Correcting experimental data for spatial trends in a common bean breeding program

Abstract

In common bean (Phaseolus vulgaris L.) breeding, several trials are carried out in field conditions to predict the genotypic values, but experimental designs may not be sufficient to capture the field heterogeneity in the experimental area. The objective of this work was to evaluate the potential of spatial models to correct data from a common bean breeding program for spatial trends and improve the prediction of genotypic values. We used real data from 19 field trials from a common bean breeding program and three experimental designs. The traditional statistical model with design effects and independent errors was fitted and used as the basic model. Later, we fitted a sequence of spatial models to include different residual (co)variance structures for local trends and fixed and random effects based on plot position information to capture global and extraneous trends. The basic model and the best-fit spatial model were compared regarding the estimates of heritability, accuracy, prediction error variance, and discordance in the top-ranking genotypes. In most cases, the use of spatial models improved the estimates of heritability and accuracy or, at least, reduced the estimates of prediction error variance. Also, changes in the genotypic values classification were observed. Because no single model presented the best fit for all trials, some of the tested models were recommended for future trials based on the patterns of spatial trends observed. Thus, the use of spatial models helped to improve the data analysis and the prediction of genotypic values by capturing the field heterogeneity in our common bean field trials.