4:45 PM - 5:00 PM
[G06-2-02] Optimum stochastic modeling for GNSS tropospheric delay estimation in real-time
In GNSS data processing, the station height, receiver clock and tropospheric delay (ZTD) are highly correlated to each other. Although the zenith hydrostatic delay of the troposphere can be provided with sufficient accuracy, zenith wet delay (ZWD) has to be estimated, which is usually done in a random walk process. Since ZWD temporal variation depends on the water vapor content in the atmosphere, it seems to be reasonable that ZWD constraints in GNSS processing should be geographically and/or time dependent.
Rather than performin a prior empirical testing for each station, we propose to take benefit from numerical weather prediction (NWP) models to define optimum random walk process noise. In the first approach, we used archived VMF1-G data to calculate a grid of yearly means of the difference of ZWD between two consecutive epochs divided by the root square of the time lapsed, which can be considered as a random walk process noise. Alternatively, we used the Global Forecast System NWP model from National Centres for Environmental Prediction to calculate random walk process noise dynamically in real-time, by performing ray-tracing through the two shortest available forecasts.
We performed two representative experimental campaigns with 20 globally distributed International GNSS Service (IGS) stations and compared real-time ZTD estimates with the official ZTD product from the IGS. We have shown that a single random walk processing noise (RWPN) value should not be applied globally to all stations, because it may lead to significant degradation of solution quality. With RWPN yearly grid we were able to reconstruct the wet RWPN value obtained from empirical testing with a mean error of 1mm/sqrt(h). A superior result was obtained with dynamical NWP-based apporach, in which we obtained an improvement of up to 10% in accuracy of the ZTD estimates compared to any uniformly fixed random walk process noise applied for all stations.
Rather than performin a prior empirical testing for each station, we propose to take benefit from numerical weather prediction (NWP) models to define optimum random walk process noise. In the first approach, we used archived VMF1-G data to calculate a grid of yearly means of the difference of ZWD between two consecutive epochs divided by the root square of the time lapsed, which can be considered as a random walk process noise. Alternatively, we used the Global Forecast System NWP model from National Centres for Environmental Prediction to calculate random walk process noise dynamically in real-time, by performing ray-tracing through the two shortest available forecasts.
We performed two representative experimental campaigns with 20 globally distributed International GNSS Service (IGS) stations and compared real-time ZTD estimates with the official ZTD product from the IGS. We have shown that a single random walk processing noise (RWPN) value should not be applied globally to all stations, because it may lead to significant degradation of solution quality. With RWPN yearly grid we were able to reconstruct the wet RWPN value obtained from empirical testing with a mean error of 1mm/sqrt(h). A superior result was obtained with dynamical NWP-based apporach, in which we obtained an improvement of up to 10% in accuracy of the ZTD estimates compared to any uniformly fixed random walk process noise applied for all stations.