Higher accuracy estimation of the weighted mean temperature (Tm) using GPT3 model with new grid coefficients over China

Abstract

The weighted mean temperature (Tm) is an essential parameter in the field of Global Navigation Satellite System (GNSS) meteorology, as it enables the conversion of zenith wet delay to atmospheric precipitable water vapor (PWV). The existing Tm models, such as the GPT3 model, utilize the empirical annual and semi-annual amplitudes of Tm at grid points along with corresponding trigonometric functions to estimate Tm, making it difficult to provide a more detailed description of the daily variation of Tm. Therefore, the measured surface temperature and water vapor pressure were introduced to calculate the new series of grid coefficients using the least-square method, which enables the GPT3 model to estimate Tm with higher accuracy. In this process, the ERA5 reanalysis data derived from ECMWF from 2001 to 2010 was utilized to compute Tm values, and the new grid coefficient obtained by introducing only surface temperature was considered as scheme #1 and that achieved by introducing both surface temperature and surface water vapor pressure was regarded as scheme #2. The Tm data of 2023 calculated both by the radiosonde and ERA5 reanalysis data were utilized to assess the new grid coefficients for Tm estimation, and the results show that the grid coefficients based on these two surface parameters outperformed those only using the temperature. In the compassion using the ERA5 data, the bias/MAE/RMSE with the two types of new grid coefficients are −0.09/1.89/2.41 K and − 0.07/1.82/2.31 K, respectively, and the improvements reach to 50%/32.5%/34.2% and 61.1%/35%/36.9% compared with GPT3 model. In the comparison with radiosonde data, the scheme #1 and #2 performed a 0.55 K(13.8%) and 0.68 K(17.1%) improvement in the mean RMSE of stations compared with the GPT3 model. Note that the proposed two schemes can effectively improve the accuracy of estimating Tm in the comparison of regions, epochs and seasons with the Bevis and GPT3 model. Furthermore, the impact of Tm on PWV can be improved by the proposed two schemes, which was demonstrated using the theoretical RMSE and relative error of PWV.