Modeling Of Zenith Troposphere Wet Delay And Weighted Average Temperature In China

The tropospheric zenith wet delay(ZWD)and weighted mean temperature(T_m)are the key parameters of the troposphere and are important factors affecting high-precision GNSS positioning and GNSS atmospheric water vapor retrieval.Aiming at the problems of the current ZWD and T_m models,such as the low spatial and temporal resolution of the model,and the lack of consideration of the ZWD/T_m value changing with the elevation direction,based on the ERA5 reanalysis data,the variation characteristics of ZWD and T_m are studied,combined with traditional modeling methods.and neural network,respectively,to carry out research on the construction of ZWD and T_m models in China.The main research contents and conclusions of this paper are as follows:1.The accuracy of the ZWD/T_m data calculated from the ERA5 reanalysis data in the Chinese region in this study was evaluated using radiosonde data.The results show that the annual mean RMS of ZWD and T_m calculated by the ERA5 reanalysis data used in this paper are 1.79cm and 1.51K,indicating that the ZWD and T_m data calculated from this data have high accuracy and stability,and can be used as the basic data source for modeling the key parameters of the troposphere in China.This paper not only analyzed the spatiotemporal characteristics of ZWD,T_m and other factors at the surface grid points,but also analyzed the vertical variation of ZWD/T_m with vertical of elevation,which provided the basis for the next model construction.2.Based on the periodic characteristics of the AN model and surface ZWD,combined with the ZWD elevation scaling factor,this paper constructs two ZWD grid empirical models(ANZWD-H model and ZWD-H model),and based on the ZWD-H model,using the BP neural network optimized by genetic algorithm to correct the residual error of the surface ZWD provided by the ZWD-H model.Finally,a meteorological parameter model(GBZWD-H model)is obtained.The accuracy of the built model was verified by combining the ERA5 data and sounding data in 2018,and the accuracy was compared with the UNB3m model,the GPT2w model and the Saastamoinen model.The results show that:(1)Taking the ERA5 surface ZWD data as the reference value,the RMS corresponding to the ANZWD-H model,the ZWD-H model and the GBZWD-H model are 4.14cm,4.03cm and 3.06cm,respectively.Compared with other models,its RMS value is small;(2)Taking the sounding data as the reference value,the RMS corresponding to the ANZWD-H model,the ZWD-H model and the GBZWD-H model are 5.04cm,4.93cm and 3.62cm,respectively,which are all smaller than the RMS corresponding to the UNB3m and Saastamoinen models.The RMS of the ANZWD-H model is comparable to that of the GPT2w-1 model,while the RMS of the ZWD-H model and the GBZWD-H model are reduced by 0.11 cm and1.42 cm,respectively,relative to the GPT2w-1 model.3.Based on the periodic characteristics of surface T_m and the vertical decline rate of T_m,this paper constructs the T_m grid empirical model(Tm-H model).On this basis,a BP neural network based on genetic algorithm optimization is used to correct the residual error of the surface T_m provided by the Tm-H model,and then a new meteorological parameter model(GBTm-H model)is obtained.The accuracy of the built model is verified by combining the ERA5 data and sounding data in 2018,and the accuracy is compared with the UNB3m model,the GPT2w model and the Bevis formula.The results show that:(1)Taking the ERA5 surface T_m data as the reference value,the RMS of the Tm-H and GBTm-H models are 3.73K and 2.89K,respectively,and their accuracy is better than other models;(2)Taking the sounding data as the reference value,the RMS of the Tm-H and GBTm-H models are 4.31K and 3.32K,respectively.Compared with other models,their RMS values are smaller and have higher accuracy.