Robust and high-precision vehicle positioning information is crucial for Internet of Things (IoT) applications like autonomous driving and intelligent transportation. While global navigation satellite system (GNSS)/inertial navigation system (INS) integration ensures reliable positioning in open areas, signal interruptions in urban environments cause rapidly accumulating INS errors. Deep neural networks (DNNs) are integrated into the Kalman filter (KF) framework to mitigate INS errors through learning-based data fusion. If the DNN provides unreliable pseudo-measurement information, the positioning results may deteriorate. To enhance DNN estimation reliability, this study introduces a 1-D convolution-based spatiotemporal attention mechanism for implicit modeling of temporal dependencies and spatial correlations in sensor data. This mechanism is utilized to construct a multitask odometry network (MT-ONet) for accurate forward velocity and motion state estimation of vehicles. Building upon this, an adaptive fusion vehicle positioning algorithm is proposed. This algorithm combines the MT-ONet and an improved adaptive KF (AKF) to dynamically adjust measurement noise to enhance robustness. Experimental results indicate that MT-ONet offers high estimation accuracy and low computational complexity, making it suitable for deployment on resource-constrained IoT devices. During simulated 180-s GNSS outages, the proposed method decreases horizontal root mean square (RMS) and maximum errors by 14.47% and 16.09% compared to hardware odometer-assisted scheme, and by 33.88% and 40.28% versus the nonholonomic constraint (NHC)-assisted scheme.

Recent decades have witnessed the emergence of the Global Navigation Satellite System (GNSS) Zenith Tropospheric Delay (ZTD) as a pivotal element within the domain of GNSS meteorology. The spatial density of ZTD data is contingent on the number of regional GNSS stations, thereby influencing its application in regional meteorology. This research presents a novel model, GWR-CNN (Geographically Weighted RegressionConvolutional Neural Networks), which utilizes convolutional neural networks to address geographically weighted regression residuals. Utilizing the French region as an illustrative case, this model necessitates solely the acquisition of positional information (comprising precise longitude, latitude, and height) and the ZTD values of the GNSS stations to execute high-precision interpolation calculations for regional ZTD. It is noteworthy that the outcomes of high-precision interpolation for ZTD can closely approximate actual observations, thereby augmenting the predictive capability of the Weather Research and Forecasting (WRF) model for regional rainfall through the process of data assimilation.

Global Navigation Satellite System (GNSS) meteorology utilizes the Zenith Tropospheric Delay (ZTD) experienced by signals traversing the troposphere, which reflects atmospheric conditions. The Weather Research and Forecasting (WRF) model, a mature numerical weather prediction system, often exhibits biases in surface meteorological parameter forecasts (temperature, humidity, pressure). To address this, we developed the GeoAwareCorrector deep learning model, which directly corrects WRF forecasts using GNSS station ZTD data. Testing on hourly first-day forecasts per month (2021-2023) in Northwest France showed GeoAwareCorrector outperformed Linear Regression and Multilayer Perceptron models. It significantly improved forecasts for all three variables, reducing WRF Mean Absolute Error by 66.5% (pressure), 22.5% (temperature), and 6.1% (humidity). This demonstrates GNSS ZTD’s potential to enhance surface meteorological numerical forecasting.

Rapid and accurate retrieval of vertical water vapor distributions is important for numerical weather forecasting, climate research, and disaster management. Traditional methods, such as radiosondes and microwave radiometers, often suffer from limited spatial and temporal resolution, high operational costs, and delayed data availability, leading to variable accuracy. The FY-4A meteorological satellite has introduced the Infrared Hyperspectral Atmospheric Vertical Sounder sensor to geostationary orbit for the first time, which offers significant advantages in terms of observational efficiency and cost compared to ground-based GNSS. This study employs a 1-D convolutional neural network (1D-CNN) to reconstruct specific humidity at 21 pressure levels, ranging from 1000 to 100 hPa. The 1D-CNN integrates two data sources: radiative data from 560 water vapor channels on FY-4A/geostationary interferometric infrared sounder (GIIRS) and ground-based GNSS-derived precipitable water vapor (GNSS-PWV) data. The combined use of FY-4A/GIIRS and GNSS-PWV data significantly improves the accuracy of water vapor vertical profile retrievals compared to radiosonde data from Hong Kong. Under clear-sky conditions, this combination achieves a root mean square error (RMSE) reduction of 25.64% overall, with a 30.68% reduction at heights from the surface to 600 hPa, as well as an average bias (BIAS) reduction of 32.98%, compared to retrievals using FY-4A/GIIRS data alone. Under cloudy conditions, the results showed an average reduction of 23.66% in RMSE and 50% in BIAS. The results demonstrate that integrating ground-based GNSS-derived PWV data can effectively increase the accuracy of FY-4A/GIIRS water vapor vertical profile retrieval.

The application of Global Navigation Satellite System (GNSS) meteorology has gained prominence in recent years, and the GNSS-derived zenith wet delay (ZWD) has become a focus of research due to its intrinsic connection to atmospheric water vapor dynamics. However, the inherent spatiotemporal variability of ZWD poses significant challenges in achieving precise estimations, thereby constraining its data utility. To address this limitation, we propose a deep geographically weighted regression network (DeepGWR-Net) model designed for high spatiotemporal resolution (0.1°, 1h) ZWD estimation. This model calculates ZWD using geographic coordinates (longitude, latitude, elevation), temperature, and pressure parameters. The model demonstrated excellent performance, with MAE and RMSE values of 6.46 mm and 8.30 mm, respectively, representing improvements of 41.2% and 23.9%, and 39.5% and 24.3% over the multiple linear regression method and convolutional neural network model, respectively. Leveraging ERA5-Land reanalysis data (0.1°, 1h), we generated high-resolution ZWD grids by incorporating three-dimensional positional information, temperature, and pressure data. A comprehensive spatiotemporal coupling analysis was conducted using a persistent precipitation event over mainland France and adjacent regions from October 18 to November 18, 2023, employing correlation analysis, lag analysis, and geographically weighted regression techniques. Key findings reveal predominantly positive ZWD-precipitation correlations across most regions, with localized areas influenced by the Mediterranean climate, showing a mixture of positive and negative correlations. It is worth noting that ZWD exhibits a characteristic 3-12 hour window of dynamic changes before precipitation events. Furthermore, the response of ZWD to short-duration intense rainfall is significantly more pronounced across diverse rainfall types, and extreme rainfall events show a synchronized surge effect between ZWD and precipitation intensity. Collectively, these results substantiate significant regional spatiotemporal covariation and spatial coupling mechanisms between ZWD and precipitation patterns.

This study analyzed variations of the global tropopause structure using radiosonde products from 2010 to 2020. First, the covariance transformation of the logarithm of the refractivity (CTLR) method was successfully applied to radiosonde data to determine tropopause, which also contributed to solve the problem that CTLR cannot be widely utilized on radiosonde data due to low top height. Then, nearly 1000 radiosonde stations’ observed data from 2010 to 2020 were collected to determine global tropopause with the CTLR method. Finally, the study analyzed the characteristics of changes in global tropopause in the last decade. The results revealed a continuous rise in tropopause from 2010 to 2020 and it increased at a rate of 11–13 m/year. The measurements from the lapse rate tropopause (LRT) method are consistent with those results. The continuous rise of tropopause (predominantly mid-latitude) is primarily due to tropospheric warming and stratospheric cooling. This increasing trend remains after removing natural variability, which can provide further observational evidence for global warming due to human activities.

Three-dimensional water vapor data are now being used for numerical weather prediction, which is effective for monitoring extreme weather events and improving forecast quality. This study focuses on reconstructing the 3-D water vapor field using Global Navigation Satellite System (GNSS) water vapor tomography techniques, addressing two main aspects: 1) Achieving high-precision real-time 3-D water vapor predictions as initial values. In this study, a novel high-precision water vapor prediction model, the Informer-WV model, is introduced, and its predictions can be served as the initial values for tomography. We trained the Informer-WV model using 5 years of historical ERA5 reanalysis data in Hong Kong (HK) region to obtain the real-time values from sliding-window predictions. The model demonstrated a remarkable prediction accuracy, with an annual root mean square error (RMSE) better than 0.80 g/m3 compared to the actual ERA5 values. 2) The upper boundary height of the 3-D tomography grid is determined by the vertical precision of initial values, which is adjusted to 5.2 km in this study, and the reconstructed slant water vapor (SWVs) are calculated with the predictions. By benchmarking against radiosonde data, we analyzed the near-real-time tomography inversion results for the two weakest prediction periods of the model. The RMSE of the water vapor inversion values derived from the optimized method was reduced from 1.55 to 1.26 g/m3, and the most significant improvement is at about 2–5 km. This approach not only improved the accuracy by 19% relative to the initial predictions but also significantly outperformed the traditional tomography method.

Using the global navigation satellite system (GNSS) data from the Hong Kong region, this study comprehensively investigates the ionospheric irregularities responses to strong geomagnetic storms over the past two solar cycles 2001–2020. Based on the geomagnetic index Dst, a total of 64 strong storms are confirmed during 2001–2020. Statistical results indicate that for the total 64 strong storms, only 20 storms are considered to trigger irregular occurrences. When the occurrence local time (LT) of the minimum dDst (dDst$_{\mathrm {min}}$) is in 10:00–14:00 LT, no ionospheric irregularities occurred at nighttime although there is a total of 14 strong storms, while that of dDst min is in the nighttime of 18:00–21:00 LT, ionospheric irregularities are detected in ten out of 12 strong storms. For the two special storms on 19 April 2002 (dDst min occurred at 21:00 LT) and 23 May 2002 (dDst min occurred at 20:00 LT), they did not trigger the generation of ionospheric irregularities although their dDst min occurred in 18:00–21:00 LT. Based on vertical total electron content (VTEC) derived from global positioning system (GPS) measurements, it is found that the westward electric fields during two storms should play a vital role to inhibit the nighttime ionospheric irregularities (NIIs) occurrence. This study suggests that caution should be taken when the dDst min determined LT is used to decide the occurrence of nighttime irregularities.

This article analyzes the contribution of the delay-Doppler map (DDM) observation noise to the uncertainty of global navigation satellite system reflectometry (GNSS-R) ocean scatterometry observables and the retrieved wind speeds. For this purpose, the parameter$K^{p}$, which is commonly used in the traditional microwave scatterometer, is introduced to characterize the relative standard deviation (RSD) of the GNSS-R normalized bistatic radar cross section (NBRCS) measurement. Based on the noise covariance of the DDM measurements, the analytic expressions of RSD are derived for two cases, i.e., the NBRCS computed with one single DDM bin at the specular point and the NBRCS computed from$M\,\,\times \,\,N$DDM bins around the specular point. By analyzing the dependence of the RSD on different system, geometry, and instrument parameters, the simplified models of RSD are derived empirically. As a simple application of the proposed models, the wind speed retrieval errors are computed using parameters from the Cyclone GNSS (CYGNSS) mission. It shows that the wind speed retrieval error due to thermal noise and speckle can be an important error source for the overall wind speed retrieval performance, especially at high wind speed and with the high incidence angle GNSS-R measurements.