Agaricia spp. recruitment failure and subsequent recovery in the wake of mass bleaching and disease outbreak on reefs of St. Thomas, US Virgin Islands

Caribbean coral reefs have declined severely under cumulative stressors including hurricanes, thermal anomalies, disease outbreaks, and local anthropogenic impacts. In early 2019, stony coral tissue loss disease (SCTLD) emerged in the northern U.S. Virgin Islands (USVI), followed later that year by anomalously high temperatures and mass bleaching across the territory, providing an opportunity to assess the resilience of recruitment to acute disturbance. This study quantified recruitment at three shallow reef sites (7–14 m) in St. Thomas from 2019–2022 and compared results with pre-disturbance data (2015–2017). Recruitment tiles (15 × 15 cm) were deployed across eight time periods and photographed to determine recruit density and taxonomic identity. Bleaching and disease prevalence were calculated from coral health assessments along six 10-m transects as part of the Territorial Coral Reef Monitoring Program. All recruits observed were Agaricia spp. Recruitment rates in 2019 and 2020, following these disturbances, declined by an order of magnitude relative to pre-disturbance surveys. Because the six environmental predictors vary to differing degrees among versus within sites, their correlations with recruitment were decomposed into between-site and within-site components. Coral cover and Agaricia spp. abundance varied predominantly among sites (86.9% and 51.9% of variance between sites, respectively); their apparent positive associations with recruitment reflected differences among three site means rather than temporal variation within sites, and neither persisted once site means were removed. In contrast, Degree Heating Weeks (DHW) and combined bleaching and paling prevalence varied almost entirely within sites (8.7% and 1.3% between-site variance), and both showed marginal negative within-site relationships with recruitment (DHW: r = -0.43, p = 0.054; bleaching/paling: r = -0.41, p = 0.064). Thermal stress was therefore the only environmental variable with a detectable within-site signal, consistent with the temporal concordance between the 2019 disturbance and recruitment collapse. By 2021–2022, recruitment rebounded toward pre-disturbance levels at most sites, indicating recovery of recruitment potential within approximately two years. These findings demonstrate the sensitivity of Agaricia spp. recruitment to disturbance and underscore monitoring recruitment as a key indicator of reef resilience, as continued thermal stress and persistent disease threaten the natural recovery capacity of Caribbean reefs.

Caribbean reef-building corals have faced dramatic declines in recent decades due to a combination of multiple, interacting pressures. Global stressors, including ocean warming and acidification, combine with local stressors such as sedimentation, eutrophication, pollution, and overfishing to threaten coral survival (Zaneveld et al., 2016; Hughes et al., 2017; Lesser, 2021; Bhuyan et al., 2025). Some coral species have experienced reductions in coverage of up to 80% (Gardner et al., 2003; Mudge and Bruno, 2023), and disease outbreaks following thermal stress have driven further losses, including reductions in coral cover exceeding 50% in the USVI after the 2005 bleaching event (Cróquer and Weil, 2009). As anthropogenic stressors continue to escalate, the loss of these vital ecosystem engineers will likely continue, unless they are offset by the regrowth of remaining populations and recruitment of new individuals through reproduction. Coral reefs provide critical ecosystem services, such as essential fish habitat and coastal protection, but without living corals to maintain the reef structure, the economic benefits they provide will be severely compromised (Barbier et al., 2011).

Adult corals depend on an obligate symbiotic relationship with photosynthetic dinoflagellates in the family Symbiodiniaceae (known as zooxanthellae) for the majority of their energy. These symbionts can provide up to 90% of a coral’s nutritional needs via photosynthesis (Muller-Parker et al., 2015). When corals experience stress, the symbiosis is disrupted, causing the corals to expel or degrade their Symbiodiniaceae, resulting in coral bleaching. During bleaching, the density of these algal symbionts decreases, causing corals to lose their vibrant colors and reveal their underlying white skeletons. This loss of Symbiodiniaceae decreases the amount of photosynthetically derived energy available to the corals and, if prolonged, can result in high mortality rates (Eakin et al., 2010).

Thermal stress, primarily driven by rising global temperatures due to climate change, is the leading cause of mass coral bleaching. Prolonged bleaching can result in high coral mortality and the disruption of the reef ecosystem (Hoegh-Guldberg et al., 2007; Hughes et al., 2017). If they survive, thermally stressed corals often divert their limited energy reserves away from reproduction and toward survival. Because coral gametes are lipid rich and thus energetically costly to produce (Leinbach et al., 2021), bleached corals may resorb developing gametes to recoup energetic resources. The need to survive a bleaching event is a priority over reproductive investment during a single spawning season (Leinbach et al., 2021; Rodrigues and Padilla-Gamiño, 2022). Suppressed spawning can also persist for several years after a thermal anomaly rather than resolving in the following season: a decade of consistent spawning in Orbicella spp. was followed by repeated spawning failure after bleaching (Levitan et al., 2014). Therefore, mass coral bleaching can have impacts beyond direct mortality, affecting coral reproductive output and potentially limiting recruitment in the subsequent seasons.

Coral diseases also contribute to reef decline (Weil et al., 2009; Weil and Rogers, 2011). In January 2019, stony coral tissue loss disease (SCTLD) was first detected at Flat Cay, off of St. Thomas in the USVI (Brandt et al., 2021). SCTLD affects numerous Caribbean coral species, including Agaricia spp (Brandt et al., 2021; Meiling et al., 2021; Papke et al., 2024), and has caused substantial declines in coral cover and diversity (Costa et al., 2021). These losses may reduce the abundance of sexually mature coral colonies, though direct effects of SCTLD on gamete production remain poorly understood. Significant coral loss can also trigger a phase shift from coral-dominated reefs to reefs dominated by other organisms, such as macroalgae, sponges, or gorgonians, creating an environment that may inhibit coral settlement and recruitment (Alvarez-Filip et al., 2022). Further research is needed to understand the direct and indirect impacts of SCTLD on coral early life stages, as sustained natural recruitment is critical to continued reef existence. Coral reefs must be able to replace lost colonies with new coral recruits to combat continued degradation.

Coral reproduction can occur through both sexual and asexual mechanisms. The most common type of asexual reproduction in corals is fragmentation, which can facilitate population expansion, but it produces genetically identical colonies and does not increase genetic diversity (Highsmith, 1982). Sexual reproduction provides a mechanism for genetic recombination, thereby increasing genetic diversity and providing opportunities for enhanced adaptive potential. Corals have two main mechanisms of sexual reproduction: brooding and broadcast spawning. Brooding is a process in which sperm are released into the water column and drawn into maternal colonies, where internally retained eggs are fertilized and develop into larvae. In some Caribbean brooders this step may not be obligate: parent-offspring genotyping has confirmed parthenogenetic larval production in Porites astreoides, indicating that larvae can be produced without fertilization and that locally retained genotypes accumulate within a reef (Levitan et al., 2025). In broadcast spawning species, both sperm and egg are released simultaneously into the water column, where external fertilization occurs. For both strategies, successful coral recruitment depends on three processes: 1) larval availability, 2) settlement ecology, and 3) post-settlement ecology (Ritson-Williams et al., 2009). Larval availability includes gamete production, fertilization success, and larval survival and dispersal, which vary dramatically between brooders and broadcast spawners. Settlement ecology is a two-step process in which planktonic coral larvae cease dispersal and actively seek a suitable settlement substrate. Species differ in how selective that search is. In a comparison of three brooding and four broadcast spawning Caribbean corals, crustose coralline algae facilitated settlement of the broadcast spawners but not of most of the brooders, indicating that brooding taxa such as Agaricia spp. settle across a broader range of substrata (Ritson-Williams et al., 2016). Metamorphosis follows settlement and is a morphological and physiological transition into a single polyp recruit. Post-settlement ecology encompasses the factors that influence survival and growth of coral recruits. Coral recruits can experience high rates of mortality following settlement due to many factors, including their small size, vulnerability to predation and competition, and adverse environmental conditions such as thermal stress.

The agariciids are among the most abundant and widely distributed scleractinian corals on Caribbean reefs, occurring on shallow fringing reefs to mesophotic depths. Agaricia spp. now dominate many Caribbean reefs following the decline of larger frame-building taxa (Darling et al., 2012; Alvarez-Filip et al., 2022). They are classified as weedy coral species, with a life-history strategy characterized by small colony size, rapid early growth, short generation time, and brooding reproduction (Darling et al., 2012). Newly settled agariciids grow relatively rapidly, adding roughly 1 to 3 mm in diameter per month in the months following settlement (van Moorsel, 1988), but they remain poor competitors for space and are readily overgrown by larger, slower-growing frame-builders and/or by macroalgae and sponges (Vermeij, 2006). Sexual maturity is attained at small colony sizes, with Agaricia humilis reproductive at approximately 28 mm in diameter (van Moorsel, 1983), so colonies can begin contributing larvae within a few years of settlement. Unlike broadcast spawners, which release gametes during one or a few synchronized annual events, agariciids brood internally fertilized larvae and release competent planulae repeatedly through the year. Agaricia humilis planulates continuously, whereas A. agaricites releases larvae principally in spring and summer as water temperatures rise (van Moorsel, 1983). Brooded planulae are released at an advanced developmental stage and are competent to settle soon afterwards, giving a short planktonic phase that favors retention near natal reefs and local self-seeding rather than long-distance dispersal (Vermeij, 2005). Together, these traits make agariciids the taxon most likely to be captured by recruitment tiles deployed over intervals of several months, but also mean that patterns in agariciid recruitment cannot be assumed to represent recruitment of the wider coral community.

Marine benthic communities rely on recruitment of new individuals to maintain their populations and recover from disturbances, yet both adult corals and their larval stages are vulnerable to thermal stress and disease (Polato et al., 2010). Given the increasing frequency of bleaching events (IPCC, 2021; Hughes et al., 2018), and the emergence of new virulent diseases like SCTLD, it is crucial to understand how these threats impact coral recruitment, a process already known to vary substantially in space and time across Caribbean reefs (Irizarry-Soto and Weil, 2009). In the USVI in 2019, the emergence and spread of SCTLD coincided with a mass bleaching event driven by thermal stress. Coral recruitment rates were monitored around the island of St. Thomas before, during, and after this period of co-occurring stressors, and relationships between recruitment rates and these disturbances are presented here.

Methodology

This study quantified coral recruitment rates at three shallow sites off St. Thomas, USVI (Figure 1), across eight independent sampling periods between 2015 and 2022 (Table 1). No tile deployment occurred between Spring 2017 and Fall 2019, a gap of approximately 2.5 years, due to widespread infrastructure damage and logistical constraints following Hurricanes Irma and Maria, which made landfall in the USVI in September 2017 and caused extensive reef damage across the territory (Ennis et al., 2019). Recruitment rates during four of the sampling periods were previously reported in Brandt et al. (2019), and these rates are compared with four additional periods. At each study site, 15 x 15 cm terracotta tiles were haphazardly installed across the coral reef and left for 5–8 months (Figure 2). Tiles were removed and coral recruits on the tiles were counted, identified to the lowest possible taxon, and photographed using a microscope camera (Figure 3). Study methods, including the length of tile deployment and tile installation methods, were modeled after Green and Edmunds (2011) and Brandt et al. (2019).

Site selection

Multiple sites distributed throughout the USVI have been monitored annually since 2002 by the U.S. Virgin Islands Territorial Coral Reef Monitoring Program (TCRMP) administered through the University of the Virgin Islands Center for Marine and Environmental Studies (CMES) using standard benthic assessment methods (Ennis et al., 2019). This study focused on three TCRMP sites around St. Thomas: Black Point, Buck Island St. Thomas (hereafter referred to as Buck Island), and Magens Bay (Table 2). Black Point is a shallow nearshore fringing reef on the south shore of St. Thomas at 9 m depth, lying close to the developed southern coastline and thus exposed to terrestrial runoff and sedimentation. Magens Bay, at 7 m, is the shallowest site and sits on the north shore within a semi-enclosed embayment that affords it reduced wave exposure. Buck Island is the deepest and most oceanic site at 14 m, an offshore cay separated from the main island of St. Thomas, where greater flushing and reduced terrestrial influence are associated with higher water clarity (Figure 1). These differences in depth, wave exposure, and proximity to land are relevant to the interpretation of the among-site differences in recruitment reported below.

Tile installation

To prepare the 15 x 15cm terracotta tiles for installation on the reefs, a ¼ in hole was drilled into the center of each tile, wide enough for a 3 x ¼ in stainless steel lag screw. All tiles were conditioned for one month prior to deployment. Tiles for Periods Fall 2015 – Spring 2017 and Fall 2020 – Spring 2022 were conditioned in Brewers Bay under the CMES dock (Figure 1). Prior to installation for Fall 2019, tiles were conditioned at Coculus Rock (Figure 1) because SCTLD was already affecting Brewers Bay at the time of conditioning, and there was a concern that if the tiles were conditioned in a diseased location (e.g., CMES dock), they could act as reservoirs or vectors for the pathogen. However, by Fall 2020, SCTLD had affected all sites surrounding St. Thomas, so the CMES dock was used to condition tiles for that and all subsequent periods. Tiles installed by divers on SCUBA were distributed haphazardly across the sites. At each site, tiles were installed within a narrow depth band around the depth of the site’s permanent TCRMP monitoring transects (Table 2). Once a suitable location was located, the diver drilled a hole in the reef substrate using a 5 / 16 x 6” masonry drill bit. The conditioned tile was then manually screwed into the substrate with the stainless-steel lag screw fitted into a 1 ½” #14–16 ribbed plastic anchor with an individually numbered tag attached to the screw. Areas of living substrate were avoided. Finally, the diver took a photograph of the tile and surrounding area.

Recruit tile processing

At the end of each deployment time period, recruitment tiles were collected by divers on SCUBA and placed in individually numbered zip-lock bags, along with their identifying numbered tags. They were then transported in coolers with seawater to CMES where they were kept in a flow-through seawater table until they could be processed. Processing included photographing all six sides of the tile and examining each side for coral recruits using a magnifying glass and dissecting microscope. Coral recruits were counted, identified to the lowest possible taxon, and photographed using a digital microscope camera (Hayear HY-2307) (Figure 3).

TCRMP methods

TCRMP is an annual coral reef monitoring program managed and implemented by CMES, administered by the Division of Coastal Zone Management, USVI Department of Planning and Natural Resources. TCRMP monitors both benthic and fish resources across the USVI at permanent monitoring sites, but for the purpose of this project, only data from benthic monitoring from the three study sites were used.

At each site, benthic percent cover surveys are conducted annually along six fixed 10 m-long transects marked with metal stakes at each end. During each monitoring survey, a transect tape is stretched between the stakes. A diver traverses each transect while videotaping benthic cover using a high-definition digital video recorder held at a consistent height above the benthos. After recording, 20–50 non-overlapping images from each video transect are captured and imported into Rstudio, where 20 randomly generated points are superimposed on each image. This results in 400–1000 points being evaluated per transect. Analysis consists of identifying the substrate beneath each point. For each transect, percent cover of coral, epilithic algae (EAC), macroalgae, sponges, gorgonians, and sand/sediment are calculated by dividing the number of points of the given substrate by the total number of points evaluated per transect.

Coral health surveys are also conducted by TCRMP annually along the same six transects at each site. All coral colonies with any portion located directly beneath the transect lines are assessed in situ for signs of disease, bleaching and mortality. Coral diseases are categorized into recognized Caribbean scleractinian diseases and syndromes that include bleaching, black band disease, dark spot disease, white plague, yellow band disease, and SCTLD. More specific details about the survey methods can be found in Ennis et al. (2019).

At each site, benthic temperatures were recorded with a calibration-checked HoboTemp™ thermistor data logger (Onset Computer Corporation, Bourne, Massachusetts). These data were used to calculate Degree Heating Weeks (DHW), a measure of sustained coral thermal stress over a 12-week period when temperatures exceed 1 °C above the maximum monthly mean. Significant bleaching is likely to occur at 4 DHWs, and widespread bleaching and mortality may occur at 8 DHWs or higher (Eakin et al., 2010).

Statistical analysis

Coral recruitment rates were calculated as recruits m⁻² day⁻¹ by taking the number of recruits on a tile and dividing by the area of the tile and the number of days that the tile had been deployed. Where data did not meet assumptions of normality, a Friedman rank sum test was used as a non-parametric alternative to a repeated-measures ANOVA to assess differences in recruitment rates across periods. A Scheirer-Ray-Hare test was used as a non-parametric alternative to a two-way ANOVA to assess the effects of site and period on recruitment rates while accounting for non-normality. Dunn’s post-hoc test with Bonferroni correction was applied to identify specific pairwise differences among sites and periods. These tests were conducted on individual tiles, which are true replicates within each site and period.

Relationships between recruitment and environmental variables were assessed at the site-period level. All six environmental predictors (DHW, coral cover, Agaricia spp. abundance, combined bleaching and paling prevalence, SCTLD prevalence, and white disease prevalence) are derived from site-level TCRMP surveys or from site-level temperature loggers and are represented by a single value per site per period; they do not vary among tiles. Recruitment rates were therefore averaged to site-period means (N = 24 site-period means derived from 429 tiles, df = 22), and Pearson’s correlation coefficients were calculated with 95% confidence intervals obtained by Fisher z transformation. Confidence intervals are reported in place of a post-hoc power analysis, which is a deterministic function of the observed p-value and conveys no information beyond it.

Because the 24 site-period means are repeated measures at three sites, correlations computed across all observations confound variation among sites with variation within sites over time. Each predictor was therefore decomposed into between-site and within-site components by subtracting site means, and within-site correlations were tested with degrees of freedom reduced by the three estimated site means. Relationships were additionally modelled as weighted linear models of the form recruitment ~ predictor + site, weighted by the number of tiles contributing to each site-period mean, which controls for site and downweights means estimated from fewer tiles. Equivalent linear mixed models with site as a random intercept are reported in Supplementary Table 2 but are not used for inference, because the random-intercept variance is poorly identified with only three grouping levels.

A reduced model containing DHW and coral cover, chosen to represent the thermal stress and adult stock pathways, respectively, was fitted in place of a full six-predictor multiple regression, which would be overfitted at this sample size. Collinearity among predictors was assessed using Pearson correlations and variance inflation factors (Supplementary Table 3).

TCRMP benthic and coral health surveys are conducted annually whereas tiles were deployed twice yearly, so paired Fall and Spring deployments within the same survey year share values for the five TCRMP-derived predictors, which make 15 unique combinations across the 24 site-period observations. DHW is not subject to this constraint: benthic temperature was logged continuously at 15-minute intervals throughout each year-long logger deployment, and DHW was calculated independently for every tile deployment window. Data analyses were conducted and graphs constructed in RStudio Version 2026.05.1 + 225.

Recruitment rates

All coral recruits observed across all sites and time periods were identified as Agaricia spp. Recruitment rates varied significantly by period, site, and their interaction (Scheirer-Ray-Hare test: Site H = 49.33, df = 2, p < 0.0001; Period H = 84.13, df = 7, p < 0.0001; Site × Period H = 50.50, df = 14, p < 0.0001).

There was a significant effect of sampling period on recruitment rates (Friedman rank sum test, p < 0.0001). Fall 2019 and Fall 2020 rates were significantly lower than Fall 2015 (p < 0.0001 for both comparisons) and Fall 2016 (p < 0.01 for both comparisons). Fall 2019 and Fall 2020 did not differ significantly from each other (p = 1.000). Recruitment rates during the recovery periods of Fall 2021 and Spring 2022 were both significantly higher than Fall 2019 and Fall 2020 (all p < 0.0001) and did not differ significantly from the pre-disturbance periods of Fall 2015 and Fall 2016 (all p > 0.05) (Figure 4; per-period and per-site means with standard errors are given in Supplementary Table 5). Spring sampling periods showed a significant effect of site (Friedman rank sum test, p = 0.019) but no significant effect of period.

There was an overall significant effect of site on recruitment rates (Dunn’s test with Bonferroni correction). Black Point supported significantly higher recruitment than both Buck Island (Z = 6.05, p < 0.0001) and Magens Bay (Z = 4.83, p < 0.0001). Recruitment rates for Buck Island and Magens Bay did not differ significantly from each other (p = 0.638).

Relationships between recruitment and environmental variables

Across all 24 site-period means, recruitment was positively correlated with coral cover (r = 0.496, 95% CI 0.116 to 0.750, p = 0.014) and with Agaricia spp. abundance (r = 0.559, 95% CI 0.200 to 0.785, p = 0.005). Correlations with DHW (r = -0.260, p = 0.220), SCTLD prevalence (r = -0.185, p = 0.387), white disease prevalence (r = -0.239, p = 0.261), and combined bleaching and paling prevalence (r = -0.399, p = 0.053) were negative but not significant (Figure 5; Supplementary Table 1). Confidence intervals for the non-significant relationships are wide and encompass moderate effects in both directions, indicating an absence of a detectable linear relationship at this sample size rather than evidence of no effect.

These correlations pool variation among sites with variation within sites over time, and the two components behaved differently (Supplementary Table 4). Coral cover and Agaricia spp. abundance varied predominantly among sites, with 86.9% and 51.9% of their variance, respectively, occurring between sites rather than within them. Their apparent relationships with recruitment reflect the contrast among three site means, in which Black Point demonstrated the highest coral cover (15.3%), the highest Agaricia spp. relative abundance (24.9%), and the highest recruitment (0.488 recruits m⁻² day⁻¹), while Buck Island (5.7%, 9.0%, 0.107, respectively) and Magens Bay (3.3%, 14.5%, 0.168, respectively) were lower. Once site means were removed, neither relationship persisted: coral cover showed no within-site relationship with recruitment (r = -0.045, p = 0.846), and Agaricia spp. abundance was not significant (r = 0.264, p = 0.248). With site included as a fixed effect and observations weighted by number of tiles, neither predictor approached significance (coral cover p = 0.386; Agaricia spp. abundance p = 0.675; Supplementary Table 2b).

The thermal stress variables showed the opposite pattern. DHW and combined bleaching and paling prevalence vary almost entirely within sites, with only 8.7% and 1.3% of their variance occurring between sites, and were therefore estimated from the temporal variation that the sampling design was structured to capture. Within sites, DHW showed the strongest relationship with recruitment of any variable examined (r = -0.427, p = 0.054), and combined bleaching and paling prevalence was comparable (r = -0.411, p = 0.064). These were also the only predictors approaching significance once site was included as a fixed effect (DHW slope = -0.043, 95% CI -0.086 to 0.000, p = 0.050; bleaching and paling slope = -0.008, 95% CI -0.017 to 0.000, p = 0.061; Supplementary Table 2b). DHW and coral cover were effectively uncorrelated (r = 0.024), and the reduced two-predictor model was therefore unaffected by collinearity (VIF = 1.00 for both predictors); the variance inflation factors in Supplementary Table 3 are those from the full six-predictor model. In the reduced two-predictor model, DHW approached significance (slope = -0.042, p = 0.054) whereas coral cover did not (slope = 0.020, p = 0.370).

Thermal stress was therefore the only environmental variable with a detectable within-site relationship to recruitment, although that relationship remained marginal. Relationships involving the five TCRMP-derived predictors should be interpreted in light of their annual resolution, which yields 15 unique predictor combinations across the 24 site-period observations; DHW, calculated independently for each deployment window, was not subject to this constraint.

Recruitment rates observed in this study were consistent with previous surveys conducted in the USVI when disturbance periods were excluded. Green and Edmunds (2011) recorded recruitment rates of 0.42 recruits m⁻² day⁻¹ in St. John, Brandt et al. (2019) recorded 0.21 recruits m⁻² day⁻¹ in St. Thomas, and Rogers et al. (1984) recorded 0.26 recruits m⁻² day⁻¹ in St. Croix, comparable to rates recorded in this study in Fall 2021 (0.50 recruits m⁻² day⁻¹) and Spring 2022 (0.40 recruits m⁻² day⁻¹). Longer records from the region show that such rates are spatially structured and persistent: recruitment across five St. John sites monitored for 14 years sustained a hotspot that did not translate into increased coral cover, indicating a demographic bottleneck between settlement and adult size (Edmunds, 2021). However, recruitment rates during the 2019 bleaching and SCTLD outbreak (Fall 2019: 0.013 recruits m⁻² day⁻¹) and the following year (Fall 2020: 0.033 recruits m⁻² day⁻¹) were an order of magnitude lower than both historical and post-disturbance rates. Recruitment rates recorded during spring periods did not differ from each other, but spring periods were only monitored before and after the bleaching and disease events and not during the disturbances themselves. It is also worth noting that the pre-disturbance baseline period (Fall 2015 – Spring 2017) preceded Hurricanes Irma and Maria, which caused significant physical damage to USVI reefs in September 2017. Thus, entering the 2019 bleaching and SCTLD events, these reefs had already experienced considerable disturbance (Gochfeld et al., 2020, 2025), and these cumulative impacts likely reduced reef resilience and influenced subsequent recruitment dynamics.

The near-complete failure of Agaricia spp. recruitment at all monitored sites during Fall 2019 and Fall 2020 coincided with the emergence of SCTLD in January 2019, a mass bleaching event driven by anomalously high thermal stress later that year, and the continued legacy of hurricane damage. Recruitment rates declined by an order of magnitude relative to pre-disturbance baselines (Fall 2015 – Spring 2017) and recovered as these co-occurring stressors diminished (Figure 4). The temporal concordance between this period of compounding disturbance and recruitment collapse, followed by signs of recovery as conditions improved, provides strong evidence that environmental stressors were the primary drivers of reduced reproductive output and recruitment failure.

This pattern is consistent with the well-documented capacity of acute disturbance, particularly thermal stress, to disrupt coral reproductive processes for multiple years, with effects persisting well beyond the initial bleaching event (Szmant and Gassman, 1990; Levitan et al., 2014). These lagged effects help explain why recruitment failure extended into subsequent spawning seasons, emphasizing that bleaching should be viewed as a chronic disturbance to coral population replenishment, rather than as a singular acute event.

Despite this pronounced decline, recruitment rates rebounded to pre-bleaching levels within approximately two years post-bleaching (Fall 2021 and Spring 2022). Similar trajectories of reproductive recovery from thermal stress have been documented on other reefs where stressors have subsided and suitable settlement conditions remain (Gilmour et al., 2013). This rebound suggests that while Agaricia spp. recruitment is highly sensitive to thermal anomalies, some degree of reproductive resilience persists within USVI Agaricia spp. populations. However, full demographic recovery will depend on the persistence and resilience of reproductive capacity of adult Agaricia spp. colonies at these sites (Hughes and Tanner, 2000).

Recovery trajectories varied notably among sites. Black Point not only rebounded to pre-disturbance levels but exceeded them in Fall 2021 (1.17 recruits m⁻² day⁻¹), suggesting that reproductively active adult colonies at this shallow nearshore site were largely maintained through the disturbance. In contrast, Buck Island, the deepest and most offshore site, showed the weakest recovery, remaining below pre-disturbance rates through Fall 2021 and Spring 2022, possibly reflecting reduced connectivity to nearshore larval sources or greater impacts of SCTLD on local adult coral abundance. Magens Bay showed the most variable pattern, with complete recruitment failure in Fall 2019 followed by the highest single-period rate at any site in Spring 2022 (0.668 recruits m⁻² day⁻¹), indicating that while thermal stress initially suppressed recruitment entirely, the reproductive capacity at this site recovered rapidly once conditions improved. These site-level patterns highlight the importance of local factors, including depth, connectivity, and adult colony survivorship, in affecting the magnitude and timing of recruitment recovery following mass bleaching events.

Thermal stress and bleaching fundamentally alter coral energetic budgets. During bleaching events, corals expel their symbiotic dinoflagellates, drastically reducing energy acquisition and forcing colonies to prioritize survival over reproduction. Reduced gametogenesis, gamete resorption, and suppressed spawning have all been documented under such conditions (Szmant and Gassman, 1990; Grottoli et al., 2006; Leinbach et al., 2021). These energetic tradeoffs directly limit larval availability, one of the critical bottlenecks in the recruitment process. In addition to reduced gamete quantity, thermal stress negatively affects gamete quality. Elevated temperatures can impair sperm morphology and motility, resulting in lower fertilization success even when spawning occurs (Randall and Szmant, 2009a; Henley et al., 2021). Thermal stress has also been shown to increase larval mortality and reduce settlement competency, compounding recruitment failure during and after bleaching events (Negri et al., 2007; Polato et al., 2010). Thus, even if spawning occurred during the thermal stress event, elevated temperatures might have directly inhibited successful fertilization, larval development, and/or settlement, independently of parental energy depletion. Elevated temperatures narrow the range of sperm concentrations that yield successful fertilization, effectively increasing sperm limitation under thermal stress (Negri et al., 2007). For Caribbean brooding corals, such as Agaricia spp., temperatures of 31 °C have been shown to reduce larval survivorship and increase settlement-stage mortality by more than 25% compared to ambient conditions (Randall and Szmant, 2009b). For broadcast spawning species, temperatures between 30 and 31.5 °C produce an eightfold reduction in larval survivorship and reduce settlement success by nearly half (Randall and Szmant, 2009c). These mechanisms suggest that recruitment failure during Fall 2019 and Fall 2020 likely resulted from simultaneous impairment across the entire recruitment pathway, from suppressed gametogenesis and spawning failure to direct mortality of larvae and reduced competency of settlers. Together, these mechanisms provide a physiological framework for understanding how the co-occurring bleaching and disease events of 2019 could suppress recruitment across multiple stages, from gametogenesis through larval settlement, even in the absence of a statistically detectable correlation between DHWs and recruitment at the site-period level.

Recruitment failure represents a critical bottleneck for coral reef recovery. When recruitment remains persistently low, reefs are unable to replace lost colonies, increasing the risk of long-term coral decline and phase shifts toward alternative benthic states (Hughes and Tanner, 2000; Mumby et al., 2007). The near-complete recruitment failure observed during Fall 2019 and Fall 2020 therefore has significant implications for future reef structure and function, particularly if thermal stress events continue to intensify.

SCTLD was present throughout the later study periods, when recruitment rates returned to their pre-disturbance levels. Coral recruits and juveniles are susceptible to SCTLD (Cróquer et al., 2022; Williamson et al., 2022) and the dominant recruiting taxa in this study, Agaricia spp., exhibited both susceptibility to SCTLD and declines in abundance following SCTLD emergence in the USVI (Brandt et al., 2021). However, this group is comparatively less susceptible and impacted by SCTLD than other coral taxa (Brandt et al., 2021). The temporal alignment between recruitment collapse and peak thermal stress, combined with the absence of significant correlations between disease prevalence and recruitment rates (Supplementary Table 1), suggests that disease was not the primary driver of recruitment failure, though it cannot be excluded as a contributing factor. The recovery of recruitment to pre-disturbance levels in Fall 2021 and Spring 2022, despite continued SCTLD presence on USVI reefs, further supports the dominant role of thermal disturbance in driving the observed recruitment decline. Nonetheless, interactions between chronic disease pressure and episodic thermal stress may still reduce reproductive capacity over longer timescales by reducing the abundance of mature colonies.

Analyses of the relationships between recruitment and benthic and thermal metrics support thermal stress as the environmental correlate of recruitment variability at these sites. Recruitment was positively correlated with coral cover and with Agaricia spp. abundance when all 24 site-period means were considered together, but neither relationship persisted once variation among sites was accounted for (Supplementary Tables 2b and 4). Both variables differ far more among the three study sites than within any one site over time, so their apparent associations with recruitment rest on a comparison of three locations rather than on the temporal contrast this study was designed to resolve. We therefore do not interpret these correlations as evidence for a stock-recruitment relationship at the scale sampled here, although the positive association between adult Agaricia spp. abundance and Agaricia spp. recruitment across sites is consistent with the localized larval retention expected of a brooding taxon with a short planktonic phase.

DHW and bleaching/paling prevalence varied almost entirely within sites across deployment periods. Within sites, DHW showed the strongest relationship with recruitment of any predictor examined and combined bleaching and paling prevalence was comparable. We believe the marginal p-values reported for these relationships reflect the limited number of independent site-period observations and the simultaneous influence of multiple co-occurring stressors, rather than an absence of a biological relationship. Mechanistic evidence linking bleaching to reproductive failure remains strong and is consistent with the direction and timing of the recruitment declines observed here (Szmant and Gassman, 1990; Levitan et al., 2014; Leinbach et al., 2021). DHW and combined bleaching and paling prevalence vary predominantly within sites across periods, and both showed marginal negative relationships with recruitment once site was accounted for. Although neither reached significance, they were the only environmental variables with a detectable within-site signal, and their direction and timing align with the collapse of recruitment during Fall 2019 and Fall 2020, and its subsequent recovery. This is consistent with the physiological mechanisms outlined above, in which thermal stress suppresses recruitment through impaired gametogenesis, reduced fertilization success, and/or elevated larval and early post-settlement mortality. The limited number of independent site-period observations constrains the power of these analyses, and the absence of a significant relationship with disease prevalence does not exclude disease as a contributing factor.

The exclusive recruitment of Agaricia spp. was consistent with their reproductive strategy as year-round brooders, which produce and release planula larvae rather than spawning during discrete annual spawning events (Ritson-Williams et al., 2009). Multi-year recruitment monitoring across the Florida Reef Tract shows the same contrast, with broadcast spawners recruiting in a boom-or-bust pattern while brooders recruit consistently between years (Harper et al., 2023). The six-month tile deployment intervals used in this study were therefore more likely to catch recruitment by brooding species, whereas broadcast spawners typically reproduce during a single annual mass spawning event and may have been underrepresented within the deployment windows. The dominance of Agaricia spp. across all sites and sampling periods, including during recovery, likely reflects this deployment bias. The recruitment failure and subsequent recovery documented here may represent a conservative estimate of the broader effects of disturbance on the coral community. Whether broadcast spawning species experienced similar recruitment failure cannot be determined from this dataset, and their response during this period remains unknown. Thermal stress is well documented to disrupt mass spawning synchrony (Harrison, 2011), but the deployment design used here cannot test whether it did so at these sites.

The increasing prevalence of the thermally tolerant symbiont Durusdinium trenchii adds further complexity to coral reproductive dynamics. While association with D. trenchii can enhance coral survival during thermal stress, this symbiont has been linked to reduced host growth and reproductive output, reflecting a tradeoff between thermal stress tolerance and long-term fitness (Cunning et al., 2015; Chan et al., 2026). Such tradeoffs may partially explain suppressed recruitment following bleaching events, even when adult colonies persist.

Study limitations

Terracotta tiles provide a standardized settlement surface that can be retrieved and examined under magnification, allowing recruits to be counted and identified at sizes that are effectively undetectable in situ on natural reef substrate. This gain in detectability and replication comes at a cost in ecological realism. Settlement tiles differ from adjacent reef substrate in texture, microtopography, mineralogy, orientation, and in the biofilm and crustose coralline algal communities that develop on them, all of which influence larval settlement choice and early post-settlement survival (Ritson-Williams et al., 2009). No paired natural-substrate surveys were conducted, so the rates reported here are best interpreted as a relative index of recruitment that is internally comparable across sites and periods rather than as an absolute measure of recruitment onto the reef. The central result of this study, an order-of-magnitude decline during 2019 and 2020, followed by recovery, is robust to this limitation, because tile material, dimensions, conditioning duration, deployment length, and processing protocol were held constant across all eight periods, so any bias introduced by the artificial surface applies equally to every comparison. Absolute rates may nonetheless differ from those on natural substrate, and comparisons with studies using different substrate types, orientations, or deployment durations should be made cautiously. Retaining the methods of Brandt et al. (2019) was a deliberate trade-off that preserved comparability with the pre-disturbance baseline at the expense of incorporating a natural-substrate control.

All tiles were conditioned for one month prior to deployment to accrue a biofilm, but conditioning was not identical across periods. Fall 2019 tiles were conditioned at Coculus Rock rather than at the CMES dock in Brewers Bay, because SCTLD was already present in Brewers Bay and there was concern that tiles conditioned there could act as reservoirs or vectors for the pathogen. Because Fall 2019 was also one of the two recruitment-failure periods, conditioning location and disturbance history are confounded for that period and their contributions cannot be separated statistically. Several observations argue against conditioning location as the primary driver of the observed failure. First, Fall 2020 tiles were conditioned at the CMES dock yet produced recruitment rates statistically indistinguishable from Fall 2019 (0.033 and 0.013 recruits m⁻² day⁻¹ respectively, p = 1.000), and the recovery documented in Fall 2021 and Spring 2022 occurred under that same standard protocol. Conditioning location therefore cannot account for the two-period failure, although a contribution specific to Fall 2019 cannot be excluded. Conditioning timing also differed systematically between fall and spring deployments, each of which was conditioned in the month preceding its deployment date (Table 1); any seasonal difference in biofilm composition or maturity is therefore shared by all deployments within a season and is confounded with season itself, though this is less likely to affect the fall-to-fall comparisons on which the main conclusions rest.

Several features of the sampling design constrain the strength of inference. Environmental predictors use a single value per site per period, so relationships with recruitment were necessarily tested on 24 site-period means rather than on individual tiles; this is the appropriate unit of analysis but yields limited statistical power, and confidence intervals for the non-significant relationships are correspondingly wide. Because only three sites were monitored, site was modelled as a fixed effect and between-site variation is estimated from three means, so predictors that vary principally among sites, namely coral cover and Agaricia spp. abundance, cannot be evaluated independently of site identity at this scale. The five TCRMP-derived predictors are resolved annually, so paired fall and spring deployments within a survey year share values and the 24 observations represent only 15 unique predictor combinations. DHW, calculated independently for each deployment window from continuous logger records, is not subject to this constraint. The 2.5-year gap between Spring 2017 and Fall 2019 following Hurricanes Irma and Maria means no recruitment data are available for the intervening period, and the pre-disturbance baseline (Fall 2015 – Spring 2017) predates those storms. Finally, recruits were counted after 5–8 months on the reef, so the rates reported here reflect settlement and early post-settlement survival combined and cannot distinguish reduced larval supply or settlement failure from elevated post-settlement mortality. This distinction is consequential: a 14-year record from Moorea found that early post-settlement events, rather than settlement itself, governed rates of coral recruitment and subsequent change in coral cover (Edmunds et al., 2024).

Continued research

Ongoing monitoring of coral recruitment, thermal stress, and disease prevalence remains essential for detecting early signs of reproductive decline and evaluating reef resilience. Long-term datasets such as those presented here are critical for distinguishing natural variability from persistent ecological change and for informing adaptive management strategies in an era of increasing environmental disturbance.

Active restoration approaches may help accelerate reef recovery where natural recruitment is insufficient to support population recovery. Assisted reproductive techniques, including induced spawning and larval propagation, have demonstrated promise for enhancing coral recruitment under controlled conditions (Craggs et al., 2017; Omori, 2019). Such approaches may be particularly valuable following mass bleaching events or other large-scale disturbances that severely depress larval supply.

In situ interventions aimed at protecting reproductively valuable colonies may also support reef recovery. Shading has been shown to reduce bleaching severity and thermal stress exposure, while supplemental heterotrophic feeding can enhance energy reserves and resilience (Grottoli et al., 2014; Anthony et al., 2017; Tagliafico et al., 2022). Targeted application of these strategies could help preserve large, fecund colonies that disproportionately contribute to larval production.

Without mitigation of climate-driven thermal stress, coral reefs may lose their capacity for natural reproduction and self-replacement. Although the recruitment recovery observed in this study demonstrates some resilience, continued warming is likely to exceed this capacity. Sustained local management efforts, combined with global reductions in greenhouse gas emissions, will be necessary to preserve coral reef ecosystems and the ecological and societal benefits they provide.

Agaricia spp. recruitment on three St. Thomas reefs declined by an order of magnitude during the 2019 emergence of SCTLD and the mass bleaching event that followed, remained low through 2020, and returned by 2021 and 2022 to rates statistically indistinguishable from the 2015 to 2017 baseline. Of the six environmental predictors examined, thermal stress was the only one with a detectable relationship to recruitment once differences among sites were accounted for. These results describe a brooding genus at three sites over eight sampling periods. They do not establish recovery of the coral community, coral cover, or reef function, and the response of broadcast-spawning taxa over the same period remains unknown.

Several management actions follow from these findings. First, recruitment tile monitoring should be maintained as a standing component of TCRMP rather than as episodic, project-based approach: the 2019 and 2020 results are interpretable only because a 2015 to 2017 baseline happened to exist, and no comparable baseline exists for the taxa this design does not sample. Second, deployments timed to bracket the late-summer broadcast spawning window, in addition to the current six-month intervals, would allow recruitment by broadcast spawners to be measured rather than inferred. Third, sites such as Black Point, which sustained recruitment through the disturbance and exceeded its pre-disturbance rate in Fall 2021, warrant consideration as candidate larval source populations, both for protection and as a source of material for restoration. Fourth, the approximately two-year lag between disturbance and recruitment recovery is a useful planning parameter: restoration outplanting in the window immediately following a bleaching event should not assume natural replenishment across that interval. Fifth, in situ interventions such as shading and supplemental feeding are best targeted at large, fecund colonies during forecast thermal stress, given their disproportionate contribution to larval supply.

Two monitoring gaps would need to be closed before a future recruitment failure could be attributed to a mechanism rather than inferred from its timing. Water-column larval sampling would distinguish failure of larval supply from failure of settlement, and histological assessment of gametogenesis in adult colonies would establish whether reproductive output was reduced. Neither was performed here, and both are tractable additions to an existing monitoring programme.