Detecting environmental trends to rethink soybean variety testing programs

Abstract

Variety testing programs (VTPs) use multi-environment trials (MET) to evaluate and report the performance of commercially available and pre-commercial soybean (Glycine max L. Merr.) varieties targeting a specific set of environments. Adequate modeling of the environmental variability and genotype–environment interactions (G × E) within the VTP would help farmers and seed companies decide which variety to choose or recommend. We propose an approach to characterize environments using the soybean data from the University of Missouri VTP. We modeled an environmental trend (EnvT) based on the phenotypic mean performance and the observed phenotype in each environment. The environments were classified into four different EnvT environment types, and soil and climate data were used as predictors of the EnvT through eXtreme Gradient Boosting (XGBoost) model. Temperature on late vegetative and flowering, soil-saturated hydraulic conductivity, and silt content were key drivers of EnvT. The approach identified overrepresented environments (62%) and increased the ratio between variety and G × E variance. A simulation case study verified that the random removal of overrepresented sites from the dataset quickly degraded G × E analysis, implying that increasing the number of underrepresented sites is recommended. Our results demonstrate that environmental characterization is essential for optimizing resource allocation within VTP, thereby supporting the end goal of aiding farmers to utilize the best varieties for their production environment.