Coral Reefs

, Volume 32, Issue 1, pp 1–12

Relationships between temperature, bleaching and white syndrome on the Great Barrier Reef

Authors

    • ARC Centre of Excellence for Coral Reef StudiesJames Cook University
  • N. A. J. Graham
    • ARC Centre of Excellence for Coral Reef StudiesJames Cook University
  • S. R. Connolly
    • ARC Centre of Excellence for Coral Reef StudiesJames Cook University
    • School of Marine and Tropical BiologyJames Cook University
Report

DOI: 10.1007/s00338-012-0944-6

Cite this article as:
Ban, S.S., Graham, N.A.J. & Connolly, S.R. Coral Reefs (2013) 32: 1. doi:10.1007/s00338-012-0944-6

Abstract

Coral bleaching and disease have often been hypothesized to be mutually reinforcing or co-occurring, but much of the research supporting this has only drawn an implicit connection through common environmental predictors. In this study, we examine whether an explicit relationship between white syndrome and bleaching exists using assemblage-level monitoring data from up to 112 sites on reef slopes spread throughout the Great Barrier Reef over 11 years of monitoring. None of the temperature metrics commonly used to predict mass bleaching performed strongly when applied to these data. Furthermore, the inclusion of bleaching as a predictor did not improve model skill over baseline models for predicting white syndrome. Similarly, the inclusion of white syndrome as a predictor did not improve models of bleaching. Evidence for spatial co-occurrence of bleaching and white syndrome at the assemblage level in this data set was also very weak. These results suggest the hypothesized relationship between bleaching and disease events may be weaker than previously thought, and more likely to be driven by common responses to environmental stressors, rather than directly facilitating one another.

Keywords

Coral reef ecologyMultiple stressorsSynergyResilienceCoral bleachingCoral disease

Introduction

Many studies have posited a relationship between two of the most prevalent causes of large, episodic declines in coral cover: coral bleaching and coral diseases (e.g., Brandt and McManus 2009; Croquer and Weil 2009). Understanding the strength and causal direction of any such relationship is important because increasing ocean temperatures are expected to lead to more frequent and extensive coral bleaching episodes (Nicholls et al. 2007), and possibly also to increased frequency and intensity of coral disease outbreaks through factors such as increased pathogen growth rates with warmer ocean temperatures and increased host susceptibility due to environmental stress (Sokolow 2009; Mydlarz et al. 2010). Thus, if the two kinds of events are self-reinforcing, any projections of the effects of coral cover that do not account for such synergies may underestimate likely declines in coral cover. However, there are few comprehensive assessments of interactions between bleaching and disease on regional scales, in comparable habitats and over long time periods.

Bleaching and disease events may co-occur simply because they are responses to common environmental stresses, such as temperature. However, several mechanisms for a direct causal relationship between bleaching and disease have been suggested. For example, according to the microbial hypothesis of coral bleaching (Ben-Haim et al. 2003; Rosenberg et al. 2009), bleaching is pathogenically induced and reduces the coral host’s ability to defend against other infections. Alternatively, according to the coral probiotic hypothesis (Reshef et al. 2006), the symbiotic bacterial community associated with corals is disrupted during bleaching, or portions of the symbiotic bacterial community normally residing in the gastrodermis penetrate the coral epithelial layer (Ainsworth and Hoegh-Guldberg 2009), increasing the host’s vulnerability to infection by reducing the competitive exclusion of pathogens. Finally, bleaching has been proposed to compromise immune competence (Banin et al. 2003; Mydlarz et al. 2009), by reducing protective enzyme [e.g., prophenoloxidase (PPO)] activity.

All of the above hypotheses suggest that disease may be more likely to occur coincident with, or in the aftermath of, bleaching events, but there are few instances where the possibility of a direct causal link has been tested explicitly. Indeed, the pattern of co-occurrence of coral bleaching and disease outbreaks, although frequently hypothesized to exist, is poorly documented. Much of the evidence for a link between bleaching and disease has been qualitative or anecdotal, and the lag at which disease outbreaks have been proposed to follow bleaching events has ranged from several months (Miller et al. 2006; Brandt and McManus 2009) to over a year (Mydlarz et al. 2009). Some studies have noted increased mortality relative to bleaching-only episodes when bleaching and disease co-occur (Harvell et al. 2001; Miller et al. 2006), but few to date have explicitly tested the hypothesis that the occurrence of disease is affected by bleaching (or vice versa). For instance, previous work (Bruno et al. 2007; Heron et al. 2010; Maynard et al. 2011) on the relationship between bleaching and white syndrome disease outbreaks has used common environmental predictors of bleaching and disease (specifically, temperature anomalies), rather than explicitly testing whether, under a given set of environmental conditions, disease outbreaks were more likely when bleaching had previously occurred. Thus, the available evidence does not allow us to distinguish between the possibility that bleaching and disease events share common physical environmental drivers, and the possibility that the occurrence of bleaching events makes disease outbreaks more likely (or vice versa), as has been hypothesized.

Of the environmental variables that have been used to predict both bleaching and disease, temperature appears to be the strongest and most common. However, other potential common environmental predictors include nutrient loading and pollution (Hayes and Goreau 1998; Wooldridge 2007), high irradiance (Boyett et al. 2007; Richier et al. 2008; Muller and van Woesik 2009), and sedimentation (Anthony et al. 2007; Harvell et al. 2007). Predictive models for mass bleaching are well-established for shallow reef flat habitats (Goreau and Hayes 1994; Gleeson and Strong 1995; Lough 2000; McClanahan et al. 2007; Maina et al. 2008; Donner 2011), but reef slope bleaching models are far less common (Glynn 1996). This is important because even during mass bleaching events, incidence and severity vary widely between inshore and offshore areas, and between different reef zones (Berkelmans and Oliver 1999). Because most coral species that occupy reef flats also occupy other habitats, the extent to which coral assemblages in habitats such as the shallow reef slope respond similar to the environmental stressors like temperature has important implications for the likely capacity for heavily bleached habitats to recover.

To date, the most successful models for predicting bleaching use multiple temperature metrics, including duration, rate, and magnitude of temperature anomalies (Maynard et al. 2008). In contrast, predictive models for disease are relatively recent and still being refined. Although several coral diseases have been hypothesized to have a link with temperature stress, white syndrome (WS) has been the focus of most research in this regard. In the Indo-Pacific context, WS describes a variety of conditions (including white pox, white band, and white plague) with similar symptoms, the cause(s) of which are still unknown (Willis et al. 2004). Models used to predict WS have incorporated at least one temperature metric, such as weekly sea surface temperature anomaly (WSSTA) (Bruno et al. 2007), Mean Positive Summer Anomaly (MPSA) (Maynard et al. 2011), and Hot Snap (Heron et al. 2010), in addition to a measure of coral cover. All of these temperature metrics measure short-term (intra-annual) deviations from a climatological mean value, but differ in how the magnitude and duration of those deviations are integrated to generate a cumulative predictor of risk. These models have shown remarkable ability to hindcast the occurrence of white syndrome outbreaks, but the definition of an outbreak has varied in each study (e.g., Heron et al. (2010)—50 cases per 1,500 m2; Maynard et al. (2011)—60 cases per 1,500 m2 with a “severe” outbreak constituting 100 cases per 1,500 m2).

In this paper, we provide a comprehensive, regional-scale assessment of whether observations of coral bleaching and disease confer any additional mutual predictability over and above common environmental drivers, using a long-term monitoring data set of the reef slope in the Great Barrier Reef region, Australia. This data set assesses both bleaching and disease occurrence at the assemblage level using a consistent depth, reef aspect, and methodology. Specifically, we (1) examine whether traditional physical environmental predictors of mass bleaching in shallow water reef flat habitats are also useful for predicting bleaching events on the reef slope; (2) test the utility of established predictors of white syndrome for these coral assemblages; (3) determine whether, once common physical drivers are accounted for, the occurrence of bleaching is an informative predictor of white syndrome (and vice versa) at the assemblage level; and (4) examine the spatial patterns of white syndrome and bleaching events at the assemblage level for evidence of overlap or spatial clustering. We focus on white syndrome specifically to facilitate comparison with previous studies (e.g., Bruno et al. 2007; Heron et al. 2010; Maynard et al. 2011), and because it is the most commonly reported disease.

Methods

Bleaching and disease surveys

We used observations of bleaching and disease from the Australian Institute of Marine Science (AIMS) Long-Term Monitoring Program (LTMP) and Representative Areas Program (RAP). The total number of reefs visited in a given year ranged from a minimum of 27 in 2003 to a maximum of 112 in 2006. At each reef, five 2-m wide, 50-m long fixed transects were surveyed at each of three sites by trained observers using SCUBA along a depth contour of 6–9 metres, providing a total of 1,500 m2 area surveyed per location. Benthic cover estimates were obtained using point sampling from video recordings of transects. Sites were visited annually or biennially; for full survey protocols, see Sweatman et al. (2008). We also included data from 39 sites in the AIMS inshore monitoring program. Since the inshore survey design differs, disease counts were segregated by depth (2 or 5 m) and normalized for area (inshore monitoring transects are 40 m2 each; LTMP transects are 100 m2 each) to be comparable with the LTMP and RAP counts. The mean values for white syndrome counts and average bleaching across all transects on each reef were compared between the inshore and LTMP data sets using an independent samples t test and not found to be significantly different for either bleaching or disease; thus, the two data sets were combined and analysed together. Although both bleaching and disease observations were collected simultaneously, bleaching data were only collected starting in 1999; thus in both the inshore and LTMP surveys, complete disease and bleaching data existed from 1998 to 1999 onwards, respectively. Between the inshore and LTMP data, there were a total of 93 locations spanning latitudes from 23.91 to 12.23°S that were visited a minimum of twice and a maximum of 14 times in the 1998–2010 period, giving a total of 961 reef and year replicates. In all surveys, disease observations were recorded as number of diseased colonies per transect, whereas bleaching observations were recorded as percent area of the transect bleached. Although genus and species information were not recorded for either bleaching or disease observations (these data were at the transect level only), the proportion of coral cover represented by each genus in each transect was recorded. This allowed us to use taxonomic composition as a statistical predictor for transect-level bleaching or disease responses.

Environmental variables

We used or calculated six temperature metrics for subsequent inclusion in analyses of bleaching and disease (Table 1). In previous studies, these metrics have been found to be useful predictors of either bleaching or disease (or both) (e.g., Bruno et al. 2007; Heron et al. 2010; Maynard et al. 2011). We obtained the first two metrics, weekly sea surface temperature anomaly (WSSTA) and degree-heating-weeks (DHW) from version 3 of the global CoRTAD data set (Selig et al. 2010), which contains data until the end of calendar year 2009; thus, we constrained the analyses that used environmental data to this time period, even though the bleaching and disease data extend to November 2010. WSSTA counts the frequency of warm anomalies greater than 1 °C from the climatological mean (1985–2004) during 52 weeks prior to the survey date. A degree-heating week is the sum of the previous 12 weeks where the temperature exceeded the climatological maximum temperature by at least 1 °C.
Table 1

Summary of temperature metrics used in this study and how they are calculated

Index

Accumulation period (prior to survey date)

Formula

Weekly sea surface temperature anomaly (WSSTA)

52 weeks

\( {\text{WSSTA}}, {\text{week}}_{i + 51} = \sum\nolimits_{i}^{i +51} {\left\{ {\begin{array}{*{20}c} {1, {\text{Week}}_{\text{i}}{\text{SST}} > {\text{Climatological Week}}_{i} {\text{SSTMean}} + 1^\circ {\text{C}}} \\ {0, {\text{Week}}_{i} {\text{SST}}\le {\text{Climatological Week}}_{i} {\text{SST Mean}} + 1^\circ{\text{C}}} \\ \end{array} } \right.} \)

Degree-heating week (DHW)

12 weeks

\( {\text{DHW}}, {\text{week}}_{ i + 11} = \sum\nolimits_{i}^{i +11} {\left\{ {\begin{array}{l} {{\text{Week}}_{i} {\text{SST}} -\left( {{\text{Clim}} . {\text{Max SST}} + 1^\circ {\text{C}}}\right), {\text{Week}}_{i} {\text{SST}} \ge {\text{Clim}} .{\text{Max SST}} + 1^\circ {\text{C}}} \\ {0, {\text{Week}}_{i}{\text{SST}} < {\text{Clim}} . {\text{Max SST}} + 1^\circ{\text{C}}} \\ \end{array} } \right.} \)

Climatology maximum is highest monthly SST value from the climatology period.

 Hot snap

Variable length; from start-of-preceding spring

\( {\text{Hot snap}} = \mathop \sum \nolimits \left\{ {\begin{array}{l} {{\text{SST}} - \left( {{\text{Clim}} . {\text{Summer Mean SST}} + 1\sigma } \right), {\text{SST}} > ({\text{Clim}} . {\text{Summer Mean SST}} + 1\sigma )} \\ {0, {\text{SST}} \le \left( {{\text{Clim}} . {\text{Summer Mean SST}} + 1\sigma } \right)} \\ \end{array} } \right. \)

 Cold snap

39 week period prior to most recent summer

\( {\text{Cold snap}} = \sum\nolimits_{i}^{i + 39} {\left\{ {\begin{array}{l} {{\text{SSTweek}}_{i} - \left( {{\text{Clim}} . {\text{Winter Mean SST}} - 1\sigma } \right), {\text{SSTweek}}_{i} < ({\text{Clim}} . {\text{Winter Mean SST}} - 1\sigma )} \\ {0, {\text{SSTweek}}_{i} \ge \left( {{\text{Clim}} . {\text{Winter Mean SST}} - 1\sigma } \right)} \\ \end{array} } \right.} \)

 Winter condition

39 week period prior to most recent summer

\( {\text{Winter condition}} = \mathop \sum \nolimits \left\{ {\begin{array}{l} {{\text{Weekly SST}} - \left( {{\text{Clim}} . {\text{Winter Mean SST}}} \right), {\text{winter weeks}}} \\ {{\text{Weekly SST}} - \left( {{\text{Clim}} . {\text{Winter Mean SST}}} \right) , {\text{nonwinter week and SST}} \le ({\text{Clim}} . {\text{Winter Mean SST}} + 1\sigma ) } \\ \end{array} } \right. \)

Mean positive summer anomaly (MPSA)

Up to 12 weeks, summer months only

\( {\text{MPSA, week}}_{i} = \sum\nolimits_{\text{First\,summer\,week}}^{i} {\frac{{{\text{DHW week}}_{i} }}{{{\text{No}} . {\text{ of weeks where DHW}} > 0}}} \)

where

\( {\text{DHW week}}_{i} = \mathop \sum \nolimits \left\{ {\begin{array}{l} {({\text{SSTweek}}_{i} - \left( {{\text{Clim}} . {\text{month SST}}} \right),{\text{SSTweek}}_{i} > {\text{Clim}} . {\text{month SST}}} \\ {0, {\text{SSTweek}}_{i} < {\text{Clim}} . {\text{month SST}}} \\ \end{array} } \right. \)

We calculated the next three metrics—Hot Snap, Cold Snap, and Winter Condition—for each survey location according to Heron et al. (2010) using the CoRTAD gap-filled temperature data. The Hot Snap metric accumulates when temperatures exceed the climatological (1998–2005) summer mean plus one standard deviation, for a period of accumulation that begins 3 months prior to the summer preceding the survey date and ends at the survey date. The Cold Snap index accumulates when temperatures are more than one standard deviation below the climatological winter mean over a period of accumulation for the 9 months proceeding the most recent summer. Finally, the Winter Condition index records unusually cold periods (more than one standard deviation below the climatological mean) outside of the winter months, as well as unusually mild (more than one standard deviation above the climatological mean) winters, thus accumulating both positive and negative values. The Winter Condition index thus records both unusually mild winters and unusually cold periods during other times of the year. A difference of note is that the CoRTAD database uses daytime–nighttime averages, whereas Heron et al. (2010) used nighttime temperature data only. The spatial (4 km) and temporal (weekly) resolution of the data were otherwise identical.

The final temperature metric was mean positive summer anomaly (MPSA). We calculated the MPSA values as per Maynard et al. (2008) using 4-km Pathfinder SST data, but using weekly instead of daily values so as to maintain a consistent temporal resolution for all of the temperature predictors.

Data analyses

We performed several steps in analyzing the data. First, because bleaching data were encoded using an eight-category system (absent; individual colonies; 1–5 %; 6–10 %; 11–30 %; 31–50 %; 51–75 %; 76–100 %) that recorded percentage of bleached hard coral cover for each transect, we used the midpoint of each area category as a weighting factor when determining the average bleaching severity for a given reef (i.e., 0 for absent, 1 for individual colonies, 3 for 1–5 %, 8 for 6–10 %, 20.5 for 11–30 %, 40.5 for 31–50 %, 63 for 51–75 %, and 88 for 76–100 %). We aggregated transect data to the reef level (15 transects per reef); any reefs that were surveyed only once were excluded from the temporal analysis but not the spatial analysis. For each reef, we summed white syndrome counts and calculated an average % bleaching using the weighting described above. White syndrome counts were normalized for area in the rare instances where the total area surveyed was less than 1,500 m2 due to missed transects. We also considered normalizing white syndrome counts by the amount of coral cover in each transect, but found the normalized diseased count to be highly correlated (r = 0.725) with the raw counts, and thus conducted all analyses using the raw count data. Because the transect-level data contained no additional spatial information (i.e., there was no information on the position of transects relative to each other), these reef-level data were also used for the spatial analyses.

Second, we examined all of the potential predictor variables for simple correlations using Pearson’s r and the variance inflation factor (VIF). VIF provides an estimate of how much of the increase in variance of a regression coefficient for a particular variable is due to collinearity with another variable; VIF values above 5 are generally considered to indicate a problem with multicollinearity (Menard 1995). Neither metric indicated collinearity at a high enough level to require exclusion of variables from the baseline models. Potential predictor variables were standardized using z-scores prior to inclusion in the statistical models, to facilitate comparison of effect magnitudes within and between models.

Third, we used logistic models for both bleaching and white syndrome because both appear to exhibit a threshold-type response (Fitt et al. 2001; Bruno et al. 2007; Jones 2008). Since logistic models require a binary-dependent variable, both bleaching and white syndrome data needed to be recoded as presence/absence, and thus we needed to set a threshold for counts (in the case of WS) or area (in the case of bleaching) to identify bleaching and disease “events.” For bleaching, all non-zero observations were considered to be bleaching events. For disease, rather than defining a single threshold a priori, we evaluated model performance using thresholds that varied from 0 to 50 counts per reef for WS. We present only results of the disease threshold that resulted in the highest Peirce skill scores (a threshold of 5 observed cases per reef). For the spatial analysis, the raw count/area data were used rather than thresholds.

To examine the utility of each of the six temperature metrics in predicting bleaching or disease, we first used each of the metrics in isolation in a logistic regression model. Then, to examine the effect of incorporating multiple abiotic predictors, we constructed baseline models for both bleaching and white syndrome using a backward-stepwise removal process based on the likelihood-ratio statistic for variable removal, with initial models containing all uncorrelated temperature predictors. For models predicting white syndrome, % acroporid cover (i.e., the proportion of each transect composed of acroporids) and interaction terms between abundance and each of the temperature metrics were also included, as per Bruno et al. (2007) and Heron et al. (2010). We did not include interactions between temperature metrics due to the lack of a plausible mechanism and corresponding meaningful physical interpretation of such effects. Bleaching was not used in any interaction terms so as to facilitate direct comparison with the baseline models without bleaching as a predictor. Acroporid cover was used from the year previous to the surveys where white syndrome was reported, as the correlation with white syndrome abundance was higher. This takes into account the possibility that acroporid cover may have already declined between the point of initial infection and when the survey detecting disease was carried out. Three alternate models for predicting bleaching were evaluated against the baseline model: one incorporating white syndrome from the same survey year, one incorporating white syndrome from the previous year, and one that used just white syndrome as a predictor without using any temperature variables. Similarly, for predicting white syndrome, we considered models that included bleaching. We evaluated models with and without an AR(1) covariance structure for white syndrome counts to account for the possibility of temporal autocorrelation. Furthermore, we also ran the models using transect-level data to verify whether the parameter estimates were sensitive to data aggregation. For the transect-level logistic models, a threshold value of 1 white syndrome case per transect was used due to the relative rarity of high counts. We also compared the results of models considering only presence/absence of white syndrome with those using the raw counts using a negative binomial error distribution with fits obtained through generalized estimating equations in SPSS.

For each model, we report the hit rate (% of cases in which the model predicts presence where presence is observed), false positive (% of cases in which the model predicts presence where absence is observed), false negative (% of cases in which the model predicts absence where presence is observed), and overall % classification (total # of correct classifications divided by the total number of cases). Overall % classification can be a misleading metric of model utility in cases where, for example, a “constant” model predicting that events never occur may have high apparent predictive power when events are rare. Thus, we used the Peirce Skill Score (PSS) (Peirce 1884; van Hooidonk and Huber 2009) as the primary indicator of model performance; standard errors for the scores were calculated according to Stephenson (2000). The PSS ranges from −1 (for a model where the predicted state is exactly the opposite of the observed state) to 1 (for a model where all cases are predicted correctly), with random or constant models standardized to a score of 0.

Finally, a spatial analysis of the reef-level bleaching and disease count data (i.e., using actual counts rather than threshold values) was conducted using Moran’s I (Moran 1950), Ripley’s (Ripley 1976, 1977), and semivariograms to check for potential spatial autocorrelation in either bleaching or white syndrome cases at broad scales. Moran’s I was run at several distance bands ranging from <1 to ~40 km to examine trends in spatial autocorrelation and to verify that non-spatially explicit statistical models were appropriate to use. Local non-random clustering of high or low values was quantified using the Getis-Ord Gi* (Ord and Getis 1995) and Anselin Local Moran’s I statistic (Anselin 1995). To verify that temporal aggregation was not obscuring patterns of spatial clustering, a year-by-year analysis of bleaching and white syndrome events was also conducted using the Getis-Ord Gi* statistic (see Electronic Supplementary Material, ESM). These analyses were conducted using ArcGIS 10.

Results

Between 1999 and 2010, a total of 914 surveys were conducted (i.e., 129 reefs visited an average of 6 times over the 11-year period). The actual amount of data available for each analysis varied slightly due to data being missing for certain variables (see table legends for sample sizes for each analysis). A total of 8,792 white syndrome-affected colonies were observed during that time, with the mean bleaching category ranging between 0.02 and 1.43 (meaning that the maximum amount of observed bleaching when averaged across all reefs surveyed in a year was less than 5 % per transect). Within these reefs, there was a pronounced spike in white syndrome cases in 2002 which coincided with increased bleaching (Fig. 1); however, a much larger increase in bleaching in 2006 was not matched by an accompanying increase in white syndrome cases. Aside from the large outbreak in 2002, the number of white syndrome cases since 1999 has been relatively constant.
https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0944-6/MediaObjects/338_2012_944_Fig1_HTML.gif
Fig. 1

Observations of average white syndrome and area-weighted (extent category reweighted by area) bleaching counts over time (1999–2010) across all surveyed reefs (normalized to 15 transects/1,500 m2 survey area per reef). Error bars represent one standard error. Numbers above bars represent sample size (# of reefs surveyed) for each year

The baseline model for white syndrome included Cold Snap index, % acroporid cover, and an interaction term of the Hot Snap index and % acroporid cover (Table 2). Overall model classification success was 74 %, with a PSS of 0.329. The addition of bleaching as a predictor did not improve model performance (Fig. 2a), and the bleaching term was non-significant (Table 2). Consistent with this, the model using only bleaching and % acroporid cover had higher false-positive and false-negative rates than the baseline model, as well as a lower PSS score. The bleaching term itself was not significant in this model either, indicating that acroporid cover alone provided most of the predictive utility. Finally, the model that included observations of bleaching from the previous survey year had an only marginally higher PSS score than the baseline model, and again, the bleaching term of the model was non-significant. Since the PSS is dependent on the presence/absence threshold percentage, we verified that our conclusions were not sensitive to our choice of PSS as our model diagnostic by examining two other diagnostics (ROC AUC score and Nagelkerke R-square value), both of which also indicated that the baseline model without bleaching was the best model.
Table 2

Comparison of logistic models for the presence of white syndrome with and without bleaching in the same/previous year as a predictor

Parameter (standardized)

Model without bleaching

Model with bleaching

Model with bleaching, w/o temperature predictors

Model with previous year bleaching

Estimate

Significance

Estimate

Significance

Estimate

Significance

Estimate

Significance

Cold snap

0.553 ± 0.124

0.000

0.558 ± 0.125

0.000

n/a

n/a

0.576 ± 0.127

0.000

Hot snap*acroporid cover

−0.260 ± 0.072

0.000

−0.262 ± 0.072

0.000

n/a

n/a

−0.255 ± 0.073

0.000

% Acroporid cover

1.146 ± 0.113

0.000

1.152 ± 0.114

0.000

0.958 ± 0.094

0.000

1.131 ± 0.115

0.000

Proportion bleached

n/a

n/a

0.078 ± 0.085

0.359

0.051 ± 0.082

0.535

0.097 ± 0.080

0.228

Constant

−0.647 ± 0.091

0.091

−0.650 ± 0.091

0.000

−0.587 ± 0.080

0.000

−0.609 ± 0.092

0.000

Hit rate (H)

41.4 %

41.4 %

38.4 %

43.4 %

False positive % (F)

8.6 %

8.6 %

10.1 %

8.4 %

False negative % (1-H)

58.6 %

58.6 %

61.6 %

57.6 %

Overall %

74.0 %

74.0 %

71.2 %

74.2 %

PSS (H–F) ± SE

0.329 ± 0.034

0.329 ± 0.034

0.283 ± 0.031

0.350 ± 0.034

Values in bold are statistically significant at p < 0.05

(Absence: 0–5/Presence: >5)

The baseline model (first column) was derived by backward-stepwise selection starting with all potential predictor variables. For all effects, the effect size and standard errors are shown. n/a = term not included in model. There were a total of 718 reef and year combinations with data available for this analysis

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0944-6/MediaObjects/338_2012_944_Fig2_HTML.gif
Fig. 2

Peirce skill scores for models predicting a bleaching and bwhite syndrome. Error bars represent 1 standard error

Hierarchical logistic models that used disaggregated (transect-level) data yielded parameter estimates with the same sign and very similar magnitude as the logistic models that used the aggregated (reef level) data, but performed no better than chance according to their ROC scores (ESM Table 1); additionally, bleaching was not a significant predictor in any of these models. The inclusion of an AR (1) covariance structure for white syndrome also did not improve model fits, indicating that there was no temporal autocorrelation of white syndrome. Hierarchical models of white syndrome counts produced similar results in terms of predictors and relative model performance as the reef-level logistic models. In particular, bleaching in the same or previous year was not a significant predictor of white syndrome count (ESM Table 2).

The baseline model for bleaching included three temperature indices: Hot Snap, Cold Snap, and Winter Condition (Table 3, Fig. 2b). However, this model had a relatively low PSS of 0.053. Adding white syndrome counts from the same year to the baseline model decreased the false-positive rate with no change to the false-negative rate, although the PSS was largely unchanged from the baseline model and the WS term was non-significant. Using white syndrome counts from the previous survey year rather than the same year led to a substantial increase in the PSS, but, again, the WS term was non-significant. Finally, the model that used only white syndrome to predict bleaching predicted virtually no bleaching: it had the highest false-negative rate of all the models (99.5 %), the lowest false-positive rate (0.5 %), and had a very low PSS. Again, in this model, the WS term itself was non-significant. Bleaching models based on transect-level data were uninformative, producing uniform predictions and Peirce skill scores of 0 in all cases (ESM Table 3).
Table 3

Comparison of logistic models for presence/absence of bleaching with and without white syndrome in the same/previous year as a predictor

Variable (standardized)

Model w/o WS disease

Model with WS disease

Model with WS disease (previous year)

Model with WS disease, w/o temperature variables

Estimate ± SE

Significance

Estimate ± SE

Significance

Estimate ± SE

Significance

Estimate ± SE

Significance

Hot snap

0.237 ± 0.071

0.000

0.234 ± 0.071

0.001

0.181 ± 0.077

0.018

n/a

n/a

Cold snap

0.172 ± 0.077

0.002

0.170 ± 0.078

0.028

0.306 ± 0.084

0.000

n/a

n/a

Winter condition

−0.183 ± 0.073

0.013

−0.184 ± 0.073

0.012

−0.253 ± 0.082

0.002

n/a

n/a

WS count

n/a

n/a

0.024 ± 0.067

0.726

−0.141 ± 0.099

0.154

0.044 ± 0.065

0.501

Constant

−0.444 ± 0.070

0.000

−0.444 ± 0.070

0.000

−0.146 ± 0.078

0.061

−0.416 ± 0.066

0.000

Hit rate (H)

10.8 %

10.8 %

42.4 %

0.5 %

False positive % (F)

5.5 %

4.9 %

29.7 %

0.2 %

False negative % (1-H)

89.2 %

89.2 %

57.6 %

99.5 %

Overall %

61.6 %

62.0 %

57.4 %

60.4 %

PSS (H–F) ± SE

0.0533 ± 0.020

0.0589 ± 0.019

0.127 ± 0.036

0.00351 ± 0.0041

Values in bold are statistically significant at p < 0.05

The baseline model (first column) was derived by backward-stepwise selection starting with all temperature variables. n/a = term not included in model. There were a total of 870 reef and year combinations with data available for this analysis

Overall, bleaching occurrence and white syndrome prevalence do not appear to be correlated (Fig. 3) at the regional scale based on the assemblage-level data available to our study. Pooling across all years, no spatial autocorrelation of either bleaching or disease was detected with semivariograms or Moran’s I statistic, that is, the occurrence of bleaching or disease does not decay as a predictable function of distance from a given point (ESM Fig. 1). However, the results of the Getis-Ord Gi* statistic calculated on the average prevalence of bleaching and white syndrome across all years (Fig. 4a, b) did suggest that white syndrome cases tended to be clustered at the latitudinal extremes of the GBR (Fig. 5a). Bleaching was also patchy, but significant clustering of high values occurred in different regions than disease, specifically between Townsville and Cairns, and near Heron Island in the south (Fig. 5b). No clustering of unusually low (but non-zero) values was detected for either bleaching or white syndrome. The results of the Anselin Local Moran’s I analysis showed the same patterns as the Getis-Ord Gi* statistic and thus are not shown. The year-by-year Getis-Ord Gi* analysis also did not show any overlap of bleaching and white syndrome clusters for any year (ESM Fig. 2a, b). Clustering of the two events thus does not appear to be spatially congruent.
https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0944-6/MediaObjects/338_2012_944_Fig3_HTML.gif
Fig. 3

White syndrome counts versus bleaching severity (area-weighted) across all years and reefs. Axes are log (x + 1)

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0944-6/MediaObjects/338_2012_944_Fig4_HTML.gif
Fig. 4

Cumulative area-weighted bleaching and white syndrome frequencies across all years (1999–2010) for the 129 reefs surveyed at least once during the study. Blue represents lowest numbers; red represents highest numbers. Each point represents a sampled reef. Categories are quantiles (i.e., bins with an equal number of records in each). a Average  % bleaching. In this data set, recorded instances of bleaching are generally low apart from some sites between Cairns and Townsville and in the far south of the GBR. b Average number of white syndrome cases per reef; the main areas of high white syndrome occurrence are in the far north and the far south of the GBR

https://static-content.springer.com/image/art%3A10.1007%2Fs00338-012-0944-6/MediaObjects/338_2012_944_Fig5_HTML.gif
Fig. 5

Getis-Ord Gi* p values for average number of white syndrome and bleaching cases across all years (1999–2010) for the 129 reefs surveyed at least once during the study. Significant p values (p < 0.05) in red indicate non-random clustering of high values. Non-significant p values in gray indicate no clusters of high or low values. a Bleaching. Statistically significant clusters of high bleaching observations occurred at two inshore sites between Cairns and Townsville, and around Heron Island in the southern GBR. b White syndrome cases. Statistically significant clustering of high values occurred in the far north and far south of the GBR. No clusters of low values were detected

Discussion

While many studies have suggested a direct causal relationship between coral bleaching and disease (Jones et al. 2004; Muller et al. 2008; Croquer and Weil 2009), we found no evidence of a correlation between observations of white syndrome and observations of bleaching, at the assemblage level. Moreover, over the past decade, the spatial patterns of bleaching and white syndrome on the GBR have not generally coincided. This is consistent with recent findings by Roff et al. (2011), who suggest that bleaching and white syndrome occurrence could be negatively correlated, possibly due to density-dependence of white syndrome. Given that including the occurrence of bleaching—whether in the same year or the previous year—did not significantly improve model performance, bleaching does not appear to be a useful predictor of white syndrome prevalence when using assemblage-level data, except perhaps insofar as it may indirectly capture the presence of physical environmental stresses that may cause bleaching. Even then, however, the large-scale environmental variables themselves were better predictors of WS.

Our results indicate that there have been localized hotspots of white syndrome outbreaks, which is consistent with white syndrome resulting from a spreading pathogen or another multiple-point-source or clustered risk factor exposure in addition to—or instead of—being a purely environmentally driven phenomenon (e.g., Ainsworth et al. 2007; Kvennefors et al. 2010). This contrasts with Roff et al.’s (2011) findings, who concluded that the pathogenicity of white syndrome was low in both aquaria and field settings and found a lack of spatial aggregation of white syndrome cases at the colony level. However, we do find that bleaching and disease share some temperature-related environmental predictors—although these variables were relatively weak predictors of bleaching risk in these data—even though these temperature indicators were at extreme values in some years, such as during the 2002 mass bleaching event on the GBR (Liu et al. 2003).

Unlike Heron et al. (2010), in all models, an increasing Cold Snap index was associated with an increasing probability of white syndrome occurrence, but an increasing Winter Condition was associated with a decrease in white syndrome occurrence. However, while the Hot Snap term by itself was not significant, the interaction of Hot Snap and % acroporid cover was significant in all three of the models where it was included, indicating that the Hot Snap index has a greater influence on white syndrome occurrence when acroporid cover is low than when it is high. While our results differ from Heron et al.’s (2010), a key difference between the approaches is that we were predicting presence/absence of white syndrome rather than linear severity above an outbreak threshold. Additionally, our analyses included an additional 4 years of environmental data and five additional years of disease/bleaching data compared to Heron et al. (2010). These findings may indicate that the temperature and white syndrome relationship is more complicated than previously suspected; for example, Roff et al. (2011) found that initiation and progression of white syndrome occurred under both summer and winter conditions and that there was only a weak relationship between white syndrome lesion progression and thermal stress.

Using multiple temperature indices to predict bleaching improved Peirce Skill Scores, however, all of the skill scores were low compared with the DHW-based models assessed by van Hooidonk and Huber (2009), which had an average PSS of 0.55. Although in our analysis the model that incorporated white syndrome occurrence from the previous year had the highest PSS score of all models, the bleaching term itself was non-significant. Given the large change in the coefficient of the Cold Snap term when WS is added as a predictor (from 0.172 to 0.306), this could indicate that an interaction exists between white syndrome and another unknown variable that the model does not include. Interestingly, all three of the temperature indices developed by Heron et al. (2010) for predicting white syndrome remained as significant predictors, with unusually cold winters (as measured by Cold Snap) and hot summers (as measured by Hot Snap) increasing, and unusually mild winters (as measured by Winter Condition) decreasing the likelihood of bleaching. The direction of the bleaching relationship with the Cold Snap and Winter Condition indices is opposite to that found by Heron et al. (2010) for white syndrome. This provides further support that the bleaching in our data set was predominantly environmentally stress-linked rather than causally linked with white syndrome occurrence. Furthermore, our findings support the hypothesis that unusually cold winters (i.e., temperatures below the lower limit of corals’ thermal optimum range) may result in a physiological stress that persists long enough to affect susceptibility to later heat stress, while a mild winter may pre-condition corals to an ensuing warm summer—similar to a concept first suggested by Berkelmans and Willis (1999).

None of the commonly used temperature metrics on their own proved to be good predictors of bleaching in the LTMP survey data. While this result was unexpected, Berkelmans and Oliver (1999) reported that, even during the 1998 mass bleaching event on the GBR, bleaching was most severe on the reef flat and at depths shallower than 4 m, although bleaching was observed to as much as 20 m depth on some mid- and outer-shelf reefs. Since the LTMP transects are on the reef slope at depths of 6–9 m, they may not bleach substantially even if mass bleaching is occurring at shallower depths and/or on the reef flat. While the LTMP bleaching data do not encompass the 1998 mass bleaching event, there was a second mass bleaching event on the GBR in 2002, in which a greater proportion of offshore reefs bleached (41 %; Berkelmans et al. 2004). This event is also not reflected in the LTMP data (Fig. 1), although there was a pronounced spike in white syndrome cases that year. Bruno et al. (2007) also proposed there was little or no spatial overlap between the 2001/2002 mass bleaching event and white syndrome severity during the same period. However, Bruno et al. (2007) were only able to compare bleaching observations from aerial surveys conducted by Berkelmans et al. (2004) (and thus likely dominated by reef flat habitats) with disease data from in situ observations of the reef slope. Our study confirms the conjecture of Bruno et al. (2007), at least for reef slope habitats, by direct comparison of bleaching and disease outbreaks observed on the same transects.

The apparent lack of co-occurrence of bleaching and white syndrome in our study could have been influenced by the survey design. As others (e.g., Jones 2008) have pointed out, sampling frequency should ideally be matched to the temporal scale of the events being monitored, and thus annual surveys may be missing episodes that are too temporally fleeting or localized to be detectable weeks or months after the event. Although lags of as long as a year between a bleaching episode and disease onset have been reported (Mydlarz et al. 2009), the annual or biennial frequency of sampling means that short-term lags that leave no lasting visible effects may be missed between survey visits. Studies that have used longitudinal monitoring of individual coral colonies (e.g., Brandt and McManus 2009; Bruckner and Hill 2009; Croquer and Weil 2009) have generally found stronger correlations between bleaching and disease. Thus, patterns of bleaching and disease that are readily apparent at the colony level scales may not be manifest at the assemblage level, highlighting the need for long-term, regional monitoring studies that track the progression of bleaching and disease at the level of individual colonies [similar to Roff et al. (2011), but replicated across a larger area and over a longer time period].

Whether or not bleaching and disease have a direct causal link at the level of individual colonies, information about one could still, in principle, be useful for predicting the other indirectly. In particular, where environmental data are lacking, or coarse in scale, bleaching and disease may serve as useful surrogate measures of localized environmental stress. Indeed, it is in this context that the lack of a strong relationship between these two variables in our analysis was most surprising. Our temperature data are remotely sensed, and thus are more reflective of conditions prevailing near the ocean surface. Temperatures on the reef slope are likely to be partially decoupled from surface temperatures due to such influences as tidal bores, mixing, and bottom topography (Wolanski and Hamner 1988; Jiménez 2001). We expected that bleaching and disease would be particularly useful co-predictors under these circumstances. Instead, we found that our measurements of the physical variables themselves were much more reliable, and moreover, they worked well for predicting disease. This tends to suggest that the remotely sensed temperature data actually provided a reasonably good index of the thermal conditions on the reef slope, but that bleaching susceptibility is simply much lower in those habitats, perhaps due to lower irradiance or higher flow.

Disease and bleaching are just two of the many stressors at work in coral reef ecosystems that pose a complex problem for ecologists and resource managers. The response of an ecological community to these stressors is at least partially dependent on whether the community has a positive or negative co-tolerance (Vinebrooke et al. 2004), and thus regions or habitats in which disease and bleaching are not strongly associated or cannot be used as reliable co-predictors are likely to pose a more difficult management problem than those areas in which they are tightly coupled. It remains to be seen whether the findings of this study are indicative of biogeographical (e.g., Caribbean vs Indo-Pacific) or habitat-specific (e.g., reef flat vs. reef slope) differences in bleaching-disease susceptibility and responses, or whether these relationships may have been obscured through the temporal and spatial aggregation necessary with our data set. Effect sizes that are readily apparent at the colony level may become more difficult to detect when scaled up to the assemblage and community level. While we did not find any spatial or temporal correlations at this regional scale, we cannot rule out the existence of a bleaching-white syndrome connection at fine spatial scales or in different reef habitats.

Acknowledgments

We thank Scott Heron and Jeff Maynard for assistance with temperature metric calculations, the members of the Australian Institute for Marine Science (AIMS) Long-term Monitoring Program for collecting these data, Angus Thompson and Hugh Sweatman for kindly providing the AIMS data sets and answering our questions about them, and feedback from Jonathan Belmaker and two anonymous reviewers that helped to improve this manuscript. This research was supported by the Australian Research Council and James Cook University.

Supplementary material

338_2012_944_MOESM1_ESM.docx (28 kb)
Supplementary material 1 (DOCX 28 kb)
338_2012_944_MOESM2_ESM.eps (2 mb)
Supplementary material 2 (EPS 2051 kb)
338_2012_944_MOESM3_ESM.eps (13.2 mb)
Supplementary material 3 (EPS 13529 kb)
338_2012_944_MOESM4_ESM.eps (15.3 mb)
Supplementary material 4 (EPS 15617 kb)

Copyright information

© Springer-Verlag 2012