Estimation of tropospheric wet delay from GNSS measurements

Abstract

The determination of the zenith wet delay (ZWD) component can be a difficult task due to the dynamic nature of atmospheric water vapour. However, precise estimation of the ZWD is essential for high-precision Global Navigation Satellite System (GNSS) applications such as real-time positioning and Numerical Weather Prediction (NWP) modelling.The functional and stochastic models that can be used for the estimation of the tropospheric parameters from GNSS measurements are presented and discussed in this study. The focus is to determine the ZWD in an efficient manner in static mode. In GNSS, the estimation of the ZWD is directly impacted by the choice of stochastic model used in the estimation process. In this thesis, the rigorous Minimum Norm Quadratic Unbiased Estimation (MINQUE) method was investigated and compared with traditional models such as the equal-weighting model (EWM) and the elevationangle dependent model (EADM). A variation of the MINQUE method was also introduced. A simulation study of these models resulted in MINQUE outperforming the other stochastic models by at least 36% in resolving the height component. However, this superiority did not lead to better ZWD estimates. In fact, the EADM provided the most accurate set of ZWD estimates among all the models tested. The EADM also yielded the best ZWD estimates in the real data analyses for two independent baselines in Australia and in Europe, respectively.The study also assessed the validity of a baseline approach, with a reduced processing window size, to provide good ZWD estimates at Continuously Operating Reference Stations (CORS) in an efficient manner. Results show that if the a-priori station coordinates are accurately known, the baseline approach, along with a 2-hour processing window, can produce ZWD estimates that are statistically in good agreement with the estimates from external sources such as the radiosonde (RS), water vapour radiometer (WVR) and International GNSS Service (IGS) solutions. Resolving the ZWD from GNSS measurements in such a timely manner can aid NWP model in providing near real-time weather forecasts in the data assimilation process.In the real-time kinematic modelling of GNSS measurements, the first-order Gauss- Markov (GM) autocorrelation model is commonly used for the dynamic model in Kalman filtering. However, for the purpose of ZWD estimation, it was found that the GM model consistently underestimates the temporal correlations that exist among the ZWD measurements. Therefore, a new autocorrelation dynamic model is proposed in a form similar to that of a hyperbolic function. The proposed model initially requires a small number of autocorrelation estimates using the standard autocorrelation formulations. With these autocorrelation estimates, the least-squares method is then implemented to solve for the model’s parameter coefficients. Once solved, the model is then fully defined. The proposed model was shown to be able to follow the autocorrelation trend better than the GM model. Additionally, analysis of real data at an Australian IGS station has showed the proposed model performed better than the random-walk model, and just as well as the GM model. The proposed model was able to provide near real-time (i.e. 30 seconds interval) ZTD estimates to within 2 cm accuracy on average.The thesis also included an investigation into the several interpolation models for estimating missing ZWD observations that may take place during temporary breakdowns of GNSS stations, or malfunctions of RS and WVR equipments. Results indicated marginal differences between the polynomial regression models, linear interpolation, fast-Fourier transform and simple Kriging methods. However, the linear interpolation method, which is dependent on the two most recent data points, is preferable due to its simplicity. This result corresponded well with the autocorrelation analysis of the ZWD estimates where significant temporal correlations were observed for at most two hours.The study concluded with an evaluation of several trend and smoothing models to determine the best models for predicting ZWD estimates, which can help improve real-time kinematic (RTK) positioning by mitigating the tropospheric effect. The moving average (MA) and the single-exponential smoothing (SES) models were shown to be the best-performing prediction models overall. These two models were able to provide ZWD estimates with forecast errors of less 10% for up to 4 hours of prediction.