USVI Public Records

Document text

LETTER doi:10.1038/nature14140 Predicting climate-driven regime shifts versus rebound potential in coral reefs Nicholas A. J. Graham1, Simon Jennings2,3, M. Aaron MacNeil1,4, David Mouillot1,5 & Shaun K. Wilson6,7 Climate-induced coral bleaching is among the greatest current threats tocoralreefs,causingwidespreadlossoflivecoralcover1.Conditions under which reefs bounce back from bleaching events or shift from coral to algal dominance are unknown, making it difficult to predict and plan for differing reef responses under climate change2. Here we document and predict long-term reef responses to a major climate- induced coral bleaching event that caused unprecedented region- wide mortality of Indo-Pacific corals. Following loss of .90% live coral cover, 12 of21reefs recovered towards pre-disturbancelive coral states,while ninereefs underwentregimeshiftstofleshymacroalgae. Functional diversity of associated reef fish communities shifted sub- stantiallyfollowingbleaching,returningtowardspre-disturbancestruc- ture on recovering reefs, while becoming progressively altered on regime shifting reefs. We identified threshold values for a range of factors that accurately predicted ecosystem response to the bleach- ing event.Recoverywas favoured when reefs werestructurallycomplex and in deeper water,when density of juvenile corals and herbivorous fisheswas relativelyhighand whennutrientloads werelow. Whether reefs were inside no-take marine reserves had no bearing on ecosys- tem trajectory. Although conditions governing regime shift or recov- erydynamicswere diverse,pre-disturbancequantification ofsimple factorssuchasstructuralcomplexity and water depth accurately pre- dicted ecosystem trajectories. These findings foreshadow the likely divergent but predictable outcomes for reef ecosystems in response toclimatechange,thusguidingimprovedmanagementandadaptation. Some mass bleaching events have resulted in the loss of almost all live coral within individual nations3. Examples of coral recovery follow- ing severe bleaching have been documented4,5, yet theory predicts that regime shifts to new benthic assemblages, such as macroalgae, are also likely2. To date, there are no documented coral reef regime shifts attrib- uted specifically to climate change. Evidence for coral reef regime shifts due to other causes comes almost exclusively from the Caribbean6,7, with limited knowledge from the more extensive and diverse Indo-Pacific reef province8. Ongoing uncertainty about Indo-Pacific coral reef res- ponses to climate impacts has generated considerable debate regarding appropriate management and adaptation plans, especially those related tono-takemarinereserves9,10,thatcanberesolvedthroughunderstanding siteandecosystemcharacteristicsthatdictatereefecosystemtrajectories. Using a 17-year data set, spanning a majorclimate-induced bleaching event, we assess thelong-term ecosystem dynamics of 21 reef sites across the inner islands of Seychelles. Seychelles reefs were the most severely affected globally by the 1998 coral bleaching event, in which a strong El Nin˜o coincided with the Indian Ocean dipole11. Across all sites before the bleaching event, average hard coral cover was 28% and macroalgal cover 1% (Fig. 1a), within average bounds for the Indian Ocean region at the time12. The mass bleaching event was severe across all reefs in the inner Seychelles, with coral cover reduced by .90%13. Both hard coral cover and macroalgal cover steadily increased between 2005 and 2011, with high heterogeneity reflecting markedly different trajectories among sites (Fig. 1a, Extended Data Figs 1 and 2). Using four complimentary metrics (Methods), we defined a recover- ing reef as having greater post-disturbance coral cover than macroalgae cover, with coral cover remaining high or increasing. A regime shifting reef was defined as having greater post-disturbance macroalgal cover than coral cover, with macroalgal cover remaining high or increasing. We used an index, based on the Euclidian distance from pre-bleaching benthic composition, to visualize the differing benthic trajectories of 1Australian Research Council Centre of Excellence for Coral Reef Studies, James Cook University, Townsville, Queensland 4811 Australia. 2Centre for Environment, Fisheries and Aquaculture Science, Pakefield Road, Lowestoft NR33 OHT, UK. 3School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UK. 4Australian Institute of Marine Science, PMB 3 Townsville MC, Townsville, Queensland 4810, Australia. 5ECOSYM, UMR CNRS-UM2 5119, Universite´ Montpellier 2, 34095 Montpellier Cedex, France. 6Department of Parks and Wildlife, Kensington, Perth, Western Australia 6151, Australia. 7School of Plant Biology, Oceans Institute, University of Western Australia, Crawley, Western Australia 6009, Australia. 1994 2005 2008 2011 0 20 40 60 80 Per cent cover Coral Macroalgae a 0 20 40 60 80 7 6 5 4 3 2 Per cent cover Distance from pre−disturbance b 0 20 40 60 80 7 6 5 4 3 2 Per cent cover Distance from pre−disturbance 2005 2008 2011 c Coral dominated Algal dominated R e gi m e s h if t R e c o v e r y Figure 1 | Recovery and regime shift dynamics on Seychelles coral reefs. a, Box (median and 50% quantile) and whisker (95% quantile) plots of coral cover and macroalgal cover in 1994 and through time after the 1998 bleaching event (n 5 84). b, Conceptual diagram for visualizing recovery and regime shift dynamics, using post-disturbance cover of coral (blue) and macroalgae (red) and Euclidian distance of benthic composition from pre-disturbance values. c, Recovery and regime shifts on Seychelles reefs. Lines are multinomial model fits for percent coral and macroalgae on each of the 21 reef sites (years distinguished by shading), with Euclidian distance from pre-disturbance composition as a predictor (n 5 63). Each reef site is represented by two points, one for coral cover (blue dots) and one for macroalgal cover (red dots). Twelve reefs (top of figure) are recovering, moving towards pre- disturbance (1994) compositions, with increasing coral cover and low macroalgal cover, whereas 9 reefs are regime shifting, with increasing distance from 1994 compositions, increasing macroalgal cover and low coral cover (Extended Data Table 1, Extended Data Fig. 4). 0 0 M O N T H 2 0 1 5 | V O L 0 0 0 | N A T U R E | 1 Macmillan Publishers Limited. All rights reserved ©2015 regime shifting and recovering reefs (Fig. 1b). Of the 21 reefs surveyed, 12 followed a post-bleaching recovery trajectory, with live coral cover increasing through time to an average cover of 23% by 2011, macro- algal cover remaining low (,1%), and benthic composition moving towards that observed in1994 (Fig.1c, Extended Data Table1, Extended Data Figs 3 and 4). Annual increases in coral cover on recovering reefs were low for 7–10 years following bleaching, but then increased sub- stantially, probably reflecting increasing local recruitment levels neces- sary for rapid recovery in isolated coral communities4. In contrast, 9 of the 21reefsfolloweda post-bleaching regimeshift trajectory, withaver- age macroalgal cover steadily increasing to 42% by 2011, coral cover remaininglow (,3%), and benthic composition diverging substantially from the pre-bleaching state (Fig. 1c, Extended Data Table 1, Extended Data Figs 3 and 4). Percent cover of coral and macroalgae did not differ betweenthese twogroupsofreefs in1994(Extended Data Table 1), pro- vidingthefirstclearevidencethatregimeshiftstomacroalgae-dominated states occur in response to major coral bleaching events on Indo-Pacific reefs. Divergent ecosystem trajectories following bleaching are likely to have major implications for reef associated organisms. The functional structure of communities, which captures species’ roles based on their biologicaltraits,is predicted to showdeterministic links todisturbance14, but predictions are rarely tested with data. Here we show changes to functional structure of reef fish assemblages are stronglytied to thepost- bleaching benthic response (Fig. 2, Methods). Seven years after bleach- ing, assemblages on all reefs had fewer small bodied species and a less complexfunctionalstructure,typifiedbyincreasingdominanceofinver- tebrate feeding fishes. On recovering reefs, a smaller initial change in functional structurewas followed bya return towards the pre-bleaching state (Fig.2b), whereas themuch greater change on reefs that underwent regime shifts progressed in post-bleaching years, suggesting important functions are being eroded (Fig. 2c). We assessed the association between 11 reef-level factors and eco- system trajectories post-disturbance, a priori selected for their generic importance in structuring coral reef ecosystems globally(Extended Data Table 2). Five of the factors correctly (97% of the time) characterized post-bleaching reef trajectories as ‘recovery’ or ‘regime shift’: density of juvenilecorals,initialstructuralcomplexity ofthe reef,waterdepth,bio- mass of herbivorous fishes (also reflecting herbivore diversity, which was collinear), and nutrient conditions of the reef (Fig. 3). Estimates of critical values for each factor, where recovery was more likely to occur thanaregimeshift,showedthatjuvenilecoraldensities.6.2perm2reduce the probability of a regime shift (Fig. 3a), consistent with the expected roleofcoralrecruitmentandsurvivalinreefrecovery4.Corallarvalsupply does not differ among recovering and regime shifting reefs in Seychelles, but post-settlement survival is lower on regime shifted reefs15, likely due to unsuitable settlement substrate, or enhanced post-settlement mortality due to macroalgal overgrowth or alleopathic mechanisms2,16. Structural complexity, which captures the structure provided by corals and the underlying reef matrix, decreased the risk of a regime shift tra- jectory when values before disturbance were .3.1 (widespread mod- eratelycomplexrelief)(Fig.3b).Structuralcomplexityinfluences arange of ecological processes, being a substantial contributor to diversity and productivity of many reef associated organisms17,18. Sites deeper than 6.6 m depth were less likely to undergo a regime shift (Fig. 3c), possibly reflecting the relationship between light penetration and algal growth19, orgreatervulnerabilityof shallowerreefsto disturbancessuchasrecur- rent coral bleaching or storm damage20. Experimental and modelling studies will be necessary to clarify the principal mechanisms by which structural complexity and depth influence ecosystem trajectories. Reef fish herbivory is a key process mediating competition between corals and algae21, and herbivore biomass,implicitlylinked to bodysize, is a good proxy for overall rates of herbivory22. Interestingly, a relatively low biomass of herbivores (177kg ha21)—below average values for the Indian Ocean23—reduced the risk of a regime shift occurring (Fig. 3d). Algalproliferation and dominance overcoralsisalsoinfluencedbynutri- ent input to reef systems24. Here, lower C:N ratios in macroalgal fronds, indicative of higher nutrient loads25, were linked to a greater likelihood of regime shifts, with the likelihood of regime shifts reducing below 50% when ratios passed 38 (Fig. 3e). Previous debate has focused on whether levels of herbivory or nutrients mediate coral reef regime shifts or recov- ery26,27. Our results suggest that although both variables relate to ecosys- temtrajectory,theyareweakerandlesscertainpredictorsthanstructural complexity, depth, and the density of juvenile corals. Reefs within no-take marine reserves were no more likely to recover than reefs in fished areas. Althoughmarine reserves may have a positive long-term influence on coral cover in the absence of disturbance28, our results suggest they may have little influence on post-disturbance ben- thic trajectories. In Seychelles, marine reserves do enhance herbivore biomass (2005 values: 279 kgha216 21.5 s.e.) compared to fished areas (mean163 kg ha216 58.6 s.e.),but biomass in fished areas was still close to the threshold we identified for recovery. Marine reserves may have a greater role in aiding coral recovery in nations where herbivorous fishes are more heavily exploited or where fishing gears that reduce structural complexity are used. Collectingdata on nutrientsand juvenile coral densities can bechal- lenging for the often resource-limited agencies charged with monitor- ing and managing coral reef ecosystems, especiallyover large areas.We therefore assessed the predictive power of structural complexity and waterdepthalonetopredictreeftrajectories.Bothvariablescanberapidly recorded over large areas and are relatively stable through time, allowing extensive pre-disturbance data to be compiled. Using only these two –0.2 –0.1 0 0.1 0.2 1994 2005 2008 2011 b c –0.2 –0.1 0 0.1 0.2 –0.2 –0.1 0 0.1 0.2 PCoA axis 1 PCoA axis 1 –0.2 –0.1 0 0.1 0.2 PCoA axis 1 PCoA axis 2 a 3 5 4 1 6 2 Figure 2 | Trajectories in the functional structure of fish assemblages. a, Functional trait space underlying the analyses, with arrows depicting increasing body size amongst each fish functional group, depicted by fish outlines: corallivores (1), grazing herbivores (2), invertebrate feeders (3), planktivores (4), piscivores (5), scraping and excavating herbivores (6). b, c, Position of recovered (b) and regime shifted (c) reef sites in the functional space. Black dots represent the mean trait values in 1994, pale to darker colours represent 2005–2011 (n 5 84). Large rings indicate centroid for sites within each year (2005 and 2008 overlap in b). Dashes on the axes indicate centroid positions for each year. Year and ecosystem trajectory were significant factors (Methods, ANOVA model: axis 1 year P , 0.001, trajectory P , 0.01; axis 2 year P , 0.001, trajectory P , 0.05). Fish graphics by T. Saxby, D. Kleine and J. Woerner (Integration and Application Network, University of Maryland Center for Environmental Science, http://ian.umces.edu/imagelibrary/). RESEARCH LETTER 2 | N A T U R E | V O L 0 0 0 | 0 0 M O N T H 2 0 1 5 Macmillan Publishers Limited. All rights reserved ©2015 variables, we correctly predicted reef trajectory 98% of the time. This suggests that, when combined, structural complexity and depth can effectively identify reefs with a high likelihood of recovery or regime shift, thereby informing reef risk assessment and guiding marine spa- tial planning initiatives (Fig. 4). We have demonstrated persistent and divergent responses of coral reefstoclimate-inducedbleaching withpotentialconsequencesforreef fish functioning.With predictionsthatmasscoral bleachingeventswill increase in frequency1, our findings foreshadow the likely divergent tra- jectoriesexpectedonotherIndo-Pacificreefs.Weshowthatseveralfactors can affect reef ecosystem trajectories following bleaching but, where necessary, just depth and structural complexity may be useful predic- tors of ecosystem fate. The predictor factors we identify are important for coral reefs globally (Methods) and the range of values reported for SeychellesisconsistentwithotherlocationsintheIndo-Pacific(Extended Data Table 3). Further, depth and initial structural complexity evaluated across 6 other countries from East Africa to the South Pacific were con- sistentpredictorsofcoralormacroalgaldominatedreefspost-disturbance, witheffectsizesoverlappingthosefromSeychelles(ExtendedDataFig.5). However, there may be regional variations in the relative contribution of factors and threshold values we identified that require more invest- igation. Uncovering the predictors that dictate reef trajectories following major bleaching events can inform ecosystem management and strat- egies for human adaptation. Mapping probable ecosystem trajectories can enable limited management resources to target actions, such as phas- ingoutfishinggearsthatcausehabitatdamage29,onreefswhereincreased structural complexity could help promote recovery29. Spatial under- standing of ecosystem vulnerability to climatic impacts also holds great potential to be linked with social vulnerability assessments to develop thehumanadaptationstrategiesneededtocopewithanticipatedregime shifts and associated changes in ecosystem services30. Online Content Methods, along with any additional Extended Data display items andSource Data, are available in the onlineversion of the paper; references unique to these sections appear only in the online paper. Received 26 August; accepted 3 December 2014. Published online 14 January 2015. 1. Hoegh-Guldberg, O. et al. Coral reefs under rapid climate change and ocean acidification. Science 318, 1737–1742 (2007). 2. Hughes, T. P., Graham, N. A. J., Jackson, J. B. C., Mumby, P. J. & Steneck, R. S. Rising to the challenge of sustaining coral reef resilience. Trends Ecol. Evol. 25, 633–642 (2010). 3. Goreau, T., McClanahan, T., Hayes, R. & Strong, A. Conservation of coral reefs after the 1998 global bleaching event. Conserv. Biol. 14, 5–15 (2000). 4. Gilmour, J. P., Smith, L. D., Heyward, A. J., Baird, A. H. & Pratchett, M. S. Recovery of an isolated coral reef system following severe disturbance. Science 340, 69–71 (2013). 5. Sheppard, C. R. C., Harris, A. & Sheppard, A. L. S. Archipelago-wide coral recovery patterns since 1998 in the Chagos Archipelago, central Indian Ocean. Mar. Ecol. Prog. Ser. 362, 109–117 (2008). 6. Hughes, T. P. Catastrophes, phase shifts, and large-scale degradation of a Caribbean coral reef. Science 265, 1547–1551 (1994). Juvenile coral density 1 3 7 20 55 0 0.2 0.4 0.6 0.8 1.0 a 2.6 2.8 3.0 3.2 3.4 3.6 0 0.2 0.4 0.6 0.8 1.0 Initial structural complexity b 3 4 5 6 7 8 9 10 0 0.2 0.4 0.6 0.8 1.0 Depth (m) c 50 100 500 0 0.2 0.4 0.6 0.8 1.0 d Probability of regime shift 25 30 35 40 45 50 0 0.2 0.4 0.6 0.8 1.0 Carbon:nitrogen Herbivore biomass (kg ha–1) Standardized effect size e Bayesian HPD estimates –10 –6 –2 Carbon:nitrogen loge (herb. biomass) loge (juvenile corals) Initial complexity Depth Intercept f Figure 3 | Bayesian hierarchical logistic regression of predictor variables for the probability of a regime shift. a–e, Marginal plots for predictor variables included in the full model: solid black line represents the mean model fit; grey lines are random resamples of the parameter estimates, representing variability of the expected model fit. Grey dots represent replicates from regime shifted (1) or recovering (0) reefs (n 5 184). The blue dotrepresents the point at which regime shifts and recovery are equally likely. Blue shading is beyond upper 95% uncertainty interval for P (shift)5 0.5, where regime shifts are decreasingly likely. f. Effect size medians (black dot), with 50% (thick line) and 95% (thin line) highest posterior density (HPD) uncertainty intervals. 3 4 5 6 7 8 9 Depth (m) Initial complexity 2.6 2.8 3.0 3.2 3.4 3.6 0.1 0.4 0.2 0.6 0.8 0.9 10 Recovery Regime shift Figure 4 | Contour biplot of the probability of regime shift (red shading) or recovery (blue shading) based on initial structural complexity and water depth. For example, for a complexity value of 3, there is a 0.9 probability of a regime shift occurring for sites ,4.5 m depth (red dot), and only sites deeper than ,8 m are more likely to recover. For a complexity value of 3.2, sites deeper than 6.3m are highly likely to recover (blue dot). LETTER RESEARCH 0 0 M O N T H 2 0 1 5 | V O L 0 0 0 | N A T U R E | 3 Macmillan Publishers Limited. All rights reserved ©2015 7. Mumby, P. J. et al. Fishing, trophic cascades, and the process of grazing on coral reefs. Science 311, 98–101 (2006). 8. Roff, G. & Mumby, P. J. Global disparity in the resilience of coral reefs. Trends Ecol. Evol. 27, 404–413 (2012). 9. McCook, L. J. et al. Adaptive management of the Great Barrier Reef: A globally significant demonstration of the benefits of networks of marine reserves. Proc. Natl Acad. Sci. USA 107, 18278–18285 (2010). 10. Mora, C.& Sale, P. F. Ongoing global biodiversity loss and the need to movebeyond protected areas: a review of the technical and practical shortcomings of protected areas on land and sea. Mar. Ecol. Prog. Ser. 434, 251–266 (2011). 11. Saji, N. H., Goswami, B. N., Vinayachandran, P. N. & Yamagata, T. A dipole mode in the tropical Indian Ocean. Nature 401, 360–363 (1999). 12. Ateweberhan, M., McClanahan, T. R., Graham, N. A. J. & Sheppard, C. R. C. Episodic heterogeneous decline and recovery of coral cover in the Indian Ocean. Coral Reefs 30, 739–752 (2011). 13. Graham, N. A. J. et al. Dynamic fragility of oceanic coral reef ecosystems. Proc. Natl Acad. Sci. USA 103, 8425–8429 (2006). 14. Mouillot, D., Graham, N. A. J., Ville´ger, S., Mason, N. W. H. & Bellwood, D. R. A functional approach reveals community responses to disturbances. Trends Ecol. Evol. 28, 167–177 (2013). 15. Chong-Seng, K. M., Graham, N. A. J. & Pratchett, M. S. Bottlenecks to coral recovery in the Seychelles. Coral Reefs 33, 449–461 (2014). 16. Smith, J. E. et al. Indirect effects of algae on coral: algae-mediated, microbe- induced coral mortality. Ecol. Lett. 9, 835–845 (2006). 17. Graham, N. A. J. & Nash, K. L. The importance of structural complexity in coral reef ecosystems. Coral Reefs 32, 315–326 (2013). 18. Rogers, A., Blanchard, J. L. & Mumby, P. J. Vulnerability of coral reef fisheries to a loss of structural complexity. Curr. Biol. 24, 1000–1005 (2014). 19. McCook, L. J. Macroalgae, nutrients and phase shifts on coral reefs: scientific issues and management consequences for the Great Barrier Reef. Coral Reefs 18, 357–367 (1999). 20. Bridge, T. C. L., Hughes, T. P., Guinotte, J. M. & Bongaerts, P. Call to protect all coral reefs. Nature Climate Change 3, 528–530 (2013). 21. Bellwood, D. R., Hughes, T. P., Folke, C. & Nystrom, M. Confronting the coral reef crisis. Nature 429, 827–833 (2004). 22. Lokrantz, J., Nystro¨m, M., Thyresson, M. & Johansson, C. The non-linear relationship between body size and function in parrotfishes. Coral Reefs 27, 967–974 (2008). 23. McClanahan, T. R. et al. Critical thresholds and tangible targets for ecosystem- based management of coral reef fisheries. Proc. Natl Acad. Sci. USA 108, 17230–17233 (2011). 24. Lapointe, B. E. Nutrient thresholds for bottom-up control of macroalgal blooms on coral reefs in Jamaica and southeast Florida. Limnol. Oceanogr. 42, 1119–1131 (1997). 25. Atkinson, M. J. & Smith, S. V. C:N:P ratios of benthic marine plants. Limnol. Oceanogr. 26, 1074–1083 (1983). 26. Smith, J. E., Smith, C. M. & Hunter, C. L. An experimental analysis of the effects of herbivory and nutrient enrichment on benthic community dynamics on a Hawaiian reef. Coral Reefs 19, 332–342 (2001). 27. Burkepile, D. E. et al. Chemically mediated competition between microbes and animals: microbes as consumers in food webs. Ecology 87, 2821–2831 (2006). 28. Selig, E. R. & Bruno, J. F. A global analysis of the effectiveness of marine protected areas in preventing coral loss. PLoS ONE 5, e9278 (2010). 29. Cinner, J. E. et al. Gear-based fisheries management as a potential adaptive response to climate change and coral mortality. J. Appl. Ecol. 46, 724–732 (2009). 30. Adger, W. N., Hughes, T. P., Folke, C., Carpenter, S. R. & Rockstro¨m, J. Social-ecological resilience to coastal disasters. Science 309, 1036–1039 (2005). Acknowledgements This research was principally supported by the Australian Research Council (DP1094932, DE130101705), the Leverhulme Trust (F/00 125/M), and the Western Indian Ocean Marine Science Association. The Natural Environment Research Council (GR3/1154) funded work in Fiji. We thank the Seychelles Fishing Authority, Seychelles Marine Parks Authority, Nature Seychelles, and Seychelles National Meteorological Services for technical andlogistical assistance. Many thanks to N. Polunin for support early in the project, to N. Cariglia for collecting the sea urchin data, to K. Chong-Seng for collecting the juvenile coral data, to C. Huchery for helping develop the wave exposure model, to J. Turner for photos a and b in Extended Data Fig. 1, and T. McClanahan and N. Dulvy for sharing data used in Extended Data Table 3 and Extended Data Fig. 5. J. Cinner, C. Hicks, K. Nash, and three anonymous referees provided useful comments on the manuscript. Author Contributions N.A.J.G. conceived of the study with S.K.W. and M.A.M.; N.A.J.G. S.J., and S.K.W. collected the data; N.A.J.G., M.A.M., and D.M. developed and implemented the analyses; N.A.J.G. led the manuscript with S.J., M.A.M., D.M., and S.K.W. Author Information Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. Readers are welcome to comment on the online version of the paper. Correspondence and requests for materials should be addressed to N.A.J.G. (nick.graham@jcu.edu.au). RESEARCH LETTER 4 | N A T U R E | V O L 0 0 0 | 0 0 M O N T H 2 0 1 5 Macmillan Publishers Limited. All rights reserved ©2015 METHODS Field Surveys and indices. TwentyonereefsthroughouttheinnerSeychellesIslands were surveyed using identical methods in 1994, before the mass coral bleaching of 1998, and in 2005, 2008 and 2011, following the bleaching event. The surveys incorporated three different reef habitat types; carbonate fringing reefs; granitic rockyreefswithcoralgrowth; andpatchreefhabitats onsand,rubble, orrockbase. Eight to sixteen replicate 7 m radius point counts were surveyed along the reef slope on each reef, covering up to 0.5 km of reef front and 2,500m2 of reef habitat. The underlying substrate type and water depth were recorded at each site. Within each point count area, the percent cover of different growth forms of live hard coral, soft coral, macroalgae, sand, rubble, and rock was quantified, and the structural complex- ity of the reef was visually estimated on a 6 point scale13. This structural complexity measure captures landscape complexity, including the complexity provided by live corals, that of the underlying reef matrix and other geological features, and has been shown to correlate well to other measures of complexity, such as measures of reef height and the linear versus contour chain method31. The complexity of each individualcountareawasassignedtooneofthefollowingcategories:0 5novertical relief, flat or rubbly areas; 1 5 low (,30 cm high) and sparse relief; 2 5 low but widespread relief; 3 5 widespread moderately complex (30–60cm high) relief; 45 widespread very complex (60–100 cm high) relief with numerous fissures and caves; 5 5exceptionally complex (.1m high) relief with numerous caves and over- hangs. Estimates canbe standardised among observers with minimal training, thereby providing reliable and comparable data of ecological relevance31. The density and individual sizes of 134 species of diurnally active, non-cryptic, reef associated fish were recorded within each point count area. Length estimation was calibrated at the beginning of each day by estimating the length of known lengths of PVC pipe. Accuracy was within4% of actuallengths32. Weconverted data onfishcountstobio- mass with publishedlength–weight relationships33,34. Fish were assigned to feeding groupsbasedontheirdominantdietsandfeedingbehaviour35.Thedensityofjuvenile corals (,10cm diameter) was quantified in 2011, within 8 replicate 33cm by 33cm randomly placed quadrats in each point count area. Juvenile coral data were also collected in 2008, and correlated strongly across sites with 2011 data (r 5 0.85), suggesting that patterns in juvenile coral densities are fairly consistent through the post-disturbance years surveyed. Sea urchin density (family: Diadematidae) was counted in 2008 in each point count area. Meta-data from these surveys will be deposited in the James Cook Universityresearchdata repository,the Tropical Data Hub (https://eresearch.jcu.edu.au/tdh). To assess local nutrient levels ten fronds from each of ten different Sargassum thalli were taken from each reef site in 2014. These samples were dried for 48 h at 60 uC, powdered and analysed for nitrogen and carbon abundance. Carbon and nitrogen weight percent (%C and %N, respectively), and C:N ratios were deter- mined using a Costech Elemental Analyzer fitted with a zero-blank auto-sampler atJamesCookUniversity’sAdvancedAnalyticalCentre,Cairns.Theratioofcarbon to nitrogen in macroalgae is a stable, longer-term indicator of ambient nutrient regimes at a given location and should reflectstable spatial patterns among sites25,36. Specifically, nutrient inputs incorporated into algal tissue are averaged over the activegrowingperiod(3–6 monthsbeforeourcollectioninthisregion37),reflecting nitrogen availability during this time38. This time period spans the main rainy season in Seychelles (December to February), so the nitrogen in these algal tissues should reflect land-based nutrient inputs. Although temporal variability in nutri- ent input is likely, due to differences in rainfall among years, the spatial patterns associated with adjacent land-based input among sites (that is, the relative differ- encesinnutrientregimesweareinterestedinhere)arethoughttobestablethrough time38,39. Wave exposure was calculated based on the uninterrupted (by land or other reefs) distance winds can blow over the ocean to generate waves (fetch), coupled with data on wind speed and direction40,41. Larger waves develop with greater fetch and stronger winds. We added reef crest polygons to a base map of land mass42, and rasterised the map at a spatial resolution of 55 m. Fetch values for each of our 21reefswascalculatedin32compassdirections(eachwithanangularwidthof 11.25) using the Waves Toolbox for ArcGIS 10.2 (ref. 43). We restricted fetch calculations to a maximum distance of 500km around the Seychelles islands, reflecting wind generated wave energy for this isolated archipelago. We used hourly readings of both wind speed and direction from the Seychelles National Meteorological Service for every day of the post-bleaching period (1998 to 2011). Wave energy (in Joules) was calculated as a function of fetch, wind speed and direction40,41. We calculated average wave exposure for each reef based on hourly wave exposure estimates for theentirepost-bleachingperiod,capturinginformationonstrongandsporadicwinds. Benthic trajectory. Changesinbenthiccompositionamong sitesandthroughtime were examined using correlation-based principle components analysis, based on Euclidian distance. Data were log(x11) transformed to account for some right skew- nessdetectedindraftsman’splotsandnormalizedtoensureallmetricsintheanalysis were on a common scale. Eigenvectors were overlaid to identify direction and contribution of the different variables to the patterns. We defined a regime shifting reef as one where post-disturbance macroalgal cover becomes greater than coral cover and trajectories through time indicate that cover of macroalgae remains high or is increasing. Recovering reefs, conversely, aredefined as those reefs where post- disturbance coral cover becomes greater than macroalgal cover and trajectories through time indicate that cover of hard coral remains high or is increasing. To determine whether sites were regime shifting or recovering we used four metrics, the first based on a static cover estimate, while the other three reflect cover traject- ories. These four metrics provide a comprehensive assessment of site status and trajectory. Sites must conform to metric 1 and at least one of the trajectory metrics (2–4) to be classified as either recovering or regime shifting. (Metric1) Percent cover ofcoraland macroalgaeatlastdata point (2011). Ifcoral cover is higher than macroalgal cover, and greater than the first post-disturbance survey (2005), the site is classified as recovering. If macroalgal cover is higher than coral cover and greater than pre-disturbance macroalgal cover (1994), the site is classified as regime shifting (Extended Data Table 1). (Metric 2) The Euclidian distance between the pre-disturbance (1994) benthic composition, andthat of2005, 2008and2011wascalculatedatasitelevel,toquantify if this distance was increasing or decreasing through time. A substantial change from 1994 and increasing distance through post-disturbance years indicates a regime shifting site, whereas declining distance through post-disturbance years is indi- cative of recovery (Fig. 1b, c, Extended Data Fig. 3). (Metric 3) The rate of change in coral or macroalgal cover post-disturbance (2005–2011) indicates the trajectory of the site in terms of increasing coral versus macroalgae. If rate of coral cover increase remains stable or increases faster than rate of macroalgal cover the site is classified as recovering,andvice versaforregime shifting (Extended Data Table 1). (Metric 4) Change in cover (coral and macroalgae) between 1994 and 2005, and 1994 and 2011 was calculated. If the net decline in coral cover becomes smaller between these two time periods, and change in macroalgae is negligible and static, the site is classified as recovering. If the netdeclinein coral cover remains large and static, and increases in macroalgal cover are becoming greater, the site is classified as regime shifting (Extended Data Fig. 4). To visualize the regime shift and recovery dynamics, we quantified the site level relationships between distance from the pre-disturbance benthic composition and proportions of coral and macroalgae using a multinomial model of benthic com- position (j),whereby percentages ofcoral,macroalgae,andothersubstrates (1003 pij) at a given site (i) were predicted based on Euclidian distances (ED) for a given year: pij~ eb0jzb1jED PJ k~1 (eb0kzb1kED) ð1Þ Parameters include an intercept(b0) and slope (b1) for ED in each of the habitat categories (j).This is a standard multinomial logit model, in this case run using the nnet package in R (http://www.R-project.org). To illustrate the progression of reef states, we ran the same model (1) for each post-disturbance survey (2005, 2008, and 2001) and represented uncertainty in model fit by sampling values from the maximumlikelihoodandestimatedstandarddeviationofeachparameterin(1)and plotting the resultingmodelfits(Fig.1cand ExtendedDataFig. 3). To betterconvey the progressionofrecovering andregimeshiftingreefsthroughtime werotated the ED predictor onto the y axis. Fish functional structure. The functional structure of fish assemblages was cal- culated based ontwodominant traits that capture a large proportion ofthe implicit functional roles playedby reef fishes: dietary group andbody size. We classified the fish into well-established feeding groups (corallivores, invertivores, planktivores, grazing herbivores, scraping and excavating herbivores, and piscivores) based on the literature35 and Fishbase34. These groupings cover some of the main feeding func- tions performed by fishes on coral reefs, including mediation of coral:algae inter- actions, removal of sediments, and trophic control through predation. Body size is a trait that captures information related to the feeding ecology, energetic demands, and movement patterns in reef fish. For example, there are well established non- linearrelationships between body size and area of reef grazedby parrotfishes22, while predators feed on prey in accordance to their gape size7. Furthermore, home range size and functional range size in reef fish is tightly correlated to body size, with larger fish feeding over a greater area44,45. Body size was coded into four categories: ,20 cm, 20–40 cm, 41–60 cm and .60 cm. We focused on these two biological traits to specifically target functional traits46. Based on these two functional traits, we performed a principal coordinate ana- lysis (PCoA) on a Gower distance matrix between species pairs to provide two inde- pendent synthetic axes that summarize species distribution within a trait functional space47,48. ThesetwoindependentfunctionalaxesfromPCoA,incombinationwith the species abundance matrix for all reefs in each year of sampling, were used to measure functional structure through biomass-weighted mean values for each LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015 community. We used these biomass-weighted mean values for each PCoA axis to assesshowthefunctionalstructureofa givenassemblagechangedthrough timefor both recovering and regime shifting reefs. Year and trajectory (regime shift versus recovery)hadasignificantinfluenceonthefunctionalstructureoffishassemblages on both PCoA axes (ANOVA model: axis 1 year P , 0.001,trajectory P ,0.01; axis 2 year P , 0.001, trajectory P , 0.05), whereas management (marine reserve ver- sus fished) had no significant influence (ANOVA model: management axis 1 P 5 0.533, axis 2 P 5 0.106). Predicting recovery versus regime shift dynamics. We assessed the ability of eleven different site level factors (habitat type, initial branching coral cover, juvenile coral density, depth, herbivore biomass, herbivore diversity, marine reserve status, nutrient regime, initial structural complexity, sea urchin density, wave exposure) to predict whether a site went on a recovery or regimeshift trajectory. Each factor was a priori chosen based on a pre-existing rationale from the literature, to specifically targetfactorsknown torepresent key processes orto structure coral reef ecosystems globally (Extended Data Table 2). Indeed, the generic importance of the selected variablesoncoralreefsishighlightedbyspecificreviewpapers,forexample:branch- ing corals49; juvenile coral survivorship2; depth20; marine reserve status50; herbivor- ous fish biomass and diversity21; nutrientregime51; structuralcomplexity17; and sea urchins2. Furthermore, the range of values found on Seychelles are consistent with values from other reef locations in the Indo-Pacific (Extended Data Table 3), indi- cating that the relationships we find should be generic, although we might expect location-specific variation in thresholds. We could not include some factors, such asturbidityandwatercurrents(typicallyweakinSeychelles),howeverwewereable toaccurately predictecosystemtrajectory with the covariates includedinthe model. Although some of the factors are not influenced over the time frames investi- gated here(forexample, depth),or wereconstructedacross time (for example,wave exposure), we had to make decisions on which time period to use for other factors. Branching coral cover,important for a rangeofreeffishes andother organisms49, is particularly vulnerable to coral bleaching52, so reefs with a high branching coral coverpre-disturbancemaybeexpected toundergoextensivedegradation,andthus we used data from 1994. We used pre-disturbance data for structural complexity because it can be maintained for many years post disturbance, ensuring the con- tinuation of reef processes13. For both herbivorous fish biomass and diversity, we used data from 2005 to first account for any changes in these variables associated withthedisturbanceeventitself53,andsecondtoreflecttheamountof post-disturbance herbivory available to influence subsequent benthic responses. We were interested inpost-settlementsurvivalbasedonthejuvenilecoraldensityvariable,souseddata from 2011 to allow more (often slow growing) corals to make it through early life history stages. Sea urchin density (family: Diadematidae), would ideally be taken from 2005 to reflect herbivory available for post-disturbance benthic responses, however the first year of available data was 2008, which correlated well (coefficient 0.57) to subsequent years, suggesting spatial consistency through time. As noted above, spatial differences in our nutrient regime data should reflect long-term pat- terns. Collinearity among these predictor covariates was assessed (Extended Data Fig. 6), and herbivore diversity was excluded from the analysis due to collinearity with herbivore biomass. The spatial distribution of sites following a regime shift or recovery trajectory was to some extent geographically clustered, but some sites in close proximity fol- lowed different trajectories (Extended Data Fig. 7). We used a Bayesian hierarchical logistic regression model to assess which site-scale variables were best at predicting thetrajectoryofthesitespostbleaching(05recovered,15regimeshifted).Discarding variables with posterior densities centred on zero (suggesting no effect), our full model for regime shifts among Seychelles sites was: yij *Bern pij   (2) logit pij   *N mj,s   (3) mj ~b0zb1LHBjzb2JCDjzb3DEPjzb4ISTjzb5CNRj (4) b0,:::,5 *N 0:0,1000 ð Þ (5) s *U 0,100 ð Þ (6) which included log(herbivore biomass 1 1) (LHB; kg ha21); juvenile coral density (JCD; m2); depth (DEP; m); initial structural complexity (IST; 0–5 scale); and the carbon nitrogen ratio of sampled algae (CNR). Covariates were then standardized by subtracting their mean and dividing by 2 of their standard deviations to gauge theirrelative importance asthe relativemagnitudeoftheireffectsizes.Modelswere run using the PyMC package54 for the Python programming language (http://www. python.org). For each of these variables, we calculated marginal 95% uncertainty intervals (UI) around the point at which the probability of a regime shift was half (P(shift) 50.5). The lower 95%UI was taken as the point at which a regime shift is more likely than not, with all other variables held at their average, while the upper 95%UI was taken as the point at which recovery is more likely to occur. The values within these uncertainty bounds reflect the unstable area where recovery or regime shift out- comes are equally likely. In order to assess the predictive ability of the full model, we ran a leave-one-out cross validation (Loo-CV) analysis, a permutation procedure whereby the ability of the model to successfully predict the ecosystem trajectory (recovery or regime shift)ofanygivensiteisrepeatedlytestedwithdifferentsites.Todothiswerandomly selected one site and predictedits recovery or regime shift status using a modelbased on the other 20 sites, repeating the procedure 999 times. Wetestedthepredictivepowerofthemodelifreducedtoincludeonlytwometrics (initial complexity of the reef and water depth), both of which can be easily quan- tified in surveys and collected on reefs before major disturbance events, making them very strong candidates for management planning and informing develop- ment of adaptation strategies. This gave a reduced parameter model: yij *Bern pij   (7) logit pij   *N mj,s   (8) mj ~b0zb3DEPjzb4ISTj (9) We re-ran the Loo-CV analysis with this subset of predictor variables to assess the predictivepower thatwas lostby removing othervariables from the analysis. Despite the substantial increase in deviance information criterion (DIC) scores between models (Mfull DIC5129; Msubset DIC5145), the subset model performed equally well in correctly predicting the trajectory of our observed reefs. We sampled from the posteriors of the two variables from the subset model across their observed range to model probability surfaces of regime shift risk when these two metrics are both quantified, presented as a contour plot in Fig. 4. To assess whether these two simple predictor variables (depth and initial struc- tural complexity) are likely to be useful predictors of post-disturbance reef con- dition elsewhere in the Indo-Pacific, we compiled data from 6 countries (Australia (Great Barrier Reef), British Indian Ocean Territory (Chagos), Fiji,Kenya, Maldives, Tanzania)55–58 where before disturbance site depth and structural complexity, and post-disturbance (average 8.4 years 60.4 post-disturbance) coral and macroalgal cover were recorded. The disturbances were mostlyclimate induced bleaching, but incombinationwithcrown-of-thornsstarfish (COTS)inFiji andCOTS andstorms in Australia. For depth we used data from all 51 reef sites. However, because struc- tural complexity can vary based on reef zone (for example, the reef slope versus the reef flat), or geomorphology (for example, atolls versus coastal fringing reefs), we used a paired design for assessing the role of structural complexity where a site that had become dominated by macroalgae was paired with a nearby similar site (for example, same zone, geomorphology, and depth) that had recovered to high coral cover. Thisresultedin7 pairs(n 5 14), from4 ofthe countries.We ranthe Bayesian hierarchical logistic regression model used for the Seychelles data, with reef sites assigned as having either higher macroalgal cover (1) or coral cover (0) post dis- turbance. The Bayesian effect size posterior density distributions for these wider Indo-Pacific data were compared to those for the Seychelles reefs (Extended Data Fig. 5). Considerable overlap in both the depth (Seychelles: 24.77 [26.92, 22.62]; Indo-Pacific: 21.67 [23.74, 0.15]; median and 95% highest posterior density uncertainty intervals) and complexity (Seychelles: 25.66 [29.13, 22.69]; Indo- Pacific: 23.29 [26.94, 20.25]) posterior distributions from the six Indo-Pacific countries with those from Seychelles, and more than 95% of the density in the posterior distributions falling below zero in all cases, suggests that both variables are important, generic predictors of macroalgal dominance post-disturbance on Indo-Pacific reefs. Interestingly, although the structure of reefs may vary based on zone or geomorphology, in 6 of the 7 pairwise comparisons, the reef with the greateststructurerecovered.Onaverage,reefs thatrecoveredhadcomplexity scores 0.85 (0.09–1.62 UI) greater than reefs than became dominated by macroalgae. 31. Wilson, S. K., Graham, N. A. J. & Polunin, N. V. C. Appraisal of visual assessments of habitat complexity and benthic composition on coral reefs. Mar. Biol. 151, 1069–1076 (2007). 32. Graham, N. A. J. et al. Lag effects in the impacts of mass coral bleaching on coral reef fish, fisheries, and ecosystems. Conserv. Biol. 21, 1291–1300 (2007). 33. Letourneur, Y., Kulbicki, M. & Labrosse, P. Length–weight relationship of fishes from coral reefs and lagoons of New Caledonia: an update. Naga, the ICLARM Quarterly 21, 39–46 (1998). 34. Froese, R. & Pauly, D. FishBase (http://www.fishbase.org) (2011). 35. Wilson, S. K. et al. Exploitation and habitat degradation as agents of change within coral reef fish communities. Glob. Change Biol. 14, 2796–2809 (2008). 36. Burkepile, D. E. et al. Nutrient supply from fishes facilitates macroalgae and suppresses corals in a Caribbean coral reef ecosystem. Sci. Rep. 3, 1493 (2013). 37. Ateweberhan, M., Bruggemann, J. H. & Breeman, A. M. Seasonal dynamics of Sargassum ilicifolium (Phaeophyta) on a shallow reef flat in the southern Red Sea (Eritrea). Mar. Ecol. Prog. Ser. 292, 159–171 (2005). RESEARCH LETTER Macmillan Publishers Limited. All rights reserved ©2015 38. Umezawa, Y., Miyajima, T., Yamamuro, M., Kayanne, H. & Koike, I. Fine-scale mapping of land-derived nitrogen in coral reefs by d15N in macroalgae. Limnol. Oceanogr. 47, 1405–1416 (2002). 39. Wooldridge, S., Brodie, J. & Furnas, M. Expoure of inner-shelf reefs to nutrient enriched runoff entering the Great Barrier Reef Lagoon: Post-European changes and the design of water quality targets. Mar. Pollut. Bull. 52, 1467–1479 (2006). 40. Ekebom, J., Laihonen, P. & Suominen, T. A GIS-based step-wise procedure for assessing physical exposure in fragmented archipelagos. Estuar. Coast. Shelf Sci. 57, 887–898 (2003). 41. Chollett, I. & Mumby, P. J. Predicting the distribution of Montastrea reefs using wave exposure. Coral Reefs 31, 493–503 (2012). 42. Wessel, P. & Smith, W. H. F. GSHHG – A Global Self-consistent, Hierarchical, High- resolution Geography Database. http://www.ngdc.noaa.gov/mgg/shorelines/ gshhs.html (2014). 43. Rohweder, J. et al. Application of wind fetch and wave models for habitat rehabilitation and enhancement projects – 2012 update. http://www.umesc.usgs. gov/management/dss/wind_fetch_wave_models_2012update.html (2012). 44. Kramer, D.L.& Chapman, M.R.Implicationsoffish homerange sizeand relocation for marine reserve function. Environ. Biol. Fishes 55, 65–79 (1999). 45. Nash, K.L., Graham, N. A.J. & Bellwood, D.R. Fish foragingpatterns, vulnerability to fishing,andimplicationsfor the managementofecosystemfunctionacrossscales. Ecol. Appl. 23, 1632–1644 (2013). 46. McGill, B. J., Enquist, B. J., Weiher, E. & Westoby, M. Rebuilding community ecology from functional traits. Trends Ecol. Evol. 21, 178–185 (2006). 47. Legendre, P. & Legendre, L. Numerical Ecology 2nd edn (Elsevier, 1998). 48. Ville´ger, S., Mason, N. W. H. & Mouillot, D. New multidimensional functional diversity indices for a multifaceted framework in functional ecology. Ecology 89, 2290–2301 (2008). 49. Coker, D., Wilson, S. K. & Pratchett, M. S. Importance of live coral habitat for reef fishes. Rev. Fish Biol. Fish. 24, 89–126 (2014). 50. Lester, S. E. et al. Biological effects within no-take marine reserves: a global synthesis. Mar. Ecol. Prog. Ser. 384, 33–46 (2009). 51. Fabricius, K. E. Effects of terrestrial runoff on the ecology of corals and coral reefs: review and synthesis. Mar. Pollut. Bull. 50, 125–146 (2005). 52. Loya, Y. et al. Coral bleaching: the winners and the losers. Ecol. Lett. 4, 122–131 (2001). 53. Pratchett, M. S. et al. Effects of climate-induced coral bleaching on coral-reef fishes: ecological and economic consequences. Oceanogr. Mar. Biol. Annu. Rev. 46, 251–296 (2008). 54. Patil, A. et al. PyMC: Bayesian stochastic modelling in Python. J. Stat. Softw. 35, 1–81 (2010). 55. Cheal, A. J., Emslie, M., MacNeil, M. A., Miller, I. & Sweatman, H. Spatial variation in the functional characteristics of herbivorous fish communities and the resilience of coral reefs. Ecol. Appl. 23, 174–188 (2013). 56. Graham, N. A. J. et al. Climate warming, marine protected areas and the ocean- scale integrity of coral reef ecosystems. PLoS ONE 3, e3039 (2008). 57. Wilson, S. K. et al. Exploitation and habitat degradation as agents of change within coral reef fish communities. Glob. Change Biol. 14, 2796–2809 (2008). 58. McClanahan, T. R., Ateweberhan, M., Darling, E. S., Graham, N. A. J. & Muthiga, N. A. Biogeography and change among regional coral communities across the western Indian Ocean. PLoS ONE 9, e93385 (2014). 59. Jennings, S., Grandcourt, E. M. & Polunin, N. V. C. The effects of fishing on the diversity,biomass and trophic structureofSeychelles’reef fishcommunities.Coral Reefs 14, 225–235 (1995). 60. Wilson, S. K. et al. Effect of macroalgal expansion and marine protected areas on coral recovery following a climatic disturbance. Conserv. Biol. 26, 995–1004 (2012). 61. Sheppard, C. R. C., Spalding, M., Bradshaw, C. & Wilson, S. Erosion vs. recovery of coral reefs after 1998 El Nin˜o: Chagos reefs, Indian Ocean. Ambio 31, 40–48 (2002). 62. Mumby, P. J. et al. Empirical relationships among resilience indicators on Micronesian reefs. Coral Reefs 32, 213–226 (2013). 63. Cheal,A. J.etal.Coral-macroalgal phase shifts orreefresilience:links withdiversity andfunctional roles of herbivorous fishes on the GreatBarrier Reef. Coral Reefs 29, 1005–1015 (2010). 64. Burkepile, D. E. & Hay, M. E. Herbivore species richness and feeding complementarity affect community structure and function on a coral reef. Proc. Natl Acad. Sci. USA 105, 16201–16206 (2008). 65. Rasher, D. B., Hoey, A. S. & Hay, M. E. Consumer diversity interacts with prey defenses to drive ecosystem function. Ecology 94, 1347–1358 (2013). 66. Mumby, P. J. & Harborne, A. R. Marine reserves enhance the recovery of corals on Caribbean reefs. PLoS ONE 5, e8657 (2010). 67. Burkepile, D. E. & Hay, M. E. Herbivore vs. nutrient control of marine primary producers: context-dependent effects. Ecology 87, 3128–3139 (2006). 68. Verge´s, A., Vanderklift, M. A., Doropoulos, C. & Hyndes, G. A. Spatial patterns in herbivory on a coral reef are influenced by structural complexity but not by algal traits. PLoS ONE 6, e17115 (2011). 69. Carpenter, R. C. & Edmunds, P. J. Local and regional scale recovery of Diadema promotes recruitment of scleractinian corals. Ecol. Lett. 9, 271–280 (2006). 70. Madin, J. S. & Connolly, S. R. Ecological consequences of major hydrodynamic disturbances on coral reefs. Nature 444, 477–480 (2006). 71. Done, T. J. in Perspectives on Coral Reefs (eds. Barnes, D. J. & Clouston, B.) Coral zonation, its nature and significance. 107–147 (Australian Institute of Marine Science, 1983). 72. Larned, S. T. & Atkinson, M. J. Effects of water velocity on NH4 and PO4 uptake and nutrient-limited growth in the macroalga Dictyosphaeria cavernosa. Mar. Ecol. Prog. Ser. 157, 295–302 (1997). 73. Graham, N.A.J. & McClanahan,T. R.The last call for marine wilderness?Bioscience 63, 397–402 (2013). 74. Graham,N.A.J.,Chong-Seng,K.M., Huchery,C.,Januchowski-Hartley, F.A.& Nash, K.L.Coralreefcommunitycompositioninthe contextofdisturbance historyonthe Great Barrier Reef, Australia. PLoS ONE 9, e101204 (2014). LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 1 | Changing condition of Seychelles coral reefs. a, b, Coral reefs of the inner Seychelles were typified by high coral cover and low macroalgal cover in 1994. c, d, The 1998 coral bleaching event caused widespread coral loss, but some reefs maintained their structural complexity (c), whileothers collapsed(d) by 2005. e, f, In 2011, many reefs had recovered to high live coral cover (e), while others had undergone a regime shift to abundant macroalgal cover (f). RESEARCH LETTER Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 2 | Principal components analysis of benthic composition on 21 reefs across the inner Seychelles 1994–2011. Reefs coloured blue are tracking back to pre-disturbance benthic composition in 2005–2011 following the 1998 bleaching event, whereas reefs coloured red are shifting to alternate benthic compositions, dominated by macroalgae (n 5 84). LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 3 | Distance from pre-disturbance benthic community composition. Euclidian distance in multivariate space, plotted against percent cover of dominant biotic benthic organisms (live coral in blue, macroalgae in red) (n 5 63). a, 2005 data. b, 2008 data. c, 2011 data. Shading represents 95%confidence boundsfor the mean trendlines ofeach habitat type. RESEARCH LETTER Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 4 | Changing coral and macroalgal cover is relation to pre-disturbance values. a, Data for recovering reefs, where the change in coral cover compared to 1994 was reducing through time, whereas change in macroalgae remained stable (n 5 42). b, Data for regime shifting reefs where the decline in coral cover persisted through time, and changes in macroalgae increased through time (n 5 42). LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 5 | Comparison of effect size posterior density distributions for depth and initial structural complexity in predicting coral versus macroalgae outcomes post disturbance in Seychelles versus 6 other countries across the Indo-Pacific. a, Depth effect size plot, dark blue posterior distributionfor Seychelles, grey for other countries (n 5 51). b, Initial structural complexity effect size plot, dark blue distribution for Seychelles, green for other countries (n 5 14). Depth and structural complexity variables were standardized in both analyses before estimation and all posterior distributions have more than 95% of their density below zero. RESEARCH LETTER Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 6 | Collinearity matrix of the eleven predictor variables. LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Figure 7 | Map of study sites around the inner Seychelles. Sites in blue are recovering from the 1998 mass bleaching event, whereas sites in red have undergone a regime shift to macroalgal cover. RESEARCH LETTER Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Table 1 | Coral cover and macroalgal cover at recovering and regime shifting sites in 2011, with the change (based on the model slope) in coral cover and macroalgae estimates from sites between 2005–2011 Coral cover (%) 1994 Coral cover (%) 2011 Coral slope 2005-2011 Macroalgae cover (%) 1994 Macroalgae cover (%) 2011 Macroalgae slope 2005- 2011 Recovering sites Mahe E Patch 19.3 12.3 3.3 0.6 0.1 -0.8 Mahe NW Carbonate 38.9 26.8 8.2 0.0 0.0 0.0 Mahe NW Granitic 10.8 19.0 4.2 0.0 0.0 0.0 Mahe NW Patch 17.3 21.3 3.3 0.0 0.0 0.0 Mahe W Carbonate 34.9 23.9 4.2 0.0 0.0 0.0 Mahe W Granitic 19.3 32.4 8.3 0.0 0.0 0.0 Mahe W Patch 20.8 47.5 9.3 0.0 0.0 0.0 Praslin NE Granitic 17.1 23.8 10.0 0.0 0.1 0.1 Praslin NE Patch 25.4 10.1 2.9 0.0 0.0 0.0 Praslin SW Granitic 17.4 17.5 6.8 0.3 0.0 0.0 Ste. Anne Granitic 38.5 20.4 3.3 1.0 1.5 0.7 Ste. Anne Patch 54.2 18.9 5.7 0.0 8.8 4.4 Average 26.1 ± 5.5 CI 22.8 ± 5.5 CI 5.8 ± 1.6 CI 0.2 ± 0.2 CI 0.9 ± 1.4 CI 0.4 ± 0.8 CI Regime shift sites Cousin Carbonate 49.7 0.4 -0.1 0.4 49.3 4.0 Cousin Granitic 23.3 2.8 0.8 0.6 37.3 16.8 Cousin Patch 38.6 2.8 1.1 0.0 26.0 -4.5 Mahe E Carbonate 28.6 1.9 0.3 12.4 52.3 2.6 Mahe E Granitic 9.3 1.0 -2.5 0.7 46.0 22.2 Praslin NE Carbonate 28.3 1.0 0.3 0.0 53.1 21.0 Praslin SW Carbonate 42.1 0.4 -0.1 3.4 77.6 19.6 Praslin SW Patch 17.3 3.4 1.3 4.7 14.5 1.3 Ste. Anne Carbonate 40.1 14.8 2.6 2.0 20.3 3.8 Average 30.8 ± 8.5 CI 3.14 ± 2.9 CI 0.40 ± 0.9 CI 2.7 ± 2.6 CI 41.8 ± 12.8 CI 9.63 ± 6.6 CI Average values for coral cover and macroalgal cover in 1994 also given for recovering versus regime shifting reefs. LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Table 2 | Rationale for predictor variables included in models determining different post-disturbance reef trajectories on Seychelles reefs Predictor Rationale References Habitat type Three habitat types were surveyed in Seychelles, including carbonate fringing reefs, granitic rocky reefs, and patch reef habitats. These habitats differ in their underlying matrix that may influence coral recruit success or likelihood of ecosystem collapse. 32,59,60 Pre-disturbance branching coral cover Branching corals are particularly vulnerable to coral bleaching events, and once dead the structure they provide erodes fairly rapidly. Therefore, reefs with a high cover of branching coral before a disturbance may be particularly vulnerable to extensive coral loss and a reduction in other organisms. 52,61 Juvenile coral density Successful settlement, survival and growth of new corals is thought to be key to coral recovery dynamics. 2,4 Depth Many threats on coral reefs are worst in shallow water, making shallow areas most vulnerable to change. Light penetration in shallow water may also favour rapid growth of fleshy macroalgae. 19,20 Herbivorous fish biomass Herbivorous fish are key to mediating competition for space among corals and algae. Biomass of these fish is a good proxy for function as the area of a reef grazed by these fish scales with both abundance and body size. 21,22,62 Herbivorous fish diversity Many types of algae have defences against herbivores, meaning that only certain species of fish can feed on some species of algae. This differential ability of fish species to control algae, can mean that a diversity of herbivorous species is required to provide the feeding complementarity necessary to control macroalgae. 63-65 Marine reserve status No-take marine reserves are expected to reduce fishing and hence enhance ecosystem processes, and may therefore promote faster rates of coral recovery. 28,66 Nutrient regime Higher nutrient loads in the waters around reefs can enhance the growth of algae and result in algae outcompeting corals, particularly when space becomes available through coral mortality. 19,67 Pre-disturbance structural complexity The structural complexity of a reef provides a great deal of the habitat variability for a diverse array of other organisms to inhabit. This in turn contributes to enhance a range of ecological processes, and provides niche space for coral settlement and survival. Structural complexity prior to a disturbance is expected to maintain ecosystem processes through the disturbance. 17,18,68 Sea urchin density Sea urchins are important herbivores on coral reefs, helping to control algal growth and promote successful coral recruitment and recovery. 2,69 Wave exposure Wave exposure influences coral distribution patterns, growth forms and colony sizes that will likely affect recovery trajectories. Similarly, algal growth can be enhanced with higher flow rates due to increased exposure to water borne nutrients, but algal dislodgement can occur where wave exposure is strong. 70-72 Including references 59–72. RESEARCH LETTER Macmillan Publishers Limited. All rights reserved ©2015 Extended Data Table 3 | Mean values with 95% confidence intervals for predictor covariates in Seychelles compared to other coral reef locations where similar data for the covariates were available Seychelles 95% CI N Other locations 95% CI N Countries Herbivore biomass 210.3 59.0 21 258.1 33.8 281 Kenya, Madagascar, Maldives, Mauritius, Mayotte, Reunion, Tanzania Pre-disturbance coral cover 28.1 5.4 21 29.0 3.9 60 Kenya, Tanzania Pre-disturbance structural complexity 3.2 0.1 21 3.1 0.2 30 British Indian Ocean Territory (Chagos), Great Barrier Reef (Australia) Urchin abundance 2.2 0.8 21 4.1 2.0 33 Kenya, Madagascar, Mozambique, Tanzania Data for regions other than Seychelles were provided directly by the authors of these studies23,73,74. LETTER RESEARCH Macmillan Publishers Limited. All rights reserved ©2015