Ionospheric structure usually changes dramatically during a strong geomagnetic storm period, which will significantly affect the short-wave communication and satellite navigation systems. It is critically important to make accurate ionospheric predictions under the extreme space weather conditions. However, ionospheric prediction is always a challenge, and pure physical methods often fail to get a satisfactory result since the ionospheric behavior varies greatly with different geomagnetic storms. In this paper, in order to find an effective prediction method, one traditional mathematical method (autoregressive integrated moving average—ARIMA) and two deep learning algorithms (long short-term memory—LSTM and sequence-to-sequence—Seq2Seq) are investigated for the short-term predictions of ionospheric TEC (Total Electron Content) under different geomagnetic storm conditions based on the MIT (Massachusetts Institute of Technology) madrigal observation from 2001 to 2016. Under the extreme condition, the performance limitation of these methods can be found. When the storm is stronger, the effective prediction horizon of the methods will be shorter. The statistical analysis shows that the LSTM can achieve the best prediction accuracy and is robust for the accurate trend prediction of the strong geomagnetic storms. In contrast, ARIMA and Seq2Seq have relatively poor performance for the prediction of the strong geomagnetic storms. This study brings new insights to the deep learning applications in the space weather forecast.

Observations with a global coverage are very important for space physics research and space weather monitoring. However, due to the technical limitations, it would be very expensive or even impossible to achieve a seamless global coverage even with advanced observational devices. It would be useful to fill missing data gaps to create a global map from the available data, but up until now this has been very challenging. Fortunately, the deep learning method, a recent breakthrough in artificial intelligence, may provide an effective way to solve this problem by making full use of data from reliable observations. In this paper, a promising deep learning algorithm, deep convolutional generative adversarial network (DCGAN), is investigated to fill the missing data of total electron content (TEC) map images. The direct use of DCGAN fails to fill missing data for the completion of TEC maps because there are always missing TEC data in some regions, such as oceans, where the features vary with time and geophysical conditions. Thus, no useful information can be utilized by DCGAN to achieve a meaningful image completion. In order to overcome this shortcoming of the original DCGAN method, a novel regularized DCGAN (R‐DCGAN) is proposed by adding an extra discriminator and some widely used reference TEC maps from the International Global Navigation Satellite Systems Service Ionosphere Working Group. The proposed R‐DCGAN method generates satisfactory ionospheric peak structures at different times and geomagnetic conditions, which demonstrate its effectiveness on filling the missing data in TEC maps. The proposed R‐DCGAN framework can be readily extended to a broad application in other fields of space sciences, particularly for addressing the missing observation data issues.

The accurate prediction of ionospheric Total Electron Content (TEC) is important for global navigation satellite systems (GNSS), satellite communications and other space communications applications. In this study, a prediction model of global IGS‐TEC maps are established based on testing several different long short‐term memory (LSTM) network (LSTM)‐based algorithms to explore a direction that can effectively alleviate the increasing error with prediction time. We find that a Multi‐step auxiliary algorithm based prediction model performs best. It can effectively predict the global ionospheric IGS‐TEC in the next 6 days (the mean absolute deviation (MAD) and root mean square error (RMSE) are 2.485 and 3.511 TECU, respectively) compared to the IRI (the MAD and RMSE are 4.248 and 5.593 TECU). The analyses of four geomagnetic storm events are completely separate from the time range of the training set, so as to further validate the performance of the model. The International Reference Ionosphere model is used as a reference for the performance of our predictive model, and a rotated persistence is estimated by time‐shift algorithm of IGS‐TEC. The result suggests that the Multi‐step auxiliary prediction model has a good generalization performance and can have a relatively good stability and low error during a geomagnetic storm and quiet time.

Solar flare formation mechanisms and their corresponding predictions have commonly been difficult topics in solar physics for decades. The traditional forecasting method manually constructs a statistical relationship between the measured values of solar active regions and solar flares that cannot fully utilize the information related to solar flares contained in observational data. In this article, we first used neural-network methods driven by the measured magnetogram and magnetic characteristic parameters of the sunspot group to learn the prediction model and predict solar flares. The prediction fusion model is based on a deep neural network, convolutional neural network, and bidirectional long short-term memory neural network and can predict whether a sunspot group will have a flare event above class M or class C in the next 24 or 48 hr. The real skill statistics (TSS) and F1 scores were used to evaluate the performances of our fusion model. The test results clearly show that this fusion model can make full use of the information related to solar flares and combine the advantages of each independent model to capture the evolution characteristics of solar flares, which is a much better performance than traditional statistical prediction models or any single machine-learning method. We also proposed two frameworks, namely F1_FFM and TSS_FFM, which optimize the F1 score and TSS score, respectively. The cross validation results show that they have their respective advantages in the F1 score and TSS score.

The accurate prediction of storm‐time thermospheric mass density is always critically important and also a challenge. In this paper, an available prediction model is established by Long Short‐Term Memory (LSTM)‐based ensemble learning algorithms. However, the generalization ability of the deep learning model is often suspicious since training data and testing data are from the same data set in the conventional method. Therefore, in order to objectively validate the performance and generalization of the model, we utilize the GOCE data for training and the SWARM‐C data for testing to verify its performance mainly during the geomagnetic storm period. The results show that the LSTM‐based ensemble learning model (LELM) is robust under different geomagnetic activity levels and has good generalization ability for the different satellite data set. The prediction accuracy of the LELM is proved to be better than a common‐used empirical model (NRLMSISE‐00). Thus, our approach provides a promising way to give reliable and stable predictions of thermospheric mass density.

A solar active region is a source of disturbance for the Sun–terrestrial space environment and usually causes extreme space weather, such as geomagnetic storms. The main indicator of an active region is sunspots. Certain types of sunspots are related to extreme space weather caused by eruptive events such as coronal mass ejections or solar flares. Thus, the automatic classification of sunspot groups is helpful to predict solar activity quickly and accurately. This paper completed the automatic classification of a sunspot group data set based on the Mount Wilson classification scheme, which contains continuum and magnetogram images provided by the Solar Dynamics Observatory’s Helioseismic and Magnetic Imager SHARP data from 2010 May 1 to 2017 December 12. After applying some data preprocessing steps such as image cropping and data standardization, the features of magnetic type in the data are more obvious, and the amount of data is increased. The processed data are spliced into two frames of single-channel data for the neural network to perform 3D convolution operations. This paper constructs a variety of convolutional neural networks with different structures and numbers of layers, selects 10 models as representatives, and chooses XGBoost, which is commonly used in ensemble-learning algorithms, to fuse the results of independent classification models. We found that XGBoost is an effective way to fuse models, which is proved by the relatively balanced high scores in the three magnetic types. The accuracy of the ensemble model is above 92%. The F1 scores of the magnetic types of Alpha, Beta, and Beta-x reached 0.95, 0.91, and 0.82 respectively.