DUGi

Compositional meta-analysis of Citrus varieties in the state of São Paulo, Brazil

http://dugi.udg.edu/item/http:@@@@hdl.handle.net@@10256@@13659

Brazil is the largest orange (Citrus sinensis) producer worldwide. The nutrient management of orange orchards is designed from experiments on a limited number of varieties. This knowledge is transferred to other varieties by diagnosing tissue nutrient composition. Nutrient diagnostic tools are based on nutrient concentration (critical minimum value or CMV) and ratio (Diagnosis and Recommendation Integrated System or DRIS) norms that disregard the compositional nature of analytical data and the limited number of nutrient ratios that can be diagnosed independently in a given composition. The diagnosis of cationic micronutrients is also biased by contamination from fungicides. Compositional data analysis that can avoid such problems has been first applied to tissue analysis of agricultural crops using centered log ratios (Compositional Nutrient Diagnosis – CND-clr). The isometric log ratio (ilr) transformation is a new approach based on binary nutrient ratios and the principle of orthogonality (CND-ilr). Binary partitions can be defined and varietal nutrient profiles classified based on positive and negative nutrient interactions and meta-analysis. We analyzed 11 nutrients (N, S, P, K, Ca, Mg, B, Cu, Zn, Mn, Fe) in tissue samples across 108 orchard areas, i.e. 31 ‘Valencia’, 22 ‘Hamlin’, 20 ‘Pêra’, and 35 ‘Natal’. Nutrients were partitioned between macro- and micro-nutrients as well as anionic and cationic species. The effect size of varieties over ‘Valencia’ was quantified by the mean and standard deviation of ilr values across ilr coordinates. Specific varietal nutrient profiles and ilr norms were defined. To guide correcting nutrient deficiencies by appropriate nutrient management, compositions can be varied by a perturbation vector on nutrients with to the largest and most negative influence on ilr differences from ilr norms until the Aitchison distance falls below critical value

Universitat de Girona. Departament d’Informàtica i Matemàtica Aplicada