Scholarlink

<p>Ray-tracing using numerical weather models (NWMs) represents one of the most precise methods for deriving tropospheric delays in high-precision space geodetic applications. However, considerable computational demands, particularly those associated with input/output operations, interpolation procedures, and data transmission, constrain contemporary global tropospheric delay grid products to coarse spatial resolutions, thereby introducing centimeter-to-decimeter-level biases in large height difference areas. This study introduces TropDS, a novel downscaling framework for tropospheric delays that integrates physical residual modeling with a U-Net artificial intelligence (AI) refinement. By independently characterizing the spatiotemporal deterministic and stochastic components, TropDS produces high-precision, high-resolution products from low-resolution global tropospheric delay fields. The model was trained and validated on the TUW VMF3_OP global grid spanning 2020–2024 and subsequently tested on 2025 data. Evaluation results for the testing year demonstrate that TropDS enhances global zenith hydrostatic delay (ZHD) and zenith wet delay (ZWD) accuracies to 90.97% and 85.76%, respectively, with global root mean square error (RMSE) reduced to less than 2&#xa0;mm. Inference for a single epoch data requires only ~ 0.04&#xa0;s. Additional validation on 2025 VMF3_FC-based products confirms the model’s efficacy for forecasted tropospheric delay. Experiments further examine the application of TropDS to tropospheric delay forecasts using AI weather forecast foundation model frameworks, specifically Pangu-Weather, GraphCast, and FengWu, revealing substantial improvements in early-stage forecast precision. This research establishes a robust methodology for generating high-resolution tropospheric products, with significant implications for geodetic investigations requiring precise atmospheric corrections. The code and the pre-trained model are available at <a href="https://github.com/Sardingfish/TropDS">https://github.com/Sardingfish/TropDS</a>.</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

TropDS: downscaling to enhance tropospheric delay grid precision in space geodesy

  • 期刊论文

  • 发表日期：

    2026年7月5日

查看全文
  • Junsheng Ding,
  • Wu Chen,
  • Junping Chen,
  • Jungang Wang,
  • Yize Zhang

摘要

Ray-tracing using numerical weather models (NWMs) represents one of the most precise methods for deriving tropospheric delays in high-precision space geodetic applications. However, considerable computational demands, particularly those associated with input/output operations, interpolation procedures, and data transmission, constrain contemporary global tropospheric delay grid products to coarse spatial resolutions, thereby introducing centimeter-to-decimeter-level biases in large height difference areas. This study introduces TropDS, a novel downscaling framework for tropospheric delays that integrates physical residual modeling with a U-Net artificial intelligence (AI) refinement. By independently characterizing the spatiotemporal deterministic and stochastic components, TropDS produces high-precision, high-resolution products from low-resolution global tropospheric delay fields. The model was trained and validated on the TUW VMF3_OP global grid spanning 2020–2024 and subsequently tested on 2025 data. Evaluation results for the testing year demonstrate that TropDS enhances global zenith hydrostatic delay (ZHD) and zenith wet delay (ZWD) accuracies to 90.97% and 85.76%, respectively, with global root mean square error (RMSE) reduced to less than 2 mm. Inference for a single epoch data requires only ~ 0.04 s. Additional validation on 2025 VMF3_FC-based products confirms the model’s efficacy for forecasted tropospheric delay. Experiments further examine the application of TropDS to tropospheric delay forecasts using AI weather forecast foundation model frameworks, specifically Pangu-Weather, GraphCast, and FengWu, revealing substantial improvements in early-stage forecast precision. This research establishes a robust methodology for generating high-resolution tropospheric products, with significant implications for geodetic investigations requiring precise atmospheric corrections. The code and the pre-trained model are available at https://github.com/Sardingfish/TropDS.