Satellite signals from the Global Navigation Satellite System (GNSS) are refracted as they pass through the troposphere, owing to the variable density and composition of the atmosphere, causing tropospheric delay. Typically, tropospheric delay is treated as an unknown parameter in GNSS data processing. Given the growing need for real-time GNSS applications, accurate tropospheric delay predictions are crucial to improve Precise Point Positioning (PPP). In this paper, time-series of tomography data are used for wet refractivity prediction employing Machine Learning (ML) techniques in both Poland and California, under extreme weather conditions including sweeping rain bands and storms. The predicted wet refractivity is implemented for tropospheric delay determination through ray-tracing technique. PPP processing is conducted in both static and kinematic modes using different setups. These are: (1) common PPP, called Com-PPP, (2) Ray-PPP, which applies obtained tropospheric delay on GNSS observations and thus eliminates tropospheric parameters from unknowns, and (3) Dif-PPP, which applies the difference of estimated tropospheric delay from ray-tracing and GNSS measurements to compensate for the remaining tropospheric delay in the observations. The results show that Dif-PPP reduces the Mean Absolute Error (MAE) of the Three-Dimensional (3-D) component between 8 and 33% in static mode compared to the Com-PPP method. Additionally, it can improve the convergence time of the up component in the kinematic mode by between 6 and 17%.
Read full abstract