With the continuous advancement of deep learning algorithms and the rapid growth of computational resources, deep learning technology has undergone numerous milestone developments, evolving from simple BP neural networks into more complex and powerful network models such as CNNs, LSTMs, RNNs, and GANs. In recent years, the application of deep learning technology in ionospheric modeling has achieved breakthrough advancements, significantly impacting navigation, communication, and space weather forecasting. Nevertheless, due to limitations in observational networks and the dynamic complexity of the ionosphere, deep learning-based ionospheric models still face challenges in terms of accuracy, resolution, and interpretability. This paper systematically reviews the development of deep learning applications in ionospheric modeling, summarizing findings that demonstrate how integrating multi-source data and employing multi-model ensemble strategies has substantially improved the stability of spatiotemporal predictions, especially in handling complex space weather events. Additionally, this study explores the potential of deep learning in ionospheric modeling for the early warning of geological hazards such as earthquakes, volcanic eruptions, and tsunamis, offering new insights for constructing ionospheric-geological activity warning models. Looking ahead, research will focus on developing hybrid models that integrate physical modeling with deep learning, exploring adaptive learning algorithms and multi-modal data fusion techniques to enhance long-term predictive capabilities, particularly in addressing the impact of climate change on the ionosphere. Overall, deep learning provides a powerful tool for ionospheric modeling and indicates promising prospects for its application in early warning systems and future research.

The Global Positioning System (GPS) is a powerful atmospheric observing system for determining precipitable water vapour (PWV). In the detection of PWV using GPS, the atmospheric weighted mean temperature (T_m) is a crucial parameter for the conversion of zenith tropospheric delay (ZTD) to PWV since the quality of PWV is affected by the accuracy of T_m. In this study, an improved voxel-based T_m model, named GWMT-D, was developed using global reanalysis data over a 4-year period from 2010 to 2013 provided by the United States National Centers for Environmental Prediction (NCEP). The performance of GWMT-D was assessed against three existing empirical T_m models – GTm-III, GWMT-IV, and GTm_N – using different data sources in 2014 – the NCEP reanalysis data, surface T_m data provided by Global Geodetic Observing System and radiosonde measurements. The results show that the new GWMT-D model outperforms all the other three models with a root-mean-square error of less than 5.0 K at different altitudes over the globe. The new GWMT-D model can provide a practical alternative T_m determination method in real-time GPS-PWV remote sensing systems.

The Disturbance storm time (Dst) index has been widely used as a proxy for the ring current intensity, and therefore as a measure of geomagnetic activity. It is derived by measurements from four ground magnetometers in the geomagnetic equatorial region. We present a new model for predicting Dst with a lead time between 1 and 6 hr. The model is first developed using a Gated Recurrent Unit (GRU) network that is trained using solar wind parameters. The uncertainty of the Dst model is then estimated by using the Accurate and Reliable Uncertainty Estimate method (Camporeale & Carè, 2021, https://doi.org/10.1615/int.j.uncertaintyquantification.2021034623). Finally, a multi‐fidelity boosting method is developed in order to enhance the accuracy of the model and reduce its associated uncertainty. It is shown that the developed model can predict Dst 6 hr ahead with a root‐mean‐square‐error of 13.54 nT. This is significantly better than a persistence model or a single GRU model.

Global navigation satellite systems (GNSS) has been applied to the sounding of precipitable water vapor (PWV) due to its high accuracy and high spatiotemporal resolutions. PWV obtained from GNSS (GNSS-PWV) can be used to investigate extreme weather phenomena, such as the formation mechanism and prediction of rainfalls. In the study, a new, improved model for rainfall forecasting was developed based on GNSS data and rainfall data for the 9-year period from 2010 to 2018 at 66 stations located in the USA. The new model included three prediction factors—PWV value, PWV increase, maximum hourly PWV increase. The two key tasks involved for the development of the model were the determination of the thresholds for each prediction factor and the selection of the optimal strategy for using the three prediction factors together. For determining the thresholds, both critical success index (CSI) and true skill statistic (TSS) were tested, and results showed that TSS outperformed CSI for all rainfall events tested. Then, various strategies by combining the three prediction factors together were also tested, and results indicated that the best forecast result was from the case that any two of the prediction factors were over their own thresholds. Finally, the new model was evaluated using the GNSS data for the 2-year period from 2019 to 2020 at the above mentioned 66 stations, and the probability of detection (POD) and false-alarms rate (FAR) were adopted to measure the model performances. Over the 66 stations, the POD values ranged from 73% to 97% with the mean of 87%, and the FARs ranged from 26% to 77% with the mean of 53%. Moreover, it was also found that both POD and FAR values were related to the region of the station; e.g., the results at the stations that are located in humid regions were better than the ones located in dry regions. All these results suggest the feasibility and good performance of using GNSS-PWV for forecasting rainfall.

The height of F2 peak (hmF2) is an essential ionospheric parameter and its variations can reflect both the earth magnetic and solar activities. Therefore, reliable prediction of hmF2 is important for the study of space, such as solar wind and extreme weather events. However, most current models are unable to forecast the variation of the ionosphere effectively since real-time measurements are required as model inputs. In this study, a new Australian regional hmF2 forecast model was developed by using ionosonde measurements and the bidirectional Long Short-Term Memory (bi-LSTM) method. The hmF2 value in the next hour can be predicted using the data from the past five hours at the same location. The inputs chosen from a location of interest include month of the year, local time (LT), $K_p$ , $F_{10.7}$ and hmF2 as an independent variable vector. The independent variable vectors in the immediate past five hours are considered as an independent variable set, which is used as an input of the new Australian regional hmF2 forecast model developed for the prediction of hmF2 in the hour to come. The performance of the new model developed is evaluated by comparing with those from other popular models, such as the AMTB, Shubin, ANN and LSTM models. Results showed that: (1) the new model can substantially outperform all the other four models. (2) Compared to the LSTM model, the new model is proven to be more robust and rapidly convergent. The mew model also outperforms that of the ANN model by around 30%. (3) the minimum sample number for the bi-LSTM method (i.e., 2000) to converge is about 50% less than that is required for the LSTM method (i.e., 3000). (4) Compared to the Shubin model, the bi-LSTM method can effectively forecast the hmF2 values up to 5 h. This research is a first attempt at using the deep learning-based method for the application of the ionospheric prediction.

Two solar flares X2.2 and X9.3 erupted over the active region 2673 on 6 September 2017, and the second flare is the strongest since 2005. In order to investigate the ionospheric and thermospheric responses to the two solar flares, the global total electron content and the critical frequency of F2 layer obtained from GPS stations and ionosondes are used. The results indicate that the ionosphere in the sunlit hemisphere increased significantly with magnitudes of 0.1 and 0.5 total electron content units for the X2.2 and X9.3 solar flares, respectively. The electron density, thermospheric neutral density, and neutral temperature simulated by the Thermosphere‐Ionosphere Electrodynamics Global Circulation Model show that the behavior of ionospheric and thermospheric responses is different. The ionospheric disturbances occurred at the altitude ranges of 150−300 km, and the thermospheric responses occurred at the altitudes of 250−400 km are caused by solar extreme ultraviolet and ultraviolet photons, respectively. Both ionospheric and thermospheric responses are proportional to the height within their corresponding altitude ranges. Observations and simulations reveal that the ionospheric and thermospheric responses are nonlinearly dependent on the solar zenith angle. The disturbances caused by the X2.2 solar flare are symmetric, but the X9.3 solar flare are not. The O/N₂ density ratio simulated by Thermosphere‐Ionosphere Electrodynamics Global Circulation Model increases from lev0 to lev5.0 pressure surface with a magnitude of 0.1–1.8, while the ratio decreases in the American sector with a magnitude of −0.6 to −0.3. The longitudinal asymmetry of O/N₂ density ratio is a major contributor to the longitudinal asymmetry of ionospheric and thermospheric responses.