basis of random colonization opportunities by the component species or does the presence of a certain species, or groups of species, change the probability of successful establishment by yet other species? The resolution of the question could be approached at a larger scale to include both historical (i.e., lineage) and dispersal components, although much of the ecological literature has focused on a finer scale that emphasizes dispersal (but see Gorman 1992; Winston 1995). As natural communities are increasingly besieged by nonnative taxa, the answer to the questions of how species are added to assemblages has also taken on new importance in the efforts to develop models predicting the vulnerability of native communities to invasion by nonnative taxa (see also Chapter 15).
Understanding the formation and maintenance of assemblages also requires an understanding of the vagility of fishes. Without some degree of movement, fishes would be unable to colonize new habitats or to recolonize habitats from which they have been eliminated because of biotic or physical impacts. Movement ability varies widely among fish taxa and within a given taxon by life-history stage, physiological state, and simply by individual differences.
COLONIZATION OF ASSEMBLAGES
The Search for Assembly Rules
The question of whether or not there are ecological “rules” governing how species are added to communities has intrigued ecologists for decades. Indeed, long before that, Plato provided the first attempt in the fourth century BC when he considered how human communities were assembled (Belyea and Lancaster 1999). A seminal paper by Diamond (1975) on the addition of bird species to islands initiated a debate that continues in some measure to the present. Diamond hypothesized that species composing a community are “selected and coadjusted in their niches and abundances, so as to fit with each other and to resist invaders.” A more formal definition of assembly rules was proposed by Belyea and Lancaster (1999) in an effort to make the distinction between underlying mechanisms, rules related to the mechanisms, and the resultant patterns resulting from the rules. In their definition, assembly rules are “general and mechanistic, and operate within the case-specific constraints imposed by colonization sequence and environment.”
For several decades now since Diamond’s (1975) publication, debate has centered on issues of whether communities are structured as opposed to random. Such debate is, in part, because Diamond did not first clearly demonstrate that the bird communities he studied were actually structured rather than random collections of species before he proposed specific rules for assembly (Weiher and Keddy 1995). Later studies focused on distinguishing between structured versus random assemblages by testing observed patterns of species occurrences against null models of species assembly, resulting in the demonstration that at least some of the patterns that ecologists used to invoke assembly rules could be attributed to chance (Connor and Simberloff 1979). The use of null models increased the rigor of studies on community structure and assembly rules, although philosophical differences in constructing null models, such as the appropriate species pool for building the model, have continued to foster debate (see Gotelli and Graves 1996; Box 5.1).
BOX 5.1 • Null Models in Community Ecology
The experimental tools that are conventionally available in community ecology are field, laboratory, and so-called natural experiments (Gotelli and Graves 1996). Well-designed experiments require both tests and controls so that the effect of a potential factor can be evaluated. However, ecological communities, or even a subset such as fish assemblages, are notoriously complex, making large-scale manipulations or laboratory experiments impractical from the standpoint of logistics, time, or money. For instance, assessing the impact of Species X on the colonization probability of Species Y could require eliminating Species X from randomly chosen sites, introducing Species Y, and then comparing the colonization success of Species Y between sites with and without Species X.
Coupled with the ethical or legal issues of manipulating systems that might already be at risk, the potential of being able to “manipulate” communities statistically is certainly an attractive alternative (Connor and Simberloff 1986). In the case of Species X and Y, a null model might be constructed such that the probability of the two species co-occurring in the absence of any interaction could be calculated and then compared with the actual co-occurrence. Simply put, a null model “is an attempt to generate the distribution of values for the variable of interest in the absence of a putative causal process” (Connor and Simberloff 1986). However, the challenge is to design an appropriate null model for the situation being tested! The literature on null models, relative to testing forces involved in shaping communities, is a litany of proposed methods followed by criticisms thereof. (For a few examples see Connor and Simberloff 1986; Brown et al. 2000, 2002; Stone et al. 2000; and Fox 2001.)
The use of null models in addressing the question of community assembly has focused on the interactions of species with potential sites and the interactions among species (Gotelli and Graves 1996). As outlined by Gotelli and Graves (1996), all models that have been used to test the hypotheses generated by assembly rules have identifiable shortcomings. There are key questions to consider: (1) what species to include in the species pool from which potential assemblages are randomly, or otherwise, extracted; (2) rather than species, should some other category, such as functional groups (i.e., a collection of species all of which have a similar ecological function) be used; (3) what metric is most appropriate to determine whether or not assemblage organization follows a random pattern; and (4) what is the most appropriate way to design a null model (being that the possibilities are large)? Obviously, null models, although potentially very useful, are not without their own set of issues. The use of null models in ecological applications is treated in detail by Gotelli and Graves (1996).
A challenging issue is one of choosing a preinteraction species pool—one that has not already been subjected to long periods of interaction (such as competition) that would have caused the removal or stifling of speciation of certain species or lineages (Colwell and Winkler 1984). As a consequence of sampling a postinteraction species pool, the strength of the effect being measured is consistently underestimated. The issue of sampling a postinteraction species pool is termed the “Narcissus Effect” (Colwell and Winkler 1984), named after the Greek god Narcissus who, according to one version of the story, when gazing into a pool saw only his own image and could not discern the depth of the pool.
Clearly there are challenges to providing rigorous tests of community assembly rules. One approach that has been taken is to consider predictions that would be met by assemblages if, indeed, assembly rules were important. These have primarily focused on the random versus structured dichotomy. This approach was taken by Gotelli and McCabe (2002), who conducted a meta-analysis of 96 published studies of species presence-absence matrices, comprising a wide range of taxa and geographic locations, to determine the generality of Diamond’s (1975) model of assembly rules. Their results showed that most plant and animal assemblages do follow predictions of Diamond’s assembly rules in that there are fewer species combinations than expected by chance (i.e., communities are nonrandom), a result also shown by community assembly studies using laboratory microecosystems (Drake et al. 1993). In an effort to move studies of community assembly forward, Weiher and Keddy (1995) argued that the issue of whether there are nonrandom communities has long been resolved in the affirmative and that it is time to move on to newer questions. Although fish assemblages treated by Gotelli and McCabe (2002) showed less evidence for structuring by competition than did other groups of organisms, the low sample size (n=3) precluded any meaningful generalities.
Application to Fish Assemblages
There have been relatively few studies that specifically addressed the reality of assembly rules governing fish assemblages or included fish assemblages in more general analyses. The studies of fish assemblages that have evaluated the predictions derived from community assembly rules have had mixed results in terms of finding support.
Matthews (1982) provided an early attempt
