Dangare et al. (2026) Modelling water - use and yield of selected irrigated subtropical crops using machine learning and hybrid models in north - eastern South Africa

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-01-06
• Authors: Prince Dangare, Paul JR Cronje, Zama Eric Mashimbye, J. Masanganise, Zanele Ntshidi, Shaeden Gokool, Vivek Naiken, Tendai Sawunyama, Sebinasi Dzikiti
• DOI: 10.1016/j.agwat.2025.110113

Research Groups

• Department of Horticultural Science, Stellenbosch University, Stellenbosch, South Africa
• Citrus Research International, Mbombela, South Africa
• Department of Geography and Environmental Studies, Stellenbosch University, Stellenbosch, South Africa
• Inkomati – Usuthu Catchment Management Agency, South Africa
• Arid Lands Node, South African Environmental Observation Network (SAEON), Kimberley, South Africa
• Department of Electronics and Telecommunications, University of Zimbabwe, Harare, Zimbabwe
• Department of Physical and Earth Sciences, Sol Plaatje University, Kimberley, South Africa
• Department of Physics and Engineering, Bindura University of Science Education, Bindura, Zimbabwe
• Centre for Water Resources Research, School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, Pietermaritzburg, South Africa
• South African Environmental Observation Network (SAEON), Grasslands-Forests-Wetlands Node, Pietermaritzburg, South Africa
• Department of Agrometeorology, Schools of Agriculture and Science, University of KwaZulu-Natal, Pietermaritzburg, South Africa

Short Summary

This study developed and validated machine learning and hybrid models to accurately estimate evapotranspiration (ET), transpiration (T), crop coefficients (Kc), and yield response factors (Ky) for five irrigated subtropical crops in north-eastern South Africa. The findings provide crucial, locally derived water-use parameters to optimize irrigation management and improve water productivity in water-scarce regions.

Objective

• To model evapotranspiration (ET) for banana, grapefruit, litchi, mango, and sugarcane crops using Light Gradient Boosting Machine (LightGBM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) models to gap-fill eddy covariance ET measurements.
• To use the best performing ET models to derive highly accurate crop coefficients (Kc).
• To model transpiration (T) for grapefruit, litchi, and mango crops using LightGBM, RF, and XGBoost models, and select the best performing T model for yield modeling.
• To formulate a hybrid water-yield model for deriving grapefruit, litchi, and mango yield response factors (Ky) by combining empirical water-yield relations with the best machine learning T model.

Study Configuration

• Spatial Scale: Two farms (Riverside farm, Malelane; Welgelegen farm, Komatipoort) in Mpumalanga province, South Africa, approximately 50 km apart. Crops studied include banana, grapefruit, litchi, mango, and sugarcane.
• Temporal Scale: Microclimate data from September 2021 to September 2023 (with historical data from 2014-2021). Leaf Area Index (LAI) derived from January 2015 to December 2024. ET and T measurements conducted between March 2022 and February 2024, varying by crop.

Methodology and Data

• Models used:
  • Machine Learning (ML) models: Light Gradient Boosting Machine (LightGBM), Random Forest (RF), Extreme Gradient Boosting (XGBoost) for ET and T prediction.
  • Hybrid model: Empirical water-yield relation combined with the best performing ML T model for Ky derivation.
  • FAO-56 reference evapotranspiration (ETo) for Kc calculation.
• Data sources:
  • Microclimate: Automatic weather stations (solar radiation, air temperature, relative humidity, rainfall, wind speed and direction) at 2 meters height. Historical data from the Agricultural Research Council, South Africa.
  • Leaf Area Index (LAI): Derived from Landsat 8 Tier 1 surface reflectance data (Normalized Difference Vegetation Index - NDVI) using Google Earth Engine, interpolated to hourly data.
  • Transpiration (T): Heat Ratio Method (HRM) sap flow sensors (grapefruit, litchi, mango).
  • Evapotranspiration (ET): Eddy Covariance (EC) flux tower (grapefruit, litchi, mango) and Surface Renewal (SR) method (banana, sugarcane).
  • Yield data: 10-year historic yield data from different orchards for seasonal maximum yield (Ym) and seasonal maximum transpiration (Tc), and measured yield data for validation.

Main Results

• ET and T Model Performance: All models achieved high accuracy with R² values ranging from 0.83 to 0.96, RMSE from 0.02 to 0.10 mm/h, MAE from 0.01 to 0.06 mm/h, and KGE from 0.88 to 0.97.
  • LightGBM showed the highest accuracy for banana, grapefruit, litchi, and sugarcane ET, and for grapefruit, litchi, and mango T.
  • XGBoost showed the highest accuracy for mango ET.
• Feature Importance: Solar radiation (Rs) was identified as the most influential input feature for both ET and T models across all crops.
• Mean Annual Evapotranspiration (ET):
  • Banana: 1524 mm
  • Grapefruit: 824 mm
  • Litchi: 639 mm
  • Mango: 960 mm
  • Sugarcane: 1137 mm
• Derived Monthly Crop Coefficients (Kc):
  • Banana: 0.92–1.30
  • Grapefruit: 0.53–0.75
  • Litchi: 0.40–0.57
  • Mango: 0.63–0.84
  • Sugarcane: 0.62–1.16
• Derived Yield Response Factors (Ky):
  • Grapefruit: 2.70
  • Litchi: 2.50
  • Mango: 2.90
  • These Ky values indicate a high sensitivity of these crops to water deficits.
• Yield Model Validation: The hybrid yield models performed well, with R² values of 0.81 (grapefruit), 0.87 (litchi), and 0.83 (mango). RMSE values were 4.49 tonne/ha (grapefruit), 0.78 tonne/ha (litchi), and 2.79 tonne/ha (mango).

Contributions

• This study presents the first comparison of gradient boosting frameworks (LightGBM, Random Forest, XGBoost) for modeling evapotranspiration (ET) and transpiration (T) of economically important subtropical crops in South Africa.
• It introduces the first development of a hybrid model to estimate yield response factors (Ky) for subtropical crops in South Africa.
• The research provides accurate, locally derived crop coefficients (Kc) and yield response factors (Ky) for banana, grapefruit, litchi, mango, and sugarcane, which are crucial for improving irrigation planning and promoting productive water use in water-scarce regions.

Funding

• Water Research Commission (WRC) - Project WRC C2020/2023–00399
• Inkomati-Usuthu Catchment Management Agency (CMA)

Citation

@article{Dangare2026Modelling,
  author = {Dangare, Prince and Cronje, Paul JR and Mashimbye, Zama Eric and Masanganise, J. and Ntshidi, Zanele and Gokool, Shaeden and Naiken, Vivek and Sawunyama, Tendai and Dzikiti, Sebinasi},
  title = {Modelling water - use and yield of selected irrigated subtropical crops using machine learning and hybrid models in north - eastern South Africa},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2025.110113},
  url = {https://doi.org/10.1016/j.agwat.2025.110113}
}