Scholarlink

This study presents a novel method for the inversion of terrestrial hydrological load displacements (HYDLDs) based on the Convolutional Neural Network (CNN) integrating data from the Gravity Recovery and Climate Experiment (GRACE) and Global Navigation Satellite System (GNSS) across 36 observation stations in the Chinese mainland. The initial phase involves the correction of GNSS data for atmospheric loading and non-tidal ocean loading to obtain HYDLDs. Subsequently, water storage data provided by GRACE, along with the corresponding latitudes and longitudes, are utilized as inputs to train the CNN model, with the target being HYDLDs observed by GNSS. The trained model is then employed to invert HYDLDs, and their accuracy is also extensively analyzed. Results indicate that corrections applied to GNSS data for atmospheric and non-tidal ocean loading led to an average reduction of 18.35 and 3.12% in the annual and semi-annual amplitudes of the observation stations, respectively. Compared to the traditional Green’s function approach, the CNN algorithm demonstrated a higher correlation with the GNSS observed HYDLDs. The mean and standard deviations of the differences between the CNN method and GNSS observations were −0.038 and 1.45 mm, respectively, marking a significant decline of 13.3 and 2.1 times compared to those obtained by the Green’s function method. These findings confirm the superior efficacy of the CNN method in HYDLD inversion.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Inversion of Hydrological Load Displacements Based on Convolutional Neural Networks and GRACE Water Storage Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rui Zhang,
  • Jiaqi Liu,
  • Wei Jiang,
  • Zan Li

摘要

This study presents a novel method for the inversion of terrestrial hydrological load displacements (HYDLDs) based on the Convolutional Neural Network (CNN) integrating data from the Gravity Recovery and Climate Experiment (GRACE) and Global Navigation Satellite System (GNSS) across 36 observation stations in the Chinese mainland. The initial phase involves the correction of GNSS data for atmospheric loading and non-tidal ocean loading to obtain HYDLDs. Subsequently, water storage data provided by GRACE, along with the corresponding latitudes and longitudes, are utilized as inputs to train the CNN model, with the target being HYDLDs observed by GNSS. The trained model is then employed to invert HYDLDs, and their accuracy is also extensively analyzed. Results indicate that corrections applied to GNSS data for atmospheric and non-tidal ocean loading led to an average reduction of 18.35 and 3.12% in the annual and semi-annual amplitudes of the observation stations, respectively. Compared to the traditional Green’s function approach, the CNN algorithm demonstrated a higher correlation with the GNSS observed HYDLDs. The mean and standard deviations of the differences between the CNN method and GNSS observations were −0.038 and 1.45 mm, respectively, marking a significant decline of 13.3 and 2.1 times compared to those obtained by the Green’s function method. These findings confirm the superior efficacy of the CNN method in HYDLD inversion.