区域实时对流层网格模型构建及在精密单点定位中的应用

Establishing a high-precision real-time regional tropospheric model and its application in precise point positioning

受制于对流层延迟建模方法和建模背景场精度及时空分辨率的影响,目前实时对流层延迟模型的精度和稳定性都有待进一步改善.本文利用甘肃及周围地基共计184个GNSS(Global Navigation Satellite System)站估算的ZTD(Zenith Troposphere Delay),构建了空间分辨率为0.25°×0.25°的甘肃地区实时ZTD网格模型.针对传统的高程归化模型及水平内插模型精度低的问题,本文提出了利用高斯指数函数模型将不同高程的GNSS/ZTD归化到统一的高度,再基于BP神经网络模型从网格顶点周围统一高度后的GNSS/ZTD中内插出网格顶点处的ZTD.为了验证甘肃ZTD网格模型的精度,选取2022年甘肃地区8个未参与建模的陆态网GNSS测站的数据进行了实验.统计结果显示:与事后PPP(Precise Point Positioning)处理GNSS估算的ZTD相比,甘肃ZTD网格模型与真值偏差的RMS优于1.52 cm.此外,将构建的实时ZTD格网模型用于约束PPP处理,对于PPP浮点解施加ZTD约束后U方向精度提升22.9%,U方向收敛时间缩短26.4%.

The accuracy and stability of the current real-time tropospheric delay models are subject to further improvement due to the limitations of the tropospheric delay modeling method and the accuracy and temporal resolution of the observation data required for modeling.In this paper,a real-time high-precision ZTD(Zenith Troposphere Delay)grid model with 0.25°×0.25°spatial resolution is constructed for the Gansu region using the ZTD estimated from 184 ground-based GNSS(Global Navigation Satellite System)stations in and around Gansu.To address the problems of low accuracy of the traditional elevation normalization model and horizontal interpolation model,this paper proposes the use of Gaussian exponential function model and BP neural network model for ZTD elevation normalization and horizontal interpolation respectively.In order to verify the accuracy of the new ZTD grid model,data from eight CMONOC stations in Gansu Province that were not involved in the modelling of the land state network was selected for the experiment.The statistical results show that the RMS of the deviation between the ZTD grid model and the post-processing GNSS-ZTD is better than 1.52 cm.In addition,the ZTD real-time grid model is used to constrain PPP(Precise Point Positioning)processing.After ZTD constraints are applied to the PPP floating point solution,the U direction accuracy is improved by 22.9%,and the U direction convergence time is shortened by 26.4%.