Abstract:Using the data of 18 IGS stations in Asian region and 16 radiosonde stations in China region from 2016 to 2018, the accuracy of zenith tropospheric delay (ZTD) and precipitable water vapor (PWV) inversed by GPT3 model are studied, and compared with other GPT series models. The results show that the ZTD of the GPT3-1 model has the smallest mean and maximum deviation bias values of 1.34 mm and 14.06 mm, respectively; the accuracy of GPT3 model is slightly better than GPT2w model and better than GPT2 model. The bias and root mean square error(RMSE) of the GPT3 model-derived PWV at the radiosonde stations show strong seasonal characteristics, and the GPT3-1 model has higher accuracy and stability than the GPT3-5 model, as shown by the monthly mean values of the GPT3 model-derived PWV.