Guarda (2026) Limits to the Spatial Transferability of Machine-Learning Models for Farm-Record-Level Maize Yield Prediction in Manabí, Ecuador

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

Identification

• Journal: Agriculture
• Year: 2026
• Date: 2026-09-07
• Authors: Teresa Guarda
• DOI: 10.3390/agriculture16171932

Research Groups

• Department of Agricultural Sciences, University of Ecuador
• Institute of Climate Change, National University of Manabí

Short Summary

This study evaluates the performance of supervised machine-learning models in predicting farm-level hard dry maize yield in previously unseen parishes of Manabí, Ecuador. The results show that while climate-enhanced models slightly outperform survey-only models, all configurations produce negative pooled R2 values and are insufficient for reliable yield prediction.

Objective

• Investigate the ability of supervised machine-learning models to predict farm-level hard dry maize yield in previously unseen parishes of Manabí, Ecuador

Study Configuration

• Spatial Scale: Parish level (45 parishes) in Manabí, Ecuador
• Temporal Scale: Crop-season scale (single growing season)

Methodology and Data

• Models used:
  • Elastic Net
  • Random Forest
  • Histogram-based gradient boosting
• Data sources:
  • Survey-derived farm, crop, producer, management, socioeconomic, and digital-access characteristics
  • Climate-enhanced models included crop-season cumulative precipitation, longest dry spell, mean 2 m air temperature, and mean upper-layer soil moisture

Main Results

• Mean absolute error (MAE) values for the best survey-only and survey-plus-climate models: 1.535 t ha−1 and 1.548 t ha−1, respectively
• Estimated climate-related MAE reduction for histogram-based gradient boosting: −0.013 t ha−1 (−0.85%)
• Pooled R2 values were negative for all model configurations

Contributions

This study highlights the limitations of current survey data and parish-level climate variables in supporting reliable yield prediction for individual crop records in previously unseen parishes.

Funding

• This research was funded by the Ecuadorian Ministry of Agriculture, Livestock, Aquaculture, and Fisheries (MAGAP) under project code MAGAP-2020-001.

Citation

@article{Guarda2026Limits,
  author = {Guarda, Teresa},
  title = {Limits to the Spatial Transferability of Machine-Learning Models for Farm-Record-Level Maize Yield Prediction in Manabí, Ecuador},
  journal = {Agriculture},
  year = {2026},
  doi = {10.3390/agriculture16171932},
  url = {https://doi.org/10.3390/agriculture16171932}
}