Bourbour et al. (2025) Pre-harvest forecasting of rainfed wheat yield in Iran using multi-source remote sensing and machine learning

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-11-21
• Authors: Hanie Bourbour, Mohammad Abdolahipour, Hamid Abdollahi, Ershad Abiar, Mahmoud Mashal
• DOI: 10.1016/j.agwat.2025.110005

Research Groups

• Department of Water Engineering, Faculty of Agricultural Technology, College of Agriculture and Natural Resources, University of Tehran, Tehran, Iran
• Sharif University of Technology, Tehran, Iran

Short Summary

This study developed and compared machine learning models integrating multi-source remote sensing and meteorological data to forecast rainfed wheat yield across 22 Iranian provinces from 2001 to 2021. The XGBoost algorithm achieved superior accuracy (R²=0.64, MAE=0.25 t/ha) two months pre-harvest, outperforming Random Forest and Support Vector Regression.

Objective

• To develop and compare machine learning models integrating multi-source remote sensing and meteorological data for forecasting rainfed wheat yield across 22 Iranian provinces in Iran.
• To evaluate the performance of different machine learning algorithms (Random Forest, XGBoost, Support Vector Regression) in accurately predicting wheat yield and to analyze the impact of environmental factors on yield under rainfed conditions.

Study Configuration

• Spatial Scale: 22 Iranian provinces, covering approximately 1,648,195 square kilometers.
• Temporal Scale: 2001–2021 (21 years) for data collection; yield forecasting performed 2, 3, and 4 months pre-harvest.

Methodology and Data

• Models used: Random Forest (RF), XGBoost, Support Vector Regression (SVR).
• Data sources:
  • Satellite/Remote Sensing:
    • MODIS (MOD13Q1): Normalized Difference Vegetation Index (NDVI) (16-day, 250 m)
    • MODIS (MOD11A2): Land Surface Temperature (LST) (8-day, 1 km)
    • MODIS (MOD15A2): Leaf Area Index (LAI) (8-day, 500 m)
    • MODIS (MOD15A2): Fraction of Photosynthetically Active Radiation (FPAR) (8-day, 500 m)
    • MODIS (MOD16A2): Evapotranspiration (ET) (8-day, 500 m)
  • Reanalysis/Gridded Data:
    • TerraClimate: Palmer Drought Severity Index (PDSI) (monthly, 4638.3 m)
    • ERA5: Air temperature (monthly, 11132 m), Precipitation (monthly, 11132 m)
    • NASA/FLDAS: Soil Moisture at 0–10 cm and 10–40 cm depths (monthly, 11132 m)
  • Ground-based/Official Statistics:
    • Ministry of Jihad Agriculture: Wheat yield (tonnes per hectare), cultivated area, length of growing season, plant growth stages, land area maps.

Main Results

• The XGBoost algorithm achieved the best performance with 80% training data (352 samples) and 20% test data, yielding an R² of 0.64, a mean-normalized NRMSE of 42%, and a Mean Absolute Error (MAE) of 0.25 tonnes per hectare.
• XGBoost consistently outperformed Random Forest (R²=0.60, NRMSE=43%, MAE=0.26 t/ha) and Support Vector Regression (R²=0.55, NRMSE=46%, MAE=0.26 t/ha).
• Optimal prediction accuracy was achieved when forecasting two months prior to harvest (April).
• Key predictors influencing rainfed wheat yield were identified as: NDVI (April, November, December), precipitation (February, March), LST (April, December), and LAI (April).
• The optimal combination of input variables included air temperature, precipitation, NDVI, LAI, soil moisture at 0–10 cm depth, and LST.
• All models showed a tendency to slightly overestimate low yields and underestimate high yields.

Contributions

• First large-scale evaluation of rainfed wheat yield prediction across multiple provinces in Iran, addressing a significant regional gap in the literature.
• Integration of a diverse array of multi-source remote sensing indices (MODIS-derived NDVI, LAI, LST, FPAR, ET), meteorological datasets (ERA5 precipitation, temperature, TerraClimate PDSI), and soil moisture (FLDAS) with advanced machine learning algorithms (XGBoost, RF, SVR).
• Demonstrated the superior performance of XGBoost for operational pre-harvest yield forecasting in arid and semi-arid rainfed agricultural systems.
• Identified critical temporal windows and specific environmental variables (NDVI, LAI, precipitation, LST, soil moisture) that are most influential for accurate pre-harvest yield prediction in the region.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Citation

@article{Bourbour2025Preharvest,
  author = {Bourbour, Hanie and Abdolahipour, Mohammad and Abdollahi, Hamid and Abiar, Ershad and Mashal, Mahmoud},
  title = {Pre-harvest forecasting of rainfed wheat yield in Iran using multi-source remote sensing and machine learning},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.110005},
  url = {https://doi.org/10.1016/j.agwat.2025.110005}
}

Generated by BiblioAssistant using gemini-2.5-flash (Google API)