Adaptabilidade e estabilidade de pré-cultivares de feijão carioca no Agreste-Sertão pernambucano

Abstract

The common bean is an important source of protein in human food and show relevant socioeconomic value in Brazil, which has a preference for the commercial carioca type. In the Agreste-Sertão pernambucano, its cultivation is carried out in several municipalities, with different edaphoclimatic conditions, which influence productivity. To minimize the effects of interaction, breeders seek to identify adaptable and stable genotypes, and the selection of the adaptability and stability assessment methodology must be done in the most appropriate way to ensure effective data analysis. The objective of this work was to compare the methodologies of Eberhart & Russel, Linn & Binns modified by Carneiro, and Additive Main Effects and Multiplicative Interaction Analysis (AMMI), identifying the most efficient in the simultaneous selection of productive, adapted and stable carioca bean pre-cultivars and then to confront the frequentist and bayesian versions of AMMI analysis, to assess the predictive power. Ten pre-cultivars and four commercial were used, with a randomized block design and three replications. Grain yield was evaluated in the years of 2014 and 2015. Initially, adaptability and stability were estimated using the techniques of Eberhart & Russell, Lin & Binns modified by Carneiro and AMMI, which were then compared using the Spearman correlation. Subsequently, random imbalances were performed on the data (10% and 20% loss) and analyzes were performed with the classic AMMI and the Bayesian AMMI (BAMMI), using the EM (expectation-maximization) algorithm to impute the missing data in the classic analysis. Finally, to assess the predictive power of the proposed models, cross-validation was performed using the correlation between predicted and observed values (Cor), Spearman's Correlation (CorS) and PRESS (Prediction Error Sum Square). No correlation was observed between Eberhart & Russell and Lin & Binns. The AMMI is the most complete frequentist method for isolated use, however, BAMMI showed a better predictive capacity, as well as better performance in the study of adaptability and stability. The BAMMI shows flexibility to deal with data resulting from multi-environmental experiments, overcoming limitations of the standard analysis of the AMMI model.