Zhang et al. (2026) High-resolution global groundwater storage anomalies: a 1 km downscaled dataset

Identification

• Journal: Earth system science data
• Year: 2026
• Date: 2026-10-19
• Authors: Chen Zhang, Yifei Fan, Wenxia Han, Yingchun Ge, J ZHU, Tao Che
• DOI: 10.5194/essd-18-7593-2026

Research Groups

• State Key Laboratory of Cryospheric Science and Frozen Soil Engineering, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou, China
• Heihe Remote Sensing Experimental Research Station, Key Laboratory of Remote Sensing of Gansu Province, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou, China
• University of Chinese Academy of Sciences, Beijing, China
• Key Laboratory of Green and High-end Utilization of Salt Lake Resources, Qinghai Institute of Salt Lakes, Chinese Academy of Sciences, Xining, China
• Qinghai Provincial Key Laboratory of Geology and Environment of Salt Lakes, Qinghai Institute of Salt Lakes, Chinese Academy of Sciences, Xining, China

Short Summary

This study developed a global 1 km monthly groundwater storage anomaly (GWSA) dataset for April 2002–December 2020 by downscaling GRACE observations using machine learning and high-resolution environmental predictors. The dataset provides enhanced spatial detail, showing high consistency with original GRACE data and reasonable agreement with in-situ observations, though with regional variations in performance.

Objective

• To develop a high-resolution (1 km) global monthly groundwater storage anomaly (GWSA) dataset from April 2002 to December 2020, addressing the coarse spatial resolution limitations of GRACE observations for characterizing groundwater variability at regional and local scales.

Study Configuration

• Spatial Scale: Global, with a 1 km spatial resolution for the downscaled product, derived from original GRACE data at approximately 0.25° resolution.
• Temporal Scale: Monthly, covering the period from April 2002 to December 2020.

Methodology and Data

• Models used:
  • Singular Spectrum Analysis (SSA) for temporal gap filling of GRACE data.
  • Extreme Gradient Boosting (XGBoost) and Deep Neural Networks (DNN) for aquifer-stratified machine-learning downscaling (XGBoost selected for production).
  • Cubic spline interpolation for residual correction.
  • Theil–Sen slope estimator for long-term trend estimation.
  • Mann-Kendall Significance Test for trend significance.
  • STL (Seasonal-Trend decomposition using LOESS) for time series decomposition.
• Data sources:
  • Satellite Observations: GRACE/GRACE-FO mass concentration (mascon) solutions (CSR RL0603, JPL RL06.3Mv04 CRI-filtered mascon product) for Terrestrial Water Storage Anomaly (TWSA).
  • Land Surface Models: GLDAS-2.1 Noah V2.1 (for soil water storage, snow water equivalent, vegetation canopy water storage), H08, and WaterGAP 2.2c (for groundwater storage benchmarks).
  • In-situ Observations:
    • Global groundwater-monitoring network (Jasechko et al., 2024) comprising 152,069 wells.
    • Global-Scale Time Series Dataset for Groundwater Studies within the Earth System (GROW; Bäthge et al., 2026).
    • China Groundwater Level Yearbook for Geo-environmental Monitoring (1074 stations).
  • High-resolution Environmental Predictors (1 km):
    • Meteorological: Temperature (TAS) and Precipitation (PRE) from CHELSA.
    • Hydrological: Actual surface evapotranspiration (ET) from PML_V2, Global Daily Surface Soil Moisture Dataset for surface soil moisture (SM).
    • Topographic and Soil: Global mean elevation (DEM) from GMTED2010, Soil attributes (Proportion of Clay, Silt, Sand, Coarse Fragments (CFVO), Cation Exchange Capacity (CEC), Soil pH, Soil Organic Carbon (SOC), Total Nitrogen (TN)) from SoilGrids.
    • Biophysical: Normalized Difference Vegetation Index (NDVI) from MOD13A1, Land Surface Temperature (LST) from MOD11A2.
  • Ancillary Data: World-wide Hydrogeological Mapping and Assessment Programme (WHYMAP) for aquifer boundaries, Global Surface Water Explorer (GSWE) for permanent water body mask, The Randolph Glacier Inventory (RGI) and MEaSUREs Greenland Ice Mapping Project (GIMP) for glacier and ice sheet masks.

Main Results

• A global 1 km monthly GWSA dataset for April 2002–December 2020 was successfully generated and is openly available.
• Singular Spectrum Analysis (SSA) effectively filled GRACE data gaps, with median reconstruction errors of 1.156 cm for short gaps and 1.679 cm for long gaps.
• XGBoost consistently outperformed Deep Neural Networks (DNN) in internal model tests and cross-scale consistency evaluations. When aggregated to the original 0.25° GRACE scale, the XGBoost product showed high agreement with the original GWSA (R² = 0.972, Pearson correlation coefficient (CC) = 0.986, Root Mean Square Error (RMSE) = 2.097 cm).
• Spatially independent in-situ validation using a held-out 0.25° grid, with effective specific yield (Syeff) correction, yielded R² = 0.538, RMSE = 0.378 cm/year, and Bias = 0.106 cm/year for XGBoost. Sensitivity analysis showed Syeff uncertainty primarily affected amplitude-related metrics (e.g., regression slope) more than correlation or trend-direction.
• Conversion-free in-situ temporal comparisons showed strong agreement in climate-driven basins (e.g., Stampriet-Kalahari Basin CC = 0.82) but revealed decoupling in some anthropogenic-dominated basins (e.g., Northern Great Plains Aquifer CC = -0.23), suggesting uncaptured human impacts.
• Monthly validation across representative aquifers indicated better reproduction of long-term variability (mean CC for trend = 0.55) than seasonal fluctuations (mean CC for seasonal = 0.39).
• Regional comparisons with published estimates showed strong agreement in the North China Plain (r = 0.963, RMSE = 32.0 mm), but weaker depletion magnitudes in northwestern India and California, and a trend-sign mismatch in the High Plains Aquifer during 2003–2013.
• The 1 km downscaled product revealed predictor-informed spatial details, such as dendritic textures in the Amazon Basin, canyon-aligned patterns in southern Tibet, and localized groundwater-storage gradients aligned with agricultural boundaries in intensively managed regions.
• Global climatology showed significant groundwater depletion zones between 25° and 45° N, associated with intensive human water use and arid conditions, and accumulation zones in tropical rainforests and mid-to-high latitudes.
• Seasonality dominated GWSA dynamics in 52.30% of the global land area, particularly in regions with regular hydrological rhythms.
• Long-term trend analysis (2002–2020) indicated that approximately 38.84% of effective land area experienced significant groundwater depletion, while 49.16% showed significant recharge.
• Residual signals, representing differences from GHM-simulated backgrounds, showed localized negative patches in intensively irrigated regions (e.g., Indus Basin, North China Plain), consistent with potential anthropogenic influences not fully captured by GHMs.

Contributions

• Generation of the first global 1 km monthly groundwater storage anomaly (GWSA) dataset, providing unprecedented spatial detail for groundwater monitoring from April 2002 to December 2020.
• Development of an innovative downscaling framework combining Singular Spectrum Analysis (SSA) for temporal gap filling, aquifer-stratified machine learning (XGBoost), and 19 high-resolution environmental predictors.
• Comprehensive multi-scale validation, including internal model diagnostics, cross-scale consistency checks, and spatially independent in-situ evaluations using a global network of over 152,000 groundwater monitoring wells.
• Quantification of the sensitivity of in-situ validation results to the effective specific yield (S_yeff) conversion coefficient, highlighting its impact on amplitude-related metrics.
• Detailed analysis of global climatology, seasonal contributions, and spatiotemporal evolution patterns of GWSA at 1 km resolution, revealing predictor-informed local hydrological gradients and regional differences in groundwater dynamics.
• Identification of residual depletion signals not explained by global hydrological model (GHM) simulations, offering a valuable diagnostic for areas potentially influenced by anthropogenic activities.

Funding

• Gansu Provincial Science and Technology Program (grant no. 23ZDKA0004)
• National Natural Science Foundation of China (grant no. 42595534)
• Qinghai Province Major Science and Technology Project (grant no. 2025-SF-A2)
• Qinghai Province through the Kunlun Talent Program (for Wenxia Han)

Citation

@article{Zhang2026Highresolution,
  author = {Zhang, Chen and Fan, Yifei and Han, Wenxia and Ge, Yingchun and ZHU, J and Che, Tao},
  title = {High-resolution global groundwater storage anomalies: a 1 km downscaled dataset},
  journal = {Earth system science data},
  year = {2026},
  doi = {10.5194/essd-18-7593-2026},
  url = {https://doi.org/10.5194/essd-18-7593-2026}
}