The Model Research On Key Parameters In Regional GNSS Water Vapor Inversion

With the aggravation of the world’s environmental problems,Greenhouse gas effects are a growing concern around the world.As an important part of greenhouse gases,water vapor in the atmosphere plays a vital role in the impact of global climate.The application of GNSS technology to atmospheric water vapor content monitoring as a new means of water vapor monitoring has the advantages of low cost,high time resolution and broad application prospects.Based on the experience and methods of previous scholars,this paper analyzes the model accuracy of key parameters in the process of GNSS technology detecting water vapor.On the basis of analyzing the accuracy of the model,the modeling method of tropospheric delay and weighted average temperature is further discussed and improved.The main work and results of this paper are as follows:(1)The principle of GNSS water vapor inversion and the application of ERA5 reanalysis data in water vapor inversion are systematically introduced,mainly including tropospheric delay correction methods,common tropospheric models,weighted average temperature acquisition and ERA5 reanalysis data processing.(2)The accuracy of the Global Pressure and Temperature 3(GPT3)tropospheric model to calculate the zenith tropospheric delay(ZTD)in the Asian region was studied,and compared with the GPT2 w model,and analyze the application of GPT3 model inversion of precipitable water(PWV)in China from the aspects of season,latitude and elevation.The results show that: for the ZTD accuracy,the accuracy of the GPT3 model is slightly higher than that of the GPT2 w model at the same resolution;In terms of the inversion accuracy of the PWV,the average deviation(Bias)and the root mean square error(RMSE)of the GPT3 model at the Radiosonde station show strong seasonal characteristics.(3)Research on the construction method of regional high-precision nonlinear weighted average temperature(Tm)model in the absence of measured meteorological data.Using ERA5 reanalysis data to obtain meteorological data,combined with three machine learning methods of BP neural network,Random Forest and Extreme Learning Machine,and introduced regional relative elevation to establish a high-precision nonlinear Tm model.When testing the internal coincidence accuracy of the model,it was found that compared with the h0 Tm model as a comparison,the annual average RMSE value of the model based on the machine learning method were reduced to varying degrees;Among the three models based on the machine learning method,RFh0 Tm has higher accuracy when modeling with overall data,and ELM-h0 Tm has higher accuracy when modeling with single-layer data.For the external coincidence accuracy of the model,the model established by the machine learning method has higher accuracy than the GPT3 model and the Bevis regression model.(4)In view of the problem that the GPT3 model can only fit the general variation trend of the actual tropospheric delay,but cannot fit the detailed variation process.Based on the machine learning method,this paper discusses the GPT3 correction model using the ZTD residual calculated from ERA5 reanalysis data as input(scheme 1)and the ZTD residual calculated from the tropospheric delay reference value provided by IGS as input(scheme 2).The experimental results show that the annual average Bias value and annual average RMSE value of the zenith tropospheric delay calculated by the models established based on the scheme 1 and scheme 2 are both smaller than those of the GPT3 model,and the improvement effect of the latter is better.