Studies on changes in the local diversity of coral reef fishes following acute disturbances show that very extensive coral loss has a marked effect on the diversity of reef fishes (Figure 1). The overall relationship between percentage coral loss and changes in fish diversity was non-linear and best represented using a polynomial (r² = 0.21, n = 75), where the y-intercept is set to 0. Where coral loss exceeded 60%, diversity of fishes generally declined, although there were some examples of negligible loss of diversity when coral decline was >80%. In contrast, moderate levels of coral loss often had limited or positive effects on the local diversity of reef fishes (Figure 1a). This complex relationship may reflect differential responses of coral loss on coral-dependent compared to generalist fishes, which may benefit from some reduction in overall coral cover [63]. Increases in local diversity of fishes following moderate declines in coral cover may result from increases in overall habitat heterogeneity, whereby reefs with intermediate levels of coral cover are likely to support a diverse mix of fishes that live on live coral versus non-coral substrates [23]. However, extensive (>60%) coral loss may lead to a net decline in the diversity because increases in diversity of generalist species cannot compensate for the local extirpation of coral-dependent species. There were however, several examples where moderate declines in live coral cover had a disproportionate effect on the local diversity of fishes. During experimentally induced disturbances on the GBR, in which coral cover was reduced by 16–36%, the proportional decline in diversity of fishes was 1.8–2.3 times the proportional coral loss [76]. Studies that revealed marked declines in diversity of fishes following minor (<20%) coral loss are mostly conducted at very restricted spatial scales and tend to consider a very restricted set of fishes, mostly coral-dependent fishes. It is likely therefore, that these studies overestimate changes in local diversity of coral reef fishes.

The unexplained variation in changes in local diversity of coral reef fishes for a given level of coral loss is likely due to a number of confounding influences [23], whereby effects of acute disturbances on reef fishes may be more or less severe depending on associated changes in biological and physical structure of reef habitats. Most importantly, the effects of coral loss on coral reef fishes are expected to be much more pronounced when coral loss is combined with declines in structural complexity of reef habitats [22,82], but few studies have explicitly measured changes in structural complexity associated with acute disturbances. In this study, we assumed that biological disturbances (e.g., bleaching and outbreaks of A. planci) would have limited influence on structural complexity of reef habitats in the short-term, compared to physical disturbances (e.g., cyclones) [23,24]. However, the longer-term effects of biological disturbances may be equivalent to that of physical disturbances [83] as the skeletons of dead corals are highly susceptible to biological and physical erosion [84]. Declines in topographic complexity tend to occur 4–5 years after acute biological disturbances [24], so we tested whether short-term (<4 years) effects of coral loss are less severe compared to physical disturbances or longer-term (>4 years) effects of biological disturbances, where coral loss is likely to be compounded by declines in the physical structure of reef habitats. When considering only those studies where coral loss may be accompanied by changes in structural complexity, the relationship between coral loss and biodiversity loss for fishes is much stronger (r² = 0.36, n = 48), and at high levels of coral loss, resulting declines in diversity of fishes are more pronounced (Figure 1). Structural collapse of coral habitat is not a necessary consequence of coral loss; in some environments, such as highly exposed reef fronts, corals contribute very little to topographic complexity [64,83]. However, in the absence of direct and consistent measures of structural complexity, comparisons between biological versus physical disturbances may indicate the extent to which changes in structural collapse accounts for observed variation in the relationship between coral loss and changes in species diversity of reef fishes.

Variation in the effects of coral loss on diversity of fishes may also depend on the absolute level of coral loss [23]. When looking at absolute (rather than proportional) changes in coral cover and fish diversity, it is clear that high levels of coral loss (e.g., >40% coral loss) almost universally, lead to declines in diversity of reef fishes, especially when caused by physical disturbances (Figure 1b). In this instance, large declines in absolute coral cover are only possible when coral cover is very high prior to disturbance, such that habitats are likely to be dominated by fishes with strong-reliance on habitat-forming corals. Overall, there was a strong negative relationship between absolute coral loss and absolute changes in fish diversity (r² = 0.41, n = 74), which did not differ greatly between biological versus physical disturbances.

The Indo-Australia Archipelago (IAA) is the world's richest coral reef biodiversity hotspot, encompassing coral reefs of northern Australia, Papua New Guinea, Solomon's, Indonesia, and Philippines [85]. Species richness of coral reef fishes declines five-fold as you move from the IAA to the central Pacific, and declines even further as you move further east [85]. Reef fish diversity also declines (but much less rapidly) as you move west across the Indian Ocean. However, the lowest diversity of coral reef fishes (and corals) is found in the Caribbean [25]. Furthermore, these global diversity gradients are reflected within most major families of coral reef fishes [1]; declines in the diversity of reef fishes as you move away from the IAA are attributable to declines in the number of species within each major family, rather than loss of entire families. Consequently, some key functional groups are represented by only one (or very few) species in highly depauperate locations. These locations are thus expected to be highly susceptible to acute disturbances, whereby the loss of entire families will compromise ecosystem function and may have a range of feedbacks that ultimately result in extensive biodiversity loss [1]. Accordingly, the most pronounced changes in the structure of benthic assemblages (e.g., phase shifts from coral- to macroalgal-dominance) have occurred in depauperate and marginal reef locations, such as the Caribbean and eastern Pacific [1].

Despite regional differences in the extent of coral decline recorded, recent episodes of coral loss appear to have had a disproportionate effect on fishes in areas of high, not low, diversity (Table 2). Observed changes in the species richness of reef fishes were greatest in the central Pacific (e.g., Samoa, and French Polynesia) but these locations also experienced the greatest coral loss. When changes in species richness were considered relative to local levels of coral loss, the standardized declines in fish diversity were greatest for the IAA (Table 2). These data suggest that low diversity systems are not necessarily more susceptible to acute disturbances, though we must also consider geographical differences in disturbance history. In the Caribbean, for example, there has been a long-history of anthropogenic disturbances that have greatly altered contemporary fish communities [86], possibly favoring more generalist species that are less susceptible to environmental perturbations. Our results are however, very consistent with Mora et al [31], who conducted a comprehensive global analysis of interactions between human population density, biodiversity of fishes, and ecosystem functioning on coral reefs. This study showed strong positive relationships between biodiversity and ecosystem functioning (measured as standing biomass of coral reef fishes), but also found that more diverse locations are more (not less) susceptible to anthropogenic pressures (e.g., fishing) imposed by high human populations [31]. Mora et al. [31] argued that high diversity systems have evolved greater specialization and increased efficiency in energy transfer, such that the loss of species will have disproportionate effects on ecosystem functioning. More importantly, the relationship between biodiversity and ecosystem functioning was non-asymptotic (non-saturating) suggesting that even in high diversity systems the addition or loss of species will have potentially devastating effects.

Species within a particular functional group may differ in their responses to disturbances (and resource depletion) owing to differential reliance on highly susceptible resources (e.g., coral), differences in their ecological versatility [24], distribution relative to disturbed sites and habitats [94], or variation in the scales at which they operate (e.g., due to differences in home ranges) [7,95]. Species in strong competition for specific resources are also expected to exhibit compensatory changes in abundance during acute disturbances, which would further contribute response diversity within functional groups [95]. However, given that most functional groups are based on dietary composition and feeding behavior, it is also possible that entire functional groups will respond similarly during disturbances [7], especially highly specialist groups, such as corallivores [8]. If so, high redundancy and high biodiversity will contribute little to ecological stability and ecosystem resilience [25]. Previous studies have demonstrated contrasting responses to disturbance between functional groups (e.g., herbivores versus corallivores) [24,39], but how variable is the response of species within a functional group and how does the within-group variation compare to differences among groups?

To compare variation in responses (changes in the abundance of fishes relative to local coral loss) within versus among functional groups, we used a three-level nested ANOVA. We also calculated variance components to assess the relative contributions of (i) primary functional groups (four-levels), (ii) secondary functional groups (19-levels), and (iii) species (403 species) in accounting for differential responses of fishes to coral loss. If there is very high response diversity among fishes within functional groups, we would expect to find a disproportionate amount of variance explained by species. If however, responses of fishes are strongly influenced by their specific ecological functions (e.g., herbivores versus carnivores) and there is relatively limited response diversity within groups, then most of the variation is likely to be explained by either primary of secondary functional groups. Variance components in responses of fishes from 30 different studies and 403 species revealed very limited response diversity within functional groups (Figure 2). Despite significant variation in responses of individual species (nested within primary and secondary functional groups), it was the distinction between primary functional groups (corallivores, herbivores, omnivores and carnivores) that accounted for the greatest proportion of variation (62.4%) in responses of fishes. Differences among secondary primary groups (which provides greatly increased resolution in separating functionally-important groups of fishes) accounted for much greater variation compared to individual species, but differences in mean responses between these groupings were not significant (Figure 2). This analysis shows that fishes with fundamentally different trophic functions vary in their responses to coral loss (which is already known [24]), but also that there is relatively low response diversity within most functional groups. Perhaps more important, however, is the relative levels of response diversity apparent within each functional group (Figure 3, Table 4).

To compare within-group variation across different functional groups we constructed frequency distributions of species responses (Figure 3). These plots revealed a wide range of responses for planktivores, carnivores and herbivores, but relatively limited response diversity among corallivorous fishes (Figure 3). Importantly, modal responses were much more pronounced for corallivores, compared to all other groups, whereby most species studied exhibited a > one-fold decline in abundance following local coral loss. The strong modality in the distribution of responses for corallivorous fishes is reflected in the low value for kurtosis (Table 3). Kurtosis is a commonly used descriptor of size-frequency distributions, which accounts for both the concentration and spread of data [96]; high levels of kurtosis generally reflect “fatter tails” or an over-representation of extreme values. As such, we used kurtosis as a proxy for response diversity, and compared values across primary and secondary functional groups, wherever there was sufficient data (>30 observations) to construct meaningful frequency distributions of individual responses (Table 3). Kurtosis was very low (4.3–5.3) across all groups of corallivores, but was generally much higher (>30.0) for herbivores, omnivores and carnivores. However, this does not in itself guarantee the persistence of key ecological functions during natural or anthropogenic disturbances. Despite marked differences in response diversity, it was apparent that more species declined in abundance than increased in abundance across all four groups (Figure 3), such that the median response was always negative.

Overall, there were strong negative effects of coral loss on the individual abundance of coral reef fishes. In 60% (815 out of 1360) of all observations, local abundance of fishes declined following declines in live coral cover. Moreover, declines in the abundance of fishes were often (in 499 out of 1360 cases) disproportionately (up to five-fold) greater than proportionate loss of live coral cover. For example, Syms and Jones [76] documented the local extirpation of Chaetodon trifascialis following a 14% decline in live coral cover. As such, the decline in abundance of C. trifascialis was >five-fold greater than declines in the abundance of live coral. Disproportionate declines in the abundance of fishes are most likely attributable to selective effects of coral loss. In the former example, marked declines in the abundance C. trifascialis probably relate to extensive loss of preferred coral prey even though overall coral loss was very moderate. Chaetodon trifascialis is among the most highly specialized of coral reef fishes and feeds almost exclusively on only one type of coral, tabulate Acropora [60]. Accordingly, spatial and temporal changes in the abundance of this species are strongly related to changes in the abundance of tabulate Acropora, rather than overall changes in live coral cover [97]. Similarly, for other fishes (e.g., among diurnal planktivores that often rely on coral for shelter) the extent to which coral loss affects local population size, probably relates to the changes in the local availability of very specific coral-types [98], or even the relative size and complexity of different coral hosts. However, the high proportion of fishes that decline in abundance following coral depletion supports the notion that many different fishes benefit from extensive cover (and probably diversity) of corals [33], even though relatively few species may be explicitly reliant on coral for food or shelter [24]. However, widespread declines in abundance of fishes were sometimes offset by very pronounced increases in the abundance for the small number of species that seemingly benefit from local coral loss. In extreme examples, the abundance of fishes (often herbivorous species) increased 10-fold following coral depletion, such that there was an overall increase in the mean abundance of herbivores, omnivores and carnivores (Table 4).

Most studies ascertain effects of coral loss on reef fishes by quantifying changes in their distribution and abundance [99], which implicitly assumes that such disturbances will lead to widespread mortality and/or migration. It is possible, however that fishes that do not exhibit immediate and pronounced changes in abundance are nonetheless negatively affected by severe coral loss, whereby coral loss leads to declines in growth or condition [100,101]. Coral feeding fishes are generally considered to be among the first and worst affected fishes following extensive coral loss. On the GBR however, local densities of all coral feeding butterflyfishes were unchanged for >2 years after 90% loss of live coral, caused by severe bleaching in 2001–02 [77]. These butterflyfishes did eventually (after five years) succumb to local coral depletion, and several species went locally extinct, but short-term effects of coral loss on these species were only apparent by quantifying changes in the physiological condition of individual fishes [101]. It is also possible, that some species may fluctuate in abundance independent of local changes in habitat structure. For example, rapid and pronounced increases in the abundance of herbivorous fishes following localized coral loss may be caused by aggregation of adult fishes at highly disturbed locations [23] or pronounced pulses of recruitment, but it is unclear to what extent these are attributable to habitat changes. Many fishes, especially short-lived species, exhibit random fluctuation in local abundance due to marked inter-annual variation in recruitment [102]. There are also likely to be apparent changes in the abundance of some species due to vagaries associated with random sampling, whereby there may be as much variation in abundance of species between successive surveys conducted minutes versus years apart [103]. As a consequence, our meta-analyses of data from different studies are likely to have significant overestimated response diversity within functional groups. Moreover, the concept of functional redundancy implicitly assumes that all species contribute equally in fulfilling necessary functions, whereas common or large species often exert a disproportionate role in ecosystem stability and function [104]. We must also consider the relative importance of different functional groups, relative to their apparent level of response diversity. For example, it is apparent that corallivorous fishes (especially, obligate coral feeding species) have limited response diversity (Figure 3), but are coral-feeding fishes critically important to ecosystem function? The remainder of this study focuses on corallivorous fishes, which are the worst affected groups of fishes, and herbivorous fishes, which are potentially the most important functional group.

Corallivorous fishes have the most apparent and direct reliance on live corals [105,106] and are often among the worst affected fishes following distinct episodes of coral loss [24]. Accordingly, the response diversity among corallivorous fishes was much lower than for all other functional groups (Figure 3). It is also possible that the strong consistency in responses of corallivorous fishes is attributable to the extensive research that has been undertaken to clarify and confirm the trophic function and dietary composition of nominal corallivores [106], whereas some other functional groups (e.g., herbivores and carnivores) are less well defined. However, the extent to which corallivorous fishes feed on corals (vs. other non-coral prey) is highly variable [60], such that most studies distinguish between those fishes that feed exclusively on live corals (obligate corallivores) versus those fishes that only partly or occasionally consume corals (facultative corallivores) [60]. There is also a third group of coral-feeding butterflyfishes, which specialize on soft (alcyonarian) corals as opposed to hard (scleractinian) corals [60]. Corallivores within each of these secondary functional groups tend to decline in response to coral loss, however the magnitude of this response is highest, and the response diversity (kurtosis) is lowest for obligate corallivores (Table 3). These differences were not large (particularly when compared to differences among primary groups), but are consistent with expected differences in their responses to coral loss, whereby obligate coral feeders are much more reliant on live corals and likely to be much more affected by coral loss compared to facultative coral feeders. Detailed research on Chaetodon butterflyfishes [72,77], shows that facultative corallivores are relatively unaffected by moderate decline in abundance of corals, presumably because they can compensate by increasing intake of non-coral prey. However, corals may still represent an important and necessary component of their diet, such that even facultative corallivores are adversely affected by extensive coral loss [77].

Coral feeding represents a relatively unique and potentially important trophic link between scleractinian corals and higher consumers [106]. Corallivorous fishes make this carbon immediately available to higher consumers, and greatly increase transfer efficiency of energy from corals to fishes. While many early researchers suggested that there were few fishes capable of feeding on corals [105], recent reviews of corallivory reveal that there are at least 128 species of reef fishes from 11 families that feed on coral [106,107]. Despite their diversity, the functional-importance of corallivorous fishes is equivocal. One interesting observation, which suggest that corallivorous fishes are not overly important, is that obligate corallivores are one of the only functional groups of fishes (e.g., among all secondary functional groups listed in Table 3) that are not represented across all coral reef regions. Although recently degraded, coral reef assemblages in Caribbean have been isolated from Indo-Pacific reefs for >three million years and persisted despite the lack of obligate coral feeding fishes [107,108]. This lack of obligate coral-feeding fishes may be compensated by the local diversity of obligate coral-feeding invertebrates [109], whereby there are a very large number of invertebrate species known to feed (both obligately and facultatively) on live coral tissues. However, overall effects of corallivores on habitat-forming corals and entire reef ecosystems are generally considered to be damaging [107], reducing growth, fitness, or ultimately abundance of live corals.

Concern regarding the loss of key functional groups on coral reefs mostly centers on herbivorous fishes, and particularly grazing species [25,94]. Herbivorous fishes (and invertebrates) are believed to play a critical role in controlling macroalgae, which might otherwise monopolize reef substrates [25] and prevent population replenishment by corals and other habitat-forming species [110]. This is especially important given the increasing diversity, frequency and severity of disturbances that threaten corals, because continual grazing of reef algae is fundamental to the recovery and resilience of coral assemblages [26]. Extensive coral mortality provides space on the reef that is rapidly colonized by turf algae [111]. On reefs where herbivorous fishes and/or sea urchins are abundant, algal assemblages may remain as cropped turf forms following coral bleaching and coral loss. However, where grazing pressure is low, or if coral mortality is extensive, herbivores may be unable to counter increased algal abundance and macroalgal blooms can develop [27,111,112], bringing about a phase shift from coral- to macroalgal-dominated reef habitats.

Although our results show that there is a degree of response diversity among herbivorous fish species as a whole (Figure 2), there was moderate response diversity within each of the herbivorous functional groups, with low response diversity among the territorial farmers (Table 4). This distinction is important as inferences from the primary ‘herbivore’ functional group imply that all herbivorous species perform a similar role. There is, however, considerable variation in the feeding behavior of these fishes [113], and this variation is directly related to the impact on reef processes [25]. Whilst excavating, scraping, and cropping taxa generally consume algal turfs, they perform different and complimentary roles in helping reefs to resist shifts to alternate states and reassemble following disturbances. Croppers remove the upper portions of the algae (and associated epiphytic material), and in doing so prevent the growth and development of macroalgal species; scraping taxa remove pieces of the substratum together with the algae, clearing space for the settlement of other organisms; and excavating taxa remove large amounts of carbonate material when feeding and play an important role in external bioerosion. Furthermore, the removal of adult macroalgae appears to be restricted to a limited suite of species (i.e., macroalgal browsers [30,89,114]) and represents a separate but critical process in the reversal of phase-shifts [115]. Differential responses of herbivorous fishes to coral loss probably reflects their differential reliance on coral structures; territorial farmers showed the most consistent responses, whereby all species were negatively affected by severe coral loss. This finding may relate to their smaller size and increased site fidelity, making them more prone to habitat modification from disturbances compared to other larger more mobile herbivorous species. It is also apparent that many territorial farming damselfishes favor sites with high coral cover in which to establish territories [116].

Clearly, consideration needs to be given to the role of the species in ecosystem processes rather than broad trophic groups (e.g., herbivores) when assessing response diversity and functional redundancy. Recent studies have shown that even the functional redundancy of herbivores appears to have been overestimated, with single species performing a disproportionate role in ecosystem processes [30,117]. In such instances, the loss of a single species may compromise the entire process. For example, overfishing of the excavating parrotfish, Bolbometopon muricatum, has resulted in a near total loss of external bioerosion on several Indo-Pacific reefs [117]. This loss of function has occurred despite the presence of several other excavating species (four-eight species) on these reefs. It has also been suggested that functional redundancy and response diversity may be influenced by the scales at which different species operate, with species operating at different scales providing mutual reinforcement for function maintenance [94]. Among reef fishes, however, there are limited differences in the scales at which species operate, and nearly all species operate at scales below the spatial extent of most major disturbances (e.g., cyclones, bleaching and outbreaks of A. planci). Recent studies have demonstrated that herbivorous coral reef fish are relatively site attached, at least for those species examined to date, and have a more restricted home range than previously assumed (<10 ha; [118,119]).