由 张 良承 A refined zenith tropospheric delay model for Mainland China based on the global pressure and temperature 3 (GPT3) model and random forest
基于GPT3模型与随机森林的中国大陆天顶对流层延迟优化模型
引文格式 | Citation:
Jiang N, Wu Y H, Li S, Xu Y, Wang Y B, Xu T H. First PWV retrieval using MERSI-LL onboard FY-3E and cross validation with co-platform occultation and ground GNSS[J]. Geophysical Research Letters, 2024, 51(8): e2024GL108681. DOI: 10.1029/2024GL108681.
Geophysical Research Letters（中科院1区Top，IF：4.6）
GNSS
Zenith tropospheric delay
Empirical model
Random forests algorithm
Prediction
Zenith tropospheric delay
Empirical model
Random forests algorithm
Prediction
Abstract | 摘要
Zenith Tropospheric Delay (ZTD) plays a vital role in Global Navigation Satellite System (GNSS) navigation, positioning, and meteorology. The generally accepted empirical models that are now in use can only reflect the periodic changes in ZTD. Nevertheless, capturing its subtle nonlinear changes, like rapid ZTD variations, is challenging as its accuracy needs further improvement. To overcome these drawbacks, the relationship between the residuals of GPT3 ZTD minus GNSS ZTD and the spatiotemporal information was fitted by using random forests (RF). Consequently, a refined model of GPT3 ZTD was established in Mainland China (named RGPT3), and the performance of the proposed model was compared to two accepted empirical models and another model based on a popular algorithm of machine learning (backpropagation neural network algorithm). Based on the results of the study, the RMSE of RGPT3 is 1.83 cm, which has an improvement of 28.0, 16.8, and 34.4% over the three compared models. The RGPT3 performs better in capturing the instantaneous ZTD changes than the empirical models. The result of RGPT3 ZTD constraint GNSS precise point positioning (PPP) is also superior to that of GPT3 ZTD, with the U direction convergence time reduction of 12.3% and accuracy improvement of 7.9%. The new model can offer higher-precision ZTD predictions in the study area.
天顶对流层延迟（ZTD）在全球导航卫星系统（GNSS）导航、定位与气象学中具有重要作用。当前普遍采用的经验模型仅能反映ZTD的周期性变化，但在捕捉其快速变化等细微非线性特征方面存在局限，精度有待进一步提升。为克服上述不足，本研究利用随机森林（RF）算法拟合了GPT3模型ZTD相对GNSS ZTD残差与时空信息之间的关系，在此基础上建立了适用于中国大陆地区的GPT3 ZTD优化模型（RGPT3），并与两种常用经验模型及另一种基于主流机器学习算法（反向传播神经网络）的模型进行了性能比较。研究结果表明，RGPT3模型的均方根误差为1.83厘米，相较于三种对比模型分别提升了28.0%、16.8%和34.4%。RGPT3在捕捉瞬时ZTD变化方面表现优于经验模型。将其应用于GNSS精密单点定位（PPP）中作为ZTD约束时，RGPT3的结果也优于GPT3，U方向收敛时间缩短12.3%，精度提升7.9%。该模型能够为研究区域提供更高精度的ZTD预测值。