Dataset: Edmunds 2020 L&O: sea surface temperature

Variation in coral recruitment

Unglazed terracotta tiles (15 ×15 × 1 cm) were used to evaluate coral recruitment, and these were secured horizontally and independently to the substratum with their rough side downwards. Each tile was attached to the reef using a stainless steel stud epoxied into non-living reef rock, and a ~ 1 cm gap was maintained beneath each tiles as a microhabitat favored by coral recruits. In St. John, 15 tiles were secured in clusters at each of five sites at ~ 5-m depth on fringing reefs between White Point and Cabritte Horn. In Mo'orea, 15 tiles were secured in clusters at each of 10-m and 17-m depth at two fore reef, and three back reef sites (~ 2-m depth), on the north shore. In each cluster, tiles were scattered haphazardly across the reef with spacing between them varying from a few centimeters to about a meter.

Tiles were seasoned in seawater (i.e., beneath the laboratory dock) for 6-12 months prior to use, and were deployed in fixed seasons. In Mo'orea, tiles were exchanged at the end of January (or early February) and the end of August (or early September), with the first deployment in August 2005 (back reef), August 2006 (10-m depth), and August 2008 (17-m depth); tiles were immersed for ~ 6 months. Freshly collected tiles were cleaned of organic material in dilute bleach in freshwater, dried, and scored for coral recruits using a dissecting microscope (× 40 magnification). Analyzed tiles were cleaned in hydrochloric acid, washed, and placed in seawater for seasoning until the next deployment. In St. John, tiles were exchanged in late July (or early August), with the first deployments in August 2006. Tiles were exchanged in January and August for the first two years, but after 2008, were exchanged annually. Tiles were immersed ~ 6 months in the first two years, but ~ 12 months thereafter, and were processed in an identical manner to those in Mo'orea. The sample size of tiles varied among years as a few were broken by storms, notably following Hurricanes Irma and Maria (September 2017) in St. John. Broken tiles were replaced at the next scheduled exchange date.

The top, bottom, and sides of tiles were screened by the author, and recruits were identified to family using primary literature and field guides. Recruits in Mo'orea were assigned to Pocilloporidae, Poritidae, Acroporidae, or "other", and in St. John to Poritidae, Agaricidae, Siderastreidae, Acroporidae, Faviidae, or "other"; these taxa reflect the resolution that was possible based on skeletal morphology. In Mo'orea, annual recruitment was estimated by summing mean recruitment between times within each year and site, and averaging between sites by habitat. Annual recruitment in St. John was estimated the same way for the first two years, but thereafter, was estimated from tiles immersed for 12 months.

Environmental conditions

In situ seawater temperature was measured using loggers. In Mo'orea, Sea-Bird 39s (± 0.002°C, Sea-Bird Electronics, Bellevue, WA) recording at 0.0083 Hz were deployed on the north shore (10-m depth, 20-m depth, and back reef), but the results from 10-m depth were augmented with results from a Hobo logger (U22-001, Onset Computer Corp., Bourne, MA) for 136 days from 18 August 2019. Data were average by day and summarized across years (2005-2019) by daily means (± 95% confidence interval). These records were characterized the thermal regime over the ~ 365 days sampled by the two batches of tiles deployed annually (1 September to 31 August). The thermal regime was described by the yearly mean, and the yearly mean variation calculated between consecutive days (day-day, D-D), weeks (week-week, W-W), or months (each lasting 4 weeks, month-month, M-M). Records from 20-m depth were used as a proxy for seawater temperature at 17-m depth (where tiles were located, but temperature was not recorded), and the two data sets were tested for congruence. Evidence of strong association of temperature between 10-m and 20-m depth, and differences between depths that were trivial with respect to coral performance, were used as a rationale to characterize the temperature of the fore reef by records from 10-m depth.

In St. John, temperature was recorded at Yawzi Point (9-m depth) using a variety of loggers from 1989 to 2019. Most records were obtained using Hobo loggers (± 0.2°C) (U22-001) sampling at 0.0011 Hz. Data were summarized as above for Mo'orea, except that yearly rates were calculated between 1 August and 31 July, and D-D variation was not considered because it could not be resolved with the resolution of the loggers.

To gain insight into temporal variation in seawater temperature at a spatial scale larger than the locations of the temperature loggers, sea surface temperature (SST) recorded through remote sensing was evaluated. SST was obtained from the NOAA Coral Reef Watch (CRW) web site (https://coralreefwatch.noaa.gov/, accessed 28 September 2020) using Regional Virtual Stations for the Society Archipelago and the Virgin Islands. Each Virtual Station consists of a 5 × 5 km cell in which nighttime SST is reported, as calibrated to 20 cm depth, with the Society Archipelago station centered at -16.9500°, -151.3750°, 167 km northwest of Mo'orea, and the Virgin Islands station centered at 18.200°, -64.5500°, 22 km southeast of St. John. Daily SST was accessed from January 1985 to September 2020, and records were used evaluate the 95% confidence intervals for daily values across a year.

Statistical Analysis

One-way PERMANOVAs were used to compare recruitment, and year-year changes in recruitment, among habitats/regions; post-hoc, pair-wise contrasts were completed with permutational t-tests. Repeated measures (RM) PERMANOVAs were used to compare recruitment among years (RM factor) within habitats/regions. PERMANOVAs were prepared using square-root transformed data and resemblance matrices containing Bray-Curtis dissimilarities, and results are reported as Pseudo-F values and their permutational probabilities (pperm). Contingency tables were used to test for independence between outcomes of year-year changes in recruitment (increases versus decreases) for pairs of years with delays of 1, 2, and 3 years. Each contingency table was tested for independence using ??2 tests with Yates correction for small sample sizes. To evaluate whether short-term variation in recruitment (i.e., over 1 year) was an accurate predictor of future trends in recruitment, Pearson correlations were used to test for association between year-year changes in recruitment and subsequent linear trends for variation in recruitment over time.

Before testing the second hypothesis that motivated the present study (that recruitment is associated with temperature), the relationships between coral recruitment and coral cover were explored using least squares regression for linear and quadratic functions, with the best model selected using AICc. Analyses were completed for all corals, Poritidae and Pocilloporidae in Mo'orea, and for all corals and Poritidae in St. John, with these taxa selected based on their high abundance. Although it was reasonable to expect that the relationship between coral recruitment and coral abundance would not be well developed, in part because of the capacity for widespread dispersal of corals, under some conditions it can be well developed. Therefore evaluating the extent to which coral recruitment was associated with coral cover was a necessary step to evaluating the relationships between recruitment and temperature. The statistical approaches were similar to those used to explore the relationships between recruitment and temperature. Analyses were completed using recruitment and temperature recorded over concurrent 12-month periods, and were repeated with temperature lagged by one year (i.e., to test for associations with temperature recorded 12-24 months before). The lagging of temperature provided insight into the effects on adult corals that would produce larvae in the following year.

Statistical analyses were completed using Systat 13 software for parametric tests, and PERMANOVA+ for PRIMER for permutational tests.