- Email: [email protected]
Have coral calcification rates slowed in the last twenty years?
Marine Geology 346 (2013) 392–399
Contents lists available at ScienceDirect
Marine Geology journal homepage: www.elsevier.com/locate/margeo
Have coral calcification rates slowed in the last twenty years? Peter V Ridd a,⁎, Eduardo Teixeira da Silva a, Thomas Stieglitz a,b a b
Marine Geophysical Laboratory, School of Engineering and Physical Science, James Cook University, Townsville 4811, Australia Laboratoire des sciences de l'environnement marin CNRS UMR 6539, Institut Universitaire Européen de la Mer, 29280 Plouzané, France
a r t i c l e
i n f o
Article history: Received 14 February 2012 Received in revised form 30 July 2013 Accepted 3 September 2013 Available online 14 September 2013 Communicated by G.J. de Lange Keywords: coral calcification Great Barrier Reef ocean acidification ocean pH
a b s t r a c t This paper reports a reanalysis of calcification rates of 328 Porites cores from the Great Barrier Reef from which previous workers have concluded that a 14% reduction in calcification rates has occurred between 1990 and 2005. In this reanalysis it is shown that the apparent reduction in the Porites spp. calcification rate in the last two decades is at least partly due to a combination of (a) ontogenetic effects (disregarded in the previous analysis), combined with a highly variable age distribution of the coral growth bands with time, and (b) a systematic data bias clearly evident in the last growth band of each core. When the outermost growth band in addition to bands which have record age less than 20 years was excluded from the analysis, the dramatic fall in calcification after 1990 was no longer evident. © 2013 Elsevier B.V. All rights reserved.
1. Introduction There is widespread concern that a hitherto unperceived consequence of global carbon dioxide emissions is a decrease in ocean pH which will have dire consequences for the calcification of calcareous marine organisms such as corals. A recent analysis (De'ath et al., 2009) of coral calcification data extracted from 328 Porites corals collected from 69 reefs over the Great Barrier Reef (GBR) that span the last 400 years has indicated that there has been a dramatic decline in coral calcification by as much as 14% between 1990 and 2005. It was suggested that a tipping point was reached in 1990 when declining ocean pH due to increased atmospheric CO 2 combined with increasing temperature stress caused rapid reductions in calcification. The 14% decline in calcification rate between 1990 and 2005 (De'ath et al., 2009) is prima facie a surprising result because a previous comprehensive study (Lough and Barnes, 2000), using a subset of the data used in De'ath et al. (2009), demonstrated a statistically significant 4% increase in GBR coral growth over the 20th century. In addition, it is notable that a more recent paper on calcification rates on Australia's north western coastline does not indicate any significant decline in calcification rates after 1990 (Cooper et al., 2012). However, laboratory experiments show that calcification decreases under increasing pH for a variety of reef organisms (Hoegh-Guldberg et al., 2007), suggesting
⁎ Corresponding author. Tel.: +61 747814978; fax: +61 747815880. E-mail address: [email protected] (P.V. Ridd). 0025-3227/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.margeo.2013.09.002
that modern coral reefs may be facing major challenges due to environmental change (Carpenter et al., 2008). Clearly the precise nature of the trend in coral calcification is of considerable importance for scientists, managers and policy makers alike. The study of De'ath et al. (2009) used a linear mixed effect (LME) model, an analysis technique that aims to compensate for the many challenges involved with analysing complex data such as coral growth time series. The different number of corals from each location and a large latitudinal sampling range, which implies varying environmental factors leading to variable calcification rates, are examples of confounding variables that can be accounted for by LME. However there are additional important aspects of the data set that were not taken into account in the analysis of De'ath et al. (2009). These include (a) the use of three different types of coral samples (long cores, short cores and colonies) which produce samples of different lengths and therefore record ages (record age is defined as the time between a particular yearly calcification record and the first data record in the series), (b) a strong temporal variation in the average age of the corals over the sampling period, and (c) the likelihood of systematic sampling problems. These three confounding variables intrinsic to the data set are discussed in this paper. At the centre of the analysis by De'ath et al. (2009) is the assumption that calcification rates for a particular coral do not change with the age of the coral (if environmental conditions remain constant), i.e., there are no ontogenetic effects. Here we reanalyse the coral calcification data from the Great Barrier Reef and show that the apparent decline over the last two decades may be the result of a combination of ontogenetic effects and measurement artefacts
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
rather than an abrupt calcification reduction caused by environmental conditions. 2. Material & methods 2.1. Data The coral calcification data of 328 Porites spp. from the Great Barrier Reef (GBR) presented in De'ath et al. (2009) can be subdivided into 3 broad corals categories: 61 long cores, 135 short cores and 132 colonies. Long cores were collected using a large drill rig coring device, from a wide range of locations, and their time span is on average 155 years. Short cores were collected using a small drill, mostly from 5 reefs in the central section of the GBR between 18° and 20°S, and their time span is on average 28 years. Colonies were sampled by taking a saw slice through the coral, and they have on average a time span of 23 years (Fig. 1). Before 1940, only long core data contributes to the overall calcification rate and most of the long core data ends by 1990. Short core data starts around 1940 and mostly ends around 2005. Colonies also start to contribute data around 1940 and roughly 2/3rds of the colony records terminate around 1990. The other third are generally very short records which end around 2005 (Fig. 1).
393
Also evident at ca. 1990 is the discontinuity of records especially in the colony data (Fig. 1b). The discontinuity of records at 1990 occurs because the field work when the coral cores were recovered cover two distinct periods: (a) the late 1980's and early 1990's, and (b) the early part of the 2000's, especially in 2003, 2004 and 2005. The short core data (Fig. 1c) does not show a discontinuity at around 1990 as this sampling technique was only used in the later sampling period. For this reason, all the short core data ends on or after 2003. The number of contributing corals changes dramatically with time (Fig. 2). Before 1940 the total number of corals was around 50 and rose to 269 in the early 1980's before declining to 189 in 1999, and then sharply reducing to only 21 in 2005. 2.2. Data analysis methods LME analysis for the calcification data was performed in R scripting language (Team, R.D.C., 2009), with the package mgcv, using exactly the same R code as provided to us by G. De'ath (pers. comm.). When running this code it was necessary to select the k value of the gamm function, which gives the dimension of the basis used to represent the smooth term in the spline model (Team, R.D.C., 2009). The smaller value of k (fewer degrees of freedom) causes some of the detailed temporal changes in the calcification curve to be masked. In short, a
Long Cores
a Number of corals
121 101 81 61 41 21 1 1900
1910
1920
1930
1940
1960
1970
1980
1990
2000
2010
1960
1970
1980
1990
2000
2010
1960
1970
1980
1990
2000
2010
Colonies
b Number of corals
1950
121 101 81 61 41 21 1
1900
1910
1920
1930
c
1940
1950
Short Cores
Number of corals
121 101 81 61 41 21 1 1900
1910
1920
1930
1940
1950
Fig. 1. Time span of each coral record included in the reanalysis. Each horizontal bar represents the time period over which a particular coral contributed to the data set for (a) long cores, (b) colonies, and (c) short cores. Dark Grey represents coral growth bands with record age N20 years. Light grey represents the growth band of corals with record age ≤20 years. Black represents the last growth bands which have record age N20 years.
394
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
Fig. 2. Temporal variation of the (a) mean record age of the corals and (b) number of contributing corals, for all the corals together and for colonies, long cores and short cores individually. Red lines represent the cases for growth bands with record age N20, 10 and 5 years when the outer growth bands have also been removed.
higher k value will generally make the curve more “wiggly”. Following De'ath et al. (2009), in this paper we adopted the AIC (Akaike Information Criterion) value to select the k value, choosing the lowest but conservative AIC for each fitted model. In order to investigate potential problems with data from the outermost growth band, a simple data normalisation procedure, which was similar to that used by Lough (2008) was adopted. The normalised calcification rate in year t for the mth coral is calculated by: norm
Cm
ðt Þ ¼
C m ðt Þ Cm
;
ð1Þ
where Cm(t) is the raw calcification rate of the coral growth band in year t and C m is the average calcification rate of all its growth bands
(i.e., over the entire time span of the mth coral). Then, the aggregate normalised calcification rate for year t, was calculated by: Xi C ðt Þ ¼
norm C 0 ðt Þ i0 ¼1 i
i
ð2Þ
where i is the number of corals contributing to the data in year t, and the summation is over the i contributing corals in year t. This procedure was used to look at the last band of the series that ended in 2004 and 2005. These two years were chosen because of the large numbers of corals which terminated in these years. It should be noted that this normalisation procedure was not used with the LME analysis which used the raw calcification data.
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
2.3. Definition of terminology
395
remains sufficiently large so that uncertainties in the results do not dominate.
In this paper, we adopted the term record age to describe the time between a particular yearly calcification record and the first data record in the series (i.e. within one core). It should not be confused with the ‘true’ age of the coral which is the total time period that it lived before being collected. True age could be considerably greater than record age, if the core taken from a coral did not include its initial growth point. Measurements of the physical size of the corals which could be used to estimate true age are not available for this data set. The time span of a coral is defined as the period of time from which calcification data was available for that coral. To demonstrate these different parameters, consider a coral colony that started growing in 1910, was collected in 2005, and from which calcification data is available from 1930 to 2004. The final true age of the colony is 95 years but this is generally not known because the date of initiation of growth is unknown. The time span of the record is 75 years, and the record age varies from 1 year for the 1930 record to 75 years for the 2004 record. 3. Results The basic approach in this paper was to repeat the analysis with subsets of the original data after the exclusion of the following two classes of data. Firstly, data from the outermost growth band of each individual core is disregarded; because it is apparent that this data has a systematic bias, most likely due to instrumental artefacts. Secondly, growth bands with record age less than (a) 5 years, (b) 10 years and (c) 20 years were disregarded i.e., removal of data from corals when they are relatively young. The rationale of this exercise is to remove growth bands (in this case the younger bands) which are more likely to experience an ontogenetic variation in growth rate. If the analysis of De'ath et al. (2009) is robust, the removal of both these classes of data should make little difference to the result. In the following sections we explore these two scenarios and show how they can affect the original data interpretation presented in De'ath et al. (2009). It is worthwhile to mention that the analysis herein was carried out using statistical methodology and data provided to us by G. De'ath, and is identical to that used by De'ath et al. (2009). Thus results from both analyses are directly comparable. The data set size (i.e. numbers of corals and annual growth bands) is sufficiently large to allow testing of results using subsets of the data. If the original analysis is sound, it should be expected that a similar result should be found from a subset of the data provided that the subset
3.1. Systematic low bias in calcification rate of the last growth band Because a large number of calcification time series ends in 2004 and 2005, a detailed analysis of the bias in the outer growth band can be made on these two sets of corals. Fig. 3 shows the aggregated calcification rate (using Eq. (2)) for coral series ending in 2004 and 2005. Both series show a considerable decline in calcification rate in the last growth band. A One-way ANOVA test indicates that the average aggregated calcification rates for the final year on both series are significantly different from the previous years (F(24,1634) = 4.923, p b 0.001 and F(25,490) = 2.06, numbers between brackets stand for degrees of freedom and number of data points, respectively, p = 0.002 for 2004 and 2005 series, respectively). Importantly, the series ending in 2005 did not show any significant decline in 2004 — the year in which the series that ended in 2004 did show a significant reduction. If there was no data bias in the outer bands, the conclusion must be drawn that in 2004, the corals of the 2005 series somehow avoided the environmental conditions that so affected the 2004 series. It is difficult to see how this could occur. Instead, it is highly likely that the last annual growth bands have a systematic low bias, perhaps due to difficulty in identifying the extension of this band. It should be noted that this analysis of potential problems with the last growth band has concentrated on the corals that end in 2004 and 2005. A very large number of corals are in this category, and thus a useful comparison can be made due to relatively small error margins. Similar problems may also occur with data series ending in other years, but due to a smaller numbers of corals producing higher error margins, potential problems in the last band of these series may be more difficult to detect. In order to be consistent, all of the last growth bands were removed in the analysis shown in the next section, not just the last bands of series ending in 2004 and 2005. To demonstrate the considerable influence of the bias due to these last growth bands, the LME model was used both with and without the final growth bands included (Fig. 4) (note: the final bands of all corals were removed in Fig. 4b, not just the final growth bands for the series ending in 2004 and 2005). It is evident that the fall in the calcification rate at the end of the series is considerably reduced when the last band is removed. Using the LME model of De'ath et al. (2009) the
Aggregated calcification rate (g/cm2/y)
1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 1979
2004 series 2005 series 1984
1989
1994
1999
2004
Year Fig. 3. Annual average aggregated calcification rate calculated using Eq. (2) for corals with end band corresponding to 2004 (opened circle, dashed line) and 2005 (opened square, solid line). To make error bars distinguishable, data points were displaced −0.1 and +0.1 of the year for the 2004 and 2005 series, respectively. Vertical bars indicate 95% confidence intervals.
396
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
Fig. 4. Temporal variation of the calcification rate (g cm−2 y−1) obtained using the R code as provided by G. De'ath, with k = 11, for (a) all the data together, and (b) when the outermost growth band was excluded (bold solid line with bold dotted lines as error limits).
Fig. 5. Temporal variation of the calcification rate obtained for (a) all the data together as presented in De'ath et al. (2009), and when coral growth bands younger than (b) 5 years, (c) 10 years, or (d) 20 years are removed. The data set also does not include the outermost growth band for the cases b, c and d. Dotted lines indicates the 95% confidence intervals for comparison between years. The number of corals used for each of the four cases is shown in Fig. 2b.
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
removal of this data alone resulted in a reduction in the post 1990 fall from 14% to ca 9% (Fig. 4). 3.2. Ontogenetic effect When the growth bands of corals with record age less than 5 years, 10 years and 20 years were removed from the analysis (together with the outermost growth band), the post 1990 drop in the calcification rate was reduced from 14% to 8%, 6% and 1%, respectively (Fig. 5). The post 1990 reduction in calcification is thus no longer evident once the uncertainties in the calculation are taken into account (wide error bars in Fig. 5). Notably, the calcification rate at the final year of the time series is unmeasurably different from the calcification rates of the early 1900's, even when only the bands representing the youngest 5 years of growth are removed, and uncertainties are taken into account (Fig. 5b). For all the fitted models, the selection of the k-value was based on the lowest AIC. For Fig. 5a we used k = 10, as used in De'ath et al. (2009), even though k = 11 produces the lowest AIC. For Fig. 5b and c, k = 11 produced the lowest AIC. For Fig. 5d, the lowest AIC value (i.e., 8273) corresponded to a k = 5 which was marginally lower than the AIC value (8275) for k = 11. However for comparability with Fig. 5b and c, the value of k for Fig. 5d was set to 11. Use of different k values (either 11 or 5) does not affect the conclusion that the removal of data with record age less than 20 years (plus the final record of each coral) removes the post 1990 fall in calcification. Fig. 6 shows the calcification time series for k = 5, in which the post 1990 fall is not evident. All the fitted models exhibited a p b 0.001. 4. Discussion The considerable change in the result due to the exclusion of the youngest growth bands of each coral record is concerning. It is likely to be due to a combination of two factors: (a) a slowing of growth rate with age of a coral, and (b) the record age distribution of the coral growth bands changes with time. These factors are discussed in detail as follows. Whilst De'ath et al. (2009) assume absence of any ontogenetic effects on coral calcification, the presumption of constant growth rates of Porites with age is not established in the scientific literature. Indeed many authors allude to the possibility of significant ontogenetic variations (Buddemeier and Kinzie, 1976; Anthony et al., 2002;
397
Langmeada and Sheppard, 2004; Holmes and Johnstone, 2010). In particular Holmes & Johnstone (Holmes and Johnstone, 2010) state that small colonies (b 20 cm diameter) can grow considerably faster than large colonies (N50 cm diameter). Porites exhibit an average growth rate of 1.4 cm y−1 (De'ath et al., 2009), which suggest that they may experience ontogenetic variations up to 30 years old. An ontogenetic reduction of the calcification rate would not affect the data analysis if the age distribution of the coral data (i.e., record age of growth bands) does not change with time. However, as happens in this data set, if younger corals are included in the latter part of the data set, slowing growth curves of these corals would result in a reduction in the aggregated calcification rate. In this data set, the number of coral growth bands and their average record age varies dramatically with time (Fig. 2). The average record age of growth bands contributing to each year in the time series exhibits a clear reduction after 1940, with the inclusion of short cores and colonies into the data set, falling from around 110 years in 1940 to around 30 years towards the end of the sampling period (Fig. 2a). The relatively short record age of the colony data is particularly evident when compared with the long core and short core data. For example in 2003, the average colony age was only 14.2 years compared with 143 years and 27.9 years for long and short core data respectively (Fig. 2a). This dramatic change in the average colony age means that any ontogenetic effect on calcification rate could dominate the analysis. The change in result after removal of young growth bands is strong evidence that there is an ontogenetic artefact in the data set. The fact that there is any change in result by changing the age structure of the data is by definition proof that the age of the coral is important. In contrast De'ath et al. (2009) have stated that an ontogenetic effect associated with the outermost annual growth bands is unlikely to exist. They point out that of the 189 corals (from 13 reefs) in the period from 1990 to 2005, 12 reefs (92%) and 137 corals (72.5%), i.e., a large majority of reefs and corals, showed a negative trend in calcification rate. For the 139 corals (from 56 reefs) sampled before 1990, only 29 reefs (52%) and 68 corals (49%) showed a decline in calcification rate, i.e., an equal number of reefs and corals increased or decreased calcification rate. This however does not provide much insight into ontogenetic effects as the coral samples that ended around 1990 are disproportionately older than those that end in the early part of this century (Fig. 2a), and thus may be displaying considerably less slowing of
Fig. 6. Temporal variation of the calcification rate obtained when corals growth bands b20 years are removed in addition to the outermost growth band. (a) k = 11, (b) k = 5.
398
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
growth rate. For example almost all long cores ended around 1990 at an average record age of 150 years (Fig. 2). Post 1990, without the contribution of the long cores, the average record age drops to less than 30 years at the end of the sampling series. It should be noted that the problems posed by the ontogenetic effects do not prove that there is no environmental effect that may be causing decreased (or increased) calcification, but it makes it considerably more challenging to extract the environmental signal from the ontogenetic artefact. In early correspondence about his paper, G. De'ath (pers. comm.) raised a strong argument that there may be a general environmental signal at play in addition to the ontogenetic effect, i.e., they demonstrated that the colonies which start before 1985 have slow calcification at a reduced rate than to those colonies which start after 1985. In our analysis above we show that excluding all the young layers causes the post 1990 drop to disappear, i.e., the older corals are not experiencing a calcification decline. However it is possible that only the younger corals are susceptible to the potentially changing environmental conditions. To shed further light on the possibility that the younger corals are showing a calcification reduction after 1990 we have plotted the calcification rate determined only from corals that have record ages less than 20 years (see Fig. 7) which shows a clear, though, modest (between 4% and 11%), reduction in calcification rate after 1990. There is no measurable difference between the calcification rates in 2004 compared with 1960 once the error margins are considered. It should be noted that the data before 1950 is of little value because of high errors and the fact that most of the growth bands considered are likely to be from corals much older than 20 years. This is because they are taken from long cores and short cores where there is no guarantee that the core has passed through the oldest part of the coral. Fig. 7 uses only that data which was excluded from Fig. 5(d) (except the outer layer which is excluded in both cases) and the result is very different. Fig. 7 shows a drop in calcification in the last 15 years and Fig. 5(d) shows no measureable change. This reaffirms our proposition that there is an ontogenetic effect, and it indicates that there may be a modest change in calcification in the last 15 years in the younger corals. One potential problem with this conclusion is that even Fig. 7 has a changing age distribution with time which may influence the results.
4.1. Is using data subsets a valid method to show that the post 1990 decline may be a data artefact? An important question that must be asked about the removal of data from the analysis is: what effect does data removal have on the error of the analysis? Fig. 5 shows that the error in the analysis degrades only marginally with the exclusion of data. This error estimate was calculated with the identical methodology as used by De'ath et al. (2009). The crucial parameter influencing the error estimate is the number of corals and reefs that contributes to the data in a given year. For the whole data set, with the exception of the last two years, the smallest numbers of corals sampled occur in the pre 1940 period (Fig. 2b) not in the post 1990 period which is of most interest here. It is important to note that when the growth bands younger than 20 years are disregarded, the number of contributing corals during the post 1990 period, generally still exceeds the number of corals contributing to the period up to 1960 (Fig. 2b, red line). Thus it cannot be argued that so much data was removed that the post 1990 result would become invalid. Moreover, the greatest number of bands was removed from data between 1970 and 1990, not the post-1990 period where the reduction in calcification occurs (Fig. 2b, difference between red and black curves). It might be argued that we have thrown away most of the data used to detect the change, however as mentioned above, the majority of the data removed was from the pre-1990 period where for much of the time calcification increased (Fig. 4). If the conclusion about the sudden post1990 decline is to be accepted, there needs to be a good explanation of why removing the younger bands from all of the corals should make such a large difference to the result. 4.2. Can the problems with the data be overcome by modelling the ontogenetic effect and the end-of-series problem? By looking at different classes of data it has been shown that there is considerable doubt that the fall in calcification after 1990 is a real signal or at least partly an artefact of the changing age distribution of the corals and the end-of-series problem. To overcome these problems, one approach could be to attempt to model both the end of series problem and the ontogenetic effect. For example, if the end of series problem is due to an instrumental problem that could be measured, this could be modelling and the analysis corrected. Alternatively, the end of series problem could have been due to a change in procedure over the 20 years that these corals were analysed. If this effect could be measured, a correction could be applied to the data. The ontogenetic effect could be corrected if more information became available about the change in calcification rates as the coral grows with all other factors remaining fixed. However modelling away the problems with the data is fraught with danger if the physical or biological cause of the problem cannot be clearly identified and quantified. Great care would certainly be needed if a purely statistical reanalysis is used to correct the problem as more variables will now be required to contribute to the model reducing its reliability. It would be particularly difficult to correct the ontogenetic effect by a purely statistical reanalysis because the true age of the coral is not known but would be required in such a reanalysis. Any statistical approach to correct for the ontogenetic effect would need to be carefully scrutinised. 5. Conclusion
Fig. 7. Temporal variation of the calcification rate obtained for all data with record age less than 20 years, i.e., it uses the data that was excluded in Fig. 5(d) except the data from the outermost growth band which is similarly excluded. Dotted lines indicates the 95% confidence intervals for comparison between years, in the model, k = 9, n = 4960 and p b 0.01.
Despite the caveats discussed in this paper, the data set analysed by De'ath et al. (2009) is a highly valuable resource of coral calcification rates. Some of the problems associated with ontogenetic effects could be addressed if more information about the true age of the coral was available. The approximate dimension of the individual Porites would be very valuable in assessing the approximate age of the sample. Presumably this data would be retrievable for the colony data sets as these individual corals have been collected in one piece. However for
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
short core and long core data, it may be necessary to attempt to find the original coral to measure its dimensions. In summary, the reanalysis of the calcification data presented in De'ath et al. (2009) suggests that the 14% reduction in the Porites calcification rate in the last two decades has been overestimated due to a combination of (a) ontogenetic effects combined with a dramatically changing age distribution of the coral growth bands with time, and (b) systematic data bias in the last growth band of the cores. When the outermost growth bands in addition to bands with record age less than 20 years were excluded from the analysis, the dramatic fall in calcification post 1990 was no longer evident. Whilst large data sets of calcification rates of massive corals are an excellent archive of environmental effects on reef building organisms, currently available data of coral calcification in the GBR cannot reliably resolve a systematic dramatic reduction in calcification in recent years. In order to assess potential environmental effects on calcification in the future, it is imperative to continue the collection of calcification data, and to continue research into ontogenetic growth effects in order to obtain a better understanding of growth patterns, both spatially and temporally. Possibly the most important outcome of this re-analysis is that, for future work, care must be taken to ensure that the age distribution of the coral layers that contribute data to a particular year should remain constant as far as it's possible. In addition it is possible that the young corals may be more susceptible to environmental changes and the focus of future work should be collection of small colonies where it can be guaranteed that the early years of growth can be recovered from the measurements. Evidently, the most convincing proof of either analysis (De'ath et al., 2009 or herein) would be to carefully analyse calcification rates in the years 1990–2005 measured in coral cores sampled today.
399
Acknowledgements We thank G De'ath and J Lough (Australian Institute of Marine Science) who kindly made the coral growth source available for reanalysis. Also thanks to Mel Boer who assisted with the figures. References Anthony, K.R.N., Connolly, S.R., Willis, B.L., 2002. Comparative analysis of energy allocation to tissue and skeletal growth in corals. Limnology and Oceanography 47, 1417–1429. Buddemeier, R., Kinzie, R., 1976. Coral growth. Oceanography and Marine Biology: An Annual Review 183–225. Carpenter, K.E., Abrar, M., Aeby, G., Aronson, R.B., Banks, S., Bruckner, A., Chiriboga, A., Cortés, J., Delbeek, J.C., DeVantier, L., Edgar, G.J., Edwards, A.J., Fenner, D., Guzmán, H.M., Hoeksema, B.W., Hodgson, G., Johan, O., Licuanan, W.Y., Livingstone, S.R., Lovell, E.R., Moore, J.A., Obura, D.O., Ochavillo, D., Polidoro, B.A., Precht, W.F., Quibilan, M.C., Reboton, C., Richards, Z.T., Rogers, A.D., Sanciangco, J., Sheppard, A., Sheppard, C., Smith, J., Stuart, S., Turak, E., Veron, J.E.N., Wallace, C., Weil, E., Wood, E., 2008. One-third of reef-building corals face elevated extinction risk from climate change and local impacts. Science 321, 560–563. Cooper, T.F., O'Leary, R.A., Lough, J.M., 2012. Growth of Western Australian corals in the Anthroppocene. Science 335, 593–596. De'ath, G., Lough, J.M., Fabricius, K.E., 2009. Declining coral calcification on the Great Barrier Reef. Science 323, 116–119. Hoegh-Guldberg, O., Mumby, P.J., Hooten, A.J., Steneck, R.S., Greenfield, P., Gomez, E., Harvell, C.D., Sale, P.F., Edwards, A.J., Caldeira, K., Knowlton, N., Eakin, C.M., IglesiasPrieto, R., Muthiga, N., Bradbury, R.H., Dubi, A., Hatziolos, M.E., 2007. Coral reefs under rapid climate change and ocean acidification. Science 318, 1737–1742. Holmes, G., Johnstone, R.W., 2010. Modelling coral reef ecosystems with limited observational data. Ecological Modelling 221, 1173–1183. Langmeada, O., Sheppard, C., 2004. Coral reef community dynamics and disturbance: a simulation model. Ecological Modelling 175, 271–290. Lough, J.M., 2008. Coral calcification from skeletal records revisited. Marine Ecology Progress Series 373, 257–264. Lough, J.M., Barnes, D.J., 2000. Environmental controls on growth of the massive coral Porites. Journal of Experimental Marine Biology and Ecology 245, 225–243. Team, R.D.C., 2009. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria.
Contents lists available at ScienceDirect
Marine Geology journal homepage: www.elsevier.com/locate/margeo
Have coral calcification rates slowed in the last twenty years? Peter V Ridd a,⁎, Eduardo Teixeira da Silva a, Thomas Stieglitz a,b a b
Marine Geophysical Laboratory, School of Engineering and Physical Science, James Cook University, Townsville 4811, Australia Laboratoire des sciences de l'environnement marin CNRS UMR 6539, Institut Universitaire Européen de la Mer, 29280 Plouzané, France
a r t i c l e
i n f o
Article history: Received 14 February 2012 Received in revised form 30 July 2013 Accepted 3 September 2013 Available online 14 September 2013 Communicated by G.J. de Lange Keywords: coral calcification Great Barrier Reef ocean acidification ocean pH
a b s t r a c t This paper reports a reanalysis of calcification rates of 328 Porites cores from the Great Barrier Reef from which previous workers have concluded that a 14% reduction in calcification rates has occurred between 1990 and 2005. In this reanalysis it is shown that the apparent reduction in the Porites spp. calcification rate in the last two decades is at least partly due to a combination of (a) ontogenetic effects (disregarded in the previous analysis), combined with a highly variable age distribution of the coral growth bands with time, and (b) a systematic data bias clearly evident in the last growth band of each core. When the outermost growth band in addition to bands which have record age less than 20 years was excluded from the analysis, the dramatic fall in calcification after 1990 was no longer evident. © 2013 Elsevier B.V. All rights reserved.
1. Introduction There is widespread concern that a hitherto unperceived consequence of global carbon dioxide emissions is a decrease in ocean pH which will have dire consequences for the calcification of calcareous marine organisms such as corals. A recent analysis (De'ath et al., 2009) of coral calcification data extracted from 328 Porites corals collected from 69 reefs over the Great Barrier Reef (GBR) that span the last 400 years has indicated that there has been a dramatic decline in coral calcification by as much as 14% between 1990 and 2005. It was suggested that a tipping point was reached in 1990 when declining ocean pH due to increased atmospheric CO 2 combined with increasing temperature stress caused rapid reductions in calcification. The 14% decline in calcification rate between 1990 and 2005 (De'ath et al., 2009) is prima facie a surprising result because a previous comprehensive study (Lough and Barnes, 2000), using a subset of the data used in De'ath et al. (2009), demonstrated a statistically significant 4% increase in GBR coral growth over the 20th century. In addition, it is notable that a more recent paper on calcification rates on Australia's north western coastline does not indicate any significant decline in calcification rates after 1990 (Cooper et al., 2012). However, laboratory experiments show that calcification decreases under increasing pH for a variety of reef organisms (Hoegh-Guldberg et al., 2007), suggesting
⁎ Corresponding author. Tel.: +61 747814978; fax: +61 747815880. E-mail address: [email protected] (P.V. Ridd). 0025-3227/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.margeo.2013.09.002
that modern coral reefs may be facing major challenges due to environmental change (Carpenter et al., 2008). Clearly the precise nature of the trend in coral calcification is of considerable importance for scientists, managers and policy makers alike. The study of De'ath et al. (2009) used a linear mixed effect (LME) model, an analysis technique that aims to compensate for the many challenges involved with analysing complex data such as coral growth time series. The different number of corals from each location and a large latitudinal sampling range, which implies varying environmental factors leading to variable calcification rates, are examples of confounding variables that can be accounted for by LME. However there are additional important aspects of the data set that were not taken into account in the analysis of De'ath et al. (2009). These include (a) the use of three different types of coral samples (long cores, short cores and colonies) which produce samples of different lengths and therefore record ages (record age is defined as the time between a particular yearly calcification record and the first data record in the series), (b) a strong temporal variation in the average age of the corals over the sampling period, and (c) the likelihood of systematic sampling problems. These three confounding variables intrinsic to the data set are discussed in this paper. At the centre of the analysis by De'ath et al. (2009) is the assumption that calcification rates for a particular coral do not change with the age of the coral (if environmental conditions remain constant), i.e., there are no ontogenetic effects. Here we reanalyse the coral calcification data from the Great Barrier Reef and show that the apparent decline over the last two decades may be the result of a combination of ontogenetic effects and measurement artefacts
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
rather than an abrupt calcification reduction caused by environmental conditions. 2. Material & methods 2.1. Data The coral calcification data of 328 Porites spp. from the Great Barrier Reef (GBR) presented in De'ath et al. (2009) can be subdivided into 3 broad corals categories: 61 long cores, 135 short cores and 132 colonies. Long cores were collected using a large drill rig coring device, from a wide range of locations, and their time span is on average 155 years. Short cores were collected using a small drill, mostly from 5 reefs in the central section of the GBR between 18° and 20°S, and their time span is on average 28 years. Colonies were sampled by taking a saw slice through the coral, and they have on average a time span of 23 years (Fig. 1). Before 1940, only long core data contributes to the overall calcification rate and most of the long core data ends by 1990. Short core data starts around 1940 and mostly ends around 2005. Colonies also start to contribute data around 1940 and roughly 2/3rds of the colony records terminate around 1990. The other third are generally very short records which end around 2005 (Fig. 1).
393
Also evident at ca. 1990 is the discontinuity of records especially in the colony data (Fig. 1b). The discontinuity of records at 1990 occurs because the field work when the coral cores were recovered cover two distinct periods: (a) the late 1980's and early 1990's, and (b) the early part of the 2000's, especially in 2003, 2004 and 2005. The short core data (Fig. 1c) does not show a discontinuity at around 1990 as this sampling technique was only used in the later sampling period. For this reason, all the short core data ends on or after 2003. The number of contributing corals changes dramatically with time (Fig. 2). Before 1940 the total number of corals was around 50 and rose to 269 in the early 1980's before declining to 189 in 1999, and then sharply reducing to only 21 in 2005. 2.2. Data analysis methods LME analysis for the calcification data was performed in R scripting language (Team, R.D.C., 2009), with the package mgcv, using exactly the same R code as provided to us by G. De'ath (pers. comm.). When running this code it was necessary to select the k value of the gamm function, which gives the dimension of the basis used to represent the smooth term in the spline model (Team, R.D.C., 2009). The smaller value of k (fewer degrees of freedom) causes some of the detailed temporal changes in the calcification curve to be masked. In short, a
Long Cores
a Number of corals
121 101 81 61 41 21 1 1900
1910
1920
1930
1940
1960
1970
1980
1990
2000
2010
1960
1970
1980
1990
2000
2010
1960
1970
1980
1990
2000
2010
Colonies
b Number of corals
1950
121 101 81 61 41 21 1
1900
1910
1920
1930
c
1940
1950
Short Cores
Number of corals
121 101 81 61 41 21 1 1900
1910
1920
1930
1940
1950
Fig. 1. Time span of each coral record included in the reanalysis. Each horizontal bar represents the time period over which a particular coral contributed to the data set for (a) long cores, (b) colonies, and (c) short cores. Dark Grey represents coral growth bands with record age N20 years. Light grey represents the growth band of corals with record age ≤20 years. Black represents the last growth bands which have record age N20 years.
394
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
Fig. 2. Temporal variation of the (a) mean record age of the corals and (b) number of contributing corals, for all the corals together and for colonies, long cores and short cores individually. Red lines represent the cases for growth bands with record age N20, 10 and 5 years when the outer growth bands have also been removed.
higher k value will generally make the curve more “wiggly”. Following De'ath et al. (2009), in this paper we adopted the AIC (Akaike Information Criterion) value to select the k value, choosing the lowest but conservative AIC for each fitted model. In order to investigate potential problems with data from the outermost growth band, a simple data normalisation procedure, which was similar to that used by Lough (2008) was adopted. The normalised calcification rate in year t for the mth coral is calculated by: norm
Cm
ðt Þ ¼
C m ðt Þ Cm
;
ð1Þ
where Cm(t) is the raw calcification rate of the coral growth band in year t and C m is the average calcification rate of all its growth bands
(i.e., over the entire time span of the mth coral). Then, the aggregate normalised calcification rate for year t, was calculated by: Xi C ðt Þ ¼
norm C 0 ðt Þ i0 ¼1 i
i
ð2Þ
where i is the number of corals contributing to the data in year t, and the summation is over the i contributing corals in year t. This procedure was used to look at the last band of the series that ended in 2004 and 2005. These two years were chosen because of the large numbers of corals which terminated in these years. It should be noted that this normalisation procedure was not used with the LME analysis which used the raw calcification data.
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
2.3. Definition of terminology
395
remains sufficiently large so that uncertainties in the results do not dominate.
In this paper, we adopted the term record age to describe the time between a particular yearly calcification record and the first data record in the series (i.e. within one core). It should not be confused with the ‘true’ age of the coral which is the total time period that it lived before being collected. True age could be considerably greater than record age, if the core taken from a coral did not include its initial growth point. Measurements of the physical size of the corals which could be used to estimate true age are not available for this data set. The time span of a coral is defined as the period of time from which calcification data was available for that coral. To demonstrate these different parameters, consider a coral colony that started growing in 1910, was collected in 2005, and from which calcification data is available from 1930 to 2004. The final true age of the colony is 95 years but this is generally not known because the date of initiation of growth is unknown. The time span of the record is 75 years, and the record age varies from 1 year for the 1930 record to 75 years for the 2004 record. 3. Results The basic approach in this paper was to repeat the analysis with subsets of the original data after the exclusion of the following two classes of data. Firstly, data from the outermost growth band of each individual core is disregarded; because it is apparent that this data has a systematic bias, most likely due to instrumental artefacts. Secondly, growth bands with record age less than (a) 5 years, (b) 10 years and (c) 20 years were disregarded i.e., removal of data from corals when they are relatively young. The rationale of this exercise is to remove growth bands (in this case the younger bands) which are more likely to experience an ontogenetic variation in growth rate. If the analysis of De'ath et al. (2009) is robust, the removal of both these classes of data should make little difference to the result. In the following sections we explore these two scenarios and show how they can affect the original data interpretation presented in De'ath et al. (2009). It is worthwhile to mention that the analysis herein was carried out using statistical methodology and data provided to us by G. De'ath, and is identical to that used by De'ath et al. (2009). Thus results from both analyses are directly comparable. The data set size (i.e. numbers of corals and annual growth bands) is sufficiently large to allow testing of results using subsets of the data. If the original analysis is sound, it should be expected that a similar result should be found from a subset of the data provided that the subset
3.1. Systematic low bias in calcification rate of the last growth band Because a large number of calcification time series ends in 2004 and 2005, a detailed analysis of the bias in the outer growth band can be made on these two sets of corals. Fig. 3 shows the aggregated calcification rate (using Eq. (2)) for coral series ending in 2004 and 2005. Both series show a considerable decline in calcification rate in the last growth band. A One-way ANOVA test indicates that the average aggregated calcification rates for the final year on both series are significantly different from the previous years (F(24,1634) = 4.923, p b 0.001 and F(25,490) = 2.06, numbers between brackets stand for degrees of freedom and number of data points, respectively, p = 0.002 for 2004 and 2005 series, respectively). Importantly, the series ending in 2005 did not show any significant decline in 2004 — the year in which the series that ended in 2004 did show a significant reduction. If there was no data bias in the outer bands, the conclusion must be drawn that in 2004, the corals of the 2005 series somehow avoided the environmental conditions that so affected the 2004 series. It is difficult to see how this could occur. Instead, it is highly likely that the last annual growth bands have a systematic low bias, perhaps due to difficulty in identifying the extension of this band. It should be noted that this analysis of potential problems with the last growth band has concentrated on the corals that end in 2004 and 2005. A very large number of corals are in this category, and thus a useful comparison can be made due to relatively small error margins. Similar problems may also occur with data series ending in other years, but due to a smaller numbers of corals producing higher error margins, potential problems in the last band of these series may be more difficult to detect. In order to be consistent, all of the last growth bands were removed in the analysis shown in the next section, not just the last bands of series ending in 2004 and 2005. To demonstrate the considerable influence of the bias due to these last growth bands, the LME model was used both with and without the final growth bands included (Fig. 4) (note: the final bands of all corals were removed in Fig. 4b, not just the final growth bands for the series ending in 2004 and 2005). It is evident that the fall in the calcification rate at the end of the series is considerably reduced when the last band is removed. Using the LME model of De'ath et al. (2009) the
Aggregated calcification rate (g/cm2/y)
1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 1979
2004 series 2005 series 1984
1989
1994
1999
2004
Year Fig. 3. Annual average aggregated calcification rate calculated using Eq. (2) for corals with end band corresponding to 2004 (opened circle, dashed line) and 2005 (opened square, solid line). To make error bars distinguishable, data points were displaced −0.1 and +0.1 of the year for the 2004 and 2005 series, respectively. Vertical bars indicate 95% confidence intervals.
396
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
Fig. 4. Temporal variation of the calcification rate (g cm−2 y−1) obtained using the R code as provided by G. De'ath, with k = 11, for (a) all the data together, and (b) when the outermost growth band was excluded (bold solid line with bold dotted lines as error limits).
Fig. 5. Temporal variation of the calcification rate obtained for (a) all the data together as presented in De'ath et al. (2009), and when coral growth bands younger than (b) 5 years, (c) 10 years, or (d) 20 years are removed. The data set also does not include the outermost growth band for the cases b, c and d. Dotted lines indicates the 95% confidence intervals for comparison between years. The number of corals used for each of the four cases is shown in Fig. 2b.
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
removal of this data alone resulted in a reduction in the post 1990 fall from 14% to ca 9% (Fig. 4). 3.2. Ontogenetic effect When the growth bands of corals with record age less than 5 years, 10 years and 20 years were removed from the analysis (together with the outermost growth band), the post 1990 drop in the calcification rate was reduced from 14% to 8%, 6% and 1%, respectively (Fig. 5). The post 1990 reduction in calcification is thus no longer evident once the uncertainties in the calculation are taken into account (wide error bars in Fig. 5). Notably, the calcification rate at the final year of the time series is unmeasurably different from the calcification rates of the early 1900's, even when only the bands representing the youngest 5 years of growth are removed, and uncertainties are taken into account (Fig. 5b). For all the fitted models, the selection of the k-value was based on the lowest AIC. For Fig. 5a we used k = 10, as used in De'ath et al. (2009), even though k = 11 produces the lowest AIC. For Fig. 5b and c, k = 11 produced the lowest AIC. For Fig. 5d, the lowest AIC value (i.e., 8273) corresponded to a k = 5 which was marginally lower than the AIC value (8275) for k = 11. However for comparability with Fig. 5b and c, the value of k for Fig. 5d was set to 11. Use of different k values (either 11 or 5) does not affect the conclusion that the removal of data with record age less than 20 years (plus the final record of each coral) removes the post 1990 fall in calcification. Fig. 6 shows the calcification time series for k = 5, in which the post 1990 fall is not evident. All the fitted models exhibited a p b 0.001. 4. Discussion The considerable change in the result due to the exclusion of the youngest growth bands of each coral record is concerning. It is likely to be due to a combination of two factors: (a) a slowing of growth rate with age of a coral, and (b) the record age distribution of the coral growth bands changes with time. These factors are discussed in detail as follows. Whilst De'ath et al. (2009) assume absence of any ontogenetic effects on coral calcification, the presumption of constant growth rates of Porites with age is not established in the scientific literature. Indeed many authors allude to the possibility of significant ontogenetic variations (Buddemeier and Kinzie, 1976; Anthony et al., 2002;
397
Langmeada and Sheppard, 2004; Holmes and Johnstone, 2010). In particular Holmes & Johnstone (Holmes and Johnstone, 2010) state that small colonies (b 20 cm diameter) can grow considerably faster than large colonies (N50 cm diameter). Porites exhibit an average growth rate of 1.4 cm y−1 (De'ath et al., 2009), which suggest that they may experience ontogenetic variations up to 30 years old. An ontogenetic reduction of the calcification rate would not affect the data analysis if the age distribution of the coral data (i.e., record age of growth bands) does not change with time. However, as happens in this data set, if younger corals are included in the latter part of the data set, slowing growth curves of these corals would result in a reduction in the aggregated calcification rate. In this data set, the number of coral growth bands and their average record age varies dramatically with time (Fig. 2). The average record age of growth bands contributing to each year in the time series exhibits a clear reduction after 1940, with the inclusion of short cores and colonies into the data set, falling from around 110 years in 1940 to around 30 years towards the end of the sampling period (Fig. 2a). The relatively short record age of the colony data is particularly evident when compared with the long core and short core data. For example in 2003, the average colony age was only 14.2 years compared with 143 years and 27.9 years for long and short core data respectively (Fig. 2a). This dramatic change in the average colony age means that any ontogenetic effect on calcification rate could dominate the analysis. The change in result after removal of young growth bands is strong evidence that there is an ontogenetic artefact in the data set. The fact that there is any change in result by changing the age structure of the data is by definition proof that the age of the coral is important. In contrast De'ath et al. (2009) have stated that an ontogenetic effect associated with the outermost annual growth bands is unlikely to exist. They point out that of the 189 corals (from 13 reefs) in the period from 1990 to 2005, 12 reefs (92%) and 137 corals (72.5%), i.e., a large majority of reefs and corals, showed a negative trend in calcification rate. For the 139 corals (from 56 reefs) sampled before 1990, only 29 reefs (52%) and 68 corals (49%) showed a decline in calcification rate, i.e., an equal number of reefs and corals increased or decreased calcification rate. This however does not provide much insight into ontogenetic effects as the coral samples that ended around 1990 are disproportionately older than those that end in the early part of this century (Fig. 2a), and thus may be displaying considerably less slowing of
Fig. 6. Temporal variation of the calcification rate obtained when corals growth bands b20 years are removed in addition to the outermost growth band. (a) k = 11, (b) k = 5.
398
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
growth rate. For example almost all long cores ended around 1990 at an average record age of 150 years (Fig. 2). Post 1990, without the contribution of the long cores, the average record age drops to less than 30 years at the end of the sampling series. It should be noted that the problems posed by the ontogenetic effects do not prove that there is no environmental effect that may be causing decreased (or increased) calcification, but it makes it considerably more challenging to extract the environmental signal from the ontogenetic artefact. In early correspondence about his paper, G. De'ath (pers. comm.) raised a strong argument that there may be a general environmental signal at play in addition to the ontogenetic effect, i.e., they demonstrated that the colonies which start before 1985 have slow calcification at a reduced rate than to those colonies which start after 1985. In our analysis above we show that excluding all the young layers causes the post 1990 drop to disappear, i.e., the older corals are not experiencing a calcification decline. However it is possible that only the younger corals are susceptible to the potentially changing environmental conditions. To shed further light on the possibility that the younger corals are showing a calcification reduction after 1990 we have plotted the calcification rate determined only from corals that have record ages less than 20 years (see Fig. 7) which shows a clear, though, modest (between 4% and 11%), reduction in calcification rate after 1990. There is no measurable difference between the calcification rates in 2004 compared with 1960 once the error margins are considered. It should be noted that the data before 1950 is of little value because of high errors and the fact that most of the growth bands considered are likely to be from corals much older than 20 years. This is because they are taken from long cores and short cores where there is no guarantee that the core has passed through the oldest part of the coral. Fig. 7 uses only that data which was excluded from Fig. 5(d) (except the outer layer which is excluded in both cases) and the result is very different. Fig. 7 shows a drop in calcification in the last 15 years and Fig. 5(d) shows no measureable change. This reaffirms our proposition that there is an ontogenetic effect, and it indicates that there may be a modest change in calcification in the last 15 years in the younger corals. One potential problem with this conclusion is that even Fig. 7 has a changing age distribution with time which may influence the results.
4.1. Is using data subsets a valid method to show that the post 1990 decline may be a data artefact? An important question that must be asked about the removal of data from the analysis is: what effect does data removal have on the error of the analysis? Fig. 5 shows that the error in the analysis degrades only marginally with the exclusion of data. This error estimate was calculated with the identical methodology as used by De'ath et al. (2009). The crucial parameter influencing the error estimate is the number of corals and reefs that contributes to the data in a given year. For the whole data set, with the exception of the last two years, the smallest numbers of corals sampled occur in the pre 1940 period (Fig. 2b) not in the post 1990 period which is of most interest here. It is important to note that when the growth bands younger than 20 years are disregarded, the number of contributing corals during the post 1990 period, generally still exceeds the number of corals contributing to the period up to 1960 (Fig. 2b, red line). Thus it cannot be argued that so much data was removed that the post 1990 result would become invalid. Moreover, the greatest number of bands was removed from data between 1970 and 1990, not the post-1990 period where the reduction in calcification occurs (Fig. 2b, difference between red and black curves). It might be argued that we have thrown away most of the data used to detect the change, however as mentioned above, the majority of the data removed was from the pre-1990 period where for much of the time calcification increased (Fig. 4). If the conclusion about the sudden post1990 decline is to be accepted, there needs to be a good explanation of why removing the younger bands from all of the corals should make such a large difference to the result. 4.2. Can the problems with the data be overcome by modelling the ontogenetic effect and the end-of-series problem? By looking at different classes of data it has been shown that there is considerable doubt that the fall in calcification after 1990 is a real signal or at least partly an artefact of the changing age distribution of the corals and the end-of-series problem. To overcome these problems, one approach could be to attempt to model both the end of series problem and the ontogenetic effect. For example, if the end of series problem is due to an instrumental problem that could be measured, this could be modelling and the analysis corrected. Alternatively, the end of series problem could have been due to a change in procedure over the 20 years that these corals were analysed. If this effect could be measured, a correction could be applied to the data. The ontogenetic effect could be corrected if more information became available about the change in calcification rates as the coral grows with all other factors remaining fixed. However modelling away the problems with the data is fraught with danger if the physical or biological cause of the problem cannot be clearly identified and quantified. Great care would certainly be needed if a purely statistical reanalysis is used to correct the problem as more variables will now be required to contribute to the model reducing its reliability. It would be particularly difficult to correct the ontogenetic effect by a purely statistical reanalysis because the true age of the coral is not known but would be required in such a reanalysis. Any statistical approach to correct for the ontogenetic effect would need to be carefully scrutinised. 5. Conclusion
Fig. 7. Temporal variation of the calcification rate obtained for all data with record age less than 20 years, i.e., it uses the data that was excluded in Fig. 5(d) except the data from the outermost growth band which is similarly excluded. Dotted lines indicates the 95% confidence intervals for comparison between years, in the model, k = 9, n = 4960 and p b 0.01.
Despite the caveats discussed in this paper, the data set analysed by De'ath et al. (2009) is a highly valuable resource of coral calcification rates. Some of the problems associated with ontogenetic effects could be addressed if more information about the true age of the coral was available. The approximate dimension of the individual Porites would be very valuable in assessing the approximate age of the sample. Presumably this data would be retrievable for the colony data sets as these individual corals have been collected in one piece. However for
P.V. Ridd et al. / Marine Geology 346 (2013) 392–399
short core and long core data, it may be necessary to attempt to find the original coral to measure its dimensions. In summary, the reanalysis of the calcification data presented in De'ath et al. (2009) suggests that the 14% reduction in the Porites calcification rate in the last two decades has been overestimated due to a combination of (a) ontogenetic effects combined with a dramatically changing age distribution of the coral growth bands with time, and (b) systematic data bias in the last growth band of the cores. When the outermost growth bands in addition to bands with record age less than 20 years were excluded from the analysis, the dramatic fall in calcification post 1990 was no longer evident. Whilst large data sets of calcification rates of massive corals are an excellent archive of environmental effects on reef building organisms, currently available data of coral calcification in the GBR cannot reliably resolve a systematic dramatic reduction in calcification in recent years. In order to assess potential environmental effects on calcification in the future, it is imperative to continue the collection of calcification data, and to continue research into ontogenetic growth effects in order to obtain a better understanding of growth patterns, both spatially and temporally. Possibly the most important outcome of this re-analysis is that, for future work, care must be taken to ensure that the age distribution of the coral layers that contribute data to a particular year should remain constant as far as it's possible. In addition it is possible that the young corals may be more susceptible to environmental changes and the focus of future work should be collection of small colonies where it can be guaranteed that the early years of growth can be recovered from the measurements. Evidently, the most convincing proof of either analysis (De'ath et al., 2009 or herein) would be to carefully analyse calcification rates in the years 1990–2005 measured in coral cores sampled today.
399
Acknowledgements We thank G De'ath and J Lough (Australian Institute of Marine Science) who kindly made the coral growth source available for reanalysis. Also thanks to Mel Boer who assisted with the figures. References Anthony, K.R.N., Connolly, S.R., Willis, B.L., 2002. Comparative analysis of energy allocation to tissue and skeletal growth in corals. Limnology and Oceanography 47, 1417–1429. Buddemeier, R., Kinzie, R., 1976. Coral growth. Oceanography and Marine Biology: An Annual Review 183–225. Carpenter, K.E., Abrar, M., Aeby, G., Aronson, R.B., Banks, S., Bruckner, A., Chiriboga, A., Cortés, J., Delbeek, J.C., DeVantier, L., Edgar, G.J., Edwards, A.J., Fenner, D., Guzmán, H.M., Hoeksema, B.W., Hodgson, G., Johan, O., Licuanan, W.Y., Livingstone, S.R., Lovell, E.R., Moore, J.A., Obura, D.O., Ochavillo, D., Polidoro, B.A., Precht, W.F., Quibilan, M.C., Reboton, C., Richards, Z.T., Rogers, A.D., Sanciangco, J., Sheppard, A., Sheppard, C., Smith, J., Stuart, S., Turak, E., Veron, J.E.N., Wallace, C., Weil, E., Wood, E., 2008. One-third of reef-building corals face elevated extinction risk from climate change and local impacts. Science 321, 560–563. Cooper, T.F., O'Leary, R.A., Lough, J.M., 2012. Growth of Western Australian corals in the Anthroppocene. Science 335, 593–596. De'ath, G., Lough, J.M., Fabricius, K.E., 2009. Declining coral calcification on the Great Barrier Reef. Science 323, 116–119. Hoegh-Guldberg, O., Mumby, P.J., Hooten, A.J., Steneck, R.S., Greenfield, P., Gomez, E., Harvell, C.D., Sale, P.F., Edwards, A.J., Caldeira, K., Knowlton, N., Eakin, C.M., IglesiasPrieto, R., Muthiga, N., Bradbury, R.H., Dubi, A., Hatziolos, M.E., 2007. Coral reefs under rapid climate change and ocean acidification. Science 318, 1737–1742. Holmes, G., Johnstone, R.W., 2010. Modelling coral reef ecosystems with limited observational data. Ecological Modelling 221, 1173–1183. Langmeada, O., Sheppard, C., 2004. Coral reef community dynamics and disturbance: a simulation model. Ecological Modelling 175, 271–290. Lough, J.M., 2008. Coral calcification from skeletal records revisited. Marine Ecology Progress Series 373, 257–264. Lough, J.M., Barnes, D.J., 2000. Environmental controls on growth of the massive coral Porites. Journal of Experimental Marine Biology and Ecology 245, 225–243. Team, R.D.C., 2009. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria.