To solve the problems of low accuracy and poor stability in the traditional estimation of zenith delay, a method of adding annual and semi-annual periodic terms on the basis of Hopfield model is proposed. Based on the atmospheric data of the global geodetic observation system at 45 stations from 2012 to 2014 in China, the time series and spectrum distribution of zenith tropospheric delay (ZTD) and residual Hopfield model are analyzed. With the introduction of annual and semi-annual cycle term, the GHop model suitable for China region is established, and the accuracy and adaptation of the two models are evaluated. The results show that the deviation and middle error of Hopfield model represent obvious seasonal variation in time, while GHop model is small and stable. In terms of spatial distribution, the Hopfield model varies greatly with the increase of elevation and latitude, GHop model can adapt to different latitudes and elevation ranges. Annual deviation of coincidence accuracy in GHop is 28% higher than that of Hopfield, and 76 radiosensory data in China are used for external coincidence test. The statistical results are better than those of Hopfield model. The ZTD results calculated by the proposed method are more reliable and have high practical value.