In order to improve the positioning and timing capacity of global navigation satellite system (GNSS), and make it convenient to obtain the tropospheric delay estimation with high accuracy without meteorological parameters detection equipment, this paper summarizes the development of the main tropospheric delay models, analyzes the estimation precision of the global pressure and temperature 2 wet (GPT2w) model in two different spatial resolutions at the global geodetic observing system (GGOS) stations, develops a strategy combining measured temperature and experience fitting meteorological parameters according to the relationship between estimation error of Zenith Tropospheric delay and that of meteorological parameters in GPT2w model, establishes a GPT2w improved model based on BP neural network. The simulation results show that the accuracy of Zenith Tropospheric delay estimation of the improved model at the global GGOS stations in 2017 is nearly 32% higher than that of the GPT2w model, and the accuracy of estimation of other global positions is also improved. The improvement is related to the density of GGOS stations.