Fear effects and group size interact to shape herbivory on coral reefs
Bauman, Andrew et al. (2021), Fear effects and group size interact to shape herbivory on coral reefs, Dryad, Dataset, https://doi.org/10.5061/dryad.vdncjsxtf
1. Fear of predators (‘fear effects’) are an important determinant of foraging decisions by consumers across a range of ecosystems. Group size is one of the main behavioural mechanisms for mitigating fear effects but also provides foraging benefits to group members. Within coral reef ecosystems, fear effects have been shown to influence the feeding rates of herbivorous fishes (i.e. browsers), a key functional group that prevent macroalgal overgrowth. Yet, how fear effects and group size interact to shape macroalgal removal on coral reefs remain unclear.
2. Here, we conducted field-based experiments using models of a common piscivorous fish, the leopard coral grouper (Plectropomus leopardus), and a series of macroalgal (Sargassum ilicifolium) assays positioned at increasing distances from the models (1, 2, 3 and 4 m) on two Singaporean coral reefs to investigate how acute fear effects shape the intensity of herbivory, and whether these effects were influenced by variation in the group size of herbivorous fishes feeding on the assays.
3. We found acute fear effects strongly influenced the foraging behaviour of herbivorous fishes over small spatial scales. Rates of Sargassum biomass removal, feeding rates and the total number of individual feeding events were all lower near the predator model. These effects dissipated rapidly with increasing distance from the predator model, and were undetectable at a distance of 4 m. We also found generally larger group sizes of herbivorous fishes further from the predator model presumably reflecting decreased risk. Further, the number of individual bites-event-1 increased significantly with increasing group size for two common browsing fishes, Siganus virgatus and Siganus javus.
4. Our findings highlight that acute fear effects influence the distribution and intensity of herbivory over small spatial scales. Fear effects also interacted with herbivore group size resulting in changes in the number of individual feeding events and bite rates that collectively shape the realised ecosystem function of macroalgal removal on coral reefs. Group size is an important context-dependent factor that should be considered when examining fear effects on coral reefs.
We conducted field-based experiments between September and October 2016 on Pulau Satumu and Kusu, two off-shore islands in Singapore with well-developed fringing reefs. Both reefs have a clearly defined reef crest at 3–4 m depth, and have the highest coral cover, the lowest macroalgal cover, and highest rates of herbivory in Singapore . Each experimental replicate consisted of a series of individual Sargassum ilicifolium assays positioned at increasing distances (1, 2, 3 and 4 m) from models of the piscivorous leopard coral grouper (Plectropomus leopardus, 53 cm total length, TL) to simulate different levels of acute predation risk, together with two experimental controls (i.e. object control and herbivore exclusion). Plectropomus leopardus was selected because this species is common on both Pulau Satumu and Kusu and have broad diets that include herbivorous fishes. The size of the models (53 cm TL) was selected to represent the maximum size of serranids (including P. leopardus) observed on Singaporean reefs. Sargassum ilicifolium was selected because it is the most abundant and widespread Sargassum species on Singapore reefs.
Sargassum ilicifolium thalli of similar heights (ca. 40 cm) were collected daily from a nearby shallow reef flat on Pulau Hantu, Singapore. Individual thalli were spun in a salad spinner for ~20 s to remove excess water and the wet weight recorded to the nearest 0.1 g. The initial mass (mean ± se) of each thalli was 44.7 ± 8.4 g. For each experimental replicate, six Sargassum assays were allocated randomly to one of three treatments: a predator model treatment (four assays positioned 1, 2, 3 and 4 m away from the predator model), one object control treatment (53 cm length of PVC pipe, 8 cm in diameter) with one assay positioned 1 m away where the largest effect on browsing was theorised to occur, and a herbivore exclusion treatment (one assay placed inside a 30 cm radius, 100 cm height, 0.5 cm plastic mesh cage). The object control was used to account for the effect of introducing a novel object in the water while the herbivore exclusion cage was used to account for the autogenic losses due to handling and translocation. A negative control treatment (i.e. a series of four assays separated by 1 m without a predator or novel object) was not included in this study because, with replication, there was no conceivable reason why browsing would consistently vary within a 4 m scale in the absence of any object.
Each morning (09:30–10:30) we transplanted two replicates of six Sargassum assays (total of 12 assays) haphazardly along the reef crest at ~3–4 m depth at one site (i.e. either Pulau Satumu or Kusu). Predator models were secured ~50 cm above the reef substratum. Individual Sargassum assays were subsequently attached to the reef substratum at increasing distances (1, 2, 3 and 4 m) from the predator model. The two additional assays were positioned approximately 20 m (object control) and 30 m (herbivore exclusion control) away from the predator models within the same habitat (i.e. ~3–4 depth along the reef crest). Within each site, experimental replicates were separated by a minimum of 30 m to facilitate independence. This procedure was replicated over four non-consecutive days on each reef (n = 8 experimental replicates, with n = 4 per reef).
To identify herbivorous fish species feeding on the Sargassum assays, a small video camera (GoPro) mounted on a dive weight (2kg) was positioned approximately 1m from each of the assays in the predator exposure treatment (i.e. 1, 2, 3 and 4 m from the predator model). Filming commenced immediately after the assays and predator models were deployed, with a small scale bar (10 cm) placed adjacent to each assay for 10 s to allow calibration of fish sizes on the videos. All cameras, macroalgal assays and predator models were collected after 4.5 h. Thus on each day of the experiment there were 8 cameras per reef, resulting in 144 h of video observations for each reef (288 h in total).
Following retrieval, each individual Sargassum thalli was spun and re-weighed as above to calculate biomass loss per thallus (section 2.3). To minimize potential diver interference the first 20 min and last 10 min of each video were discarded. From the video footage, we recorded the total number of bites, species and estimated TL to the nearest cm for each fish feeding, group size per feeding event, and total bites per feeding event. Size estimates for each fish were converted to biomass using published length-weight relationships. A feeding event was recorded every time a fish entered the video frame and fed on Sargassum, and the bites from each individual fish were counted until each fish left the video frame. If other fishes entered during the feeding event, bites taken by those individuals were counted and included within the same feeding event. Group feeding was defined as 2 or more fishes feeding simultaneously feeding during an event. To account for variation in the feeding impact of individual fishes related to body size, mass-standardized bite impact was calculated as the product of the number of bites and the estimated body mass (kg) for each individual following Hoey & Bellwood (2009).
Individual assays positions within each predator exposure treatment replicate (i.e., at 1, 2, 3 and 4 m) were considered non-independent due to their close proximity, and hence potential exposure to the same individual herbivorous fishes. To account for non-independence, we used a Bayesian mixed modelling approach employing Markov chain Monte Carlo (MCMC) methods for fitting generalized linear mixed models (Hadfield, 2010) with experimental replicate defined as the random effect. To examine the response of herbivorous fishes to the predator model, we compared: i) changes in Sargassum biomass at each position away from the predator model and the object control, ii) herbivorous fish species from the video footage feeding at each assay position from the predator model. For all analyses, assay position was considered an ordinal factor rather than a continuous covariate and the five positions were modelled for analyses of biomass removal (i.e. 1 m from the object control and 1, 2, 3, and 4 m from the predator model).
To examine biomass (g) loss due to herbivory at each assay position, data were first standardized to control for autogenic loss during handling following Cronin & Hay (1996). For individual assays in each replicate, the reductions in macroalgal biomass attributed to herbivory was calculated using the following formula: [(Ho x Cf/Co) - Hf] where Ho and Hf were the initial and final wet weights, respectively, of the macroalgal assay exposed to browsing, and Co and Cf were the initial and final masses of the corresponding assays from the herbivore exclusion treatments. Changes in Sargassum biomass were compared by modelling the absolute (g) and relative (proportion) reduction in biomass of replicate assays. In the latter case, proportions were logit transformed (Warton & Hui, 2011). Changes in biomass data were modelled using a Gaussian error structure with site, position and their interaction as fixed effects in initial models.
From the video feeding observations, we modelled the following three response variables: (i) counts of bites per feeding event (bites.event-1), (ii) feeding rates (mass-standardized bites.hour-1 hereafter ‘ms-bites’) and (iii) group size per feeding event (group-size.event-1). Bites.event-1 was modelled to assess whether individual foraging events were affected by distance to the predator models, whereas feeding rates indicated the overall effect of predator on macroalgal removal at each position. Bites.event-1 were modelled for the four most common herbivores (Siganus virgatus, Kyphosus vagiensis, Scarus rivulatus, and Siganus javus) using a Poisson error structure. Group sizes >4 were excluded from the analysis due to lack of cases across other explanatory variables. The initial model included the explanatory terms site, group size, assay position, their three-way interaction and pairwise two-way interactions and terms for species and species/group size interaction. Feeding rates were only analysed for S. virgatus because this species was responsible for most of the feeding (see Results). Feeding rates (ms-bites) were rounded to whole integers to employ a Poisson error structure, and site, assay position and their interaction were used as explanatory variables in the initial model. Analysis of group size (group-size.event-1) was performed for the entire dataset, including group sizes >4, with the initial model including the explanatory terms site, assay position and species, with site/assay position and site/species interactions, using a Poisson error structure.
Models were fit using the MCMCglmm package, which provides parameter estimates, parameter 95% highest posterior density (HPD) credible intervals and a p-value (pMCMC) corresponding to the smaller of two times the probability that the MCMC parameter estimate is either >0 or <0 (Hatfield, 2010). Model terms were considered significant where one or more levels (for factors) or a co-variate had a pMCMC value <0.05 and parameter estimate 95% HPD’s did not include zero. Diffuse, uninformative inverse gamma priors were used for variance components and default priors for fixed effects. Backward model selection was applied from initial models (defined above) by comparing reduced nested models with deviance information criterion (DIC) and model weight. Top ranking models were compared, and the most parsimonious selected as the simplest model with explanatory terms significant. Presented results are predicted posterior means and their 95% HPD intervals unless otherwise stated. Comparisons between factor levels in final models were considered statistically significant when prediction 95% HPD’s did not overlap prediction estimates of other factor levels. All models were fit using 130,000 iterations, a burn-in of 30,000 iterations and a thinning interval of 50, except for feeding rates and group size, which required longer iterations, burn-ins and thinning intervals due to poor mixing and unacceptably high autocorrelation between thinned samples in models with fewer iterations. Diagnostics were performed by visual inspection of trace plots and ensuring autocorrelation between thinning intervals was low (i.e. at least <0.05). Model comparisons were made using the MuMIn package (Barton, 2015). All data were analysed in R (R Development Core Team, 2017). The R code, full model selection details, parameter estimates, 95% HDP’s and pMCMC values are provided in electronic supplementary materials.
AXA Research Fund, Award: 154-000-649-507
National Research Foundation Singapore, Award: MSRDP-P03