Ali et al. (2025) Weight-Supported Random Forest Downscaled GRACE (-FO) Data Uncovers Groundwater Depletion Linked to Winter Wheat Cultivation in the North China Plain

Identification

• Journal: Earth Systems and Environment
• Year: 2025
• Date: 2025-12-22
• Authors: Shoaib Ali, Jiangjun Ran, Natthachet Tangdamrongsub, Behnam Khorrami, Vagner G. Ferreira, Haiyun Shi, W. M. Zhang
• DOI: 10.1007/s41748-025-00976-6

Research Groups

• Department of Earth and Space Sciences, Southern University of Science and Technology, Shenzhen, China
• Water Engineering and Management, School of Engineering and Technology, Asian Institute of Technology, Pathum Thani, Thailand
• Department of Remote Sensing & GIS, Faculty of Planning & Environmental Sciences, University of Tabriz, Tabriz, Iran
• School of Earth Sciences and Engineering, Hohai University, Nanjing, China
• School of Environmental Science and Engineering, Southern University of Science and Technology, Shenzhen, China
• Department of Geosciences and Natural Resource Management, University of Copenhagen, Copenhagen, Denmark

Short Summary

This study introduces a novel spatially weighted random forest (RF_SW) model to downscale GRACE (-FO) groundwater storage anomaly (GWSA) data to a high resolution (0.1°) across the North China Plain (NCP) from 2003 to 2023, revealing significant groundwater depletion directly linked to the expansion of winter wheat cultivation.

Objective

• To address the limitations of aspatial machine learning models in capturing spatial heterogeneity when downscaling GRACE (-FO) data, this study proposes a novel spatially weighted random forest (RF_SW) model to downscale GWSA to a high resolution (0.1°) across the North China Plain (NCP) from 2003 to 2023, and to uncover the relationship between groundwater depletion and agricultural practices, specifically winter wheat cultivation.

Study Configuration

• Spatial Scale: North China Plain (NCP), including Piedmont Plain (PP) and East-Central Plain (ECP) sub-regions. Downscaling from 0.25° to 0.1° spatial resolution.
• Temporal Scale: 2003 to 2023 for GRACE/GRACE-FO data; 2005 to 2018 for in-situ well data.

Methodology and Data

• Models used:
  • Spatially Weighted Random Forest (RFSW)
  • Global Random Forest (RFG)
  • Seasonal-Trend decomposition based on Loess (STL)
  • Generalized Three-Cornered Hat (GTCH) method
  • NOAH model (for soil moisture and snow water equivalent)
  • Random Forest Classifier (RFC)
  • Sen’s slope and Mann-Kendall test
  • Statistical metrics: Root Mean Square Error (RMSE), Coefficient of Determination (R²), Kling-Gupta Efficiency (KGE), Nash-Sutcliffe Efficiency (NSE)
• Data sources:
  • Satellite/Remote Sensing:
    • GRACE and GRACE-FO (CSR, JPL, GSFC Mascon products, specifically CSR-mascon RL06.2) for Terrestrial Water Storage Anomaly (TWSA)
    • Landsat-5/7/8 and Sentinel-2 multispectral imagery (via Google Earth Engine) for winter wheat mapping and Enhanced Vegetation Index (EVI)
    • MODIS (for Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), snow cover)
  • Reanalysis/Model Products:
    • Famine Early Warning Systems Network Land Data Assimilation System (FLDAS-NOAH) for hydro-meteorological parameters (precipitation, evapotranspiration, soil moisture, runoff, snow water)
    • Catchment Land Surface Model (CLSM) (for TWSA validation)
  • Observation/In-situ:
    • Monthly groundwater level anomaly from 559 observational wells (Chinese Groundwater Management Authority and ICGEM 2018)
    • Specific Yield (Sy) values (0.025 to 0.290)

Main Results

• The RFSW model significantly outperformed the RFG model, reducing RMSE by 44.44% and residuals by 43.57%. During testing, RF_SW achieved an R² of 0.951, NSE of 0.947, and RMSE of 31.81 mm.
• The downscaled GWSA data showed strong correlation with in-situ measurements from 559 monitoring wells (correlation coefficients ranging from 0.52 to 0.85).
• Significant groundwater depletion was observed in the Piedmont Plain (PP) and East-Central Plain (ECP) sub-regions of the NCP. The most severe losses were in Shijiazhuang (-17.08 mm/yr), Xingtai (-16.67 mm/yr), and Handan (-16.02 mm/yr).
• The winter wheat cultivation area in the NCP doubled from 2.5 million hectares in 2003 to 5.8 million hectares in 2022.
• This expansion of winter wheat cultivation was linked to a reduction in GWSA from -180 mm to -480 mm in the southern NCP, particularly in the PP region near Shijiazhuang.
• The South-to-North Water Diversion Project (SNWDP) and increased precipitation contributed to a partial recovery of GWSA in some areas after 2015, with an increase in wells showing an upward trend. However, RF_SW downscaled data still indicated a slight decline (-1.13 mm/yr) during the recovery period, while in-situ data showed an increase (+1.88 mm/yr).

Contributions

• Introduction of a novel spatially weighted random forest (RF_SW) model for GRACE (-FO) GWSA downscaling, which explicitly accounts for spatial heterogeneity and significantly outperforms traditional aspatial RF models.
• Generation of high-resolution (0.1°) GWSA data for the North China Plain, enabling more accurate assessment of local-scale groundwater dynamics.
• Clear quantification and spatial mapping of the link between winter wheat cultivation expansion and groundwater depletion in the NCP over a two-decade period (2003-2023).
• Improved understanding of local groundwater dynamics and their relationship to agricultural practices, providing valuable insights for targeted water management strategies in water-stressed regions.
• Validation of downscaled data with a large network of in-situ wells, demonstrating high correlation and reliability.

Funding

• National Key Research and Development Program of China (2021YFB3900604)
• National Natural Science Foundation of China (42322403, 42174096)

Citation

@article{Ali2025WeightSupported,
  author = {Ali, Shoaib and Ran, Jiangjun and Tangdamrongsub, Natthachet and Khorrami, Behnam and Ferreira, Vagner G. and Shi, Haiyun and Zhang, W. M.},
  title = {Weight-Supported Random Forest Downscaled GRACE (-FO) Data Uncovers Groundwater Depletion Linked to Winter Wheat Cultivation in the North China Plain},
  journal = {Earth Systems and Environment},
  year = {2025},
  doi = {10.1007/s41748-025-00976-6},
  url = {https://doi.org/10.1007/s41748-025-00976-6}
}