Hung et al. (2025) Downscaled global 60-meter resolution estimates of irrigation water sources (2000–2015)

Identification

• Journal: Scientific Data
• Year: 2025
• Date: 2025-10-09
• Authors: Fengwei Hung, Davide Danilo Chiarelli, J. S. Famiglietti, Marc F. Müller
• DOI: 10.1038/s41597-025-05920-x

Research Groups

• Environmental Change Initiative, University of Notre Dame, Notre Dame, Indiana, USA
• School of Sustainability, Arizona State University, Tempe, Arizona, USA
• Department of Civil and Environmental Engineering, Politecnico di Milano, Milano, Italy
• Eawag, Swiss Federal Institute of Aquatic Science and Technology, Dübendorf, Switzerland

Short Summary

This study generates high-resolution (60 meters) global maps of irrigation water sources (rainfed, groundwater, surface water) for 2000-2015 by downscaling existing datasets using spatial patterns and crop water requirements, demonstrating significantly improved accuracy in distinguishing groundwater use in the U.S. compared to previous global maps.

Objective

• To generate global maps of primary crop water sources (rainfall, groundwater, and surface water irrigation) at 60-meter resolution for 2000, 2005, 2010, and 2015, addressing the need for fine-resolution global groundwater irrigation maps.

Study Configuration

• Spatial Scale: Global, with output resolution of 60 meters. Input data included 5-arcmin (approximately 10 kilometers) for GMIA and WATNEEDS, and 30 meters for GLAD cropland data. Validation was conducted at resolutions from 300 meters to 5 kilometers. Major surface water features with a contributing drainage area greater than 25 square kilometers were considered.
• Temporal Scale: Reference years 2000, 2005, 2010, and 2015. Irrigation water requirements (BWR) were averaged over five-year periods surrounding each reference year (e.g., 1998–2002 for 2000). Cropland data was used in four-year intervals (e.g., 2012–2015 for 2015).

Methodology and Data

• Models used:
  • WATNEEDS (global crop water use model for blue water requirements)
  • Random Forest (machine learning algorithm for harmonizing AEI percentages)
  • Machine learning models (used for generating GLAD cropland dataset)
• Data sources:
  • Global Map of Irrigated Areas (GMIA) AEI dataset (5-arcmin resolution)
  • GMIA groundwater irrigation dataset (5-arcmin resolution)
  • Global Land Analysis and Discovery (GLAD) cropland dataset (30-meter resolution)
  • HydroRIVERS dataset (global river networks and major surface water features)
  • Global Human Settlement Layers (GHSL) dataset (urban areas, 1-kilometer resolution)
  • Validation data:
    • Landsat-based Irrigation Dataset (LANID) for the U.S. (30-meter resolution)
    • United States Groundwater Well Database (USGWD) (1,505,371 active irrigation well records)
    • Indian rice farmer survey (8,355 field locations with irrigation source information)
    • U.S. Geological Survey (USGS) county-level water use data

Main Results

• Generated a global dataset of irrigation water sources (rainfed, surface water, groundwater) at 60-meter resolution for 2000, 2005, 2010, and 2015.
• Validation against LANID (U.S. AEI): The downscaled AEI prediction (Pred) showed a national underestimation of -4% compared to LANID, significantly outperforming GMIA (+23% overestimation). Pred achieved an overall accuracy of 0.88 in cropland-masked scenarios, surpassing GMIA (0.86) and GMIA-C (0.85).
• Validation against USGWD (U.S. groundwater irrigation): At a 2-kilometer resolution, Pred achieved a user accuracy of 0.44 and an overall accuracy of 0.85, representing improvements of 0.20 and 0.60, respectively, over GMIA. Pred maintained a 20–30 percentage point lead in overall accuracy over GMIA across all tested resolutions (300 meters to 5 kilometers).
• Validation against Indian farmer survey (India groundwater irrigation): Both Pred and GMIA performed well, with producer accuracy consistently above 0.85 and user/overall accuracies generally exceeding 0.70. Pred showed modest improvements over GMIA, with user and overall accuracies a few percentage points higher, reflecting more precise targeting.
• The dataset includes estimates of uncertainty (e.g., PctPtErrorGWDownscale, PctPtErrorSWDownscale) and bias (PctPtErrorAEI) for each 10-kilometer pixel.

Contributions

• Provides the first global, high-resolution (60 meters) dataset that distinguishes between rainfed, surface water, and groundwater irrigation sources.
• Significantly enhances the spatial detail and accuracy of global irrigation water source mapping, particularly for groundwater irrigation, compared to existing coarse-resolution products like GMIA.
• Integrates diverse high-resolution geospatial data (cropland extent, surface water features) with crop water requirements and machine learning techniques for robust downscaling.
• Offers a valuable resource for improved analysis of agricultural water use, hydrological modeling, water resource management, and sustainability assessments at local to global scales.
• The complete downscaling methodology is openly available on Google Earth Engine and GitHub, facilitating replication, modification, and future updates.

Funding

• National Science Foundation (grants ICER1824951 and EAR 2142967)

Citation

@article{Hung2025Downscaled,
  author = {Hung, Fengwei and Chiarelli, Davide Danilo and Famiglietti, J. S. and Müller, Marc F.},
  title = {Downscaled global 60-meter resolution estimates of irrigation water sources (2000–2015)},
  journal = {Scientific Data},
  year = {2025},
  doi = {10.1038/s41597-025-05920-x},
  url = {https://doi.org/10.1038/s41597-025-05920-x}
}