Abstract:Based on the radiosonde records at 81 sounding stations from integrated global radiosonde archive (IGRA) throughout 2017, we conduct a comprehensive evaluation and analysis of the precision of four tropospheric delay models over China. Results show that GPT2w model outperforms the other models, including the Saastamoinen model that is dependent on meteorological elements, GZTD, as well as UNB3m with the format of spherical harmonic function. Mean bias (MB) and root mean square error (RMSE) of GPT2w model are -0.8 cm and 4.1 cm, respectively, and they range from -2 to 2 cm, and from 1.3 to 7.9 cm at all sites, respectively. The UNB3m model has the largest MB and RMSE among the four models, with the maximal RMSE reaching 10.2 cm. In addition, the four models have consistent sensitivity to the latitude of the stations, presenting as decreasing accuracy with increasing latitudes. The precision of the models has a prominent seasonal variability, presenting as different seasonal sensitivities among the four models. The low precision in modeling of tropospheric wet delay causes the lower accuracy of the models in summer (RMSE: 6~9 cm), in contrast with that in winter (RMSE: 2~2.5 cm).