Yu et al. (2025) Soil MoistureRetrieval from TM-1 GNSS-R Reflections with Auxiliary Geophysical Variables: A Multi-Cluster and Seasonal Evaluation

Identification

• Journal: Land
• Year: 2025
• Date: 2025-12-24
• Authors: Jin Yu, Min Ji, Naiquan Zheng, Zhihua Zhang, Penghui Ding, Qian Zhao
• DOI: 10.3390/land15010036

Research Groups

• College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao, China
• Shandong Engineering Research Center for Beidou Navigation and Intelligent Spatial Information Technology Application, Qingdao, China
• Qingdao Key Laboratory of Beidou Navigation and Intelligent Spatial Information Technology Application, Qingdao, China
• Qingdao Surveying & Mapping Institute, Qingdao, China
• Shandong Engineering Research Center of Digital Intelligence Technology in Underground Space, Qingdao, China

Short Summary

This study develops a 9 km resolution global soil moisture retrieval model using Tianmu-1 (TM-1) GNSS-R reflectivity combined with auxiliary geophysical variables and a Random Forest algorithm. It demonstrates that a land-cover-based spatial clustering and seasonal temporal partitioning strategy significantly improves retrieval accuracy and stability, achieving a correlation coefficient (R) of 0.8155 and an unbiased RMSE (ubRMSE) of 0.0689 cm³/cm³ at the cluster level.

Objective

• Develop a global 9 km resolution soil moisture retrieval model using Tianmu-1 (TM-1) GNSS-R reflectivity and multi-source auxiliary variables.
• Systematically evaluate the impact of different spatial clustering schemes (including land-cover-based) and temporal partitioning (seasonal vs. chronological) strategies on retrieval performance.
• Compare TM-1 soil moisture estimates with SMAP and ISMN to quantify their consistency and assess TM-1's potential added value in regions where SMAP retrievals are sparse or quality-flagged.

Study Configuration

• Spatial Scale: Global coverage, 9 km spatial resolution (EASE-Grid 2.0).
• Temporal Scale: 2023 data, daily temporal resolution (hourly TM-1 observations averaged to daily, SMAP two daily overpasses averaged, ISMN minute/hour aggregated to daily).

Methodology and Data

• Models used: Random Forest (RF) algorithm. (CatBoost, LightGBM, XGBoost used for comparison).
• Data sources:
  • Tianmu-1 (TM-1) Level-1 reflectivity data (GNSS-R, hourly sampling frequency, DDM statistics/specular point reflectivity).
  • Soil Moisture Active Passive (SMAP) enhanced Level-3 radiometer soil moisture product (SPL3SMP_E, 9 km, daily).
  • ERA5-Land hourly dataset (precipitation, ~9 km, hourly).
  • VIIRS Global NDVI product (10-day compositing, 1 km).
  • GTOPO30 global digital elevation model (DEM, ~1 km).
  • MODIS MCD12C1 Climate Modeling Grid (CMG) product (Land cover, 17 IGBP classes, 0.05°).
  • International Soil Moisture Network (ISMN) in situ soil moisture observations (0–5 cm depth, hourly, aggregated to daily).
  • Auxiliary variables: VWC (Vegetation Water Content), Clay content, Surface Roughness, Slope, LST (Land Surface Temperature).

Main Results

• The model combining LC-cluster with seasonal partitioning achieved the best performance at the cluster level (R = 0.8155, ubRMSE = 0.0689 cm³/cm³).
• Strongest performance was observed in barren and shrub ecosystems (cluster-level R = 0.94 and 0.87, respectively; minimum ubRMSE = 0.043 cm³/cm³ for barren).
• At the grid level, the LC-cluster with seasonal partitioning yielded median R = 0.5251 and ubRMSE ≈ 0.0499 cm³/cm³.
• TM-1 demonstrated strong temporal fitting at selected ISMN sites, often outperforming SMAP in dynamic trend tracking and bias control (e.g., PSA6Plaenterwald: TM-1 R=0.8335, ubRMSE=0.03902 cm³/cm³ vs. SMAP R=0.7568, ubRMSE=0.04635 cm³/cm³).
• Overall comparison with ISMN showed TM-1 R = 0.560 and ubRMSE = 0.0895 cm³/cm³, comparable to SMAP (R = 0.605, ubRMSE = 0.0865 cm³/cm³).
• Feature importance analysis identified VWC, precipitation, and surface roughness as the dominant drivers for soil moisture prediction.
• TM-1 consistently exhibited stronger wetness responses than SMAP, particularly under heavy rainfall conditions, suggesting greater sensitivity to short-term precipitation events.

Contributions

• First study to develop a global 9 km soil moisture retrieval model using the Tianmu-1 (TM-1) constellation's GNSS-R reflectivity data.
• Systematic evaluation of spatial clustering (especially land-cover-based) and seasonal temporal partitioning strategies, demonstrating their effectiveness in improving accuracy and generalization.
• Demonstrates TM-1's capability to provide more balanced temporal sampling and availability in cloudy, rainy, and mid-to-high latitude regions, complementing SMAP's limitations.
• Shows TM-1's potential for temporal gap filling and rapid event response (e.g., moisture transitions post-precipitation) where SMAP data might be masked or intermittent.
• Confirms that TM-1 retrievals achieve mainstream performance levels for spaceborne GNSS-R soil moisture products, particularly in arid and semi-arid regions.

Funding

• Shandong Province Key R&D Program (Competitive Innovation Platform) Project, Grant No. 2024CXPT101
• Qingdao Science and Technology for the Benefit of People Project, Grant No. 25-1-5-xdny-12-nsh
• China Postdoctoral Science Foundation (76th General Program), Grant No. 2024M761845
• Open Fund of the Key Laboratory of Marine Environmental Survey Technology and Application, Ministry of Natural Resources, Grant No. MESTA-2024-B001
• Qingdao Natural Science Foundation (Young Scientists Program), Grant No. 25-1-1-54-zyyd-jch
• Shandong Provincial Postdoctoral Innovation Project, Grant No. SDCX-ZG-202502046
• National Program for Funding Postdoctoral Researchers (Category B), Grant No. GZB20250067

Citation

@article{Yu2025Soil,
  author = {Yu, Jin and Ji, Min and Zheng, Naiquan and Zhang, Zhihua and Ding, Penghui and Zhao, Qian},
  title = {Soil MoistureRetrieval from TM-1 GNSS-R Reflections with Auxiliary Geophysical Variables: A Multi-Cluster and Seasonal Evaluation},
  journal = {Land},
  year = {2025},
  doi = {10.3390/land15010036},
  url = {https://doi.org/10.3390/land15010036}
}