Introduction

In a famous conversation with Euthyphro, Socrates posed the question: Is the pious loved by the gods because it is pious, or is it pious because it is loved by the gods? (Plato ~399 BCE cited in Tredennick 1954). One interpretation of this question is: does piety come from the bottom-up (from man) or top-down (from the gods)? Here we pose a similarly circular question about the relationship between a major group of benthic grazing fish on coral reefs, the parrotfish (Family Labridae; tribe Scarinae) and the benthic habitat of these fish: Does parrotfish abundance change because the benthic habitat changes, or does the benthic habitat change because abundance of the parrotfish changes? The first alternative implies bottom-up control of the parrotfish by benthos (e.g. Russ 1984; Cheal et al. 2012; Tzadik and Appeldoorn 2013; Taylor et al. 2014), the second, top-down control of the benthos by parrotfish (e.g. Hughes 1994; Bellwood et al. 2004; Mumby et al. 2006; Bonaldo et al. 2014; Adam et al. 2015).

Despite the bivalent phrasing of our question, these are not opposite, mutually exclusive alternatives with equal complexity and consequences. Firstly, top-down control is far more complex than its alternative. It involves four separate steps to cause benthic change: (1) fishing reduces parrotfish abundance, (2) reducing grazing pressure, (3) increasing cover of macroalgae, (4) decreasing coral cover (algae outcompetes adult corals for space or inhibits coral recruitment) (Hughes 1994; Bellwood et al. 2004; Mumby et al. 2006; Bonaldo et al. 2014). Bottom-up control simply implies that if habitat changes, fish abundance changes. Secondly, the consequences of altering top-down processes are usually considered more dire than those of altering bottom-up processes (Bellwood et al. 2004; Mumby et al. 2006; Hughes et al. 2007, 2010; Bruno et al. 2009). In bottom-up control, effects of environmental disturbances to the benthos (e.g. typhoons, coral bleaching) are generally assumed to be reversible. As coral recovers after the disturbance, the fish are presumed to recover accordingly (Wilson et al. 2006, 2008; Pratchett et al. 2008; Graham et al. 2015). In top-down control, the change in benthic habitat is often perceived as a “phase shift” from coral to macroalgal dominance. This shift is proposed to be stable, as long as parrotfish abundance is low (Hughes 1994; Bellwood et al. 2004; Mumby et al. 2006; Bonaldo et al. 2014; Adam et al. 2015). There is considerable debate as to whether recovery of parrotfish abundance, for example in No-Take marine reserves (NTMRs), will reverse such benthic phase shifts (Aaronson and Precht 2006; Mumby et al. 2006; Hughes et al. 2007, 2010; Graham et al. 2015; McClanahan et al. 2011a; Cheal et al. 2013; Toth et al. 2014). Thirdly, the alternatives are not strictly mutually exclusive. Both may occur at different places and/or times, depending on the circumstances (e.g. Smith et al. 2010), and the dichotomy between the two forms of regulation is somewhat artificial (Hunter and Price 1992). Furthermore, overfishing is suggested to reduce resilience of coral reefs to other environmental disturbances (Mumby et al. 2006; Hughes et al. 2010; Scheffer et al. 2015), providing a potential interaction between the top-down and bottom-up effects.

In the past 20 years, it has become almost paradigmatic in the coral reef literature that fishing-induced reductions of parrotfish abundance cause benthic phase shifts from coral to macroalgal dominance (Hughes 1994; Bellwood et al. 2004; Mumby et al. 2006; Bonaldo et al. 2014). Mumby and Steneck (2008) highlight reduced herbivory on coral reefs, mostly caused by fishing, as a rapidly evolving ecological paradigm. The classic example of this paradigm comes from Jamaica where overfishing of parrotfishes and declines of herbivorous urchins due to a disease led to a benthic phase shift from coral to macroalgal dominance (Hughes 1994).

In many tropical regions, parrotfish are important fishery targets, particularly in developing nations (Edwards et al. 2013). Increased fishing pressure has resulted in widespread declines of these fishes in the last few decades (Comeros-Raynal et al. 2012; Edwards et al. 2013). Thus, there is concern that the loss of parrotfishes that can potentially exert top-down control on primary producers will eventually result in detrimental macroalgal-dominated states (Bellwood et al. 2004; Hughes et al. 2007, 2010; Mumby and Steneck 2008; Adam et al. 2015).

An extensive review of the status and trends of Caribbean coral reefs from 1970 to 2012 (Jackson et al. 2014) concluded that overfishing of herbivores, particularly parrotfish, was a major driver of decline of coral cover in the Caribbean. This view is supported by research showing that a Caribbean NTMR had higher biomass of, and grazing rates by, large parrotfish, resulting in lower cover of macroalgae and enhanced coral cover in the NTMR relative to fished areas (Mumby et al. 2006; Mumby and Harborne 2010). Jackson et al. (2014) recommended reduction of fishing, including the possibility of total fishing bans for parrotfish. An extensive review of Indian Ocean coral reef fisheries suggested that fishing below a threshold proportion of herbivorous fish in the fishable biomass had major consequences to coral reef ecosystems (McClanahan et al. 2011a). Some authors thus advocate NTMRs as a potential means of increasing cover of hard corals (e.g. Mumby et al. 2006; Mumby and Harborne 2010; Selig and Bruno 2010).

Other research suggests that overfishing of herbivorous fish, causing a shift from coral to algal dominance on Caribbean coral reefs (Aronson and Precht 2006, Bruno et al. 2009; Cote et al. 2013; Toth et al. 2014; Adam et al. 2015) and coral reefs generally (Bruno et al. 2009; Cheal et al. 2010; Carassou et al. 2013), might be an over-simplification. A global review of the incidence of benthic phase shifts from coral to algal dominance (Bruno et al. 2009) concluded that such events were still relatively rare, particularly on Indo-West Pacific coral reefs. Carassou et al. (2013) concluded that NTMRs did not affect either coral recovery or macroalgal development following major environmental disturbances. Cheal et al. (2010) pointed out that the prevailing view that coral–macroalgal phase shifts commonly occur due to insufficient grazing by fishes is based largely on correlation with overfishing rather than on long-term quantitative field experiments. In addition, an emerging literature stresses the need for greater understanding of the dietary composition and nutrient utilization of marine herbivorous fishes and how such factors affect grazing impacts on coral reefs (Choat et al. 2002, 2004; Crossman et al. 2005; Clements et al. 2009; Cheal et al. 2012). Thus, there is some debate about the regional applicability of the concept that top-down, fishing-induced reductions of parrotfish result in benthic phase shifts from coral to algal dominance (Roff and Mumby 2012), and a dearth of long-term (decadal-scale) field studies of the parrotfish–coral interaction.

Philippine coral reefs are heavily fished (Alcala and Russ 2002; Newton et al. 2007). Parrotfish are targeted, mostly by subsistence fishers using traps, spear and gill nets, and often account for 5–10 % of reef fish yields in the country (Alcala and Russ 2002). Parrotfish often have significantly higher densities in Philippine NTMRs (Russ and Alcala 1998; Stockwell et al. 2009). In addition to fishing pressure, Philippine coral reefs have been subjected to major coral bleaching and typhoon events causing substantial coral loss in recent decades (Arceo et al. 2001). Environmental disturbances that cause loss of coral cover and habitat complexity affect density, species richness and assemblage structure of coral reef fishes significantly and usually negatively (reviewed by Wilson et al. 2006, 2008; Pratchett et al. 2008). Thus, Philippine coral reefs are excellent candidates for testing the relative importance of top-down effects of parrotfish on benthos and bottom-up effects of benthos on parrotfish.

This study investigates the relative effects of top-down control of the benthos by parrotfish and bottom-up control of parrotfish by the benthos on coral reefs on four small offshore islands in the central Philippines. At two of these islands, Sumilon and Apo, parrotfish density and benthic habitat were monitored almost annually for 30 years (1983–2013) inside and outside NTMRs. During this time, many environmental disturbances [destructive fishing, coral bleaching, local storms, typhoons, crown-of-thorns-starfish (COTS) outbreaks] caused major reductions in live coral cover. The study was long enough to also document coral recovery after disturbances, and subsequent responses of parrotfish to this coral recovery. As such, this study represents a unique “natural experiment” over considerable timescales. Specifically, we address the following questions:

  1. 1.

    What is the evidence for top-down control of the benthos by parrotfish density?

  2. 2.

    What is the evidence for bottom-up control of parrotfish density by the benthos?

Materials and methods

Study sites, history of protection and disturbances

This study was conducted at four islands located in the central Philippines: Apo, Mantigue, Selinog and Sumilon (Fig. 1). Each island had a community-managed NTMR that had good compliance throughout the study, with the exception of brief non-compliance events at Sumilon during the 1980s and 1990s (see Electronic Supplementary Material—ESM). Each NTMR and a paired fished (control) site were monitored for reef fish and benthos. All islands were affected by the coral bleaching event of 1998, and Apo Reserve was heavily damaged by typhoons in 2011 and 2012. All of the islands surveyed had been subject to intensive fishing before implementation of NTMRs (Alcala and Russ 2006). Some islands had been subjected to fishing with explosives and drive-net fishing (muro ami). Sumilon Reserve was pulse fished (1984–1986 with explosives and drive nets; 1992–1994 without explosives and drive nets) and received only partial protection in 1995–1999 (see ESM). Conversely, Sumilon Non-Reserve (fished site) had full protection from fishing in 1987–1991 and partial protection from fishing in 2009–2013. Apo and Sumilon islands were surveyed for reef fish and benthos almost annually 25 times in 30 years during the period 1983–2013 (see ESM). Selinog and Mantigue were surveyed 5 and 4 times, respectively, during the period 1999–2007 (see ESM).

Fig. 1
figure 1

Map showing location of four small offshore islands in the central Philippines (Apo, Mantigue, Selinog and Sumilon - all four study islands fall within latitude 8.5–9.5 degrees North and longitude 123–125 degrees East)

Full size image

Surveys of parrotfish

Underwater visual censuses (UVC) were performed on the reef slope (3–17 m) to estimate reef fish density, with six 50 m × 20 m (1000 m2) replicates in each of the eight sites (four NTMRs, four control sites) per sampling year. The methods of UVC, the observer (GRR), and the positions of all replicates were the same throughout the entire sampling period (1983–2013). All surveys were performed between November and December of each year. Density estimates were made for 20 species of parrotfish (Family Labridae; tribe Scarinae; Table 1) at all eight study sites. All species surveyed were vulnerable to and targeted by fish traps, spearfishing and gill nets at the four islands. Length estimates of individual fish were not made, and juveniles (<10 cm TL) were not counted. Maximum body sizes vary within species, but range from 30 to 50 cm TL in most species.

Table 1 List of 20 species of parrotfish surveyed at Apo, Sumilon, Mantigue and Selinog sites
Full size table

At Sumilon and Apo, parrotfish were placed into log4 abundance categories (Category 1 = 1 fish, Cat. 2 = 2–4, Cat. 3 = 5–16, Cat. 4 = 17–64, Cat. 5 = 65–256 fish 1000 m−2) to quantify fish densities from 1983 to 1998 (Russ and Alcala 1998). From 1999 onwards, actual counts (no. 1000 m−2) were made (including at Mantigue and Selinog). To obtain estimates of actual numbers of fish per replicate from log abundance categories for the period 1983–1998 (15 years), the modal number of fish per abundance category was estimated from actual count data collected from 1999 to 2013 (15 years). The frequency of each actual count was placed into each log abundance category (1–5) and displayed as frequency distributions. The mode of each distribution was taken as the best estimate of actual density.

Surveys of benthos

For the period of 1983–1999, benthic habitat surveys were conducted using the line-intercept method and were performed in the same areas as fish surveys and immediately after the fish surveys. Percent benthic cover was estimated at 20-cm intervals along a 50-m transect tape, with six to nine 50-m transects sampled per site. During benthic line-intercept transects, eight benthic categories were recorded: (1) branching and table coral; (2) massive and encrusting coral; (3) soft coral (SC); (4) rubble (R); (5) sand (S); (6) hard dead substratum (HDS); (7) algae; and (8) other.

Benthic cover from 1999 onwards was estimated within each 50 × 20 m transect used to census fish, and immediately after the fish survey. Each transect was divided into ten 10 × 10 m quadrats, and percent cover of each of the eight benthic categories was estimated by eye to the nearest 5 %. Benthic cover was then averaged across all 10 quadrats, providing the percent cover in each 50 × 20 m transect (Russ et al. 2005). Benthic categories 1 and 2 (see above) were combined to make a live hard coral (HC) category. Benthic categories 4, 5 and 6 were combined to make a dead substratum (DS) category. In some years, only fish data (not benthic cover data) were collected (1988, 1990, 1991 and 1992) in Apo Reserve and Non-Reserve and Sumilon Non-Reserve. These years were excluded from any statistical analyses.

Analysis of data

To investigate the effect of fishing on parrotfish density, the effect of parrotfish density on benthic cover [live hard coral (HC)] and the effect of benthic cover on parrotfish density at each of the four islands, generalized linear mixed modelling (GLMM) was performed in R using the lme4 and MuMIn packages (R Development Core Team 2012). Total parrotfish density (all 20 species combined) was related to reserve status (NTMR or fished), time and cover of HC, soft coral (SC) and DS, and cover of HC was related to reserve status (NTMR or fished), time and parrotfish density, at each of the four islands separately, using GLMM with a Poisson error distribution and a log link function. Demonstration of a fishing effect on parrotfish density or HC cover requires a significant NTMR Status × Time interaction plus an increasing density of parrotfish in the NTMR relative to the fished control over time, or a decreasing HC cover in the control relative to the NTMR over time. The GLMM approach was able to account for the non-independence of replicate transects and non-normal error distribution characteristic of count data and percent cover data. Model selection was based on minimization of the Akaike information criterion corrected for small sample sizes (AICc). The top three models for each variate (HC or parrotfish density) based on AICc values were examined, changes in AICc with respect to the top-ranked model noted (∆AICc) and AICc weights recorded (wAICc). Significance values and parameter estimates were presented for the top-ranked model for each variate at each of the four islands. Second- and third-ranked models were also included for variables that were highly significant but not included in top-ranked models.

Percentage cover of HC and total parrotfish density (no. 1000 m−2) were each plotted against duration of NTMR protection (in years) for Apo, Sumilon, Mantigue and Selinog NTMR and fished control sites. Given its complex history, Sumilon was the only island where duration of protection did not correspond to chronological time (see ESM). NTMR protection at Sumilon Reserve was applied and removed on multiple occasions. Parrotfish density estimates at Sumilon Reserve were not included in the plots of duration of protection against parrotfish density for 1985, 1992 and 1993. In all 3 years, NTMR protection had just ceased (after 10 years of protection in 1985 and 5 years of protection in 1992–1993) and parrotfish density had not declined to levels typical of zero years of NTMR protection. For Sumilon Non-Reserve, years where the site was closed to fishing (1987–1991, 2009–2013) were also excluded from the plots. Substantial changes were observed in parrotfish densities in 2012 and 2013 (post-typhoons) in Apo Reserve, and these data were also removed from the NTMR effect plots at this island to obtain a more realistic long-term estimate of the NTMR effect on parrotfish density over time. Since modal points of log4 abundance categories were used to estimate parrotfish density prior to 1999 for Sumilon and Apo, standard errors could not be estimated for this time period and were displayed only from 1999 onwards. Trends in cover of HC and parrotfish density were fitted with polynomials to emphasize temporal relationships between NTMR and fished sites over time.

Total parrotfish density and benthic cover were plotted against chronological time (in years) at the two long-term study islands—Apo and Sumilon (NTMRs and control sites). Due to a general lack of significant results of benthos on parrotfish density at Mantigue Island (Table 2), and the relatively short timescales of monitoring, temporal trends of benthic and fish variables are not presented in detail for Mantigue and Selinog islands. Both benthic and fish variables were fitted with polynomials to display long-term temporal trends. Dead substratum (DS: dead coral rock, sand, rubble), hard coral (HC: branching, massive and encrusting combined) and soft coral (SC) cover were selected as key benthic habitat variables and displayed with changes of parrotfish density in NTMR and control sites. Macroalgal cover (MA) was very low to non-existent at all eight study sites. It was included in the graphs of parrotfish density and live hard coral (HC) cover at Sumilon and Apo, to emphasize its lack of relationship with parrotfish density or cover of HC, but was not included in statistical analyses.

Table 2 Parameter estimates for the top-ranked generalized linear mixed-effects models (GLMMs)
Full size table

To highlight major changes in benthic cover and parrotfish density over time, specific time periods were selected which corresponded to either major environmental disturbances (human-induced or natural) or recovery of the benthos at Apo and Sumilon Islands. To display these patterns, a standardized index of percent change was calculated for each of these events and included the variables total parrotfish density, density of the 5 most abundant species (Scarus dimidiatus, Chlorurus bleekeri, C. spilurus, S. niger and S. tricolor) and cover of key benthic variables (HC, DS). This was done by firstly adding a constant value of 2 to both initial and final values for the specific time period to produce a set of non-negative data and standardized magnitudes of change. Fish variables were then log10 transformed to account for the overly abundant species, and benthic variables were arc-sine transformed. Transformed values were then used to calculate the standardized percentage change using the following formula:

$${\text{Standardized index of percent change}} = {\text{final}}-{\text{initial/initial value.}}$$

Changes in benthic cover and fish density were then represented in bar plots for each individual environmental disturbance.

Results

Fishing effects on parrotfish density

Significant NTMR × Time interactions were detected in the GLMMs for parrotfish density (all islands except Selinog, Table 2). Temporal trends in density of parrotfish were consistent with an NTMR effect at all islands except Apo (Fig. 2). At Sumilon, density in the NTMR increased, to be up to 3-times that in the fished site, during the first 8 years of NTMR protection, converged with the fished site at years 9–11 of protection, and then diverged again to be 2 times higher than the fished site at years 12–14 of protection (Fig. 2). Density appeared to increase in the NTMR relative to the fished site at both Mantigue and Selinog islands (Fig. 2), but a statistically significant interaction between NTMR status and time was detected only at Mantigue (Table 2). The significant interaction at Mantigue likely reflects similar parrotfish densities at the NTMR and fished sites before NTMR implementation, but densities consistently ~2 times higher in the NTMR than the fished site after implementation (Fig. 2). Similarly, at Selinog, parrotfish density was similar at the NTMR and fished sites at 0.5 years of protection and was consistently ~2 times higher in the NTMR site for almost a decade (Fig. 2). Collectively, data in Fig. 2 and Table 2 are consistent with significant fishing effects on parrotfish density at Sumilon and Mantigue Islands, a possible (but not statistically significant) effect of fishing at Selinog, but no effect at Apo.

Fig. 2
figure 2

Total parrotfish density (no. 1000 m−2) inside NTMRs (black dots, solid lines) and at fished control sites (white diamonds, broken lines) versus duration of protection (years) at Apo, Sumilon, Selinog and Mantigue Islands. Polynomial regressions are fitted to data from both reserve and fished control sites to show temporal trends. Error bars are standard errors

Full size image

Effects of parrotfish density on benthos

If fishing can affect density of parrotfish at some islands, does this affect cover of HC or MA? It should be noted that, a priori, fishing is far more likely to affect fish than benthos, since parrotfish are targeted by fishing, the benthos is not. Significant NTMR × Time interactions were detected in the GLMMs for cover of HC at all islands (Table 2). A benthic phase-shift scenario would predict that in fished areas, fishing-induced reductions in density of parrotfish would increase cover of MA and thus decrease cover of HC. However, cover of MA never differed between NTMR and fished sites at any island, since cover of MA was invariably zero (Fig. 3). Furthermore, a decrease of HC cover in the fished control site relative to the NTMR over time was seen only at Mantigue Island (Fig. 3). The divergence of the NTMR and control (fished) lines for HC cover at Mantigue (Fig. 3) was due mostly to HC cover declining in the Mantigue fished site in 2003 due to a crown-of-thorns-starfish (COTS) outbreak that affected only the fished site. Thus, no islands showed any clear indication of a fishing effect on cover of HC (Fig. 3). Furthermore, at three of four islands parrotfish density had no significant effect on cover of HC (Table 2).

Fig. 3
figure 3

Mean cover (%) of live hard coral (HC) inside NTMRs (black dots, solid lines) and at fished control sites (white diamonds, broken lines) versus duration of protection (years) at Apo, Sumilon, Selinog and Mantigue Islands. Mean cover (%) of macroalgae (grey squares, dotted lines) for NTMR and fished sites (single line, no difference between sites) at each island included for comparison with coral cover. Polynomial regressions are fitted to coral data from both reserve and fished control sites to show temporal trends. Error bars are standard errors

Full size image

Effects of benthos on parrotfish density

Examination of the long-term temporal trends in parrotfish density and cover of HC, DS, MA and SC at the two long-term study islands of Apo and Sumilon (Figs. 4, 5) suggest a generally negative relationship of parrotfish density with cover of HC (and a generally positive relationship with cover of DS), the exact opposite of that predicted by a classic benthic phase shift on coral reefs. To test for bottom-up control, we examined how benthos may affect parrotfish density. Cover of dead substrata (DS), hard coral (HC) and soft coral (SC) affected parrotfish density significantly at all islands except Mantigue, where only cover of HC had a significant positive effect (Table 2). At Apo (Figs. 4, 6) density of parrotfish displayed a strong positive relationship with cover of dead substratum (DS: dead coral rock, rubble, sand; Fig. 4a, d; Table 2), a negative relationship with cover of live hard coral (HC) (Fig. 4b, e; Table 2) and a positive relationship with cover of soft coral (SC) (Fig. 4c, f; Table 2). At Sumilon (Figs. 5, 7), density of parrotfish appeared positively related to cover of DS (Fig. 5a, d), but a delay in the fish response to change in cover of DS in the first but not the second 15 years at Sumilon Reserve (Fig. 5a) resulted in a negative statistical relationship (Table 2). At Sumilon (Figs. 5, 7), a negative relationship of parrotfish density with cover of both live hard coral (HC) (Fig. 5b; Table 2) and soft coral (SC) (Fig. 5c, f; Table 2) was clear. There was a significant, relatively short-term positive effect of hard coral cover on parrotfish density at Selinog, and to a lesser extent at Mantigue (Figs. 2, 3; Table 2). The strong positive relationship of parrotfish density with cover of DS was particularly clear when major environmental disturbances affected the benthos by killing corals (and thus increasing cover of DS) at Apo and Sumilon (Figs. 6, 7). Such disturbances included coral bleaching, crown-of-thorns-starfish (COTS) outbreaks, destructive fishing with explosives and drive nets, local storms and typhoons.

Fig. 4
figure 4

Temporal variations in mean total parrotfish density and cover of dead substrata (hard dead substrata, rubble and sand combined) (a, d); hard coral cover (branching, tabular; massive and encrusting combined) and macroalgal cover (b, e); soft coral cover (c, f) at Apo Reserve and Non-Reserve from 1983 to 2013. Black triangles represent total parrotfish density (no. 1000 m−2). White or grey markers represent % cover of benthic variables [white square: dead substrata (a, d); white circle hard coral and grey circle macroalgae (b, e); white diamond soft coral (c, f)]. Polynomial regressions are fitted to benthic (dashed or dotted lines) and fish (solid lines) data. Under the x-axis, the white arrows indicate the 1998 coral bleaching event, and the grey arrow indicates the typhoon event in Apo Reserve

Full size image
Fig. 5
figure 5

Temporal variations in mean total parrotfish density and cover of dead substrata (hard dead substrata; rubble; sand) (a, d) hard coral cover (branching, tabular; massive and encrusting) and macroalgal cover (b, e) soft coral cover (c, f) at Sumilon Reserve and Non-Reserve from 1983 to 2013. Black triangles represent total parrotfish density (no. 1000 m−2). White or grey markers represent % cover of benthic variables [white square dead substrata (a, d); white circle hard coral and grey circle macroalgae (b, e); white diamond soft coral (c, f)]. Polynomial regressions are fitted to benthic (dashed or dotted lines) and fish (solid lines) data. Under the x-axis, the black arrows indicate destructive fishing, white arrows indicate the 1998 bleaching, and the grey arrows indicate the typhoon event

Full size image
Fig. 6
figure 6

Average response of fish density (grey bars) to environmental disturbances that changed benthic cover (white bars; DS dead substratum, HC hard coral) at Apo reserve and Apo Non-Reserve, presented on a standardized index of % change (−1 = 100 % decrease, +1 = 100 % increase)

Full size image
Fig. 7
figure 7

Average response of fish density (grey bars) to environmental disturbances that changed benthic cover (white bars; DS dead substratum, HC hard coral) at Sumilon Reserve and Sumilon Non-Reserve, presented on a standardized index of % change (−1 = 100 % decrease, +1 = 100 % increase)

Full size image

At Apo Reserve a long-term (1983–1995) increase in cover of HC following implementation of NTMR protection resulted in long-term (1983–1995) declines in cover of dead substratum (DS) and density of all parrotfish (Figs. 4a, b, 6a). Over the same period, there were declines in the density of four of the five most abundant species: Scarus dimidiatus and S. niger (both scrapers) Chlorurus bleekeri and C. spilurus (both excavators) (Fig. 6a). The severe coral bleaching event in 1998 reduced cover of HC and increased cover of DS substantially in Apo Reserve (Figs. 4a, b, 6b). This led to increased density of parrotfish and density of most of the common species over the next 4 years (1997–2001) (Figs. 4a, b, 6b). Cover of HC increased, and cover of DS declined, from 1999 to 2008 in Apo Reserve, resulting in a decrease in density of all parrotfish and most common species except the scraper S. dimidiatus (Figs. 4a, b, 6c). A severe local storm in Apo Reserve, coupled with a coral bleaching event in 2010, reduced cover of HC, increased cover of DS and increased density of parrotfish, including all common species (Figs. 4a, b, 6d). The two typhoons that hit Apo Reserve in 2011 and 2012 reduced cover of HC and increased cover of DS and density of parrotfish, including all common species, substantially (Figs. 4a, b, 6e). There was no clear relationship between parrotfish density and cover of SC at Apo Reserve, except for a clear negative relationship in 2012–2013 due to the typhoons (Fig. 4c). The long-term (1997–2013) replacement of SC with HC at Apo Non-Reserve reduced the cover of DS and resulted in a long-term decline in density of all parrotfish except C. bleekeri (Figs. 4d–f, 6f). This dramatic increase in hard coral cover over 15 years at the Apo Non-Reserve followed the 1998 coral bleaching event. This bleaching event reduced soft coral cover, which had previously dominated the substratum at this site for 15 years (1983–1997—see Fig. 4e, f), opening up space for rapid colonization of branching Acropora corals.

At Sumilon Reserve parrotfish density was also positively related to cover of DS (with a delayed response of fish to cover of DS in the first 15 years at Sumilon Reserve—Fig. 5a) and negatively related to cover of HC and SC (Fig. 5b, c). Use of explosives and drive nets to fish Sumilon Reserve (1983–1985) reduced cover of HC and increased cover of DS and parrotfish density (1983–1991; Figs. 5a, b, 7a). This increase in parrotfish density from 1983 to 1991 was due to both the increased availability of dead substratum and the re-closure of Sumilon Reserve to fishing for 5 years from 1987 to 1991 (Fig. 5a, ESM). Recovery of the HC from this event (1985–1994) resulted in strong declines in cover of DS and parrotfish density (1991–1997; Figs. 5a, b, 7b). The decline in parrotfish density between 1991 and 1994 was also affected strongly by the reopening of Sumilon Reserve to fishing at this time (Fig. 5a, ESM). In Sumilon Reserve, the coral bleaching and COTS events in 1998 contributed to a decline in cover of HC from 1994 to 2001 and increased cover of DS and parrotfish density (1998–2006; Figs. 5a, b, 7c). Recovery of the HC cover from these environmental disturbances (2006–2013) resulted in declines in cover of DS and parrotfish density (2006–2011; Figs. 5a, b, 7d). What is particularly notable about these relationships at Sumilon Reserve is the temporal delay in response of density of parrotfish to changes in cover of HC and DS in the first 15 years. The response of the fish sometimes occurred more than 5 years after the change in benthos (Fig. 5a–c). Parrotfish density appeared to have a strong negative relationship with cover of SC at Sumilon (Fig. 5c, f). At the Sumilon Non-Reserve, there was a suggestion that, over the long term, parrotfish density was positively related to cover of DS and negatively related to cover of SC (Fig. 5d, f). Unlike previous results, in 2012–2013, a strong decline in cover of HC and an increase in cover of DS (particularly sand cover with low structural complexity in this case) at Sumilon Non-Reserve, caused by the effects of a typhoon in 2012, resulted in a sharp decrease in density of parrotfish (Fig. 5d, e).

The responses of both scrapers and excavators to changes in the benthos were generally the same (Figs. 6, 7). That is, as cover of HC declined and DS increased, the most common scrapers and excavators tended to increase in density in both the short (1–2 years) and longer (5–8 years) term (Figs. 6, 7). The reverse was observed as cover of HC slowly increased and cover of DS decreased (Figs. 6, 7), with decreases in density of both scraping and excavating feeding guilds occurring over the long term (5–15 years).

Discussion

We found no evidence for top-down control of the benthos by parrotfish density, but strong evidence for bottom-up control of parrotfish density by the benthos. Why is the evidence for top-down control so limited? Fishing reduced parrotfish density, sometimes to half or a third that in NTMRs. These fishing-induced reductions in parrotfish density presumably reduced grazing rates on the benthos. However, macroalgal cover never increased, nor coral cover decreased, in fished relative to NTMR sites. Thus there was no long-term evidence for fishing-induced phase shifts from coral to macroalgal dominance. These results may seem surprising, given that Philippine coral reefs are classified as overfished (Alcala and Russ 2002; Newton et al. 2007). Below we provide a brief overview of top-down/bottom-up theory and then evaluate evidence for and against such forms of regulation of the parrotfish–coral relationship.

Top-down and bottom-up regulation of population numbers and community structure

The concepts of top-down and bottom-up regulation of population numbers and community structure derive from food web theory (Hairston et al. 1960; Hunter and Price 1992; Menge 2000). Regulation is assumed to be by predators (top-down) or resources (bottom-up). Resources are usually trophic (food), often driven by availability of nutrients and light. In this study, the resource regulating local parrotfish density was benthic habitat, specifically cover of hard dead substratum, sand and rubble. Such habitats are a proxy for food availability since they are favoured feeding substrata of parrotfish (Choat 1991). “Bottom-up” in this study thus has both a figurative and a very literal meaning. There is consensus that the top-down/bottom-up dichotomy is somewhat artificial, with both likely operating in species-specific ways at different times/places (Hunter and Price 1992). Environmental variations (e.g. productivity gradients or environmental “stresses”) modulate the relative strengths of top-down/bottom-up drivers on a case by case basis (Hunter and Price 1992). In the present study, environmental disturbances (e.g. typhoons, coral bleaching) modulated availability of the benthic resource and created a strong bottom-up driver of parrotfish density. Note also that our study documents regulation of “local” density, not size of whole populations, the latter the focus of most theory.

Top-down, bottom-up regulation and the parrotfish–coral relationship

Top-down regulation of the benthos by parrotfish is more complex and indirect than bottom-up regulation. Top-down regulation requires four steps: fishing reduces parrotfish abundance, reducing grazing, increasing macroalgal cover and decreasing coral cover (Hughes 1994; Bellwood et al. 2004; Mumby et al. 2006). Fishing and grazing are both top-down processes involving trophic interactions. However, macroalgal abundance may increase due to reduction of grazing (top-down) or increased availability of nutrients, space or macroalgal propagules (bottom-up). Displacement of corals by algae also depends on competitive interactions among corals and algae (bottom-up) (McCook et al. 2001). Bottom-up regulation of the parrotfish by benthos is far simpler. Any environmental disturbance that kills corals directly can create a suitable benthic and/or feeding habitat for parrotfish, increasing local abundance of parrotfish that persists until coral recovers.

How does top-down or bottom-up regulation of the parrotfish–coral relationship vary from place to place and why? The main causes of such variation are likely to be differences in fishing intensity on parrotfish, feeding biology and functional composition of parrotfish, characteristics of the coral reef benthos (types of coral, availability of macroalgae, coral-algal competitive relationships) and the type, frequency and intensity of environmental disturbances to the coral reef benthos.

Evidence for and against top-down regulation of the benthos by parrotfish

The highest profile case of fishing-induced reductions in herbivore density leading to benthic phase shifts came from Jamaican coral reefs (Hughes 1994). The level of overfishing of parrotfish was not quantified, but a recent review by Jackson et al. (2014) makes a good case for heavy fishing pressure on parrotfish and shifts from coral to algal dominance on many Caribbean coral reefs. This is supported by research in a Bahamian NTMR which had higher biomass of, and grazing rates by, large parrotfish, lower cover of macroalgae and enhanced coral cover and coral recruitment than in fished areas (Mumby et al. 2006; Mumby and Harborne 2010). The latter are, however, single-point-in-time or short-term studies. Adam et al. (2015) provide an extensive overview of the evidence for and against fishing-induced trophic cascades and benthic phase shifts on Caribbean coral reefs.

Some evidence for fishing-induced trophic cascades and benthic phase shifts also comes from Indo-Pacific coral reefs. The extensive studies of NTMRs in Kenya (e.g. McClanahan 1994, 2014, McClanahan et al. 2007) demonstrated increases in biomass of herbivorous fish (including parrotfish) and decreases in abundance of urchins over almost 40 years of NTMR protection. These studies reported lower cover of fleshy erect algae in NTMRs than in fished areas (Babcock et al. 2010) or, more recently, no long-term trends in cover of fleshy algae or corals in NTMRs over 40 years (McClanahan 2014). Rasher et al. (2013), in a one-point-in-time study, reported higher biomass of herbivorous fish (including parrotfish), higher grazing rates, lower cover of macroalgae and higher coral cover in two Fijian NTMRs than in fished sites. Contrary to the present study, fishing-induced reductions of parrotfish were found to influence the benthos of Philippine inshore coastal coral reefs (Stockwell et al. 2009). Parrotfish biomass was higher, cover of macroalgae lower, but coral cover the same, in NTMRs compared to fished sites. This may be due in part to more intense fishing pressure inshore, as densities of fishers are higher on inshore than offshore reefs in the central Philippines (Abesamis pers. comm.).

In other places, fishing of parrotfish cannot be a major contributor to benthic change and coral reef decline. In the Florida Keys, the Great Barrier Reef (GBR) and New Caledonia, large-scale degradation of coral reefs has occurred without fishing of parrotfish reducing herbivory (e.g. De’ath et al. 2012; Cheal et al. 2010, 2013; Carassou et al. 2013; Toth et al. 2014). Decadal-scale NTMR protection in the Florida Keys had little effect on either parrotfish density (Bohnsack et al. 2009; Smith et al. 2011) or coral cover (Toth et al. 2014). In Belize, NTMR protection did not affect coral or algal cover (McClanahan et al. 2011b). Globally, differences in cover and recovery rate of hard corals between NTMRs and fished sites are generally small (Selig and Bruno 2010), and evidence for benthic phase shifts from coral to algal dominance is limited, and confined mostly to the Caribbean (Bruno et al. 2009).

The reversal of benthic phase shifts caused by recovery of parrotfish abundance in NTMRs, or by other means, may be problematic if parrotfish do not eat macroalgae (Choat et al. 2002, 2004; Clements et al. 2009; Adam et al. 2015). Two rare long-term examples of the reversal of a benthic phase shift from macroalgal to coral dominance provide contrasting results. Such a reversal occurred independent of the abundance of scraping and excavating parrotfish on inshore coral reefs of the GBR (Cheal et al. 2010). Contrarily, recovery from algal to coral dominance in the Seychelles (following coral bleaching) was more likely to occur at locations where densities of herbivorous fish were high and nutrients low (Graham et al. 2015). Whether recovery of parrotfish abundance can reverse benthic phase shifts will depend substantially on what parrotfish eat. We agree with calls for greater understanding of the dietary composition of, and nutrient extraction by, marine herbivorous fishes and how these factors affect the impact of such fish on coral reefs (Choat et al. 2002, 2004; Crossman et al. 2005; Clements et al. 2009; Cheal et al. 2012).

The parrotfish–coral relationship in the Caribbean and Indo-Pacific

Evidence for wide-spread benthic phase shifts on coral reefs caused by overfishing of parrotfish comes predominantly from the Caribbean and is far less common on Indo-Pacific coral reefs (Bruno et al. 2009; Roff and Mumby 2012; Carassou et al. 2013; Bonaldo et al. 2014; Adam et al. 2015). There are clear differences in taxonomic and functional composition of parrotfish on coral reefs of the Caribbean and Indo-Pacific (Bonaldo et al. 2014). Caribbean taxa are dominated by the “seagrass” or Sparisoma clade (formerly Sparisomatinae), with relatively higher proportions of browsers that feed on seagrass and macroalgae, while Indo-Pacific taxa are dominated by the “reef” or Scarus clade (formerly Scarinae), with relatively higher proportions of grazers and excavators (Bonaldo et al. 2014). Furthermore, Caribbean coral reefs tend to have higher cover of seagrass and sponges, and lower diversity and cover of scleractinian corals, relative to Indo-Pacific coral reefs (Spalding et al. 2001; Bonaldo et al. 2014). These differences in functional composition of the parrotfish and in benthic cover may well affect the relative importance of top-down and bottom-up relationships between parrotfish and benthos in these regions (Bruggemann et al. 1994; Burkepile and Hay 2008; Bonaldo et al. 2014). The corollary of this is that care should be taken when generalizing parrotfish–coral relationships between regions (see also Roff and Mumby (2012); Bonaldo et al. 2014).

Evidence for bottom-up regulation of parrotfish by the benthos and how it is modulated by environmental disturbances

In this study, bottom-up regulation of local parrotfish density was modulated by environmental disturbances to the benthos. Environmental disturbances to coral reefs can be both acute, such as typhoons (De’ath et al. 2012), coral bleaching (Hoegh-Guldberg et al. 2007; Graham et al. 2006; De’ath et al. 2012), crown-of-thorns-starfish (COTS) outbreaks (De’ath et al. 2012) or chronic, such as increased nutrients and sediments (Fabricius 2005), coral diseases (Hoegh-Guldberg et al. 2007) and prolonged fishing pressure (Edwards et al. 2013). A consensus about causes of coral reef degradation globally now emphasizes multiple drivers of benthic change (Aronson and Precht 2006; Hoegh-Guldberg et al. 2007; Mumby and Steneck 2008; Bruno et al. 2009; Hughes et al. 2010; Toth et al. 2014), particularly cumulative impacts of multiple stressors over long periods of time (Scheffer et al. 2015), as opposed to reductions in herbivory being the principal cause (Hughes 1994; Mumby et al. 2006; Jackson et al. 2014).

Differences in water quality and availability of macroalgal propagules (bottom-up effects) may partly explain the virtual lack of any macroalgae on our offshore study reefs and explain why our results differ from those of Stockwell et al. (2009) on inshore Philippine coral reefs. Inshore coastal reefs of the central Philippines generally have higher sediment and nutrient loads than offshore reefs (Hodgson and Dixon 2000; Kamp-Neilsen et al. 2002) due mainly to the effects of forestry, agriculture and urbanization, features common to many of the world’s coastal coral reefs (Fabricius 2005). Furthermore, our four study islands were 2–15 km from the nearest mainland island, distances that may limit availability of propagules of some macroalgae (McCook et al. 2001).

The lack of evidence for top-down control of benthos by parrotfish in this study contrasts with the compelling evidence for bottom-up control of parrotfish density by the benthos. Parrotfish generally have strong associations with particular coral reef zones and benthic habitats (Russ 1984; Choat 1991; Cheal et al. 2012; Tzadik and Appeldoorn 2013). Coral reef benthos has a very strong effect on coral reef fish, with coral loss generally having negative effects on reef fish assemblages (Wilson et al. 2006, 2008; Pratchett et al. 2008; Graham et al. 2015). However, some reef fish appear to benefit from coral loss (e.g. parrotfish, wrasses and goatfish—Wilson et al. 2006, 2008; Pratchett et al. 2008; Emslie et al. 2008, 2014; Russ et al. 2015), because they often feed preferentially over dead coral substrata, rubble and sand. This was clearly the case in this study, with parrotfish density having a strong positive relationship with cover of dead substrata and a strong negative relationship with cover of hard corals.

Increases in parrotfish density following decreases in hard coral cover likely occurred via three mechanisms, sometimes working together. In order of speed of response, these were (1) migration of adults, (2) settlement of recruits and (3) settlement and subsequent growth of recruits into disturbed areas. Part of the rapid increase in density at Apo Reserve following the typhoons in 2011 and 2012 may have been due to adult parrotfish migrating off the shallow reef flat onto the reef slope where the surveys were conducted, that is, by mechanism one. However, many of the increases in parrotfish density occurred over periods of 5–8 years, often with a delay of ~5 years after decreases in hard coral cover (e.g. see Fig. 5a). These slow, long-term increases of parrotfish density suggest that mechanisms two and three were the most likely causes. Both Wilson et al. (2006) and Pratchett et al. (2008) speculated that rapid (months to a few years) increases in herbivorous fish in areas where coral cover had been reduced substantially were more likely due to migration than to real population increases. The long-term nature of our study suggests that the increases in parrotfish density did result in real, but local, population increases, as opposed to temporary migration effects. Arrival of adult or recruit parrotfish into areas where coral cover has declined will depend upon the proximity of, and degree of connectivity with, areas that already have parrotfish. Thus, rates of increase of parrotfish density following disturbances may be higher in areas closer to nursery habitats and higher for contiguous (e.g. coastal fringing reefs) than for isolated habitats and reefs (e.g. atolls and remote islands).

Declines in parrotfish density also occurred gradually, often over 5–15 years, as hard coral cover increased slowly. These declines in density of parrotfish may be due to a decrease in availability of suitable feeding habitat—hard dead substratum, sand and rubble—causing fish to move to more suitable habitats. That changes in the benthos drove changes in parrotfish density is supported on practical and statistical grounds. Virtually none of the species of parrotfish eat corals directly and those that do (B. muricatum, C. bicolor and C. microrhinos—see Table 1) were rare in this study. Thus, no mechanism exists to cause dead substratum to increase because parrotfish density increased. On the other hand, most species of parrotfish in this study feed on algal turfs, detritus and sediments, all common on dead substrata, so a clear mechanism exists to cause parrotfish density to increase as cover of dead substrata increases. Statistical analyses support this since parrotfish density rarely affected hard coral cover significantly, whereas parrotfish density was significantly and negatively affected by hard coral cover and significantly and positively affected by cover of dead substratum at Sumilon and Apo. The significant positive effect of hard coral cover on parrotfish density at Selinog, and to a lesser extent at Mantigue, resulted from fish recovering in the NTMRs from 1999 onwards at the same time as hard coral cover was recovering after the 1998 coral bleaching event.

Conclusions

This study produced no evidence for top-down control of the benthos by parrotfish density, but compelling evidence for bottom-up control of parrotfish density by the benthos. These results are not consistent with an almost paradigmatic view in the coral reef literature that fishing-induced reductions of parrotfish abundance cause benthic phase shifts from coral to macroalgal dominance (Hughes 1994; Bellwood et al. 2004; Mumby and Steneck 2008; Jackson et al. 2014). In recent decades, the cover of hard corals on coral reefs has declined globally due to a broad range of environmental disturbances (Carpenter et al. 2008; De’ath et al. 2012, Jackson et al. 2014), only one of which is overfishing. The top-down, overfishing of parrotfish hypothesis is an indirect effect on the benthos, requiring unequivocal demonstration that fishing reduces parrotfish density and grazing rates, thus increasing macroalgal cover and decreasing coral cover. When scientists observe changes from coral to macroalgal dominance on coral reefs that appear to persist over time, rather than simply inferring that the “fishing-induced, trophic cascade benthic phase-shift” sequence of events has occurred, they should do two things. Firstly, attempt to demonstrate all steps in the sequence above, unequivocally. Secondly, consider multiple drivers of benthic change, not just reduced herbivory (see also Mumby and Steneck 2008; Hughes et al. 2010; Roff and Mumby 2012; Adam et al. 2015; Scheffer et al. 2015). Other drivers of benthic change on coral reefs may include macroalgal productivity (driven by nutrients) and availability, or environmental disturbances (e.g. typhoons, coral bleaching) that kill corals directly. What is clear from this study is that when such environmental disturbances reduce coral cover, they often increase local parrotfish density.

Soon after his discussion with Euthyphro, Socrates was found guilty of atheism and of corrupting the minds of Athenian youth, and sentenced to death. We trust that our colleagues in coral reef science are a little more compassionate with us. Our point here is simple. Top-down control of benthos by parrotfish no doubt occurs, but we should not be too quick to apply this “paradigm” to all coral reefs at all times. On the contrary, when we do observe changes from coral to macroalgal dominance on coral reefs, we should consider the possibility of multiple drivers of benthic change, not just reduced herbivory.