Logo image
Back
Modeling the tracking and prediction of grape phenology using machine learning
Thesis   Open access

Modeling the tracking and prediction of grape phenology using machine learning

Nathan Balcarcel
Master of Science (MS), Washington State University
12/2025
DOI:
https://doi.org/10.7273/000008293
pdf
Balcarcel, Nathan Thesis3.29 MBDownloadView
Open Access CC BY V4.0

Abstract

Growing degree days Phenological development Crop phenology Air temperature
Accurate forecasting of crop phenology is crucial for time-sensitive farm management decisions and for implementing mitigation strategies to protect crops during suboptimal conditions. In grapevines, phenological development involves complex interactions between environmental factors and cultivar-specific physiology, making prediction inherently challenging. Traditional process-based models rely primarily on growing degree days (GDD) derived from air temperature alone. For each cultivar, these models require independent parameters, typically derived via regression. While fairly straightforward, this approach makes assumptions about the simplicity of crop phenology, overlooking other influential factors that may affect the exact timings of the phenology cycle. Our work distinguishes itself in three main ways: i) we leverage expanded weather data inputs (i.e., air temperature, relative humidity, dew point, precipitation, and wind speed); ii) we replace the traditional process-based approach with a gated recurrent unit (GRU); and iii) training and prediction for different cultivars are handled by the same model. Using a 34-year dataset spanning 20 grape cultivars, our model outperforms GDD-based baselines in predicting budbreak, bloom, and veraison growth stages for grapevines. Furthermore, a post-processing step is introduced to generate adaptive confidence intervals for stage forecasts, offering users a quantifiable measure of prediction uncertainty.

Metrics

2 File views/ downloads
7 Record Views

Details

Logo image