|Abstract:||Weighted average temperature (Tm) is one of the key parameters to determine the accuracy of GNSS water vapor inversion. Aiming at the problem that the local weighted average temperature model based on machine learning method considers less influence of regional relative elevation, it is difficult to obtain meteorological data in the area without radiosonde observations. Taking three radiosonde stations in Poyang Lake as an example, the accuracy of nonlinear local weighted average temperature modeling in Poyang Lake area is analyzed by using the meteorological data provided by ERA5 data and introducing the relative elevation as the input of machine learning modeling. Two machine learning methods of BP neural network and Random Forest are used as the modeling method. When checking the internal coincidence accuracy of the model, the experimental results show that for the accuracy of the whole troposphere, the accuracy of BP-h0Tm model at Anqing station, Nanchang Station and Quzhou station is 15.12%, 11.28% and 14.22% higher than that of h0Tm model respectively, and the accuracy of RF-h0Tm model is 16.73%, 14.60% and 16.16% higher than that of h0Tm model respectively. In the two Tm models based on machine learning algorithm, RF-h0Tm has higher accuracy in the selected overall troposphere range. BP-h0Tm has higher accuracy in the single layer with the sounding stations. When using radiosonde data to analyze the external coincidence accuracy of the model, it is found that compared with the results of Bevis regression formula and GPT3 model, the Bias absolute value and RMSE values of BP-h0Tm model and RF-h0Tm model established considering relative elevation are reduced accordingly, and the system error and model accuracy are improved. Through this nonlinear modeling method, it provides an idea for obtaining a higher-precision Tm method in areas without measured meteorological data or radiosonde observations.|
Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022)
September 19 - 23, 2022
Hyatt Regency Denver
|Pages:||3098 - 3111|
|Cite this article:||
Huang, Cong, Guo, Hang, Yu, Min, Xiong, Jian, Wan, Min, "Study on Nonlinear Modeling Method of Local Weighted Average Temperature in Poyang Lake Region," Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 3098-3111.
ION Members/Non-Members: 1 Download Credit