Gao et al. (2026) High-resolution daily surface soil moisture mapping over the Qinghai–Tibet Plateau via predictors fusion and machine learning

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-01-18
• Authors: Xiao Gao, Ping Lu, Yonghong Yi
• DOI: 10.1016/j.jhydrol.2026.134982

Research Groups

College of Surveying and Geo-Informatics, Tongji University, Shanghai 200092, China

Short Summary

This study developed a hyperparameter-optimized Random Forest downscaling framework to generate HRPSSM, a seamless 500 m daily surface soil moisture product for the Qinghai–Tibet Plateau from 2015–2023, demonstrating superior accuracy and spatial consistency compared to existing products.

Objective

• To develop a high-accuracy, high-spatiotemporal-resolution (500 m daily) surface soil moisture (SM) product for the Qinghai–Tibet Plateau (QTP) by downscaling coarse-scale SMAP retrievals using a machine learning framework and regionally calibrated high-resolution predictors.

Study Configuration

• Spatial Scale: Qinghai–Tibet Plateau (QTP), 500 m resolution.
• Temporal Scale: Daily, spanning 2015–2023.

Methodology and Data

• Models used: Hyperparameter-optimized Random Forest (RF) downscaling framework.
• Data sources: Coarse-scale SMAP retrievals, regionally calibrated high-resolution predictors (e.g., Enhanced Vegetation Index (EVI), soil organic carbon, soil texture, diurnal surface temperature difference, near-surface humidity), in-situ measurements (for validation), multiple benchmark datasets (for validation).

Main Results

• A seamless 500 m daily surface soil moisture product (HRPSSM) for the Qinghai–Tibet Plateau was generated, covering 2015–2023.
• HRPSSM demonstrated superior accuracy and spatial consistency compared to existing products, with a mean R of 0.78 and an unbiased Root Mean Square Error (ubRMSE) of 0.0369 m³/m³.
• The product effectively captures seasonal dynamics and rainfall responses across the QTP.
• Variable importance analysis indicated that EVI and soil organic carbon were the most significant predictors for downscaling performance, followed by soil texture and diurnal surface temperature difference. Meteorological factors, such as near-surface humidity, played greater roles during non-growing seasons and across subregions.

Contributions

• Fills a critical gap in high-resolution surface soil moisture products over the Qinghai–Tibet Plateau, a region highly susceptible to climate change.
• Provides a robust foundation for regional hydrological forecasting and climate–ecosystem interaction studies.
• Introduces a novel hyperparameter-optimized Random Forest downscaling framework for generating high-resolution soil moisture data.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Gao2026Highresolution,
  author = {Gao, Xiao and Lu, Ping and Yi, Yonghong},
  title = {High-resolution daily surface soil moisture mapping over the Qinghai–Tibet Plateau via predictors fusion and machine learning},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.134982},
  url = {https://doi.org/10.1016/j.jhydrol.2026.134982}
}