Accuracy Evaluation Of COSMIC-2 Occultation Inversion Data And Construction Method Of Tm Model

Radio occultation technology has been widely used in the detection and research of the Earth’s atmospheric environment.As a new generation of occultation mission,COSMIC-2occultation program has significantly improved the quantity and quality of data,which has great application value for GNSS real-time water vapor inversion.With the increasing demand for real-time and high-precision water vapor detection,the high-precision modeling of weighted average temperature Tm,a key parameter in GNSS water vapor inversion,has become a research hotspot in recent years.Considering that the weighted average temperature model constructed by traditional algorithms cannot simulate Tm fine synoptic scale changes,and the existing Tm model constructed by machine learning method only adopts the single data of sounding station for model construction,this paper,based on the neural network algorithm of self-optimization of hyperparameters,integrates COSMIC-2 occultation data.The construction of global Tm model is studied.The main research contents and achievements of this paper are as follows:1.By using sounding station data,the index of refraction,atmospheric temperature,atmospheric pressure and vapor pressure profile data of COSMIC-2 occultation secondary product wet Pf2 are compared,verified and analyzed in terms of height,latitude and season.The accuracy analysis results show that the atmospheric refractive index,temperature,atmospheric pressure,vapor pressure and radiosonde data from COSMIC-2 inversion are consistent with each other at different altitude levels,and the highest correlation coefficients are 0.981,0.98,0.977,0.943,respectively.The highest relative deviation of atmospheric refractive index is 4.05%,and the highest RMSE of atmospheric temperature,pressure and vapor pressure are 1.82 K,2.04 h Pa and 2.82 h Pa,respectively.From the perspective of sublatitude samples,the accuracy of COSMIC-2 inversion of meteorological parameters samples is significantly affected by latitude,and the accuracy of high latitude samples is lower than low latitude samples.From the analysis of seasonal accuracy,the four samples have the same seasonal trend,and seasonal factors have no significant influence on meteorological data retrieved from COSMIC-2 occultation.In general,the high SIGNal-to-noise ratio and observation amount of COSMIC-2 in low-latitude regions still maintain high accuracy in lowlatitude regions with more water vapor in the atmosphere,which can be used as a data source for global Tm model construction.2.Based on the integrated Tm data of sounding stations in the middle and low latitudes of the world as the reference value,the error distribution of Tm calculated by COSMIC-2occultation data is analyzed.The results show that,influenced by the number of observations,the accuracy of Tm obtained by integrating COSMIC-2 data is better than that in the middle latitudes.The Tm obtained by integrating COSMIC-2 occultation data has high accuracy and stability.3.Bohb-bp neural network algorithm based on BOHB hyperparameter self-optimization,GPT2w-NN-1(radiosonde data),GPT2W-NN-2(occultation data and radiosonde data),and Tm-NN-1 and Tm-NN-2(BOHB-BP algorithm only)were constructed.The results show that the BIAS and RMSE of GPT2w-NN-2 model are 0.06 K and 3.85 K,respectively.The accuracy of GPT2w-NN-1 model is better than that of GPT2w-NN-1 model,and RMSE is0.18 K less than that of GPT2w-1 model.The RMSE of the constructed meteorological parameter model Tm-NN-2 is 2.80 K,which is 0.5K less than that of the Bevis model.Therefore,COSMIC-2 occultation data can be used as a powerful supplement for GNSS inversion of atmospheric water vapor.The high-precision and real-time weighted average temperature of the world can be obtained by integrating occultation data with high temporal and spatial resolution and high precision through machine learning algorithm,which is of great significance for realizing real-time GNSS water vapor remote sensing.