Enhancing the Accuracy of GNSS-IR Sea Surface Height Retrieval with a Novel Deep-learning Composite Atmospheric Delay Correction Inversion Model

Abstract

Abstract Accurate sea surface height (SSH) is critical for marine research and is an important basis for establishing the ocean gravity field. The Global Navigation Satellite System Interferometry Reflectometry (GNSS-IR) monitors SSH changes around the station, but it includes troposphere, ionosphere and other errors. Hence, it is crucial to remove these errors for accurate GNSS-IR sea surface altimetry. This study introduces a new Deep-learning Composite atmospheric delay Correction Inversion Model (DCCIM), which integrates a long short-term memory network based on the traditional GNSS-IR algorithm and the factor-driven dataset. This approach considers the atmospheric delay for factor-driven dataset to improve the accuracy of GNSS-IR SSH inversion. The Pearson's correlation coefficient (PCC) between the DCCIM and tide gauge data is 0.92, with a maximum of 0.99 at GOM1 GNSS station. The root mean square error (RMSE) ranged from 4.35 cm (TRRG) to 7.13 cm. This strongly suggests that the DCCIM can be used to effectively monitor SSH changes. To objectively demonstrate the superiority of the DCCIM over traditional GNSS-IR, the DCCIM and GNSS-IR are used to invert SSH changes and then compared with tide gauge data. In addition, this finding also shows that the DCCIM significantly promote the SSH inversion accuracy compared with that of conventional GNSS-IR altimetry. The RMSE was 61.74% lower on average, and the PCC was 67.44% higher. This highly valuable study provides effective SSH monitoring and a coastal SSH inversion technique for high-precision ocean gravity field construction.