Large scale intraspecific trait variation
In addition to mapping phylogenetic and environmental trait signals, another essential reason for a combined approach is that it allows for predictions of intraspecific variation (Figure 5). Models based on CWMs, by definition, do not account for intraspecific variation. Although environment-only models implicitly include intraspecific variation, it is not decoupled from trait variation due to shifts in species composition (Hulshof & Swenson, 2010). This interspecific variation must be separated from intraspecific variation to quantify the latter. Our combined modeling approach makes this separation possible for hundreds of species-trait combinations. Quantifying intraspecific variation in a comprehensive manner (i.e., for all species and across a large geographic region) is a necessary step towards testing hypotheses about the role of individual trait variation in species coexistence and interspecific competition (Hart et al., 2016). We explore the potential for our approach to enable research understanding the role of intraspecific variation in community dynamics by (i) comparing our estimates of intraspecific variation to field data for three widespread species, (ii) describing how the predicted intraspecific variation varies among eastern US species, and (iii) describing how trait variation may affect trends in one of the trait-tradeoffs described by the LES.
The combined model produced realistic ranges of intraspecific variation when compared with available independent data for widely distributed species. Specifically, the intraspecific variation of N% predicted from the combined model showed ranges similar to those observed from the field by independent datasets for three widespread and abundantly sampled species in the NEON, FIA and TRY datasets (Abies balsamea, Acer rubrum , and Fagus grandifolia ). For these species (Figure 5b-d), predictions from the combined model showed N% covering a large range of values (1.12 N% on average), comparable to the average difference between evergreen needleleaf and deciduous broadleaf species in the eastern US (1.13 N% based on NEON field data). Such a wide range of intra-species variation further supports the idea that adaptation to the environment accounts for a large proportion of community-level variation and may be driving community-level shifts across environmental gradients (Albert et al. 2010, Violle et al. 2012, Siefert et al. 2015, Fajardo & Siefert, 2018). The contribution of the environment to intra-species variation may also explain altitudinal (e.g. Acer rubrum) and latitudinal (e.g. Abies balsamea ) gradients observed for these three species.
Predicted patterns of intraspecific variation varied widely across species. For N%, the ratio of predicted intraspecific to total observed interspecific variation ranged from less than 10% for species with limited geographic ranges (e.g., Populus heterophylla ,Sabal palmetto , and Gleditsia aquatica ) to over 60% for broadleaf species with broad geographic ranges (e.g., Cercis canadensis, Betula lenta , and Carpinus caroliniana ). Intraspecific responses of N% to temperature also showed a variety of patterns among species (Figure 5a, Figure S.15), including 1) bell-shaped patterns (n=85) like those observed in compilations of field data (Reich & Oleksyn, 2004, Laughlin et al., 2012); 2) negative relationships (n = 7); and 3) cases with no significant relationship (n=108). We observed these different patterns despite the model not including species-by-environment interactions. This is possible due to differences in how the environmental drivers are jointly related in different subregions of the posterior predictive distribution that represent different geographic regions and different environments by trait combinations. Therefore, while the underlying relationship between traits and temperature in the model contains a single dominant mode with an uptick at very high temperatures, analyzing patterns of trait distributions from predictions across large geographic areas shows that focusing only on the subregions of the posterior that are relevant to the species of interest can yield different species level relationships.
Variation in environmental drivers may also impact trends in trait trade-offs, including those described in the LES. Since these relationships affect nutrient and carbon use efficiency in plants (Reich, 2014), understanding how they change geographically under different environmental conditions is fundamental for improving how carbon dynamics are simulated by earth system models (Weng et al., 2017). Our results suggest that within the Eastern US, the relationship between N% and LMA is weaker than the global inter-species average, but stronger than local within-species relationships (Figure 6). This relationship, first observed from field data from the western US (Anderegg et al., 2018), generalizes to the Eastern US and holds when extrapolating trait values continuously across species ranges. The scale-dependent change in the strength of the relationship may be driven by different nitrogen to mass allocation strategies between maximizing short term productivity (higher foliar photosynthetic mass) or long-term defense (higher foliar structural mass) (Osnas et al., 2018). According to this conceptual model, we expect stronger negative N%-LMA relationship for datasets with high variation in LMA, due to variation in leaf structural components. This is the case of the global LES, calculated on species sampled from all biomes and many ecosystems (slope of the logarithmic N%-LMA linear relationship ~-0.7, Wright et al., 2004). At local scales, where macro-environmental drivers have a much smaller effect on LMA variation, adaptations to local conditions may drive higher variation in photosynthetic mass leading to weaker N% to LMA relationships (slope of the logarithmic N%-LMA linear relationship ~-0.2; Osnas et al. 2018). At the intermediate scale of the Eastern US, both strategies may contribute significantly to the relationship, resulting in N%-LMA trends falling halfway in between the two extremes (slope of the logarithmic N%-LMA linear relationship ~-0.43). This slope was consistent across broadleaves, needleleaves, deciduous and evergreen species, providing support that this trait-tradeoff can be represented the same across plant functional types used in earth system models.