Simeón et al. (2026) Rice Yield Prediction Model at Pixel Level Using Machine Learning and Multi-Temporal Sentinel-2 Data in Valencia, Spain

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Agriculture
• Year: 2026
• Date: 2026-01-13
• Authors: Rubén Simeón, Alba Agenjos-Moreno, C. C. Rubio, Antonio Uris, Alberto San Bautista
• DOI: 10.3390/agriculture16020201

Research Groups

Researchers associated with the Albufera rice area, Valencia, Spain.

Short Summary

This study assessed the potential of multitemporal Sentinel-2 imagery and machine learning to estimate rice yield at pixel level, finding that XGBoost models using all spectral bands accurately predicted within-field yield variability for two Japonica varieties.

Objective

• To assess the potential of multitemporal Sentinel-2 imagery and machine learning to estimate rice yield at pixel level, characterizing within-field variability, in the Albufera rice area.

Study Configuration

• Spatial Scale: Pixel-level (10 m resolution) within rice fields, covering the Albufera rice area.
• Temporal Scale: Five growing seasons (2020–2024), with predictions made at 110 and 130 days after showing (DAS).

Methodology and Data

• Models used: Random Forest (RF), XGBoost (XGB).
• Data sources: Multitemporal Sentinel-2 imagery (visible, near-infrared, and short-wave infrared bands) at 10 m resolution; combine harvester yield maps for ‘JSendra’ and ‘Bomba’ Japonica rice varieties.

Main Results

• XGBoost systematically outperformed Random Forest models.
• At 110 DAS, XGBoost achieved an R² of 0.74 and a root mean square error (RMSE) of 0.63 t·ha⁻¹ for ‘JSendra’.
• At 130 DAS, XGBoost achieved an R² of 0.85 and an RMSE of 0.28 t·ha⁻¹ for ‘Bomba’.
• Prediction accuracy increased as the season progressed.
• Models utilizing all spectral bands significantly outperformed configurations based solely on spectral indices, with near-infrared (NIR) reflectance identified as the dominant spectral contribution.
• Spatial error analysis revealed higher errors at field edges and headlands, while central pixels were predicted more accurately.

Contributions

• Provides accurate, spatially explicit rice yield maps at pixel level, effectively capturing within-field variability.
• Supports both end-of-season yield estimation and early season forecasting.
• Enables the identification of potentially low-yield zones to facilitate targeted management decisions.
• Demonstrates the superior performance of XGBoost and the importance of using all spectral bands (especially NIR) over just spectral indices for pixel-level rice yield prediction.

Funding

No funding information was provided in the text.

Citation

@article{Simeón2026Rice,
  author = {Simeón, Rubén and Agenjos-Moreno, Alba and Rubio, C. C. and Uris, Antonio and Bautista, Alberto San},
  title = {Rice Yield Prediction Model at Pixel Level Using Machine Learning and Multi-Temporal Sentinel-2 Data in Valencia, Spain},
  journal = {Agriculture},
  year = {2026},
  doi = {10.3390/agriculture16020201},
  url = {https://doi.org/10.3390/agriculture16020201}
}