The error of the influence of the Earth’s neutral atmosphere (tropospheric errors) is one of the main sources of the errors in radio dimentional measurements. Considering this error were applied troposphere models that use real, predicted or statistical meteorological information. The most used models in geodesy are Hopfield, Saastamoinen, Black, GCAT and MOPS. In this paper, we consider a methodology of using predicted meteorological data obtained by the GFS digital weather model in GRIB format. A numerical experiment is described, the purpose was to obtain an estimate of the accuracy of calculating the zenith tropospheric errors. The results of the estimation of the zenith tropospheric errors at the stations of the IGS network were used as true definitions. Based on the data obtained for several months of 2018, it was concluded that the accuracy in applying digital models of meteorological data, compared with statistical ones, is 1.5 to 2 times higher, and UPC models of Hopfield, Saastamoynen, and Black’s using predicted meteorological parameters differ from each other no more than 5% (0.063 m, 0.061 m, 0.062 m, respectively).