Trait-Environment Relationships and Tiered Forward Model Selection in Linear Mixed Models

To understand patterns of variation in species biomass in terms of species traits and environmental variables a one-to-one approach might not be sufficient, and a multitrait multienvironment approach will be necessary. A multitrait multienvironment approach is proposed, based on a mixed model for species biomass. In the model, environmental variables are species-dependent random terms, whereas traits are fixed terms, and trait-environment relationships are fixed interaction terms. In this approach, identifying the important trait-environment relationship becomes a model selection problem. Because of the mix of fixed and random terms, we propose a novel tiered forward selection approach for this. In the first tier, the random factors are selected; in the second, the fixed effects; in the final tier, nonsignificant terms are removed using a modified Akaike information criterion. We complement this tiered selection with an alternative selection method, namely, type II maximum likelihood. A mesocosm experiment on early community assembly in wetlands with three two-level environmental factors is analyzed by the new approach. The results are compared with the fourth corner problem and the linear trait-environment method. Traits related to germination and seedling establishment are selected as being most important in the community assembly in these wetland mesocosms