Temporal comparison and predictors of fish species abundance and richness on undisturbed coral reef patches

Study site

Ras Mohammed National Park is located at the southern tip of the Sinai Peninsula in the Red Sea, Egypt (Fig. 2). Established as a protected area in 1983, the park spans an area of 480 km² including coral reef, coastline, mangrove and desert habitats (SEAM, 2005). Threats to the area include shipping traffic, oil spills, invasive species (e.g., the crown-of-thorns starfish, Acanthaster planci), climate change (e.g., increased temperatures causing coral bleaching), and tourism (Ormond et al., 1997; SEAM, 2005; Tilot et al., 2008). However, Mersa Bareika (27°47′23.9″N, 34°13′02.8″E), located inside the Ras Mohammed National Park, is a designated research area, which restricts human activity occurring in the region and was closed to tourists during the study period. No natural or human disturbance is known to have occurred in the study area during the ten years between surveys. An outbreak of A. planci in 1998 affected some of the reefs in Ras Mohammed National Park. However, the Egyptian Environmental Affairs Agency reacted quickly to the threat, thus minimizing the impact at this site (PERSGA/GEF, 2003). No A. planci were observed on the patch reefs in Mersa Bareika throughout the outbreak during surveys by RB in 1998. Similarly, the 1998 global coral bleaching event, which caused substantial damage in the southern Red Sea and Gulf of Aden, did not affect the northern Red Sea (Kobt et al., 2004). Permission to carry out research in Mersa Bareika was granted to RB by the Egyptian Environmental Affairs Agency, locally verified by the authorities at Ras Mohammed National Park.

Habitat surveys

Mersa Bareika is a narrow fringing reef along the coast of a well-protected bay (Fig. 2). Sand enters the water through wadis channels that flood during the rainy season, creating small patch reefs separated by at least 5 m of sand. Forty seven patch reefs (depth 0.68 m to 7.57 m) spanning 450 m of shoreline were surveyed in November 1997 (see Bshary, 2003) and again between September and November 2007 (Fig. 2 and Table S1). Average depth below the surface was estimated for each patch with a dive computer. Patch volume was estimated following methods described in Bshary (2003). Maximum patch length, width, and average height above the substrate were used to calculate volume according the closest geometrical shape of the patch (e.g., rectangle, ellipse). Percentage live coral cover on each patch was estimated using a modified line-intercept method (English, Wilkinson & Baker, 1997): a transect tape was laid out across the longest section of each patch and coral cover estimated as the proportion of the transect tape that was laid over live coral. The tape was then set perpendicular to this axis, across the widest area of the patch, and the process repeated. The total percentage area of live coral cover was calculated as the mean of these two measures. The total percentage of holes and caves on each patch was also estimated in this way. Holes were defined as small cavities (typically 1–3 cm wide) on the surface of the reef matrix (excluding spaces in between coral). Caves were defined as openings at the base of the reef where the reef matrix connected to the substrate.

Fish surveys

Coral reef fishes were categorized into four groups depending on their movement patterns and habitat use (Table S2; Bshary, 2001). Resident fishes have small home ranges restricted to a single patch reef and include damselfishes (Pomacentridae), cardinalfishes (Apogonidae), squirrelfishes (Holocentridae) and some wrasses (Labridae). Visitors have larger home ranges that overlap several patches, and include parrotfishes (Scaridae), goatfishes (Mullidae) and fusiliers (Caesionidae). Butterflyfishes (Chaetodontidae) remain on one patch for long periods (hours to days) but also move between patches, and were considered facultative visitors (i.e., intermediate home ranges). Species that we could not classify into any of the three categories were categorized as undetermined (e.g., Cyrrithidae, Tetraodontidae).

Reef fish abundance and species richness were surveyed by two SCUBA divers three times in 1997 and five times in 2007, at two week intervals. One observer identified species and recorded their abundance at each patch. The second diver also recorded species’ identity to ensure the primary observer had not missed a species. Upon arrival at a patch, observers recorded fishes in the following order: visitors, facultative visitors, residents and undetermined. The abundance of species occurring in shoals or schools was estimated in multiples of ten. Fishes that swam to patches after the count had begun were not recorded. Cryptic species including most blennies (Blenniidae) and gobies (Gobiidae) were excluded from the surveys except for the scale-eating blenny (Plagiotremus tapeinosoma), the bluestriped blenny (Plagiotremus rhinorhynchos), the mimic blenny (Aspidontus taeniatus), and the blackfin dartfish (Ptereleotris microlepis). Species that primarily occupy sandy areas such as pipefishes (Syngnathidae) and lizardfishes (Synodontidae) were also excluded.

Statistical analyses

We examined differences in percent live coral cover between 1997 and 2007 on the 47 patches using a paired t-test. Percent live coral cover was logit transformed (Warton & Hui, 2010). We used the Information Theoretic approach (Burnham & Anderson, 2002; Johnson & Omland, 2004; Grueber et al., 2011) to determine which variables best explained variation in fish abundance (number of individuals) and species richness (number of species) among patch reefs and years. Response variables were averaged across fish surveys within year. Therefore, the data were not Poisson distributed or zero inflated. Two full models were constructed using general linear mixed-effects models (LMM; lmer function in R). Model assumptions were checked with plots of residuals vs. fitted values and qqplots of residuals for fixed and random effects. Fish abundance was log transformed to meet model assumptions. Seven predictor variables were included in each model: survey year, depth below surface, patch volume, percent live coral cover, percent caves, percent holes and cleaner wrasse abundance. Cleaner wrasse abundance was not included in the fish abundance counts. We graphically examined all two-way interactions and excluded those that had no apparent effect on the response variables for model simplicity and parsimony following recommendations in Burnham & Anderson (2002). We examined non-linear relationships using pairs plots and plots of generalized additive models (GAM). Quadratic terms were included where necessary. Patch ID was included as a random factor in both models. Continuous predictors were z-standardized (mean = 0, SD = 1) and a correlation plot used to check for collinearity. All correlation coefficients were <0.35. See Tables S3 and S4 for details of the main effects and interaction terms included in the full models.

We used the Akaike Information Criterion modified for small sample sizes (AICc) to select the best candidate models for each response variable with the “dredge” function in the R package MuMIn. We performed model averaging using the “model.avg” function (MuMIn) when the normalized Akaike weight value (w_im) of the best model was <0.9 (Burnham & Anderson, 2002). The confidence set of candidate models selected for model averaging included all models for which the w_im fell within 10% of the maximum normalized weight, suggesting that these models had substantial support in explaining the data (Burnham & Anderson, 2002). We calculated the marginal R² (variance explained by the fixed factors; $R_{\mathrm{GLMM}\left(m\right)}^2$) and conditional R² (variance explained by the fixed and random factors; $R_{\mathrm{GLMM}\left(c\right)}^2$) following Nakagawa & Schielzeth (2013). All analyses were done in R 3.1.2 (R Development Core Team , 2014). The data and code for this study are deposited in the public respository figshare (Wagner et al., 2015) following best practices (White et al., 2013; Roche et al., 2015).