Investigations on the Use of Numerical Weather Predictions, Ray Tracing, and Tropospheric Mapping Functions for Network RTK

Real-time kinematic positioning (RTK) using a network of reference stations (network RTK) is a well-known technique for high accuracy positioning. One of the major technical challenges in network RTK is the large distances between reference stations that make it difficult to compensate for the errors within the network. The atmospheric effects on the satellite signals are the most dominant spatially correlated errors. The ionospheric error can, most of the time, be mitigated using dual frequency receivers, and compensation for the tropospheric error is often carried out using global tropospheric delay models. During abnormal weather conditions and rapid weather changes, the global models are not, however, sufficiently accurate for RTK, especially for longer baselines. Numerical Weather Predictions (NWP) form the basis for weather forecasts, and give a good representation of the conditions in the neutral atmosphere. Data from a NWP can provide an estimate of the tropospheric delay in GPS signals, and the estimate will often be better than a delay determined using a global tropospheric delay model. This paper compares three different ways of using a NWP for predicting the tropospheric delay for a GPS station. One solution is to estimate the zenith delay from the weather model and then use a standard mapping function for mapping the delay down to the elevation angle where the signal is received. The second approach is to use a ray tracer to simulate the signal path through the NWP considering the combined effect of refraction and bending of the media. For the third approach the signal path is simulated through the NWP anticipating a straight line and thus ignoring the signal bending. The three methods are evaluated, and their suitability for network RTK is discussed. For the evaluation a NWP provided by the Danish Meteorological Institute is used together with GPS data collected at 14 permanent GPS reference stations located in Denmark and Southern Sweden. The paper concludes that the first approach is most feasible for signals received with elevation angles down to about 10 degrees. For signals received at lower elevation angles, the difference between using a mapping function and the ray tracer becomes more significant, and the selection of a method will be a trade off between computation time and accuracy. The linear approximation performs better than the mapping functions for the lowest elevation angles, but is not as good, in terms of accuracy, as the ray tracer. It is, however, computationally faster than the ray tracer and might therefore be a suitable alternative.