Research On The Key Models Of Atmospheric Water Vapor Inversion Using Precise Point Positioning

Due to the variation of atmospheric composition with altitude,the GNSS signal will be refracted when it passes through the neutral atmosphere,which leads to bending and delay of the radio signal.Tropospheric refraction delay is an important error source in GNSS positioning,which needs to be eliminated by model constraint,observation combination and parameter estimation.However,from the opposite point of view,the tropospheric delay also contains the information of water vapor in the lower atmosphere.The high-accuracy and high-reliability water vapor obtained by tropospheric delay is of great significance for the analysis of climate causes and the prediction of weather changes in GNSS meteorology.With the rapid development of GNSS,the stability,reliability,convergence speed and positioning accuracy of GNSS precise point positioning(PPP)have been improved.Compared with the traditional water vapor detection methods,the precipitable water vapor(PWV)retrieved by GNSS PPP has the advantages of all-weather,high temporal resolution,high precision and low cost.It can play an important role in disaster monitoring,rainfall forecasting and detecting rainfall information.In order to obtain high-accuracy and high-reliability GNSS PWV,this paper studies the key models of GNSS PPP water vapor inversion.In PPP data processing,a new multipath mitigation strategy is proposed.Based on the new generation reanalysis data of ECMWF,a high-precision and regional zenith tropospheric delay(ZTD)model is constructed.In the calculation of PWV using PPP ZTD,this paper establishes a weighted mean temperature(T_m)modeling strategy based on sparse kernel learning.And the process of water vapor retrieval using GNSS real-time PPP based on measured meteorological parameters is studied preliminarily.The main work and contents are as follows:(1)Aiming at the periodic characteristic of PPP multipath error,this paper establishes a new strategy of sidereal filtering based on the carrier phase residuals and the multipath error model using sparse regularization.Results show that the carrier phase residuals can be improved by about 49.8%when the new strategy of sidereal filtering is used.For kinematic PPP float solution,it can be learned that the smoother coordinate series are obtained with multipath mitigation.And the positioning errors in X,Y and Z directions can be improved by about 49.5%,48.9%and 63.0%,respectively.The final coordinate accuracy can be improved by about 54.0%on average.And the mean accuracy of PPP ZTD is improved by about 55.4%.(2)The accuracy and reliability of ERA5 ZTD data are verified by using radiosonde data and GNSS products.Results show that the mean bias and root mean squared error(RMS)of ERA5 ZTD data are 0.86 cm and 1.95 cm,respectively.The ERA5 ZTD data has high accuracy and reliability,and can be used as an effective modeling data for ZTD model.In addition,this paper proposes an improved ZTD modeling method whereby the piecewise model of the atmospheric refractivity is introduced.Based on this improved modeling method and ERA5 ZTD data from 2013to 2018,a China's new regional gridded ZTD model(RGZTD)is established with spatial resolution of 2.5°×2.5°,which ranges from 70°E to 135°E and 15°N to 55°N,respectively.In this paper,the ERA5 data,the radiosonde data,and the GNSS ZTD data are used as external compliance check data.Results show that the overall accuracy of RGZTD model is better than exponential model,UNB3m model and GPT3 model.Compared with the exponential model,the average accuracy of RGZTD model is improved by about 8.9%.(3)This paper proposes a new T_m modeling strategy without measured meteorological parameters through using the Gaussian radial basis function to model the residuals of seasonal model.To solve the problem of model complexity,the L1-norm regularization is introduced and the highly efficient fast iterative shrinkage thresholding algorithm is employed to obtain sparse solution.Results show that the accuracy of T_m obtained by the sparse kernel learning method can be improved by about 52.9%when that of the seasonal model is compared.And the accuracy of GNSS PWV using sparse kernel learning T_m modeling strategy is improved by about 63.0%.In addition,this paper tests the accuracy of PPP ZTD and PPP PWV by International GNSS Service(IGS)ZTD products and radiosonde data,respectively.And the relationship between PPP PWV and total daily rainfall is analyzed by using rainfall data of the rainfall station.Results show that the mean bias and RMS of PPP PWV is-1.90 mm and 2.41 mm,respectively.And the short-term change of PPP PWV can provide some references for rainfall prediction.(4)Based on the radiosonde data in China,the relationship between T_m and other meteorological parameters are analyzed in this paper.Results show that the related coefficients between T_m and surface temperature,water vapor pressure,surface pressure and relative humidity are 0.895,0.785,-0.590 and 0.004,respectively.The correlation between T_m and other meteorological parameters in tropical and high-altitude regions of China is weaker than that in other regions.In this case,the accuracy of calculated T_m values by single parameter or multi parameters model may be poor.Based on the data of 86 radiosonde stations from 2013 to 2018,this paper establishes the new one parameter model,two parameters model and three parameters model in China.Results show that the accuracy of one parameter,two parameters and three parameters T_m model is better than that of Bevis model,and the average RMS is4.06 K,3.66 K and 3.50 K,respectively.At the same time,two parameters model and three parameters model can improve the accuracy of T_m in high-altitude and inland areas when the one parameter model is compared.Compared with the two parameters model,the accuracy of the three parameters model with air pressure added is not significantly improved.(5)Based on the IGS final products,the accuracy of position and ZTD of static real-time PPP are tested.Results show that GPS real-time PPP has a similar accuracy with GPS/BDS real-time PPP.The accuracy of GPS real-time PPP are 2.19 cm,1.27cm and 2.02 cm in east,north and up directions,respectively.The average convergence time of each direction is 24.06 min.The mean bias and RMS of real-time PPP ZTD are-0.85 mm and 10.40 mm.It can be known that the ZTD estimated by Real-time GPS PPP has a good accuracy,which can be used as the effective data source of GNSS water vapor inversion.(6)Based on the rainfall data,this paper analyses the relationship between real-time PPP PWV and hourly rainfall.Results show that the PWV in the rainfall period is higher than that in the period without rainfall.Due to the effect of airflow and the time of cloud passing through station,rainfall occurs before the peak of PWV or at the initial stage when PWV decreases from the peak.And the real-time PPP PWV can be used as an important meteorological factor in the rainfall forecast model to assist the weather forecast when the air pressure and temperature are compared.This paper includes 69 pictures,17 tables and 243 references.