Monitoring of Caspian Sea-level changes using deep learning-based 3D reconstruction of GRACE signal

The gravity recovery and climate experiment (GRACE) satellites detect changes in the distribution of water on the Earth surface based on the gravity anomaly. A correct reconstruction of the GRACE signal is still a challenging problem due to signal attenuation and noise levels. In this contribution, the GRACE data are analysed using a deep neural network to investigate changes caused in the Caspian Sea level (CSL). The novelty is to reconstruct the three-dimensional GRACE signal for reducing stripe errors, leakage error and gap filling. The reduction of the CSL was approximately 70 ? 0.2 cm from 2005 to 2016, with an annual trend of -6.77 ? 0.2 cm/year. The northern regions of CSL have a smaller annual amplitude than other regions. The proposed method has an average significant correlation of 82% with satellite altimetry and data collected by three tide gauge stations, thus showing good compatibility.

Automated summary

The study uses a deep neural network to analyze GRACE data and investigate changes in the Caspian Sea level, reconstructing the GRACE signal to reduce errors. The research found that the Caspian Sea level decreased by approximately 70 cm from 2005 to 2016, with an annual trend of -6.77 cm/year, and the northern regions had a smaller amplitude of change compared to other regions.