Geometry-Aware PPP for Reliable GNSS Tropospheric Sensing in Dense Urban Environment

Global Navigation Satellite System (GNSS), based tropospheric sensing provides valuable, high-temporal-resolution observations for numerical weather modeling, but its application in dense urban environments remains challenging due to severe multipath interference and non-line-of-sight (NLOS) signal reception. These effects introduce geometry-dependent biases that destabilize Precise Point Positioning (PPP) and significantly degrade Zenith Tropospheric Delay (ZTD) estimation, limiting the usability of crowdsourced and low-cost GNSS data in cities. This study presents a ray-tracing-assisted method for urban GNSS multipath mitigation that combines ray-tracing with PPP processing. Using (Level-Of-Detail) LOD1 3D city models and raytracing, GNSS signal propagation is explicitly simulated to classify satellite observations into line-of-sight (LOS), Echo, reflected, diffracted, mixed multipath, and NLOS components. 
First, a simulation is performed to develop a city-scale “healthy zone” identification strategy by mapping LOS satellite availability across dense urban areas. Locations exhibiting sufficient unobstructed LOS visibility are identified as favorable sites for crowdsourced data collection for ZTD estimation. This strategy enables systematic and reliable collection of GNSS observations while mitigating multipath effects, thereby improving the spatial coverage and quality of urban ZTD.
Second, a ray-tracing–assisted PPP framework is developed, in which multipath contaminated observations are adaptively excluded or down-weighted based on their physically modeled propagation characteristics derived using raytracing. This raytracing-assisted PPP approach is evaluated using real urban GNSS data collected at a stationary location in Hong Kong. The results demonstrate that conventional, unmitigated PPP suffers from large code residuals (50–100 m), meter-level positioning errors, and strongly biased ZTD estimates. In contrast, the proposed method reduces code and phase residuals to approximately 2 m and 0.02 m, respectively, achieving sub-meter positioning accuracy, and improves ZTD precision by more than two orders of magnitude.
The results indicate that geometry-aware, physics-based multipath modeling is a critical enabler for reliable urban ZTD estimation. By jointly leveraging ray tracing and adaptive filtering in PPP and extending the framework toward potential mobile GNSS deployment, this work lays the foundation for ZTD retrieval in dense urban environments. Such an approach facilitates the future assimilation of crowdsourced GNSS observations into next-generation numerical weather prediction systems, supporting enhanced atmospheric monitoring in cities.