Several denialists have sort to deliberately confuse the readership over the important evidence gathered by De’ath et al. (2009) on slowing coral calcification on the Great Barrier Reef.  Given the recent resurgence in this misinformation, I thought it would be a good idea to post Dr Glenn De’ath, Dr Janice M. Lough and Dr Katharina E. Fabricius’s recent reply  to Dr Peter Ridd’s confused and misleading claims.

The maintenance of coral calcification rates is critical for the future of coral reefs and it is, therefore, important to identify spatial patterns and temporal trends in the rates of coral calcification. Our recent report showed that substantial declines in coral calcification have occurred on the Great Barrier Reef in the last 20 years (De’ath et al., 2009), and similar reports are now emerging from other parts of the world (Tanzil et al., 2009). Ridd et al. here suggest that (1) ontogenetic effects, and (2) the last data points at the end of the recent cores, largely explain the ~14% decline in coral calcification we have shown across the Great Barrier Reef. We believe the assertions of Ridd et al. are erroneous due to: (1) their invalid assumptions about the data, and (2) their inappropriate statistical analyses.

Ontogenetic effects

Ridd et al. argue that we ignored the possibility that ontogentic effects contributed to the reported decline, namely that corals in their youngest years calcify at a faster rate than later in life. However, their main underlying assumption, that age of each short core is given by its number of growth records is wrong. Thus their Fig 2b derived from this assumption, is also wrong. Short cores are ~50 cm long (the length of the coring barrel), whereas the median height of the corals from which the short cores were taken was 1.5 m. The innermost year bands in short cores do not thus reflect early years of the corals’ life in the colonies sampled. Rather, corals were on average ~50 years old (rather than 1 year old, as Ridd et al assume) when the innermost year ring of the short cores was deposited.

In contrast, ontogenetic effects can be accurately assessed in whole colonies where the first years of the corals are preserved. However, Ridd et al. do not include year as a covariate factor, so their analysis is unable to disentangle the two potentially confounded effects of age and temporal trends in environmental conditions.

In the Report, we also investigated ontogenetic effects by comparing calcification in the last 15 years in the life of a coral (the outermost bands) in the 189 colonies collected from 1990 to 2005, and the 139 colonies sampled prior to 1990. We showed that for the cohort prior to 1990, the number of colonies and the number of reefs with increasing and declining rates were approximately equal in number, with 29 of the 56 reefs (51.7%) declining at an average rate of 0.11% yr-1 (SE=0.18%). However, in the 1990–2005 period, 12 of the 13 reefs (92.3%) declined at an average rate of 1.44% yr-1 (SE=0.31%), indicating a strong decline specific to that period, rather than reflecting ontogenetic properties of the outermost annual growth bands in coral skeletons.

End of core data

Ridd et al. argue that the last annual growth layer for each coral of the 2004 and 2005 series are negatively biased estimates of growth due to unspecified problems of measurement and should, therefore, be discarded. Such specific measurement problems are only likely if those corals were measured separately from the remainder. This was not the case as the data are based on the re-measuring and re-dating of all the material using the same methods and the same instrument, and conducted by one person (JML) within the past 5 years.

Ridd et al also argue that the series ending in 2005 (21 corals) did not show a significant decline in 2004, when the series ending in 2004 (containing 77 corals) showed a decline. It is perhaps also worth noting that Ridd et al use the term “significant” on six occasions without any statistical reference or justification. This statement was neither supported by their Fig 1D, nor by any form of statistical analysis or significance tests.

However, we also re-ran the temporal change model excluding the records of corals in 2004-5 (Fig. 1). The decline in calcification from 1990 – 2005 reduces to ~77% of that predicted when all data were included; still a decline of ~11.0%.

Figure 1. Decline in calcification based on all data and with the final years records for 2004-5 removed. The predicted reduction in the current decline for 1990-2005 is reduced from ~14.2% to ~11.0%.

Statistical analyses

Ridd et al. standardise the measurements of individual calcification records, average them for each year, and then analyse the temporal trends using an antiquated smoothing technique (Savitzky-Golay, 1964). There are three major problems with their approach:

(1)   It fails to account for the sampling structure whereby coral colonies are sampled from different reefs in highly variable numbers. There are between 1 and 46 colonies per reef, and the analyses in De’ath et al (2009) accounts for this structure by including random effects of reef and colony nested in reef in their generalized additive models (GAMs). The latter approach also takes into account the correlation across time due to repeated measures on colonies.

(2)   The fitted curves of Ridd et al. have no basis for the selection of smoothness (such procedures did not exist in 1964) and are mostly over-fitted (i.e. they are too wriggly), in particular in the last few years at which time Ridd et al. claim the anomalies exist. For example, the rapid increase in the last year or so of the truncated series is extreme. This contrasts with the failure of their fit to capture a rapid rise in the period 1940-45. All efforts to recapture their fits (no details were provided in Ridd et al.) failed despite using the Savitzky-Golay procedure with a wide range of smoothing.

The analyses of De’ath et al (2009) [SOM] used widely accepted model selection procedures for both random and fixed effect components of the GAMs, within which the smoothness of the temporal profiles was based on cross-validation.

(3)   None of Ridd et al.’s analyses use an inferential statistical model other than linear regression in their Fig. 2., and in that instance no confidence intervals or significance of the regressions are provided. It is also clear from inspection of those plots that strong serial correlation is present, which is not catered for in their analyses.

Conclusion

For the above reasons, we disagree with Ridd et al that the observed declines in coral calcification on the Great Barrier Reef are due to ontogenetic effects in corals, and that the last two years of record should be omitted from the data set. The predicted decline in calcification would drop from ~14.2% to ~11.0% were the last two records omitted; still a major decline. We maintain that this decline in calcification, probably due to synergistic effects of prolonged and repeated temperature stress and ocean acidification in tropical waters, is a real and serious issue for massive Porites on the Great Barrier Reef, and indeed for coral reefs around the world (Tanzil et al., 2009).

Dr Glenn De’ath, Dr Janice M. Lough and Dr Katharina E. Fabricius

Australian Institute of Marine Science, PMB 3, Townsville Qld 4810, Australia.