Multi-model Fusion Method to Estimate Zenith Tropospheric Delay

This paper proposes a method fusing GPT3, Saastamoinen and Askne models to improve the estimation accuracy of zenith tropospheric delay (ZTD) based on the conventional modeling strategy. In this method, Saastamoinen and Askne models are used to estimate zenith hydrostatic and wet delay respectively, and GPT3 model is employed to provide the meteorological parameters including temperature, pressure, water vapor pressure, weighted temperature and vertical gradient of water vapor required by the two models. The ZTD data in the the International GNSS Service (IGS) Asia stations from 2016 to 2018 published by the Global Geodetic Observing System (GGOS) atmosphere and IGS are used to evaluate this proposed method. The results indicate that the accuracy (RMS:4.53 cm) of the Sas+Ask+GPT3 model is about 29% and 19% higher than that of Sas+Ask+UNB3m and Sas+GPT3 models respectively when the ZTD data from the GGOS atmosphere are served as a reference. When the IGS ZTD products are used as a reference, the accuracy (RMS:4.35 cm) of the model is about 25% and 14% higher than that of the other two models, respectively. The bias and RMS of the ZTD estimates is more larger in summer than in winter, showing that the error has a seasonal characteristics. However, the error in summer is significantly lower than the other two models. In addition, it is also shown that the bias and RMS decrease with the increase of altitude or latitude. Especially, the error in high-altitude stations is substantially lower than that of the Sas+Ask+UNB3m model. It is concluded that the proposed method is more potential to accurately estimate ZTD in Asia areas and therefore can be used for real-time ZTD correction over Asia.