Supplementary material
Conditioning Machine Learning Models to Adjust Lowbush Blueberry Crop Management to the Local Agroecosystem
by Serge-Étienne Parent, Jean Lafond, Maxime C. Paré, Léon Etienne Parent and Noura Ziadi
Plants 2020, 9(10), 1401; https://doi.org/10.3390/plants9101401
Agroecosystem conditions limit the productivity of lowbush blueberry. Our objectives were to investigate the effects on berry yield of agroecosystem and crop management variables, then to develop a recommendation system to adjust nutrient and soil management of lowbush blueberry to given local meteorological conditions. We collected 1504 observations from N-P-K fertilizer trials conducted in Quebec, Canada. The data set, that comprised soil, tissue, and meteorological data, was processed by Bayesian mixed models, machine learning, compositional data analysis, and Markov chains. Our investigative statistical models showed that meteorological indices had the greatest impact on yield. High mean temperature at flower bud opening and after fruit maturation, and total precipitation at flowering stage showed positive effects. Low mean temperature and low total precipitation before bud opening, at flowering, and by fruit maturity, as well as number of freezing days (<−5 °C) before flower bud opening, showed negative effects. Soil and tissue tests, and N-P-K fertilization showed smaller effects. Gaussian processes predicted yields from historical weather data, soil test, fertilizer dosage, and tissue test with a root-mean-square-error of 1447 kg ha−1. An in-house Markov chain algorithm optimized yields modelled by Gaussian processes from tissue test, soil test, and fertilizer dosage as conditioned to specified historical meteorological features, potentially increasing yield by a median factor of 1.5. Machine learning, compositional data analysis, and Markov chains allowed customizing nutrient management of lowbush blueberry at local scale.
The ellipse represents a model conditioned to local variables (e.g. soil texture, weather, etc.). The initial position of the squirrel represents a field observation. The path followed by the squirrel is a Markov chain sequentially improving the outcome (e.g. yield) by scanning then modifying managable variables (e.g. soil nutrients, plant nutrients, fertilizers, etc.) until the optimal condition is reach for given field conditions.
