Hussainzada et al. (2026) Comprehensive framework for agricultural water management in data-scarce regions: Integration of hydrological models and remotely sensed crop type data

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-02-03
• Authors: Wahidullah Hussainzada, Han Soo Lee, Ahmad Tamim Samim
• DOI: 10.1016/j.jhydrol.2026.135073

Research Groups

• Luhdorff & Scalmanini Consulting Engineers, Woodland, CA, USA
• Transdisciplinary Science and Engineering Program, Graduate School of Advanced Science and Engineering, Hiroshima University, Japan
• Center for Planetary Health and Innovation Science (PHIS), The IDEC Institute, Hiroshima University, Japan
• Smart Energy, Graduate School of Innovation and Practice for Smart Society, Hiroshima University, Japan

Short Summary

This study proposes a comprehensive framework for agricultural water management in data-scarce regions by integrating hydrological modeling (WRF-Hydro) with remotely sensed crop type data and machine learning, demonstrating significant potential for water savings through improved irrigation efficiency in the Amu River Basin.

Objective

• To develop and apply a comprehensive framework for sustainable agricultural water management in data-scarce regions, specifically the Amu River Basin, by quantifying water supply through hydrological modeling and agricultural water demand using remotely sensed crop type data and the FAO Penman-Monteith method, and assessing the impact of irrigation efficiency improvements.

Study Configuration

• Spatial Scale: Amu River Basin (ARB), northeastern Afghanistan, covering three subcatchments: Kokcha (22,367.7 km²), Khanabad (11,993.5 km²), and Kunduz (28,023 km²). Hydrological modeling was performed at a 3 km spatial resolution, and crop type mapping at 250 m resolution, with an elevation range of 308 to 6847 m.
• Temporal Scale: The study period for simulations and predictions was from 2014 to 2019, with model calibration from 2014 to 2016 and validation from 2017 to 2019. Data inputs included daily discharge, 16-day Normalized Difference Vegetation Index (NDVI), and daily meteorological parameters.

Methodology and Data

• Models used:
  • Hydrological: WRF-Hydro stand-alone model coupled with the Noah Multi-Physics (Noah-MP) Land Surface Model.
  • Machine Learning (ML) for crop type mapping: Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Model (GBM), and a multi-model ensemble using a maximum voting classifier.
  • Irrigation Water Requirement (IWR) estimation: UNFAO Penman–Monteith method.
• Data sources:
  • Atmospheric forcing: Global Land Data Assimilation System (GLDAS) Version 2.1 gridded data (0.25° spatial, 3-hourly temporal resolution).
  • Land use: MODIS combined Terra + Aqua land cover product (International Geosphere-Biosphere Program (IGBP) 2-classes, 30 arcsec).
  • Soil texture: United Nation Food and Agriculture Organization (UNFAO) soil map (5 km).
  • Elevation: Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Digital Elevation Model (DEM) (30 m).
  • Observed river discharge: Ministry of Energy and Water (MEW) (point data, daily).
  • Crop type mapping input: MODIS Aqua (MYD13Q1) and Terra (MOD13Q1) NDVI products (250 m spatial, 16-day temporal resolution).
  • Crop type training data: UNFAO 2020 crop type map (10 m spatial resolution, based on Sentinel-2 imagery).
  • In-situ meteorological data for Crop Water Requirement (CWR)/Reference Evapotranspiration (ET0): 13 Automated Weather Stations (AWSs) from MEW (point data, daily).
  • Missing meteorological data imputation: NASA Prediction of Worldwide Energy Resources (POWER) dataset.

Main Results

• The WRF-Hydro model, after calibration and optimization of Land Surface Model (LSM) parameters, significantly improved the simulation of snow accumulation and melt, and peak flow, aligning well with ground observations. Statistical indicators for daily discharge showed varying performance across five stations, with correlation coefficients (R) ranging from 0.42 to 0.85, Nash-Sutcliffe Efficiency (NSE) from -8.64 to 0.52, and Kling-Gupta Efficiency (KGE) from -0.56 to 0.74.
• A multi-model ensemble of Random Forest, Support Vector Machine, and Gradient Boosting models, trained on MODIS NDVI data and UNFAO crop maps, successfully predicted spatial crop type maps for five major crops (wheat, rice, cotton, barley, maize) across eight elevation zones from 2014 to 2019. The ensemble approach yielded higher accuracy than individual models, with an average total cultivated area of 249,400 ha.
• Irrigation Water Requirements (IWRs) were estimated for major crops using the Penman–Monteith method. Analysis of water balance revealed that available river discharge generally exceeded irrigation demand from June to August, but demand closely matched supply at the beginning and end of the irrigation season.
• Current irrigation systems in the ARB were found to be highly inefficient, with an overall efficiency of 48%. Improving irrigation canal conditions (lining) could increase water availability by 9% (to 57% efficiency), and further adopting efficient field application methods like sprinkler irrigation could increase water availability by up to 23.25% (to 71.25% efficiency).

Contributions

• Proposes a comprehensive and flexible framework for agricultural water resource management in data-scarce regions by integrating hydrological modeling, remote sensing-based crop mapping, and conventional irrigation assessment methods.
• Quantifies both water supply (via WRF-Hydro) and agricultural water demand (via IWR estimation and spatially explicit crop maps), explicitly linking these two critical aspects of water resource management.
• Improves the performance of the WRF-Hydro model in snowmelt-influenced mountainous watersheds through targeted Land Surface Model (LSM) parameterization.
• Enhances the accuracy of crop type mapping through the utilization of multi-model ensemble machine learning techniques, incorporating spatially explicit crop type maps to capture temporal and spatial variability in agricultural land use.
• Offers a time-effective and low-cost alternative for preliminary investigations and development of water resource management plans in data-limited regions, supporting informed decision-making for sustainable water use.

Funding

The first author was supported by the Japanese Government Scholarship MEXT, Japan. The authors declare that no other funds, grants, or support were received during the preparation of this manuscript.

Citation

@article{Hussainzada2026Comprehensive,
  author = {Hussainzada, Wahidullah and Lee, Han Soo and Samim, Ahmad Tamim},
  title = {Comprehensive framework for agricultural water management in data-scarce regions: Integration of hydrological models and remotely sensed crop type data},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135073},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135073}
}