The application of deep learning technology to ionospheric prediction has become a new research hotspot. However, there are still some gaps, such as the prediction effect with different input solar and geomagnetic activity parameters, and the forecast accuracy with different prediction methods as well as the validation of long period data results, to be filled. We developed an ionospheric long short‐term memory network (Ion‐LSTM) with multiple input parameters to predict the global ionospheric total electron content (TEC). Two solutions with different ionospheric data based on Ion‐LSTM were assessed, namely spherical harmonic coefficients (SHC) and vertical TEC (VTEC) prediction solution. The results show two solutions, both perform well in accuracy and stability. The input of the geomagnetic activity index improves the prediction effect of the model in the storm period. For the 1‐ and 2‐day‐predicted global ionospheric maps (GIMs) from 2015 to 2020, the root mean square error (RMSE) of SHC prediction solution is 1.69 TECU and 1.84 TECU while that of the VTEC prediction solution is 1.70 TECU and 1.84 TECU, respectively. Over 70% of the absolute residuals are within 3 TECU in high solar activity and over 96% in low solar activity. Further, by comparing the predicted results between Ion‐LSTM and conventional methods (e.g., Center for Orbit Determination in Europe (CODE) predicted GIMs), the evaluation results show that the RMSE of Ion‐LSTM is 0.7 TECU lower than that of CODE predicted GIMs under different solar and geomagnetic activities. Additionally, the accuracy of the Ion‐LSTM prediction results decreases slightly as the input time span increases.

In recent years, deep learning has been extensively used for ionospheric total electron content (TEC) prediction, and many models can yield promising prediction results, particularly under quiet conditions. Owing to the ionosphere's intricate and dramatic changes during geomagnetic storms, the high‐reliable prediction of the storm‐time ionospheric TEC remains a challenging problem. In this study, we developed a new deep learning‐based multi‐model ensemble method (DLMEM) to forecast geomagnetic storm‐time ionospheric TEC that combines the Random Forest (RF) model, the Extreme Gradient Boosting (XGBoost) algorithm, and the Gated Recurrent Unit (GRU) network with the attention mechanism. Seven features in 170 geomagnetic storm events, including the three components Bx, By and Bz of interplanetary magnetic field (IMF), the Kp and Dst indices of geomagnetic activity data, the F_10.7 index of solar activity data and global TEC data, were used for modeling. The test set results showed that the DLMEM model can reduce the root mean square errors (RMSE) by an average of 43.6% in comparison to our previously presented model Ion‐LSTM, especially during the recovery period of geomagnetic storms. Furthermore, compared to Ion‐LSTM, the RMSE values of the low‐, middle‐ and high‐latitude single‐station forecast TEC can be greatly decreased by 33%, 53% and 59%, respectively. It was shown that the new model allows for more precise short‐term global ionospheric forecasting during geomagnetic storms, enabling real‐time monitoring of ionospheric changes.

Precise extraction of ionospheric total electron content (TEC) observations with high precision is the precondition for establishing high-precision ionospheric TEC models. Nowadays, there are several ways to extract TEC observations, e.g., raw-code method (Raw-C), phase-leveled code method (PL-C), and undifferenced and uncombined precise point positioning method (UD-PPP); however, their accuracy is affected by multipath and noise. Considering the limitations of the three traditional methods, we try directly to use the phase observation based on zero-difference integer ambiguity to extract ionospheric observations, namely, PPP-Fixed method. The main goal of this work is to: 1) deduce the expression of ionospheric observables of these four extraction methods in a mathematical formula, especially the satellite and receiver hardware delays; 2) investigate the performance and precision of ionospheric observables extracted from different algorithms using two validation methods, i.e., the co-location experiment by calculating the single difference for each satellite, and the single-frequency PPP (SF-PPP) test by two co-location stations; and 3) use the short arc experiment to demonstrate the advantages of the PPP-Fixed method. The results show that single-difference mean errors of TEC extracted by PL-C, UD-PPP, and PPP-Fixed are 1.81, 0.59, and 0.15 TEC unit (TECU), respectively, and their corresponding maximum single-difference values are 5.12, 1.68, and 0.43 TECU, respectively. Compared with PL-C, the precision of the TEC observations extracted by the PPP-Fixed method is improved by 91.7%, while it is 67.3% for UD-PPP. The SF-PPP experiment shows that PPP-Fixed is the best among these methods in terms of convergence speed, correction accuracy, and reliability of positioning performance. Moreover, the PPP-Fixed method can achieve high accuracy even when the observed arc is short, e.g., within 40 min.

It is important to confirm the accuracy and reliability of commonly used ionosphere models climatologically. In this contribution, International Global Navigation Satellite System Global Ionospheric Maps (IGSG) and two empirical models, that is, NeQuick2 and IRI‐2016, are assessed in detail by applying different assessment methods, for example, Jason2/3 ionospheric data, difference of Slant Total Electron Content (dSTEC) data derived from Global Navigation Satellite System (GNSS) phase observation, and single‐frequency precise point positioning. Compared with IGSG, the biases mainly range from −10 to 10 Total Electron Content Unit (TECU), while they are between −5 and 5 TECU on solar low‐level days for empirical models. The hourly mean biases are about −2.7 and −2.5 TECU for IRI‐2016 and NeQuick2 models, respectively. Over the oceanic region, the mean biases for IRI‐2016 and NeQuick2 models relative to Jason2/3‐Vertical Total Electron Content (VTEC) are smaller than that of IGSG. The Root Mean Square (RMS) values are 4.8 and 6.0 TECU for IGSG and empirical models relative to Jason2‐VTEC while the values are 5.4 and 6.0 TECU with relative to Jason3‐VTEC. Compared with dSTEC values derived from the selected stations, the RMS values are about 1.8 and 2.6 TECU for IGSG and empirical models, respectively. In the positioning domain, the accuracy of single‐frequency precise point positioning corrected by the three models can reach 1 m in three‐dimensional direction. The positioning accuracy is 0.17 m corrected by IGSG and 0.50 m corrected by IRI‐2016 and NeQuick2 in the horizontal direction.

Global ionosphere model with high accuracy and resolution is of great importance for ionospheric research and global navigation satellite system (GNSS) precise positioning applications. In the last 20 years, the accuracy and reliability of global ionospheric model benefitted from the development of GNSS technology have been improved significantly, but its performance is still not good over many regions (e.g., oceans and polar region) due to the lack of ground GNSS stations. Fortunately, the rapid development of low earth orbit (LEO) satellite constellations provides a potential opportunity to address this issue. However, there is still no optimal approach to make use of the LEO-based ionospheric observations because of the different observation ranges compared with that of ground-based GNSS ionospheric observations. In this article, two approaches are proposed to combine GNSS and LEO observation data for ionosphere modeling, which are single-layer normalization (SLN) method and dual-layer superposition (DLS) method, respectively. The results exhibit a significant improvement of ionospheric model accuracy by combining LEO and GNSS observation data based on our proposed methods compared with that using GNSS data only, with a reduction in root mean square (rms) error of about 25% and 21% for SLN method and DLS method, respectively. The article also highlights the relations between the performance of ionospheric model estimated by the SLN method and LEO ionospheric observations with different observation accuracy and different satellite cut-off elevations. The results indicate that ionospheric model estimated by GNSS/LEO using SLN method improves at least 25% compared with that by GNSS only. The improvement of ionospheric model estimated with the cut-off elevation of 50° is the best, followed by 70°, and then 20°.