Scholarlink

<p>This study proposes a novel approach to refine zenith tropospheric delay (ZTD) products firstly, using Feedforward Neural Networks (FNN). In our previous work, two advanced ZTD products, provided by TUW-VMF3 and GFZ-VMF3, were comprehensively assessed using 12,552 GNSS sites worldwide, which provides a detailed reference for the use of the above-mentioned ZTD products both in GNSS community and other space geodesy fields. Meanwhile, it is found that spatial and temporal interpolation is inevitably required when computing the ZTD at any location and any time, limiting their accuracy. Therefore, this article is dedicated to improving the accuracy of the ZTD products for the first time, in which a refinement method was proposed based on FNN. The refined models were extensively validated, showing significant improvements in both TUW-VMF3 and GFZ-VMF3 products. For TUW-VMF3, regional and global models improved accuracy by 34.13% and 27.58%, respectively; for GFZ-VMF3, improvements reached 33.51% and 26.72%. Regional models consistently outperformed global ones, with the most notable enhancements observed in Europe, followed by Japan, Australia, North America, South America, and China. In the EU region and AU region, 100% of the stations achieved an average RMSE below 20 mm. Notably, lower performance in South America and China is attributed to complex terrain and sparse GNSS coverage. This study demonstrates the potential of machine learning for refining ZTD products, offering potential support for the development of GNSS-meteorology and geodesy applications.</p>

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

Improving the accuracy of zenith tropospheric delay in TUW-VMF3 and GFZ-VMF3 using feedforward neural networks

  • 期刊论文

  • 发表日期：

    2025年6月27日

查看全文
  • Jian Li,
  • Fei Yang,
  • Lei Fan,
  • Debao Yuan,
  • Guifei Jing,
  • Zhuangzhuang Wang,
  • Jia Xie,
  • Yanru Yang

摘要

This study proposes a novel approach to refine zenith tropospheric delay (ZTD) products firstly, using Feedforward Neural Networks (FNN). In our previous work, two advanced ZTD products, provided by TUW-VMF3 and GFZ-VMF3, were comprehensively assessed using 12,552 GNSS sites worldwide, which provides a detailed reference for the use of the above-mentioned ZTD products both in GNSS community and other space geodesy fields. Meanwhile, it is found that spatial and temporal interpolation is inevitably required when computing the ZTD at any location and any time, limiting their accuracy. Therefore, this article is dedicated to improving the accuracy of the ZTD products for the first time, in which a refinement method was proposed based on FNN. The refined models were extensively validated, showing significant improvements in both TUW-VMF3 and GFZ-VMF3 products. For TUW-VMF3, regional and global models improved accuracy by 34.13% and 27.58%, respectively; for GFZ-VMF3, improvements reached 33.51% and 26.72%. Regional models consistently outperformed global ones, with the most notable enhancements observed in Europe, followed by Japan, Australia, North America, South America, and China. In the EU region and AU region, 100% of the stations achieved an average RMSE below 20 mm. Notably, lower performance in South America and China is attributed to complex terrain and sparse GNSS coverage. This study demonstrates the potential of machine learning for refining ZTD products, offering potential support for the development of GNSS-meteorology and geodesy applications.