Tropospheric delay error is independent of signal frequency and has strong temporal and spatial variation. It is one of the most important sources of error in satellite navigation and positioning. The common tropospheric empirical model is affected by the temporal and spatial characteristics of meteorological parameters, which cannot meet the needs of precision positioning. In this paper, we use the tropospheric delay data obtained from the GNSS continuous operation reference station in 2013-2017 years to analyze its spatiotemporal variation characteristics, and analyze the applicability of several commonly used tropospheric empirical models in Australia. At the same time, a ZTD spatiotemporal modeling method based on improved polyhedral function is proposed, and a regional non-meteorological delay model (A_ZTD) is established for the zenith tropospheric delay calculated by the Australian base station. After testing, the internal coincidence accuracy of the region model (A_ZTD) is 4.1cm, and the fitting effect is better. The ZTD provided by the reference station, which is not involved in the modeling, the tropospheric delay provided by the GGOS Atmosphere and the ZTD derived from the sounding data solution are used as the external coincidence check. The results show that the accuracy of the A_ZTD model in Australia is better than that of the GZTD model, UNB3, UNB3m and UNB4 models. Compared with the empirical model, the accuracy of the A_ZTD model is improved in the region. This model can describe the spatial variation of the troposphere without meteorological parameters. It is more suitable for the precise correction of the regional tropospheric delay.