Global Navigation Satellite Systems (GNSS) provide a promising opportunity for real‐time precipitable water vapor (PWV) sensing. However, relying on meteorological information restrains the implementation of real‐time inversion from tropospheric zenith total delays (ZTD) into PWV. In this study, a stacked machine learning model for mapping ZTD into PWV in the absence of meteorological parameters is developed. The model performance is assessed and validated with information offered by fifth generation European Centre for Medium‐Range Weather Forecasts reanalysis (ERA5) and radiosondes. An accuracy of better than 2.5 mm is achievable for the PWV values. Compared to the physical model which applies GPT3‐derived meteorological parameters, the proposed model reveals an enhanced performance, especially in the high‐latitude regions, with improvements of 28.1% and 22.2% when validated with ERA5 and radiosondes. This model is capable of fulfilling the demands of time‐critical meteorological applications and is also promising for real‐time PWV retrieval of other techniques that own the capability to sense ZTD.

Ocean surface winds play an essential role in regulating the earth’s weather and climate, and the Cyclone GNSS (CYGNSS) mission launched in 2016 is designed specially to monitor the ocean wind speed. In this study, an innovative model is developed based on a deep learning method to retrieve the ocean wind speed by making full use of the spatiotemporal information of CYGNSS observations. The proposed model named CNN-LSTM is established based on two modules, i.e., the Convolution Neural Network (CNN) module that extracts the spatial features around the Specular Point (SP) from a Two-Dimensional matrix of delay-Doppler Map (DDM) and the Long Short-Term Memory (LSTM) module which extracts the temporal features over a time series. The performance of the ocean wind speed derived from CNN-LSTM is assessed with the fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA5) products. The results show that the wind speed derived from CNN-LSTM reveals an accuracy of 1.34 m/s in terms of root mean square error (RMSE) values, showing an improvement of about 36.8%, 14.6%, 6.3%, when compared to the official retrieval algorithm called Minimum Variance Estimator (MVE), Multilayer Perceptron (MLP) net, and the CNN, respectively, confirming the feasibility and effectiveness of the designed method. Among all the experiments in this study which apply machine learning-based algorithms, the wind speed achieved by CNN-LSTM presents the smallest RMSE value. Furthermore, the error analyses of the wind speed retrieval in spatial and temporal scale are also discussed, which indicate the robust performance of CNN-LSTM model. The results show that the CNN-LSTM model proposed in this study contributes to offering efficient processing of Global Navigation Satellite Systems Reflectometry (GNSS-R) observations and fully exploits the capabilities of high-accurate ocean wind speed retrieval on a global scale.

Atmospheric water vapor plays an essential role in climate change and weather forecasting. However, monitoring water vapor with high spatial and temporal resolutions remains a challenge, especially over ocean regions where observations are insufficient. Shipborne global navigation satellite systems (GNSSs) contribute to enriching water vapor measurements over oceans and also can help validate satellite observations. Due to the lack of long-time serial observations, the performance of shipborne GNSS-derived precipitable water vapor (PWV) is inadequately evaluated on the global ocean scale. In this study, an overall assessment of shipborne GNSS PWV over global oceans is performed based on six voyages from 2014 to 2018. In coastal areas, the PWV differences of shipborne GNSS with respect to (w.r.t.) ground-based GNSS and ground-launched radiosonde data are 2.64 and 2.85 mm in the root mean square (rms), respectively. In open oceans, compared to ship-launched radiosonde profiles and satellite measurements, shipborne GNSS PWV shows the rms of differences of 2.54 and 2.53 mm, respectively. In addition, the rms of PWV differences between the whole track of shipborne GNSS PWV and National Centers for Environmental Prediction (NCEP) Climate Forecast System Version 2 (CFSv2) products is 2.96 mm. The intertechnique validations demonstrate that the accuracy of shipborne GNSS PWV is superior to 3 mm, which meets the requirements of climate research and numerical weather prediction (NWP).

Ocean surface wind is vital to the Earth’s meteorological system, and their properties can be detected by spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) measurements. With the growing number of GNSS-R signal sources, machine learning technology exhibits prominent advantages in wind speed estimation. Currently, the deep-learning techniques that establish relationships between GNSS-R measurements and ocean surface wind speeds generally apply grids and sequence structures and lack flexibility and robustness. Additionally, constructing models with individual GNSS-R observations results in the loss of valuable temporal correlation within Delay-Doppler Maps (DDMs). Therefore, this study proposes a novel spatiotemporal graph-based deep neural network (STG-DNN) for retrieving wind speed, which incorporates a graph module with a transformer module to fully exploit the spatial-temporal dependencies of DDMs. Results demonstrate that the graph module significantly improves both the accuracy and reliability in wind speed retrieval. Meanwhile, the transformer module effectively captures temporal features from various DDMs. Validations with Cyclone GNSS (CYGNSS) test data support the superior accuracy of STG-DNN, revealing a correlation coefficient of 0.92 for the wind speeds. The results indicate that the root mean square error (RMSE) of STG-DNN for wind speed is 1.27 m/s, representing improvements of approximately 33.2%, 20.6%, and 13.6% over the minimum variance estimator (MVE), convolutional neural network (CNN), and Vision Graph Neural Networks (VIG), respectively. Additionally, a promising agreement is observed between STG-DNN and ERA5 in the spatial distributions of wind speed retrieval, indicating a robust spatial performance in STG-DNN. As for the temporal scale, the daily variations in retrieval accuracy of STG-DNN exhibit smaller fluctuations compared to both CNN and VIG wind data in the test dataset.