Abstract

Zenith tropospheric delay (ZTD) is one of the main error sources in space geodetic techniques, such as Global Positioning System (GPS) and satellite altimetry. To qualitatively and quantitatively determine the most suitable model for Antarctica, we analyze the accuracy and applicability of nine models (UNB3m, EGNOS, GPT2 + Saastamoinen, GPT2w + Saastamoinen, GPT3 + Saastamoinen, GPT2 + Hopfield, GPT2w + Hopfield, GPT3 + Hopfield, and IGGtropSH) in Antarctica using 8 years of GPS‐derived ZTD time series from 65 stations. The results show that the GPT2/2w/3 + SAAS models are better than the other six models, with a bias of 0.2, −0.22, and −0.29 cm and root mean square (RMS) of 2.33, 2.31, and 2.36 cm. Based on the decimeter bias and RMS, the UNB3m model and EGNOS model present the worst performance in Antarctica. There are regional characteristics of bias and RMS in the nine models. The GPT2/2w/3 + SAAS models have the smallest regional deviation, and the bias and RMS between subregions (Antarctic Peninsula, Amundsen Sea Embayment, Ross Ice Shelf, Inland area of West Antarctica, Filchner‐Ronne Ice Shelf, and coastal East Antarctica) are all at the 0.2 and 0.7 cm levels, respectively. The GPT2/2w/3 + HOP models have the largest regional deviation, with regional bias and RMS at the levels of 8 and 6 cm, respectively. Our results suggest that the uncertainty of ice sheet elevation derived from satellite altimetry may be partly caused by the spatial‐related bias and error in the ZTD corrections. The bias and RMS of six GPT combined models and the IGGtropSH model present limited seasonal changes, indicating that these models can simulate the seasonal characteristics of ZTD better. The time series of the bias and RMS values of the EGNOS and UNB3m models show obvious seasonal characteristics, which may contaminate the annual ice sheet elevation by approximately 5 cm if used as ZTD corrections.