Accurate identification of sea ice transition zones, where sea ice and open water gradually transform, remains a critical challenge for Arctic navigation and climate monitoring. Existing sea ice boundary detection efforts are primarily based on satellite observations, which are limited in spatial and temporal resolution. This study addresses this limitation by presenting the first Arctic application of shipborne global navigation satellite system reflectometry (GNSS-R) for real-time detection of sea ice transition zones. The method passively receives L1-band GNSS signals reflected from the sea surface, using power fluctuations-linked to changes in surface dielectric properties-as indicators of ice-water transitions. Two field experiments aboard the Chinese icebreaker Xuelong 2 demonstrate that GNSS-R can stably and repeatedly capture transition zone signatures with minute-level temporal resolution. Within each similar to 1 km experimental track, the retrieved sea ice concentration time series quantitatively reflects fluctuations corresponding to an effective spatial resolution of approximately 31 m, thereby enabling the detection of distinct signal characteristics between continuous ice edges and fragmented ice. Synchronized onboard video provides independent validation, showing overall detection accuracies exceeding 80%. This work establishes GNSS-R as a low-cost, passive, and high-resolution technique for operational monitoring of sea ice transition zones, supporting safer navigation and improved understanding of polar processes.

Global navigation satellite system reflectometry (GNSS-R) has emerged as a promising remote sensing technique for sea surface wind speed monitoring, owing to its high spatial and temporal resolution and cost-effectiveness. However, existing methods still face challenges in achieving high inversion accuracy, particularly under high wind speed conditions where errors tend to increase significantly. To address these limitations and improve the overall accuracy of sea surface wind speed retrieval, a novel hybrid model, MACT-RF, is proposed to enhance the accuracy of wind speed inversion using GNSS-R data. The model constructs a dual-path attention enhancement architecture: the SE-Convolutional Block Attention Module is employed to strengthen the convolutional neural network ability to extract detailed local features from delay Doppler maps (DDMs), while a Transformer network with self-attention captures global dependencies within auxiliary parameter sequences. The extracted features are fused and input into a random forest regression model for final wind speed prediction. Experiments based on Cyclone Global Navigation Satellite System satellite and ERA5 reanalysis data show that MACT-RF achieves a root-mean-square error of 1.15 m s-1 for overall wind speed inversion. Moreover, in the relatively high-wind-speed range of the dataset used in this study, the model predictions follow a trend consistent with the ERA5 reference winds, but they remain jointly constrained by the systematic biases inherent in ERA5 and the scarcity of high-wind samples. These results demonstrate that the proposed model offers strong stability and adaptability, and that the integration of multiple attention mechanisms effectively enhances the feature coordination capability of GNSS-R in challenging wind scenarios.

This paper presents GNSS radio occultation (RO) observational analyses on deducing the relationships and dependences between post-sunset EPB occurrences and EIA strength variability. The RO data were acquired from the FS7/COSMIC2 Program from 2020 to 2025. In this study, we incorporate both effects from crest peak electron density (Nemax) and crest-to-trough Nemax ratio and propose a new EIA strength parameter defined as the mean of northern and southern crest-to-trough Nemax differences to recognize and characterize the post-sunset EIA features. Both seasonal–longitudinal appearances of intense post-sunset EPB occurrences and strong EIA events occurred on more or less 30 days expanded from when and where magnetic flux tubes align with the sunset terminator at the magnetic equator but have more intense EPB and/or strong EIA days during southern (northern) hemispheric summers in the South American area (the Central Pacific area and the Africa area). It is well consistent with Tsunoda’s hypothesis during the evening pre-reversal enhancement (PRE) and reveals more informationt on day-to-day variability, intensities and extents of post-sunset EPB occurrences and EIAs subject to seasonal, longitudinal, and solar cycle variability. Moreover, the local-time evolutions of peak post-sunset EIAs occurred during 19~20 LT which is earlier than that of the obtained experimental peak (i.e., 20:20 LT) of post-sunset EPB occurrences. We expect that the post-sunset EIA detection could be a potential precursor for post-sunset EPB occurrence.

Global Navigation Satellite System Interferometric Reflectometry (GNSS-IR) provides scientists with an efficient method of measuring snow depth data with high spatial and temporal resolution, complementing existing snow products. The interferometric phase of the signal-to-noise ratio (SNR) is strongly correlated with the additional path delay, which is a key factor in altimetry applications. Existing altimetry applications are realized by analyzing the rate of change of the phase (i.e., the dominant frequency). However, there is a lack of in-depth exploration of the direct application of phase. Therefore, the initial focus of this study is to develop a geometric model between the interferometric phase derived from SNR and snow depth and to evaluate its performance at different stations. The secondary aim is to analyze the sensitivity of the phase to different snow states (accumulation and melting). The SNR data from three GNSS stations (PRDN, SG27, and P351) and four constellations (GPS, GLONASS, BDS, and Galileo) were analyzed. The results show that the retrieval results of the proposed method at different stations and at the same station but using different datasets are highly consistent with the conventional method, which demonstrates the feasibility and generalizability of the proposed method. For station P351, the root mean square error of the proposed method is also reduced by more than 20

In current conventional precise point positioning (PPP) processing strategies, the tropospheric zenith wet delay (ZWD) is usually dynamically estimated as a stochastic parameter. During the convergence period, ZWD estimates can appear to be negative or unrealistically large due to the low estimation precision, which adversely affects the estimation of other state parameters, especially the Up component of coordinates. To address this issue, we propose a method that incorporates physical constraints on ZWD in PPP processing. This method employs the inequality constrained least squares (ICLS), utilizing Karush–Kuhn–Tucker (KKT) conditions to add boundary conditions on ZWD. The boundary conditions of ZWD are calculated based on the relation between ZWD and relative humidity (RH). The use of physical constraints does not rely on external products or space state representation (SSR) corrections for ZWD during PPP processing and can improve the short-term accuracy of ZWD and coordinate Up component. The efficiency of this approach has been validated using GNSS data and products from GFZ operational networks. For real-time PPP solutions, there is a 30% improvement in short-term accuracy of Up component; for post-processing solutions, the short-term RMSE improvement is about 20% . After convergence, the ZWD upper bound is no longer applied as an ICLS constraint, but is instead used as a diagnostic indicator to identify ZWD anomalies. This indicator demonstrates high sensitivity and reliability under extreme weather conditions, highlighting its potential for application in meteorological hazard early-warning systems.

With growing concerns about climate change, increasing natural hazards, and extreme weather events, monitoring Earth’s surface parameters has become a critical area of interest for both the scientific community and society. Global Navigation Satellite System Reflectometry (GNSS-R) is an innovative and low-cost technique that exploits existing Global Navigation Satellite System (GNSS) signals after reflection from Earth’s surface. GNSS-R constellations offer unique observations with unprecedented data volume, temporal resolution, and spatial coverage across the entire globe under all-weather conditions. As the data volumes are continuously accumulating, the trend in applying Artificial Intelligence (AI) is expanding. However, current AI models rely heavily on labelled data, feature engineering, and extra fine-tuning, leading to high computational and labor costs. To address these issues, we propose the project EcoGEM: Energy-efficient Multimodal GNSS Reflectometry Models for Generalist Earth Surface Monitoring and Hazard Response.EcoGEM develops cutting-edge Earth observation foundation models using GNSS-R measurements and integrates them with other remote sensing data. It pioneers the first general-purpose GNSS-R foundation models and curated multimodal datasets to support climate science, hazard detection, and environmental monitoring. Unlike task-specific methods, the proposed models adapt across applications such as soil moisture, vegetation water content, and ocean wind speed. Uniquely, EcoGEM emphasizes energy-efficient AI through model pruning, knowledge distillation, and dynamic architectures, enabling deployment on edge devices and small satellite platforms. This collaborative project of GFZ and DLR advances sustainable AI and promotes novel and open-access tools for Earth scientists, environmental policymakers, and global users.

The diurnal cycle of Integrated Water Vapor (IWV) plays a key role in radiation, convection, and land–atmosphere interactions, yet its global characteristics and representation in atmospheric reanalysis products remain insufficiently quantified. Using a decade of GNSS observations from more than 6,000 stations, this study characterizes the global IWV diurnal variability and assesses the performance of ERA5 in reproducing these signals. GNSS IWV exhibits a coherent diurnal cycle dominated by the S₁ (24 h) harmonic, with annual-mean amplitudes decreasing from 3 kg m⁻² in the tropics to near zero at high latitudes, a global mean of 0.40 kg m⁻², and peak times clustered around 17.4 local time; the semidiurnal S₂ (12 h) component is weaker but similarly systematic. In contrast, ERA5 exhibits sharp artificial discontinuities at the transition times (09–10 and 21–22 UTC) of its 12-hour assimilation windows, affecting 54

Abstract Real-time GNSS precise point positioning (PPP) is an essential tool for numerous applications in the Earth sciences, navigation, surveying, as well as in early warning systems for geo-hazards. It requires precise information about satellite orbits and clock offsets that have to be distributed to the user as correction streams with a delay of only a few seconds. While satellite orbits can be predicted with high precision for a few hours, satellite clocks have to be estimated from observation data in real-time. In this contribution, we introduce the new GFZ in-house real-time GNSS network analysis software. GPS, GLONASS, and Galileo satellite clocks are determined every 5 s with a recursive least-squares estimator, making use of a sequential scalar implementation of the Kalman filter. For the detection of cycle slips, the computationally efficient concept of single-receiver, single-channel integrity is applied. The generated products are evaluated by means of a comparison to the post-processed CODE rapid solutions and with real-time PPP examples. The new GFZ corrections streams are available through the Real-Time Service (RTS) of the International GNSS Service (IGS).

Abstract. This study evaluates the Weather Research and Forecasting (WRF) model, with and without WRFDA 3D-VAR data assimilation (DA), for precipitation and temperature forecasts in Luxembourg and the Greater Region during the July 2021 flood event. Conventional observations (CONV), Global Navigation Satellite System (GNSS) Zenith Total Delay (ZTD), and their combination (CONV + ZTD) were assimilated using a 6-hour rapid-update cycle over a full month (20 June–20 July 2021), and verified against independent surface stations withheld from the assimilation, together with radar and GPM IMERG data. For precipitation, DA improved categorical skill: CONV yielded the largest gains in bias (+37.4 %) and probability of detection (+18.3 %), at the cost of a higher false alarm ratio, while CONV + ZTD gave more moderate but balanced improvements. Absolute-error metrics (RMSE, MAE, SMAPE) changed little and mostly not significantly, indicating that DA improves event detection rather than magnitude. For temperature, all configurations reduced bias, ZTD being the most effective (+97.4 %). Station-level Wilcoxon signed-rank tests confirm that the bias and detection gains are statistically significant (p < 0.001). The results demonstrate the complementary roles of conventional and GNSS-based observations for regional numerical weather prediction.

Global Navigation Satellite Systems (GNSS) are based on measuring signal propagation time, so that clock information is required for both the transmitting satellite and the receiving ground station. For Precise Point Positioning (PPP), synchronization errors of the receiver clock are usually estimated as epoch-wise biases with white noise as stochastic behavior. A major issue for the receiver clock estimates is the high correlation with the station height and the tropospheric zenith delay that results from the observation geometry. Introducing models to reduce the number of unknown clock parameters is one way to mitigate these correlations. However, adequate modeling requires a high degree of stability for the corresponding clock.In this contribution, we show the results of modeling highly stable GNSS receiver clocks in PPP using piece-wise linear representations. We discuss different options to control the deterministic and stochastic part of the piece-wise linear model (interval length, parameter weighting, etc.). For 56 highly stable hydrogen maser (H-maser) stations of the International GNSS service (IGS), the impact of piece-wise linear clock modeling on correlated parameters, especially the sub-daily height estimates, is presented. The differences in the modeling impact of the individual receiver clocks are explained by categorizing the stations based on statistical and observational quality measures. In addition, the effects of individual processing options (used GNSS constellations, elevation cutoff angle) are shown.

The ionosphere is the layer of the Earth’s upper atmosphere containing free electrons, where solar radiation ionizes atoms and affects GNSS signals. Airborne GNSS campaigns offer a unique opportunity to monitor ionospheric variations. This study presents a comprehensive analysis of Total Electron Content (TEC) using data acquired during the GEOHALO Mission over Italy and adjacent parts of the Mediterranean Sea. The dataset consists of GNSS observations collected by the HALO (High Altitude Long Range) research aircraft during flights conducted on 6th, 8th, 11th and 12th June 2012, at an approximate altitude of 3500 metres over sea areas.The methodology is based on RINEX data extracted from binary receiver data (JAVAD GNSS) using RTKLIB. The GOPI Tool, a free software for the retrieval of TEC using dual-frequency GNSS data is used to estimate preliminary Slant TEC (STEC) and Vertical TEC (VTEC). However, GOPI assumes static position which is not sufficient for geo-referencing airborne results. Therefore, an additional processing step is introduced to compute Ionospheric Piercing Points (IPP) along the aircraft trajectory. A geometric ray tracing is applied to determine the IPP for the aircraft’s position assuming an ionospheric shell (~350km, F-layer). The analysis gives spatially referenced profiles of STEC and VTEC, for corresponding IPP along the flight trajectory.Preliminary results confirm the expected enhancement of STEC at lower elevation angles and local maximum of VTEC is observed in the early afternoon on 8th,11th and 12th June 2012. However, in further steps, these results need to be validated, for example, using TEC Maps or the Neustrelitz Electron Density Model (NEDM).

Artificial intelligence (AI) models developed for Global Navigation Satellite System Reflectometry (GNSS-R) observations have demonstrated competitive performance in estimating geophysical parameters, especially ocean surface wind speeds. However, the transition from transparent physical scattering models to complex deep learning architectures raises concerns regarding reduced model transparency and trust. Understanding the decision-making processes of these "black-box" models is essential for assessing model behavior, detecting anomalies, and ensuring reliability in AI-based Earth observations. In this study, we investigate the role of explainable artificial intelligence (XAI) in addressing the transparency gap for hybrid deep learning models designed for GNSS-R observations. Focusing on ocean wind speed retrieval as a well-characterized benchmark, our study is structured around three primary objectives: first, assessing the robustness and efficiency of XAI explainers, second, interpreting a benchmark hybrid model trained using a manually selected feature set with Shapley additive explanations (SHAP) and gradient-weighted class activation mapping (Grad-CAM), which provide quantitative branchwise attribution and qualitative spatial saliency, and third, proposing an XAI-based feature selection pipeline that leverages SHAP-based ranking and exclusion, comparing its efficacy against conventional statistical methods. The results demonstrate that SHAP is effective not only for model interpretation but also for supporting computationally efficient feature selection and model debugging. Meanwhile, Grad-CAM offers complementary spatial interpretability by highlighting salient regions in the delay-Doppler map inputs. This study demonstrated the potential of integrating XAI as a diagnostic and validation tool into the model development cycle, enabling more transparent, robust, and trustworthy AI models for upcoming GNSS-R missions and future applications.

This paper presents all-sky airglow image analyses from OI 630-nm fisheye lens (FEL) photographs on the equatorial plasma bubble (EPB) observations over Taiwan and the nearby area. A FEL image calibration procedure of using single image with concentric circle and radial line control features has been described and applied to transform the raw FEL images into the expected perspective projection images. Two approximations on FEL image transformation have been proposed and evaluated to model the varying spatial resolution using polynomial and cubic spline interpolation functions. The validity and error analyses of the FEL image transformations are also demonstrated by testing a synthetic spiral coil image. For further applications, a low-cost miniature all-sky imaging system using Sony IMAX462-mode FEL camera and open-source indi-allsky operation software was set up and operated at the Lulin Mountain (23.47°N, 120.87°E; 2842 m altitude), Taiwan. The obtained field of view is from around 19.2°N to 27.8°N latitude (geomagnetic latitude 10.12°N ∼ 18.74°N) and 116.6°E to 125.2°E at an assumed airglow emission height of 250 km. We demonstrate the experimental all-sky airglow observations and obtain bifurcated and parallel plasma depletions separately during two events of post-sunset and midnight EPBs occurred before and after the spring equinox of 2025, respectively. The FEL image distortion calibration technique yields a “consistent” determination of nighttime plasma drifts from the transformed all-sky images on EPB observations and motion analyses. The derived eastward plasma drifts were higher at lower latitudes than those at higher latitudes and showed negative latitudinal gradients of the zonal plasma velocities during both events.

Modeling highly stable GNSS receiver clocks helps to increase the stability of correlated parameters, especially the station height. The use of a piece-wise linear receiver clock model, which is a common method for clock parametrization in Very Long Baseline Interferometry, has not yet been extensively investigated for GNSS. In this work, we examine piece-wise linear receiver clock modeling for GNSS. Based on a one-month series of kinematic Precise Point Positioning analyses for 56 highly stable hydrogen maser stations, the effects of piece-wise linear clock modeling are demonstrated. For the stochastic part of the model, the process for determining the optimal relative constraint weight is explained in detail, whereas the deterministic component consists of piece-wise linear segments with interval lengths of 8, 4 and 2 hours to represent the clocks of highly stable hydrogen maser stations. Compared to the epoch-wise clock estimation, there are no significant differences for the tropospheric zenith delay. However, applying a piece-wise linear model leads to more stable height estimates. The larger the modeling interval, the more stable the sub-daily height estimates get: the median standard deviation decreases by -1.6 mm (-10%) for 2-hour and -3.0 mm (-20%) for 8-hour intervals. Stations with a higher error of unit weight and a smaller number of observations tend to show larger improvements in height stability (up to 45%). When observations are limited by satellite system exclusion or high elevation cutoff angles, piece-wise linear clock models become especially advantageous in terms of height stability.

A method for estimating Antarctic sea ice concentration using shipborne GNSS-R technology is presented, achieving a high temporal resolution of 20 min. By analyzing the changes in the power of GNSS-reflected signals interacting with different ice surfaces, the sea ice concentration estimated by this method shows a strong correlation with NOAA's sea ice data and the AMSR-E/AMSR2 datasets. NOAA's dataset provides daily data with a spatial resolution of 25 km & times; 25 km, while the AMSR-E/AMSR2 dataset offers daily data with a spatial resolution of 12.5 km & times; 12.5 km. This method was validated by comparing it with these existing high-resolution datasets, demonstrating the reliability of the shipborne GNSS-R estimates and highlighting its higher temporal resolution. This study was conducted aboard the Tianhui Ship, a support vessel for China's 40th Antarctic Expedition, starting on 18 November 2023. The ship's global navigation satellite system (GNSS) antenna was positioned 33 m above the waterline. This approach compares reflected signals to direct GNSS measurements, including key variables, such as sea ice concentration and satellite elevation angle. The method is highly sensitive in detecting dynamic sea ice conditions, accurately identifying ice boundaries and changes in sea ice concentration. Its strong performance highlights the potential of shipborne GNSS-R in polar research, providing valuable tools for environmental monitoring and ice dynamics analysis. Future improvements, including the use of multiconstellation GNSS systems and additional remote sensing datasets, are expected to expand the method's applicability, reinforcing its importance in polar environmental studies.

Grazing Global Navigation Satellite System Reflectometry (GNSS-R) altimetry, while promising for precise surface height retrieval, faces challenges from tropospheric refraction. Beyond path delay errors, tropospheric refraction can cause an asymmetry in the incident and reflected signal paths due to tropospheric gradients and the asymmetric observation geometry. The impact of this refraction asymmetry on grazing GNSS-R altimetry remains unclear. This study systematically quantifies this impact and characterizes the resulting errors in grazing observations. We propose a new surface height retrieval model that accounts for refraction asymmetry and applies to both spaceborne and airborne scenarios. Through a 30-day global GNSS-R simulation, in conjunction with European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis data and a ray tracing technique, our results show that decimeter-level systematic bistatic delay biases can be observed in airborne scenarios (3.5 km), increasing exponentially with decreasing elevation angles. In contrast, these errors are millimeter-level in spaceborne scenarios (530 km). A comparison with the traditional height retrieval model reveals that uncorrected refraction asymmetry in airborne observations can introduce decimeter- to meter-level altimetric errors at extremely low elevation angles (2–7 degrees). This finding highlights the critical importance of accounting for the effects of refraction asymmetry on airborne grazing GNSS-R altimetry, especially for continuous coherent observations covering a wide range of low elevation angles.

An improved numerical and climatological model, named the Taiwan Ionospheric Model-II (TWIM2), of global ionospheric electron density (Ne) has been investigated in this study. Based on the near-vertical Ne profiles retrieved from the FS7/COSMIC2 radio occultation observations, the TWIM2 exhibits vertically fitted oh-Chapman layers, with distinct F2, F1.5, F1, E, and D layers, and has been improved on the surface spherical harmonics approaches of the fitted Chapman-layer peak density (Nemax), peak density height (hm), and scale height (H) from geographic coordinate system to geomagnetic coordinate system. On average the mean error values from the NemaxF2 modeling in geomagnetic coordinate system are around 10 % less than those in geographic coordinate system. The result is consistent with those of surface spherical harmonic coefficient analyses and implies that the surface spherical harmonic analyses applied to three-dimensional Ne modeling in geomagnetic coordinate system are more effective. The TWIM2 results have been evaluated by the investigations of equatorial ionospheric anomaly (EIA) specification with diurnal and seasonal variations in a universal-time modeling mode or with latitudinal and longitudinal variations in a local-time modeling mode. We have applied a new proposed EIA strength (EAS) parameter, defined as the mean crest-to-trough Nemax difference, to identify and separate the post-sunset EIAs from the daytime EIAs and obtained daytime and post-sunset EIA peaks occurring at around mid-afternoon and 20.5 LT, respectively. Moreover, based on the derived TWIM2 NemaxF2-and hmF2-modeling maps at the local times with daytime or post-sunset EIA peaks, both the EAS and average low-latitude hmF2 profiles reveal wavenumber-4 longitudinal structures with similar longitudes at wave peaks. (c) 2025 The Author(s). Published by Elsevier B.V. on behalf of COSPAR. This is an open access article under the CC BY license (http:// creativecommons.org/licenses/by/4.0/).

Under ongoing Arctic warming, robust monitoring of coastal sea ice freeze–thaw dynamics requires indicators that are not only physically interpretable but also quantitatively validated. Although Global Navigation Satellite System Interferometric Reflectometry (GNSS-IR) is a promising ground-based approach, existing studies primarily focus on qualitative interpretation of time-series patterns, with limited quantitative assessment of metric performance under heterogeneous coastal environments. To address this gap, this study introduces the Spectral Area Factor (SAF), derived from spectral analysis of GNSS signal-to-noise ratio data, and systematically quantifies its monitoring capability using a Bayesian classification framework. Using five years (2018–2022) of observations from the Arctic coastal station TUKT in Canada, we estimate class-conditional statistics from 2018–2020 and evaluate the performance on independent data from 2021–2022. The proposed SAF shows strong discriminative capability between sea ice and sea water compared to conventional feature factors and provides a stable response during transitional periods. A multi-frequency Bayesian fusion model using SAF at GPS L1/L2/L5 bands achieves an accuracy of 92.72% for near-real-time arc-level classification and 98.63% for daily-scale post-processing classification. More importantly, the model outputs class probabilities rather than conventional binary labels, allowing it to accommodate the inherent spatial heterogeneity of coastal sea surface states. The resulting probabilistic time series supports near-real-time coastal sea ice state inference with hourly-scale updates and provides quantitative information on local sea ice conditions within the study area. In addition, the framework is operationally flexible, enabling adaptive fusion of available frequency combinations and offering a scalable path for future extension to multi-constellation GNSS observations. Overall, these results demonstrate a practical route from feature design to quantitative, application-oriented coastal sea ice monitoring using existing ground-based GNSS infrastructure.

Abstract. In this study, we investigate the impact of assimilating Global Navigation Satellite System (GNSS) Zenith Total Delays (ZTDs) and tropospheric gradients (TGs) on the simulation of Super Typhoon Koinu (2023) using the Weather Research and Forecasting (WRF) model. Four data assimilation experiments are conducted: a control experiment assimilating conventional observations only, experiments separately assimilating ZTDs and tropospheric gradients on top of conventional observations, and a combined experiment assimilating both ZTDs and tropospheric gradients on top of the conventional observations. A two-month assimilation experiment using observations from approximately 250 GNSS stations demonstrates the overall benefit of GNSS data, with reductions in root-mean-square error (RMSE) of up to 60 % in ZTD when both ZTDs and tropospheric gradients are assimilated. Assimilation of tropospheric gradients alone also reduces the ZTD RMSE, demonstrating that gradients provide independent and complementary information beyond the vertically integrated moisture constraint of ZTDs. The impact of Typhoon Koinu is then examined as it approaches Taiwan between 3 and 6 October 2023. Continuous assimilation of GNSS observations improves the representation of the atmospheric moisture field surrounding the cyclone, leading to changes in the spatial distribution of integrated water vapor and the azimuthal moisture asymmetry around the storm. These moisture adjustments produce small but systematic improvements in the environmental deep-layer steering flow and, consequently, modest reductions in typhoon track error after landfall over Taiwan. The combined assimilation of ZTDs and tropospheric gradients also provides the most accurate simulation of cyclone intensity. The results demonstrate that tropospheric gradients complement ZTDs by improving the representation of horizontal moisture variability, thereby enhancing the simulation of the tropical cyclone environment. This study highlights the potential of assimilating multiple GNSS tropospheric products to improve tropical cyclone prediction in numerical weather prediction systems.

As Global Navigation Satellite System (GNSS) signals are affected by the atmospheric refraction when transporting through the neutral atmosphere, the resulted tropospheric delay is highly linked to the atmospheric water vapor, promoting the establishment of GNSS as a reliable meteorological observation means of high accuracy and high temporal resolution. Over the past few decades, the advancement of GNSS constellations including GPS, BeiDou, Galileo, and GLONASS has driven progress in tropospheric parameters retrieval, modeling and applications. This paper presents a detailed review of the state of the art in ground-based GNSS troposphere monitoring and applications, covering fundamental concepts, theory, and algorithms. The advances in GNSS tropospheric parameters estimation for both high-precision and low-cost processing are outlined, followed by the progress in tropospheric delay modeling and forecasting. Meanwhile, a recent development on the applications of the external tropospheric models in augmenting GNSS precise positioning is illustrated. Thereafter, progress in the assimilation of GNSS tropospheric data into numerical weather prediction (NWP) models is presented in detail, accompanied by the advances in its direct applications in rainfall forecasting. Furthermore, we outline the key challenges and opportunities ahead, including leveraging Low Earth Orbit (LEO) and smart terminals for advanced tropospheric products, establishing more enhanced delay models in challenging environments particularly cooperating with real-time data transmission, developing more sophisticated data assimilation operators, and coupling deep learning models with physical mechanisms for improved atmospheric simulation. Finally, the paper concludes with some comments and the prospects for future research.