to 30 solar radii. The artificial intelligence (AI) generated 3D electron densities from this study are quite consistent with the simulated ones from lower radii to higher radii, with an average correlation coefficient 0.97. The computing time of testing data sets up to 30 solar radii of 152 deep-learning models is about 45.2 s using the NVIDIA TITAN XP graphics-processing unit, which is much less than the typical simulation time of MAS. We find that the synthetic coronagraphic images estimated from the deep-learning models are similar to the Solar Heliospheric Observatory (SOHO)/Large Angle and Spectroscopic Coronagraph C3 coronagraph data, especially during the solar minimum period. The AI-generated coronal density distribution from this study can be used for space weather models on a near-real-time basis.

 In this study, we present the application of deep reinforcement learning to the forecasting of major solar flares. For this, we consider full-disk magnetograms at 00:00 UT from the Solar and Heliospheric Observatory/Michelson Doppler Imager (1996–2010) and the Solar Dynamics Observatory/Helioseismic and Magnetic Imager (2011–2019), as well as Geostationary Operational Environmental Satellite X-ray flare data. We apply Deep Q-Network (DQN) and Double DQN, which are popular deep reinforcement learning methods, to predict "Yes or No" for daily M- and X-class flare occurrence. The reward functions, consisting of four rewards for true positive, false positive, false negative, and true negative, are used for our models. The major results of this study are as follows. First, our deep-learning models successfully predict major solar flares with good skill scores, such as HSS, F1, TSS, and ApSS. Second, the performance of our models depends on the reward function, learning method, and target agent update time. Third, the performance of our deep-learning models is noticeably better than that of a convolutional neural network (CNN) model with the same structure: 0.38 (CNN) to 0.44 (ours) for HSS, 0.47 to 0.52 for F1, 0.53 to 0.59 for TSS, and 0.09 to 0.12 for ApSS. 

 In this study, we suggest a pixel-to-pixel image translation method among similar types of filtergrams such as solar extreme-ultraviolet (EUV) images. For this, we consider a deep-learning model based on a fully connected network in which all pixels of solar EUV images are independent of one another. We use six-EUV-channel data from the Atmospheric Imaging Assembly (AIA) on board the Solar Dynamics Observatory (SDO), of which three channels (17.1, 19.3, and 21.1 nm) are used as the input data and the remaining three channels (9.4, 13.1, and 33.5 nm) as the target data. We apply our model to representative solar structures (coronal loops inside of the solar disk and above the limb, coronal bright point, and coronal hole) in SDO/AIA data and then determine differential emission measures (DEMs). Our results from this study are as follows. First, our model generates three EUV channels (9.4, 13.1, and 33.5 nm) with average correlation coefficient values of 0.78, 0.89, and 0.85, respectively. Second, our model generates the solar EUV data with no boundary effects and clearer identification of small structures when compared to a convolutional neural network-based deep-learning model. Third, the estimated DEMs from AI-generated data by our model are consistent with those using only SDO/AIA channel data. Fourth, for a region in the coronal hole, the estimated DEMs from AI-generated data by our model are more consistent with those from the 50 frames stacked SDO/AIA data than those from the single-frame SDO/AIA data. 

 In this study, we generate He i 1083 nm images from Solar Dynamic Observatory (SDO)/Atmospheric Imaging Assembly (AIA) images using a novel deep learning method (pix2pixHD) based on conditional Generative Adversarial Networks (cGAN). He i 1083 nm images from National Solar Observatory (NSO)/Synoptic Optical Long-term Investigations of the Sun (SOLIS) are used as target data. We make three models: single-input SDO/AIA 19.3 nm image for Model I, single-input 30.4 nm image for Model II, and double-input (19.3 and 30.4 nm) images for Model III. We use data from 2010 October to 2015 July except for June and December for training and the remaining one for test. Major results of our study are as follows. First, the models successfully generate He i 1083 nm images with high correlations. Second, Model III shows better results than those with one input image in terms of metrics such as correlation coefficient (CC) and root mean square error (RMSE). CC and RMSE between real and synthetic ones for model III with 4 by 4 binnings are 0.88 and 9.49, respectively. Third, synthetic images show well observational features such as active regions, filaments, and coronal holes. This work is meaningful in that our model can produce He i 1083 nm images with higher cadence without data gaps, which would be useful for studying the time evolution of the chromosphere and transition region. 

 In this Letter, we generate realistic high-resolution (1024 × 1024 pixels) pseudo-magnetograms from Ca II K images using a deep learning model based on conditional generative adversarial networks. For this, we consider a model "pix2pixHD" that is specifically devised for high-resolution image translation tasks. We use Ca II K 393.3 nm images from the Precision Solar Photometric Telescope at the Rome Observatory and line-of-sight magnetograms from the Helioseismic and Magnetic Imager (HMI) at the Solar Dynamics Observatory from 2011 January to 2015 June. 2465 pairs of Ca II K and HMI are used for training except for January and July data. The remaining 436 pairs are used for an evaluation of the model. Our model shows that the mean correlation coefficient (CC) of total unsigned magnetic flux between AI-generated and real ones is 0.99 and the mean pixel-to-pixel CC after 8 × 8 binning over the full disk is 0.74. We find that the AI-generated absolute magnetic flux densities are highly consistent with real ones, even to the fine scale structures of quiet regions. On the other hand, the mean pixel-to-pixel correlations of magnetic flux densities strongly depend on a region of interest: 0.81 for active regions and 0.24 for quiet regions. Our results suggest a sufficient possibility that we can produce high-resolution solar magnetograms from historical Ca II data.