Within-species patterns challenge our understanding of the causes and consequences of trait variation with implications for trait-based models

Functional traits could facilitate ecological prediction by provide scale-free tools for modeling ecosystem function. Yet much of their utility lies in three key assumptions: 1) that global patterns of trait covariation are the result of universal trade-offs independent of taxonomic scale, so empirical trait-trait relationships can be used to constrain vegetation models 2) that traits respond predictably to environmental gradients and can therefore be reliably quantified to parameterize models and 3) that well sampled traits influence productivity. We use an extensive dataset of within-species leaf trait variation in North American conifers combined with global leaf trait datasets to test these assumptions. We examine traits central to the `leaf economics spectrum', and quantify patterns of trait variation at multiple taxonomic scales. We also test whether site environment explains geographic trait variation within conifers, and ask whether foliar traits explain geographic variation in relative growth rates. We find that most leaf traits vary primarily between rather than within species globally, but that a large fraction of within-PFT trait variation is within-species. We also find that some leaf economics spectrum relationships differ in sign within versus between species, particularly the relationship between leaf lifespan and LMA. In conifers, we find weak and inconsistent relationships between site environment and leaf traits, making it difficult capture within-species leaf trait variation for regional model parameterization. Finally, we find limited relationships between tree relative growth rate and any foliar trait other than leaf lifespan, with leaf traits jointly explaining 42% of within-species growth variation but environmental factors explaining 77% of variation. We suggest that additional traits, particularly whole plant allometry/allocation traits may be better than leaf traits for improving vegetation model performance at smaller taxonomic and spatial scales.