Conditioning Machine Learning Models to Adjust Lowbush Blueberry Crop Management to the Local Agroecosystem

Area of Science:

• Agricultural Science
• Agronomy
• Horticulture

Background:

• Lowbush blueberry (Vaccinium angustifolium) productivity is constrained by agroecosystem conditions.
• Optimizing crop management requires understanding environmental influences on yield.

Purpose of the Study:

• Investigate agroecosystem and crop management effects on lowbush blueberry yield.
• Develop a recommendation system for adjusting nutrient and soil management based on local meteorological conditions.

Main Methods:

• Utilized 1504 observations from N-P-K fertilizer trials in Quebec, Canada.
• Applied Bayesian mixed models, machine learning, compositional data analysis, and Markov chains.
• Employed Gaussian processes for yield prediction and a Markov chain algorithm for optimization.

Main Results:

• Meteorological indices, including temperature and precipitation, were the primary drivers of yield.
• Specific temperature and precipitation patterns at different growth stages significantly affected berry yield.
• Soil and tissue tests, along with N-P-K fertilization, had lesser impacts on yield.

Conclusions:

• Meteorological data is crucial for predicting and optimizing lowbush blueberry yields.
• A customized nutrient and soil management system, informed by weather, can enhance berry production.
• Advanced statistical and machine learning techniques enable localized agricultural management strategies.