Scholarlink

<p>Snow Water Equivalent (SWE) is a critical parameter for characterizing snow water resources. However, its continuous and fine-scale retrieval remains challenging due to uneven snow distribution, rapid changes, and sparse ground observation stations. This study proposes a Multi-branch Feature Fusion for SWE retrieval Network (MBF-SWENet) integrating Transformer and Convolutional Neural Network to achieve high-precision SWE retrievals at a spatial resolution of approximately 5&#xa0;km, utilizing multi-mode Global Navigation Satellite Systems Reflectometry (GNSS-R) signals acquired from the Chinese Tianmu-1 (TM-1) satellite constellation. First, GNSS-R data collected from 22 operational TM-1 satellites were processed and spatiotemporally matched with the Copernicus SWE products to construct an experimental dataset. The MBF-SWENet model employs a Transformer module to capture global spatial dependencies and a CNN module to extract local texture features from Delay-Doppler Maps (DDMs). A feature fusion module is developed to incorporate both GNSS-R characteristics and geometric parameters to enhance feature representation for the establishment of an SWE retrieval model. Five experiments were then conducted to comprehensively evaluate the SWE retrieval performance and influencing factors of multi-mode GNSS-R data. Compared with the Copernicus SWE product, the SWE retrieved based on BeiDou Navigation Satellite System Reflection (BDS-R) data achieved a root mean square error (RMSE) of 0.84&#xa0;cm for SWE ranges from 0 to 35&#xa0;cm and a correlation coefficient of 0.977. Cross-system modeling and retrieval tests have demonstrated that the strong generalization capability of the MBF-SWENet model across multi-system GNSS, with Lin’s concordance correlation coefficient ranging from 0.944 to 0.978. A fusion strategy employing the union of multi-system GNSS-R data not only improved spatiotemporal coverage by approximately 240% but also reduced the SWE retrieval RMSE by up to 33%, achieving a minimum RMSE of 0.56&#xa0;cm. This study is the first to systematically validate the effectiveness of TM-1 multi-mode GNSS-R data for SWE retrieval, offering a novel, cost-effective solution with a high spatiotemporal resolution for global snow hydrology monitoring.</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

Preliminary validation of Tianmu-1 multi-GNSS reflections for snow water equivalent retrieval using a MBF-SWENet model

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • Yongchao Zhu,
  • Xinyu Ma,
  • Maorong Ge,
  • Xiaohong Zhang,
  • Kefei Zhang,
  • Kegen Yu,
  • Xiaochuan Qu

摘要

Snow Water Equivalent (SWE) is a critical parameter for characterizing snow water resources. However, its continuous and fine-scale retrieval remains challenging due to uneven snow distribution, rapid changes, and sparse ground observation stations. This study proposes a Multi-branch Feature Fusion for SWE retrieval Network (MBF-SWENet) integrating Transformer and Convolutional Neural Network to achieve high-precision SWE retrievals at a spatial resolution of approximately 5 km, utilizing multi-mode Global Navigation Satellite Systems Reflectometry (GNSS-R) signals acquired from the Chinese Tianmu-1 (TM-1) satellite constellation. First, GNSS-R data collected from 22 operational TM-1 satellites were processed and spatiotemporally matched with the Copernicus SWE products to construct an experimental dataset. The MBF-SWENet model employs a Transformer module to capture global spatial dependencies and a CNN module to extract local texture features from Delay-Doppler Maps (DDMs). A feature fusion module is developed to incorporate both GNSS-R characteristics and geometric parameters to enhance feature representation for the establishment of an SWE retrieval model. Five experiments were then conducted to comprehensively evaluate the SWE retrieval performance and influencing factors of multi-mode GNSS-R data. Compared with the Copernicus SWE product, the SWE retrieved based on BeiDou Navigation Satellite System Reflection (BDS-R) data achieved a root mean square error (RMSE) of 0.84 cm for SWE ranges from 0 to 35 cm and a correlation coefficient of 0.977. Cross-system modeling and retrieval tests have demonstrated that the strong generalization capability of the MBF-SWENet model across multi-system GNSS, with Lin’s concordance correlation coefficient ranging from 0.944 to 0.978. A fusion strategy employing the union of multi-system GNSS-R data not only improved spatiotemporal coverage by approximately 240% but also reduced the SWE retrieval RMSE by up to 33%, achieving a minimum RMSE of 0.56 cm. This study is the first to systematically validate the effectiveness of TM-1 multi-mode GNSS-R data for SWE retrieval, offering a novel, cost-effective solution with a high spatiotemporal resolution for global snow hydrology monitoring.