Enhancing Genomic Prediction Models for Forecasting Days to Maturity in Soybean Genotypes Using Site-Specific and Cumulative Photoperiod Data

Abstract

Genomic selection (GS) has revolutionized breeding strategies by predicting the rank performance of post-harvest traits via implementing genomic prediction (GP) models. However, predicting pre-harvest traits in unobserved environments might produce serious biases. In soybean, days to maturity (DTM) represents a crucial stage with a significant impact on yield potential; thus, genotypes must be carefully selected to ensure latitudinal adaptation in this photoperiod-sensitive crop species. This research assessed the use of daylength for predicting DTM in unobserved environments (CV00). A soybean dataset comprising 367 genotypes spanning nine families of the Soybean Nested Association Mapping Panel (SoyNAM) and tested in 11 environments (year-by-location combinations) was considered in this study. The proposed method (CB) returned a root-mean-square error (RMSE) of 5.2 days, a Pearson correlation (PC) of 0.66, and the predicted vs. observed difference in the environmental means (PODEM) ranged from −3.3 to 4.5 days; however, in the absence of daylength data, the conventional GP implementation produced an RMSE of 9 days, a PC of 0.66, and a PODEM range from −14.7 to 7.9 days. These results highlight the importance of dissecting phenotypic variability (G × E) based on photoperiod data and non-predictable environmental stimuli for improving the predictive ability and accuracy of DTM in soybeans.