The Heat Is Online

Predicted recurrences of mass coral mortality in the Indian

Nature, Sept. 18, 2003 V. 425 -- pp. 294-297

Predicted recurrences of mass coral mortality in the Indian Ocean

CHARLES R. C. SHEPPARD

Department of Biological Sciences, University of Warwick, Coventry CV4 7AL, UK

Correspondence and requests for materials should be addressed to C.R.C.S. (csheppard@bio.warwick.ac.uk).

In 1998, more than 90% of shallow corals were killed on most Indian Ocean reefs1. High sea surface temperature (SST) was a primary cause2, 3, acting directly or by interacting with other factors3-7. Mean SSTs have been forecast to rise above the 1998 values in a few decades2, 3; however, forecast SSTs rarely flow seamlessly from historical data, or may show erroneous seasonal oscillations, precluding an accurate prediction of when lethal SSTs will recur. Differential acclimation by corals in different places complicates this further3, 7, 8. Here I scale forecast SSTs at 33 Indian Ocean sites where most shallow corals died in 1998 (ref. 1) to identify geographical patterns in the timing of probable repeat occurrences. Reefs located 1015° south will be affected every 5 years by 20102025. North and south from this, dates recede in a pattern not directly related to present SSTs; paradoxically, some of the warmest sites may be affected last. Temperatures lethal to corals vary in this region by 6 °C, and acclimation of a modest 2 °C by corals could prolong their survival by nearly 100 years.

Timing of recovery from the 1998 massive coral mortality on Indian Ocean reefs (refs 1, 2, 6 and Fig. 1) and how frequently rising SSTs will cause repeat mortalities are issues of practical urgency for many countries because of the high value of reefs to shoreline protection, biodiversity, protein supply and tourism6, 9, 10. Raw, modelled SSTs cross supposed thresholds of coral bleaching in a few decades2, but scaling problems in forecast data, coral acclimation3, 7, 8 and different absolute SSTs and rates of SST rise vary markedly between sites, which greatly affects estimates of when rising temperatures will reach values that proved lethal before. Exact dates remain unattainable, but a probability approach proves very revealing in terms of both timing and geographical pattern. Using the Indian Ocean, whose reefs were worst affected in the warm SSTs of 1998, I have 'blended' forecast SSTs seamlessly onto historical SST data, with appropriately scaled forecast seasonal cycles, for 33 sites.

Historical SST data from 18711999 from the HadISST1 data set11, 12 were combined with surface ('skin') temperature from 19502099 from the HadCM3 model for each site; the latter equates with SST (see http://www.cru.uea.ac.uk/cru/info/modelcc/). Both series are monthly; HadISST1 cells are 1° latitude and longitude, whereas the HadCM3 cells used are 2.5 3.75° (http://www.cru.uea.ac.uk/cru/info/modelcc/). HadISST1 comprises the latest historical data11 and HadCM3 is the most recent 'coupled' model from the Hadley Centre13. Other data sets can be substituted with minimal changes to the pattern described (Supplementary Information). With regard to warming schemes, more than 40 exist today; the one chosen is the thoroughly tested IS92a scheme14, which follows a median path13, 14 and has been most used for inter-comparisons (Supplementary Information).

'Raw' SST data require transformation before use (Fig. 2). In typical oceanic atolls (Chagos), historical and forecast series are discontinuous by more than 1.5 °C. In an exceptionally variable site (Saudi Arabian Gulf), measured SST amplitude is the highest known for coral reefs (15 °C) but its forecast SST series shows an erroneously small seasonality, which would mislead because it is the annual high SST that causes mortality15.

From each SST series (Methods), the probability of repeat critical SSTs can be determined. Four sites illustrate this (Fig. 3): those noted above plus those with the fastest and slowest rates of rise among these 33 sites. These curves integrate: the absolute SST at a site, its rate of rise, and the temperature that was lethal to more than 90% of the shallow corals there in 1998, which is a function of acclimation. The coral species are all from the same western Indian Ocean subset of fauna, with relatively few endemic exceptions16, 17.

 

Mortality in 1998 was triggered by SST rises lasting as short as the warmest month, although warming lasted 3 months in many areas1, 2, 18-22. As it is not yet known exactly how much warming triggers bleaching leading to mortality (or for how long), probability of recurrence curves are computed with warmest month, warmest 3 months and averaged warmest quarter. To compare these across sites, a uniform 'extinction date' is required. Although almost any point will suffice to show the geographical pattern, the date chosen is the year when there is a 0.2 probability of the warmest month (or warmest 3 months or averaged warmest quarter) hitting the 1998 value. The value 0.2 is ecologically fairly realistic: most corals do not mature until 5 years old2, 23 and today, 5 years after the 1998 event, most of these sites have recovered only marginally19, with coral cover rising in shallow water from 12% immediately after 1998 to about 35% today, as compared with pre-1998 values of 4075% (refs 1922).

Whether warmest month, warmest 3-month period or averaged warmest quarter values are used, earliest vulnerability will occur between 1015° south (Fig. 4) between 2010 and 2025 along all three northsouth transects. These 'extinction dates' recede southwards, but also recede northwards, initially towards the equator. The Arabian sites have a confused pattern owing partly to a "pseudo-high latitude effect"24 caused by cold summer upwelling in the Arabian Sea, such that some sites with highest temperatures have the most distant extinction dates; in others, upwelling precluded determination of the 'extinction date' (Supplementary Information). Many Arabian corals annually survive temperatures that killed the same species elsewhere in 1998 (refs 15, 19, 20). The northern Red Sea remains relatively unaffected but, even with their marked temperature acclimation, most corals in the Arabian Gulf were killed by the 1998 peak SSTs. The fact that most sites between 0° and 15° south will have a 1 in 5 probability annually of suffering a month as warm as that of 1998 within 1015 years means that several of the world's poorest countries, for which reefs provide essential resources6, 9, 10, will be affected soonest.

A modest acclimation or adaptation by corals7, 8 would greatly prolong time before their 'extinction date'. In 1998, lethal SSTs varied by 5 °C (from <29 °C to >34 °C) depending on location. By raising the SST presumed to be lethal at a site by 2 °C, nearly a century will be gained (Fig. 5). A value of 2 °C seems modest compared with that already achieved by today's Arabian corals15, 17, but the latter had millennia to acclimate. In these 33 sites today, total coral cover has improved little since 1998 (refs 1820). The few decades before probable recurrence of lethal SST values may not be sufficient for recovery to become well established.

Diverse physiological and pathological factors are triggered by a rise in SST. Most act with, or are triggered by, temperature7. Some physical factors such as increased cloudiness4 and water exchange5 oppose these effects. SST should not be taken as the only important factor, although it is at worst a quantitative, measurable surrogate.

Methods
Scaling of forecast data series SSTs forecast from climate models rarely flow seamlessly from historical series, and errors in forecast seasonal amplitudes further prevent accurate estimation of when lethal mortalities might recur. The first transformation simply adjusts each forecast data series by the mean difference in values in the overlapping data between 1950 and 1999 (n = 600 months).

The second transformation scaled the seasonal amplitude of each forecast series to match that of each site's historical data. Fourth-order polynomial fits were computed for each historical series. Fourth order was chosen by extensive trial and error; orders up to fourth significantly increased R2, whereas higher orders added no significant further improvement. Predicted values were subtracted from their corresponding monthly data points to obtain series of residuals. This was repeated for forecast data at each site. Correlations between residuals of the historical and forecast SSTs at each site in overlapping years (19501999) are always highly significant; in the examples plotted, (Fig. 2), r = 0.973 for the Gulf and r = 0.762 for Chagos (n = 600).

By using the 'normdist' function of Excel (Microsoft), residuals of forecast series were expressed as standard deviations. By substituting the standard deviation of the historical data residuals in place of that of the forecast data, Excel's 'norminv' function was then used to compute forecast temperature residuals whose annual oscillation matches in magnitude that of the historical series. Adding these scaled residuals back to the polynomial curves gives, for every site, an SST series (18712099) with no disjunction and, where they overlap, the same seasonal amplitude.

Standard deviations of HadISST1 residuals are stable with time11. HadCM3 residuals increase in amplitude by 325% with time, reflecting the climate model's increasing uncertainty into the future, and this is carried through in the transformed series. Other historical SST series could be substituted; the changes to the pattern are minor to negligible (Supplementary Information).

Computing probabilities of recurrence Subsets of data were extracted for warmest months, three warmest months and averaged warmest quarters of each year. Residuals of all but one (Alphonse atoll) warmest month series have normal distributions (KolmogorovSmirnov tests). Warmest quarters' residuals are also normally distributed in all sites except one (granitic Seychelles). As time proceeds, the difference between the lethal 1998 SST value (also expressed as a residual) and the normally distributed population of SSTs decreases. For each month, '1 - normdist' determines the probability that each site's lethal temperature is part of the site's population of temperatures. This yields probability curves of repeat recurrences of the peak temperature of 1998.

In the warmest 3-month data sets, residuals in only half of the sites have normal distributions (they lack extended 'tails'). For these, 'bootstrap tests' were used instead of the normdist function to compute probability; probability was the number of residuals in the whole data set with a value greater than the test value, divided by the total number. Curves almost exactly match those obtained by the normdist method. This test was also used for the north Seychelles site for the quarterly series to extend that transect northwards; the test differed by less than 1 year from that obtained with the normdist function at that site.

'Lethal' SSTs and timing Peak SSTs ranged from February in the south to September in the northwest. For 27 sites, the warmest quarter was the peak month with the preceding and the following months. For the other six sites, it was the peak month with the two preceding months. For the warmest month and 3-month tests, the test SST value was the warmest 1998 HadISST1 temperature. For the warmest quarter test, the average SST of the warmest 3 months was used. These temperatures were generally only less than 0.2 °C warmer than any earlier recorded temperatures at that site.

Supplementary information accompanies this paper.

Received 13 May 2003;accepted 5 August 2003

------------------

References

1.

Wilkinson, C. R. in Seas at the Millennium, an Environmental Evaluation vol. 3 (ed. Sheppard, C.) 43-57 (Elsevier, Amsterdam, 2000)

2.

Hoegh-Guldberg, O. Climate change, coral bleaching and the future of the world's coral reefs. Mar. Freshwater Res. 50, 839-866 (1999) | ISI |

3.

Pittock, A. B. Coral reefs and environmental change: Adaptation to what? Am. Zool. 39, 10-29 (1999) | ISI |

4.

Turner, J., et al. in Coral Reef Degradation in the Indian Ocean, Status Report 2000 (eds Souter, D., Obura, D. & Linden, O.) 94-107 (CORDIO, Stockholm, 2000)

5.

Nakamura, T. & Woesik, R. Water-flow rates and passive diffusion partially explain differential survival during the 1998 bleaching event. Mar. Ecol. Prog. Ser. 212, 301-304 (2001) | ISI |

6.

Sheppard, C. R. C. The main issues affecting coasts of the Indian and western Pacific Oceans: a meta-analysis from seas at the millennium. Mar. Pollut. Bull. 42, 1199-1207 (2001) | Article | PubMed | ISI | ChemPort |

7.

Douglas, A. W. Coral bleaching--how and why? Mar. Pollut. Bull. 46, 385-392 (2003) | Article | PubMed | ISI | ChemPort |

8.

Gates, R. D. & Edmunds, P. J. The physiological mechanisms of acclimatisation in tropical reef corals. Am. Zool. 39, 30-43 (1999) | ISI |

9.

Wilkinson, C. R. et al. Ecological and socio-economic impacts of 1998 coral mortality in the Indian Ocean: an ENSO (El Niño-Southern Oscillation) impact and a warning of future change? Ambio 28, 188-196 (1999) | ISI |

10.

Cesar, H. S. J. (ed.) Economics of Coral Reefs 244 (CORDIO, Kalmar, Sweden, 2000)

11.

Rayner, N. A. et al. Global analyses of SST, sea ice and night marine air temperature since the late nineteenth century. J. Geophys. Res. Atmos. (in the press)

12.

Sheppard, C. R. C. & Rayner, N. Utility of the Hadley Centre sea ice and surface temperature data set (HadISST1) in two widely contrasting coral reef areas. Mar. Pollut. Bull. 44, 303-308 (2002) | Article | PubMed | ISI | ChemPort |

13.

McAvaney, B. J. et al. Climate Change 2001: The Scientific Basis. Intergovernmental Panel on Climate Change (ed. Houghton, J. T. et al.) 471-523 (Cambridge Univ. Press, Cambridge, UK, 2001)

14.

Cubasch, U. et al. Climate Change 2001: The Scientific Basis. Intergovernmental Panel on Climate Change (ed. Houghton, J. T. et al.) 525-582 (Cambridge University Press, Cambridge, UK, 2001)

15.

Sheppard, C. R. C. & Loughland, R. Coral mortality and recovery in response to increasing temperature in the southern Arabian Gulf. Aquat. Ecosyst. Health Mgmt 5, 395-402 (2002) | Article |

16.

Veron, J. E. N. Corals of the World vols 1-3 (Australian Institute of Marine Science, Townsville, Australia, 2000)

17.

Sheppard, C. R. C. & Sheppard, A. L. S. Corals and coral communities of Arabia. Fauna Saudi Arabia 12, 7-192 (1991)

18.

Sheppard, C. R. C., Spalding, M., Bradshaw, C. & Wilson, S. Erosion vs. recovery of coral reefs after 1998 El Niño: Chagos reefs, Indian Ocean. Ambio 31, 40-48 (2002) | PubMed | ISI |

19.

Linden, O., Souter, D., Wilhemsson, D. & Obura, D. Coral Reef Degradation in the Indian Ocean, Status Report 2002 284 (CORDIO, Kalmar, Sweden, 2002)

20.

Riegl, B. Effects of the 1996 and 1998 positive sea-surface temperature anomalies on corals, coral diseases and fish in the Arabian Gulf (Dubai, UAE). Mar. Biol. 140, 29-40 (2002) | Article | ISI |

21.

McClanahan, T. R. Bleaching damage and recovery potential of Maldivian coral reefs. Mar. Pollut. Bull. 40, 587-597 (2000) | Article | ISI | ChemPort |

22.

Spencer, T., Teleki, K. A., Bradshaw, C. & Spalding, M. D. Coral bleaching in the southern Seychelles during the 1997-1998 Indian Ocean warm event. Mar. Pollut. Bull. 40, 569-586 (2000) | Article | ISI | ChemPort |

23.

Harrison, P. L. & Wallace, C. C. in Coral Reefs (ed. Dubinsky, Z.) 133-206 (Elsevier, Amsterdam, 1990)

24.

Sheppard, C. R. C. & Salm, R. V. Reefs and corals of Oman, with a description of a new species of coral (Scleractinia, Acanthastrea). J. Nat. Hist. 22, 263-279 (1988) | ISI |

Acknowledgements. The HadCM3 data were provided by the Hadley Centre for Climate Research through D. Viner, who also provided information on the data's characteristics. I thank M. Keeling and G. Medley for advice on analyses; N. Rayner of the Hadley Centre for information on the HadISST1 data and for communicating results before publication; and O. Langmead and A. Edwards for assistance with data extraction.

Competing interests statement. The authors declare that they have no competing financial interests.