International Journal of Marine Science, 2016, Vol.6, No.21, 1-20
5
Images of each cinder block and its associated coral fragments, were taken monthly over the course of two years
with a Nikon D500 in a Fantasea underwater housing. Illumination was provided by the built-in camera flash and
housing diffuser. Each image was analyzed with Image J to determine the mean fragment brightness. The
brightness of gray-scale images has been shown to be correlated (R
2
= 0.63) with symbiont density (Siebeck et al.,
2006) and thus, colony darkening (decreased brightness) is an indication of increased concentration of the coral
endosymbiont (Siebeck et al., 2006). All brightness values were determined relative to a clean area of the cinder
block (the brightest object in each image). Analysis proceeded by first transforming an image to 8-bit format.
The scale of each image was set using the average width of a cinder block (15.24 cm). Regions were drawn
around each fragment including only live coral tissue using the polygon tool and minimum, maximum, and mean
brightness values obtained. Differences in the background brightness between different blocks and/or site
dependent water quality (turbidity) was corrected for by adding the corresponding value or subtracting the
corresponding value to reach that of the first sampling period for each block. For analysis, mean colony brightness
values were either used as is or grouped into one of three bins of equal width spanning the range of brightness
values observed.
2.4 Statistical Analyses
Statistical analyses were performed with R version 3.0.2 (R Core Team, 2013) and all figures were created with
ggplot2 (Wickham, 2009).
We first analyzed the CREMP dataset for variation in coral cover among sites and zones (inshore vs. offshore)
across the 16-year sampling period
via
two-way repeated measures analysis of variance (ANOVA). The aov()
function within the statistical package R was applied with reef zone and sites as predictors of total coral cover and
an error term described by sample year to create a repeated measures design. Assumptions of ANOVA were
visually evaluated with the plot() function, which produces figures including residuals
vs
. the fitted model,
standardized residuals
vs
. theoretical quantiles (qq-plot), square root of the standardized residuals
vs
. the fitted
values, and the standardized residuals at each factor level. Non-linearity in these figures is indicative of a
deviation from a normal distribution. Homogeneity of variances was assessed using the Bartlett test (bartlett.test()),
which tests the null hypothesis that variances are equal at all levels.
Site community richness and evenness was assessed across the 16-year CREMP dataset and the 2 year dataset
from our field sites using Rényi diversity profiles (Jost, 2006). The renyi () function from the vegan package
served as the platform for this analysis (Oksanen et al., 2013). Mean relative abundances of coral species across
the 16-year period were applied to this function. The shape of each profile provides information with respect to
the evenness of a community. Horizontal lines represent perfectly even communities where all species are present
in equal abundances. Alpha values of 0, 1, 2, and infinity indicate species richness, Shannon diversity index,
Simpson index (reciprocal), and the proportion of the most abundant species present, respectively.
Coral community differences were identified using a multivariate approach. Principal coordinate analysis (PCoA)
was performed on the annual relative abundances of coral species at each site. To accomplish this, Bray-curtis
dissimilarities were calculated with the vegdist () function from the vegan package (Oksanen et al., 2013). Eigen
analysis was performed by the application of the cmdscale () function on this dissimilarity matrix in 2 dimensions
to produce eigen values. Confidence intervals (95%) were calculated from these eigen values across years for each
site. The correlations between the original relative abundances and eigen values were used as loadings to
determine the association of species with particular sites.
Principal component analysis was carried out on abiotic factors including ammonium (NH
4
+
) total organic
nitrogen (TON), chlorophyll a (Chl a), total organic carbon (TOC), silicate (SiO
2
), salinity, SWT and dissolved
oxygen (DO) from the WQMP. The WQMP sites mirrored the selected CREMP sites. Data were standardized by
subtracting each measurement by the mean and dividing by the standard deviation. The covariance of this matrix
was then calculated and eigen analysis performed on the covariance matrix with the eigen () command (Jackson,
