Abstract: We calculated the key parameters of ground-based GPS water vapor inversion in each grid point and got the time series of the conversion factor Π^-1by using the global zenith wet delay (ZWD) grid data with a spatial resolution of 2.5°×2°(longitude × latitude) provided by global geodetic observing system (GGOS) Atmosphere and precipitable water vapor (PWV) products with a same spatial resolution provided by European Centre for Medium-Range Weather Forecasts (ECMWF) from 2005 to 2011, followed by the analysis of temporal and spatial characteristic. We establish a global empirical model of Π which is related to the site's latitude, longitude, altitude and the day of year (doy) without meteorological data. Then we used the GGOS Atmosphere and ECMWF grid data that not involved in modeling in 2012, 661 radiosonde stations' data in 2012 to test accuracy of the model. The results show that the mean deviation (Bias) is -0.179 mm and root mean square error (Root Mean Square Error, RMS) is 1.806 mm from the grid data; the Bias is 0.465 mm, RMS is 0.789 mm from radiosonde data. Both of them indicate the accuracy and stability of the model is high with a small systematic bias on a global scale.