Drivers of recovery and reassembly of coral reef communities

Abstract

Understanding processes that drive community recovery are needed to predict ecosystem trajectories and manage for impacts under increasing global threats. Yet, the quantification of community recovery in coral reefs has been challenging owing to a paucity of long-term ecological data and high frequency of disturbances. Here we investigate community re-assembly and the bio-physical drivers that determine the capacity of coral reefs to recover following the 1998 bleaching event, using long-term monitoring data across four habitats in Palau. Our study documents that the time needed for coral reefs to recover from bleaching disturbance to coral-dominated state in disturbance-free regimes is at least 9–12 years. Importantly, we show that reefs in two habitats achieve relative stability to a climax community state within that time frame. We then investigated the direct and indirect effects of drivers on the rate of recovery of four dominant coral groups using a structural equation modelling approach. While the rates of recovery differed among coral groups, we found that larval connectivity and juvenile coral density were prominent drivers of recovery for fast growing Acropora but not for the other three groups. Competitive algae and parrotfish had negative and positive effects on coral recovery in general, whereas wave exposure had variable effects related to coral morphology. Overall, the time needed for community re-assembly is habitat specific and drivers of recovery are taxa specific, considerations that require incorporation into planning for ecosystem management under climate change.

1. Introduction

Natural ecosystems are frequently impacted by acute disturbances, removing major habitat providers and releasing competition for space [1]. Following disturbance, entire communities may (i) enter alternate states [2], (ii) undergo succession until recovering to a state that is similar to the original pre-disturbance condition (figure 1) [4], or (iii) recover towards a state that is similar to pre-disturbance but dominated by a different community [5]. The fate of disturbed communities is complex because it is driven by processes interacting at multiple scales: population traits (demographic), within environments (ecological conditions), and through space (site characteristics, biogeography) [6] (figure 1).

Figure 1.

Conceptual diagram illustrating in the lower half, a coral reef (bottom right) impacted by an acute disturbance, transitioning (grey arrows) to a non-coral dominated reef (bottom left) [3] and recovering to a high coral state going through the phases of coral populations growth. Boxes are ecological processes influencing recovery rate. The upper half of the diagram represents the coral recovery structural equation model (SEM). Each path (direct or connected) ending with an arrow head represents a model component. Black circles show interactions among variables. Coral legacies represent the coral percentage cover after the disturbance.

During recovery, populations of habitat providers in terrestrial and marine ecosystems may transition through lag, growth and climax phases, following logistic growth (figure 1) [7]. A recovering and climax population reflects on both the characteristics of the disturbance and the biological legacies that remain [8]. Biological legacies include the species traits of survivors and the remaining abiotic structure, often aiding in the re-establishment of structural complexity through ecosystem engineering [9].

During lag and early growth phases, propagule dispersal can play a major role in population recovery, especially in marine systems [10]. The heterogeneity of disturbances often creates a mosaic of areas in different states, with more mature patches aiding recovery of the most impacted areas through connectivity of larval supply [11]. Areas with enhanced retention of propagules (sinks) are thought to have higher rates of initial recovery and overall resilience, but only a few studies have demonstrated that in marine systems or have focused on specific populations [12,13].

From the establishment of new recruits to the adult communities, facilitation, competition and predation processes shape community recovery trajectories [14]. After large disturbances in coral reefs for example, bare substrate is rapidly covered with early successional algae and microbial communities. Subsequent coral recolonization often depends on grazing activity that decreases fleshy algal cover and increases crustose coralline algal (CCA) cover—a facilitator of coral settlement [15]. The re-assembly of the community during the growth phase can also have a high degree of stochasticity as a result of different functional roles among species [16] and great variability in recruitment patterns [17]. Environmental characteristics where communities reside such as levels of wave exposure [18], rugosity [19], or water quality [20], also greatly influence recovery trajectories.

As populations recover, growth asymptotes towards a limit that defines the carrying capacity (K) of the community. For a system to be considered stable in its climax state, it has to remain the same when undisturbed [21]. It is challenging to detect prolonged community stability because of (i) too frequent disturbance regimes [22], and/or (ii) a lack of data with high taxonomic resolution to investigate shifts in species composition through time [23].

Recovery of disturbed coral reefs has been shown to occur over a few years to few decades [24,25] but is overall poorly studied owing to the frequent occurrence of disturbances, making reefs extremely dynamic systems [26]. However, in the absence of large disturbances over long periods, community stability can be explored to elucidate the time needed for a community to fully recover from different types of disturbances.

Throughout the Palau archipelago, nearly 50% of the live corals were totally bleached by the 1998 El-Niño mass coral bleaching event [27], and by 2001 more than 80% of 315 reefs around Palau only had 0–5% Acropora cover [28]. Palauan reefs then experienced 14 years without any major disturbance, until two category-5 (Simpson scale) typhoons inflicted an average 60% loss in live coral cover along its eastern outer reefs in 2012 and 2013 [29]. Between these major disturbances, there was a minor thermal stress event in 2010 that induced little mortality [30].

We took advantage of a unique long-term monitoring dataset from Palau, where reefs experienced ≥14 years free of mass disturbance following the 1998 bleaching event. We first assess the differences in benthic community structure and temporal change among distinct habitats. Second, we quantify coral community trajectories and investigate stability following recovery. Third, we calculate the instantaneous growth rate of the four most dominant coral groups in the study system and use this to assess recovery. Last, we use a structural equation modelling (SEM) approach to investigate the direct and indirect effects of the demographic, environmental, and spatial drivers that influenced the recovery rates of the four dominant coral groups.

2. Material and methods

(a). Study sites

The Palau International Coral Reef Centre (PICRC) long-term coral reef monitoring surveys began in 2002 and were conducted throughout the archipelago within four habitats: western outer reefs, eastern outer reefs, inner reefs, and patch reefs. Through time the number of sites increased in all habitats (electronic supplementary material, table S1). Since 2006, surveys covered six sites in the eastern and western outer reefs, the inner reefs, and three sites in the patch reefs (figure 2). Surveys were conducted approximately every two years, unless a disturbance occurred.

Figure 2.

Map of Palau with the location of long-term coral reef monitoring sites in each reef habitat.

(b). Reef assessment methodology

Coral reef monitoring was conducted as previously described in Golbuu et al. [31]. Five 50 m transects were placed haphazardly at the global positioning system location of each site and depth (3 m and 10 m), following the depth contour of the reef, leaving ≥3 m between transects. Along each transect, data on benthic coverage, fish abundance and size, and juvenile coral density and size were recorded. Benthic cover was determined from underwater images using 40–50 images per transect. Benthic substrates underneath five randomly placed crosses per image were recorded by trained PICRC staff to calculate percentage cover of live corals and macroalgae to genus, sponges, ascidians, CCA, turf algae and non-living substrate. Juvenile coral density was defined as colonies ≤5 cm in maximum diameter, and surveyed along the first 0.3 m × 10 m of each transect, to genus or family. The abundance of commercially targeted fishes, including all parrotfish, was recorded within a 5 m wide belt along each transect. A proxy for herbivory was extracted based on Scarinae abundances as this was the herbivorous fish group that was consistently surveyed over time. Abundances were used instead of biomass to avoid errors associated with fish size estimations owing to changes in observers.

(c). Other environmental variables

The coral-larval oceanographic model of Golbuu et al. [13] for Palau was used to extract long-term averaged rates of coral self-seeding and larval imports for each study site and is detailed in the electronic supplementary material. The variables represent the relative difference in larval density among study sites, whether larvae are self-seeded or imported from elsewhere and are used in the SEM analyses described below.

Mean wave exposure at each site was calculated from wind speed, fetch distance to the reef and angle of exposure to the wind using geographical information system methods detailed in Houk et al. [32]. Satellite derived surface wind data were downloaded from the National Oceanic and Atmospheric Administration (NOAA) ERD THREDDS data server from 1999 to 2016 (http://ferret.pmel.noaa.gov/uaf/las/UI.vm). Differences in relative wave energy between sites and observation intervals were estimated based on surface wind data from the preceding two years at each time point.

Fine-scale reef rugosity was measured at each site using a 2 m chain (1 cm chain link) every 10 m along each transect in survey year 2007, giving values ranging from approximately 1 m (high rugosity) to 2 m (low rugosity). Values were inversely scaled (low to high) for ease of interpretation.

(d). Data analyses

To estimate trends in the cover of benthic groups through time for each habitat and depth, corals were grouped into six major groups that together represent greater than 75% of coral cover: Acropora spp., Montipora spp., Porites spp., Merulinidae spp., Pocillopora spp., Agariciidae spp. (only for inner reefs), and ‘other corals’. Other substrata were grouped into four broad categories, including macroalgae, turf and CCA, rubble and sand, and other invertebrates. A general additive mixed effects model (GAMM) was performed to analyse the effect of time, habitat and depth on the coverage of these benthic groups. Spline smoothing was applied to time, with depth, habitat, and their interactions as fixed factors. Sites were incorporated as a random term nested within habitats.

As major differences were found among habitats and depths across time for all benthic groups (see Results section: ‘Changes in benthic communities’), multivariate analyses were performed for each habitat and depth combination separately. To visualize coral community changes through time within each habitat by depth combination, non-metric multidimensional scaling (nMDS) was conducted on averaged log + 1 transformed Bray–Curtis dissimilarity matrices of coral cover at the highest taxonomic level. Only coral genera with a cover greater than 5% within each habitat, depth and time point were included to avoid misinterpretation of ordination bias towards zero values. To visualize which coral taxon correlated with community changes, correlation vectors were overlaid on nMDS plots using the ‘envifit’ function. To investigate coral-community stability, log + 1 transformed coral community matrices were compared through time, using permutational multivariate analyses of variance (adonis test). Permutations were constrained within the group ‘site’ to account for the hierarchical structure of the dataset [33]. For any significant main effects, pairwise comparison among distinct time intervals were then conducted [34]. Communities were assumed to have reached stability when no significant differences were found between at least two subsequent time points or more through time.

(e). Rates of coral recovery

To assess the direct and indirect effects of demographic, spatial and environmental processes on coral recovery rates, instantaneous growth rate (IGR) [7], was used as a proxy for recovery of each of the major taxa (Acropora, Montipora, Porites and Merulinidae) for each site and depth. Under this model, coral populations are assumed to follow exponential growth (equation (2.1)) until density-dependence occurs, representing the growth phase in figure 1:

$$\mathrm{cove}{\mathrm{r}}_{t_a}=\mathrm{cove}{\mathrm{r}}_{t_0}\hspace{0.17em}{\mathrm{e}}^{\mathbf{I}\mathbf{G}\mathbf{R}.\mathrm{time}}.$$ 2.1

IGR for each coral taxon, at each site and depth was calculated using equation (2.2)

$$\mathrm{IGR}\hspace{0.17em}=\frac{{\ln ((\mathrm{cover}}_{t_a}+5)/({\mathrm{cover}}_{t_0}+5))}{t_a-t_0},$$ 2.2

where cover at t₀ is the coral cover post-bleaching disturbance and cover at t_a is either the cover from the period when there was no significant increase through time (assessed by t-tests), before the occurrence of another disturbance, or the last data point in the dataset (electronic supplementary material, figure S1). As ln(x) is only defined for x > 0, a value of 5% was added to coral cover to avoid very low or high IGR values when the cover was close or equal to zero.

In complex natural systems, various processes are interdependent and hierarchically structured and are known to influence recovery trajectories and the resilience of natural communities (figure 1) [4]. To test these, we developed an SEM to examine the causal effects of hypothesized pathways that influence rates of coral recovery using a suite of demographic, environmental, and spatial variables. Each path is explicitly depicted in figure 1 and detailed in the electronic supplementary material, and was formulated using known relationships from the literature and the results of our analyses of the individual system components.

CCA was not consistently categorized on benthic images and therefore could not be included. Turf algae was categorized only when appearing dense and visible on benthic images and was therefore included in the ‘algal cover’ category. Relative larval imports, relative larval self-seeding, and rugosity represent site characteristics, whereas all other continuous variables in the SEM were averages from the growth phase for each coral group at each site and depth. One SEM analysis per main coral group was conducted. All component models were linear models and response variables were checked for normality and log-transformed accordingly. Explanatory variables were scaled during the analytic procedure. Homogeneity of variance was checked visually using residuals plots. The goodness of fit of SEM was assessed using directional separation test (d-separations test) Fisher's C statistic [35].

All analyses were conducted in R v. 3.4.1 [36], using the vegan [37], gamm4 [38] and piecewiseSEM [35] packages.

3. Results

(a). Changes in benthic communities

In 2002, three years after bleaching, benthic community structure differed among habitats and depths (figure 3). Mean live coral cover was moderate on the western and eastern outer reefs (approx. 11.5% at 3 m and approx. 20% at 10 m), low on the patch reefs (7.5% at 3 m and 2.6% at 10 m), and relatively high on the inner reefs (approx. 45% at 3 m and 34% at 10 m).

Figure 3.

Temporal dynamics in benthic communities in the four habitats and two depths. n = 20 to 30 × 50 m transects for outer reefs, n = 10 to 30 for inner reefs, and n = 10 to 15 for patch reefs. The red star represents qualitative coral cover range estimation from 20 m depth to the shallows [31,39]. Black bars represent degree heating weeks (DHW) in 1 July–30 Sept 1998, 2010 and 2014 [40]. Black lines in the eastern outer reef habitat represent two typhoons occurrence (December 2012 and November 2013, [29]). Grey arrows show survey times (electronic supplementary material, table S1).

Coral community trajectories were distinct among habitats and depths (GAMM, p < 0.001; figure 3). Live coral cover gradually increased in all habitats through time but at different rates. In western outer reefs, cover started to asymptote after 2010 at approximately 30% (±2.8%) at 3 m, and approximately 50% (±1.7%) at 10 m. The eastern outer reefs had reached 35% (±2.5%) mean live coral cover by 2010, but this was reduced to approximately 6% (±1.6%) at both depths in 2013 because of the two typhoons. In the patch reefs, mean coral cover reached 31% (±5.3%) at 3 m and 24% (±3.5%) at 10 m in 2016. In inner reefs, mean live coral cover started to asymptote after 2008 at approximately 60% (±1.9%) cover at 3 m, and approximately 40% (±3.6%) at 10 m.

Mean macroalgal cover was continually low (less than 12%) on outer reef and inner reefs, but approximately 15% on the patch reefs at 10 m. Macroalgae significantly increased through time in the inner reefs (p < 0.01), albeit remaining less than 10%. Changes in turf algae, CCA and carbonate cover were tightly correlated with changes in coral cover. Cover of rubble and sand were stable in the inner and western outer reefs, increased to 35% in the eastern outer reefs following the typhoons, and was highest (greater than 50%) in the patch reefs.

Throughout time, coral genus richness was highest in the western outer reefs (22 recorded coral genera) and lowest in the patch reefs (16 genera), increasing linearly through time (electronic supplementary material, figure S2). Despite relatively high genus richness, coral communities were largely dominated at greater than 75% or more by Porites spp. (mostly massive and Porites rus), Merulinidae and Agariciidae in inner reefs; Acropora spp. (tabular and digitate morphologies), Montipora spp. (encrusting morphology), Porites spp. (massive and P. rus) and Merulinidae in outer reefs; and Acropora (branching morphology), Montipora spp., and Porites spp. in patch reefs.

(b). Re-assembly and stability of coral communities

Coral communities' re-assembly trajectories were visualized and investigated separately among habitats and depths. Within each habitat, communities at the two depths were different but followed similar re-assembly trajectories through time (figure 4). In the western outer reefs, communities transitioned through time to greater diversity and greater representation of Acropora, Favites, Favia and Goniastrea at both depths from 2010 to 2016 (figure 4a). Despite a small divergence in 2013, probably attributed to minor typhoon-related wave damage, coral communities did not statistically differ from 2010 onwards to 2016, and demonstrated community stability. In the eastern outer reefs, corals gradually transitioned to a more diverse community up to and including 2010, with increasing representation of Acropora, Echinopora, Montipora and several other coral genera at both depths. They did not exhibit evidence of stabilization prior to the typhoon disturbances (figure 4b). Post-typhoons, communities had very low live coral cover (5–7%) and did not exhibit assemblage stability by 2016. Patch reefs transitioned from 2002 to 2005 to communities mostly characterized by Fungiidae, Montipora and Porites corals, and from 2007 to 2016 by Acropora spp. (figure 4c). There were no significant differences in coral community structure from 2007 to 2016, indicating that even though coral cover was still gradually increasing, the community structure was relatively stable within this habitat. Inner reef communities changed from 2002 to 2007, and then stabilized with minimal changes from 2007 to 2016 (p > 0.05; figure 4d). There was a small divergence in 2010 (p < 0.05), probably attributed to the minor thermal stress event that may have affected Merulinidae corals within this habitat [30].

Figure 4.

Non-metric multidimensional scaling plots (nMDS) of coral community matrices within each habitat and depth. Overlaid vectors show coral taxa that correlate mostly with the change of communities. Photographs show coral reef communities in their most coral abundant state ((a) © Tane H. Sinclair-Taylor, (b–d) © PICRC).

(c). Coral taxa recovery

IGR differed among coral groups and habitats. Montipora had the highest IGR compared to all other coral groups (p < 0.05). Across the four coral groups, IGR in the western outer reefs was significantly higher than the IGR in the patch reefs (p < 0.05) but not in the eastern outer reefs and inner reefs (electronic supplementary material, figure S5). There was no difference in IGR between the two studied depths.

Findings from SEM analyses supported the hypothetic pathways inferred by our conceptual diagram (figure 1). For Acropora, Porites and Montipora SEMs, Fisher's C statistic p-values were greater than 0.15 suggesting that no important pathways among variables of the model framework were omitted. Merulinidae SEM showed a poor fit (Fisher's C = 95.96, p = 0.014; electronic supplementary material, figure S6) with no effects on the rate of recovery implying that significant drivers were not detected by our hypothetic model framework. All significant pathways are displayed in figure 5 and corresponding partial regression and bar plots are in the electronic supplementary material, figure S7.

Figure 5.

Coral recovery structural equation models (SEM) showing direct and indirect effects of coral demographics, spatial and environmental variables on the rate of recovery (IGR) of three coral groups. Continuous variables are averages during the recovery phase of each coral group. The thickness of paths is proportional to the given standardized path coefficients. Black and red arrows indicate positive and negative pathways, respectively. Non-significant pathways are not shown.

Acropora IGR was directly and positively affected by larval imports and juvenile density, yet negatively affected by increasing wave exposure. Indirect negative effects to IGR included increasing algal cover through its negative effect on juvenile coral density. Algal cover was higher at 10 m than 3 m and weakly positively related with increasing wave exposure. Available coral settlement substrate was negatively related to algae cover and rugosity. The western outer reefs had the fastest recovery rate owing to the highest level of larval imports, lowest cover of algae, and highest rugosity. Although the patch reef habitat had a high Acropora cover legacy and rugosity, it also had the highest cover of algae that indirectly reduced recovery.

Montipora IGR was directly and positively affected by increasing wave exposure and was greater at 10 m than 3 m depth. Juvenile coral density was lower at 10 m where there was less settlement substrate available, more algal cover, and low reef rugosity. The western outer reefs had higher larval imports, and the abundance of parrotfish was positively affected by the legacy cover of Montipora corals and rugosity.

Porites IGR was directly and negatively affected by algal cover, and indirectly affected by the availability of settlement substrate, which was negatively correlated with algal cover. Available substrate was affected positivity by parrotfish abundance, wave exposure and negatively by rugosity and depth (10 m). Inner reefs had the highest Porites cover legacy, the highest reef rugosity, the lowest cover of algae (together with western outer reefs), but the lowest abundance of parrotfish. Western and eastern outer reefs had the highest abundance of parrotfish.

Merulinidae SEM showed a poor fit as IGR was not affected by any variable (electronic supplementary material, figure S6). However, juvenile coral density was positively affected by the availability of substrate, which was higher on the outer reefs and at 3 m across all habitats. Algal cover was negatively affected by the abundance of parrotfish, which were found in higher abundance on reefs with high rugosity and the outer reefs.

4. Discussion

Disturbances to coral reefs are increasing worldwide and the occurrence of acute disturbances such as mass coral bleaching events are becoming more frequent, often leaving insufficient time for corals to fully recover before the next event [41]. Our analysis documents that the time needed for shallow coral communities to recover from bleaching disturbance to coral-dominated states is at least 9–12 years and only during periods free of major disturbances, as has been shown in other reefs worldwide [19,24,25,42]. This study is among the few to show that relative coral community stability in a climax phase can be achieved in the absence of major disturbance [43]. Palauan reefs were fortunate to not suffer from any major disturbance (mass bleaching events, storms, crown of thorns outbreaks) for more than a decade, in contrast to many other reefs worldwide [22], allowing a full investigation into the drivers of community recovery. With predicted increasing frequency of large-scale disturbance caused by global climate change, the full recovery of reefs to climax phases will be rare, apart from those locations that act as climate-refugia such as those predicted by Cacciapaglia & van Woesik [44]. Palau's reefs are listed as such. Our study is also among few to investigate the direct and indirect effects of drivers on coral recovery. Most previous work has focused on direct drivers [7,19], however, the complexity and hierarchy of ecological processes needs to be incorporated into investigations of diverse systems. With the use of SEM, we address these complexities and hierarchies, we show that differences in the rates of recovering of coral groups were predominantly explained by coral larval connectivity, juvenile coral density, algal coverage, parrotfish abundance, substrate characteristics and wave exposure. However, the strength of these relationships varied among coral taxa and habitats, highlighting the complexity of predicting community recovery trajectories in different reef settings.

(a). Coral community dynamics and stability

In the periods between disturbances on coral reefs, studies have shown that live coral cover can increase and communities re-assemble to pre-disturbance states [24] or have divergent recovery trajectories to different coral community states [42]. To the best of our knowledge, few studies have shown that recovered coral communities can persist in a relatively stable coral dominated state in disturbance free regimes [43]. Disturbances are common and coral reefs are often referred to as disturbance driven systems [26], hence communities are often in recovery phases. Moreover, the lack of long-term monitoring datasets has hindered broad-scale investigations. Our findings demonstrate that during a disturbance-free period, live coral cover gradually increased and communities re-assembled into stable climax states in two of the four studied habitats (western outer reefs and the inner reefs) within 9–12 years. In the absence of quantitative data at this spatial scale from before the bleaching disturbance, we are not able to verify that recovered communities were similar to pre-bleaching ones, however, they had similar total mean live coral cover to that observed qualitatively in 1992 [39]. The stability of coral communities within these two habitats suggest that potential chronic disturbances (e.g. overfishing, terrestrial run-off) may be occurring at low levels and/or managed adequately.

By contrast, recovery on the eastern outer reefs and patch reefs remained incomplete. In 2010, 12 years following mass bleaching, the live coral cover in the eastern outer reefs approached 30%. It is unknown whether coral communities fully reassembled because in 2012 the occurrence of two super typhoons decreased live coral cover from approximately 30% to approximately 6%. Three to four years after the typhoons, live coral cover was lower (5–7%) than three years after the bleaching disturbance (11–18%), implying that their recovery will be slower. Interestingly, patch reef communities appear to have reassembled by 2007, but total live coral cover is still slowly increasing. Patch reefs were the most severely impacted habitat by the mass bleaching (less than 8% coral cover in 2002), and had the highest macroalgal and loose substrate coverage. Lagoonal patch reefs are often silty and/or affected by terrestrial run-off and are easily accessible to fishing. The combination of these factors may have slowed down recovery within this habitat.

(b). Direct and indirect effects on coral recovery rates

During succession, community recovery rates differ greatly among natural systems and despite some overlap and redundancy of theories explaining system recovery, there is no universal theory that encompasses all concepts [4]. In this study, we made use of all available variables and examined their role in regulating taxa recovery while considering the complexity and hierarchical structure of processes occurring within the system. The application of IGR combined with SEM analyses addressed this complexity and identified key parameters that had direct and/or indirect roles in the rates of coral taxa recovery.

Acropora recovery was high at sites with high coral larval imports and high juvenile coral density, implying that coral reproduction and recruitment were the key processes replenishing Acropora populations following bleaching. Such direct effects are not always obvious as several other processes such as density-dependence, competition, and predation affect coral recruitment [45]. The broadcast spawning trait of this coral group allows for high dispersal of larvae [46], which were found in their highest concentrations along the western outer reefs, making this sub-region well-connected and highly resilient. High juvenile Acropora densities also imply that larval settlement and post-settlement growth and survival at these reefs is high. Our finding coincides with Golbuu et al. [13] who showed that high water retention played an important role in determining the density of Acropora juvenile corals in Palau, as well as other studies which demonstrate that enhanced larval supply [12] and recruitment [17] drive rapid habitat recovery. In addition, SEM detected that the densities of juvenile Acropora, Montipora and Merulinidae corals were negatively affected by high algal cover and low reef rugosity, and were generally more abundant where substrate availability was high, supporting empirical relationships found from previous studies in Micronesia [45] and other parts of the world [19,47].

Macroalgae and turf algae are known to have negative effects on coral settlement, recruitment and coral growth rates [48]. Many reefs worldwide have failed to recover and transition to macroalgal-dominated states when some ecological processes are disrupted (e.g. herbivory) or environmental driver degraded (e.g. water quality). In this study, in addition to the negative effect of algae on juvenile coral density, the recovery rate of Porites spp. was negatively affected by algal cover, and substrate availability was negatively related to algal cover. From recruits to adult colonies, corals compete for space [49]. Turf and macroalgae are some of the first benthic species to establish post-disturbance and are controlled largely by herbivores [50]. In Micronesia, Mumby et al. [51] showed that increasing acanthurid biomass was correlated with decreasing algal cover. While acanthurids were not consistently surveyed during monitoring surveys, our study showed positive and negative relationships between substrate availability and parrotfish abundance, and algal cover and parrotfish abundance, respectively. Despite our study limitations using abundance as opposed to biomass or bite rates as a proxy for herbivory [52], our findings are still highlighting the contribution of Scarinae for creating space for coral recruits to settle and to release algal competition for adult corals to recover, as also shown in other regions [53,54],

Reefs that are structurally complex have been shown to provide refugia for coral recruits from grazing, predation and overgrowth, increasing their survival [45]. Our results showed that reef rugosity played a direct role at partly explaining densities of Acropora juvenile corals. In addition, we found that parrotfish abundance positively related to reef rugosity. These findings coincide with the findings of Graham & Nash [55], whereby reefs with high structural complexity provide more refuges and habitat space for fishes. Thus, increased grazing activity at structurally complex reefs probably contributed to faster recovery through freeing space for coral settlement.

A key environmental driver of coral taxa recovery rates was the level of wave exposure, and the direction of the relationship contrasted between populations of Acropora (negative) and Montipora (positive). It is recognized that wave energy influences coral species zonation, coral growth, primary productivity, CCA distribution, and partitioning of herbivorous fish niches [56–59]. While coral morphology information was not available in our dataset, tabular and branching forms of Acropora corals are dominant in Palau [60], and are comparatively more fragile than other coral groups [61]. Mechanical stress from wave exposure affected their rates of recovery. By contrast, Montipora populations had fast recovery at sites with high levels of wave exposure, where they mostly occur as encrusting morphologies (M. Gouezo 2016, personal observation). In this case, fast coral recovery at turbulent sites may be the result of combined effects of (i) grazing by a group of herbivorous fishes, diminishing turf algae and facilitating CCA, aiding coral recruitment, (ii) low sediment deposition, and/or (iii) fast growth of encrusting corals.

Despite climate-mediated increases in the frequency and intensity of disturbances worldwide, this study provides empirical evidence that stability and persistence in coral-dominated states can occur, highlighting the resilience of these reefs and their future potential for acting as refugia [44]. Our study provides new evidence on the critical role of coral reproduction, larval connectivity, post settlement success and herbivory in coral recovery, and how these can differ among major functional groups. Physical characteristics of reef environments such as wave exposure and structural complexity also contributed to recovery dynamics. Developing such understanding on how biophysical drivers contributed to coral recovery is necessary information for managers to plan for maximizing recovery potential. Depending on the focal taxa, management options arising from this study include mapping and managing resilient reef areas based on their larval-connectivity level, structural complexity, and wave exposure and diminishing herbivory fishing pressures following future acute disturbances to facilitate faster recovery.

Data accessibility

Data are available via the Micronesian Challenge database: https://micronesiareefmonitoring.com.

Authors' contributions

M.G., K.F. and C.D. conceived the ideas and generated research questions. M.G., Y.G., D.O., G.M. and V.N. contributed to data collection, extraction and entry. E.W. and M.G. ran the coral-larvae dispersal model. M.G. and C.D. performed data analyses. K.F., Y.G., P.H., E.W. and C.D. provided useful insights to the interpretation of analysis and contributed to reviewing the manuscript critically. M.G. prepared the manuscript.

Competing interests

We declare we have no competing interests.

Funding

This study was supported by The David & Lucile Packard Foundation, NOAA Coral Reef Conservation Program, NOAA Coastal Oceans Programs, and Coral Reef Initiative Grant (US Department of Interior).