Three decades of coral reef community dynamics in St. John, USVI: a contrast of scleractinians and octocorals

Transcription

1 Three decades of coral reef community dynamics in St. John, USVI: a contrast of scleractinians and octocorals GEORGIOS TSOUNIS AND PETER J. EDMUNDS Department of Biology, California State University, Nordhoff Street, Northridge, California USA Citation: Tsounis, G., and P. J. Edmunds Three decades of coral reef community dynamics in St. John, USVI: a contrast of scleractinians and octocorals. Ecosphere 8(1):e /ecs Abstract. To better understand phase shifts on Caribbean reefs, we quantified community structure on shallow reefs over 27 yr in St. John, U.S. Virgin Islands, contrasted the community dynamics of scleractinians and octocorals, and evaluated the extent to which community structure was associated with rainfall, temperature, and hurricanes. To gain insight into the likely abundance of octocorals on future reefs with low scleractinian cover, we compared two sites dominated by the major Caribbean reef-building coral Orbicella annularis. Between 1987 and 2013, scleractinian cover declined from 45% to 6% at Yawzi Point, but remained at ~30% at Tektite. We compared changes in community structure using four benthic assemblage constructs scleractinian-focused (cover of scleractinians, macroalgae, and CTB i.e. crustose coralline algae, algal turf, and bare space), octocoral-focused (abundance of octocorals, cover of macroalgae, and CTB), octocoral genera (abundance by genus), and a complete approach (all taxa) to reveal how a consideration of octocoral abundance influenced the interpretation of coral reef community dynamics. Overall, temporal variation in community structure differed among the four assemblage constructs at both sites and was associated with rainfall and mean seawater temperature. These results suggest that: (1) scleractinian- and octocoral-focused communities in the same location responded differentially to the same environmental conditions, (2) their communities generally were influenced more by the chronic effects of rainfall and temperature than acute effects of storms, and (3) octocoral-focused communities were more resilient to environmental conditions than scleractinian-focused communities. With further declines in cover of scleractinians, octocoral communities are likely to become more common throughout the Caribbean. Key words: bleaching; Caribbean; conservation; environmental factors; gorgonians; phase shifts; rainfall; resilience; soft corals; temperature. Received 12 May 2016; revised 9 November 2016; accepted 14 November Corresponding Editor: Hunter S. Lenihan. Copyright: 2017 Tsounis and Edmunds. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. georgios.tsounis@csun.edu INTRODUCTION Many Caribbean coral reefs provide examples of complex communities that have undergone dramatic changes over decades (Jackson et al. 2014) that have been defined by declines in cover of scleractinians (Hughes 1994, Gardner et al. 2003), and reductions in population size of macroscopic vertebrates (Jackson et al. 2014). These changes have been caused by a variety of natural and anthropogenic disturbances, but for scleractinians the most profound large-scale threats arise from global climate change and ocean acidification (Hoegh-Guldberg 2011). In the Caribbean, the degradation of benthic reef communities has largely been characterized by a transition from high cover of scleractinians to low-profile coral rubble, macroalgae, algal turf, and sometimes other invertebrates including octocorals (McClanahan and Muthiga 1998, 1 January 2017 Volume 8(1) Article e01646

2 Pandolfi et al. 2005, Alvarez-Filip et al. 2011, Ruzicka et al. 2013). Collectively, the changes affecting Caribbean reefs over the last 40+ years provide examples of phase changes in community structure (Done 1992), the best-known of which involves a transition in spatial dominance from scleractinians to macroalgae (Hughes 1994, Mumby et al. 2007, Jackson et al. 2014). Although much is known about the roles of disturbances in mediating these transitions (Hughes et al. 2007, Roff and Mumby 2012), there remains a need to evaluate how disturbances have affected other benthic taxa (e.g., Graham et al. 2014). Studies of taxa other than scleractinians on coral reefs are important, because they have the potential to support a broader interpretation of the notion of ecological winners (or losers) on contemporary reefs (sensu Loya et al. 2001), and can provide insight into the community structure that is likely to prevail in the future. As it is unknown whether the selective environment on future reefs will favor the taxa that are winning on contemporary reefs (Van Woesik et al. 2011, Brown and Phongsuwan 2012), it remains possible that novel taxa will emerge as ecologically important community members. Amidst efforts to understand the factors favoring declining cover of scleractinians (Riegl et al. 2009, Roff and Mumby 2012), octocorals have received little consideration. Yet octocorals are successful on tropical reefs (Cary 1918, Goldberg 1973, Kinzie 1973, Lasker and Coffroth 1983, Fabricius and Alderslade 2001), where they are important components of fringing, back reef, and outer reef habitats to at least 72 m depth (Kinzie 1973), and can create dense canopies of branches serving as habitat for many organisms (Goldberg 1973, Kinzie 1973). In contrast to the recent history of scleractinians in the Caribbean (Gardner et al. 2003, Edmunds 2014, Jackson et al. 2014), three studies have reported stable (or increasing) population sizes of octocorals in the Florida Keys and St. John, U.S. Virgin Islands (Ruzicka et al. 2013, Lenz et al. 2015, Edmunds and Lasker 2016). The causes of these trends remain unclear, but upward trajectories of changing abundance suggest that octocorals are more resistant and more resilient to select contemporary perturbations than scleractinians (sensu Gunderson 2000). Contextualized by the aforementioned trends, the objective of this study was to evaluate the effects of disturbance on coral reef community structure in the Caribbean from a more taxonomically inclusive perspective than can be obtained from studying scleractinians and macroalgae. We focused on two reefs in St. John, U.S. Virgin Islands, which have shown different trends in stony coral cover since At one site, Tektite, coral cover remained similar at the start and end of the study (2013), while at the other site, Yawzi Point, coral cover declined from 45% to 6% (Edmunds 2013). Based on the dominance of Orbicella annularis on these reefs in 1987, Yawzi Point and Tektite represent coral reef habitats that have remained important throughout the region over ecological (Knowlton 1992, Hughes and Tanner 2000, Williams et al. 2015) and paleo timescales (Mesolella 1967, Jackson 1992). Moreover, they provide a contrast of reefs on which coral cover either has precipitously declined (Yawzi Point) or remained relatively stable (Tektite; Edmunds 2013). Time-series analyses with planar images (photoquadrats) have been used to describe scleractinian community structure at Yawzi Point and Tektite since 1987 (Edmunds 2013). In contrast to our previous work, here we use the same ~1600 photoquadrats to analyze octocoral community structure in a classic Caribbean habitat (i.e., dominated by O. annularis [Williams et al. 2015]). To reveal the bias of neglecting octocorals in coral reef community analysis, we compare the covariance of a commonly used scleractinian-focused assemblage with equivalent assemblages in which octocorals are substituted for scleractinians, or combined with them. The scleractinianfocused assemblage included the cover of scleractinians, macroalgae, and crustose coralline algae, algal turf, and bare space (combined as CTB); the octocoral-focused assemblage consisted of the abundance of octocorals (pooled by genus), and the cover of macroalgae and CTB; the octocoral genera assemblage consisted of the abundance of the 10 octocoral genera found at these sites; and the complete assemblage contained all categories of benthic taxa (scleractinian cover, octocoral abundance by species, cover of macroalgae, and CTB cover). We then describe the extent to which the four community assemblages are associated with physical 2 January 2017 Volume 8(1) Article e01646

3 Fig. 1. Map of study area in Lameshur Bay, St. John, U.S. Virgin Islands, showing Yawzi Point (9 m depth, N, W) and Tektite (14 m depth, N, W). Asterisk marks the Virgin Islands Ecological Resource Station. environmental conditions to gain insight into the factors that might causally be related to changes in community composition. Using this framework, we address two questions: 1. Does the interpretation of long-term changes in community structure differ when described using different community assemblage constructs? 2. Do physical environmental factors influence contrasting community assemblages in different ways? Based on the answers to these questions, we evaluate the general implications of our findings for understanding how octocorals might respond to the recent (and ongoing) trend for region-wide declining abundance of Orbicella spp. MATERIALS AND METHODS General approach The study sites at Yawzi Point (9 m depth) and Tektite (14 m depth) are in the Virgin Islands National Park on the south shore of St. John (Fig. 1; Appendix S1: Fig. S1), where they are exposed to low fishing pressure and a fully protected watershed, due to their location in an MPA (Marine Protected Area) that has been 3 January 2017 Volume 8(1) Article e01646

4 in place for decades. They provide the opportunity to study the local-scale response (sensu Mittelbach et al. 2001) of shallow coral reefs to environmental conditions such as seawater temperature, rainfall, and hurricane intensity (described in Glynn 1993, Rogers 1993, Fabricius 2005). Twenty-five years of change in benthic community structure of the dominant space holders on these reefs (scleractinians, macroalgae, and CTB) are summarized in Edmunds (2013, 2015) and are placed in a larger spatial context in Edmunds (2014). We refer to the selection of benthic taxa in these earlier studies as scleractinian-focused, because the intellectual attention of these studies centered on scleractinians, as is found in the majority of studies of benthic coral reef communities. All data reported in this paper can be accessed through doi / 1912/8504. Biological variables Analyses are based on ~1600 photoquadrats recorded annually since Photoquadrats were recorded with a Nikonos V camera fitted with Kodachrome 64 slide film from 1987 to 1999, but from 2000, digital cameras were used ( : Nikon Coolpix 990, 3.3 megapixels; 2007: Nikon D70, 6.1 megapixels; 2011, Nikon D90, 12.3 megapixels; and : Nikon D7000, 16.2 megapixels, Shinagawa, Tokyo, Japan). Cameras were fitted with a strobe (Nikonos SB105) and mounted on a quadrapod holding them perpendicular to the substratum (Edmunds 2002, 2013). The camera framer remained identical throughout the study and, together with the cameras, resolved objects 10 mm diameter in a 1 9 1m framer. At each site, photoquadrats were recorded at ~10 contiguous locations along each of the three transects that are parallel to one another at constant depth (1 m), and 5 m apart (30 images/yr at each site). The same transects were resampled every year. Sampling occurred in December 1987, March 1988, July 1988, December 1988, April 1989, October 1989, March 1991, May 1992, June 1993, August 1994, May , and July or August thereafter. Images are archived online ( Images were analyzed for percent cover of benthic organisms using CPCe version 3.6 software (Kohler and Gill 2006), or for abundance of octocoral colonies (number of colonies). First, percent cover was determined using 200 dots randomly scattered on each image and scored by their occurrence on scleractinians, macroalgae (algae 1 cm high, consisting mostly of Halimeda, Lobophora, Padina, and Dictyota), and CTB. Scleractinians were scored as a single functional group as the fauna was dominated by O. annularis complex (of which 85% was O. annularis at Yawzi and 58% at Tektite in 1987; see also Appendix S1: Fig. S1), and resolution in the m photoquadrats made it difficult to resolve small colonies such as those of Agaricia and juvenile Porites spp. Second, colony abundance of octocorals (individuals/m 2 ) was quantified with annual resolution, and colonies were counted when their holdfasts were visible in the photoquadrats (Lenz et al. 2015). Erythropodium spp. and encrusting Briareum spp. were represented at low cover and abundance and were quantified based on the number of discrete areas of colonies. We preferred numerical abundance of octocorals to percent coverage, as the ecological meaning of planar cover is equivocal for arborescent taxa that construct canopies from swaying colonies. Cover data with 5-yr resolution were, however, included in Appendix S1: Fig. S2 for comparison with other studies. Octocorals are challenging to resolve to species underwater, because identification typically requires analysis of sclerites (Bayer 1961) in voucher specimens, the collection of which is restricted in the Virgin Islands National Park. Identification is even more challenging in photoquadrats where lighting and resolution can be limiting, and therefore, our analysis focused on the 9 genera found at these sites: Briareum, Erythropodium, Plexaura, Pseudoplexaura, Eunicea, Plexaurella, Muriceopsis, Antillogorgia, and Gorgonia. Small colonies of Eunicea, Plexaurella, Pseudoplexaura, andplexaura spp. were scored as unknowns as they could not be distinguished in the photographs. The height of small colonies could not be determined in planar images, but they were ~12 cm tall. Pterogorgia and Muricea were found in the region, and either was not detected in the sampling areas or could not be resolved in the photographs. Community structure was characterized for four assemblage constructs that employed annual means for dependent variables (cover or abundance). First, the scleractinian-focused assemblage was quantified using the percent cover of 4 January 2017 Volume 8(1) Article e01646

5 scleractinians (pooled among taxa), macroalgae, and CTB. Second, the octocoral-focused assemblage was quantified using octocoral abundance (pooled among taxa) together with cover of macroalgae and CTB. Third, the octocoral genus assemblage focused on octocoral abundance resolved to genus, or unknowns. Fourth, a complete assemblage was used, containing scleractinians (all taxa), octocorals (abundance by genus), macroalgae, and CTB. Data for each benthic group were presented untransformed as means SE by year on scatterplots. Physical variables Physical environmental conditions were characterized using three features that are well known to affect coral reef community dynamics (described in Glynn 1993, Rogers 1993, Fabricius 2005): seawater temperature, rainfall, and hurricane intensity. Together, these were used to generate seven dependent variables describing physical environmental features. Seawater temperature was recorded at each site every min using a variety of logging sensors (see Edmunds 2006 for detailed information on the temperature measurement regime). Seawater temperature was characterized using five dependent variables calculated for each calendar year: mean temperature, maximum temperature, and minimum temperature (all averaged by day and month for each year), the number of days hotter than 29.3 C ( hot days ), and the number of days with temperatures 26.0 C ( cold days ). The upper temperature limit was defined by the local bleaching threshold, and the lower limit defined the 12th percentile of local seawater temperature records (see Edmunds 2006 for details). Rainfall was measured at various locations around St. John (see but often on the north shore (courtesy of R. Boulon; see Edmunds and Gray 2014). To assess the influence of hurricanes, a categorical index of local hurricane impact was employed, with the index based on qualitative estimates of wave impacts in Great Lameshur Bay as a function of wind speed, wind direction, and distance of the nearest approach of each hurricane to the study area (see Gross and Edmunds 2014). Index values of 0 were assigned to years with no hurricanes, 0.5 to hurricanes with low impacts, and 1 to hurricanes with high impacts, and years were characterized by the sum of their hurricane index values. Temporal trends of physical parameters were tested through linear regression using 3-yr centered moving averages to address the lag of response of benthic community structure to environmental conditions (resulting in the loss of 2 yr from the dataset). Statistical framework Descriptive analyses. All analyses of scleractinian-focused, octocoral-focused, octocoral genera, and complete assemblages were based on resemblance matrices using Bray Curtis similarities. Data for the scleractinian-focused assemblage consisted of percent cover and were square-root-transformed; data for the octocoralfocused and complete assemblage consisted of both percent cover and numerical abundance and therefore were z-score-standardized (Sokal and Rohlf 2012); and data for the octocoral genera were z-score-standardized to optimize the performance of principal coordinates analysis (PCoA) for the zero-inflated data. A dummy value of 3 was added to z-score-standardized data to create positive values that could be analyzed in this statistical framework. Non-metric multidimensional scaling (nmds) was used to visualize multivariate trends in community structure for the four assemblages. To prepare nmds plots, multiple restarts of 999 iterations were used until stress stabilized and ordinations were repeatable (after Clarke and Warwick 2001). In these plots, years were represented as circles scaled to scleractinian cover in the scleractinian-focused analysis, and to pooled abundance of octocorals in the octocoral-focused and octocoral genera analyses. Sampling years were clustered using the SIMPROF routine in PRIMER-E, with 999 permutations and significant clusters identified at an alpha of SIM- PROF results were displayed as similarity contours on the respective nmds plots visualizing hierarchical similarity among years (after Clarke and Warwick 2001). To evaluate similarities between two groupings of years that became apparent during initial analysis (as in Edmunds 2013, Edmunds and Lasker 2016), we used an iterative procedure for each graph to determine the highest value of dissimilarity percentage that would describe the groups of years separated in nmds state space. To identify the contribution of 5 January 2017 Volume 8(1) Article e01646

6 each benthic group to inter-annual variability, a PCoA was performed using the cmdscale function in the R statistical package (R Development Core Team 2008). Loading scores were calculated as the Pearson correlations of each dependent variable (i.e., benthic group) against PCO1 and PCO2 and were displayed when significant (P < 0.05) as vectors scaled to a maximum length of 1. The PCoAs were based on Bray Curtis similarities that were produced using the vegan package for R (Oksanen et al. 2015). Question 1. To test whether the description of community dynamics differs when described with the four assemblage constructs, we used a multivariate correlation procedure with significance determined within a permutational framework using a Mantel test (Legendre and Legendre 1998). First, we compared the scleractinian-focused assemblage with the octocoralfocused assemblage; second, we compared the scleractinian-focused assemblage with the octocoral genera assemblage; and third, we tested whether the scleractinian-focused assemblage differed from the complete assemblage. The Mantel test was performed using the vegan package in R [R Development Core Team 2008, Oksanen et al. 2015]). Question 2. The seven physical environmental variables were tested for collinearity by screening variables by pairwise linear correlation. This procedure identified four variables that were independent, and these were used for subsequent analyses: hurricane index (H index ), mean seawater temperature ( C), rainfall (cm), and minimum seawater temperature ( C). The physical variables were transformed using 3-yr centered moving averages of each dependent variable to smooth short-term fluctuations arising from stochastic effects, and to address delayed effects of environmental conditions on the communities. As physical conditions were measured on different scales, they were z-score-standardized prior to analysis (Sokal and Rohlf 2012) and expressed as resemblance matrix based on Euclidean distances. Each of the four assemblages was tested for associations with all combinations of the four measures of physical conditions, using Spearman rank correlation (Clarke and Ainsworth 1993). The Bioenv function (Clarke and Ainsworth 1993) was used for correlations and was followed with a Mantel procedure (Legendre and Legendre 1998) to identify the set of physical variables most strongly associated with the biological variables, with significance evaluated in a permutational framework. The Bioenv function was performed using the vegan package for R (R Development Core Team 2008, Oksanen et al. 2015). General implications. To address general implications of our findings, the Yawzi Point vs. Tektite contrast was interpreted as a comparison of a reef dominated by living Orbicella annularis (i.e., Tektite), with one dominated by antecedent (but dead) colonies of O. annularis with the decline in cover of this species having taken place since In this format, the contrast has utility in evaluating how the regional trend for declining cover of O. annularis (e.g., Hughes and Tanner 2000, Edmunds 2015) is likely to influence the community dynamics of octocorals. Central to this interpretation was an inferential test of octocoral community dynamics at Yawzi Point vs. Tektite, and this was accomplished using the Mantel test to compare all four assemblages as described above. RESULTS Descriptive analysis Yawzi Point Mean scleractinian cover at Yawzi Point declined from 44.6% 3.4% in 1987 to 5.6% 1.4% in 2013 (SE, n = 30; Fig. 2a). This trend was driven mainly by loss of Orbicella annularis, which accounted for 85% of the scleractinian cover in 1987 and 56% in Large declines in scleractinian cover occurred after major hurricanes in 1989 (Hurricane Hugo) and 1995 (Hurricanes Marilyn and Luis), and declines continued thereafter to 2013 (Fig. 2a). Mean macroalgal cover typically varied among years, but increased from 2.2% 0.4% in 1987 to 32.9% 2.3% in 2013, with the highest cover in 1997 (52.4% 6.2%; all SE, n = 30). Mean cover of CTB was 35.4% 3.7% in 1987, but it increased to 53.0% 2.7% in 2013 (SE; Fig. 2a). Mean octocoral abundance (pooled among genera) declined following Hurricanes Hugo in 1989 and Marilyn and Luis in 1995, falling from individuals/m 2 in 1987 to individuals/m 2 in 1996, but thereafter increased to individuals/m 2 by 2013 (Fig. 2b; 6 January 2017 Volume 8(1) Article e01646

7 Fig. 2. Benthic community structure over 27 yr at Yawzi Point. (a) Percent cover (mean SE) of scleractinian corals, macroalgae, and crustose coralline algae, algal turf, and bare space combined (CTB). Gray bars display years affected by hurricanes. Corals are scaled on the right axis and macroalgae/ctb on the left axis. (b) Octocoral abundance (mean SE) of Antillogorgia spp., Gorgonia spp., Plexaura spp., and all octocorals (pooled among taxa). SE, n = 30). Of all octocorals identified in the photoquadrats each year, 1 14% were unidentifiable, 7 48% Plexaura, 10 47% Gorgonia, and 0 26% Antillogorgia (n = colonies/yr). At both Yawzi Point and Tektite, initially abundant plexaurids were replaced by gorgonids (Figs. 2a and 3a). This change was not detectable when octocorals were quantified by planar cover. On this scale, Gorgonia spp. consistently covered the largest proportion of the substratum throughout the study (Appendix S1: Fig. S1). For the scleractinian-focused assemblage, nmds revealed two groups of years that clustered within 85% similarity contours and differed significantly (P perm = 0.005). One early group largely included years from 1987 to 1995 (excluding 1989, 1990, and 1994), and a later group largely included years from 1996 to 2013 (excluding 1989, 1990, and 1994; Fig. 3a). The vectors for loadings of benthic components in the PCoA of this assemblage showed that scleractinians and macroalgae contributed most of the variation along PCO1 that described 34.1% of variation (Fig. 3b). The early group of years was characterized by high cover of scleractinians and low cover of macroalgae, and the later group was characterized by the reverse (Fig. 3b). CTB contributed most to the variation along PCO2, which described 19.1% of the variation in community structure, but did not distinguish groups of years. The communities defined from 2010 to 2013 cluster close together and further from the communities in other years for all four assemblage constructs. For the octocoral-focused assemblage, nmds revealed two groups of years that clustered within 77% similarity contours and differed significantly (P perm = 0.005). One group largely included (excluding 1989, 1990, and 1994) and the other (excluding , while 2001 and 2005 overlapped; Fig 3c). The vectors for loadings of benthic groups in the PCoA revealed that octocorals and CTB contributed most to separation of years along PCO1 that accounted for 31.1% of the variation, and separated two groups of years (Fig. 3d). Macroalgae contributed most to separation of years along PCO2 that accounted for 17.3% of the variation, but did not separate groups of years (Fig. 3d). For the octocoral genera assemblage (Fig. 3e, f), nmds revealed two overlapping groups of years that clustered within 81% similarity and differed significantly (P perm = 0.005); one group largely 7 January 2017 Volume 8(1) Article e01646

8 8 January 2017 Volume 8(1) Article e01646

9 Fig. 3. Benthic community structure over 27 yr at Yawzi Point. (a) Non-metric multidimensional scaling (nmds) plots based on scleractinian corals, macroalgae, and CTB. The results are displayed with bubbles (i.e., yr) scaled in diameter to represent the cover of Scleractinia, and similarity contours indicating hierarchy of similarity based on cluster analysis (5% level). Vectors link points based on chronology. (b) Principal coordinates analysis (PCoA) of the scleractinian-focused assemblage. (c) nmds plots based on the octocoral-focused assemblage. The results are displayed with bubbles (i.e., yr) scaled in diameter to represent total octocoral abundance. (d) PCoA of the octocoral-focused assemblage. (e) nmds plots based on octocoral genera. (f) PCoA of the octocoral assemblage. PCoA loading scores were calculated as the Pearson correlations of each dependent variable (i.e., benthic group) against PCO1 and PCO2 and were displayed when significant (P < 0.05) as vectors scaled to a maximum length of 1. included 1988 and (with extensive overlap of 1989, 1992, , ) and the other (Fig. 3e). The vectors for generic loading in the PCoA revealed that Briareum and Muriceopsis contributed most to separation of years along PCO1 that accounted for 28.8% of the variation, and distinguished two groups of years. Antillogorgia, Plexaurella, Gorgonia, plexaurids, and Eunicea contributed to separation along PCO2 that accounted for 15.2% of the variation, but did not distinguish groups of years (Fig. 3f). For the complete assemblage at Yawzi Point, the nmds revealed two overlapping groups of years that clustered within 82% similarity and differed significantly (P perm = 0.005); one group included and the other , while 2001, 2005, and overlap between both groups (Fig. 4a). The vectors for generic loading in the PCoA revealed that scleractinians and octocorals represent identical vectors, which contributed equally to PCO1 that counted for 35.1% of the variation and PCO2 that accounted for 15.6% of the variation (Fig. 4b). Macroalgae contributed mostly to PCO1, and CTB contributed equally to both components. Descriptive analysis Tektite At this site, mean scleractinian cover increased from 32.2% 2.2% in 1987 to 48.8% 2.9% in 2002, then decreased to 27.3% 2.3% in 2007, and finally increased to 28.3% 1.7% in 2013 (all SE, n = 30; Fig. 5a). Orbicella annularis accounted for 58% of coral cover in 1987 and 44% in 2013, and changes in cover of this species were responsible for most of the overall change in coral cover. Mean macroalgal cover remained <12.8% until 1998, then increased to 36.1% 1.9% in 2006, but declined to 29.9% 1.4% in 2013 (SE, n = 26). CTB covered 25.8% 2.0% of the reef in 1987, 9.9% 1.0% by 1998, and 49.9% 2.0% by 2005 and finally declined to 37.6% 1.7% in 2013 (Fig. 5a). Overall mean abundance of octocorals declined from 3.55% 0.43 individuals/m 2 in 1988 to individuals/m 2 in 2002, but subsequently increased threefold to individuals/m 2 by 2013 (Fig. 5b). The decline in abundance was mainly due to Antillogorgia, Gorgonia, and Plexaura (Fig. 5b). Overall, each year 0 6% of colonies were categorized as unidentifiable, 3 59% as plexaurids, 10 44% as Gorgonia, and 3 21% as Antillogorgia (n = colonies/yr). For the scleractinian-focused assemblage, nmds revealed two groups of years that clustered within 90% similarity contours and differed significantly (P perm = 0.050). The early group largely included and the later group (Fig. 6a). The vectors for loadings of benthic groups in the PCoA showed separation along PCO1 that was driven to a similar extent by scleractinians, macroalgae, and CTB (Fig. 6b), and distinguished early years (high scleractinian cover) from later years (high cover of macroalgae and CTB). nmds for the octocoral-focused assemblage identified three clusters of years that clustered within 76% similarity contours and differed significantly (P perm = 0.005): , , and (Fig. 6c). PCoA of the octocoralfocused assemblage showed that macroalgae and CTB distinguished early years and octocorals distinguished later years and that these taxa contributed most of the separation along PCO1 (31.2%). CTB and octocorals contributed in a similar manner to PCO2 (18.7%), while macroalgae made a negligible contribution (Fig. 6d). nmds for the octocoral genera assemblage identified three loose clusters composed of , , and (Fig. 6e). 9 January 2017 Volume 8(1) Article e01646

10 Fig. 4. Benthic community structure over 27 yr at Yawzi (a, b) and Tektite (c, d). (a, c) Non-metric multidimensional scaling (nmds) plots of the complete assemblage, based on scleractinian corals, octocorals, macroalgae, and CTB. The results are displayed with bubbles (i.e., yr) scaled in diameter to represent the cover of Scleractinia, and similarity contours indicating hierarchy of similarity based on cluster analysis (5% level). Vectors link points based on chronology. (b, d) Principal coordinates analysis (PCoA) of the complete assemblage. PCoA loading scores were calculated as the Pearson correlations of each dependent variable (i.e., benthic group) against PCO1 and PCO2 and were displayed when significant (P < 0.05) as vectors scaled to a maximum length of 1. However, the 80% similarity contour distinguished two groups composed of (including 2013) and 2007, vs that differed significantly (P perm = 0.050). The vectors for generic loading in the PCoA of the octocoral community assemblage showed that Gorgonia, Antillogorgia, Briareum, and Plexaura contributed most to the separation along PCO1 (22.8%) and distinguished two groups, with characterized by a high abundance of these taxa, and the other group by lower abundance. Unidentified species and Gorgonia contributed most of the variation to PCO2 (12.8%; Fig. 6f). The nmds for the complete assemblage at Tektite identified two clusters of years within 85% similarity contours, and these differed significantly (P perm = 0.005): and (Fig. 4c); 2006 and 2008 are not included in either cluster. The vectors for generic loading in the PCoA of the complete assemblage (Fig. 4d) showed that scleractinians and octocorals contributed equally to PCO1 (34.7%), but in opposing direction along PCO2 (18.8%). Descriptive analysis physical environmental conditions The physical environmental conditions varied among years with long-term increasing trends for some variables. Overall, mean seawater temperature between 1989 and 2013 was C (SE, n = 25 yr), in 1989 (the first year records were available) was C, and increased significantly to 2013 ( C; all SE, n = yr 1 ; F 1,21 = 11.74, P = 0.003, r 2 =0.36) January 2017 Volume 8(1) Article e01646

11 Fig. 5. Benthic community structure over 27 yr at Tektite. (a) Percent cover (mean SE) of scleractinian corals, macroalgae, and crustose coralline algae, algal turf, and bare space combined (CTB). Gray bars display years affected by hurricanes. Corals are scaled on the right axis and macroalgae/ctb on the left axis. (b) Octocoral abundance (mean SE) of Antillogorgia spp., Gorgonia spp., Plexaura spp., and all octocorals (pooled among taxa). The highest mean (SE) seawater temperatures were recorded in 1998 ( C) and 2010 ( C). The maximum daily temperature each year increased over time (F 1,21 = 34.22, P < 0.001), although time was not a strong predictor of maximum temperature (r 2 = 0.62). The hottest day during the study was 28 September 2005 (30.6 C), and the coldest day was 16 March 2009 (24.9 C). The number of days categorized as hot each year ranged from 0 to 19 prior to 1998, but from 4 to 82 thereafter, and the number increased over time (F 1,21 = 20.51, P < 0.001, r 2 = 0.49; data not shown). In contrast, 93 d were categorized as cold in 1989, no days were categorized as cold in 1998, and thereafter, between 0 and 83 d were categorized as cold; the number of cold days declined over time (F 1,21 = 8.21; P = 0.009, r 2 = 0.28). Annual rainfall varied from 68 cm/yr in 1994 to 186 cm/yr in 2010, and overall rainfall increased over time (F 1,21 = 63.78, P < 0.001, r 2 = 0.75; Fig. 7). Major hurricanes affected the study area in 1989 (Hurricane Hugo), 1995 (Hurricanes Luis and Marilyn), and 2010 (Hurricane Earl). Hurricanes with lesser impacts occurred in 1996 (Hurricanes Bertha and Hortense), 1998 (Hurricane George), and 2000 (Hurricane Debby). The hurricane index did not change over time (F 1,25 = 2.07, P = 0.17, r 2 = 0.07). Question 1. Does the interpretation of long-term changes in community structure differ when described by different community assemblage constructs? This question evaluated the extent to which a traditional description of coral reef community structure (i.e., scleractinian-focused assemblage) differed from that described by the octocoralfocused and octocoral genera assemblages in the same location. For Yawzi Point (Fig. 3a, c), Spearman rank correlations based on resemblance matrices revealed differences between the scleractinian-focused and octocoral-focused assemblages (r s = 0.73, P perm = 0.001, df = 1, 25), and between the scleractinian-focused assemblage and octocoral genera assemblage (r s = 0.30, P perm = 0.002, df = 1, 25). These differences reflect dissimilar relationships among years between assemblage types. For Tektite (Fig. 6a, c), Spearman rank correlations based on resemblance matrices revealed differences between scleractinian-focused and octocoral genera assemblages (r s = 0.22, P perm = 0.007, df = 1, 25), and between the scleractinianfocused and octocoral-focused assemblages 11 January 2017 Volume 8(1) Article e01646

12 12 January 2017 Volume 8(1) Article e01646

13 Fig. 6. Benthic community structure over 27 yr at Tektite. (a) Non-metric multidimensional scaling (nmds) plots based on scleractinian corals, macroalgae, and CTB. The results are displayed with bubbles (i.e., yr) scaled in diameter to represent the cover of Scleractinia, and similarity contours indicating hierarchy of similarity based on cluster analysis (5% level). Vectors link points based on chronology. (b) Principal coordinates analysis (PCoA) of the scleractinian-focused assemblage. (c) nmds plots based on the octocoral-focused assemblage. The results are displayed with bubbles (i.e., yr) scaled in diameter to represent total octocoral abundance. (d) PCoA of the octocoral-focused assemblage. (e) nmds plots based on octocoral genera. (f) PCoA of the octocoral assemblage. PCoA loading scores were calculated as the Pearson correlations of each dependent variable (i.e., benthic group) against PCO1 and PCO2 and were displayed when significant (P < 0.05) as vectors scaled to a maximum length of 1. (r s =0.83, P perm = 0.001, df = 1, 25). Spearman rank correlations based on resemblance matrices revealed differences between the scleractinianfocused and the complete assemblage at Yawzi (r s = 0.57, P perm = 0.001, df = 1, 25) and Tektite (r s = 0.90, P perm = 0.001, df = 1, 25). Question 2. Do physical environmental factors influence contrasting community assemblages in different ways? The objective of these analyses was to evaluate the extent to which the long-term multivariate trends in community assemblages were associated with physical environmental conditions. Overall, the long-term multivariate structure of the scleractinian-focused and octocoral-focused assemblages was associated in similar ways with Fig. 7. Total annual rainfall (cm; SE) as measured in various locations around St. John, U.S. Virgin Islands, and maximum annual seawater temperature ( C) at Yawzi Point, Great Lameshur Bay, in St. John, U.S. Virgin Islands. environmental variables, notably with rainfall and temperature. At Yawzi Point, variation in the scleractinianfocused assemblage and octocoral-focused assemblages was best explained by the combined variation in rainfall and mean annual temperature (r s = 0.57, P perm = 0.001, df = 1, 21; and r s = 0.38, P perm = 0.001, df = 1, 21, respectively), and for the octocoral genera assemblage, variation was best explained by annual rainfall (r s = 0.67, P perm = 0.001, df = 1, 21). Variation in the complete assemblage was best explained by annual rainfall and mean annual temperature (r s = 0.39, P perm = 0.001, df = 1, 21). At Tektite, the scleractinian-focused assemblage was best explained by rainfall (r s = 0.55, P perm = 0.001, df = 1, 21), variation in the octocoral-focused assemblage was best explained by rainfall and mean temperature (r s = 0.54, P perm = 0.001, df = 1, 21), and variation in the octocoral genera assemblage was best explained by rainfall (r s = 0.52, P perm = 0.001, df = 1, 21). Variation in the complete assemblage was best explained by annual rainfall (r s = 0.29, P perm = 0.001, df = 1, 21). General implications Community dynamics differed between Yawzi Point and Tektite for each of the four assemblages (described previously). For the scleractinian-focused assemblage, temporal community structure differed between Yawzi Point and Tektite (r s = 0.374, P perm = 0.001, df = 1, 25), which reflects the differences between the nmds plots for this assemblage at the two sites (Fig. 3a vs. Fig. 6a). For this assemblage, two clusters of similar years characterized each site, but the break between clusters occurred in the mid-1990s at Yawzi Point, but ~2005 for Tektite. For the octocoral-focused assemblage, community dynamics 13 January 2017 Volume 8(1) Article e01646

14 differed between sites (r s = 0.312, P perm = 0.001, df = 1, 25), with the two nmds plots revealing two clusters of years that separated earlier (1995) at Yawzi Point than at Tektite (2005; Figs. 3c and 6c). Finally, for the octocoral genera assemblage (r s = 0.265, P perm = 0.002, df = 1, 25) and the complete assemblage (r s = 0.36, P perm = 0.001, df = 1, 25), community dynamics also differed between sites, again reflecting differences in the approximate timing of separating of two clusters of similar years at each site (Figs. 3e and 6e). At Yawzi Point, separation occurred in the mid- 1990s, and at Tektite, separation occurred between 2000 and DISCUSSION Frequent disturbances affecting ecosystems usually promote changes in community structure (Connell 1978, Connell et al. 1997, Walther et al. 2002, Grotolli et al. 2014). On tropical coral reefs, these changes can favor taxa other than scleractinians, for example, sponges and octocorals (Ruzicka et al. 2013, Williams et al. 2015). However, most time-series analyses of coral reefs are not well suited to detecting such changes, because they focus on a subset of organisms populating the reef, often just scleractinians and macroalgae (Hughes 1994, Jackson et al. 2014, but see: Ruzicka et al. 2013, Toth et al. 2014, Williams et al for exceptions). To expand the taxonomic scope of analyses of Caribbean reefs, we have started to augment our studies of scleractinian-focused communities (e.g., Edmunds 2015) with a consideration of octocorals (Lenz et al. 2015, Edmunds and Lasker 2016, Edmunds et al. 2016). Here, we continue this theme for two reefs that have been dominated by the scleractinian Orbicella annularis (Edmunds 2013), but which have displayed different recent ecological histories as evaluated from the perspective of scleractinians (Edmunds 2013, 2015). The quantification of community structure on two reefs that initially were similar in coral cover, but subsequently displayed different community dynamics (Edmunds 2013), created the opportunity to consider temporal covariance of octocoral and scleractinian communities under contrasting changes in abundance of O. annularis. We reasoned that characterizing this contrast would improve predictions of impending phase shifts (sensu Hughes 1994) in community structure, by focusing on the ways in which octocorals respond to reductions in scleractinian cover in a habitat (i.e., Orbicella reefs [Williams et al. 2015)] that is common throughout the region. Question 1. Does the interpretation of long-term changes in community structure differ when described by different community assemblage constructs? Analyses of community structure defined by the scleractinian-focused assemblage at Yawzi Point revealed changes that distinguished two clusters of years ( vs ), separating around 1995 when St. John was impacted by Hurricanes Luis and Marilyn. A similar pattern emerged when the same community was analyzed using the octocoral-focused assemblage, and the common pattern suggests the processes driving the changes in scleractinian- and octocoral-focused assemblages are similar. This is further underlined by the complete assemblage, which shares lesser separation and overlapping years with the octocoral-focused assemblage, and is characterized by similar vectors for scleractinians and octocorals in the PCoA (Fig. 4c). Both taxa initially responded to Hurricane Hugo in 1989 with similar declines in population size, while their long-term responses displayed dissimilar trajectories defined by the quick (i.e., within ~10 yr) recovery to pre-disturbance densities by octocorals (but not scleractinians) during the second half of the study ( ). A possible explanation for the changes observed in the octocoral community at Yawzi Point is that members of this taxon exploited space created by declining scleractinian cover, which was driven first by a hurricane and then by chronic effects of bleaching and diseases attributed to elevated seawater temperature (see below). Space liberated through the mortality of Orbicella annularis is topographically complex as a result of the columnar morphology of the antecedent framework, and this three-dimensional surface is well suited to the settlement of octocoral larvae (Yoshioka 1996). This explanation is supported by observations on the shallow reefs of Jamaica where octocoral abundance was inversely related to scleractinian cover (Kinzie 1973). Furthermore, the increase in macroalgal cover that has occurred on this reef (Edmunds 2013) is likely to 14 January 2017 Volume 8(1) Article e01646

15 reduce the survival of juvenile scleractinians (Tanner 1995), whereas recruits of arborescent octocorals are able to quickly grow above the macroalgae canopy, where they gain access to light and seawater flow (Jackson 1979). Although the changes in scleractinian cover at Tektite were small compared to those at Yawzi Point, the multivariate changes in community structure also defined two groups of years for both scleractinian- and octocoral-focused assemblages. In this case, however, the two clusters of years separated around 2005 (cf at Yawzi Point). Again, the complete assemblage is strongly influenced by the octocoral abundance, as it shares with the octocoral-focused assemblage the separation of communities in 2006 and 2008 from the communities in the cluster of years from 2005 to 2013 (Fig. 4a). The distinct separation of clusters of years by community structure at both sites around slightly different times draws attention to events occurring around this time as agents affecting community structure. The scleractinian- and octocoral-focused community dynamics at both sites were associated with rainfall and temperature, but the discontinuity in community structure between sites is defined by the timing of separation of the two clusters of years. This discrepancy may reflect the reduced exposure of the Tektite site to hurricanes that affected this region during the study period, largely due to its location in the lee of Cabritte Horn and the greater depth of the reef. These factors probably contributed to the limited damage to the coral reef at Tektite by Hurricane Hugo in 1989 (cf. Yawzi Point [Edmunds and Witman 1991, Rogers et al. 1991]). In addition, corals at the shallower site (Yawzi Point) might be subject to a positive synergy between temperature and light in causing bleaching (Mumby et al. 2001), which together compounded the damaging effects of Hurricane Hugo on this reef. The cluster of years from 2006 to 2013 based on the similarity of community structure at Tektite may reflect the negative effects of biological agents (e.g., Birkeland 1982) on the growth and survival of octocorals and scleractinians. For example, the high scleractinian cover at this site increases the likelihood of negative densityassociated processes mediating changes in scleractinian community structure, with one such process being disease (Croquer and Weil 2009), a variant of which killed large numbers of O. annularis at Tektite in late 2005 (see Miller et al. 2009). Such processes probably contributed to the decline in scleractinian cover during at Tektite. As occurred with the octocoral community at Yawzi Point from 1996 to 2013, at Tektite, octocoral abundance increased while scleractinian cover decreased. The opposing contribution to PCO2 in the PCoA of the complete assemblage underlines these dynamics (Fig. 4d). Overall, this trend may indicate a higher resilience of octocorals to elevated temperatures or disease, and potentially the exploitation of microhabitats that are suitable for the settlement of octocoral larvae. At each site, analyses of the octocoral genus assemblage revealed two clusters of years that were similar in membership to the two clusters of years defined by the scleractinian- and octocoral-focused assemblages. Edmunds and Lasker (2016) recently described long-term changes in coral reef community structure at sites close to the ones studied herein, but slightly shallower (7 9 m depth) and generally shoreward. They reported that variation in octocoral assemblages was temporally segregated to vs , which suggests that aspects of the changes reported here for Yawzi Point and Tektite have general application to the south shore of St. John. Overall, in the present study the multivariate changes in community structure defined by the octocoral assemblage differed from those defined by the scleractinian-focused assemblage, which likely reflects changes over time in the abundance of individual octocoral genera. Further, the clusters of years were more diffuse (i.e., less clearly separated in two-dimensional space) for the octocoral assemblage vs. the scleractinianand octocoral-focused assemblages at both sites, likely indicating a higher similarity of species composition before and after disturbance. For example, the nmds for the octocoral assemblage demonstrates for Yawzi Point that genus-level resolution identifies and as years with similar community structure (Fig. 3e). In contrast, analysis of the scleractinian-focused assemblage (Fig. 3a) revealed differences in the communities among these years. Insights into the organisms and functional groups driving the dynamics of scleractinianand octocoral-focused assemblages are provided 15 January 2017 Volume 8(1) Article e01646

16 by PCoA. These analyses reveal for the scleractinian-focused assemblage at Yawzi Point that much of the separation among years was caused by macroalgae (more in recent years) and scleractinians (less in recent years), but at Tektite, by scleractinians (less in recent years) and CTB (more in recent years). When described by the octocoral-focused assemblage, the causes of separation among years were more equivocal, being largely driven by CTB (less in recent years) at Yawzi Point, but by macroalgae (less in recent years) at Tektite. At both sites, octocorals played a smaller role in distinguishing clusters of years, probably because of the rapid decline and then recovery of their population densities. The complete assemblage reveals that separation among years in community structure at Yawzi Point was caused mainly by scleractinians, octocorals, and CTB (less in recent years), and at Tektite by scleractinians and octocorals (less in recent years), as well as macroalgae and CTB (more in recent years). PCoA of the octocoral assemblage at Yawzi Point revealed strong effects of Gorgonia, Plexaura, and Eunicea in distinguishing years (more in recent years). At Tektite, the comparable PCoA showed strong effects of Antillogorgia, Gorgonia, Briareum, and Plexaura and unidentified octocorals in distinguishing years (with relatively low densities in the middle of the study [ca. 2005] and more in both early and late years). The causes of the differences in the genera driving changes in octocoral communities at each site cannot be resolved with the present data, as they may reflect the effects of different environmental factors dissimilarly affecting individual genera (sensu Grottoli et al. 2014), or the result of differential representation and relative abundance of octocoral genera at each site. Furthermore, the unidentified octocoral colonies at Tektite (Fig. 6f) represent mainly small colonies, masking potential shifts in octocoral species assemblages. For example, the more easily identifiable Gorgonia spp. have surpassed initially more abundant genera in abundance at both sites by 2013 (Figs. 3f and 6f), which highlights that the octocoral communities at our sites underwent a large change in community structure and did not recover to their pre-disturbance community composition, at least during the study. The nmds plots for all assemblages at both sites show more closely clustered than , possibly highlighting a stasis of community structure over , in a state that differs from the original community structure in Together, our results demonstrate that the interpretation of long-term changes in coral reefs differs depending on the assemblage of organisms with which they are described. Using the most commonly employed assemblage scleractinians, macroalgae, and CTB (e.g., Aronson and Precht 2000) changes in community structure at Yawzi Point and Tektite were similar to those reported from other Caribbean locations (Aronson et al. 2002, Nugues and Bak 2006), namely showing large declines in cover of scleractinians (Yawzi Point) and increases in cover of macroalgae (Yawzi Point and Tektite). Viewed from the perspective of scleractinian-focused assemblages, the prognosis for long-term survival of coral reefs is poor (cf. Bellwood et al. 2004, Hooidonk et al. 2014). This assertion is based in large part on the unique role of scleractinians in building massive, wave-resistant platforms (Moberg and Folke 1999), which means that their decline in abundance has serious implications for the maintenance of reef structure (Alvarez-Filip et al. 2011). However, there is more to coral reefs than to scleractinians (and macroalgae), and as our analyses of octocorals reveal, a broader interpretation of community structure differs from that construed from the scleractinian-focused community. Critically, octocoral communities in St. John were resilient (sensu Nystr om et al. 2000) over three decades in response to conditions that reduced their population sizes, and this ability is likely to be an important factor contributing to their current success in at least some Caribbean locations (Ruzicka et al. 2013, Lenz et al. 2015). The perturbations to which these communities responded appear to have their strongest effects around 1995/2005, which marked the nadir of octocoral abundance at the study sites, and distinguished two clusters of similar years for all four assemblages considered. The segregation of multivariate community structure into two clusters of years also characterizes slightly shallower areas in St. John (Edmunds and Lasker 2016), potentially reflecting a regime shift in octocoral community structure January 2017 Volume 8(1) Article e01646