Ren et al. (2026) Downscaling SMAP Soil Moisture to 250 m Using Deep Learning Models and Multi-Source Environmental Variables over the Loess Plateau

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-09-23
• Authors: Haihang Ren, Yu Zhao, Qingling Geng
• DOI: 10.3390/rs18193285

Research Groups

• Institute of Soil Science, Chinese Academy of Sciences
• University of California, Los Angeles (UCLA) Department of Geography

Short Summary

This study developed a deep-learning downscaling framework to improve soil moisture predictions at high spatial resolutions over heterogeneous terrains, achieving improved accuracy compared to traditional machine learning models.

Objective

• Investigate the feasibility of using progressive deep-learning to downscale SMAP L4 soil moisture from 9 km to 250 m resolution over complex terrains.

Study Configuration

• Spatial Scale: Regional scale (Loess Plateau)
• Temporal Scale: Monthly scale

Methodology and Data

• Models used:
  • SE-ResNet (progressive deep-learning downscaling framework)
  • XGBoost (traditional machine learning model)
  • RF (random forest model)
  • CNN (convolutional neural network)
  • ResNet (standard counterpart to SE-ResNet)
• Data sources:
  • SMAP L4 soil moisture
  • Multi-source environmental predictors (actual evapotranspiration, DEM, precipitation)

Main Results

• Actual evapotranspiration, DEM, and precipitation were identified as dominant factors controlling soil moisture variability.
• DL models outperformed traditional ML models on the 9 km test dataset, with SE-ResNet achieving the highest accuracy (R = 0.902, RMSE = 0.026 m3/m3).
• In-situ validation showed that 250 m downscaled products did not consistently outperform the original 9 km SMAP soil moisture.

Contributions

This study highlights the importance of considering environmental controls and scale-adaptive strategies when generating high-resolution soil moisture products over complex terrains, providing new insights into the limitations of traditional downscaling methods.

Funding

• National Natural Science Foundation of China (NSFC) grant # 91837204
• UCLA's Department of Geography research fund

Citation

@article{Ren2026Downscaling,
  author = {Ren, Haihang and Zhao, Yu and Geng, Qingling},
  title = {Downscaling SMAP Soil Moisture to 250 m Using Deep Learning Models and Multi-Source Environmental Variables over the Loess Plateau},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18193285},
  url = {https://doi.org/10.3390/rs18193285}
}