Relationships between meteorological variables and the productive performance of soybean lines

Abstract

The objective of this work was to identify the relationships between meteorological variables capable of predicting the productive performance of soybean lines using stepwise regression and unsupervised machine learning. The used experimental design was of augmented blocks with interspersed controls. The following regular and irregular treatments were evaluated, respectively: 3 F₄ lines (87.5% homozygous), 138 F₅ lines (93.75% homozygous), and 88 F₈ lines (99.22 homozygous) of soybean, totaling 230 segregating lines; and four commercial cultivars in three replicates, totaling 242 experimental units. The weather data were obtained from the NASA POWER and SISDAGRO platforms. Positive linear associations were observed between maximum temperature and vapor pressure deficit and vapor saturation pressure curve (r = 0.9), as well as a negative correlation between relative humidity and potential evapotranspiration (r = -0.7). These associations had a direct influence on the dynamics of soil water storage capacity and modulated all other correlations between meteorological variables. Stepwise regression and unsupervised machine learning are effective in identifying relationships between meteorological variables that predict the productive performance of soybean lines.