Coral Reefs

, Volume 31, Issue 4, pp 977–990

Abiotic and biotic controls of cryptobenthic fish assemblages across a Caribbean seascape

Authors

    • Marine Spatial Ecology Laboratory, Biosciences, College of Life and Environmental Sciences, Hatherly LaboratoryUniversity of Exeter
    • Marine Spatial Ecology Laboratory, School of Biological SciencesUniversity of Queensland
    • ARC Centre of Excellence for Coral Reef StudiesUniversity of Queensland
  • H. L. Jelks
    • Southeast Ecological Science CenterUS Geological Survey
  • W. F. Smith-Vaniz
    • Florida Museum of Natural HistoryUniversity of Florida
  • L. A. Rocha
    • Section of IchthyologyCalifornia Academy of Sciences
Report

DOI: 10.1007/s00338-012-0938-4

Cite this article as:
Harborne, A.R., Jelks, H.L., Smith-Vaniz, W.F. et al. Coral Reefs (2012) 31: 977. doi:10.1007/s00338-012-0938-4

Abstract

The majority of fish studies on coral reefs consider only non-cryptic species and, despite their functional importance, data on cryptic species are scarce. This study investigates inter-habitat variation in Caribbean cryptobenthic fishes by re-analysing a comprehensive data set from 58 rotenone stations around Buck Island, U.S. Virgin Islands. Boosted regression trees were used to associate the density and diversity of non-piscivorous cryptobenthic fishes, both in the entire data set and on reef habitats alone, with 14 abiotic and biotic variables. The study also models the habitat requirements of the three commonest species. Dead coral cover was the first or second most important variable in six of the eight models constructed. For example, within the entire data set, the number of species and total fish density increased approximately linearly with increasing dead coral cover. Dead coral was also important in multivariate analyses that discriminated 10 assemblages within the entire data set. On reef habitats, the number of species and total fish density increased dramatically when dead coral exceeded ~55 %. Live coral cover was typically less important for explaining variance in fish assemblages than dead coral, but live corals were important for maintaining high fish diversity. Coral species favoured by cryptobenthic species may be particularly susceptible to mortality, but dead coral may also provide abundant food and shelter for many fishes. Piscivore density was a key variable in the final models, but typically increased with increasing cryptobenthic fish diversity and abundance, suggesting both groups of fishes are responding to the same habitat variables. The density of territorial damselfishes reduced the number of cryptobenthic fish species on reef habitats. Finally, habitats delineated by standard remote sensing techniques supported distinct cryptobenthic fish assemblages, suggesting that such maps can be used as surrogates of general patterns of cryptic fish biodiversity in conservation planning.

Keywords

Boosted regression treesDamselfishesMarine reservesPiscivoryRotenoneWave exposure

Introduction

Fishes provide a wide range of ecosystem services (Holmlund and Hammer 1999), so it is important to understand the ecology, functional role and conservation of all species within assemblages. However, the vast majority of studies consider only non-cryptic species, and data on cryptic fish species are scarce (Munday and Jones 1998; Depczynski and Bellwood 2003). This bias in the literature is largely a function of the logistical and taxonomic difficulties of surveying cryptobenthic fishes, typically defined as fishes with a maximum size of <5–10 cm, which are visually and/or behaviourally cryptic, and maintain a close association with the benthos (Depczynski and Bellwood 2003, 2005b). Non-cryptic fishes can be surveyed relatively easily and rapidly using underwater visual censuses, but these techniques significantly underestimate the diversity and abundance of cryptobenthic species (Brock 1982; Ackerman and Bellwood 2000; Willis 2001; Smith-Vaniz et al. 2006). Consequently, more specialised and time-consuming methods are required to accurately sample cryptobenthic fish species, typically involving ichthyocides. Therefore, cryptobenthic fishes are frequently and explicitly ignored by researchers, limiting our knowledge of their ecology and the ecosystem services they provide.

Despite a paucity of data, small and cryptic species are recognised as an abundant and important component of fish assemblages on coral reefs (Munday and Jones 1998). For example, cryptic species constitute >50 % of individuals and >40 % of all reef fish species at a Great Barrier Reef site (Ackerman and Bellwood 2000), and 72 and 68 % of species on reefs in the Gulf of California and Red Sea, respectively (reviewed by Barlow 1981). In addition to this numerical importance, the contribution of cryptobenthic fishes to ecological processes is increasingly apparent. Such contributions include a key role in the trophodynamics of reef food webs, with fishes <5 cm potentially using >25 % of the total energy required by reef fishes (Ackerman and Bellwood 2000; Depczynski and Bellwood 2003). Comb-toothed blennies may have a significantly underestimated role in the key ecological process of grazing on reefs (Townsend and Tibbetts 2000), and detritivorous blennies appear to be important secondary consumers (Wilson 2001). Furthermore, cryptic fishes are typically small and a food source for many piscivores (Ackerman and Bellwood 2000; Steele and Forrester 2002; Depczynski and Bellwood 2005a). Cryptobenthic fishes also produce approximately one-third of the carbonate precipitated by teleosts, which is an important component of marine mud and tropical carbonate budgets (Perry et al. 2011).

As the vital ecological role of cryptobenthic fish species in tropical habitats becomes clearer, it is increasingly important to understand how and why their assemblages vary amongst and within habitats (Depczynski and Bellwood 2005b). Although limited by the lack of comprehensive data sets, a Great Barrier Reef study demonstrated clear associations between different species and particular microhabitats, such as caves and sand areas (Depczynski and Bellwood 2004). More generally, another Great Barrier Reef data set suggested wave exposure as an important driver of cryptobenthic fish assemblages, along with the composition of the substratum in more sheltered areas (Depczynski and Bellwood 2005b). Further descriptions of microhabitat preferences of individual Pacific cryptobenthic fish species can be gleaned from studies of particular ecological processes and interactions, such as competition (e.g. Munday et al. 1997). In the Caribbean, a series of papers demonstrated distinct assemblages of gobies and blennies, but less clear patterns in cardinalfishes, across Central American reef habitat types (Greenfield and Johnson 1990a, b, 1999). Links between cryptobenthic fishes and biotic and abiotic habitat characteristics were also found in Venezuela (Rodríguez-Quintal 2010), Curaçao (Luckhurst and Luckhurst 1978) and Puerto Rico (McGehee 1994). These studies suggest that cryptobenthic fishes have relatively well-defined habitat requirements, with niches often partitioned amongst ecologically similar species.

The associations between habitat variables and fish assemblages described in tropical cryptobenthic fish assemblages are similar to the well-defined habitat requirements in temperate species (e.g. Macpherson 1994; Syms 1995; La Mesa et al. 2006; Santin and Willis 2007). Temperate marine reserve studies also provide some evidence that the abundance of predators can reduce densities of cryptic fishes (Willis and Anderson 2003; but see La Mesa et al. 2006). In contrast, the effect of piscivores on tropical cryptobenthic fishes is poorly documented. The effects of interactions of cryptobenthic fishes with other components of tropical fish assemblages are also rarely investigated. For example, some damselfishes aggressively defend territories (Robertson 1996) and consequently alter the behaviour and abundance of many other species (e.g. Foster 1985; Sweatman 1985). Such competitive interactions could either decrease the abundance of some cryptobenthic fishes through direct agonistic encounters or indirectly increase the abundance of some species by damselfishes driving away competitors and predators.

This study investigates associations between Caribbean cryptobenthic fishes and a range of habitat and inter-specific interaction variables by re-analysing a data set from the U.S. Virgin Islands (Smith-Vaniz et al. 2006). The Smith-Vaniz et al. (2006) study represents one of the most thorough data sets of Caribbean cryptobenthic fish species, having used the ichthyocide rotenone to sample 58 sites along a series of biophysical gradients around Buck Island. The fish data are supported by the measurements of a range of biotic and abiotic habitat variables at each site. However, the original treatment for these data focused on characterising entire fish assemblages (both cryptic and non-cryptic) within pre-defined habitat types, documenting previously unrecognized biodiversity within the area, and comparing rotenone and visual census techniques to accurately determine the composition and relative abundances of the ichthyofauna.

As there are logistical and permitting issues associated with using ichthyocides to sample cryptobenthic fishes, combined with the need to describe and understand their habitat requirements, researchers should strive to analyse data sets as thoroughly as possible. We use an additional statistical technique (regression trees) to associate just the non-piscivorous cryptobenthic fishes with each of the habitat variables provided in the original study, a more detailed metric of wave exposure, and two additional factors representing potential interactions with other fish species (the density of territorial damselfishes and cryptic and non-cryptic predatory fishes). This study hypothesised that: (1) fish assemblages would be strongly influences by the abundance of live coral and habitat complexity, as demonstrated in other studies (e.g. Jones et al. 2004; Pratchett et al. 2008); (2) additional abiotic variables, particularly wave exposure, would be important drivers of cryptobenthic fish assemblages (e.g. Depczynski and Bellwood 2005b); and (3) cryptobenthic fish density would be negatively correlated with the density of territorial damselfishes and predators. Furthermore, we examine whether cryptobenthic fish assemblages can be distinguished amongst habitat types delineated by standard remote sensing techniques. Remote sensing is unable to map many of the detailed habitat variables recorded at the sampling site scale, but can generate large-scale habitat maps for use in conservation initiatives, such as planning networks of marine protected areas (McNeill 1994). If distinct cryptobenthic fish assemblages can be reliably correlated with habitats in large-scale seascape maps, this offers the possibility of using habitat types as surrogates of cryptobenthic fish diversity. Therefore, this study aims to establish new links between cryptobenthic fish assemblages and a range of abiotic and biotic variables in order to complement existing knowledge for larger species and increase understanding of the factors that shape entire fish assemblages on reefs.

Materials and methods

Study sites

Fish data were gathered around Buck Island, northeast of St. Croix. Buck Island Reef National Monument (BIRNM) was established in 1961 and contained a poorly enforced no-take zone that was extended to 7,339 ha in 2001 (Rogers and Beets 2001; Smith-Vaniz et al. 2006). For further details about the island, its reefs and the sampling strategy employed refer to Smith-Vaniz et al. (2006). Cryptobenthic fish assemblages at 58 sites were sampled around the BIRNM (Fig. 1) and were stratified across habitat types using habitat maps. However, one site (site 8) was removed from the analyses presented here because a heavy surge limited collections (Smith-Vaniz et al. 2006). The sampling sites were taken in six geomorphological zones: shoreline, nearshore reef, lagoonal patch reefs, backreef, forereef and reefs on the bank shelf.
https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0938-4/MediaObjects/338_2012_938_Fig1_HTML.gif
Fig. 1

Maps of Caribbean Sea, St. Croix, USVI, and Buck Island Reef National Monument, showing locations of the 57 rotenone collection stations from six geomorphological zones analysed in this study. Contour line indicates linear reef delineated in the NOAA habitat map (Kendall et al. 2001). Modified from Fig. 1 in Smith-Vaniz et al. (2006)

Fish sampling and definitions of individual assemblage components

Each sampling site was surrounded by a 3-mm mesh blocknet that was weighted at the bottom, had floats at the top and encompassed an area from 7.6 to 18.0 m2 (mean 15.1 m2). Between 0.5 and 1 kg of 8 % rotenone powder, mixed with sea water and liquid detergent, was then introduced into the net and dispersed into all microhabitats. Two divers collected all dead fishes within the net over the subsequent 30–45 min. Fishes were also collected outside blocknets, but only those from inside the nets were used here to allow analyses of fish densities. Specimens were subsequently identified, and the range of sizes of each species in each net measured to the nearest 0.1 mm (standard length). Full details of fish sampling and subsequent identification are reported in Smith-Vaniz et al. (2006). The resulting data set of 197 species contains both cryptobenthic and non-cryptic fishes. As the aim of this study was to examine non-piscivorous cryptobenthic species, 96 species were removed because they were judged to be non-cryptic or piscivores, resulting in 101 non-piscivorous cryptobenthic species being considered in this study (see Electronic Supplementary Material, ESM, for categorisation of each species).

Predator biomass is often considered the most appropriate metric for analysing trophic relationships (Bohnsack and Harper 1988), but whether this metric is more appropriate than predator density is not clear for cryptobenthic species. Furthermore, the data set does not report the size of each fish, which hinders the use of allometric scaling relationships. Predator density generated better fitting models during exploratory analyses, and therefore, the density of predatory fishes within each net was used as the measure of piscivory. Finally, territorial damselfish (Microspathodon chrysurus and Stegastes spp.) densities per site were calculated from abundance data and standardised for blocknet area.

Habitat characterisation

Videos and in situ surveys of the substratum in each blocknet were used to establish the per cent areal coverage of seagrass (Thalassia), sand, cobble, boulder, bedrock, dead coral, and live coral at each site (Smith-Vaniz et al. 2006). Relief was calculated as the height difference in metres between the shallowest and deepest points within the blocknet. Gorgonians were recorded as being present or absent. The predominant algal canopy height was evaluated as encrusting, medium height (up to 10 cm tall) or tall (>10 cm). Algal cover was assessed as sparse (<20 % areal coverage), moderate (21–80 % cover) or complete (>80 % cover).

Wave exposure at each site was calculated by measuring the fetch in each of eight directions (north, north-east, east, south-east, south, south-west, west and north-west) and calculating wave energy using linear wave equations (Ekebom et al. 2003). The switch point from fetch-limited to fully developed seas and equations to calculate wave energy at depth were based on Harborne et al. (2006a). Annual mean wind speeds from each direction, and the proportion of time that the wind blows from each direction, were calculated for the U.S. Virgin Islands as described in Harborne et al. (2006a, see also ESM). Depths at each site were used to calculate the wave energy experienced by benthic communities (i.e. estimated wave energy at the seafloor following attenuation through the water column), and the wave energy at sites behind the reef crest were reduced by 97 % to account for energy dissipation (Harborne et al. 2006a).

Data analysis

The cryptobenthic fish assemblage of 101 species was used to calculate number of species per m2 (subsequently referred to as number of species), total fish density and Shannon diversity at each of the 57 sample sites. We also derived the density at each site of the three commonest species (highest total density at the 57 sites) to investigate the autecology of individual species. In order to model these parameters in terms of the 14 putative abiotic and biotic controls (habitat complexity, percentage cover of seven substrata categories, presence or absence of gorgonians, algal height and cover, exposure, territorial damselfish density and predator density), we used boosted regression trees (BRTs). BRTs have a number of advantages over more traditional regression-based approaches that make them appropriate for use in this study, including improved explanatory power, being insensitive to irrelevant predictors and outlying data points, and the automatic modelling of interactions (see De’ath 2007; Elith et al. 2008). All models were fitted in (R Development Core Team 2008) using the gbm package (Ridgeway 2006) plus customised code written and described by Elith et al. (2008). In order to identify the best combination of the parameters required by BRTs (learning rate, tree complexity and bag fraction), we used cross-validation (described in detail in Elith et al. 2008). Cross-validation was automatically repeated for learning rates from 0.001 to 0.05 (steps of 0.002), tree complexities of 1–5 and bag fractions of 0.5 and 0.75, which span the range of likely optimal values (Elith et al. 2008), and the combinations which generated the lowest mean cross-validation deviances (but calculated from at least 1,000 trees) were used for the final models. Following derivation of full models with all variables, models were investigated to establish whether irrelevant predictors could be removed (procedure as detailed in Elith et al. 2008). With small data sets, such as in this study, simplification may improve model performance (Elith et al. 2008). Finally, we derived partial dependence functions for each predictor in the final, simplified model and, if tree complexity was greater than 1, identified the most important interactions.

The full data set of cryptobenthic fishes spans soft-bottom habitats very close to shore to deeper offshore forereef sites (Smith-Vaniz et al. 2006). In order to investigate whether different variables affect fish assemblages within reef habitats, compared to across the entire range of sites, we constructed additional BRTs for the number of species, total density and Shannon diversity on reef habitats alone. Reef habitats are likely to represent the most biodiverse areas in the seascape (Mumby et al. 2008), and so provide perhaps the best insights into factors affecting intra-habitat variation. To ensure that the reef subset of data was from a relatively homogeneous group of sites, exploratory analysis was conducted on the 14 predictor variables using non-metric multidimensional scaling within PRIMER (Clarke 1993). Categorical data were converted to values of 0 or 1 (absence/presence of gorgonians) or 1, 2 and 3 (increasing algal canopy heights and coverage) for this and subsequent PRIMER analyses. The non-metric multidimensional scaling identified two main groupings (ESM), roughly relating to inshore habitats (sites 1–12, 15, 16) and offshore reef habitats (sites 18–27, 29–58). Fish assemblage data from the latter set of sites were used for further analyses. The lower deviance within the data for reef sites and the smaller number of samples necessitated testing minimum learning rates of 0.0001 (to 0.005 in steps of 0.0002) and only using bag rates of 0.75. Furthermore, seagrass was not present at any of the reef sites and was excluded as a predictor variable within the BRTs.

In addition to the analyses of univariate assemblage parameters, we also constructed nonparametric multivariate regression trees of the entire data set using LINKTREE within PRIMER (Clarke et al. 2008). LINKTREE uses similarity profile (SIMPROF) tests to establish whether groups of sites have additional multivariate structure that needs explaining (Clarke et al. 2008). We used an additional constraint that groups consisting of fewer than four sites were not split further. Fish species characteristic of significantly different groups of sampling sites identified by LINKTREE were determined using Similarity Percentage (SIMPER) analysis (Clarke 1993).

Combining cryptobenthic fish assemblages with remotely sensed habitat maps

Buck Island Reef National Monument has been mapped by the National Oceanic and Atmospheric Administration (NOAA) National Ocean Service as part of a wider programme (Kendall et al. 2001). Briefly, 1999 aerial photographs were digitised and ground-truthed to assign one of 21 benthic habitat types to each polygon visible on the imagery and then reviewed for accuracy. Final maps were considered to be >90 % accurate. Using the GPS co-ordinates recorded in the field, we established the NOAA habitat type for each cryptobenthic sampling site (ESM). Note that the Kendall et al. (2001) maps were used to stratify the sites used in the Smith-Vaniz et al. (2006) study, ensuring that cryptobenthic fishes in a wide range of habitat types were sampled. ANOSIM (Clarke 1993) was then used to test whether the cryptobenthic fish assemblages in each NOAA habitat were significantly different.

Results

Overview of data set

The 101 cryptobenthic species, of which 72 species were cardinalfishes, blennies or gobies (Apogonidae, Blenniidae, Chaenopsidae, Labrisomidae and Gobiidae), consisted of >7,400 individuals. Overall, there was a mean of 1.45 species m−2 (SE = 0.09, range of 4–35 species within individual blocknets), mean total density was 9.28 fishes m−2 (SE = 0.88) and mean Shannon diversity was 2.39 (SE = 0.06). For the reef sites, there were means of 1.72 species m−2 (SE = 0.09, range of 13–35 species within individual blocknets) and 11.19 fishes m−2 (S.E. = 1.06), and a mean Shannon diversity of 2.55 (SE = 0.046). The three commonest species were Phaeoptyx conklini (freckled cardinalfish; mean density of 1.13 fish m−2, SE = 0.22), Stathmonotus stahli (eelgrass blenny; 0.92, SE = 0.31) and Coryphopterus tortugae (sand goby; 0.69, SE = 0.11), with densities of <0.5 fish m−2 for all other species. There was a mean of 0.38 territorial damselfishes m−2 (SE 0.049) and mean piscivore density was 0.68 m−2 (SE = 0.07) for all sites, and 0.44 damselfishes m−2 (SE = 0.049) and 0.73 piscivores m−2 (SE = 0.07) for reef sites. Mean exposure was 0.11 joules m−3 (SE = 0.02). Raw data for all other predictor variables are presented in the original study (Smith-Vaniz et al. 2006).

Univariate fish: habitat associations

The optimal parameters and properties for the final BRT models for each fish assemblage parameter varied, but learning rates were typically low (<0.01), tree complexities were either 1 or 2 and bag fractions were always optimal at 0.75 (Table 1). All models of parameters from the entire data set were improved by simplification, and the final models contained between 2 and 11 explanatory variables (Table 1). There were too few samples to permit simplification of reef habitat models. Furthermore, the predictor variables explained very little of the deviance in the Shannon diversity of reef assemblages, and optimal BRT parameters could not be derived to facilitate modelling. Models of assemblage parameters for the entire data set explained >36 % of the variation in assemblage parameters, and models for reef habitats explained >17 % of the variation in the number of species and total fish density (Table 1).
Table 1

Optimal parameters and properties for the final boosted regression tree models for each cryptobenthic fish assemblage parameter

Parameter

All sites

Reef sites

Number of species

Total fish density

Shannon diversity

Phaeoptyx conklini density

Stathmonotus stahli density

Coryphopterus tortugae density

Number of species

Total fish density

Learning rate

0.003

0.001

0.001

0.007

0.003

0.003

0.0011

0.0015

Tree complexity

1

1

1

2

2

1

1

1

Bag fraction

0.75

0.75

0.75

0.75

0.75

0.75

0.75

0.75

Minimum mean cv deviance

0.250

0.362

2.132

0.191

0.121

0.073

0.092

35.326

Mean optimal number of trees

1,020

4,920

3,040

1,200

2,780

1,140

5,200

5,380

Final number of model variables

8

2

8

12

11

8

No reduction

No reduction

Final cv deviance

0.242

0.347

2.082

0.179

0.111

0.069

% of deviance explained

50.1

36.3

41.8

46.1

59.3

68.4

17.9

22.7

cv cross-validation

Different combinations of biotic and abiotic variables were contained in the final models for each assemblage parameter, and variables had a range of partial dependence functions (Table 2, Figs. 2–4, ESM). The four most important variables in the final model (Figs. 2, 3, 4) always incorporated all variables explaining >10 % of the variation in the data set (Table 2). The cover of dead coral was the first or second most important variable in six of the eight models, and for the entire data set, the number of species and total fish density increased approximately linearly with increasing dead coral. However, there was a clear threshold for Shannon diversity, which increased dramatically when dead coral cover exceeded ~25 %. Similarly, the number of species and total fish density on reef habitats increased from close to minimum values to close to maximum values when dead coral exceeded ~55 and ~60 %, respectively (Fig. 4). An approximate linear increase in the density of S. stahli and C. tortugae with increasing dead coral cover was also one of the four most important variables in the models for these species, but dead coral was less important for predicting the density of P. conklini (Fig. 3). In contrast, live coral was only one of the four most important variables in models of the number of species and Shannon diversity of the entire data set, and total fish density on reef habitats decreased with increasing live coral cover. The presence of gorgonians did not generally influence fish assemblage structure, but did increase the number of fish species in all habitats. The density and diversity of fishes typically decreased with the cover of boulders and cobbles that characterised the shallow, inshore sites, but the density of C. tortugae and number of species and total fish density on reef habitats increased with increasing sand cover. P. conklini and S. stahli densities were low in very sheltered areas. All fish parameters tended to increase with increasing predator density, with the exception of densities of P. conklini, which were roughly negatively correlated. The effect of damselfish density was less than piscivore abundance, but the density of S. stahli and the number of species on reef habitats decreased with increasing damselfish densities. Because the tree complexity of most final models was 1, there were few interactions amongst variables. However, the effect of the most important interactions for the two models with a tree complexity of 2 (density of P. conklini and S. stahli) is shown in the ESM (Fig. 5).
Table 2

Percentage of deviance explained by each abiotic and biotic factor within the final boosted regression tree models for each cryptobenthic fish assemblage parameter

Parameter

All sites

Reef sites

Number of species

Total fish density

Shannon diversity

Phaeoptyx conklini density

Stathmonotus stahli density

Coryphopterus tortugae density

Number of species

Total fish density

Complexity

3.8 (+)

2.4 (+)

8.2 (c)

1.9 (c)

3.8 (+)

7.9 (+)

% Seagrass

% Sand

17.8

3.9 (c)

3.6 (+)

12.3

5.9

16.6

% Cobbles

13.7

0.7 (−)

15.4

0

0

% Boulder

12.4

12.8

45.3

0

0

% Bedrock

1.1 (−)

4.1 (−)

3.2 (+)

4.1 (−)

1.8 (+)

3.3 (+)

% Dead coral

44.5

77.2

34.8

6.9 (c)

13.9

13.9

28.5

19.4

% Live coral

5.5

32.7

6.9 (+)

7.4 (+)

2.5 (c)

4.5 (−)

Presence of gorgonians

11.9

1.2 (+)

0

0

Algal canopy height

0.2 (-)

0

0

Algal cover

8.1 (c)

3.9 (c)

3.1 (c)

1.2 (c)

Exposure

5.1 (c)

4.3 (−)

17.3

37.2

3.4 (−)

11.8

12.0

Density of damselfishes

1.5 (c)

0.5 (−)

8.2 (c)

10.3

3.7 (c)

4.4 (−)

9.5 (c)

Density of piscivores

26.6

22.8

6.2

10.1

5.2 (+)

2.1 (+)

38.3

25.7

Approximate relationship for each partial dependence function not shown in Figs 2, 3 and 4 indicated: + positive correlation, negative correlation, (c) complex correlation. Full partial dependence functions in ESM

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0938-4/MediaObjects/338_2012_938_Fig2_HTML.gif
Fig. 2

Partial dependence functions for the most important abiotic and biotic factors (up to four shown) influencing cryptobenthic fish assemblage parameters across all sampled sites. a Number of species, b total fish density and c Shannon diversity

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0938-4/MediaObjects/338_2012_938_Fig3_HTML.gif
Fig. 3

Partial dependence functions for the most important abiotic and biotic factors (up to four shown) influencing the density of the three most common cryptobenthic fishes across all sampled sites. aPhaeoptyx conklini, bStathmonotus stahli and cCoryphopterus tortugae

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0938-4/MediaObjects/338_2012_938_Fig4_HTML.gif
Fig. 4

Partial dependence functions for the most important abiotic and biotic factors (up to four shown) influencing cryptobenthic fish assemblage parameters across reef sites only. a Number of species and b total fish density

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0938-4/MediaObjects/338_2012_938_Fig5_HTML.gif
Fig. 5

Results of LINKTREE analysis highlighting the abiotic and biotic factors best explaining the 10 multivariate assemblages apparent within the cryptobenthic fish data collected at the 57 sites. Figures in parentheses indicate assemblage group, and associated figures indicate the sampling stations contained within that grouping. Lower section of the figure indicates the three most characteristic species within each grouping as identified by SIMPER analysis, and figures in parentheses indicate percentage contribution to the pairwise differences (percentage contribution = average squared distance/average dissimilarity between assemblages)

Multivariate fish: habitat associations

LINKTREE identified 10 distinct cryptobenthic fish assemblages driven by significant thresholds within the environmental data (Fig. 5, ESM). The initial major separation in fish assemblages was between inshore areas with little dead coral and reef areas further offshore. Shoreline sites were then split depending on whether they had high or low algal cover and had more or less topographic complexity. Reef sites were separated depending on their exposure, the amount of dead coral, the density of damselfishes, density of piscivores and bedrock cover. SIMPER analyses demonstrated that a range of species were the characteristic of the 10 assemblages within the entire data set, but Gobiesox punctulatus was the most characteristic species of shoreline habitats with low algal cover, and Paraclinus nigripinnis was the most characteristic species of shoreline areas with higher algal cover (Fig. 5, ESM). The three cryptobenthic fishes considered at species level during the univariate data analyses (P. conklini, S. stahli and C. tortugae) were amongst the three most characteristic species of each of the reef habitats. However, Apogon townsendi was the most characteristic species in sheltered reef areas with <50 % dead coral cover, and Starksia lepicoelia and Enneanectes altivelis were the characteristic of relatively sheltered areas with high dead coral cover and high and low damselfish densities, respectively.

Cryptobenthic fish assemblages spanned eight different NOAA habitat types (ESM). There were significant differences in the multivariate cryptobenthic fish assemblages amongst these eight habitat types (ANOSIM Global R = 0.405, P = 0.001). Furthermore, there were significant (P < 0.05) differences in cryptobenthic fish assemblages amongst 16 of the 28 (57.1 %) pairwise comparisons amongst habitats. The habitats of ‘Colonized bedrock’ and ‘Seagrass’ are close to shore (sites 1–15, 22, 28, 30, 32) and their assemblages were distinct from all other habitat types further offshore, except ‘Uncolonized bedrock’ that was only represented by a single station (site 16). Similarly, the cryptobenthic fish assemblage in ‘Colonized pavement with sand channels’ was significantly different from all other habitat types, with the exception of ‘Uncolonized bedrock’.

Discussion

A paradox in the literature on cryptobenthic fishes is that many micro-habitat requirements are better understood than larger-scale patterns (Depczynski and Bellwood 2005b), but our analyses provide new insights into inter-habitat variability in Caribbean assemblages. The major role of dead coral cover in affecting the density, diversity and multivariate structure of cryptobenthic fish assemblages was surprising. Live coral cover has been repeatedly linked to the structure of reef fish assemblages (e.g. Luckhurst and Luckhurst 1978), and the impacts of coral mortality also have negative effects on the diversity of fishes (e.g. Jones et al. 2004; Pratchett et al. 2008). Conversely, the link between fish and dead coral is rarely reported. Caribbean reefs have lost much of their live coral cover since the 1970s because of factors including over-fishing, disease and coral bleaching (Gardner et al. 2003). Although this loss has led to an increase in macroalgal cover and decreased rugosity (Hughes 1994; Alvarez-Filip et al. 2009), dead coral still provides more shelter for cryptobenthic fishes than is available in soft-bottom habitats. However, sampling sites had live coral cover ranging from 0 to 60 %, and it was expected that this variable would have a more important role than dead coral in affecting cryptobenthic fish assemblages. The metric of habitat complexity also varied by an order of magnitude, but was of minor importance in all final models. Therefore, we suggest that dead coral has some intrinsic properties that increase the abundance of small fishes, rather than just providing the only available structure in the absence of live coral.

Definitively identifying why dead coral is important to cryptobenthic fishes requires more research, but a combination of at least six factors is likely to be important. Firstly, if dead coral is primarily comprised of structurally complex species, such as Montastraea annularis or Acropora palmata that are susceptible to mortality from bleaching (McField 1999) or disease (Aronson and Precht 2001), then cover of dead coral may simply be an excellent proxy of the reef rugosity and shelter preferred by cryptobenthic fishes. Thus, fishes may not actively seek corals because they are dead. Consequently, the link between cryptobenthic fishes and dead coral cover may be a recent phenomenon on Caribbean reefs, and historically this link may have been much weaker. Secondly, cryptobenthic fishes include herbivores, and dead coral is quickly overgrown by turf and macroalgae that are not able to establish on live coral. Thirdly, macroalgae can support high densities of invertebrate prey (Stoner 1985), and food for invertivores may be more abundant than amongst the branches of live corals. Fourthly, many cryptobenthic species are detritivores, and dead corals may be sites of detrital accumulation (Wilson 2001). Fifthly, coral bioerosion is most prevalent in dead coral (Tribollet and Golubic 2011), and the additional holes may provide extra shelter for cryptobenthic fishes. Finally, live coral may harbour more predators than dead coral, although live coral cover was not correlated with piscivore density within the sampling sites (P > 0.05).

Such hypotheses suggest that cryptobenthic species may have benefited from the widespread loss of coral cover in the Caribbean, but identifying such trends requires carefully controlled monitoring studies that integrate consideration of factors such as reduced recruitment and increased predation of fish settling on dead coral (e.g. Feary et al. 2007). Furthermore, our study was conducted in a protected area that may have higher grazing pressure and lower macroalgal cover than equivalent reefs elsewhere in the region, as demonstrated in other reserves (Mumby et al. 2006). On reefs with lower grazing pressure, higher algal cover may have more detrimental effects on cryptobenthic fishes. Reefs without live coral growth will also eventually lose topographic complexity through bioerosion (Alvarez-Filip et al. 2009), reducing hiding places for cryptobenthic fishes. Clearly, live coral cover is also vitally important for maintaining other components of fish assemblages and a range of ecological processes (Harborne et al. 2006b).

The presence of live coral did increase the number of species and Shannon diversity of cryptobenthic fishes, and the major effect was an increase in diversity when live coral cover exceeded ~5 %. Presumably, this threshold represents sufficient coral colonies to support specialist species that are found exclusively in live coral. Little is known about the natural history of many of the species studied here, but specialist coral-dwelling cryptobenthic species have previously been described in Indo-Pacific assemblages (Munday et al. 1997; Jones et al. 2004). Our data suggest that the presence of live coral is also important for maintaining the diversity of Caribbean cryptobenthic fish assemblages. The increased number of species seen at sites where gorgonians are present suggests that gorgonians also support specialist fishes, or that both gorgonians and specialist fishes are associated with some other habitat feature. Close associations between fish species and single gorgonian species have been documented in the Pacific, and a reliance on this patchy microhabitat has led to adaptations in reproductive behaviour (Munday et al. 2002). Despite their abundance on most Caribbean reefs and reports of close associations between gorgonians and species such as Monacanthus tuckeri (Lieske and Myers 1994), there are few data on the interactions between gorgonians and fishes (but see Rilov et al. 2006). However, it appears that large amounts of dead coral increase the density and diversity of cryptobenthic fishes more than live coral and gorgonians lead to the presence of additional specialist species. Indeed, the total density of fishes at reef sites decreases with increasing live coral cover, although this factor is of limited importance within the final model.

While the role of habitat variables on tropical cryptobenthic fish assemblages is poorly documented, the interaction between cryptic and non-cryptic fishes is almost totally unknown. Our data suggest that, where they are most abundant in the reef habitats, damselfishes decrease the number of cryptobenthic species and the density of some individual species. Damselfishes aggressively defend their territories against a wide range of species, but agonistic interactions are strongest with direct competitors, such as species that graze on the algae ‘farmed’ within their territories or threaten their egg masses (Thresher 1976). Some cryptobenthic species are herbivores or are likely to feed on fish eggs, so agonistic interactions with damselfishes seem inevitable. However, our sample sites cover reef habitats containing multiple territory and non-territory areas, and further work should focus on any differences in cryptobenthic fishes inside and outside individual damselfish territories. In contrast to some negative effects of damselfishes, any effects of piscivore density were typically positive, although there is some evidence that densities of P. conklini were decreased when predator densities were high. It seems likely that predators do not have a significant negative impact on most cryptobenthic fish populations, and their density is simply correlated with the same habitat variables that affect the abundance of smaller species. Additionally, predators may be attracted to areas with high densities of cryptobenthic fish prey. Alternatively, trophic cascades of larger piscivores within the guild may decrease the activity of smaller predators that feed on the cryptobenthic species so that piscivory is reduced in areas with high densities of predators. Such trophic cascades have been documented in reef fishes (Stallings 2008), but seem unlikely in our study because even the larger piscivores recorded in the blocknets (e.g., small-bodied serranids), feed on gobies and blennies (Randall 1967). Assessing the effects of piscivory on cryptobenthic fishes is hampered by a lack of a clear understanding of which species are the major predators and size data for each individual predatory fish. Additionally, the areas surrounded by blocknets are much smaller than the home ranges of many predators (Popple and Hunte 2005), and the actual predation experienced by fishes at each site may be significantly different than that indicated by the density of predators within our samples. It is also likely that the predator–prey interactions include subtle, density-dependent effects that are not captured in our analyses (e.g. Forrester and Steele 2004).

Little is known about the autecology of cryptobenthic species (Santin and Willis 2007), which makes it difficult to suggest mechanistic links between species’ densities and the different biotic and abiotic variables contained in the models for P. conklini, S. stahli and C. tortugae. Wave energy was previously highlighted as an important influence on tropical cryptobenthic fishes (McGehee 1994; Depczynski and Bellwood 2005b) and was the major factor influencing the density of P. conklini and S. stahli. Presumably, the low densities of these two species in very calm water and then the small decline of their densities with increasing exposure reflect similar factors to those suggested in the Pacific (Depczynski and Bellwood 2005b). These are the direct effects of water motion on the swimming performance of many fishes, and the indirect effects of wave energy on benthic communities and habitat structure. P. conklini is a shallow reef dweller (Greenfield and Johnson 1990a; Froese and Pauly 2010), and the strong negative correlation with boulders and cobbles probably reflects the abundance of these microhabitats in shoreline, non-reef sites. C. tortugae also appeared to avoid areas with high cover of boulders and cobbles, while being more abundant in sandy areas with large amounts of dead coral.

The logistical constraints of sampling cryptobenthic species limit the number of locations and times that can be sampled for this group of reef fishes. While there will obviously be geographical variation and temporal fluctuations driven by factors such as the seasonality and stochasticity of larval recruitment, we suggest that the distinct assemblages in shoreline versus reef habitats, the abundance of cryptobenthic fishes in areas with high areal cover of dead coral, and the variation in habitat requirements amongst species apply to most Caribbean reefs. Despite these general patterns, many of the requisite environmental parameters are difficult to map at large scales if researchers wish to infer the distribution of cryptobenthic species for objectives such as capturing fish biodiversity within reserve networks, although new remote sensing platforms can generate large-scale environmental data layers that are important predictors of fish assemblages (e.g. Pittman et al. 2009; Pittman and Brown 2011). Critically, different habitats identified on a map generated through standard remote sensing techniques contained distinct cryptobenthic fish assemblages, although only in situ sampling will reveal the identity and relative abundances of the actual species within these assemblages. Thus, habitat maps can provide insights into general patterns of cryptobenthic fish diversity across a seascape, and the generic planning principle of including each habitat type within marine protected areas is likely to capture cryptobenthic biodiversity as it does for larger fish species (Roberts et al. 2003). Although such insights are important, many questions remain outstanding. Establishing what mechanisms drive the intra- and inter-habitat variation in the composition and function of cryptobenthic fishes, and how they are impacted by anthropogenic threats, is vital for ensuring that their role on reefs is both understood and conserved.

Acknowledgments

ARH is grateful to the Marine Spatial Ecology Laboratory for discussions which improved the paper, especially Y.-M. Bozec and P. Sutcliffe. We thank N. Wolff for preparation of Fig. 1 and Figure E4.1 in the ESM. ARH was supported by Fellowship NE/F015704/1 from the Natural Environment Research Council. We are grateful to NOAA/NOS, C. Caldow, J. Christensen, M. Monaco and Z. Hillis-Starr for generous assistance with the original study and permitting process (permit # BUIS-2001-SCI-0003). Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Supplementary material

338_2012_938_MOESM1_ESM.doc (224 kb)
Supplementary material 1 (DOC 224 kb)
338_2012_938_MOESM2_ESM.doc (28 kb)
Supplementary material 2 (DOC 28 kb)
338_2012_938_MOESM3_ESM.doc (24 kb)
Supplementary material 3 (DOC 24 kb)
338_2012_938_MOESM4_ESM.doc (190 kb)
Supplementary material 4 (DOC 190 kb)
338_2012_938_MOESM5_ESM.doc (984 kb)
Supplementary material 5 (DOC 984 kb)

Copyright information

© Springer-Verlag 2012