Quintanilla-Albornoz et al. (2025) Almond yield prediction at orchard scale using satellite-derived biophysical traits and crop evapotranspiration combined with machine learning

Identification

• Journal: Frontiers in Agronomy
• Year: 2025
• Date: 2025-11-20
• Authors: Manuel Quintanilla-Albornoz, Xavier Miarnau, Magí Pàmies-Sans, Joaquim Bellvert
• DOI: 10.3389/fagro.2025.1667674

Research Groups

• IRTA (Institute of Agrifood Research and Technology), Program for the Efficient Use of Water in Agriculture, Fruitcentre, Lleida, Spain.
• IRTA, Fruit Production Program, Fruitcentre, Lleida, Spain.
• University of California-Davis, Department of Viticulture and Enology, Davis, CA, United States.

Short Summary

This study develops a machine learning framework to predict almond yield at the orchard scale across Spain using Sentinel-2 biophysical traits and Sentinel-3 derived evapotranspiration. The results demonstrate that remote sensing-based models achieve predictive accuracy comparable to ground-truth data, with the best model reaching a Root Mean Square Error (RMSE) of 399.1 kg ha⁻¹.

Objective

• To develop and evaluate machine learning-based almond yield prediction models at the orchard scale by integrating actual evapotranspiration (ETa) estimates, biophysical traits, and meteorological data.

Study Configuration

• Spatial Scale: Orchard scale across 64 commercial sites in Spain, primarily in the Ebro and Guadalquivir basins, as well as Albacete and Badajoz.
• Temporal Scale: 2017–2022 (6 growing seasons).

Methodology and Data

• Models used: Two-Source Energy Balance model with the Priestley-Taylor approach (TSEB-PT) for ETa estimation; Machine Learning algorithms including Random Forest (RF), Stochastic Gradient Boosting (SGB), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM).
• Data sources:
  • Satellite: Sentinel-2 (spectral indices and biophysical traits: fAPAR, LAI, FVC, Cab, CWC) and Sentinel-3 (Land Surface Temperature).
  • Reanalysis: ERA5-Land (meteorological parameters: Ta, Rs, VPD, ETo) and MSWEP (precipitation).
  • Ground-truth: Commercial grower records for kernel yield, irrigation volumes, orchard age, tree density, and cultivar.

Main Results

• Model Performance: LightGBM was the superior algorithm for remote sensing-based predictions. The combined remote sensing model (PMCRS) achieved an RMSE of 399.1 kg ha⁻¹ and an R² of 0.66 when using data up to July.
• Key Predictors: The most influential variables for yield prediction were the fraction of absorbed photosynthetically active radiation (fAPAR), normalized difference moisture index (NDMI), canopy chlorophyll content (Cab), and cumulative ETa.
• Historical Influence: Including data from the previous season (e.g., fAPAR-1Y, ETa-1Y) significantly enhanced model accuracy.
• Water Production Functions: The study established orchard-scale water production functions consistent with experimental data, identifying a maximum potential kernel yield of 2776 kg ha⁻¹ at approximately 1000 mm of seasonal water (irrigation + effective precipitation).
• Contextual Enhancement: Adding ground-truth data (irrigation and orchard age) to satellite models reduced RMSE by up to 10.25%.

Contributions

• Orchard-Scale Validation: First study to establish almond water production functions at the commercial orchard scale using remote sensing, bridging the gap between experimental plots and regional management.
• Methodological Integration: Demonstrates the effectiveness of combining biophysical traits (via the SNAP biophysical processor) with thermal-based ETa estimates for woody crop yield forecasting.
• Scalability: Provides a validated framework for regional yield estimation that relies on open-access satellite data, reducing dependency on difficult-to-obtain farmer records.

Funding

• DIGISPAC project (TED2021-131237B-C21), Ministry of Science, Innovation and Universities of the Spanish Government.
• Internal IRTA scholarship.
• CERCA Program, Government of Catalonia.
• European Commission Horizon 2020 Research and Innovation Program, Marie Sklodowska-Curie RISE action, ACCWA project (grant agreement No. 823965).

Citation

@article{QuintanillaAlbornoz2025Almond,
  author = {Quintanilla-Albornoz, Manuel and Miarnau, Xavier and Pàmies-Sans, Magí and Bellvert, Joaquim},
  title = {Almond yield prediction at orchard scale using satellite-derived biophysical traits and crop evapotranspiration combined with machine learning},
  journal = {Frontiers in Agronomy},
  year = {2025},
  doi = {10.3389/fagro.2025.1667674},
  url = {https://doi.org/10.3389/fagro.2025.1667674}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)