Spatial patterns of coral survivorship: impacts of adult proximity versus other drivers of localized mortality

Study site characteristics

This study was conducted on reef flats within no-take marine protected areas (MPAs) adjacent to Votua, Vatu-o-lailai, and Namada villages along the Coral Coast of Viti Levu, Fiji. These reserves are scattered along 11 km of fringing reef and are separated by ∼3–8 km. The reserves have high coral cover (38–56%), low macroalgal cover (1–3%), and a high biomass and diversity of herbivorous fishes (Rasher, Hoey & Hay, 2013; Bonaldo & Hay, 2014). The reef flats range from ∼1–3 m deep at high tide, exposed to ∼1.5 m deep at low tide, extend ∼500–600 m from shore to the reef crest, and are typical of exposed reef flats occurring throughout Fjii.

Except during low tides in calm weather, waves push water over the reef front, and water then flows laterally across the reef flats to discharge through channels bisecting the flats. This creates a relatively predictable current direction at most locations.

Survival experiments

To test whether juvenile corals experienced distance-dependent mortality near adult conspecifics, we created ∼5 mm tall fragments of P. damicornis and S. hystrix, selected suitable adult focal colonies (defined below), attached conspecific fragments 3, 12, 24 and 182 cm up- and down-current from each focal adult, and monitored fragment survival. General direction of current (east to west) was determined by in-water observations over ∼7 years of working at this site. We conducted this experiment in Votua village’s MPA, which supports a diverse assemblage of corals covering about 50% of hard substrates (Rasher, Hoey & Hay, 2013). Water flow across this reef flat is often negligible around low tide, reducing the potential for constantly dissipating microbial enemies away from focal colonies.

We used pliers to clip 16 fragments of 30–40 polyps each from the tips of each of 24 large P. damicornis and 24 large S. hystrix colonies in the Votua village MPA. We employed fragments from older colonies as proxies for ∼6 month old juveniles (Sato, 1985) because, despite these species reproducing monthly in some locations (Fan et al., 2002; Kuanui et al., 2008), neither species planulated at our site during the months of this study (August through October 2013). The fragments from each of four source colonies for a species were collected in six rounds over two days. Each round was taken to shore and four fragments (one from each source colony) were epoxied (Emerkit epoxy) onto the unglazed side of 16 2.54 × 2.54 cm tiles. Thus, each tile had fragments from four different colonies and sets of 16 tiles had fragments from the same four colonies of the same species. After epoxying, tiles were held in a tub of seawater for ∼1 h, allowing the epoxy to harden. Tiles were then cable-tied onto metal racks at ∼1 m deep in the MPA and allowed to acclimate for two weeks before deployment in the experiment. Survivorship during acclimation was 100%, producing 384 fragments on 96 tiles for each coral species.

Within the MPA, 10 adult P. damicornis and 10 adult S. hystrix colonies served as focal colonies. Focal colonies: (i) were >10 cm at their smallest diameter (10 to 35 cm for P. damicornis and 10 to 75 cm for S. hystrix), (ii) had no conspecific colonies within 4 m (so as not to confound effects of the focal colony with effects of nearby conspecifics), (iii) were 5–40 cm deep at low tide, and (iv) had space for 190 cm PVC pipes to be placed roughly east and west (the predominant current direction was west) without disturbing other corals. Focal colonies were photographed from above and their size determined using ImageJ (Rasband, 1997).

Twenty mm diameter by 190 cm long PVC pipes served as platforms to which we attached the tiles. Pipes were anchored to the reef by driving steel rebar through pre-drilled holes and cementing the rebar to the pipe. Notches 2.54 cm long allowed us to cable-tie tiles onto the pipes at distances of 3, 12, 24 and 182 cm from focal colonies (Figs. 1A and 1B). This approach secured all pipes and tiles throughout the experiment. These distances and this scale were chosen to match a previous experiment in the Caribbean that had detected distance-dependent mortality of newly settled recruits for a broadcast spawning coral (Marhaver et al., 2013).

Tiles were randomly assigned to positions on pipes. Thus, fragments at each distance and around each conspecific focal colony were random with respect to source colony. Unassigned tiles were kept on the rack as spares (64 fragments on 16 tiles for each coral species).

Every 1–2 d after deployment, we examined all fragments, recording survivorship, consumption, overgrowth by algae, bleaching, or other changes in status. The P. damicornis fragments were observed for 59 days. The S. hystrix fragments were deployed one month later and observed for 29 days.

On some P. damicornis tiles, three or four of the fragments disappeared within a 24 h period between checks on their condition, appearing to have been bitten off. To determine the agents of this localized mortality, we replaced tiles whose four fragments had been eaten with spare tiles holding four healthy fragments around three of the focal colonies that had experienced localized mortality and videotaped the tiles (GoPro II HD) from about 1 m away during the following high tides. Cameras were retrieved at the next low tide and the videos watched.

We evaluated overall survival patterns and mortality specifically due to bleaching or predation using mixed-effects Cox proportional hazards survival models (coxme package, Therneau, 2012) in R (R Core Team, 2014). Distance and direction from focal colony were fixed effects (4 levels and 2 levels, respectively) and focal colony and tile nested within focal colony were random effects because fragments were blocked by tile and focal colony. The size of the focal colony and the depth of the tiles were included as covariates. Finally, we compared the relative levels of predation and bleaching in P. damicornis and S. hystrix using a chi-square test.

Distribution surveys

We characterized the spatial distribution of P. damicornis and S. hystrix in the reef flat MPAs of Namada, Vatu-o-lailai, and Votua villages at two scales (August through October 2013). For the larger-scale survey, we mapped each colony within 8 × 8 m plots (N = 5, 5, and 10 for Namada, Vatu-o-lailai, and Votua, respectively). Each plot was divided into 256 0.5 × 0.5 m cells and each coral ≥1 cm across mapped into a cell. The location of each survey plot was determined by randomly choosing a point on shore, swimming 100, 200, or 300 kicks directly away from shore at that point, and surveying the closest bommie (coral patch) large enough to fill more than three quarters of an 8 × 8 m plot. In four of 10 surveys at Votua and in all five surveys at Vatu-o-lailai and Namada, we also measured the largest diameter of each P. damicornis colony. We did not measure S. hystrix colony size because the frequent discontinuity of colonies made accurate estimation of colony area too error-prone. To avoid confounding biotically-driven spatial distribution with patterns caused by patchiness of suitable substrate, we also recorded which cells were comprised primarily of unsuitable habitats such as sand-scoured pools or channels and bommie tops covered in rubble.

We analyzed these data using the neighborhood density function O(r) in the point pattern analysis program Programita (Wiegand & Moloney, 2004). This analysis identifies distances at which individuals are aggregated, randomly spaced, or overdispersed compared to a specified null model. Unlike the more frequently used Ripley’s K(r) statistic, each distance category is not affected by those inside it; expected aggregation at each distance is compared to the observed value independently of nearer distances. Each concentric ring centered on an individual coral is separately placed on the aggregated-overdispersed continuum and displays the spatial pattern within a different distance category. Ring width was 0.5 m extending up to 4 m. The null model for this analysis was complete spatial randomness (CSR). Because the variance in substrate types violated CSR’s assumption of uniform likelihood of coral placement, we conducted the below analyses once using the entirety of all 8 × 8 m plots and a second time excluding cells of unsuitable habitat (which should better meet CSR’s assumption of uniform likelihood).

To determine whether the observed spatial pattern was random, significantly aggregated, or overdispersed, Programita simulated placement of each plot’s colonies 999 times using CSR, calculated O(r) for each simulation, then combined replicate O(r)’s from each reef and from all three reefs. This generated a distribution of simulated O(r)’s from which we established the significance of the observed spatial patterns. The distance(s) at which significant aggregation or overdispersion occurred were determined by the distances at which the observed pattern fell above or below the 95% simulation envelopes, respectively. This analysis does not parse aggregating and overdispering processes; it shows the net resulting pattern.

In addition to analyzing all P. damicornis and S. hystrix colonies, we analyzed P. damicornis <5 cm, ≥5 cm, ≥10 cm, and ≥15 cm in diameter to see if spatial patterns changed with colony size. The <5 cm and ≥5 cm categories were mutually exclusive but because there were not enough colonies between 5 and 10 cm and between 10 and 15 cm to analyze as mutually exclusive groups, larger size categories were subsets of smaller ones.

The 8 × 8 m quadrat surveys could not resolve spatial patterns below the cell size of 0.5 × 0.5 m, meaning that patterns occurring at less than 0.25² m could be undocumented. To determine the spatial distribution of P. damicornis and S. hystrix at smaller scales, we conducted 2 m radius circular surveys around focal P. damicornis and S. hystrix colonies that (i) were the largest colony of that species within 4 m (to reduce the effects of conspecifics), and (ii) occurred where >75% of the substrate within 2 m was suitable habitat for P. damicornis and S. hystrix, again to equalize the likelihood of colonies occurring everywhere in the survey.

The distance to each surrounding (radial) P. damicornis and S. hystrix colony was the average of the distance to that colony’s near and far sides (N = 45 focal colonies for P. damicornis around P. damicornis, 10 for S. hystrix around P. damicornis, and 24 each for P. damicornis and S. hystrix around S. hystrix). We analyzed radial colony counts in 10 cm concentric rings using a generalized linear mixed effects model with Poisson errors and the canonical log link function in R (lme4 package, Bates et al., 2013). Distance was a fixed effect and focal colony was a random effect, with the log₁₀ of the ring sizes as an offset to control for unequal area sampled at each distance (i.e., ring area increased with distance from the focal colony). We repeated this analysis with just the closest 0.5 m and 1 m of the circles in case radial colonies beyond those distances were masking short-range effects of the focal colonies.

We also analyzed the P. damicornis data from the 8 × 8 m plots in the same manner as we did the circular surveys. To convert the plot data, a function was written in R to identify every surveyed P. damicornis colony ≥2 m from all edges of its plot and equal to or larger than a specified diameter (either 15 or 20 cm) as a focal colony (N = 38 and 19 focal colonies, respectively). In order to have an appreciable sample size, we did not restrict focal colonies to those that were the largest within 4 m. The script then calculated the distances to all P. damicornis colonies less than the specified focal colony diameter within 2 m and placed them into 10 cm concentric rings as above. We analyzed the resulting data using the same procedures as described for the circular surveys.

Survival experiments

In our field experiment, neither distance nor direction from focal colony significantly affected survival of P. damicornis or S. hystrix fragments (Figs. 2A and 2B, respectively). We observed two main categories of mortality: bleaching preceding death in place (potentially due to microbes (e.g., Ben-Haim, Zicherman-Keren & Rosenberg, 2003)) and partial or complete disappearance, putatively due to predation (akin to Lenihan et al., 2011). Bleaching (47 and 46 fragments out of 320 for P. damicornis and S. hystrix, respectively) of neither species was affected by distance or direction (Figs. 2C and 2D). Distance and direction did not affect the number of P. damicornis fragments that partially or fully disappeared (putative predation), and direction did not affect this for S. hystrix but distance was significant (z = 2.23, p = 0.03) (Figs. 2E and 2F), with disappearance increasing with distance from the focal colony. In contrast, 0% of the 64 extra fragments of each species remaining on the coral rack where we originally acclimated the corals bleached or disappeared despite being on the same reef at the same time (Cox proportional hazards survival analysis, likelihood ratio for P. damicornis = 16.5, likelihood ratio for S. hystrix = 24.7, p < 0.0001 for both species). Fragments on the coral rack were ∼1 m above the benthos and may have experienced more flow or fewer benthic-associated biotic or physical stressors compared to the fragments on PVC pipes, which were 5–15 cm above the benthos.

The rapid disappearance of P. damicornis fragments around some focal colonies suggested spatially localized predation. Therefore, we further divided deaths due to putative predation between isolated predation incidents (disappearance of one or two fragments on a tile in 24 h) and localized predation episodes (disappearance of three or four fragments from a tile in 24 h). We used this classification scheme because three or four fragments tended to disappear from multiple tiles within the same 24 h period at certain replicates, whereas the disappearance of one or two fragments was not often temporally coincident across tiles within a replicate. We distinguished between these two types of putative predation because their causes were potentially different and therefore either one could have been distance-dependent or masked distance-dependence in the other. Six of 10 P. damicornis replicates (23 out of 160 tiles) experienced localized predation on at least one of their eight tiles; three of those experienced localized predation on five or more tiles within 24 h. Two of 10 S. hystrix replicates experienced localized predation (on one tile each). We further investigated localized predation only for P. damicornis because localized predation on S. hystrix was infrequent.

When tiles that had experienced localized predation around three focal colonies were replaced with spare tiles holding healthy fragments, all three sets of replacement tiles again experienced localized predation and their collective survival was significantly lower than that of the replicates that did not experience localized predation in the initial run (mixed effect Cox proportional hazards, z = 3.5, p < 0.0005). Videos of these tiles showed the territorial triggerfish Balistapus undulatus consuming multiple fragments from multiple tiles around two of the three focal colonies. Balistapus undulatus feeding resulted in fragments irregularly broken at or above the top of the epoxy, as was seen for most localized predation episodes in the initial outplanting.

We next examined whether localized predation was distance-dependent and whether it masked distance-dependent mortality in replicates that did not experience localized predation. Distance and direction did not significantly affect mortality in replicates that did not experience localized predation (Fig. 3A). Considering only replicates that experienced localized predation (both original and replacement tiles), neither distance nor direction significantly affected mortality from all causes (Fig. 3B) or just from localized predation (Fig. 3C).

Pocillopora damicornis fragments were significantly more likely to die of putative predation as opposed to bleach and die in place than were S. hystrix fragments (chi-square test, χ² = 17.2, df = 1, p < 0.0001). More than three times as many P. damicornis fragments died from putative predation as bleached prior to death (169 vs. 47 out of 320, respectively), while numbers of S. hystrix fragments that died from putative predation versus bleaching did not differ significantly (58 vs. 46 out of 320, respectively). Excluding replicates with localized predation, P. damicornis and S. hystrix appeared equally susceptible to isolated predation and bleaching (χ² = 0.022, df = 1, p = 0.88).

Distribution surveys

We analyzed patterns of distribution using both entire 8 × 8 m plots and after excluding habitat deemed unsuitable for P. damicornis or S. hystrix (e.g., sand-scoured channels and pools, bommie tops covered in rubble). The analyses using only suitable habitat were quantitatively similar to those using the entire plots but were more conservative. Neighborhood density graphs using only suitable habitat are included here. Neighborhood density analysis indicated that both P. damicornis and S. hystrix were significantly aggregated at up to 1 m when all size classes were considered and surveys from all villages were pooled (Figs. 4A and 4B, respectively). When analyzed by site, the distance below which colonies were aggregated ranged from <1 m in Votua and Vatu-o-lailai to nearly 3 m in Namada. At no distance on any reef were colonies significantly overdispersed.

Identical analyses with P. damicornis separated into size categories (Figs. 4C–4F) indicated that the largest colonies (≥15 cm) were not aggregated at any scale, but all smaller size classes were strongly aggregated at scales of up to 1 m. Thus, smaller colonies appeared to drive the aggregation at up to ∼1 m when we analyzed all sizes together. However, the limited sample size for large colonies (n = 187) may have constrained our ability to detect spatial patterns for large colonies.

To resolve the spatial distribution of P. damicornis and S. hystrix more finely, we conducted separate circular surveys (radius = 2 m) around focal colonies that met specific criteria. Across all 2 m, there was a significant negative relationship between distance from focal P. damicornis colonies and radial P. damicornis count (corrected for area surveyed at each distance and henceforth called density), focal P. damicornis and radial S. hystrix density, and focal S. hystrix and radial P. damicornis density (GLM: z = − 4.4, p < 0.0001; z = − 3.9, p < 0.0005; z = − 3.6, p < 0.0005, respectively) (Figs. 5A and 5B). The relationships within the first 0.5 m or 1 m for these focal-radial combinations were not significant (see Table 1 for all values not provided in text).

Across all 2 m, there was no significant relationship between distance from focal S. hystrix colony and radial S. hystrix density (GLM, z = − 1.9, p = 0.06) (Fig. 5B). However, there was a significant positive relationship between distance and density within the first 0.5 m (GLM, z = − 12.99, p < 0.05) but not within the first 1 m.

When we converted the 8 × 8 m surveys into data analogous to the circular surveys and considered any P. damicornis colony ≥15 cm across as a focal colony and any smaller individual as a radial colony, there was a significant negative relationship between distance and radial P. damicornis density (GLM, z = − 3.6, p < 0.0005) across all 2 m but not across the first 0.5 m or 1 m (Fig. 5C and Table 2). However, when the cutoff for focal colonies was 20 cm, there was no relationship between distance and P. damicornis colony count at 0.5 m, 1 m, or 2 m (Fig. 5C and Table 2).