Abstract
Phenomic traits can support soybean breeding when genetic signal, temporal redundancy, and uncertainty are evaluated together. We assessed 30 tropical soybean genotypes in ELO_25A using chlorophyll fluorescence and SPAD measurements, UAV-derived RGB indices, and agronomic traits across developmental stages. Spatial mixed models, with PAR as the sole physiological covariate, provided adjusted means, BLUPs, Cullis generalized heritability, and prediction-error covariance. Targeted reduced-rank models assessed temporal structure, and equal-weight PCA with propagated uncertainty summarized six retained features. Genetic precision varied with development and was strongest at R3–R4. Model comparisons favored rank-one summaries for SPAD V5–R3 and GLI R3–R4–R6, with proportionality imposed by the selected structure, whereas SPAD R6 provided distinct late information. PC1 explained 39.12% of standardized variation and contrasted R3 energy regulation with chlorophyll status. It was associated with plant height and pod number (r = 0.550 and 0.560; both q = 0.006) but not yield (r = 0.056; q = 0.822). Genotype scores formed a continuum, and no cluster solution met the stability criteria. The results support stage-aware feature reduction, germplasm discrimination, and efficient phenotyping within ELO_25A, but do not validate yield prediction or indirect selection. Validation across environments and independent populations is required before use in plant breeding.