Comprehensive assessment of soybean seed composition from field trials spanning 22 US states and 24 years: Predictive insights

Abstract

Soybean seed composition has long been a subject of study, not only due to its importance to the oil market, but also because variations in seed protein and oil content impact meal quality. Previous analyses of historical data (1948–1998) examined trends in seed protein and oil concentrations over time across different maturity groups (MGs) in the United States. Our study extends the previous analysis through the more recent period of 1999–2022. We found that seed protein concentration significantly declined over time in short (MG0, MG1, and MG2) and mid (MG3, MG4, and MG5) MGs at a rate of 0.04% and 0.06% per year, respectively. In contrast, protein levels in long MGs (MG6, MG7, and MG8) increased at a rate of 0.10% per year. For seed oil concentration, however, no distinct differences among MG classes were observed; instead, oil concentration showed a consistent increase of 0.11% per year across all classes. Additionally, we identified an inverse relationship between protein and oil concentrations, with protein decreasing by 0.26% for every 1% increase in oil, regardless of MG category. These insights helped to identify predictors for modeling seed protein and oil concentrations, allowing the assessment for additional information to increase prediction accuracy. We found that a simpler model including latitude, longitude, sowing date, and mean temperature from sowing to flowering performed similarly to more complex models that included additional environmental variables. Moreover, nonlinear relationships between predictors appear to play a significant role, as models capable of capturing these interactions outperformed linear approaches.