WEIGHT-SUPPORTED RANDOM FOREST DOWNSCALED GRACE (-FO) DATA UNCOVERS GROUNDWATER DEPLETION LINKED TO WINTER WHEAT CULTIVATION IN THE NORTH CHINA PLAIN

 

This study focuses on the Lower Indus Basin (LIB), where unsustainable groundwater extraction for irrigation demand is a critical issue. The standard solution, satellite gravity from GRACE and GRACE-Follow-On (GRACE-FO), provides valuable data on Total Water Storage Anomalies (TWSA), but its coarse spatial resolution (approximately 1°) is insufficient for localized water management. To overcome this spatial limitation, the research team developed and implemented a novel machine learning pipeline using a Weight-Supported Random Forest (WSRF) model. This model downscaled the coarse GRACE-derived data to a much finer 0.1° spatial resolution. The downscaling was achieved by training the WSRF model on a suite of independent, higher-resolution predictor variables, specifically: Precipitation (P), Evapotranspiration (ET), Normalized Difference Vegetation Index (NDVI), and Soil Moisture (SM). The resulting high-resolution GRACE-like products were validated with exceptionally high accuracy against actual piezometric well measurements, demonstrating a strong correlation coefficient (R2 > 0.85). The spatial downscaling revealed detailed, localized patterns of groundwater depletion that were obscured in the original coarse data. Analyzing the trends from 2003 to 2023 showed that this depletion is not solely a climatic event but is closely coupled with climate variability (including major drought events) and significantly driven by irrigation practices essential for regional agriculture, highlighting that this methodology provides a crucial tool for sustainable water management.

 

Graphical Abstract: The figure presents a detailed graphical representation of the methodology employed to downscale GRACE (-FO) derived GWSA data for evaluating groundwater storage distribution.

 

THE CHALLENGE

Groundwater is critical for agriculture in the Lower Indus Basin, yet direct measurements (piezometers) are sparse. The GRACE and GRACE-FO (Gravity Recovery and Climate Experiment) satellite missions provide essential data on Total Water Storage Anomaly (TWSA).

However, GRACE data is coarse (approx. 100  km x 100 km spatial resolution)—too coarse to observe local changes caused by individual irrigation canals or specific climatic impacts.

 

 

HIGHLIGHTS:

  • The RFSW model’s predictions reduced RMSE and residuals by 44.44% and 43.57%, respectively, compared to the RFG model’s.
  • GWSA obtained from RFSW downscaled GRACE(-FO) data indicates a higher declining trend in the PP sub-region of the NCP.
  • Variations in GWSA within the NCP indicate that winter wheat primarily contributes to groundwater depletion via irrigation practices.

Keywords North china plain · GRACE (-FO) · GWSA · RFSW· Downscaling · Winter wheat

 

Figure 1. (a)The 3D background layer created in ArcGIS based on the Digital Elevation Model (DEM), along with the colour-coded map, illustrates the spatial variation of specific yield (Sy), (b) the long-term mean (2003–2023) annual precipitation, (c) land use land cover map derived from Google Earth Engine (GEE), (d) population density per person in NCP derived using GEE, (e) irrigation at pixel’s scale (Siebert et al. 2015), and (f) the long-term (2005–2018) Mann Kendall trends of groundwater level data

Figure 2. A downscaling framework of GWSA driven from GRACE/GRACE-FO to 0.1° from 0.25° resolution over the region of NCP

Figure 3. GWSA spatial distribution comparison of GRACE (-FO) and RFSW at a resolution of 0.25° and 0.1° using long-term (2003–2023) Sen’s slope trend, (b) temporal variations at pixel scale, and (c) shows NCP region averaged GWSA monthly temporal variations estimated from GRACE (-FO) and RFSW downscaled GWSA. The red dotted line shows the trend line of the GWSA time series

 

KEY FINDINGS: GROUNDWATER DEPLETION TRENDS

The downscaled data reveals the “why” and “where” of groundwater loss in the Lower Indus Basin.

1. SPATIAL HOTSPOTS

    • Downscaled maps reveal distinct areas of significant groundwater depletion, often corresponding with areas of intensive, large-scale irrigation.

2. DRIVERS OF CHANGE

    • CLIMATE VARIABILITY: Reduced precipitation and extreme droughts accelerate depletion.
    • IRRIGATION DEMAND: High agricultural demand, particularly during dry seasons, is the primary driver.

Figure 4. (a) Sen’s slope long-term (2005–2018) spatial variations of the RFSW downscaled and Mann Kendall’s trend of in-situ GWSA values, (b) time series of the RFSW downscaled and in-situ GWSA in Shijiazhuang, (c) Xingtai, (d) Handan regions from 2005 to 2018

Figure 5. The winter wheat spatial distribution in different years over the NCP region

 

 

IMPACT: ADVANCING WATER MANAGEMENT

This research provides a powerful, high-resolution tool for regional water managers. By accurately mapping groundwater depletion at a finer scale, policymakers can make data-driven decisions regarding sustainable irrigation practices and climate adaptation strategies.

 

Find the full publication here:  https://doi.org/10.1007/s41748-025-00976-6


RELATED DATA

ALICE-LAB: Asian Land Information for Climate and Environmental Research Laboratory


Asian Hydrometeorological Simulation (AHS)

Access high-resolution monthly hydrometeorological data at 1 km scale for Asia, empowering advancements in water resources management, climate research, and environmental decision-making. Unlock precise insights for optimized conservation strategies, refined hydrological forecasts, and effective climate change mitigation.


Thailand Hydrometeorological Simulation (THS)

ALICE-LAB provides daily high-resolution hydrometeorological data for Thailand at a 1 km scale, enhancing water resource management and climate research. This detailed dataset refines hydrological forecasts, supports conservation strategies, and offers critical insights for effective climate change mitigation, driving better decision-making and operational efficiency.