Chen Zhou

Real‐time water vapor monitoring is critical for meteorological applications such as weather forecasting and nowcasting. However, accurate real‐time retrieval of precipitable water vapor (PWV) from Global Navigation Satellite Systems (GNSS) remains challenging, especially for GNSS stations lacking meteorological sensors. This study introduces a global forecast‐based GNSS PWV retrieval method leveraging the artificial‐intelligence (AI)‐based global weather forecast model named GraphCast. We utilize GraphCast to predict key tropospheric parameters (weighted mean temperature, T _m , and zenith hydrostatic delay, ZHD), which are then combined with GNSS observations to obtain PWV. To assess the accuracy of GraphCast‐derived PWV, a comprehensive evaluation experiment is conducted through comparison with two baseline models: the Global Pressure and Temperature 3 (GPT3) model and stacked machine learning (ML) model, using global radiosonde and GNSS data sets. Evaluation results demonstrate that GraphCast‐based PWV significantly outperforms the baseline models across three metrics: root mean square error (RMSE), mean absolute error, and coefficient of determination ( R ² ). Specifically, the RMSE of GraphCast‐derived PWV is 0.73 mm when validated against radiosonde data sets, a significant improvement of more than ∼1.3 mm over the baseline models. Similarly, the RMSE is 1.7 mm when validated against GNSS data sets, reflecting an improvement of more than ∼0.7 mm. In addition, GraphCast‐based PWV also exhibits lower sensitivity to latitude, altitude, and temporal variations. This new approach demonstrates strong potential for achieving global high‐precision, real‐time PWV retrieval independent of local meteorological observations.

The hydroxyl (OH) radical is a key species in stratospheric chemistry, driving ozone loss and the coupling of other chemical families. However, global OH observations are very sparse, with a critical gap in satellite-based observations since 2009. Although alternative chemical methods attempt to mitigate such gaps, they have limitations. Here, we present DRCAT, a chemistry-informed deep learning network to predict stratospheric OH profiles using coincident satellite observations of chemically related species. Trained on only 2 years of data, DRCAT provides reconstruction of a continuous and reliable global stratospheric OH dataset from late 2004 to the present. Results show that DRCAT outperforms both a chemical transport model and an observation-constrained steady-state approximation. Moreover, DRCAT successfully predicts anomalous OH enhancements after the 2022 Hunga volcanic eruption, showing generalizability to extreme conditions outside of the training regime. DRCAT thus provides a robust pathway to fill OH observational voids. Beyond OH, this work offers a scalable framework for similar reconstruction of other key short-lived species.

The transmitting source in evaporation duct propagation modeling is often approximated by an ideal Gaussian pattern, but the influence of fine low-elevation radiation-pattern details near the horizon remains insufficiently quantified. This letter investigates this effect using a three-level source model composed of an ideal Gaussian source, an engineering reference source, and a low-elevation perturbed source. Parabolic equation simulations are performed in range-independent and range-dependent evaporation duct environments to compare source-induced propagation differences and their relative importance with respect to duct parameter uncertainty. The results show that the discrepancy between the ideal Gaussian and engineering reference sources is limited, whereas subdegree low-elevation perturbations can produce pronounced propagation loss variations. Although duct uncertainty remains the dominant contributor, source uncertainty accounts for 29.7%-36.4% of the full field error variance and 25.2%-32.9% of the receiver-height error variance. A two-metric coupling representation further characterizes the source-induced errors with coefficients of determination of 0.934-0.969.

The accurate modeling of sea clutter for over-the-horizon (OTH) radar constitutes a classical multi-scale physical problem, necessitating the precise coupling of kilometer-scale propagation links with meter-scale sea surface roughness characteristics. In order to remediate the physical inconsistencies engendered by the decoupled calculation of radio wave propagation and sea surface scattering in prevailing methodologies, this paper puts forth a proposed integrated propagation-scattering physical simulation framework. The framework's foundation is predicated on the implementation of a bidirectional propagation-scattering coupling mechanism, which integrates the propagation loss correction induced by sea surface roughness with a high-fidelity scattering kernel incorporating two-stage foam evolution and multipath effects into a unified framework. At the implementation level, a heterogeneous parallel architecture based on C++/CUDA is constructed, which effectively resolves the challenge of electromagnetic scattering calculation for massive time-varying sea surface facets. Using measured data from the South China Sea, the model's validity in describing the statistical distribution and spectral characteristics of sea clutter is substantiated. Moreover, a series of numerical experiments are conducted based on this model to investigate the impact of range-dependent atmospheric ducts on the characteristics of long-range sea clutter. The simulation framework under consideration provides critical data support and a theoretical basis for the performance analysis of OTH radar systems and the development of high-precision atmospheric duct inversion algorithms.

Path loss (PL) map prediction is an important task for wireless propagation modeling and network planning. Neural network (NN)-based approaches have been widely studied due to their higher efficiency than ray tracing (RT). However, most existing methods focus on two-dimensional (2D) PL maps. To address the three-dimensional (3D) PL map prediction problem, this letter proposes a multi-altitude PL map prediction framework based on an NN architecture. The proposed model adopts a dual-encoder design to extract heterogeneous features from the environment and the transmitter (Tx) location, and uses a height-aware feature fusion module to embed altitude conditioning into the fused features, enabling the network to differentiate PL patterns across altitude planes. Experimental results show that the proposed model improves prediction accuracy over representative baselines, and substantially reduces inference time compared with RT.

Accurate radio propagation modeling through vegetation is challenged by highly irregular three-dimensional tree geometry and uncertain electromagnetic parameters, especially when scattering is non-negligible. This letter presents a measurement-calibrated vegetation propagation framework that integrates LiDAR-based tree mesh reconstruction with ray tracing (RT)-based parameter inversion. A handheld LiDAR scan is processed to separate woody components and leaf components, followed by leaf-cluster meshing and woody mesh construction to obtain a high-fidelity tree model. Deterministic RT serves as the forward simulator, while particle swarm optimization (PSO) calibrates a set of effective vegetation parameters, including relative permittivity, conductivity, thickness, scattering coefficient, and scattering-lobe width, by minimizing the mismatch between simulations and a small set of field measurements. Experiments at 2.4 GHz on a real tree show that calibration substantially improves RT accuracy across multiple receiver locations, reducing the received-power RMSE from 4.066 dB to 1.342 dB and the MAE from 3.157 dB to 1.097 dB, and consistently outperforming ITU-R, COST235, and Weissberger baselines.

Electron density is a fundamental parameter characterizing the ionosphere. Multiple ground-based and space-based detection technologies are applied to detect ionospheric electron density using artificial electromagnetic waves, based on the ionospheric effects of reflection, refraction, incoherent scattering, and doppler shift on radio waves. Lightning-generated whistlers (LGWs) constitute a natural signal with a wide spatiotemporal distribution that can substitute for these artificial transmissions, achieving global ionospheric detection. This paper proposes a method for reconstructing ionospheric electron density profiles by comparing simulated and observed dispersion of LGWs. We develop an LGW propagation model based on the finite-difference time-domain (FDTD) algorithm, where the background electron density is derived from the International Reference Ionosphere (IRI) model. The dispersion of simulated whistlers is compared with satellite observations, and a modification factor is introduced to modify the background electron density based on the relationship between dispersion and electron density. The approach is applied to two events, and the electron density modification effect is assessed with independent data sources. The results show that the errors between the modified electron density and the true value in two events are reduced by 62.81% and 69.29%, respectively, confirming the efficacy of the proposed method.