Artificial habitats host elevated densities of large reef-associated predators

Large predators play important ecological roles, yet many are disproportionately imperiled. In marine systems, artificial reefs are often deployed to restore degraded reefs or supplement existing reefs, but it remains unknown whether these interventions benefit large predators. Comparative field surveys of thirty artificial and natural reefs across ~200 km of the North Carolina, USA coast revealed large reef-associated predators were more dense on artificial than natural reefs. This pattern was associated with higher densities of transient predators (e.g. jacks, mackerel, barracuda, sharks) on artificial reefs, but not of resident predators (e.g., grouper, snapper). Further analyses revealed that this pattern of higher transient predator densities on artificial reefs related to reef morphology, as artificial reefs composed of ships hosted higher transient predator densities than concrete reefs. The strength of the positive association between artificial reefs and transient predators increased with a fundamental habitat trait–vertical extent. Taller artificial reefs had higher densities of transient predators, even when accounting for habitat area. A global literature review of high trophic level fishes on artificial and natural habitats suggests that the overall pattern of more predators on artificial habitats is generalizable. Together, these findings provide evidence that artificial habitats, especially those like sunken ships that provide high vertical structure, may support large predators.


Introduction
Across terrestrial and aquatic ecosystems, large predators provide valuable services, such as enhancing ecosystem resilience and production [1,2], facilitating ecosystem connectivity [3], and maintaining biodiversity [4]. Numbers of many large predators across different ecosystems have declined due to factors including habitat degradation, as well as unregulated fishing or hunting pressure [5]. While approaches towards and success of predator conservation remain largely context dependent, many successful conservation efforts have focused on a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or polices of the US Government, nor does mention of trade 462 names or commercial products constitute endorsement or recommendation for use. CSS-Inc provided support in the form of salaries for author [ABP], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of this author are articulated in the 'author contributions' section.
Competing interests: Author ABP is employed by CSS-Inc. This affiliation does not alter our adherence to PLOS ONE policies on sharing data and materials.
surveyed fish along two 30-m x 4-m (120 m 2 ) belt transects seasonally established along prominent reef features, following Paxton et al. [29] (S1 Table of S1 Data). If a prominent feature did not exist, the transect direction was selected from a list of randomly generated compass headings [35]. At each reef, transect location varied among seasons. Thirty-m long transects have been demonstrated to be accurate, precise, and efficient for assessing fish density on reef habitats [36], and a prior study in the same geographic area as ours found that 77% of fish were accounted for 30 m outward from reefs. On each belt transect, divers recorded the abundance of all fishes present throughout the water column, including both conspicuous (e.g., easily discernable fishes not hiding in reef structures) and cryptic (e.g., smaller fishes often hiding in reef structures) categories of reef fishes, to the lowest taxonomic level possible. In total, 236 transects were conducted, 109 on artificial reefs and 127 on natural reefs.
We calculated the density of large reef-associated predators on each 120 m 2 transect during a sampling event. In particular, we classified fish as large reef-associated predators based on size and diet as reported in Fishbase [37] and regional studies published in the peer-reviewed literature. Following the recommendation of Heupel et al. [38] to classify marine predators by size and diet, we focused on species who can grow to 70 cm or larger and who eat other fish. We designated fish that can grow to 70 cm or larger as large predators based on our prior field surveys on offshore NC reefs, where this threshold is 'large' in the context of these reefs [29,39]. In total, we classified twenty-seven species as large reef-associated predators, including species in the families Belonidae (needlefish), Carangidae (jacks), Carcharhinidae (sand bar shark), Dasyatidae (southern stingray), Haemulidae (grunts), Lutjanidae (snapper), Odontaspididae (sand tiger shark), Paralichthyidae (flounder), Rachycentridae (cobia), Rhincodontidae (nurse shark), Scombridae (mackerel), Serranidae (grouper), Sparidae (porgy), and Sphyraenidae (barracuda) ( Table 1). We pooled density values for two species, barracuda (Sphyraena barracuda) and guachanche (Sphyraena guanchancho), at the genus level as Sphyraena sp. because similarities in appearance made it difficult for divers to distinguish them at the species level. The species that we classified as predators are likely mesopredators or top predators depending on the ecosystem specific context, as well as food web dynamics, such as whether these species exert diffuse or concentrated predation pressure [36].
We further classified the large reef-associated predators by their residency to reefs (Table 1). We designated species that more transiently associate with reefs, such as fast-swimming, highly-mobile, schooling species, which often appear in the water column above or around reefs, as 'transient.' Transients included fish, such as jacks [40], mackerel [41], barracuda [42], and sharks [43]. We classified demersal species more commonly associated with the reef structure or seafloor as 'resident.' The residents included fish, like snapper, grouper, flounder, and rays, which often exhibit high degrees of site fidelity and residence time to natural and artificial reefs [41,[44][45][46][47][48]. We made this classification based on previous studies cited above detailing residency patterns for these species, as well as our ecological knowledge from extensive observations of which species associate with the reef structure versus the water-column above and because our of personal observations that water-column associated species tended to be more prevalent on artificial versus natural reefs.

Statistical analysis
Statistical analyses were conducted in R version 3.5.3 [49]. We used generalized linear mixed models (GLMMs) [50] to model the relationship between densities of large reef-associated predators, which was the response variable, and reef and species characteristics [51] (S1 Text of S1 Data). Density values were measured as species-specific, transect-level counts observed during a single sampling event. We included reef type (artificial vs. natural) and species 'residency' status (transient vs. resident) and their interaction, as well as reef depth and sampling season (winter, spring, summer, fall), as fixed effects. We calculated reef depth as the mean depth recorded by a pressure transducer (Onset Hobo U20 Titanium Water Level Logger, U20-001-02-Ti) moved along the transects at a particular reef (see [29] for methods details). We included individual species and sampling event nested within reef as random effects, the latter to allow for reef-to-reef variation not described by the model's fixed effects and for the possible correlation of predator densities on a reef's transects visited during the same sampling event. We fit the model with the 'glmmTMB' package [52]. We used negative binomial models to allow for overdispersion in the counts relative to the Poisson model and the log link function. Likelihood ratio tests (LRTs) were used to assess strength of evidence in favor of associations between predator density and reef type, residency status and their interaction, as well as associations with season and reef depth.
We used GLMMs and generalized linear models (GLMs) to investigate potential mechanisms, such as reef morphology and vertical relief, behind observed patterns in predator densities by reef type. First, we examined the relationship between specific reef morphologies and predator density. We did this by refitting the GLMM model above (fixed effects: reef type x residency, depth, season; random effects: sampling event nested within reef, species), and conducting accompanying LRTs, with reef morphology replacing reef type as a fixed effect. Reef morphology is a factor containing four levels, two for each reef type (artificial-ship, artificialconcrete, natural-pavement-and-rubble, natural-ledge). Morphology levels for artificial reefs relate to characteristics, such as reef material and spatial extent, as ships are metal, isolated structures, whereas concrete structures are often dozens to hundreds of individual structures dispersed across a broader area. Morphology levels for natural reefs also relate to reef characteristics, as pavement-and-rubble reefs are often low-lying, ephemeral habitats that can be covered and uncovered by sand, whereas ledges are more pronounced, less ephemeral habitats. Second, we examined the relationship between reef vertical relief and predator density. We calculated vertical relief as the difference between the deepest and shallowest point on each reef transect, as recorded by a pressure transducer (Onset Hobo U20 Titanium Water Level Logger, U20-001-02-Ti) moved continuously along the transect (see [29] for methods details). Using the maximum vertical relief (m) values for each reef across sampling events and transects, we fit a negative-binomial model for total predator density to: (1) the full data set, (2) the subset of transient predators on artificial reefs, and (3) the subset of data corresponding to transient predators on artificial reefs that were ships with known length using the 'MASS' package [53]. This third model allowed us to isolate the effect of reef size from the relationship between transient predator density and reef relief. We were able to do this only for the subset of artificial reefs that were ships of known lengths prior to sinking. We do not know the habitat size of the remaining natural and artificial reefs, as most remain unmapped. We extracted information on ship lengths from historical information for shipwrecks and from the NC Division of Marine Fisheries for artificial reefs. We then scaled transient predator densities on these ships by the ship length. We converted these normalized predator values to integers and used them as the response variable regressed against the predictor variable vertical relief in our third GLM. For all GLMs, we conducted LRTs between the full model and the reduced model (e.g., model without vertical relief) to determine the effect of vertical relief on predator density, transient predator density, and scaled transient predator density, respectively.

Literature review
To test whether patterns observed in diver data were generalizable, we conducted a literature search and accompanying analysis of reef-associated predators on artificial versus natural habitats in marine and coastal ocean environments (Table 2). We set our literature search query a  [54] and screened each using specified inclusion criteria ( Table 2). Of the 479 studies, 93 were selected as potentially relevant and included in a subsequent full-text screening based on the same criteria. Of the 93 studies for which fulltext screening was conducted, fourteen met our criteria and were retained for data extraction. Data on how many fishes were present on each reef type were extracted from the fourteen studies. For each study, we extracted the scientific name of each fish species for which data existed. We then queried Fishbase [37] for the trophic level of each species. Trophic levels are assigned based on diet studies, natural history, and other related information [37] (S2 Text of S1 Data). We focused on fishes with trophic levels of 4.4 and 4.5 because these are often reefassociated predators, including some top predators and some mesopredators. There were forty-two records of fishes with trophic levels 4.4 and 4.5 with reported abundance, density, or biomass on artificial versus natural marine habitats. We further refined this list by examining prey items and body sizes to keep only those fish regarded as reef-associated predators. For example, we removed tomtate (Haemuleon aurolineatum) which have a high trophic level on Fishbase but are not reef-associated predators in many locations. Our refined list came from six unique studies, as eight of the original fourteen studies did not include large predatory fish, as defined on the basis of trophic level (S5 Table of S1 Data). The reefs in these studies did not have protection status, except for one where shrimp bottom trawling was prohibited on artificial reefs [55]. The final six studies collectively reported on eighteen predatory fish species (S6 Table of S1 Data); for each, we verified status as reef-associated predators and classified each as either resident or transient based on level of reef association. Because several studies reported on the same species of fish as other studies, our final extracted data included 26 reports of these 18 fish species of large predatory fish. Our list is likely conservative or restricted because fishes with trophic levels < 4.4 were not included yet may fill roles as large reef-associated predators.
For each reef-associated predator species from each study, we extracted metadata for the study, including the title, date, authors, geographic location, survey method (e.g., scuba belt transect vs. remotely-operated vehicle survey), artificial reef type (e.g., vessel vs. concrete), and reef depth. We then extracted the abundance, density, or biomass of each species from each study on artificial versus natural habitats. We also extracted any available measures of precision, such as standard error, standard deviation, or p-values and test statistics from which precision metrics can be calculated. Because several studies reported measures of precision whereas others did not report these values and because the measurements (e.g., abundance, density, biomass) differed among studies, we could not carry out a formal meta-analysis to estimate a global effect. Since a formal, precision-based weighting was not possible, we treated all studies with equal weight and because the abundance metrics varied by study, we summarized the data using a binary variable computing whether the species was more dense on artificial than natural habitats (= 1) versus not (= 0).
We estimated the log odds of reef-associated predators exhibiting higher values (e.g., abundance, density, or biomass) on artificial reefs than natural reefs by fitting a logit link binomial GLMM [50] with the 'glmmADMB' package [56]. We did not include residency status in the model because the sample size was too low. Because the data were extracted from six studies using different survey methods across multiple geographic areas, we included study as a random effect. The model's only fixed effect was its intercept; it is the quantity of interest because its interpretation is as the log odds of reef-associated predators being present on artificial habitats in higher values than on natural reefs. We used a one-sided test of the hypothesis that predators are equally dense on the two reef types versus more dense on artificial reefs because we turned to the literature review to validate our findings from the observational field data that reef-associated predators would be present on artificial reefs at higher values than on natural reefs. We computed point and 95% interval estimates of the odds of predator abundance being greater on artificial reefs.

Observational data
Comparative field surveys of fourteen artificial reefs and sixteen natural reefs across~200 km of the North Carolina, USA coast revealed that large reef-associated predators were 4.3 times more dense on artificial than natural reefs. Mean observed predator density across species per sampling transect conducted on artificial reefs was 1.00 ± 0.24 SE, whereas mean density across species on natural reefs was 0.23 ± 0.04 SE. This pattern is based on relative densities of large fish predators observed during diver-conducted visual surveys (236 transects) conducted seasonally on the reefs in 2013-2015 (S1 Table of S1 Data; Fig 1). The large fish predators include twenty-seven species defined on the basis of their size and diet and designated as either transient or resident by its reef association (Table 1). Of these twenty-seven species, all but eleven were rare.
When we examined patterns of predator density, we found that, although large reef-associated predators were more dense on artificial reefs than nearby natural reefs (Fig 2A; Table 1; S1 Fig and S2 Table of S1 Data; LRT reef type, NB1, χ 2 = 44.38, p < 0.0001), this pattern was driven by higher densities of transient predators on artificial reefs (Fig 2B; S3 Table; LRT reef type x residency, NB1, χ 2 = 44.42, p < 0.0001). These more transient predators, which often associate with reef structures and the water column above, include jacks (e.g., greater amberjack, almaco jack, yellow jack), mackerel (e.g., Spanish mackerel, king mackerel, little tunny), and sharks (e.g., sand tiger sharks). Our model estimated that transient predators were five times as dense on artificial than natural reefs, as the estimated mean transient predator density per transect was 0.31 ± 0.15 on artificial reefs and 0.08 ± 0.04 on natural reefs at the average reef depth (20 m) during both the summer and fall seasons (S3 Table of S1 Data). While there was a low frequency of occurrence for most of the transient predators, when they were observed, they were nearly always observed on artificial reefs ( Table 1; S1 Fig of S1 Data). For example, king mackerel were only observed on artificial reefs but with a low frequency (1.9%). Three species, nurse shark, sandbar shark, and cobia were only observed once throughout the study but always on artificial reefs (0.9%). Spanish mackerel and little tunny were observed < 5% of the time on artificial reefs. Some of the transient species were more frequently observed on artificial reefs, including almaco jack (10.1%), sand tiger sharks (17.2%), yellow jack (21.1%), barracuda (43.4%), and greater amberjack (65.2%).
In contrast, predators who are more resident to reefs, often residing closer to the seafloor, did not follow this pattern (Fig 2C; Table 1; S2 Fig and S3 Table of S1 Data). The more resident species, including grouper (e.g., gag, scamp) and snapper (e.g., red snapper), displayed similar densities on artificial and natural reefs. Our model estimated that mean resident predator density was 0.18 ± 0.08 and 0.14 ± 0.06 per transect on artificial and natural reefs, respectively, at the average reef depth (20 m) during the summer and fall (S3 Table of S1 Data). Some resident predators exhibited high frequencies of occurrence on both reef types ( Table 1; S2 Fig of S1 Data). For example, gag and scamp grouper were observed >75% and~40% of the time on both reef types, respectively. Other species, such as red snapper and black margate, were observed less frequently on both reef types (<10%). Several rare species, such as the barndoor skate and yellowtail snapper, were each observed once but only on natural reefs (0.8%), whereas one rare species, the southern stingray, did not adhere to the pattern and was exclusively observed on artificial reefs.
The model that we used to estimate the effect of reef type and fish residency on predator density accounted for depth and season as fixed effects and reef-to-reef and species-to-species variation as random effects. Depth related to the density of predators, with deeper reefs generally hosting higher predator densities (S2 and S3 Tables of S1 Data; LRT depth, NB1, χ 2 = 11.15, p < 0.001). Season did not significantly relate to predator densities, although reefs sampled during summer and fall seasons tended to have higher predator densities than reefs sampled in spring and winter (S3 Fig; S2 and S3 Tables of S1 Data; LRT season, NB1, χ 2 = 2.68, p = 0.44). The amount of variation in random effects accounted for by reefs (SD = 0.06) and sampling events (SD = 0.15) is smaller than that of species (SD = 2.32; S2 Table of S1 Data). This suggests that there are species-specific preferences. Including the predictor variable for species residency in the model reduced species variation compared to excluding this term from the model. The additional species-to-species variation unaccounted for by the residency predictor variable likely stems from differences in other species traits, such as size and schooling behavior.
Further analyses investigating potential mechanisms that may explain the pattern of elevated predator densities on artificial reefs revealed that reef morphology and reef height were associated with predator densities. Including reef morphology in the model instead of reef type improved the model fit (ΔAIC = 15.7) and suggested that higher densities of transient predators on artificial reefs were largely driven by ships (Fig 3; S2 Table of S1 Data LRT reef morphology x residency, NB1, χ 2 = 68.14, p < 0.0001). The model estimated that mean transient predator density per transect was over twice as high on ships (0.37 ± 0.18) than on concrete (0.16 ± 0.09) at the average reef depth (20 m) during the summer (S4 Table of S1 Data). The observed mean transient predator densities were even higher-eleven times greater-on ships (2.77 ± 0.79 SE) than concrete artificial reefs (0.24 ± 0.07 SE) (S4 Table of S1 Data). Additionally, vertical relief of reefs positively correlated with pooled transient and resident predator densities from both artificial and natural reefs (Fig 4A; GLM NB, LRT, χ 2 = 44.49, p < 0.0001). This positive effect of artificial reef relief held for transient predators on artificial reefs (Fig 4B; GLM NB, LRT, χ 2 = 10.33, p < 0.01). This pattern was also preserved across a subset of artificial reefs for which we could normalize transient predator density by habitat area (Fig 4C; GLM NB, LRT, χ 2 = 6.35, p = 0.01). The subset of data that we normalized was for artificial reefs that were ships of known lengths. For these ships, we used their length (length prior to sinking) as an estimate of habitat area.

Literature review
We documented twenty-six cases (reports or instances of a species within a study) in which values of large reef-associated predatory fishes were reported on both artificial and natural habitats (S5, S6 Tables of S1 Data). These twenty-six cases stemmed from six studies and represented eighteen species. Of these twenty-six cases, including observations in North America, South America, and Africa, the reef-associated predators exhibited higher values on artificial reefs in eighteen cases (69%) and on natural reefs in eight cases (31%; Fig 5). Predator residency status did not affect the outcome, as for both residents and transients, there were nine cases of higher  values on artificial reefs and four on natural reefs. When we explicitly tested the odds of large reef-associated predators occurring in higher values on artificial versus natural habitats, our model revealed large predators were significantly more likely to occur in higher abundance, density, or biomass on artificial habitats compared to natural habitats (one-sided p = 0.03). Specifically, on the basis of this analysis, we estimate that the odds of large predator species being more abundant on an artificial than natural reef is 2.25 (95% confidence interval 0.98, 5.13). Study-to-study variation in this quantity was estimated to be small (SD = 0.04).

Discussion
Extensive scuba-diver surveys of thirty reefs across~200 km of the continental shelf of coastal North Carolina revealed that large reef-associated predators were more dense on artificial reefs than nearby natural reefs and that this effect was explained by transient species, such as jacks, mackerel, and sharks. Further, this pattern of higher predator density on artificial reefs was associated with reef morphology and vertical relief, as ship-type artificial reefs hosted more transient predators than concrete artificial reefs, and higher relief reefs hosted more transient predators, even when controlling for basal area. When we tested the generality of our finding of higher predator density on artificial reefs using previous studies, we found that large reefassociated predators tend to be more abundant on artificial habitats compared to natural habitats. Based on our literature review, we estimate that the odds of artificial habitats across global marine ecosystems supporting higher abundance, density, or biomass of top consumers were over two times greater than for natural habitats. These findings provide evidence that artificial reefs, and more broadly marine artificial habitats, may have important ecological functions, including the support of large reef-associated predators, which are critical components of healthy ecosystems.
We propose two explanations for why transient predators may occur in higher densities on artificial than natural reefs. Both explanations relate the positive effect of artificial reefs on transient predator density to morphology and vertical extent of habitat, as artificial reefs are nearly three times as structurally complex (rugosity and vertical relief) as nearby reefs in our study system [29]. First, transient predators often move among habitats to find prey, so they could occur in higher numbers on artificial reefs, especially ships, simply because these vertically-extensive reefs can host high abundances of prey. Many of the transient species that we observed consume small baitfish, which have been documented to occur in high abundances on artificial reefs of various structural complexities [29,50], supporting this hypothesis. Rigorously testing this hypothesis is an avenue for future research. Second, transient predators moving from reef to reef may stop over on artificial reefs because certain artificial reefs, such as ships, are taller [29] and therefore potentially easy to distinguish from surrounding sand. In contrast, for resident predators, such as grouper and snapper, our field surveys indicate that they have similar densities on artificial and natural reefs. We do acknowledge that other habitat measures, such as area (horizontal size) or volume, could also relate to predator density.
In addition to reef morphology and vertical relief, several other reasons may explain why artificial reefs host higher densities of predators in general. First, artificial habitats may augment the capabilities of existing habitats to support large predators. For example, artificial habitats installed as replacements for degraded habitats may replace functions formerly associated with once healthy habitats. Similarly, artificial habitats deployed in locations with healthy, natural habitats may supplement, rather than replace, their function for large predators. Second, in areas that are habitat limited, artificial habitats may form otherwise unavailable sources of habitat capable of supporting large consumers. Specifically, in North Carolina, where our comparative field studies were conducted, differences in habitat availability may relate to why transient predators, but not resident predators, were more abundant on artificial than natural habitats. In this system, natural habitats are often extensive networks of ledges and flat pavements that are not degraded, so artificial habitats of both ship and concrete morphologies supplement existing functions of healthy rocky reefs. The artificial habitats of ships or fields of concrete structures are often isolated 'islands' of habitats, whereas the natural rocky reefs are more continuous [29].
The possibility remains that large predators are less likely to be detected on more extensive rocky reefs than on the more island-like artificial habitats simply based upon encounter rates in surveys relative to the overall area of natural reefs. Acoustic telemetry of large predators could help address this by investigating spatial movements and habitat use questions. Comparisons of predator density across the entire area of habitats, rather than standardized sampling transects as conducted here, would also help disentangle whether habitat area may drive the patterns in predator densities that we observed, as would examining reef spatial distributions and proximity to one another. Future research should determine whether large reef-associated predators receive net energetic or fitness benefits from artificial reefs through foraging or whether these large predators are merely resting on the reefs; answering this question will help determine whether artificial habitats provide an aggregation or production function for these fishes. Moreover, while the studied reefs are all open to fishing, we do not know whether differential fishing pressure among sites may relate to our findings. In particular, spearfishing is allowed on these reefs and may result in fish being more wary around our divers conducting fish transects and less likely to be detected on the transects [57].
While there are multiple approaches for the management of large marine predators, our findings initially indicate that strategic installation of artificial habitats may also be a potential tactic for managing these predators; however, this would not be a sound approach if artificial reef aggregate rather than produce predators. Based on the findings of our observational study, we propose that this method would be best suited for transient or mobile species, like jacks and barracuda, which move locally among habitats [40,42]. This approach may also be suitable for more highly-migratory or mobile species, such as sharks, that move across multiple habitats and geographic areas during migration [3,58]. We caution that we do not know whether artificial habitats simply aggregate large reef-associated predators or facilitate the production of these predators [23]. The broader answer to this question remains controversial, and there is likely a mixture of aggregation and production across trophic levels that occurs on marine artificial habitats [25]. If, however, artificial habitats simply aggregate predators, then using artificial habitats for predator conservation could have negative effects. For example, if predators are attracted from nearby natural habitats, then the consumptive pressure of the predators could have substantial impacts. Additionally, aggregating predators could make predator exploitation easier, possibly negating benefits of artificial reefs. These are concerns that must be addressed before artificial habitats are used to support predators. One step towards resolving these concern is testing whether predators exhibit active foraging behavior or reproductive behavior on artificial habitats that could indicate production rather than aggregation.
With these uncertainties in mind, we cannot advocate for the widespread and rapid introduction of artificial habitats globally in marine ecosystems. Instead, we suggest that in areas that may be habitat limited and also historically hosted or currently host populations of large predators that marine artificial habitats might assist predators and in such a way that these artificial habitats may facilitate conservation and management objectives.