Coral cover:
Coral cover was evaluated using photoquadrats (0.5 × 0.5 m) placed at random positions along 40 m, permanently marked transects at six sites at 7–9-m depth; these sites were randomly selected in 1992. The positions of photoquadrats were re-randomized at each sampling, and they were recorded using digital cameras (Nikon SLRs, with 6–36 Megapixel resolution) fitted with a zoom lens (Nikon DX 18–70 mm or FX 18–35 mm) and placed in a waterproof housing (Ikelite). Cameras were attached to two strobes (Nikon SB105) and mounted on a framer that held them perpendicular to the reef.
Photoquadrats were analyzed for benthic community structure by overlaying images with 200 randomly located dots and identifying the benthos beneath each dot. The full analysis resolved corals to species, macroalgae, and a combination category of crustose coralline algae, algal turf, and bare space (CTB). In the present analysis, coral cover is presented on a percentage scale, and was averaged by sites and year.

Coral recruitment:
Coral recruitment was measured using unglazed terra cotta tiles (15 × 15 × 1 cm) that were individually attached to the benthos in approximately horizontal orientations using a stainless steel stud epoxied into non-living carbonate rock. Tiles were seasoned for several months in seawater prior to each deployment, and were secured, rough surface down, with a ~ 1 cm gap beneath to create a cryptic habitat favored for coral settlement. In the initial design, each tile was an independent replicate, and in the first year (2006-2007), 10 tiles were deployed at 5-m depth at each of five sites in August 2006, and at each site were placed in a cluster with tiles ~ 25-cm apart. The first tiles were immersed for ~ 6 months, and were replaced in January 2007. When tiles were replaced in August 2007, the sample size was increased to 15 tiles site-1, and over the following year, the tiles were changed every ~ 6 months (January 2008 and August 2008). Thereafter, tiles were replaced annually in July or August.

Retrieved tiles were cleaned in dilute bleach, air dried, and screened for coral recruits using a dissecting microscope (40 x). Recruits were identified to family using field guides, keys and advice from colleagues, and were resolved to Poritidae, Faviidae, Agaricidae, Siderastreidae, and “others”. Following scoring, tiles were cleaned in ~ 10% HCl, rinsed, and stored in seawater beneath the dock until the next deployment. The number of recruits on each tile was standardized among tile surfaces (top + bottom + sides) as recruits tile-1, and the results averaged by site. To compare recruit densities in the first two years with results obtained in the following years, annual recruitment in 2007 and 2008 at each site was estimated by summing mean recruitment by site in January and August of each year. Recruitment was compared among years using sites as replicates.

Small corals:
Small corals were ≤ 40-mm diameter, and while many were juveniles based on the age of inferred sexual maturity, others were sexually mature, for example, based on analyses of Soong (1993) for Favia fragum (Esper, 1795), and Siderastrea radians (Pallas, 1766) and the recent study by Gelais et al. (2016) of S. siderea (Ellis & Solander, 1786). Small corals were counted at six sites, five at 5-m and one at 9-m depth, and separate sampling schemes were used to quantify density and dynamics (i.e., growth and mortality). The density of small corals was determined using 40 m, permanently marked transects placed along the isobath at each site, along which 40 quadrats (0.5 × 0.5 m) were randomly placed. Small corals in each quadrat were counted by genus, using calipers to ensure corals were in the desired size class, and densities were expressed at colonies 0.25 m-2, and averaged by site. Densities of small corals were compared among years using sites as replicates.

The growth and mortality of small corals was determined by marking corals with numbered aluminum tags that were epoxied to the substratum. Corals were tagged as encountered within an area ± 3 m of the first 10 m of the transects used to evaluate density (described above). All corals were tagged regardless of taxon, with some as small as ~ 2-mm diameter, but none > 40-mm diameter. Tagged corals were identified to species (or genus where the identity was uncertain), and their size measured as the mean of two planar diameters using calipers (± 1 mm). The following year, tags were relocated using a metal detector (Surfmaster PI Pro, White’s Electronics, Sweet Home, OR, or VibraProbe 580, Treasure Products Inc., Simi Valley, CA), and the marked coral located and evaluated for conditions (alive or dead) and, if alive, measured. Missing corals were assumed to have died, and their tags were removed. Additional tags were deployed annually to mark new corals that were added to replace those that died, but the number of corals tagged at each site varied among years depending on the time available for this task. Growth rates were expressed as mm y-1, and mortality as % y-1, and were averaged by site for all corals. Limited sample sizes prevented analyses with site as a replicate and taxon a factor as was used for cover and density.

Statistical analyses
Coral cover, recruitment, and density of small corals were compared among times using repeated measures (RM) PERMANOVA. For each demographic domain, the first analysis was completed with sites as replicate, time the RM factor, and abundance (i.e., cover or density of corals, pooled among taxa) the dependent variable. In the second analysis, sites were replicates, time the RM factor, and genus (for small corals: Porites, Agaricia, Favia, and Siderastrea) or family (for coral recruits: Poritidae, Agaricidae, Faviidae, and Siderastreidae) a between subjects fixed effect. PERMANOVAs were conducted using resemblance matrices prepared using square root-transformed data and Bray Curtis dissimilarities, and these were tested for differences among times. Where time was significant, multiple contrasts were conducted using pairwise t-tests in a permutational framework. The mortality and growth of small corals were compared over time with one way ANOVA, but they were not separated by site and taxon due to unequal representation (including missing data) at levels of these factors. Temporal trends in abundance, mortality, and growth also were analyzed using least squares linear regression, in which the dependent variables were averaged among sites; only significant regressions are reported.
Multivariate community structure in each domain was visualized with 2-D ordination using non-metric multidimensional scaling (nMDS) based on sites as replicates and resemblance matrices prepared using square-root transformed data and Bray-Curtis dissimilarities. Ordinations were based on results by genus (small corals) or family (recruits), and vectors were used to display the influence of dependent variables on separation along each nMDS axis as evaluated by Pearson correlations. PERMANOVA was used to test for multivariate variation in community structure over time, using an RM design with sites as replicates and time the repeated factor. Parametric statistics were completed using Systat 13 (Systat Software, Inc., San Jose, CA), and assumptions of normality and equal variance were tested through graphical analysis of residuals. nMDS and PERMANOVA were conducted using Primer-E 6.0 with PERMANOVA+ add-on (Quest Research, Ltd., Auckland, NZ).

fig3.csv