Yacoob et al. (2026) A machine learning approach for quantifying crop water stress in smallholder farms using unmanned aerial vehicle multispectral imagery

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-01-13
• Authors: Ameera Yacoob, Shaeden Gokool, Alistair Clulow, Maqsooda Mahomed, Vivek Naiken, Tafadzwanashe Mabhaudhi
• DOI: 10.1016/j.agwat.2026.110142

Research Groups

• Centre for Water Resources Research, School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, South Africa
• Discipline of Agrometeorology, School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, South Africa
• Centre for Transformative Agricultural and Food Systems, School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, South Africa
• Centre of Climate Change and Planetary Health, London School of Hygiene and Tropical Medicine, London, United Kingdom

Short Summary

This study developed a machine learning model to predict the Normalised Difference Water Index (NDWI) from UAV multispectral imagery for quantifying crop water stress in smallholder sugarcane farms. The model achieved high predictive accuracy (R² = 0.95) and effectively captured temporal variations in sugarcane water status, supporting precision water management.

Objective

• To develop and validate a machine learning model for predicting the Normalised Difference Water Index (NDWI) from UAV-acquired multispectral imagery, using satellite-derived NDWI for training, to quantify crop water stress in smallholder sugarcane farms and support precision water management.

Study Configuration

• Spatial Scale: A small-scale sugarcane field covering 7253.74 square meters in Swayimane, KwaZulu-Natal, South Africa. UAV multispectral imagery had a ground sample distance (GSD) of 0.052 meters per pixel, while Sentinel-2 satellite data ranged from 10 meters to 60 meters resolution.
• Temporal Scale: Data collection occurred from 1 July 2023 to 15 May 2024, encompassing the stalk elongation and early maturation phases of the sugarcane crop's 547-day growth cycle. Twelve UAV missions were conducted between 1 July 2023 and 15 March 2024.

Methodology and Data

• Models used:
  • Machine Learning Algorithms (MLAs): Generalised Linear Model (GLM), k-Nearest Neighbours (kNN), Classification and Regression Trees (CART), Random Forest (RF), Gradient Tree Boosting (GTB), Support Vector Machine (SVM).
  • Ensemble models (two developed).
  • Penman-Monteith equation (for reference evapotranspiration, ET0).
  • Pearson correlation analysis.
• Data sources:
  • UAV-acquired multispectral imagery (DJI Matrice 300 with MicaSense Altum camera).
  • Sentinel-2 satellite data (for training the ML model with NDWI and Structural Vegetation Indices (SVIs)).
  • In situ measurements:
    • Actual evapotranspiration (ETa) via an Eddy Covariance (EC) system.
    • Precipitation (millimeters), wind speed (meters per second), solar irradiance (megajoules per square meter), air temperature (degrees Celsius), relative humidity (from meteorological flux tower and Automatic Weather Station).
    • Soil heat flux, soil temperature, soil moisture (from EC system sensors).
    • Chlorophyll content (micromoles per square meter) using a Konica Minolta SPAD 502m.
    • Leaf Area Index (dimensionless) using an LAI-2200C sensor.
    • Crop height (meters).
  • Water Deficit Index (WDI) derived from ETa and potential ET (ETp).
  • Total Soil Water Profile (TSWP).

Main Results

• The best-performing ensemble machine learning model (Ensemble Model 2) achieved high predictive accuracy for NDWI with an R² of 0.95, a Root Mean Squared Error (RMSE) of 0.03, and a Mean Absolute Error (MAE) of 0.02.
• Predicted NDWI values effectively captured temporal variations in sugarcane water status, including post-rainfall stress recovery and increased water retention during early maturation, aligning with changes in Leaf Area Index (LAI), chlorophyll content (CC), and Total Soil Water Profile (TSWP).
• NDWI showed a strong positive correlation with actual evapotranspiration (ETa; R² = 0.60) and a strong negative correlation with the Water Deficit Index (WDI; R² = 0.62).
• In situ measurements revealed significant physiological changes from the stalk elongation (SE) to early maturation (M) phases: CC increased from 249.87 to 550.45 µmol m⁻², LAI from 3.92 to 5.13, and crop height from 2.26 to 2.73 meters.
• Predicted NDWI exhibited positive correlations with UAV-derived Structural Vegetation Indices (SVIs) such as NDVI (R = 0.51) and GNDVI (R = 0.52), and with in situ LAI (R = 0.41) and CC (R = 0.33).

Contributions

• This study is the first to develop a machine learning model to predict UAV-acquired Normalised Difference Water Index (NDWI) from multispectral UAV bands by leveraging relationships with various Structural Vegetation Indices (SVIs).
• It introduces a novel machine learning framework that enhances the utility of multispectral UAV imagery for water stress assessment in smallholder farming systems by integrating the spatial precision of UAV data with the spectral richness of satellite observations.
• The findings demonstrate the potential of ML-driven NDWI estimation to support site-specific irrigation scheduling, improve resource use efficiency, and promote sustainable sugarcane cultivation.
• The research contributes to the development of climate-resilient water management practices specifically tailored to the needs of smallholder agricultural systems in water-scarce regions.

Funding

• Water Research Commission (WRC)
• WRC Project C2021–2022–00800: "Leveraging Google Earth Engine to Analyse Very-High Spatial Resolution Unmanned Aerial Vehicle Data to Guide and Inform Precision Agriculture in Smallholder Farms."

Citation

@article{Yacoob2026machine,
  author = {Yacoob, Ameera and Gokool, Shaeden and Clulow, Alistair and Mahomed, Maqsooda and Naiken, Vivek and Mabhaudhi, Tafadzwanashe},
  title = {A machine learning approach for quantifying crop water stress in smallholder farms using unmanned aerial vehicle multispectral imagery},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110142},
  url = {https://doi.org/10.1016/j.agwat.2026.110142}
}