27 years of benthic and coral community dynamics on turbid, highly urbanised reefs off Singapore

Coral cover on reefs is declining globally due to coastal development, overfishing and climate change. Reefs isolated from direct human influence can recover from natural acute disturbances, but little is known about long term recovery of reefs experiencing chronic human disturbances. Here we investigate responses to acute bleaching disturbances on turbid reefs off Singapore, at two depths over a period of 27 years. Coral cover declined and there were marked changes in coral and benthic community structure during the first decade of monitoring at both depths. At shallower reef crest sites (3–4 m), benthic community structure recovered towards pre-disturbance states within a decade. In contrast, there was a net decline in coral cover and continuing shifts in community structure at deeper reef slope sites (6–7 m). There was no evidence of phase shifts to macroalgal dominance but coral habitats at deeper sites were replaced by unstable substrata such as fine sediments and rubble. The persistence of coral dominance at chronically disturbed shallow sites is likely due to an abundance of coral taxa which are tolerant to environmental stress. In addition, high turbidity may interact antagonistically with other disturbances to reduce the impact of thermal stress and limit macroalgal growth rates.


Methods
Long-term monitoring of benthic and coral community structure. Surveys of benthic and coral community structure were conducted at a total of 15 sites to the south of the main island of Singapore between 1986 and 2012. Not all sites were surveyed in all years, although at each site exactly the same survey method was used on each occasion. For a full description of sites and the sampling regime see Supplementary Information (Table S1, S2 and Fig. S1). Surveys were carried out at two depths, hereafter known as: "shallow" (the reef crest, approx. 3-4 m depth) and "deep" (upper reef slope, approx. 6-7 m depth) by laying five replicate 20 m transects parallel to the reef crest. Transects were laid haphazardly on each occasion to ensure independence among time points.
SCUBA divers used the line intercept transect method (LIT) 44 to estimate benthic percentage cover. The LIT method involves a diver estimating the length, to the nearest centimeter, of different benthic categories lying underneath a tape measure laid across the reef. The benthic categories used were: hard coral, macroalgae (i.e., fleshy seaweeds, e.g., Sargassum spp.), epilithic algal turfs (i.e., epilithic algal matrix (EAM) 45 ), crustose coralline algae (CCA), sponges, soft corals, zoanthids and other living benthic organisms. "Abiotic" benthic categories (i.e., substrate without conspicuous macrofauna) included recently dead coral, rock, sand, unconsolidated rubble and fine sediments (e.g., clay, silt). The genera and growth forms of all hard coral colonies encountered along each transect were also recorded. These survey data were used to quantify temporal changes in: (i) cover of benthic categories (e.g., macroalgae, hard coral etc.), (ii) overall benthic community structure and (iii) taxonomic coral community structure.
Scientific RepoRts | 6:36260 | DOI: 10.1038/srep36260 Data analysis. We performed three types of analyses on the benthic data to examine: (i) the effect of time and depth (and their interaction) on cover of eight benthic categories (hard coral, macroalgae, rubble, EAM, fine sediments, CCA, other substrata [e.g., sand, rock] and other biota [e.g., sponges, zoantharians]), (ii) the effect of time and depth on overall coral taxonomic community structure and (iii) changes in multivariate benthic and coral taxonomic community structure before and after bleaching disturbances at four selected shallow sites.
Changes in percentage cover of hard coral, macroalgae and other benthic categories were modelled as a function of time (using a splines smoothing function for years to account for potential correlation among sampling times for each site), depth ("shallow" vs "deep") and their interaction using Generalised Additive Mixed Models (GAMM) assuming a binomial distribution, with sites as random effects 46 . GAMMs were fitted instead of Generalised Linear Mixed Models (GLMM) because inspection of the residuals in the model validation showed a clear non-linear pattern 46 . Results of the GAMM analysis confirmed that the relationships between time and benthic cover of coral, macroalgae, EAM, rubble, fine sediments, CCA, other substrata and other biota at both depths were all significantly non-linear (Table 1). Analyses were done using the gamm4 package in R 47 .
To visualise changes over time in overall coral taxonomic community structure, non-metric multidimensional scaling (nMDS) ordinations (averaged over all sites) were produced based on Bray-Curtis similarities of square-root transformed data. To determine potential drivers of community patterns, vectors based on Pearson correlations > 0.7 were overlaid on nMDS plots to show which taxa correlated most with change over time (taxa contributing < 5% to overall cover were removed) 48 .
During the study period, two thermal bleaching events occurred on Singapore's reefs 40,49 . The first documented bleaching event in Singapore was in 1998 during the El Niño event that caused major thermal stress and bleaching related mortality on many reefs in the region 40,50 . The second event occurred in 2010 during a thermal anomaly that severely affected reefs throughout the Andaman Sea and parts of the South China Sea 40,51,52 . The potential of a chronically disturbed system to recover from a bleaching disturbance was analysed using benthic data from four of the shallow sites (H2, HW1, R2 and S2) (Supplementary Information, Table S1, S2, Fig. S1). Only this subset of sites was used because, among the 15 survey sites, they were the only ones surveyed multiple times within each time period, thus allowing us to decouple bleaching effects from natural temporal variability in community structure. Results from the deep sites were not analysed in this way because there was no evidence of recovery post 1998 at these sites (see Results & Supplementary Information Table S2). Overall benthic community structure and coral community structure were compared among 3 time periods using a 2-factor permutational multivariate analyses of variance in the PERMANOVA+ add-on in PRIMER v6 (factors: i] Period, fixed with 3 levels, ii] Years nested in Period) 53 . The time periods were (1) before the first bleaching event ('pre-1997-98': years 1988-1993), (2) after the first bleaching event but before the second bleaching event ('post-1998/pre-2010': years 1999-2009), and (3) after the second bleaching event ('post-2010': years 2011-2012). Bray-Curtis similarity matrices were calculated on square-root transformed data and analyses used 9999 permutations of residuals under a reduced model. SIMPER was used to determine the benthic categories and taxa contributing most to observed differences across periods 53 .

Results
Temporal dynamics of cover of hard coral, macroalgae and other benthic categories. The relationship between time and benthic cover differed between deep and shallow sites for hard coral, macroalgae, EAM, rubble and fine sediments (significant Year* Depth interaction, Table 1). Between 1988 and 2012 there was an overall ~12% decline in mean coral cover at the shallow sites, whereas at deep sites there was a decline of almost 30% between 1986 and 2012. In contrast, over the same time periods, mean cover of macroalgae only increased by ~4% at the shallow sites and by only ~1% at deep sites ( Fig. 1, Supplementary Information, Tables S1 and S2). At shallow sites, mean coral cover among years ranged from ~25% to ~49% with the highest coral cover recorded when monitoring started (1988)(1989)(1990)) and the lowest coral cover recorded in 1997-1998. At shallow sites the greatest decline in coral cover occurred during the decade between 1988 and 1998 (2.41% y −1 ), but by 2008, coral cover had recovered to approximately 1993 levels (~40%) although a second, smaller decline in coral cover of ~9% occurred between 2009 and 2011 ( Fig. 1a, Supplementary Information, Table S1). The largest increase in mean macroalgal cover (from ~3% to ~16%) occurred between 1988/1990 and 1999/2000 corresponding with the largest decline in coral cover (Fig. 1b, Supplementary Information, Table S1). Average cover of rubble increased from ~9% in 1988/1990 to ~32% in 1999/2000, but decreased to ~20% by the end of the study (Fig. 2a). At shallow sites, cover of other benthic categories (EAM, fine sediments, other substrata, CCA and other biota) varied from year to year, but did not show an overall decline or increase between the beginning and end of the study (Fig. 2b-f). At the deep sites, mean coral cover among years ranged from ~13% to ~45% with the highest coral cover recorded during the first time period (1986)(1987) and the lowest coral cover recorded in 2003. The greatest decline in mean coral cover occurred between 1986/1987 and 2003 (1.90% y −1 ), after which time coral cover remained relatively stable (ranging from ~13 to 16% total cover), with no recovery to historical levels ( Fig. 1c, Supplementary Information, Table S2). Mean cover of macroalgae generally remained very low at deep sites (< 4%) with highest cover found between 1997 & 2000 when it increased to ~8% (Fig. 1d, Supplementary Information, Table S2). Mean cover of both rubble and EAM approximately doubled during the first decade of the study, but by the end of the study, cover had declined and returned close to historical levels ( Fig. 2g,h). In contrast, mean cover of fine sediments increased markedly from < 2% in 1986/1987 to ~25% by the end of the study. Mean cover of other substrata, CCA and other biota varied annually but did not decline or decrease markedly over the course of the study (Fig. 2i-l).
Temporal changes in coral community structure. On shallow reefs, a total of 53 hard coral genera were recorded during the study period. Averaging across all years, just over half (51%) of the coral community (in terms of relative cover) was comprised of seven genera (Merulina 12%, Pectinia 11%, Montipora 7%, Pachyseris 7%, Porites 6%, Echinopora 4% and Platygyra 4%). Average cover of Acropora across all years was < 1%. The dominant coral growth forms, in terms of cover, were foliose (45%), massive (23%), sub-massive (13%) and encrusting (11%). Branching corals made up < 2% of coral cover on average. At deep sites, 54 coral genera were recorded, with 51% of the coral community dominated by five genera (Pectinia 19%, Pachyseris 14%, Merulina 6%, Porites 6% and Montipora 6%). Average relative cover of Acropora across all years was 0.25%. The coral growth forms that dominated the community were foliose (57%), encrusting (20%), massive (11%) and sub-massive (7%). Branching corals made up < 1% of coral cover on average. Coral community structure, when averaged across all sites, changed over time at both shallow and deep sites, but the pattern of change between shallow and deep sites differed (Fig. 3). At both shallow and deep sites, the genus Montipora showed the greatest decline in relative cover over time from ~14-2% and ~17-1% respectively, whereas the genus Pachyseris had the greatest relative increase from ~2-12% (in 2010) and ~8-22% respectively (Fig. 4). Relative cover of Pachyseris declined from ~12% to 6% at shallow sites following severe bleaching in 2010 (Fig. 4). Coral community structure at shallow sites changed markedly between 1988 and 2000, with a smaller shift occurring in 2011, and these correlated most strongly with reductions in cover of the genus Montipora (Fig. 4). Community structure continued to change after 2000 (correlated with increases in Goniastrea and Pachyseris) (Figs 3 and 4); however there was no return towards pre-1998 structure (Fig. 3), although community structure remained similar between 2004 and 2010 ( Fig. 3). At deep sites, change between 1986 and 2012 was driven by reductions in cover of Montipora, Platygyra and Merulina and increases in several genera, including Favites, Mycedium, Pachyseris, Dipsastraea (formerly Favia) and Echinophyllia. There was no evidence of a reverse in the direction of change towards historical structure (Fig. 3). Recovery of benthic and coral community structure at shallow sites before and after bleaching.
At three of the four selected sites (H2, HW1 & S2), benthic community structure differed significantly before and after the first bleaching event (1998; PERMANOVA, H2: pseudo-F 2,7 = 3.4, HW1: pseudo-F 2,5 = 4.4, S2: pseudo-F 2,5 = 3.0, p < 0.05 for all three sites) ( Supplementary Information, Table S3), but there were no significant differences before the first bleaching event in 1998 and after the second bleaching event in 2010. There were also no significant differences before and after the second bleaching event (PERMANOVA pairwise tests, p > 0.05) ( Supplementary Information, Table S3). Reductions in cover of hard coral and increases in cover of macroalagae and rubble made the greatest contribution to dissimilarity before and after the first bleaching event (SIMPER) ( Supplementary Information, Table S4) (Fig. 5). For coral taxonomic community structure there were significant changes at 2 of the 4 shallow sites (HW1 & R2) before and after bleaching events (PERMANOVA, HW1: pseudo-F 2,4 = 5.1, R2: pseudo-F 2,6 = 3.6, p < 0.001 for both sites) ( Supplementary Information, Table S5). Coral communities surveyed before the first bleaching event differed from those in-between bleaching events and those after the second bleaching, but there were no significant differences before and after the second bleaching event (PERMANOVA pairwise tests, Table S5).  Montipora (at R2) and Plerogyra (at HW1) made the greatest contribution to dissimilarity (in terms of decreases in relative cover) between time periods ( Supplementary Information, Table S6).

Discussion
Coral cover on Indo-Pacific reefs has declined dramatically in recent decades with some heavily disturbed reefs undergoing phase shifts from coral to algal dominance 5,6,16,54,55 . Reefs that have resisted phase shifts and returned to a coral dominated state have, in some instances, undergone major transitions in coral community structure 12,56,57 . Coral cover in Singapore also declined at both shallow and deep sites during the study period, with the largest declines occurring during the first decade of monitoring. Despite this, coral cover at shallow reef crest sites showed signs of rapid recovery (~16% increase in overall mean coral cover between 1998 and 2008), while coral cover at deep sites declined until 2003 and has still not recovered to historical levels. Despite decades of human disturbance, average coral cover at the shallow reef crest (~36%) is above the current average (~22-27%) for the Indo-Pacific 54,58,59 , although similar to estimated averages for reefs in the South China Sea region (~40%) 58 .
There was clear evidence of a temporal change in coral community structure, but communities were dominated by stress-tolerant and generalist taxa throughout the study period. Susceptible, competitive branching coral taxa were rare even at the onset of the study suggesting that they were never common or were extirpated before long term monitoring began. Furthermore, there was no evidence of phase shifts from coral to macroalgal dominance at any of the sites, despite the fact that adjacent reef flats (~0-2 m depth) have been dominated by erect fleshy macroalgae (e.g., Sargassum spp.) since at least the 1970's 60 . Cover of fine sediments increased markedly at deeper sites and was greater than that of hard corals by 2012, suggesting a shift from a coral-dominated regime to a coral/ fine sediment regime at the greater depths.
Much of the global decline on coral reefs has been attributed to coastal land use change, overfishing (leading to reduced herbivory), outbreaks of coral predators and climate change disturbances 1,2 . In Singapore we suggest that observed declines in coral cover at shallow and deep sites during the first decade of monitoring were due to a combination of increased sedimentation or turbidity and bleaching induced mortality 30,40,49 . At the time that monitoring started in the 1980s, Singapore's reefs were already highly turbid due to almost four decades of extensive coastal development 30 . Between 1988 and 1993, however, approximately 1.5 million m 3 of clay spoils were dumped on the eastern side of P. Semakau with no attempts made to contain their spread (Supplementary material Appendix 1, Fig. A1). This reportedly led to a thick layer of fine sediments being deposited on nearby reefs 61 . A parsimonious explanation for declines in coral cover prior to 1998 is that this event increased already high levels of sedimentation and turbidity well beyond the healthy range for most corals for a period of a few years 30,62 . Several additional major reclamation works have occurred since this event 63 , but they have been accompanied by extensive environmental impact assessments to identify necessary mitigation measures (e.g., use of sediment screens) 64 . To the best of our knowledge, the 1988-1993 event was the only "acute" sediment event during the study period; however, reefs were continually subjected to chronic sedimentation and turbidity throughout.
Further observed declines in coral cover at shallow sites during 1998 and 2010 coincided with two thermal bleaching events and were most likely a result of bleaching induced mortality 40 . Visual estimates in 1998 suggested 25% of coral colonies died due to bleaching (K. Tun, pers. comm) in Singapore. In contrast, surveys conducted during the 2010 bleaching event found ~60% of colonies moderately or severely bleached 40,49 , but estimated that only ~4% of colonies died due to bleaching. Other major acute disturbances present on reefs elsewhere, such as crown-of-thorns starfish or cyclonic storms, are absent in Singapore.
The marked contrast in recovery patterns of shallow reef crest sites versus deeper reef slope sites to these combined disturbances is remarkable, particularly when we consider that the reef crest and upper reef slope sites are separated spatially by only a few meters. One of the main drivers of rapid recovery of coral cover on reefs is recruitment of new corals to bare substratum 17 . Measured coral larval settlement rates are relatively low in Singapore 65 , therefore, this process is unlikely to have driven recovery of coral cover at shallow sites. In contrast, many common coral taxa in Singapore (e.g., Merulina) tend to suffer partial rather than whole colony mortality and are capable of recovering by rapid horizontal growth 49,66 . A plausible explanation therefore, is that the observed recovery at shallow sites following disturbance was driven primarily by regrowth of remnant colonies. Indeed this process is important in other locations following bleaching induced mortality 67,68 . Unfortunately we have no data on coral growth or larval settlement rates at deeper sites, making it difficult to determine why deep corals have failed to recover. Poor recovery may in part be due to low light levels as light attenuation is very rapid with < 1% of surface PAR reaching the lower reef slope (~9 m depth) resulting in corals being net heterotrophs below 4 m depth 32,69 . Furthermore, a high proportion of the benthos is now composed of unstable substrata (e.g., fine sediments, rubble) that would prevent coral larvae from settling or surviving post settlement 69 . The source of the fine sediments that replaced lost coral habitat at deeper sites has not been determined, but may result from low rates of resuspension typical of deeper reef slope habitats 70 .
Deeper reefs in other locations have been proposed as climate change refugia 51 , as corals there may escape bleaching during thermal anomalies due to cooler upwelling water or lower levels of light stress 51 . Clearly this is not the case for deeper reefs in the present study as coral cover declined markedly and was replaced by unstable substrata. Deep water refugia, therefore, may be confined to locations with sufficient light penetration at depth 51 or for species that can grow at relatively low levels of irradiance (e.g., Millepora intricata 71 ).
At shallow sites, relatively rapid recovery of coral community structure in the present study may be partly due to the dominance of stress-tolerant (e.g., massive Porites and Platygyra) or generalist coral taxa (e.g., Merulina, Pachyseris and Echinopora). Stress tolerant taxa are slow growing, have high fecundity, are long lived and relatively less susceptible to thermal stress, whereas competitive and weedy taxa tend to be faster growing, more structurally complex but less tolerant to disturbances 12 . Generalist taxa tend to have traits that overlap both life history groups (i.e., high stress tolerance and rapid growth) and these were some of the most dominant taxa in Singapore 12 .
Change in coral taxonomic community structure over the course of the study was strongly correlated with reductions in cover of Montipora (from ~15% to ~2% relative cover at both shallow and deep sites), a genus that contains competitive, stress tolerant and generalist species 72 . Currently, the most common Montipora species on Singapore's reefs are classified as generalists (e.g., M. danae and M. monasteriata) 42,72 . In the absence of species level identification, we cannot know whether this has always been the case or if competitive species (e.g., M. digitata) were more common prior to disturbances.
Most shallow coral communities in this region are dominated by Acropora, but this genus was not abundant at any time during the study. Thus there was no evidence of a shift from communities dominated by competitive taxa to dominance by stress-tolerant or generalist taxa, as has been found at other Indo-Pacific sites 21,57 . It is probable that tolerant taxa were present prior to the onset of major coastal development in the 1960s, i.e., reefs were "pre-adapted" to tolerate the reductions in water quality 73 . It is also possible that a community shift towards dominance by stress tolerant coral taxa occurred prior to the 1980s and the beginning of these surveys -as has been demonstrated from paleoecological reconstructions of other Indo-Pacific reefs subjected to anthropogenic disturbances decades before the start of ecological monitoring 74 .
There was no evidence of a phase shift from coral to macroalgal dominance at any of the sites. Cover of macroalgae was low throughout the study period and increased only marginally. EAM also remained relatively stable at both shallow and deep sites over the 27 years of study on Singapore's reefs. The average biomass of herbivorous fish in Singapore is estimated to be seven times lower (~4 g m −2 ) than the average for the broader Indo-Pacific (~29 g m −2 ) 59,75 . Densities of urchins are also low, although within the range found for Indo-Pacific reefs, but there is no relationship between macroalgal cover and urchin densities 59,75 . Combined, these data suggest that processes other than herbivory are responsible for maintenance of low macroalgal cover on the study reefs 59 . Our findings also support the contention that the EAM constitutes a stable regime on turbid reefs, possibly as a result of herbivory suppression under sediment-laden conditions 77 .
The presence of chronic human disturbances, the relatively high coral cover and rapid recovery following bleaching in 1998 suggest that disturbances may act antagonistically on Singapore's shallow reefs. For example, light attenuation in Singapore's turbid coastal waters results in an almost 50% reduction in photosynthetic efficiency of Sargassum at the reef crest 78 . If algal growth is limited by light even at quite shallow depths then relatively low rates of herbivory 59 may be sufficient to prevent macroalgae from overgrowing coral dominated areas. Similarly, while adaptation and acclimatisation following more severe thermal bleaching events (e.g. in 1998) have already been suggested as parsimonious explanations for the present level of thermal tolerance 40 , high levels of turbidity may also promote resistance to bleaching impacts by protecting colonies from stressful levels of irradiance during periods of thermal stress 4,22 . If so, these could provide examples of "ecological surprises", where negative disturbances (e.g., increased turbidity, thermal stress) interact antagonistically, resulting in a lower net impact 79 .
Our data support the notion that coral reefs will change rather than disappear entirely due to coastal land use changes 80 and provide a glimmer of hope that some heavily disturbed Indo-Pacific reefs can remain in a coral dominated state. Highly urbanised coral assemblages may be restricted to shallower areas and, if present, be dominated by stress-tolerant and generalist coral forms. Degradation does not always occur gradually and phase shifts can occur when a system has reached a tipping point 25 , therefore, the fact that the shallow reef communities in the present study have retained relatively high coral cover for almost three decades does not mean that they will remain this way indefinitely. Furthermore, despite high coral cover, we do not know if these altered shallow coral communities are providing any of the functions and ecological services normally associated with reef ecosystems (e.g., reef building, productive fisheries, diving tourism) 21 . If this is the case then future monitoring programs should consider incorporating other metrics such as net accretion, structural complexity and herbivory rates to assess reef health more broadly 81 .