Nichodemus, C. O.; Meireles, J. E.

Plant functional traits vary across leaf ontogeny and phenology, yet most trait data are snapshots from narrow time windows that miss this temporal dimension. Leaf spectra are increasingly used with empirical models to predict traits, but whether such models accurately capture phenological variation remains unclear. We monitored leaf traits and spectra weekly across a full growing season, generating 7,515 spectra from seven temperate species. Using partial least squares regression, we built three models [all-season and week-as-covariate models (both trained on full-phenology data), and a peak-season model] and evaluated them alongside a widely used model against directly measured traits. Full-phenology models predicted LMA and equivalent water thickness (EWT) with high accuracy (R2 > 0.85) and nitrogen with intermediate accuracy (R2 = 0.64); carbon accuracy was low across all models (R2 < 0.26), likely due to a small sample size. Peak-season trained models performed poorly when evaluated across the full season, often producing biologically unrealistic predictions. Traits and spectra varied significantly across phenological stages both within and among species. Ignoring phenological variation systematically biases trait estimates and ecological inference. Coupled with phenologically representative training data, spectra can capture the temporal dynamics of plant function, enabling novel research in ecology and evolution.