Predictions of herbivory in monocultures and mixtures based on plant functional traits

Background/Question/Methods

Plants can affect the capacity of herbivores to find, choose and consume them through their functional traits. Here we asked which plant traits are of importance in determining herbivory for a large pool of species interacting with natural invertebrate communities.

At the field site of the Jena Experiment (Germany), we measured standing invertebrate herbivore damage along the experimental plant diversity gradient ranging from monocultures to 60 species mixtures in May and August 2010 on a total of 82 plots. At the same time, herbivory was measured in additional monoculture plots of all species. Functional traits from four trait groups (physiological, morphological, phenological and herbivore-related traits) were collected, from literature, traits databases and own measurements for all plants of the species pool. Using the Random Forest method and multiple regressions a model was constructed predicting species specific herbivory rates in monoculture. In a second step, this model was used to predict herbivory rates measured in mixed species plots based on aggregated community traits.

Results/Conclusions

We identified seven of 42 traits to be of importance predicting herbivore damage in monocultures: leaf nitrogen and lignin concentration, number of coleopteran and hemipteran species potentially feeding on the plants, leaf lifespan, stem growth form and root architecture. The final model accounted for 63% of the variation in herbivore damage. Very different traits were selected, thus a variety of plant characteristics, including root traits, are important when assessing folivory by a diverse community of herbivores. The model successfully predicting herbivory rates in monocultures did not scale up to multispecies communities when predictions were based on abundance weighted traits averaged for the communities. This model explained only 7% of the variation in measured herbivory rates and predicted consistently lower rates of herbivory than observed. A second model with relaxed assumptions by estimating new partial slopes in mixtures for the same set of traits, thus allowing for changed importance of the traits in mixtures, explained 25% of variation, which was considerably worse compared to monocultures.

Models based on community averaged traits that assume no interactions between plant species in mixtures and mechanisms linking traits and response variable to be unaltered in communities have been used to produce valid predictions for plant physiological processes, e.g. biomass production. The low performance of a similar model in our study predicting community herbivory indicates a high degree of non-additive effects in the interaction between the herbivore and food plants community.