干旱区生态安全与可持续发展服务平台

Multi-source fusion and machine learning downscaling of soil moisture in the arid and semi-arid regions of China

Liu, Zijian ， Li, Hongrui ， Li, Mengyang ， Zhou, Peng ， Wang, Ziming

2026-05-01 JOURNAL OF ARID ENVIRONMENTS 2026   235(卷), null(期), (null页)

查看原文

  • JCR分区:

    影响因子:

    • Soil moisture (SM) is a critical variable in the global climate system and terrestrial water cycle, yet obtaining high-accuracy SM data in China's arid and semi-arid regions remains challenging due to the coarse resolution and limited accuracy of existing products. This study developed a machine learning downscaling framework integrating multi-source SM data. First, three SM products were evaluated and fused using the Extended Triple Collocation method. Then, a Random Forest model established the relationship between SM and environmental predictors, enabling downscaling from 0.25 degrees & times; 0.25 degrees to 0.05 degrees & times; 0.05 degrees resolution. Finally, XGBoost performed bias correction using ground observations. Validation showed the fused SM data achieved a higher signal-tonoise ratio, and the downscaling model attained an R of 0.9253 and RMSE of 0.0265 m3/m3. Independent verification (R = 0.708) confirmed the downscaled SM effectively captures fine spatial details influenced by micro-topography and human activities, providing a reliable high-resolution dataset for the region.

    关键词

    Soil moisture Arid and semi-arid of China Machine learning Data downscaling

    相关推荐