Testing a core assumption of trait-based ecology: Are plant traits good predictors of individual performance?

Background/Question/Methods

While the trait-based approach has taken an increasingly central role in plant ecology, some of its fundamental assumptions remain poorly tested (Shipley et al. 2016. Oecologia; Salguero-Gomez et al. 2018. Functional Ecology). At its very core, functional ecology is based on the assumption that the plants traits commonly measured are important in determining the fitness of organisms in their environment (Violle et al. 2007. Oikos; McGill et al. 2006. Trends Ecol Evol). Yet, the rare studies relating traits to performance find weak evidence for this (Paine et al. 2015 J. Ecol.; Herben et al. 2012. J. Ecol.). To test this core assumption, we ask: ‘Are functional traits are strong predictors of the performance of individual plants?’ We also explore the importance of intraspecific trait variation by asking ‘Do individual trait values outperform species mean trait values?’ On each of 360 saplings from 24 co-occurring temperate tree species, we measured plant performance as the relative growth rate of basal area, 18 commonly-measured functional traits affecting five vital physiological function, and 19 environmental variables collapsed into five principal components. Using gradient boosted models, we predicted individual performance from functional traits, environmental variables and covariates (age and species identity).

Results/Conclusions

Preliminary analyses find that the 18 most commonly measured traits are mild determinants of relative growth rate in this system; the GBM explained 36% of the variance in the training dataset and 5% in the test dataset. The low explanatory power suggests that commonly measured traits are missing important aspects of ecology. The discrepancy among training and testing data indicates that the model has some explanatory power, but little predictive power. This suggests that trait-environment-performance relationships are highly variable among individual trees and that the relationships found here are unlikely to hold in other systems. The model retained all traits and contained a very large number of interactions, indicating that the effect of individual traits on performance depend on other traits and on the environment. Further, using species mean trait values instead of individual trait values decreased the predictive power of the model to 15%, indicating that capturing intraspecific trait variation is essential to predict individual performance.In summary, we find that the most commonly measured traits are mild predictors of growth rate in this system, that considering many traits and their interactions is necessary to predict individual performance, and that individual-level measurements cannot be substituted with species mean values.