Boat anchoring contributes substantially to coral reef degradation in the British Virgin Islands

Study location

We studied the effects of anchoring on reefs in the British Virgin Islands (BVI) because the number of active vessels and their size contribute to a high risk of anchor damage to reef habitats in this territory. Approximately 1100-1500 yachts (12–16 m in length) operate in the 155 km² of BVI water. This fleet is expanding and visitation by larger mega-yachts exceeding 45 m in length is increasing (personal communication with Janet Oliver, BVI Charter Yacht Society, and Trish Baily, BVI Association of Reef Keepers and BVI ReefCheck).

Study design

We surveyed 24 reefs (each 0.5–0.75 ha in area) during the summer of 2014 to determine the effects of chronic anchoring. Sites were categorized as experiencing little or no anchoring (hereafter low, n = 11), occasional anchoring (hereafter medium, n = 3) or regular anchoring (hereafter high, n = 10) (Fig. 1). Site selection and classification was based primarily on the plausibility of use as an anchorage and expert opinion about the level of anchoring activity. We selected a set of candidate sites that were plausible anchorages because they were situated on the leeward sides of islands, usually near sand, and often in bays. We also selected candidate sites we considered unlikely to be used as anchorages because, despite being on the leeward side of islands, they were locations likely to be unsafe (e.g., near a cliff) or undesirable (e.g., subject to heavy powerboat traffic) anchorages. We consulted seven local experts who all had >10 years of experience with boating activity in the BVI (dive instructors, charter captains, and marine conservation professionals) to classify sites. High- and low-anchoring sites were those for which all 7 experts agreed on the extent of anchoring. Medium-anchoring sites were those for which there were differences of opinion among local experts about the frequency of anchoring and, although they represent a less clearly-resolved category, we retained them for our surveys.

We performed two post-hoc checks to corroborate the classification of anchoring activity and to clarify the status of medium-anchoring sites. First, we quantified the observed density of yachts using satellite images of each site (Fig. S1). We used all available images from 2004–2014 that were not obscured by clouds (n = 4–10 images per site). There were boat moorings at some sites but, because we knew the locations of moorings, we could distinguish moored boats from those at anchor. Although the images provided a far smaller sample of observations than the collective observations of local experts, we predicted that mean yacht densities in the images would differ among low-, medium-, and high-anchoring sites. Secondly, during our SCUBA surveys at each site we quantified symptoms of anchor damage (described further below) and tested whether they occurred at different frequencies among low-, medium- and high-anchoring sites.

To control for other sources of variation among sites, we selected sites that were similar in depth, wave exposure, and reef slope. In addition, we selected low- and high-anchoring sites in 10 groups that were close in proximity to further reduce the possibility that factors other than anchoring, particularly those associated with land-based human activity, differed among sites (Fig. 1). As a post-hoc check that we had not inadvertently created a bias among treatments by confounding anchoring with effects of land-based human activity, we used satellite images to measure the distance from the center of each reef site to the nearest shore and to the nearest developed land area (Lirman & Fong, 2007) (Fig. S2).

Finally, to determine what percentage of reef in the BVI is potentially vulnerable to anchoring we used GIS to classify coral reefs by exposure (leeward or windward), based on the assumption that only leeward reefs are potential anchoring sites. The GIS feature class utilized was a benthic habitat map showing areas of the sea floor covered by coral reef, seagrass, hard bottom, and algae (NOAA et al., 2001). We isolated the coral reef polygons, created a new geodatabase feature class of those areas, and edited the attribute table to include categories for leeward or windward exposure (Environmental Systems Research Institute, ESRI, 2011).

Survey methods

We sampled each site on SCUBA using 3–8 haphazardly placed 30-m transects. We used the point-intercept method to estimate the percent cover of major benthic taxa, including live hard coral; sea fans; branching soft corals; fleshy, filamentous, calcareous and crustose algae; erect and encrusting sponges; and substrates, such as sand and rubble (Almada-Villela et al., 2003). In addition, all scleractinian coral colonies intersected by the transect tape were identified to species, classified by morphology, and measured in length (L) and maximum orthogonal width (W). Colony surface area (SA) was calculated assuming colonies were elliptical (SA = 0.5L ×0.5W ×π). We calculated coral colony density using the Strong Method, in which density (organisms per m²) = Σ (1/M)(unit area/total transect length) where M is the maximum orthogonal width (Strong, 1966; Bakus, 2002). We also calculated the number of coral species intercepted per transect, as a simple estimate of species richness that is adjusted for sampling effort. All individuals of the most common sea fan (Gorgonia ventalina) within a 30 × 1-m belt transect were counted and measured for height and width.

As a post-hoc check of our classification of anchoring activity at each site, we quantified symptoms consistent with recent anchor damage. For corals, we counted the number of colonies that were newly overturned, broken, or scarred with gouges that were consistent with being scraped by an anchor or chain. For sea fans, we counted the number of colonies that were recently bent or broken at the base, or contained large rips in colony tissue with no associated symptoms of disease.

To quantify the indirect effects of anchoring, we assessed reef complexity and fish population densities. The three-dimensional structural complexity (or rugosity) was estimated for each transect using the consecutive height difference method, for which the height of the tape off the bottom is measured every 50 cm and the square root of the sum of the squared differences between successive height measurements is used to describe vertical complexity (McCormick, 1994). Reef-associated fish were counted using 30x1.5 m belt transects (45 m² in area). We counted all small- to medium-sized diurnal fish species, excluding very small cryptic benthic species (e.g., some gobies and blennies) and very mobile mid-water species (e.g., jacks). Each fish species was classified by both size (juveniles = individuals judged to be <1 month post-settlement, usually <3 cm in total length; adults = all larger individuals) and trophically based functional group (macrocarnivores, piscivores, mobile benthic invertivores, sand invertivores, coral/colonial invertivores, spongivores, diurnal planktivores, nocturnal planktivores, territorial gardeners, turf grazers, scrapers, excavator/eroders, macroalgae grazers, general omnivores) (Halpern & Floeter, 2008).

Statistical analyses

All statistical analyses were done using site means as replicates because chronic anchor damage is more appropriately assessed at the site-level than the transect-level, and because sites are a meaningful unit for management. We tested effects of two categorical factors: anchoring activity (a fixed factor with 3 levels: high (H), medium (M), and low (L)), and group (a random effect with 10 levels). Although sea fans were included in all surveys for benthic cover, specific measurements of sea fan density and size were made at fewer sites (n = 15). These 15 sites were classified as high (n_H = 8) or low (n_L = 7) anchoring.

When data conformed to the assumptions of analysis of variance (ANOVA), we used randomized block ANOVAs to assess the impact of anchoring while accounting for effects of geographic proximity (group), and pairwise comparisons between anchoring activity levels (H-L, H-M, L-M) were made using least squares means (LSM). When data did not conform to the ANOVA assumption of normality, we were sometimes able to transform the data so that the assumptions were met. In other cases, we used either nonparametric Kruskal-Wallis (K-W) tests with Nemenyi post-hoc tests or Maximum Likelihood Estimation (MLE) models using appropriate distributions to model the variance. For MLE models, changes in goodness-of-fit statistics are typically used to evaluate the contribution of subsets of explanatory variables to a particular model, so MLE models were fitted with anchoring only, group only, and both anchoring and group as factors. These models were then compared using Akaike Information Criterion (AIC) and the parameters from the best fitting model were used to calculate means and standard errors for an intuitive presentation of results (Bolker, 2008). MLE models for proportion data utilized a beta distribution with parameters ‘a’ and ‘b,’ from which we calculated the mean (a/(a+b)) and the variance (ab/((a +b)²(a +b +1)). MLE models for data that were positive and continuous utilized gamma and lognormal distributions. The gamma distribution resulted in ‘s’ (scale) and ‘a’ (shape) parameters, with which we calculated means (as) and variances (as²). The lognormal distribution provided µand σ parameters, with which we calculated means (exp(μ + σ²/2)) and variances (exp(2 μ + σ²)(exp(σ²) − 1)). Standard error was calculated as the square root of the variance. We report details of test results in a (Table S1) and list only relevant p-values for pairwise comparisons between levels of anchoring activity in the text. For ease of interpretation and comparison between anchoring intensities, we have reported parameter estimates, z-values, and p-values only from models comparing across anchoring levels. All analyses were conducted in R v. 3.0.2 (R Core Team, 2013), using packages ‘bbmle’, ‘lsmeans’ and ‘PMCMR’.

Checking the classification of anchoring activity and confounding with land-based impacts

The observed density of anchored yachts and frequency of anchor damage symptoms corroborated the expert-based classification of anchoring activity. Based on surveys of satellite imagery, there were more boats anchored at sites classified as high anchoring than either medium or low anchoring (p < 0.0001) (Fig. 2). On the reef itself, high anchoring sites also showed the most signs of apparent anchor damage. We encountered more overturned scleractinian corals (p_H-L = 0.046), broken scleractinian corals (p_H-L = 0.048), and broken gorgonians (p_H-L = 0.002) at high-anchoring sites, and these symptoms were substantially reduced at low-anchoring sites (Fig. S3). Overall, the post-hoc checks corroborated the distinction between low-and high-anchoring sites, and confirmed that, although the 3 medium-anchoring sites show intermediate levels of damage symptoms (Fig. S3), they cannot be conclusively separated from high- and low-anchoring sites.

We also confirmed that there were no apparent differences in proximity to land-based stressors. The low-, medium-, and high-anchoring sites were at similar distances from both land (p = 0.60) and development (p = 0.75) (Fig. S4).

Responses of benthic taxa associated with anchoring

Of the taxa whose benthic cover was surveyed, only hard corals and sea fans showed a significant response to anchoring activity (Table 1). Hard coral cover at highly anchored sites was reduced by a factor of roughly 1.7 relative to sites experiencing medium or little anchoring (p_M-H = 0.047, p_L-H = 0.02, p_L-M = 0.89) (Fig. 3). Sea fan cover at sites with high and medium levels of anchoring was reduced by a factor of 2.6 compared to low-anchoring sites (p_M-H = 0.9, p_H-L = 0.044, p_L-M = 0.049) (Fig. 3).

Responses of coral populations associated with anchoring

Coral colonies were approximately 39% smaller at sites with high levels of anchoring activity than at sites with medium or low levels (p_L-H = 0.046, p_M-H = 0.28, p_L-M = 0.996) (Fig. 4). In addition, coral colony density was roughly 57% lower at sites with high anchoring than at sites with little or no anchoring (p_L-H = 0.0003, p_M-H = 0.55, p_L-M = 0.068) (Fig. 4). Coral species richness was also reduced by approximately 42% at sites with higher anchoring activity (either medium or high) than those with low levels (p_H-L = 0.0002, p_H-M = 0.94, p_L-M = 0.01) (Fig. 4).

Corals with branching, mounding and plate morphologies were most strongly affected by anchoring (Table S2). Branching corals included Acropora cervicornis, Madracis decactis, Porites divaricata, P. furcata, and P. porites; mounding corals were primarily Dichocoenia stokesii, Favia fragum, Madracis pharensis, Montastraea cavernosa, Orbicella annularis, O. faveolata, O. franksii, P. astreoides, Siderastrea radians, and S. siderea, and plate corals were Agaricia agaricites, A. humilis, and A. lamarcki. Branching corals were roughly 65% smaller in size (p = 0.05) and their colony density was reduced by approximately 67% at high anchoring sites than at low-anchoring sites (p_H-L = 0.041, p_H-M = 0.97, p_L-M = 0.37). Similarly, both the colony surface area (p_H-L < 0.0001, p_H-M = 0.03, p_L-M = 0.65) and colony densities (p_L-H = 0.002, p_M-H = 0.69;, p_M-L = 0.10) of mounding corals were reduced by roughly half at sites with high anchoring. Plate coral surface area did not differ between anchoring intensities (Table S2), but colony density was 55% lower at high anchoring sites (p_L-H = 0.026, p_M-H = 0.61, p_M-L = 0.03). Brain corals, cup-like corals, and encrusting corals did not differ in surface area or colony density across anchoring regimes (Table S2).

Changes in reef rugosity associated with anchoring

Rugosity at sites with high anchoring activity was 40% lower than at sites with little or no anchoring (p_L-H = 0.00003, p_M-H = 0.009, p _M-L = 0.63) (Fig. 5).

Changes in fish density associated with anchoring

Mean total fish density was 55% lower at sites with high anchoring than at sites with little or no anchoring (p_H-L = 0.005), and sites with medium anchoring activity had intermediate densities (p_H-M = 0.4, p_L-M = 0.48) (Fig. 6). This reduction was driven by adult fish, whose density was 66% lower at low-anchoring sites than at high-anchoring sites (p_H-L = 0.002), with intermediate densities at medium-anchoring sites (p_H-M = 0.2, p_L-M = 0.58). The density of juvenile fish did not differ by anchoring level (p_H-M = 0.2, p_H-L = 0.6, p_L-M = 0.08). Accompanying the decline in total fish density was a corresponding decline in fish species richness, which was 35% lower at highly anchored sites compared to sites with little or no anchoring (p_H-L = 0.0004); and richness was intermediate at sites with medium-levels of anchoring (p_H-M = 0.6, p_L-M = 0.06) (Fig. 6).

These reductions in fish density were spread across several trophically based functional groups (Table 2). Herbivores capable of removing substratum while feeding (adult scrapers and excavators), were reduced in density by roughly 73% at high-anchoring sites relative to medium- and low-anchoring sites (p_H-L = 0.006, p_H-M = 0.4, p_L-M = 0.5). At our sites, excavators were primarily stoplight parrotfish (Sparisoma viride), and scrapers were primarily blue parrotfish (Scarus coeruleus) and queen parrotfish (Scarus vetula). Other types of adult herbivore, who vary in feeding behavior but do not excavate the substratum, were 66% lower in density at high-anchoring sites than elsewhere (p_H-L = 0.002, p_H-M = 0.14, p_L-M = 0.7). These other herbivores were primarily other parrotfish, surgeonfish, and damselfish and were functionally turf grazers, macroalgal browsers and territorial algal/detritus feeders.

Other functional groups of fish that consume benthic resources also displayed reductions in density associated with anchoring (Table 2). Spongivore densities were 95% lower at highly anchored sites than at low-and medium-anchoring sites (p_H-L = 0.003, p_H-M = 0.4, p_M-L = 0.4) and comprised primarily angelfish, filefish, spadefish, and boxfishes. Benthic carnivores were reduced by 36% at high-anchoring sites (p_L-H = 0.03, p_M-H = 0.03, p_M-L = 0.63). Benthic carnivores at our sites included species that consume mobile invertebrates or sessile invertebrates, and some may also eat fish. This was a broad group that included wrasses, butterflyfishes, hamlets, some groupers, bass, pufferfish, grunts, goatfish, drums, snappers, trunkfish, triggerfish, mojarra, porgies, and porcupine fish. Lastly, there were also 61% fewer piscivores at high-anchoring sites than at low- and medium-anchoring sites (p_M-H = 0.012, p_L-H = 0.004, p_M-L = 0.73). At our BVI sites, piscivores comprised mostly groupers, lizardfish and lionfish.

Fraction of reefs vulnerable to anchoring

From the marine habitat maps we examined, we estimated 24% of coral reef area in the British Virgin Islands is leeward in exposure, although this only a rough estimate of the fraction of reef that is safe to anchor near because exposure varies with the seasons and weather.