Overview
This study took place on the south shore of St. John, and surveys were completed in August 2019 and January 2020 at Cabritte Horn and Tektite on the eastern side of Great Lameshur Bay. These sites were selected because the abundance of Peyssonnelid Algal Crusts (PAC) has been measured in these locations since 2015, and the high abundance of PAC created a tractable system to test for the outcomes of PAC-coral interactions. The coral community structure in St. John from 1987-present is described elsewhere, but in brief, coral cover has been < 4.5% at six sites since 1992, but at two other sites, it has declined from 45% to 4% (Yawzi Point) and 32% to 27% (Tektite) from 1987–2019. Over the same period, the cover of macroalgae has increased, and the rest of the hard substratum has remained covered by crustose coralline algae, turf algae, and bare rock (combined as “CTB”). The high abundance of igneous rock on these reefs provides substratum suitable for growth of PAC. As PAC in St. John is more abundant in shallow (3–5 m) versus deep (5–9 m) water, surveys were designed to contrast PAC between depths. Sampling along a 15 m transect at each site and depth was used to evaluate PAC abundance, growth, and competitive encounters. An additional opportunity to evaluate PAC growth was provided by legacy settlement plates from adjacent areas upon which PAC was abundant during deployments extending from 2009 – 2019.

Growth of PAC
Planar growth of PAC on settlement tiles
Unglazed terracotta tiles (15 × 15 × 1 cm) originally were deployed to measure coral recruitment, and here photographs of the tiles were re-purposed to provide an additional measure of the planar growth of PAC. Tiles were seasoned for a year in seawater before use and were deployed in July of each year at five sites along 5 km of the south shore of St. John at 5–6 m depth (n = 15 tiles per site). These sites were spread over the same area of coast over which PAC was studied herein. After one year, tiles were retrieved, soaked in bleach, rinsed, dried, and scored for coral recruits; new seasoned tiles then were deployed at each site. After scoring, tiles were photographed (at ~ 10 to 34 megapixels resolution), cleaned with dilute HCl, and replaced in seawater for seasoning. The orange coloration of PAC remained following bleaching, and this provided the opportunity to quantify the coverage of PAC on the upper surface of each tile. Photographs of complete sets of tiles were available for 2009, 2011, and from 2014 – 2019 (n = 15 tiles per site per year), but images of four additional tiles were opportunistically available from 2012. These additional tiles proved valuable in timing the potential arrival of PAC in St. John, but they were not used in the statistical analysis of the rate of growth of PAC on the tiles.

The area of PAC (cm2) on the tiles was measured using the Trainable Weka Segmentation plugin for Fiji software. An image classifier (random forest model) was trained to segment out PAC using a subset of labeled tiles (10 tiles per year). The models were run across the remaining images in each year to segment out PAC and measure its area on each tile. The resulting areas for each tile were then manually reviewed to resolve any misclassifications of PAC in the image. Based on the year-long immersion times of the tiles, the area of PAC on their upper surfaces was used to provide a conservative estimate the planar growth (cm2 per year) of PAC on unglazed terracotta.

Statistical analyses
To test for variation in PAC abundance among sites and depths, a two-way fixed effects ANOVA was used in which site and depth were fixed effects, and percent cover of PAC (arcsine transformed) was the dependent variable. Differences in linear growth rate of PAC on natural substrata were analyzed using a two-way fixed effects ANOVA, with site and depth as fixed factors. As testing for an effect of depth was a primary objective of this study, growth rates were compared between depths using planned comparisons. Differences among years in the growth rate of PAC on the tiles were analyzed using a two-way fixed effects ANOVA, with site and year as fixed factors, and the area of PAC covering each tile when they were recovered as the dependent variable (i.e., cm2 y-1).

Statistical analyses were completed using the open-source software R ver. 3.5.1, with lme4 and Matrix packages for log-linear analysis, and DescTools for the G-test. Statistical assumptions of ANOVA were tested using graphical analysis of the residuals.