Local Factors Impact Accuracy of Garlic Tissue Test Diagnosis

• Autor(a) principal: Hahn, Leandro
• Data de Publicação: 2022
• Outros Autores: Parent, Léon-Étienne, Feltrim, Anderson Luiz, Rozane, Danilo Eduardo [UNESP], Ender, Marcos Matos, Tassinari, Adriele, Krug, Amanda Veridiana, Berghetti, Álvaro Luís Pasquetti, Brunetto, Gustavo
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.3390/agronomy12112714; http://hdl.handle.net/11449/246299
• topic: Adaboost; Allium sativum; compositional distance; growth-limiting factors; machine learning; perturbation vector; random forest

The low productivity of garlic in Brazil requires more efficient nutritional management. For this, environmental and fertilization-related factors must be adjusted to a set of local conditions. Our objective was to provide an accurate diagnosis of the nutrient status of garlic crops in southern Brazil. The dataset comprised 1024 observations, 962 as field tests conducted during the 2015–2017 period to train the model, and 61 field observations collected during the 2018–2019 period to validate the model. Machine learning models (MLM) related garlic yield to managerial, edaphic, plant, and climatic features. Compositional data analysis (CoDa) methods allowed classification of nutrients in the order of limitation to yield where MLM detected nutrient imbalance. Tissue analysis alone returned an accuracy of 0.750 in regression and 0.891 in classification about the yield cutoff of 11 ton ha−1. Adding all features documented in the dataset, accuracy reached 0.855 in regression and 0.912 in classification. Local diagnosis based on MLM and CoDa and accounting for local features differed from regional diagnosis across features. Local nutrient diagnosis may differ from regional diagnosis because several yield-impacting factors are taken into account and benchmark compositions are representative of local conditions.