Can knowledge of the spatial field trend improve phenotypic prediction accuracy through informed choice of experimental design?

Abstract Environmental variation is an important factor in crop yield variability, especially in large agricultural experiments such as breeding trials. This study explores whether prior knowledge of the spatial field trends can enhance phenotypic prediction accuracy by choice of experimental design and statistical analysis. Using data from three breeding trials and a simulation study with eight field trends, we compared five experimental designs and six statistical models. The results showed that combining field trend knowledge with an appropriate analysis strategy can improve predictive performance. There was no universally optimal design; but among the designs considered, the row-column design performed best for most field trends with a high degree of spatial variation. Statistical models correcting for spatial variation consistently reduced bias and increased correlation between predicted and true yield. This work emphasizes the need to integrate spatial information in the decision of both the design and analysis of breeding trials to boost prediction accuracy.