A novel approach for soil moisture retrieval via fusing multi-spaceborne GNSS-R observations in entire Tibetan Plateau
Abstract
Spaceborne GNSS-Reflectometry (GNSS-R) is a powerful technology for soil moisture (SM) retrieval. However, complex topography and steep slopes over the Tibetan Plateau lead to significant signal loss and sampling sparsity. This study proposes a novel Dirichlet-prior-guided Vision Transformer (DG-ViT) framework to fuse multi-source observations from CYGNSS, Tianmu-1, and the FY-3E/F/G constellation. The novel DG-ViT framework incorporates land cover informed prior guidance into multi-spaceborne surface reflectivity (SR) fusion, and its performance is evaluated against single-source retrievals and traditional machine learning algorithms. To assess generalization, the fused SR was coupled with six mainstream machine learning algorithms (RF, XGBoost, ANN, BP, SVM, and GBDT) for 9 km daily SM retrieval, with multi-scale validation conducted against ERA5 SM and nine in-situ stations. Results demonstrate that the DG-ViT framework achieves a daily SR coverage of 89.49 %, representing an improvement of 20.13 % and 14.32 % over standalone CYGNSS and Tianmu-1 data, respectively. In addition, this fusion strategy overcomes the coverage gap of CYGNSS north of 38°N and compensates for the inability of individual FY-3 missions to support daily SM retrieval. In comparative assessments, the DG-ViT combined with RF yields superior performance (R = 0.955, RMSE = 0.039 cm 3 /cm 3 , bias = 0.000 cm 3 /cm 3 ), markedly enhancing retrieval accuracy relative to results derived from single-source CYGNSS or Tianmu-1 SR. In-situ validation shows ubRMSE values between 0.0232 and 0.0619 cm 3 /cm 3 , a 15.1 % precision gain over ERA5 SM. The proposed DG-ViT combined with RF approach effectively exploits the complementarity among multi-spaceborne GNSS-R observations over the entire Tibetan Plateau, thereby compensating for the signal deficiencies of individual missions. This fusion approach provides a novel perspective for SM retrieval in high-altitude areas, specifically improving daily SR coverage and SM retrieval accuracy through fusion of multi-spaceborne GNSS-R observations with land cover informed prior guidance.
// Source
Authors: Wenzhou Xu, Haohan Wei, Shuanggen Jin, Dong Chen, Hongtao shi, Tianjie Zhao, Zhehui Shen, Cong Yin, Rongrong Li, Wentao Dai, Songbo Qu, Mengfei Liu, Jingyao Zheng
Institutions: Chinese Academy of Sciences, Nanjing University of Information Science and Technology, China University of Mining and Technology, Anhui University, Nanjing Forestry University, Henan Polytechnic University, Aerospace Information Research Institute, National Space Science Center