Decadal-scale time series highlight the role of chronic disturbances in driving ecosystem collapse in the Anthropocene

These data support a publication in the Journal Ecology that describes 37 years of change on the coral reefs of St. John, US Virgin Islands. In this paper, four decades of surveys from two coral reefs (9 and 14 m depth) off St. John, US Virgin Islands, are used to quantify the associations of acute and chronic events with the changes in benthic community structure. These reefs profoundly changed over 36 years, with coral death altering species assemblages to depress abundances of the ecologically important coral Orbicella spp. and elevating the coverage of macroalgae and crustose coralline algae/turf/bare space (CTB). Linear mixed models revealed the prominent role of chronic variation in temperature in accounting for changes in coverage of corals, macroalgae, and CTB, with rising temperature associated with increases in coral cover on the deep reef, and declines on the shallow reef. Hurricanes were also associated with declines in coral cover on the shallow reef, and increases on the deep reef. Multivariate analyses revealed strong associations between community structure and temperature, but weaker associations with hurricanes, bleaching, and diseases. These results highlight the overwhelming importance of chronically increasing temperature in altering the benthic community structure of Caribbean reefs.

Methods
The study began in December 1987 to evaluate the effects of coral bleaching. Three, 10-m long, permanently marked transects were established at 14-m depth at the Tektite reef, and at 9-m depth at Yawzi Point. These reefs were selected because of their high percentage cover of Orbicella spp. and ease of access from a field station with small boats.

Each transect was sampled using 10 contiguous photoquadrats (1 × 1 m), recorded annually between May and December beginning in December 1987 and ending in July 2022. The reefs were sampled twice each year in 1988, 1989, and 2017 to evaluate the effects of hurricanes, and initially were recorded using color slide film, and from 2001, digitally. A strobe was used with all cameras, each image contained a scale bar, and slides were digitized (4000 dpi) for analysis. Images were analyzed using CPCe or CoralNET software, with 200 randomly-located points image-1 that were manually annotated to resolve coral (scleractinians and Millepora), macroalgae, and crustose coralline algae + algal turf + bare space (CTB). Objects to ~ 5 mm diameter were identified, although the taxonomic resolution was occasionally restricted by shadows, leading to a consensus list of 23 space holders that were quantified for percentage cover (mean ± SE).

Environmental conditions (temperature, hurricanes, rainfall) were tested for their capacity to explain variation in coral community data, and these data are pusblished with the paper. 

Differences over time in the cover of each benthic group were evaluated from contrasts of means and non-overlapping SEs. Linear mixed models (LMM) were used to test for the effects of time, average annual temperature (°C), rainfall (mm), hurricanes (0 or 1), disease (0 or 1), and bleaching (0 or 1) (fixed effect) on the percentage cover of corals, macroalgae, and CTB. Quadrats were averaged by transect within each site (n = 3) because transects were haphazardly selected and repeatedly surveyed; transects were introduced as a random factor. Models were fit using restriction maximum likelihood methods (REML), and were also run in simplified forms with non-significant fixed effects excluded. Models were evaluated using the Akaike Information Criterion corrected for small sample sizes (AICc). Interaction terms were not included because of the potential for multicollinearity among predictors. The influence of the fixed effects was evaluated through the model estimates and 95% confidence interval about these values.

Multivariate changes were displayed with 2-dimensional ordinations using non-metric multidimensional scaling (MDS) with Bray-Curtis dissimilarities based on square-root transformed data and 999 permutations until stress stabilized. MDS plots were prepared for community structure (coral, macroalgae, and CTB), and coral taxa, with significance tested using SIMPROF.

To test for associations between the environment and multivariate benthic community (by both functional groups and coral genera) environmental conditions were screened for collinearity using Pearson correlations, and where a significant association was detected (at P ≤ 0.05 with r > 0.341 and 35 years of data), one member of each pair was excluded based on weak clarity of interpretation of the environmental condition. The association of biological and environmental ordinations was tested using the BEST routine in Primer 7.0 software, in which Spearman rank correlation (ρ) identified the best match within a permutational framework. One analysis was completed with the biological and environmental data calculated by year, and one using a centered 3-year running mean to smooth variation. Environmental data were square-root transformed and converted to a resemblance matrix using Gower distances. BIO-ENV detects multiple associations, sometimes with different combinations of variables, among which the relative role of each variable in causing changes in community structure was evaluated using distance based linear modeling (DISTLM). This procedure distinguishes among associations identified through the BEST routine using AIC, and it attributes variance to components in the model.

Descriptive statistics and LMMs were completed using Systat 13 (Inpixon, San Jose, CA), and multivariate tests were completed using Primer 7.0 and PERMANOVA+. Statistical assumptions of LMM were evaluated through graphical analyses of residuals. Although MDS, SIMPROF, and DISTLM make no explicit assumptions about the distribution of the data, attention was paid to stress and avoidance of collinearity in the DISTLM.