Site-Specific Nutrient Diagnosis of Orange Groves
Nenhuma Miniatura disponível
Data
2022-12-01
Orientador
Coorientador
Pós-graduação
Curso de graduação
Título da Revista
ISSN da Revista
Título de Volume
Editor
Tipo
Artigo
Direito de acesso
Resumo
Nutrient diagnosis of orange (Citrus sinensis) groves in Brazil relies on regional information from a limited number of studies transferred to other environments under the ceteris paribus assumption. Interpretation methods are based on crude nutrient compositions that are intrinsically biased by genetics X environment interactions. Our objective was to develop accurate and unbiased nutrient diagnosis of orange groves combining machine learning (ML) and compositional methods. Fruit yield and foliar nutrients were quantified in 551 rainfed 7–15-year-old orange groves of ‘Hamlin’, ‘Valência’, and ‘Pêra’ in the state of São Paulo, Brazil. The data set was further documented using soil classification, soil tests, and meteorological indices. Tissue compositions were log-ratio transformed to account for nutrient interactions. Ionomes differed among scions. Regression ML models showed evidence of overfitting. Binary ML classification models showed acceptable values of areas under the curve (>0.7). Regional standards delineating the multivariate elliptical hyperspace depended on the yield cutoff. A shapeless blob hyperspace was delineated using the k-nearest successful neighbors that showed comparable features and reported realistic yield goals. Regionally derived and site-specific reference compositions may lead to differential interpretation. Large-size and diversified data sets must be collected to inform ML models along the learning curve, tackle model overfitting, and evaluate the merit of blob-scale diagnosis.
Descrição
Palavras-chave
Idioma
Inglês
Como citar
Horticulturae, v. 8, n. 12, 2022.