Understanding tree functional traits and how they relate to growth and mortality can provide unique insights into how trees and forests may respond to environmental stress. A great deal has been learned about trait relationships through empirical studies that have quantified, for example, the leaf and wood economics spectra and the global spectrum of plant form and function. However, the correlative nature of these analyses can limit the predictive power of the observed trait-trait relationships. This study aimed to complement such past work by assessing functional trait relationships that emerge from a process-based model of tree growth, carbon allocation, and mortality, with the objective of assessing the effects of environmental stress (filtering) on the trait space of North American trees (irrespective of species). We fit the process-based model to 1.6 million re-measurements of tree heights and radii from the Forest Inventory and Analysis (FIA) data within a Bayesian framework, producing posterior samples of 32 functional traits. Using these posterior distributions of tree functional traits, we simulated tree growth and mortality under 62 forest gap dynamics scenarios, and subsequently assessed the effects of non-random mortality (due to light limitation) on the predicted trait space of North American trees.
Results/Conclusions
The Bayesian analysis produced 33,000 parameter sets, each a vector of 32 functional traits representing a single, simulated tree. The posterior distribution of the parameters represents a theoretical trait space for North American trees, with each set yielding predicted heights and radii that agreed with FIA data. Gap simulations produced varying degrees of light limitation leading to simulated stand-/scenario-level mortality rates of 19% to 50%, which were explained by average light level (p<0.05, R2=0.85). Logistic regressions evaluated simulated tree-level mortality as functions of average light level only, functional traits only, or traits and light, and predicted mortality correctly for 42%, 80%, and 82% of the cases, respectively, when evaluated against a hold-out dataset. Parameter set- (or phenotype-) level mortality spanned 0% to 100%; 48% of the variation was explained by 20 traits (p<0.05), but 5 traits (maximum height, radiation use efficiency, senescence rate of coarse roots and branches, and leaf maintenance respiration) accounted for >90% of the overall R2. The trait spaces of living and dead trees differed in their hyper-volumes (spread of trait values) and centroid locations (nominal trait values). Thus, environmental filtering altered the functional trait space, especially under conditions leading to low/moderate mortality rates.