Research on the modeling of tropospheric delay in virtual reference station

Atmospheric delay, including ionospheric delay and tropospheric delay, is one of the main error sources in the long distance kinematic positioning. Many authors have made great researches on this problem. Atmospheric delay can be modeled by a simulated model, such as Hopfield's model and Saastamoinen's model. Neutral atmosphere delay has badly affected precise positioning. Virtual reference station technology is an effective method to reduce the impact of atmospheric error on precise positioning. The location of VRS(usually using the navigation position) can be arbitrarily selected by user. Existing RTK processing software can be applied to long distance RTK if using VRS technology. The key problem in VRS is to model atmospheric delay precisely. For modeling the ionospheric delay, many authors have done great works based on double frequency observables on multi-reference stations[1,2]. Because of the complexity of tropospheric error, the accuracy of existing tropospheric delay models can not meet the requirement of precise positioning. After investigating the relationship between tropospheric delays and the elevations of reference stations, this paper presented seven troposphere-fitting models with height factors, and then analyzed the accuracies of proposed fitting models by two experiments. Research results show that the proposed models have better fitting accuracy than traditional method according to the data from part of SGIGN network. The best fitting model is determined by check points. The fitting accuracy depends on the number of known points and their distribution as well as the size of a network. Based on the test on part of SCIGN network with an area of 60 km × 60 km, the height of which is from -20 m to 700 m, polynomial fitting model with 4 parameters including one height parameter has the best fitting accuracy(± 7.6 mm) if using 5 known points. For an area of 500 km × 300 km, polynomial fitting model with 5 parameter including one height parameter is the best one if using 6 points, the accuracy of which is about ± 1.8 cm. For real time application, an extrapolating method was proposed by this paper. Initial test showed that the extrapolating accuracy is about ±9.0 mm. Further experiments need to be done in order to test the accuracy of proposed models in different networks with bigger average height difference than the test network from SCIGN.