The effect of reef morphology on coral recruitment at multiple spatial scales

Significance

Coral reefs are foundational species for marine ecosystems and coastal livelihoods throughout the world, yet corals are increasingly threatened by ocean heatwaves, acidification, pollution, overfishing, disease, and other interacting stressors. One pathway for rebuilding reefs is to protect habitats where coral larvae are most likely to settle and survive, yet the relationship between habitat structure and recruitment is not well known. We use airborne remote sensing and in situ photogrammetry to analyze the relationship between reef morphology and coral recruitment at multiple scales. Reef morphology (rugosity and curvature) at ≥2 m resolution has a significant relationship with coral recruitment, and thus, new technologies that expand meter-scale remote sensing can identify sites where coral restoration is most viable in the future.

Abstract

Coral reefs are in decline worldwide, making it increasingly important to promote coral recruitment in new or degraded habitat. Coral reef morphology—the structural form of reef substrate—affects many aspects of reef function, yet the effect of reef morphology on coral recruitment is not well understood. We used structure-from-motion photogrammetry and airborne remote sensing to measure reef morphology (rugosity, curvature, slope, and fractal dimension) across a broad continuum of spatial scales and evaluated the effect of morphology on coral recruitment in three broadcast-spawning genera. We also measured the effect of other environmental and biotic factors such as fish density, adult coral cover, hydrodynamic larval import, and depth on coral recruitment. All variables combined explained 72% of coral recruitment in the study region. Coarse reef rugosity and curvature mapped at ≥2 m spatial resolution—such as large colonies, knolls, and boulders—were positively correlated with coral recruitment, explaining 22% of variation in recruitment. Morphology mapped at finer scales (≤32 cm resolution) was not significant. Hydrodynamic larval import was also positively related to coral recruitment in Porites and Montipora spp., and grazer fish density was linked to significantly lower recruitment in all genera. In addition, grazer density, reef morphology, and hydrodynamic import had differential effects on coral genera, reflecting genus-specific life history traits, and model performance was lower in gonochoric species. Overall, coral reef morphology is a key indicator of recruitment potential that can be detected by remote sensing, allowing potential larval sinks to be identified and factored into restoration actions.

Corals are ecosystem engineers supporting marine life throughout the tropics, yet coral cover is decreasing due to a host of both natural and anthropogenic stresses including climate change, overfishing, and coastal runoff (1, 2). As coral mortality increases worldwide, there is a critical need to understand drivers of coral loss, survival, and regrowth in changing reef habitats. The scleractinian coral life cycle begins with a larval phase, in which motile larvae settle close to parent corals or, alternatively, disperse far from the natal site by traveling along ocean currents. Larval dispersal is thought to increase genetic diversity on reefs and improve offspring survival by allowing larvae to select favorable habitat (3). Past research has identified myriad conditions that dictate larval settlement and coral recruit survival, including favorable hydrodynamics, chemosensory cues from crustose coralline algae, algal competition, herbivory, predation, and soundscape features (4). However, the role of reef morphology, i.e., the shape of the reef, in larval recruitment is not well understood.

Reef morphology is typically represented by metrics like rugosity, slope, and curvature, where higher values indicate more complex or steep substrate. Reef morphology may affect coral recruitment by altering biotic (e.g., fish) and abiotic (e.g., ocean current) aspects of reef habitat. For example, reef rugosity may help larvae settle by generating turbulence at multiple scales (5) and by influencing the height of the reef boundary layer (6), a zone where water velocity drops dramatically near the reef surface. At large scales, the shape of reef islands can influence turbulent eddies that entrain larvae in the island wake (7), and, in some fringing reefs, steep slopes are linked to internal waves that deliver high concentrations of larvae (8). At meter or submeter scales, reef rugosity promotes high fish biomass and diversity (9) and, in turn, diverse herbivores suppress algae that may harm coral recruits, especially on reefs where dominant herbivore species have traits to overcome algal defenses (10, 11). Finally, at very small (larvae-sized or mm) scales, irregular substrate surfaces attract coral recruits by increasing settlement area and creating shelter for fragile juveniles (12, 13).

Despite the potentially important role of reef morphology in coral recruitment, few studies have directly measured the effects of morphology on recruitment at scales broader than microcrevices. Moreover, past studies linking morphology to recruitment have only considered morphology at a single scale (14). One exception was a recent study by Yanovski and Abelson (15), which showed that coral recruitment at four sites exhibited a positive association with medium-scale rugosity, where artificial boulders (dimensions 0.5 × 1 × 1 m) increased recruitment by 400 to 600%. While this study was a step in examining impacts of rugosity on recruitment, it focused on localized brooding species, was limited to a handful of experimental plots, and did not consider multiple metrics of morphology. Additional research is needed to examine the effect of reef morphology at multiple scales on coral recruitment, particularly in spawning and highly dispersive species.

New technologies allow us to measure reef structure across a broad continuum of scales. Aircraft remote sensing has produced comprehensive maps of live coral cover and reef structure at 1- to 2-m spatial resolution across the Hawaiian archipelago (16, 17). In addition, structure-from-motion (SfM) photogrammetry has become a means to generate 3D reconstructions of reefs at centimeter spatial resolution. These technologies have begun to change the methodological paradigm for measuring reef structure (18). Historically, metrics like rugosity have been measured using visual estimation or by draping a metal chain across the reef bottom, which is labor intensive and extremely limited in spatial scale. In contrast, airborne imagery and SfM allow us to measure morphology at a continuum of scales from reef crevice to reefscape. We analyzed the effect of reef morphology (rugosity, slope, curvature, and fractal dimension) on coral recruitment, where morphology was measured at multiple spatial resolutions ranging from 0.01 to 6 m and within plots ranging from 10 to 54 m². We also considered several alternative drivers of coral recruitment including larval availability and fish density. Finally, we analyzed how morphology and other variables affect recruitment differently across three coral genera with varying life-history traits.

Past research on coral recruitment has primarily used field or laboratory experiments to measure coral recruitment directly or, alternatively, hydrodynamic models to simulate larval flow along ocean currents. We used these approaches in tandem (Materials and Methods). We first measured coral recruitment on 320 in situ settlement tiles at 32 reef sites and measured reef morphology at multiple spatial scales at each site using airborne imagery from the Global Airborne Observatory (GAO) and SfM photogrammetry. We then used a hydrodynamic model to simulate larval dispersal across the study region, thus accounting for differences in larval availability (hereafter “larval import”) between sites. We also considered herbivorous fish density, adult coral cover, embayment, and depth as possible drivers of recruitment. Our research was conducted in Miloliʻi in southwest Hawaiʻi Island, which in 2022 was authorized as a Community-Based Subsistence Fishing Area (CBSFA), i.e., an indigenous Hawaiian-led management zone (Fig. 1). This region, analogous to a marine reserve, has rules of use designed to sustain culturally important species and fishing practices. As a result, the reef system in Miloliʻi is less disturbed than many coral reef regions, with low human population density and algal growth, allowing us to control for anthropogenic factors and thus focus on the effect of reef morphology on coral recruitment.

Fig. 1.

(A) Hawaiian archipelago indicating the Big Island. (B) Map of the Big Island indicating the location of the Miloliʻi CBSFA. (C) Map of the Miloliʻi CBSFA (19); polygons represent Puʻuhonua (marine sanctuaries) and other management zones, with study sites shown in red.

Results

Regional Larval Flow.

Our study took place along a 30 km stretch of coastline surrounding the commmunity of Miloliʻi (Fig. 1C) in the South Kona District of Hawaiʻi Island. The region contains several Puʻuhonua or community-managed marine sanctuaries that range from Puʻuhonua Pāpā in the north to Puʻuhonua Manukā in the south. Simulated larval flow showed a strong southerly current; our simulated larval import index (Materials and Methods) accumulated steadily from northern to southern sites until peaking north of Manukā (Fig. 2). However, larval import was lowest in several southernmost sites due to late-summer westerly flow (Fig. 3). Our hydrodynamic model was consistent with local knowledge, oceanographic data (SI Appendix, Fig. S4) and past studies (20), which demonstrate the existence of a nearshore anticyclonic eddy in South Kona driving southerly flow. This result is also consistent with data from 29 short-term drifter buoys deployed from the study region (SI Appendix, Drifter Methods and Table S5). Drifter buoys traveled at a mean bearing of 209.9° and flowed south 71.4% of the time.

Fig. 2.

(A) Coral recruitment observed on tiles per site (±SE). (B) Simulated larval import per site based on hydrodynamic flow alone, rescaled 0 to 100. Sites are ordered by latitude from south (lower y-axis) to north (upper y-axis).

Fig. 3.

Results of first 5 of 30 d of larval flow simulations beginning 24 June and 22 August 2021 (expected spawning dates) based on ocean current data from the Massachusetts Institute of Technology general circulation model [MITgcm; (21, 22)]. While there is strong southerly flow in the southwest Big Island (study region), simulations in August show higher westward flow. While larval release dates reflected the lunar calendar for 2021, the most recent ocean current data was from 2013; however, summertime flows are expected to be broadly consistent from year to year.

Coral Recruitment Patterns.

Coral recruitment on in situ reef tiles per site ranged from 17 ± 8.9 to 363 ± 39 recruits m⁻² and 0.5 ± 0.3 to 10.9 ± 1.2 recruits tile⁻¹ (mean ±SE). Significantly lower recruitment occurred in the most southern (down-current) region of Manukā compared to other regions based on 95% Confidence Intervals (CIs), while the highest recruitment occurred in the northern (up-current) region of the Pāku'iku'i Rest Area (Fig. 4 and SI Appendix, Fig. S1). Pocillopora spp. comprised the majority of recruits with a mean of 85.1 ± 4.7 recruits m⁻² (2.6 ± 0.1 recruits tile⁻¹), compared to 61.3 ± 3.6 recruits m⁻² (1.8 ± 0.1 recruits tile⁻¹) for Porites spp. and 29.3 ± 2.4 recruits m⁻² (0.9 ± 0.1 recruits tile⁻¹) for Montipora spp. Based on a one-way ANOVA (Materials and Methods), differences in recruits tile⁻¹ between genera were statistically significant with P < 0.0001.

Fig. 4.

Heatmaps of coral recruit density across the Miloliʻi CBSFA by genus. Red points represent tile sites; heatmap does not reflect tile-site density.

Regional distribution of recruitment varied by genus, with Pocillopora more concentrated in northern (up-current) sites, Porites relatively higher in central and southern (down-current) sites, and Montipora highly concentrated in two specific reefal areas (Fig. 4). Recruitment differed between sites located in close proximity to each other, as confirmed by Moran’s I > 0.05. For example, Sites 172 and 60 were located <200 m apart but showed, respectively, 113 ± 18.1 and 293 ± 30.1 recruits m⁻² (3.4 ± 0.5 and 8.8 ± 0.9 recruits tile⁻¹). There was spatial mismatch between patterns of simulated larval flow (larval import) and coral recruitment, with larval import increasing to the south and coral recruitment peaking in the north (up current; Fig. 2).

Reef Morphology.

Overall, our model explained 72% of total coral recruitment. We aggregated 13 variables of reef morphology into three principal components (PC; Materials and Methods), where PC2 represented rugosity, slope, curvature, and fractal dimension measured at fine (0.01 to 0.32 m) spatial resolution and PC3 represented rugosity and curvature measured at intermediate/coarse (2 to 6 m) resolution. Higher values of both PC2 and PC3 indicated more morphologically complex reefs. PC3 (2 to 6 m morphology) was strongly and positively linked to coral recruitment (7.29 ± 2.19, P < 0.01; Fig. 5), explaining 22% of variation in coral recruitment, the highest of any covariate. That is, coral recruitment significantly increased as intermediate/coarse-scale rugosity and curvature increased. PC2 (fine-scale morphology) was positively, but weakly, linked to coral recruitment (3.69 ± 2.68, P < 0.1). Positive values of PC2 represented, specifically, morphological metrics at 0.08 to 0.32 m resolution (SI Appendix, Fig. S2), whereas negative values represented 0.01 to 0.04 m resolution. Thus, higher rugosity, slope, and curvature at 0.08 to 0.32 m were linked to increased recruitment; however, this relationship was not significant.

Fig. 5.

Relationship between significant covariates and coral recruitment (all genera). Coral recruitment has a positive relationship to PC3 (intermediate- and coarse-scale morphology) and adult coral cover, and recruitment has a negative relationship to grazer density. Bands show 95% CI of fitted values; points represent observed data.

Additional Drivers of Recruitment.

Additional covariates showed a strong link to coral recruitment. Adult coral cover was significantly associated with higher coral recruitment (5.43 ± 2.68, P < 0.05). Conversely, both fish (grazer) density (−14.95 ± 3.01, P < 0.001) and embayment (−18.66 ± 6.08; P < 0.01) were negatively and significantly correlated with coral recruitment (Fig. 5 and Table 1). Since fishing pressure is highest in northern sites, which also showed the highest recruitment, we tested for latitude as a possible confounding variable for the effect of fish density; however, latitude was not well correlated with grazer density (Pearson’s correlation = −0.47). In sum, coral recruitment was higher in sites with many adult corals, but lower inside embayments and in areas with dense grazer fish. Larval import (estimate of larval inflow based on ocean current) and depth were not retained in the model of best fit for “all genera” but influenced genus-specific models (see below).

Table 1.

Coral recruitment model results where (a) morphology is modeled across all scales (represented by PC1–3) based on SfM and airborne remote sensing, and (b) morphology is modeled using only 2 m rugosity and curvature (aircraft remote sensing)

Covariate	All coral recruits	Pocillopora	Porites	Montipora
(a) All morphology (GAO airborne remote sensing and SfM)
Adult coral cover	5.43* (2.68)	–	–	–
Larval import	–	–	0.23* (0.10)	0.80*** (0.17)
Grazer fish density	−14.95*** (3.01)	−0.45*** (0.08)	−0.23* (0.10)	−0.31* (0.13)
PC1 (SfM dominated)	–	–	–	–
PC2 (SfM + GAO)	3.69† (1.92)	0.16** (0.05)	–	–
PC3 (GAO dominated)	7.29** (2.19)	0.31*** (0.06)	0.22** (0.08)	–
Bay enclosure	−18.66** (6.08)	−0.45** (0.16)	NA	–
Depth	–	–	–	0.23* (0.11)
R2	0.72	0.76	0.28	0.56
(b) Intermediate-scale morphology only (GAO airborne remote sensing)
Adult coral cover	6.96* (3.09)	–	–	−0.28** (0.10)
Larval import	–	–	0.23* (0.11)	0.56*** (0.12)
Grazer fish density	−16.74*** (3.32)	−0.44*** (0.10)	−0.20† (0.11)	−0.26* (0.11)
GAO 2 m rugosity	7.25* (3.16)	0.26** (0.09)	0.22* (0.12)	0.34** (0.11)
GAO curvature	–	–	–	–
Bay enclosure	−26.30*** (6.48)	−0.80*** (0.18)	NA	–
Depth	–	−0.20* (0.09)	–	–
R2	0.61	0.47	0.18	0.63

^†P < 0.1.

^*P < 0.05.

^**P < 0.01.

^***P < 0.00.

Significant variables are in bold; parentheses denote SE. A traditional R² does not apply to genus-specific models because they used a log-link function. Here, R² denotes McFadden’s R², which increases with the difference between Null Deviance (deviance with no covariates) and Residual Deviance (deviance with all covariates). See SI Appendix, Table S3, for model selection statistics.

Differences between Genera.

Each coral genus displayed distinct differences in relationships between reef morphology and recruitment (Fig. 6 and Table 1). For Pocillopora, both fine-scale morphology (PC2, as described above; 0.16 ± 0.05, P < 0.01) and intermediate/coarse-scale morphology (PC3; 0.31 ± 0.06, P < 0.001) were strongly and positively related to recruitment, though intermediate/coarse-scale morphology showed a higher and more significant effect. In the Porites model, only intermediate/coarse-scale morphology showed a significant link to recruitment (0.22 ± 0.08, P < 0.01), and in the Montipora model, there was no significant correlation between any morphological variable and recruitment.

Fig. 6.

Relationship between significant covariates and coral recruitment by genus from Table 1. Covariate significance and effect sizes differ by genus. Bands show 95% CI of fitted values; points represent observed data.

Larval import was significantly related to both Porites and Montipora recruitment, with a particularly strong effect on Montipora (0.23 ± 0.10 and 0.80 ± 0.17 for Porites and Montipora, respectively). In contrast, larval import was not significantly linked to Pocillopora recruitment. Fish density was significantly and negatively related to recruitment across all genera but had a particularly strong and significant, negative relationship with Pocillopora. Finally, depth was positively related to Montipora recruitment, i.e., recruitment increased at deeper sites, and embayment was negatively related to Pocillopora recruitment. Model performance based on a comparison of null and residual deviance was highest for Pocillopora (McFadden’s R² = 0.76), followed by Montipora (R² = 0.56) and Porites (R² = 0.28).

Discussion

While some aspects of coral recruitment are well understood, the effect of reef morphology on coral recruitment has rarely been studied at scales broader than very localized microtopography. Moreover, few studies have compared the effects of reef morphology across multiple ecological scales. We used in situ (SfM) and reef-scale (airborne) remote sensing to access a broad continuum of spatial scales in reef morphology, ranging from 0.01 to 6 m spatial resolution within plots ranging from 10 × 10 m to 54 × 54 m. By combining these methods with larval flow modeling and in situ measurements of coral recruits, we found that rugosity and curvature measured at ≥2 m spatial resolution strongly affects coral recruitment, and thus, coarse obstructions like large coral colonies, knolls, and boulders facilitate higher coral recruitment. Additionally, larval import, morphology, fish density, bay enclosure, and depth each have differential effects on recruitment among coral genera. Overall, variables explained 72% of recruitment in the study region, indicating high explanatory power for coral recruitment. This exceeds the explanatory power of many previous attempts to explain settlement on tiles, with recent studies finding R² = 18 to 33% (23) and R² = 57% (24).

Fine-Scale Morphology.

Fine-scale morphology (0.01 to 0.32 m rugosity, slope, curvature, and fractal dimension) did not show a significant relationship with coral recruitment, except in the case of Pocillopora. This result does not inherently suggest that fine morphology is inconsequential since one benefit of fine rugosity and slope is increased surface area for settlement, and our tiles provided a constant 0.1 × 0.1 m settlement area. However, our results do indicate that benthic flow regimes surrounding reefs with high fine complexity (Fig. 7 A–C) are not as important to coral recruitment as flow regimes surrounding meter-scale benthic features such as large reef faces, knolls, and drop-offs (Fig. 7 D–F). While we found little evidence of benefit from fine complexity, previous studies show that microcrevices (25–27), roughness at the 10-mm scale, and flat, exposed reef faces at the 100-mm scale (28) favor coral recruits and increase particle exchange that helps corals grow. However, in our study, fine-scale morphology affected only Pocillopora, perhaps because Pocillopora is particularly vulnerable to predation and small-scale complexity helps to shelter corals from predation (full discussion below).

Fig. 7.

Reefs with contrasting fractal dimension, shown as (Left column) diver photos, (Center) SfM 3D orthomosaics, and (Right) Digital Elevation Models. (A–C) Reef with a high fractal dimension, in which reef surface area increases primarily through small-scale complexity. (D–F) Reef with a low fractal dimension, in which coarse imagery captures most of reef complexity.

Intermediate/Coarse-Scale Morphology.

Reef morphology (rugosity and curvature) at ≥2 m spatial resolution, represented by PC3 and measured by aircraft remote sensing, was positively and significantly related to coral recruitment. Among all covariates, ≥2 m morphology explained the most variation in coral recruitment (22%). This finding aligns well with experiments on larval flow across reef structures. Hata et al. (5) used experimental flumes to demonstrate that reef complexity can generate turbulence, i.e., chaotic changes in currents that retain larvae and give them time to settle. While this study focused on small (cm) scales, authors hypothesized that larger-scale reef obstacles could also facilitate turbulent flow that promotes settlement. Boulders and large colonies may also entrain larvae by generating a thick boundary layer. When water moves across a surface, drag creates a layer where water moves slower than mainstream flow (29). In Kāneʻohe Bay, Hawaiʻi, Shashar et al. (6) studied coral knolls rising 0.8 to 1.5 m above the reef bottom and found that the height of knolls correlated with the width of the “inner benthic boundary layer” where water flow was minimal. In addition to retaining particles, turbulence behind large obstructions may be a cue in larval ontogeny. For example, in sea urchins, turbulence signals arrival in a coastal area to the larvae, accelerating the transition of precompetent larvae to competence (30).

In Hawaiʻi’s reef system, where periodic volcanic flows have generated basalt boulders throughout the reef (Fig. 7 D–F), boulder-scale complexity is key in slowing water velocity and retaining larvae. However, these processes might also retain sediment, slow food delivery, and reduce water mixing during heatwaves, which may impede coral growth and survival. In addition, turbulent flow surrounding obstructions may depend on depth given that current velocity, wind, and wave energy vary markedly with depth. The interaction of intermediate/coarse-scale morphology and depth should be examined in future research. Nonetheless, results demonstrate that rugosity and curvature ≥2 m plays an important role in recruitment. Given its rough, volcanic substrate nearshore, Hawaiʻi may be favorable ground for coral restoration.

Given the influence of ≥2 m morphology on coral recruitment, we tested whether aircraft remote sensing alone could represent morphology effects. In other words, are in situ methods like SfM necessary to help explain coral recruitment in South Kona, or do airborne data suffice as a proxy? We found that replacing all scales of morphology with aircraft remote sensing/GAO reduced our model R² from 0.72 to 0.61, i.e., our model using all morphology scales (0.01 to 6 m resolution) performed best. However, given the time and computational cost of SfMs, using aircraft-based morphology alone may help to scale up recruitment monitoring.

Additional Drivers of Recruitment.

Fish (grazer) density was negatively correlated with coral recruitment across all genera. While herbivorous fish typically increase coral recruitment by grazing down algae (31–33), herbivores and other fish can also cause recruit mortality by dislodging recruits during grazing or predation (34–36). For example, in French Polynesia, grazers caused 50% of mortality in recruits (37). The effect of grazers may have been negative and significant in our case because we examined corals within the first 1 to 4 mo of life, when they are particularly vulnerable to dislodgement (36, 38). In addition, Miloliʻi has little sedimentation, low effluent, and relatively high fish density throughout the region, and thus algae in Miloliʻi is well controlled, perhaps making fish predation/dislodgement the dominant ecological signal. Despite this result, there are demonstrable long-term benefits of herbivores to recruits that may not have been captured by our study (32). In addition to herbivores grazing algal competitors, thus supporting recruit survival, parrotfish can scrape small divots or microstructure that may support recruitment and would not have been captured by settlement tiles (39). Moreover, herbivory may result in lower direct coral contact with algae species that generate harmful, lipid-soluble metabolites (40). Previous research has shown that certain species and mixes of herbivorous fish promote algal control better than others, thus, our fish metric may be improved in the future by identifying separate effects of specific fish species and communities (10, 11).

Adult coral cover was positively related to coral recruitment in our model for all genera. However, this effect was weak relative to other model covariates and absent in genus-specific models. Thus, adult coral cover at the natal site was not a major driver of recruitment, underlining the importance of larval connectivity and environmental conditions in supporting recruitment. Note that adult coral cover did not covary with complexity measured at any scale (Pearson’s correlation = 0.06 to 0.21), and therefore was not responsible for the effect of morphology. Embayment was negatively correlated with recruitment, consistent with past literature in Hawaiʻi showing higher recruitment in sites located offshore rather than inshore (41, 42). One possible explanation is that exposed sites are frequently marked by smaller colonies (42). Coral recruitment may be higher in reefs with smaller colonies because adult coral cover has a density-dependent relationship with coral settlement; past research observed a saturation point in coral settlement at 10% coral cover (43).

Differences among Genera.

Recruitment patterns were distinctly different among coral genera. While Pocillopora represented 46% of coral recruits on tiles, Pocillopora comprised only 0.3% of adult colonies observed at tile sites. Conversely, Porites spp. made up 37% of tile recruits but 95% of adults at our sites. These patterns are not surprising given life history strategies of our focal genera. Pocillopora spp. found in Hawaiʻi have been classified as “weedy” and “competitive,” that is, R-selected species that produce high numbers of offspring, each with a low probability of surviving to adulthood (44, 45). R-selected species colonize new habitat quickly after disturbance (46), but only thrive under ideal conditions and are more susceptible to bleaching, disease, and predation (44, 47). In contrast, Porites lobata—a hyperdominant in Hawaiʻi—is a K-selected species with greater investment in fewer offspring, slow growth rates, high stress tolerance, and lower vulnerability to predation (47, 48). Therefore, we likely detected disproportionately high Pocillopora recruits due to their weedy R-selected dispersal strategy.

Porites and Montipora alone were significantly correlated with simulated larval flow (larval import) from other reefs. Larval import may have been disconnected from Pocillopora recruitment because Pocillopora spawn earliest (April–May) and thus had the longest exposure to postsettlement factors (e.g., predation) before we retrieved tiles. Conversely, the effect of larval import was highest and most significant for Montipora spp. (Fig. 6), which spawn latest in the season (SI Appendix, Table S1), i.e., closest to our time of collecting tiles. In addition, Montipora was the only coral genus that showed no significant link between reef morphology and coral recruitment. One possible reason is that Montipora eggs are fertilized at the surface whereas the eggs of Porites and Pocillopora spp. in West Hawaiʻi are fertilized in the water column (49). Therefore, Montipora larvae may have higher exposure to surface currents whereas Porites and Pocillopora are more dependent upon entrapment by rugose turbulence or the inner benthic boundary layer for successful fertilization.

While grazer density was negatively correlated with recruitment in all genera, this effect was strongest in Pocillopora and weakest in Porites (Fig. 6). This result aligns with past research indicating that Pocillopora spp. are highly vulnerable to predation, whereas Porites lobata is less vulnerable (47, 50). For example, in one experiment, Pocillopora damicornis experienced near-complete mortality after 8 d under predation pressure, while 63% of Porites lobata survived until the end of the experiment at 29 d (47). Both Porites and Montipora spp. may evade predation due to an encrusting structure (47). As noted above, the specific vulnerability of Pocillopora to predation may also explain why PC2 (fine-scale morphology) was significant for Pocillopora alone. Fine-scale rugosity provides shelter from predators, which may be more important for Pocillopora than other genera in our study.

Model performance also differed among genera. Pocillopora spp. were particularly well explained by our model (R² = 0.76), followed by Montipora (R² = 0.56) and Porites (R² = 0.28). The Porites model may have performed poorly because dominant Porites spp. on Hawaiʻi Island are gonochoric while Montipora and Pocillopora are hermaphroditic (49). Since Porites eggs and sperm do not emerge from the same colony, fertilization may be more stochastic or depend upon fine-scale hydrodynamic aggregations of eggs and sperm that our model did not capture, such as surface slicks that are common in the study region (51). Future research should account for Porites gonochorism by using finer-scale hydrodynamic modeling and estimating water residence time at coral sites, which may impact the potential for eggs to meet sperm. This step is particularly critical because Porites lobata, lutea, and compressa are dominant adult species on Hawaiian reefs. Climate stressors increasingly drive mortality among weedy species like Pocillopora, but these species still produce high numbers of offspring that can dominate settlement tiles. Therefore, future recruitment studies should be attuned to the reproductive traits of stress-tolerant corals like many Porites that are more likely to survive into the adult phase.

Importantly, coral settlement as observed on tiles does not necessarily equate to recruit survival and growth in the adult stage. Past research shows that the presence of consumers, herbivores, and coral density (adult and larval densities) have differential effects on early recruitment vs. survival (32, 39, 43, 52, 53). For example, in Palmyra Atoll, reduced herbivory promoted higher coral recruitment over 1 mo but, by 4 mo, led to increased non-CCA algae that impeded coral survival through competition (39). The presence of consumers has been shown to reduce beta diversity and stabilize reef communities over time, slowing down succession from turf to erect and corticated algae that can compete with corals (53). In later adult stages, coral cover depends on a variety of factors including depth, irradiance, herbivores, heat stress, nearshore development, and wave energy (54). While settlement in our study is a critical first step toward restoration planning, restoration managers must also account for long-term survival.

Management Implications.

Our findings have important implications for reef management under climate change. We find that local adult coral cover is not as important in coral recruitment as intermediate/coarse rugosity and curvature, which can help entrain larvae. In addition, we found that simulated larval import, i.e., larval flow to each site, was significant for two coral genera in our study. Together, these results suggest that larvae originating up-current (for example, in coral bleaching refugia) may seed degraded reefs if entrained by coarse complexity. In islands like Hawaiʻi, where coarse features like basalt boulders (Fig. 7 D–F) arise from abiotic, volcanic activity rather than biotic reef alone, corals may have high recruitment potential. Recruitment in the first 12 mo of life is strongly correlated with reef recovery after disturbance (55), thus, promoting early recruitment is a critical step toward reef restoration.

Remote sensing coupled with oceanographic modeling represents powerful tools for advancing coral restoration. Remote sensing can be used to map habitat suitability, e.g., rugosity and curvature at 2 to 6 m spatial resolution. Meanwhile, remote sensing of live coral cover can be used to seed larval flow models that indicate sources and sinks of larval dispersal. These tools, in turn, can support the following actions:

• Protect against reef flattening in larval sinks: In areas that are larval “sinks,” adult corals that sustain complexity should be protected from reef flattening after future bleaching events by reducing local conditions linked with coral cover decline during heatwaves. In Hawaiʻi, management scenarios that help retain coral cover include decreasing sediment input, urban runoff, and wastewater pollution on reefs while increasing scraper biomass (56).

• Subsidize reef complexity in larval sinks: Where complexity has been lost due to coral bleaching, anchor damage, or other human impacts, coarse complexity should be subsidized by artificial reefs, especially in larval sinks. Hawaiʻi contains a patchwork of low- and high-complexity benthic habitat due, in part, to volcanic flow that varies in structure between pahoehoe (smooth and lobed) or 'a'a [coarse rubble; (57)]. While high-complexity reefs have been linked to high fish richness and abundance, low-complexity reefs can bolster predator success (58). Further research is needed on how a blend of high- and low-complexity reefs affect predator-prey interactions, and how artificial reefs may be sited to mimic and support the natural Hawaiian reef system across multiple trophic levels.

• Site coral restoration to disperse larvae widely: Marine spatial planning is widely used to site marine reserves but seldom coral restoration sites. Where possible, active coral restoration (e.g., outplanting) should be sited in areas from which outplants and future coral generations can disperse larvae widely.

• Preserve coral refugia as larval sources: Previous research (56, 59, 60) outlines possible drivers of coral refugia from marine heatwaves. Coral refugia that are connected to high-complexity areas (larval sinks) should be protected from stressors that inhibit adult reproduction and larval dispersal, such as light pollution during critical spawning periods.

• In West Hawaiʻi specifically, results point to the need to promote CBSFA-wide connectivity using actions coordinated between embayments. Examples include protecting the high coral cover of possible “source reefs” like Alika Bay and Pāpā Bay (Fig. 1) by mitigating the effects of new development (56), which can be particularly damaging in rural areas (60), while protecting complexity in the south (e.g., Kapua Bay) by monitoring against anchor damage.

Such actions may help to enhance coral recruitment in the context of intensifying marine heatwaves.

Materials and Methods

Study Area.

Our study was conducted in Miloliʻi, a fishery and indigenous community in South Kona. Historically, small (1 to 5 km) regions within Miloliʻi restricted aquarium fishing and/or lay nets (61), but in 2022 the State of Hawaiʻi authorized an indigenous-led sanctuary that substantially expanded the geographic scale (~30 km) and indigenous leadership of reserve management (19). Miloliʻi is an ideal location for studying coral recruitment. Previous studies have indicated that there is limited larval import from other islands to Hawaiʻi Island (62), making Hawaiʻi Island a bounded hydrodynamic system. In addition, due in part to stratification in volcanic substrate and varying human pressure, South Kona contains a wide range of reef habitats within a relatively small, accessible area. While reefs in Miloliʻi vary in structural complexity, coral cover, depth, and fish density, there is very little algal cover in the region due, in part, to low effluent (63). Hawaiʻi has experienced several major bleaching events over the past decade (59) and, in response, this research provides valuable information to coral restoration activities that are accelerating statewide.

The dominant coral genera in South Kona are Pocillopora, Porites, and Montipora, which are predominantly broadcast spawners. Past, archipelago-wide genetic analysis shows evidence of isolation by distance in P. lobata and fine-scale genetic structuring with regional grouping (genetic breaks between regions) in M. capitata (64), suggesting that there is dispersal potential in both genera that is constrained by distance and regional retention. All dominant genera in South Kona spawn during summer months, with Pocillopora spawning earliest and Montipora latest (timing and references in SI Appendix, Table S1).

Coral Recruitment.

We measured coral recruitment in three dominant genera (Pocillopora, Porites, and Montipora) in the field by deploying 320 coral settlement tiles at 32 sites (10 tiles per site) across the Miloliʻi CBSFA region during the 2021 coral reproductive season in Hawaiʻi (April–September 2021) following methods adapted from Friedlander and Brown (41). We used natural unglazed limestone tiles with dimensions 2.5 × 10 × 10 cm³, electing to use limestone due to its natural microcrevices, which can enhance coral spat survival (26). Tiles were deployed at depths of 10 ± 3 m and spaced 1 to 3 m apart. After retrieving tiles from the ocean, we counted and identified coral larvae to genus level based on skeleton morphology. Tile sites were evenly balanced between embayments (18 sites) and nonembayments (14 sites) and represented varying levels of adult coral cover and structural complexity via stratified sampling. Details on site selection, tile design, and deployment can be found in SI Appendix, Settlement Tile Methods and Fig. S3.

Larval Import.

We measured hydrodynamic larval import at each of our 32 tile sites using a two-dimensional Lagrangian Particle Tracking (LPT) model, which estimated the advection and diffusion of passive particles from release (spawning) locations throughout West Hawaiʻi. We employed the Ocean Parcels framework, a collection of Python methods and classes used in numerous previous studies (e.g., ref. 65) to calculate the displacement of Lagrangian particles after particle release (spawning) based on the fourth-order Runge-Kutta implementation (66). We first initiated a cloud of virtual larvae at all possible spawning locations based on high-resolution (2 m) aircraft imagery of coral cover in West Hawaiʻi from remote sensing/GAO. GAO mapped coral cover by aircraft using high-fidelity imaging spectroscopy based on a visible-to-shortwave infrared (VSWIR) imaging spectrometer, flown on multiple dates from January 2 to February 4, 2019 (16). We created a cloud of virtual larvae where 1 to 100 larvae were released at the centroid of each pixel, proportional to coral cover within each pixel, for a total of 12,508,650 larvae. We then displaced larvae via advection and random diffusion starting on expected spawning dates (SI Appendix, Table S1) using a regional nest of the Massachusetts Institute of Technology general circulation model [MITgcm; (21)]. Finally, we calculated larval import as an index representing the relative number of simulated larvae that intercepted each tile site. Details regarding our LPT model including all simulation parameters can be found in SI Appendix, Larval Flow Simulation Methods.

Reef Morphology.

We derived reef morphology from a combination of SfM photogrammetry and GAO remote sensing, which allowed us to measure reef features at 0.01 to 6 m spatial resolution. GAO mapping is described above and SfM methods are in SI Appendix, SfM Methods. All morphology metrics are defined in SI Appendix, Table S2.

We used SfM data to calculate fine-scale reef morphology based on the following metrics: reef rugosity at spatial resolutions of 1, 2, 4, 8, 16, and 32 cm; maximum reef curvature and slope at 32 cm resolution, and fractal dimension (SI Appendix, Table S2). Rugosity measures the complexity of the seafloor, slope indicates the vertical grade of the reef surface, and curvature indicates areas where the reef slope or aspect change quickly, i.e., walls and ledges (67). Specifically, 0.32 m slope and curvature captures sheer dropoffs in substrate, rather than small vertical faces within a colony. Finally, fractal dimension is a single indicator tracking change in reef rugosity across multiple scales of measurement. While surface area always increases as cell resolution increases, this change is slight in reefs with low fractal dimension. For example, fractal dimension tends to be low in reefs with large, smooth boulders and high in reefs with many tightly packed, sinuous surface features where fine complexity contributes most to rugosity (Fig. 7).

We derived intermediate- and coarse-scale reef morphology from GAO data (17, 68). We used two rugosity metrics for morphology representing spatial resolutions of 2 m (within a 3-cell moving window) and 6 m (within a 9-cell moving window). Asner et al. (17) tested multiple resolutions and determined that these scales of rugosity best represented large colonies and basalt boulders, which are common in Hawaii’s volcanic substrate, as well as larger-scale geological patterns like shelves, ledges, spur-and-groove reef features, and other reef accretion and erosion processes. Finally, we used reef curvature (2 m resolution within 3-cell moving window) and percent sand cover from Asner et al. (68).

Other Covariates.

We measured adult coral cover as a proxy for larvae supply at the natal site, defining coral cover as the maximum percent coral cover from GAO imagery within a 10-m buffer of each site coordinate. We used coral cover maxima because coral fertility is linked to coral colony size (69) and we expected maxima to capture large, highly productive colonies at the natal site. In addition, mean and maximum coral cover were highly correlated (Pearson’s correlation = 0.85).

Grazer fish density (fish m⁻²) was measured at each tile site from Underwater Visual Counts (UVCs) conducted by divers from the South Kona Intensive Reef Survey in June–July, 2021, starting 1 mo after tile installation (SI Appendix, Fish Survey Methods). Fish in these surveys were classified as scrapers, grazers, and browsers based on trophic data in Heenan et al. (70). A list of species we classified as “grazers” can be found in SI Appendix, Fish Survey Methods. We used only grazer fish density because fish densities from other functional groups were highly correlated with grazers (Pearson’s correlation > 0.70) and grazers have demonstrated benefits for turf control on Hawaiian reefs at 0 to 30 m (71). Since turf is the dominant algal component on Hawaiian reefs (72), we expected grazers to influence coral recruitment in Hawaiʻi. We used fish density rather than the common metric of fish biomass because Foo and Asner (71) found that herbivore fish density but not biomass is significantly associated with decreased turf algae in West Hawaiʻi, our study region.

Finally, we measured depth at the center of each tile plot while installing tiles and defined bay enclosure as a binary variable (1 = site within an embayment; 0 = site outside of an embayment).

Data Analysis.

Our dataset contained 32 tile sites and a high number of morphology covariates, and therefore, we first reduced dimensionality in morphology covariates. We used a Principal Component Analysis (PCA) to reduce 13 morphology variables to 3 components that captured 81.7% of variation in morphology between sites. To determine the number of components to use, we applied 10-fold cross-validation to find the RMSE associated with using zero to all PCs; error was lowest at three PCs (SI Appendix, Fig. S5) and percent of variance explained also tapered at three PCs. Each PC was dominated by a different scale of morphology: PC1 captured all morphology metrics evenly; PC2 was heavily dominated by SfM (0.01 to 0.32 m) morphology, and PC3 was dominated by GAO (2 to 6 m) morphology (SI Appendix, Fig. S2).

After PCA, we used multiple linear regression to identify variables correlated with coral recruitment. Before analysis, we ensured that independent variables had a correlation coefficient < 0.7 (SI Appendix, Fig. S6). Regression parameters included the three morphology PCs above and five other covariates: depth, bay enclosure (binary), grazer fish density, larval import, and adult coral cover. Noninfluential covariates were trimmed through best-subset selection based on AIC, C_p, and adjusted R² values using leaps and bestglm packages in R (73, 74). Final models included three to four covariates each. Models using PCs had 30 sites since SfMs were not available for two sites; models using GAO but not SfM morphology had 32 sites. While our sample size was low, the process of maintaining tile sites and counting recruits is physically arduous, and for this reason, tile research typically uses only a few tile sites. Our study represents the higher end of tile sampling.

For all models, we compared fitted to observed values, visually determining that the relationship was linear but not overfitted. We conducted standard diagnostics on variable correlation, outliers, and residuals (SI Appendix, Fig. S6) and a Moran’s I test to detect evidence of spatial autocorrelation; there was autocorrelation only in our model for Montipora (see below). We used bootstrapping to compare mean coral recruitment between CBSFA regions. To bootstrap, we sampled with replacement from Puʻuhonua (embayments) with 10,000 iterations to estimate 95% CIs of recruits tile⁻¹ within each Puʻuhonua.

Genus-Specific Models.

We replicated the procedure above using recruits from each genus as a dependent variable, using genus-specific values for “adult coral cover” and larval import. We used generalized linear models (GLMs) with a Negative Binomial distribution for all genus-specific models, which are appropriate for overdispersed count data with low (0 to 20) counts. Moran’s I test found spatial autocorrelation in Montipora models, and thus, we applied Moran eigenvector GLM filtering to this model, adding a spatial eigenvector to regression terms to reduce residual autocorrelation below a 0.05 threshold (75). In our Porites model, embayment (binary) was negatively correlated with larval import, so we removed the embayment variable from analysis. Finally, we tested whether there were significant differences in recruit counts per genera using a one-way ANOVA, with genus as an independent variable and recruit counts tile⁻¹ as dependent variable.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix. Global Airborne Observatory data are available via Zenodo at https://zenodo.org/records/4294332 (76) and https://zenodo.org/records/4777345 (77).