The accurate estimation of electron density in the ionosphere is crucial for various applications including remote sensing systems, communication, satellite positioning, and navigation. Previous ionospheric models using artificial neural networks (ANN) are only data‐driven, whose effects are entirely affected by observational data. In this paper, we utilize the results of International Reference Ionosphere (IRI)‐2016 as prior knowledge to develop a low‐latitude (30°N–30°S) three‐dimensional ionospheric electron density model using ANN, namely ANN‐IRI, based on COSMIC radio occultation ionospheric profiles during 2006–2020. The prior knowledge helps ANN‐IRI get a better prediction effect and obtain convergence more quickly. The COSMIC data sets above the altitude 150 km are divided into three sets, a training set (2006–2014 and 2018–2020), a validation set (2016), and a test set (2015 and 2017). For the test set, ANN‐IRI shows a good performance for predicting the electron density, better than ANN (without prior knowledge) and IRI‐2016. And ANN‐IRI behaves better during quiet than disturbed times as well as during low solar activity than high solar activity years. In addition, we corrected effectively the error of ANN‐IRI in the lower ionosphere source from COSMIC data based on IRI‐2016 and spline interpolation. Compared with completely independent data sets from incoherent scatter radars (ISRs) and ROCSAT‐1 in situ observations, the electron density predicted by corrected ANN‐IRI exhibits good similarity. Generally, ANN‐IRI behaves better than ANN and IRI‐2016 for predicting electron density. Our work demonstrates a new possibility of applying deep learning methods with prior knowledge to a broader field of geosciences, particularly for problems of prediction.

Accurate forecasting of the ionospheric state is crucial for various applications including remote sensing and navigation. Total electron content (TEC) is an important ionospheric parameter to reflect ionospheric state. Consequently, there is a great interest in the prediction of TEC. In this study, we integrated the Informer deep learning algorithm and Principal Component Analysis (PCA), a dimensionality reduction technique, to achieve spatio‐temporal modeling for forecasting the global TEC maps. Our evaluation, based on test set data from 2015 to 2022, demonstrate that the PCA‐Informer model outperforms the IRI‐2016, standalone Informer, and PCA‐enhanced Long Short‐Term Memory (PCA‐LSTM) models in terms of accuracy with root mean squared error (RMSE) of 2.60 TECU and mean relative error (MRE) of 14.1%, and stability for predicting TEC maps for the subsequent 2 days. Two distinct periods (a geomagnetic quiet period and a strong storm period) in 2018 have been selected for evaluating the models' efficacy. The PCA‐Informer model has shown remarkable predictive precision during the initial and main phase of strong geomagnetic storms as well as quiet period. During 132 geomagnetic storm events on the test set, the PCA model exhibits RMSE of 3.6 TECU and MRE of 17.7%, which are lower than IRI‐2016 (6.1 TECU, 33.38%), PCA‐LSTM (4.5 TECU, 21.52%) and Informer (4.1 TECU, 20.64%). Additionally, model errors are negatively correlated with the minimum Dst, while PCA‐Informer has the best robustness.

Plasma bubbles represent notable ionospheric irregularities primarily observed in low latitudes, characterized by plasma depletions exhibiting large spatial scales, which can make a significant impact on the propagation of OTHR (over-the-horizon radar) waves. Firstly, we constructed a three-dimensional model of plasma bubbles, which is modulated by Gaussian function distribution in the horizontal direction, and then we analyzed the impact of EPBs (Equatorial Plasma Bubbles) on the ray path of OTHR shortwaves. When radio waves propagate through EPBs with different RMS ΔN/N, there is a significant difference in the propagation path of OTHR waves. For the EPB with an RMS ΔN/N of 75%, radio waves exhibit more pronounced refraction than those with lower RMS values, the focusing effect of radio waves is more obvious, and the focusing point is relatively lower. In terms of different seasons, OTHR shortwaves propagating through EPBs exhibit different degrees of refraction. In addition, radio waves show the effect of inward focusing in different seasons: the focusing effect is the most pronounced in spring, followed by autumn, then summer, and the weakest in winter. For different solar activities, the impact of EPBs on OTHR shortwaves is more significant in the high-solar-activity year.

Geomagnetic disturbances are an emerging sustainability challenge for modern, low-carbon and highly interconnected power systems, affecting both grid stability and market performance. We develop a deep causal neural network that fuses geomagnetic observatory measurements with national operational indicators and, via counterfactual inference, traces shock and no-shock trajectories to estimate instantaneous and cumulative impacts. Using Switzerland as a case, shocks significantly change national load, canton-level consumption, cross-border flows, and balancing prices. East–west disturbances have stronger effects than north–south, highlighting the role of grid topology. At the regional scale, the canton of Aargau shows pronounced cumulative consumption responses, revealing spatial heterogeneity. In cross-border exchanges, imports rise after shocks while exports contract and transit flows decline; balancing prices increase markedly, suggesting that market mechanisms can amplify physical stress into economic impacts. The approach goes beyond correlation and exposure metrics by providing system-level, decision-relevant effect sizes. The main contributions are as follows: (i) a deep causal framework that identifies and quantifies the causal effects of geomagnetic disturbances on grid operations and prices; (ii) topology-linked empirical evidence of directional and spatial asymmetries across national, canton-level, and cross-border indicators; and (iii) actionable levers for system operation and market design. These findings inform risk-aware reserve procurement, topology-aware dispatch, and cross-border coordination in highly interconnected, low-carbon grids, helping to enhance reliability, maintain affordability, and facilitate clean-energy integration.

Sporadic E (Es) layers exhibit strong intermittency and highly skewed intensity distributions, exerting significant impacts on high-frequency communication and navigation systems and posing challenges for data-driven prediction. Conventional single-stage regression models are often dominated by abundant non-event samples and therefore tend to underestimate Es intensity during occurrence periods. To address this issue, this study proposes a unified two-stage neural network framework that decouples the prediction of Es occurrence probability from the estimation of Es intensity. The model is trained using multi-station ionosonde observations, incorporating cyclic representations of seasonal and local time variations together with solar and geomagnetic indices and station-aware encoding to enable unified learning across multiple stations. Results show that the proposed two-stage framework achieves event-only MAE values of 0.53–0.76 MHz and RMSE values of approximately 1.0–1.4 MHz at most mid- and low-latitude stations, with larger errors at the high-latitude Casey station (MAE ≈ 1.45 MHz and RMSE ≈ 2.31 MHz). The consistently bounded MRE values (≈0.18–0.23) observed across multiple stations demonstrate that the framework effectively mitigates severe data imbalance and suppresses spurious high-intensity estimates under non-Es conditions.