450 Keystone Species keystone and inferring that it is the primary determinant of the structure of its community when in fact it is not As noted by Mills et al (1993), this can yield a serious loss of credibility with respect to policy and management decisions Despite limitations of spatial and temporal scales and other shortcomings, experimentation supplemented by comparison remains the most powerful tool available for revealing the dynamics of communities and ecosystems (Paine, 1994) In many cases, however, practical, legal, and ethical concerns preclude manipulations of suspected keystones (e.g., sea otters, killer whales, or prairie dogs) It is clear, therefore, that the identification of keystone species must necessarily use a variety of approaches In addition to the comparative-experimental approach, alternative approaches include path analysis, sensitivity analysis, the study of natural or accidental invasions, the study of the consequences of overexploitation, and exhaustive and detailed comparison coupled with natural history and ideally small-scale ancillary experimentation (Power et al., 1996) Inferences based solely on descriptive natural history knowledge (e.g., diet composition and frequencies, behavior, and abundance) are likely to be misleading For example, under natural conditions a keystone predator may rarely consume the competitive dominant in a system (because it has sharply reduced the prey’s availability), and thus be overlooked as a possible regulator of the dominant or the community In the sea star system studied by Paine (1966), for example, barnacles, not mussels, were the most frequently consumed prey Thus, although natural history knowledge is fundamental to the understanding of the dynamics of any ecological system, much additional evidence is necessary before the ecological role of a species can be determined Experimental Approaches Keystone species are often identified by removal or exclusion experiments; that is, the presumed keystone is deleted from a portion of the habitat, and the effect of the removal on the community is compared to a separate, control portion of the habitat One problem with this approach is that if a suspected keystone species is removed but there is no detectable response by the community, it is not possible to conclude that the community lacks a keystone species In such a case, three alternative interpretations are possible (1) Predation is weak overall, such that no predator deletion will produce an effect (2) Predation is strong and there is a keystone species in the community, but it was not the species removed (3) Predation is strong but diffuse, and multiple predators would need to be removed to produce an effect To tease apart these alternatives, an appropriate experimental protocol in testing for keystone predation (Menge et al., 1994; Navarrete and Menge, 1996) should include: Treatment Explanation Tests ỵ Predators Intact community, all predators present Predators Quantifies total predation effect, all predators removed or excluded Control or reference (‘‘natural’’ community state) Strong vs weak predation À Single predator Quantifies single species effects, deletion of each predator species singly, leaving the others present Keystones vs diffuse predation For example, if consumer species are removed both collectively and singly, but predation is weak, the total effects of predation and the effects of each species with respect to their impact on community structure will be small (Figure 8(a)) With strong predation regimes, keystone predation would be indicated if total removals and removal of just one of the predator species were similarly large (Figure 8(d)) Finally, diffuse predation would be indicated if removal of each predator species singly led to significant but substantially smaller effects than removal of all predator species (Figure 8(g)) This protocol also has important implications for understanding how communities or ecosystems might respond to loss of trophically high species (Allison et al., 1996) In a system where predation is weak, of course, loss of species should have little effect (Figure 8(a–c)) With strong diffuse predation, where each consumer has an impact on community structure, species loss may lead to little change in the system as a result of compensation by the remaining consumers, at least until most consumers have been lost (Figures and 9) In both cases, the community response to species deletion is relatively predictable (Allison et al., 1996) With strong keystone predation, however, the consequence of species loss is relatively uncertain (Figures and 9) By definition, compensation by the weakly interacting predators, while possible, is not likely to fully account for the loss of the keystone species, and if the identity of this species is not known a priori, the system response is highly uncertain (see Figure 9) This protocol has rarely been used, but two examples illustrate its efficacy Menge et al (1986) evaluated the separate and combined impacts of most combinations of four groups of consumers in rocky intertidal communities in Panama The extremely high diversity of this tropical community imposed an immediate compromise in their design: Simultaneous single-species removals were essentially impossible because 440 consumer species were relatively abundant The compromise was to remove major consumer groups (i.e., omnivorous crabs, omnivorous fishes, predatory whelks, and grazing molluscs), each consisting of several common species Their partial factorial design (difficulties in separating crab and fish effects meant only 12 of a possible 16 treatments could be conducted) included treatments assessing the effects of removing consumers in single groups and in total After years, none of the single-group removals produced an effect similar in magnitude to the exclusion of all the consumers (Figure 10) Removal of two and three groups produced changes in prey that were intermediate between single-group and total exclusion treatments Thus, total predation pressure in this system was high but the effect of individual groups was small The lack of prey responses from single-group removals compared to the increasingly strong responses from removals of two, three, and all four groups suggests three things: That, up to a point, groups can compensate for reductions in other groups; that there was no keystone species in this system; and that predation was diffuse