Philosophical Transactions of the Royal Society B: Biological Sciences
You have accessResearch article

Identifying multiple coral reef regimes and their drivers across the Hawaiian archipelago

Jean-Baptiste Jouffray

Jean-Baptiste Jouffray

Global Economic Dynamics and the Biosphere Academy Programme, Royal Swedish Academy of Sciences, PO Box 50005, Stockholm 104 05, Sweden

Stockholm Resilience Centre, Stockholm University, Stockholm 106 91, Sweden

[email protected]

Google Scholar

Find this author on PubMed

,
Magnus Nyström

Magnus Nyström

Stockholm Resilience Centre, Stockholm University, Stockholm 106 91, Sweden

Google Scholar

Find this author on PubMed

,
Albert V. Norström

Albert V. Norström

Stockholm Resilience Centre, Stockholm University, Stockholm 106 91, Sweden

Google Scholar

Find this author on PubMed

,
Ivor D. Williams

Ivor D. Williams

Coral Reef Ecosystem Division (CRED), Pacific Islands Fisheries Science Center (PIFSC), National Marine Fisheries Service, NOAA, 1125B Ala Moana Boulevard, Honolulu, HI 96814, USA

Google Scholar

Find this author on PubMed

,
Lisa M. Wedding

Lisa M. Wedding

Center for Ocean Solutions, Stanford University, Stanford Woods Institute for the Environment, 99 Pacific Street, Suite 555E, Monterey, CA 93940, USA

Google Scholar

Find this author on PubMed

,
John N. Kittinger

John N. Kittinger

Center for Ocean Solutions, Stanford University, Stanford Woods Institute for the Environment, 99 Pacific Street, Suite 555E, Monterey, CA 93940, USA

Conservation International, Betty and Gordon Moore Center for Science and Oceans, Honolulu, HI, USA

Google Scholar

Find this author on PubMed

and
Gareth J. Williams

Gareth J. Williams

Scripps Institution of Oceanography, University of California San Diego, 9500 Gillman Drive, La Jolla, CA 92093, USA

Google Scholar

Find this author on PubMed

    Abstract

    Loss of coral reef resilience can lead to dramatic changes in benthic structure, often called regime shifts, which significantly alter ecosystem processes and functioning. In the face of global change and increasing direct human impacts, there is an urgent need to anticipate and prevent undesirable regime shifts and, conversely, to reverse shifts in already degraded reef systems. Such challenges require a better understanding of the human and natural drivers that support or undermine different reef regimes. The Hawaiian archipelago extends across a wide gradient of natural and anthropogenic conditions and provides us a unique opportunity to investigate the relationships between multiple reef regimes, their dynamics and potential drivers. We applied a combination of exploratory ordination methods and inferential statistics to one of the most comprehensive coral reef datasets available in order to detect, visualize and define potential multiple ecosystem regimes. This study demonstrates the existence of three distinct reef regimes dominated by hard corals, turf algae or macroalgae. Results from boosted regression trees show nonlinear patterns among predictors that help to explain the occurrence of these regimes, and highlight herbivore biomass as the key driver in addition to effluent, latitude and depth.

    1. Introduction

    The dramatic loss of live coral cover on reefs worldwide [13] has raised serious concerns regarding their future [4,5]. Parallel to this loss, observational [68], experimental [9,10] and modelling [11] studies suggest that many coral reefs are shifting to alternative regimes (or states) with consequences for coastal societies that depend on reef resources. While transitions from coral to macroalgae dominance are most commonly described [6,7], other degraded states, characterized by sponge, soft coral or corallimorpharian dominance, have been suggested [12].

    Regime shifts in coral reefs have primarily been described in human-dominated environments [13,14] where overfishing and reduced water quality, acting in concert with climate change, have been suggested as key drivers [10,15]. Specifically, the loss of herbivores, which keep algal colonization and growth in check, has been argued to be a leading cause of regime shifts [2,7], particularly when large tracts of substratum become open for rapidly colonizing algae, e.g. following hurricanes or coral mass bleaching events [10,16,17].

    Yet, recent meta-analyses of longitudinal datasets have questioned the existence (or the stability) and the generality of alternative reef regimes in coral reefs. For example, Bruno et al. [18] looked for multimodal patterns in the frequency distribution of benthic cover in 1851 reefs worldwide, and reported that most reefs were in neither a coral-dominated state (more than 50% coral cover) nor a macroalgae-dominated state (more than 50% macroalgal cover), challenging the assumption that coral–macroalgae shifts are a common phenomenon. Similarly, using stochastic semi-parametric modelling on reef data from the Caribbean, Kenya and Great Barrier Reef, Żychaluk et al. [19] found no evidence of bimodality at a regional scale. However, Mumby et al. [20] argued that both studies make a number of unrealistic statistical assumptions with regards to the constancy of environmental variables, resolution of field data and disturbance dynamics. Similarly, Hughes et al. [14] claimed that data used are often too patchy and unable to identify the complex mechanisms or processes causing long-term change. Moreover, they contested the cut-off set by Bruno et al. [18] (i.e. more than 50% cover), arguing that few reefs globally display such abundances of dominating benthic taxa.

    Diagnosing multiple regimes from field data is problematic, because coral reefs may respond slowly to frequent pulse perturbations [20] that mask trends of recovery or decline. Given the critical consequences of regime shifts on ecosystem services [5,21] and their profound management implications, we need a better understanding of the processes that drive alternative reef regimes, and improved methods to extract evidence of their existence from field data.

    Here, relying on an extensive spatial dataset gathered in the Hawaiian archipelago across gradients of natural and anthropogenic conditions, we apply a novel approach for detecting, visualizing and defining potential multiple ecosystem regimes. We also identify the primary human and natural drivers that help to explain these regimes across a regional scale.

    2. Material and methods

    (a) Study area

    The Hawaiian archipelago (Hawai‘i, USA) is one of the most isolated archipelagos in the world (figure 1). While the heavily populated Main Hawaiian Islands (MHI) are overfished [22] and subjected to a range of anthropogenic stressors [23], the Northwestern Hawaiian Islands (NWHI) receive minimal direct human impacts and are among the most protected coral reefs globally [24].

    Figure 1.

    Figure 1. Map of the study area showing the Hawaiian archipelago (courtesy of NOAA) and the location of the survey sites (purple dots). Within the Northwestern Hawaiian Islands (ad) 118 sites were surveyed, and 184 within the main Hawaiian Islands (e).

    (b) Organization of data

    Benthic and fish data were collected in 2010 from 302 reef sites across the Hawaiian archipelago by the Coral Reef Ecosystem Division, as part of NOAA's Pacific Reef Assessment and Monitoring Programme. (For sampling methodology, see the electronic supplementary material, S1.)

    Fish species were categorized into functional groups based on their trophic level (see the electronic supplementary material, S2), and herbivores were further divided into grazers, scrapers and browsers [25]. Grazers are fish that crop on algal turf, preventing the establishment and growth of macroalgae. Scrapers also feed on turf but they remove some component of the reef substratum, which provides bare areas for coral recruitment. Finally, browsers consistently feed on macroalgae and may play a crucial role for reversing macroalgae-dominated states [5,26]. Dietary information was collected from FishBase [27] and complemented from the literature [25] when higher-resolution data were required for herbivore classification. In cases where fish could be identified only to genus level, information on the diet of close relatives from the same genus was used. Total biomass per functional group was calculated from the biomass of individual fish obtained using the formula

    Display Formula
    where W is the weight in grams, TL the total length in millimetres, and a and b species-specific conversion parameters extracted from FishBase [27].

    In addition to benthic and fish data, a set of human-use and environmental GIS (Geographic Information Systems)-derived variables—human population density, distance to potential impact (e.g. shore, stream), effluent discharge and stream disturbance data (composite metrics of many land cover variables) [28]—was compiled for the Main Hawaiian Islands except for Lāna‘i and Ni‘ihau.

    (For methodology and detailed list of the variables, see the electronic supplementary material, S3.)

    (c) Data analyses

    All statistical analyses and graphical presentations were conducted using R v. 2.15.1 [29]. Specific packages used are referred to in the text or in the figure legends.

    (i) Identification of reef regimes and categorization of sites

    First, we replicated the method applied by Bruno et al. [18] to create a phase shift index (PSI) and graphically check for multimodality in the frequency distribution of benthic states. In essence, the PSI is the first component of a principal component analysis (PCA) based on the cover of coral and macroalgae.

    Second, a correlation-based PCA was performed using six benthic habitat variables: hard coral cover (Hcoral), macroalgae cover (MA), turf algae cover (TA), structural complexity estimate (complexity), sand cover on the reef surface (sand) and crustose coralline algae cover (CCA). Structural complexity was included because it is a key aspect of reef habitat quality [30], and data were standardized to account for variables measured at different scales. A hierarchical clustering of the variables was produced with the same Euclidean distance matrix as for the PCA using pvclust package v. 1.2-2 [31], and p-values computed by 10 000 multi-scale bootstrap resampling [32] were assigned to each cluster to indicate how strongly the cluster was supported by the data. A cluster was considered to be significant when the approximately unbiased p-value was greater than 0.95. Finally, building on the number of significant clusters obtained through the previous hierarchical analysis, a K-means partitional clustering process [33] was carried out to categorize the 302 sites with regards to their benthic habitat.

    (ii) Relative influence of natural and human variables on reef regimes

    To assess what key human and environmental variables were associated with different reef regimes, a boosted regression trees (BRTs) modelling technique [34] was performed using the gbm package v. 2.0-8 [35] and the gbm.step routine described by Elith et al. [36]. For the MHI (with the exception of Lāna‘i and Ni‘ihau), 147 sites were modelled simultaneously against 15 predictor variables, whereas for the NWHI, where human settings are absent, 118 sites were modelled against a set of seven continuous and categorical predictor variables (table 1). Pairwise relationships between all variables (no Spearman rank correlation coefficient was greater than |0.75|) showed no multicollinearity. The categorization of sites into different regimes was converted to presence–absence of each regime by survey site and analysed using a binomial distribution. Partial dependency plots were used to visualize and interpret the relationships between each predictor variable and the regime after accounting for the average effect of all other predictor variables in the model. (For BRT methodology and details on model optimization, see the electronic supplementary material, S4.)

    Table 1.Predictor variables used in the boosted regression trees analysis. MHI, main Hawaiian Islands; NWHI, Northwestern Hawaiian Islands.

    variable description range
    MHI GrazerBiom grazer biomass in g m−2 0–28.3
    ScraperBiom scraper biomass in g m−2 0–13.9
    BrowserBiom browser biomass in g m−2 0–22
    latitude latitude in decimal degrees (WGS 1984) 18.97–22.24
    depth depth in metres 2–29
    DistCoast distance in metres from nearest shoreline to survey location 14–2509
    DistStream distance in metres from nearest stream to survey location 96–20 130
    effluent the total effluent in million gallons per day (mgd) discharged within parcels that intersected with a 10 km buffer from survey location 600–7 177 840
    population human population density in a 10 km buffer from survey location based on 2010 census data 232–523 576
    UrbanIndex representation of urban disturbance standardized from 0 to 1 0–0.98
    PointIndex representation of the density of sources of point pollution standardized from 0 to 1 0–0.50
    FragIndex representation of stream fragmentation standardized from 0 to 1 0–0.31
    FormplIndex representation of lands that were formerly pineapple or sugarcane plantations standardized from 0 to 1 0–0.54
    DitchIndex representation of the relative density of ditch infrastructure standardized from 0 to 1 0–0.73
    AgrIndex representation of agricultural disturbance standardized from 0 to 1 0–0.65
    NWHI GrazerBiom grazer biomass in g m−2 0–29.6
    ScraperBiom scraper biomass in g m−2 0–30.8
    BrowserBiom browser biomass in g m−2 0–146.4
    LPredBiom large predator biomass in g m−2 0–2961.8
    latitude latitude in decimal degrees (WGS 1984) 23.63–28.45
    depth depth in metres 1–28
    ReefZone zones of the reef: fore reef—back reef—lagoon n.a.

    3. Results

    (a) Identification of reef regimes and categorization of sites

    The test for multimodality in the frequency distribution of living coral and macroalgae of the 302 Hawaiian sites (figure 2) displayed a normal distribution which is similar to the findings by Bruno et al. [18], i.e. an absence of multiple regimes. By contrast, the combination of analytical approaches used in this paper highlighted the existence of three primary regimes. The results were identical for the MHI and the NWHI and, therefore, only the pattern for the archipelago as a whole is shown in figure 3.

    Figure 2.

    Figure 2. Count histogram of the phase shift index (PSI) [18] of 302 reef sites in the Hawaiian archipelago.

    Figure 3.

    Figure 3. (a) Principal component analysis (PCA) diagram showing the spatial variation in benthic habitat of 302 sites in the Hawaiian archipelago along the first two principal components. Variables are plotted as vectors and dots represent sites. The smaller the angle between two variable vectors the stronger the correlation. Hcoral, hard coral cover; MA, macroalgae cover; CCA, crustose coralline algae cover; TA, turf algae cover; sand, sand cover; complexity, structural complexity. (b) Cluster dendrogram of the benthic variables from 302 sites, with p-values given as percentages. For a cluster with p > 95%, the hypothesis that ‘the cluster does not exist’ is rejected with significance level 0.05. (c) PCA diagram with an overlaying K-means clustering of the sites. Green squares (80 sites), orange circles (153 sites) and red triangles (69 sites) represent categorization of the sites matching the previous hierarchical grouping of the benthic variables. The ellipses encompass 80% of the dots associated with each cluster.

    The PCA using six benthic parameters showed a clear pattern of three distinct regimes. The first principal component axis (PC1) and the second principal component axis (PC2) accounted for 61.4% of the total variability in the data (figure 3a). PC1 described a gradient from high coral cover, high CCA cover and high structural complexity (at negative PC1 scores), to high macroalgae cover, high sand cover and low complexity (at positive PC1 scores). PC2 best explained the variability of turf algae with higher turf cover at negative PC2 scores.

    The hierarchical cluster analysis (figure 3b) confirmed the visual impression from the PCA with the benthic variables grouping into three significant clusters: CCA being associated with hard coral and structural complexity (p = 1), turf algae being closer to macroalgae and sand but a cluster by itself since outside of any significant cluster (p = 0.79) and macroalgae being closely associated with sand (p = 0.99).

    Finally, the K-means partitional clustering differentiated the sites along these three clusters with an overrepresentation of turf algae dominated sites (figure 3c): 153 sites (51%) were categorized as turf regime (sites dominated by turf algae), 80 sites (27%) as calcifying regime (high structural complexity sites dominated by CCA and hard coral) and 69 sites (23%) as macroalgal/sand regime (low structural complexity sites dominated by macroalgae and sand). (Mapping of the categorized sites per island and average values of the benthic variables within each regime is shown in the electronic supplementary material, S5.)

    (b) Relative influence of human and natural variables on reef regimes

    For each region, only the four most influential predictor variables were reported and illustrated (figure 4). Preliminary threshold values below are given only in cases where the shape of the fitted function best matches the distribution of the fitted values. (See the electronic supplementary material, S4 for the plots of the fitted values in relation to each predictor.)

    Figure 4.

    Figure 4. Partial dependency plots for the four most influential variables in the boosted regression trees analysis of three distinct benthic regimes (ac) in the MHI and the NWHI. Number of sites within each region and relative influence of each predictor are shown in parentheses. Photo credits: NOAA.

    (i) Calcifying regime

    In the MHI, a positive relationship with scraper biomass (22.3% relative influence) was the optimal predictor of the occurrence of structurally complex coral and CCA-dominated sites (figure 4a). Other important predictors were negative relationships with higher latitudes (14.6%) and increasing effluent (13.1%, drop around 1.0 × 106 million gallons per day (mgd)), and a positive correlation with stream fragmentation (10.7%).

    In the NWHI, three variables contributed most strongly to predicting calcifying regime occurrence (figure 4a): scraper biomass (34.1%, positively correlated), depth (30.5%, peaked between 12 and 22 m) and latitude (20.7%, occurrence decreased at higher latitudes).

    (ii) Turf regime

    In the MHI, grazer biomass (18.6%) and effluent discharge (17.9%) were the two optimal predictors of turf algae occurrence, with positive nonlinear relationships for both predictors displaying thresholds around 10 g m−2 and 1.5 × 106 mgd, respectively (figure 4b). The opposite pattern (i.e. negative correlation) was observed for stream fragmentation (14.5%) and distance to stream (12.9%).

    In the NWHI, depth (30.3%) contributed most to explain the occurrence of turf algae with a drop near 10 m (figure 4b). Grazer biomass (19.8%) and latitude (16.8%) were positively correlated, followed by a negative correlation with scraper biomass (11.9%).

    (iii) Macroalgal/sand regime

    In the MHI, three key predictor variables contributed to the occurrence of macroalgae-dominated sites (figure 4c). Macroalgal dominance displayed negative nonlinear relationships with increasing grazer (24%) and browser (15.5%) biomass. Thresholds seemed to occur around 5 and 2 g m−2, respectively. Human population density represented 14.8% of the relative influence and was positively correlated, whereas scraper biomass was negatively correlated (7.1%).

    In the NWHI, herbivore biomass was again the main predictor (figure 4c): occurrence of macroalgal/sand sites decreased as the biomass of grazers (29%, drop at 5 g m−2), scrapers (22.8%, drop at 4 g m−2) and browsers (12.3%) increased. In addition, latitude (13.7%) was positively correlated.

    4. Discussion

    Identifying existing regimes and understanding their drivers has great relevance for managers, policy-makers and planners seeking to protect ecosystem services generated by coral reefs [5,37]. Based on a combination of exploratory and inferential statistics, this study provides a novel method to identify multiple regimes, and offers strong evidence of three distinct reef regimes occurring across the Hawaiian archipelago—calcifying, turf algae and macroalgal/sand regimes. It further suggests that simply testing for bimodality in the frequency distribution of present reef state [18] does not constitute a sufficient test of multiple regime existence. Our results also contribute to a deeper understanding of the relative influence of different drivers that underpin the occurrence of reef regimes, supporting the idea that they are multi-causal, driven by a combination of biotic processes, abiotic conditions and human drivers.

    All three regimes occur both in the MHI and the NWHI, despite major differences in exposure to direct human impacts. This corroborates the findings of Vroom & Braun [38], who recorded high algal abundances in the NWHI and further stressed the necessity of re-evaluating the metrics used to gauge subtropical reef health. Our results also show that most reefs in this system are dominated by turf algae (51% of all sites). This raises the question whether turf-dominated reefs constitute a stable regime or an unstable transitional state that is moving towards a coral or macroalgal attractor [17]. It has been argued that turf eventually proceeds towards macroalgae dominance if herbivore density is low [39,40], but experimental studies suggest that turf can also constitute a stable regime in sediment-rich areas, because herbivory of epilithic algal turfs is suppressed under sediment-laden conditions [4143].

    In this study, the occurrence of all three regimes was strongly predicted, for both regions, by the biomass of herbivores, confirming the important role they play in coral reef dynamics and in mediating reef regime-shifts [2,44]. In addition, categorizing herbivores into finer-scale functional groups (i.e. grazers, scrapers and browsers) allows for a better understanding of what specific ecological functions are important in supporting different regimes. The occurrence of macroalgal/sand regimes showed a negative nonlinear relationship with the density of browsers that directly consume macroalgae, as well as with increasing biomass of grazers and scrapers that limit their growth. The calcifying regime was positively associated with the biomass of scrapers that provide area of clean substratum (feeding scars) for coral recruitment, whereas the turf regime was positively correlated with grazers that prevent the transition from turf to macroalgae through top-down control. This confirms numerous studies that show how higher herbivore abundances coincide with a lower cover of macroalgae on reefs [10,11]. However, our results further suggest that an increase in herbivore biomass could lead to different regime trajectories, depending on the functional make-up of the herbivorous assemblage. Specifically, if the herbivores are predominantly grazers, then the probability of turf dominance increases, while if scrapers are abundant (e.g. Chlorurus sordidus, Chlorurus perspicillatus and Scarus rubroviolaceus), then there is an increased chance of a reef shifting to a calcifying regime.

    Human drivers also influenced the distribution of different regimes. Population density has been used as a coarse proxy for overall human influence on coral reefs [45]. However, in recent studies, the relationships between population density and reef states appear ambiguous [46,47]. This study suggests that population density is a relatively poor predictor of reef regimes across the Hawaiian archipelago, which is overwhelmed by the influence of more refined proxies, particularly those for land-based pollution and land-use change (such as effluent discharge or stream fragmentation). In the MHI, the total amount of effluent discharged had a strong positive impact on turf regime, whereas being negatively correlated with the calcifying regime. McClanahan et al. [48] showed that an increase in phosphorus stimulated growth of filamentous and turf algae but not the growth of macroalgae. Overall, there is ample literature about the negative effects of nutrient pollution on corals and how it can lead to excessive algae growth [4951, but see 52], particularly in combination with loss of herbivory [53]. However, this negative impact on corals seemed countered by a positive correlation with fragmentation of streams. The rationale behind this unexpected positive effect of human-induced stream disturbance could be that higher fragmentation reduces the natural water flow, thus allowing more of the effluent to sink or to be deposited on the stream edges before reaching the reefs.

    Finally, physical drivers also play a key role in coral reef regimes in this system. Unlike the turf and macroalgal/sand regimes, the calcifying regime was negatively correlated with latitude. Low temperatures have been shown to limit coral growth [54] and, considering the latitudes the Hawaiian archipelago encompasses, this is likely to have an influence on the results [55]. Depth was also an influential abiotic parameter, although mostly significant in the NWHI. This divergence in relative influence between both regions is not clear but could be the result of increased water turbidity that limits light gradients in the MHI because of sediments. In the NWHI, we found the occurrence of calcifying regimes to peak at mid-depths, similar to results from Williams et al. [56]. However, gradients in other oceanographic variables that are known to be important in structuring benthic communities have been identified across the region [55]. For example, additional data on sedimentation [57] or wave exposure [56,58,59] will help investigators develop a better understanding of these important physical processes in influencing reef regimes.

    These analyses provide a promising approach for investigating multiple ecosystem regimes and the processes underpinning them, which could help guide effective management. However, because our study does not account for historic disturbance events (e.g. hurricanes, disease outbreaks, bleaching events), it only represents a snapshot in time (year 2010). Therefore, the stability (or at least the trajectory of development) of any specific reef regime remains unknown and needs to be further explored. From a management perspective, the identification of nonlinear patterns and preliminary thresholds among many predictor variables offers an interesting avenue for future studies to investigate more accurately the location of these tipping points, in order to provide tangible management targets for both proactive avoidance of potential regime shifts and restoration of degraded reefs.

    Data accessibility

    The dataset supporting this article has been made available and uploaded in Dryad (doi:10.5061/dryad.rg832).

    Acknowledgements

    This paper draws on preliminary work from a Master's thesis by Jean-Baptiste Jouffray [60] . We thank Dana Infante and Bob Whittier for providing the original human dimensions data, and two anonymous reviewers for valuable comments helping to improve this paper.

    Funding statement

    This work was partly funded by the Erling-Persson Family Foundation through Global Economic Dynamics and the Biosphere at the Royal Swedish Academy of Sciences. Mistra supported this research through a core grant to the Stockholm Resilience Centre.

    Footnotes

    One contribution of 16 to a Theme Issue ‘Marine regime shifts around the globe: theory, drivers and impacts’.

    References