We suggest (Proposition 5) that a community is functionally assembled when there is at least one species representing each functional type that is adapted to the habitat. We suggest (Proposition 6) that a community is fully assembled when each functional type has the maximum number of species that can coexist. Much biological diversity lies in the lower tail of the log-normal distribution, yet ecologists frequently trim this tail of rare species to construct models. The rising tide of species extinctions requires us to revisit this procedure, and to consider how to include conservation of rare species in models for community assembly. This requires us to recognize that there are two types of “rare” species in ecological data sets: those that are merely rare in samples and those that are designated as globally rare and at risk of extinction. Most communities arise from pre-existing communities, so when filters change there is often inertia in community response, and this inertia is related to traits. The principles laid out in this book provide a guide not only to theoretical understanding but for the challenges of ecological restoration.

In order to assemble an ecological community it may be helpful to know not only how many parts there are, but what kinds of parts there are. Communities require at least two classification systems that provide simultaneous and somewhat contradictory lists of parts: phylogenetic and functional. These two classification systems can be arranged hierarchically so that many parts (species) are nested within a smaller number of groups (functional types). Even with objective classification techniques, it is difficult to know how many groups exist, and the number selected may be somewhat arbitrary. There does not seem to be a way to tell, a priori, how many functional types we can expect to find in a specified landscape or habitat. This raises difficult questions about the nature of fitness landscapes and the geometry of n-dimensional trait space.

There is a long history of describing communities in ecology. It is now time to develop a general predictive framework for this discipline. The goal is to simultaneously provide a consistent theoretical framework to guide research and a practical framework to guide conservation of wild landscapes. We propose that this framework has four key elements: the species pool vector P, the local community vector C, a vector of environmental filters E, and a vector of functional traits T. The central challenge of community ecology is to predict the species composition of any community C, using prior knowledge of P, E and T. Common filters include flooding, fire and herbivory. Each community C is a subset of the regional species pool P and is the result of filtering that matches species’ traits to the local environmental conditions. Dispersal, competition and time are also important in community assembly.

The community is a subset of the species pool created by the action of one or more environmental filters. We propose (Proposition 1) that in any habitat, the power of a filter can be measured as the proportion of species that it removes from the species pool. Filters can be abiotic or biotic in origin. Drought, frosts, hypoxia and wildfires are important abiotic (physical) filters, while predation (including herbivory) and competition are important biotic filters. We also propose (Proposition 2) that in any habitat only a small number of filters is likely important. Moreover (Proposition 3), in any region a large number of species are likely controlled by the same filters. These three propositions provide guidance for setting priorities in research and habitat management. One of the most important challenges in community ecology is to identify and rank the filters that control species composition in specified landscapes.

The CATS model (Community Assembly by Trait Selection) uses average trait values of both species and communities in a system of linear equations to predict the relative abundance of each species in a community C from the regional species pool P. This model has been tested in a variety of habitats. There is an asymptote to the number of useful traits. The asymptote is determined by the strength of covariation among the traits of the species pool. The Traitspace model uses Bayes’ theorem to combine trait–environment relationships and multidimensional trait distributions for each species to predict the relative abundance of each species in a community C from the regional species pool P. Both models can be used to determine the importance of traits in community assembly. Each model has costs and benefits, which future work will clarify.

Species pools are the result of speciation and extinction, processes that mostly act at long time scales. We make the simplifying assumption that the number of species in the pool is a relatively stable number at the time scales used in community ecology. We describe how standard reference works and classic natural history studies provide data on species pools at a variety of scales. Three types of pools should be distinguished: the regional pool (an unfiltered pool of species which takes into account geographic limits), the habitat-specific pool (a filtered pool which takes into account habitat conditions) and the potential pool (which takes into account species that are adapted to different habitats and potential invasive species). We describe how pools can be constructed by processes including accumulating or downscaling. We discuss the challenges of creating consistent species pools for research and conservation purposes, and conclude that preparing such lists is an important priority in community ecology.

It is important to be able to name the plants and animals in one’s environment, but knowing the names does not in and of itself advance the study of ecology. Frank Rigler argued that the species-oriented approach to studying ecology is intractable simply because of the time it would take to obtain enough information on each species to generalize to the community scale. Life on Earth can be named (or classified) in two complementary ways, using phylogeny and functional traits. Trait matrices provide the raw material for trait-based ecology. Compilations and screening are two distinct sources of data for trait matrices. Compilation of traits across studies is an important way of generating data for global-scale synthesis. Screening traits of local communities in the field or under standard conditions is the most effective way of generating quality data for local communities.

In order to assemble an ecological community it may be helpful to know not only how many parts there are, but what kinds of parts there are. Communities require at least two classification systems that provide simultaneous and somewhat contradictory lists of parts: phylogenetic and functional. These two classification systems can be arranged hierarchically so that many parts (species) are nested within a smaller number of groups (functional types). Even with objective classification techniques, it is difficult to know how many groups exist, and the number selected may be somewhat arbitrary. There does not seem to be a way to tell, a priori, how many functional types we can expect to find in a specified landscape or habitat. This raises difficult questions about the nature of fitness landscapes and the geometry of n-dimensional trait space.