Improving GNSS precise point positioning with tropospheric constraints from data-driven numerical weather prediction model

Accurate priori tropospheric knowledge is advantageous for Global Navigation Satellite System (GNSS) precise point positioning (PPP), which significantly influences both convergence time and the accuracy of tropospheric delay estimations. However, traditional numerical weather prediction (NWP) models, which are often used to provide this tropospheric information, rely heavily on parameterization. This reliance can introduce approximation errors and increase computational demands, limiting their effectiveness. In contrast, emerging data-driven NWP models offer enhanced forecasting capabilities with reduced computational requirements, presenting a promising alternative for improving PPP performance. This study proposes an innovative approach to improve PPP by leveraging data-driven NWP models. An evaluation involving nearly 20,000 stations reveals that these models outperform conventional NWP products, such as the Global Forecast System (GFS), achieving a 63% improvement in short-range zenith tropospheric delay (ZTD) forecast precision and a 55% enhancement in accuracy. For medium-range ZTD forecasts, data-driven NWP consistently surpasses GFS and even outperforms the empirical ZTD model GPT3 over a 15-day forecast period. Consequently, data-driven NWP facilitates a more rapid and accurate estimation of tropospheric random walk process noise (RWPN) compared to GFS. Moreover, validation with GPS kinematic positioning indicates that incorporating short-range ZTD forecasts as prior information reduces convergence time by an average of 400 s across 200 global stations, while medium-range forecasts also contribute positively when short-range data are unavailable. These findings demonstrate the potential of data-driven NWP models to improve tropospheric delay estimation and enhance PPP performance.