This study presents the observations of midlatitude plasma irregularities over Eastern Asia during a moderate magnetic storm on 16 July 2003. Multi-instrumental observations, including the ground-based ionosondes, the GNSS networks, and the CHAMP and ROCSAT-1 satellites, were utilized to investigate the occurrence and characteristics of midlatitude plasma irregularities. The midlatitude strong spread F (SSF) mainly occurred in the midnight–morning sector as observed by ionosondes over Japan during this storm. SSF was related to plasma depletions, which is also recorded by GNSS network in the form of the enhancement of the rate of total electron content (TEC) change index (ROTI). The possible mechanism for the generation of SSF is that the enhanced eastward electric fields, associated with the prompt penetration electric fields and disturbance dynamo electric fields, cause the uplift and latitudinal extension of equatorial plasma bubbles (EPBs) to generate the observed midlatitude SSF further. Meanwhile, plasma density increased significantly under the influence of this storm. In addition, other common type of spread F, frequency spread F (FSF), was observed over Japan on the non-storm day and/or at high latitude station WK545, which seems to be closely related to the coupling of medium-scale traveling ionospheric disturbances (MSTIDs) and sporadic E (Es) layer. The above results indicate that various types of midlatitude spread F can be produced by different physical mechanisms. It is found that SSF can significantly affect the performance of radio wave propagation compared with FSF. Our results show that space weather events have a significant influence on the day-to-day variability of the occurrence and characteristics of ionospheric F-region irregularities at midlatitudes.

The ionospheric sporadic E layer, the ionospheric irregularities of enhanced electron density, appears in the Earth's ionosphere at altitudes between 90 and 120 km, which supports the real‐world radio communication needs of many sectors reliant on ionosphere‐dependent decision‐making. The prediction of the occurrence of sporadic E layers has been extremely difficult due to the highly complex behavior. Conventional numerical methods are limited because of the inability to extract high‐level information from data. Deep learning is a powerful tool for mining latent features from data, which can theoretically avoid assumptions constraining physical methods. Inspired by feature extraction, we applied deep learning to explore latent relationships between mapping observable lower atmospheric data and ionospheric data from limited observations. The proposed model was trained with high‐resolution ERA5 data during 1 January 2007–30 August 2018 as input and corresponding ionospheric sporadic E data collected from COSMIC RO measurements as output. The results show that the model can learn complex relevance as bridges connecting the input and the desired output and obtain excellent performance and generalization capability by applying multiple evaluation criteria. Additionally, we established several model architecture training methods to explore the performance of the model with different input data. The statistic results show that model inference performance is proportional to the abundance of input information and is impacted by intraseasonal variability. The inference capability of the model achieves the best performance in the June–August (JJA) and December–February (DJF) seasons, which is the exact period of sporadic E layer significant occurrence, although different models are evaluated.

Abstract. The ionospheric sporadic E (Es) layer is the intense plasma irregularities between 80 and 130 km in altitude, which is generally unpredictable. Reconstructing the morphology of sporadic E layer is not only essential for understanding the nature of ionospheric irregularities and many other atmospheric coupling systems, but also useful to solve a broad range of demands for reliable radio communication of many sectors reliant on ionosphere-dependent decision-making. Despite the efforts of many empirical and theoretical models, a predictive algorithm with both high accuracy and high efficiency is still lacking. Here we introduce a new approach for Sporadic E Layer Forecast using Artificial Neural Networks (SELF-ANN). The prediction engine is trained by fusing observational data from multiple sources, including high-resolution ERA5 reanalysis dataset, COSMIC RO measurements, and integrated data from OMNI. The results show that the model can effectively reconstruct the morphology of the ionospheric E layer with intraseasonal variability by learning complex patterns. The model obtains good performance and generalization capability by applying multiple evaluation criteria. The random forest algorithm used for preliminary pro- cessing shows that local time, altitude, longitude, and latitude are significantly essential for forecasting the E-layer region. Extensive evaluations based on ground-based observations demonstrate the superior utility of the model in dealing with unknown information. The presented framework will help us better understand the nature of the ionospheric irregularities, which is a fundamental challenge in upper atmospheric and ionospheric physics. Moreover, the proposed SELF-ANN can provide a significant contribution to the development of the prediction of ionospheric irregularities in the E layer, particularly when the formation mechanisms and evolution processes of the Es layer are not well understood.

Random Forest is an ensemble of D trees {T 1 (X), • • • , T D (X)}, where X = {x 1 , • • • , x p } is a pdimensional vector of properties associated with a scintillation index. The ensemble produces D outputis the prediction for a scintillation index by the dth tree. Outputs of all trees are aggregated to produce one final prediction, Ŷ . For regression problem in this study, Ŷ is the average of the individual tree predictions. Given data on a set of n radio occulation events for training,, is a vector of descriptors and Y i is either the corresponding desired value (e.g., the S4max intensity), the training algorithm proceeds as follows. (i) From the training data of n events, draw a bootstrap sample (i.e., randomly sample, with replacement, n samples). (ii) For each bootstrap sample, grow a tree with the following modification: at each node, choose the best split among a randomly selected subset of m (m < n) descriptors. Here m is essentially the only tuning parameter in the algorithm. The tree is grown to the maximum size (i.e., until no further splits are possible) and not pruned back. (iii) Repeat the above steps until (a sufficiently large number) D such trees are grown.