Acute drivers influence recent inshore Great Barrier Reef dynamics

Abstract

Understanding the dynamics of habitat-forming organisms is fundamental to managing natural ecosystems. Most studies of coral reef dynamics have focused on clear-water systems though corals inhabit many turbid regions. Here, we illustrate the key drivers of an inshore coral reef ecosystem using 10 years of biological, environmental, and disturbance data. Tropical cyclones, crown-of-thorns starfish, and coral bleaching are recognized as the major drivers of coral loss at mid- and offshore reefs along the Great Barrier Reef (GBR). In comparison, little is known about what drives temporal trends at inshore reefs closer to major anthropogenic stress. We assessed coral cover dynamics using state-space models within six major inshore GBR catchments. An overall decline was detected in nearly half (46%) of the 15 reefs at two depths (30 sites), while the rest exhibited fluctuating (23%), static (17%), or positive (13%) trends. Inshore reefs responded similarly to their offshore counterparts, where contemporary trends were predominantly influenced by acute disturbance events. Storms emerged as the major driver affecting the inshore GBR, with the effects of other drivers such as disease, juvenile coral density, and macroalgal and turf per cent cover varying from one catchment to another. Flooding was also associated with negative trends in live coral cover in two southern catchments, but the mechanism remains unclear as it is not reflected in available metrics of water quality and may act through indirect pathways.

1. Introduction

Ecosystems are spatially heterogeneous dynamic systems, many of which alternate between periods of disturbance and recovery [1–3]. A fundamental aspect of ecosystem management is to interpret and predict system patterns and behaviours [4], therefore requiring the relative importance of different stressors on system dynamics to be unpacked and disaggregated in space and time [5,6]. The impacts of disturbance on community dynamics differ along environmental gradients [7], sometimes creating ecological surprises [8,9]. Unravelling the emergent dynamics of an ecosystem can begin with comprehensive retrospective analyses of time-series data [10,11], which are facilitated by recent technological advances and the accumulation of long-term monitoring datasets.

Analyses to investigate long-term population trends have been hampered by limitations in sampling strategies, missing values, and irregular data points [12,13]. State-space models (SSMs) can be used to estimate underlying population trends, accounting separately for natural variability in biological data (process error) and observation error inherent in all surveys. Such methods are now used routinely in fisheries research and increasingly in ecology [14,15]. Additionally, SSM allows the influence of covariates on process (process covariates) and observed time-series (observation covariates) to explain changes in population trends. Here, we use a coral reef example to investigate how a suite of biophysical drivers, including environmental gradients of water quality [16], influence a dynamic ecosystem typically characterized by prolific recovery and losses [17,18].

The majority of research into coral reef ecosystems has focused on relatively clear-water ecosystems where episodic events such as cyclones and predation [19,20], as well as thermal anomalies [21], incur major impacts to corals. However, extensive reefs also occur in areas that are either naturally or unnaturally turbid [22] and are exposed routinely to vastly different physical conditions such as higher turbidity and nutrient levels [23]. Despite extensive research on the biological properties of inshore reefs, such as effects of individual stressors [24–27], response mechanisms [28–30], and community structure over both contemporary [31,32] and palaeoecological scales [33,34], it is currently unclear how characteristics unique to inshore environments interact with ecological processes to affect coral dynamics.

Using the inshore Great Barrier Reef (GBR) as a model system, we examine a 10-year time-series of benthic community dynamics to identify trends and drivers of variability in coral cover across space and time. Our approach used multivariate autoregressive state-space models (MARSS) based on a density-dependent Gompertz population model [35]. Drivers that were explored include biological, environmental, and disturbance variables (table 1). Through an investigation of driver combinations on total coral cover, we address the following questions for the inshore GBR: (i) what are the underlying reef trends? (ii) What are the key biotic and abiotic drivers of inshore reefs? (iii) Does the modelling of observation covariates significantly improve predictions of reef dynamics in turbid environments?

Table 1.

Rationale for covariates used in the study.

predictor	rationale	methods	references
Biological
macroalgae	A direct competitor for space, an increase in macroalgae also inhibits recruitment of corals. Hypothesis: an increase in macroalgae will lead to a decrease in coral cover, caused by negative feedback mechanisms.	photo point intercept	[36–39]
turf	Turfs are favoured by nutrient enrichment and when they trap sediments they smother coral recruitment. Hypothesis: an increase in turf will have a negative impact on coral cover by reducing coral recruitment or overgrowth of coral margins.	photo point intercept	[40–43]
juvenile coral density	Juvenile corals are crucial for the recovery of coral reef systems after disturbances, but effects on coral cover can be lagged depending on genus. Hypothesis: an increase in juvenile density will lead to an increase in coral cover.	belt transects	[44–46]
Environmental
Secchi depth	Secchi depth is a proxy for water clarity and correlated with all satellite-derived water quality variables. Reduced light penetration may reduce calcification and coral growth, reducing the competitive ability of corals. Areas of low Secchi depth have lower coral biodiversity and increased macroalgae. Hypothesis: an increase in Secchi depth (better water quality) will lead to an increase in coral cover.	satellite Modis-AQUA	[47,48]
MEI	An increase in multivariate ENSO index (El Niño/Southern Oscillation) suggests abnormal warming of the ocean surface, known to cause mortality in corals. Hypothesis: an increase in MEI will lead to a decrease in coral cover.	index	[49,50]
SST	Increase in sea surface temperature causes mortality in corals. Hypothesis: an increase in SST will lead to a decrease in coral cover.	satellite	[51]
Disturbances
bleaching	Climate-induced ocean warming causes mortality in corals.	scuba transects	[52]
COTS	Crown-of-thorns starfish predation causes mortality in corals.	scuba transects	[53]
disease	Disease is a cause of coral mortality, especially in areas with high nutrient concentrations such as the inshore reefs.	scuba transects	[54,55]
flooding	Flood-induced coral mortality occurs during periods of high rainfall and river discharge and can be related to either direct exposure to freshwater or reduced water quality caused by terrestrial runoff.	scuba transects	[56]
storm	Widely known to be a major cause of coral cover decline through mechanical damage. Overall hypothesis (disturbances): an increase in the years since a disturbance will lead to an increase in coral cover.	scuba transects	[57,58]
Observation
Secchi depth	Secchi depth is a proxy for water clarity. Reduced water clarity is likely to lower the likelihood of detecting corals during monitoring. Hypothesis: a decrease in Secchi depth (lower water quality) may result in a reduction in cover estimate.	satellite Modis-AQUA	[47]
clay/silt	An increase in proportion in clay/silt signals increased sedimentation. Hypothesis: relatively high rates of sediment accumulation may result in burial of live coral and a reduction in cover estimate.	sediment sampling	[59]

2. Material and methods

(a). Data

Benthic cover (total coral, turf, and macroalgal cover), disturbance, and environmental data were collected from 15 inshore reefs (figure 1; electronic supplementary material, table S1). Surveys spanned an area from S 16°18′ to S 23°15′ along the GBR and were carried out annually between 2005 and 2014 during the austral winter. For sampling methodology, see electronic supplementary material, S1.

Figure 1.

Map of the 15 MMP reef sites and adjacent catchment-based natural resource management (NRM) regions. The Wet Tropics NRM region includes three sub-regions (Daintree: SIN, SIS; Johnstone: FIW, HIW, FGE; Tully: DIN).

(b). Multivariate autoregressive state-space modelling approach

We used a multivariate autoregressive state-space modelling approach [60] to estimate population trends and identify key drivers of coral dynamics. The MARSS approach accounts for process variability and observation error characterized by two equations: a state process that describes the underlying dynamics (2.1) and an observation process based on survey data (2.2).

2.1

and

2.2

In the above equations, x represents the estimated coral cover at each reef at time t. Coral cover data enter the model as y at the site level. The parameters w_t and v_t are the process and observation error, with a mean of 0 and covariance matrix of Q_t and R_t, respectively, following a multivariate normal (MVN) distribution. The parameter u is the estimated trend of the hidden states, and a is the trend of the observed data. B is an identity matrix and Z represents the spatial structure to which the observed data belong. C and D represent the process and observation covariate effects, respectively, where c and d are the covariate values. All covariates were centred and scaled to allow a standardized comparison of the drivers [60]. The best-fit model was selected using a derivative of Akaike Information Criterion (AICc) for small sample sizes [61]. Models with a ΔAICc of less than 2 were considered to have similar support [62]. Where competing models exist, the model with the lowest AICc was chosen for a better out-of-sample predictive ability [63]. However, any other alternative/competing models were also reported for completeness (table 2). We could have used multimodel averaging but such an approach for inference has been questioned recently [63,64] and so was not used here. All analyses were performed using the MARSS package in R [35].

Table 2.

Key drivers of coral reef dynamics from 2005 to 2014 in inshore GBR. Key drivers identified are based on model AICc values in electronic supplementary material, table S2. Drivers in parentheses represent alternative competing models of similar fit. Base represents models with no covariates and observation indicates an observation covariate.

	Daintree	Johnstone	Tully	Burdekin	Mackay	Fitzroy
2 m	storm (base)	disease and storm (disease)	turf and storm	base	base + observation: clay/silt	flood (base)
5 m	macroalgae and disease	disease and storm	storm (turf and storm)	storm (base)	juvenile coral density and flood	base/flood

(c). Modelling strategy

Models followed the design of the Marine Monitoring Program (MMP), where sites were replicated adjacent to major catchments draining to the GBR [65] (figure 1). Twelve models were constructed for the six catchments at two depths, including Daintree, Johnstone, Tully, Burdekin, Mackay, and Fitzroy. As data were collected at two specific depths, models were separated into shallow (2 m) and deeper reefs (5 m) following the modelling strategy in Harrell [66]. Twelve variables representing biological, environmental, disturbance drivers, and observation covariates were used in the MARSS analysis (table 1; electronic supplementary material, table S1). The details of modelling strategy are provided in the electronic supplementary material.

3. Results

Of the 15 reefs at two depths, almost half (14) exhibited a declining trend of coral cover, seven exhibited fluctuating dynamics with no overall long-term change in coral cover, five remained static, and four showed a trend of coral recovery from 2005 to 2014 (figure 2). Trends of decline in coral cover were strongly associated with acute environmental drivers, whereas associations with biological drivers were relatively weak (electronic supplementary material, table S2 and figures S5–S10). The importance of drivers was ordered from storm (strongest) > flood and disease > clay/silt, macroalgae, turf, and juvenile coral density (table 2 and figure 2). Shallow Burdekin reefs were the only systems where no clear drivers were identified. In the deeper (5 m) Fitzroy reefs, the model with floods and the base model are both presented because of the small difference in AICc (0.75).

Figure 2.

(a,b) MARSS coral cover predictions for 30 reefs across six catchments (north to south from top left to bottom right) at depths 2 m (a) and 5 m (b) from 2005 to 2014. Averaged catchment trends based on coral cover monitoring data are shown in colour, and MARSS predictions for individual reefs based on best-fit models are in black. Coloured bands and dotted lines represent 95% confidence intervals. Symbols represent disturbances that affected reefs during the sampling period. Text in the upper right corner shows key drivers of coral dynamics.

Storms repeatedly emerged from best-fit models in four out of the six catchments, and floods in the remaining two (table 2). Water quality (Secchi depth) was not associated with coral cover trends. However, adding clay/silt as an observation covariate significantly improved Mackay model fits (2 m) while catchment-level SST and a regional El Niño–Southern Oscillation (ENSO) index (Multivariate ENSO index (MEI)) did not. Hence, results suggest that local episodic events were more important than regional and chronic factors in driving coral dynamics along the inshore GBR for the 10 years from 2005 to 2014. Substantial increases and decreases in coral cover (2 m) were almost always driven by changes associated with the genus Acropora, though the role of this genus was less pronounced at 5 m in all catchments apart from Fitzroy (electronic supplementary material, figures S1 and S2).

Although drivers differed among models and were often specific to individual reefs (figure 2), the directionality of each driver's association with coral cover remained consistent (table 3). In general, coral cover increased as the years since disturbance term increased. Where significant, an increase in macroalgae in the previous year was associated with less coral the following year. Turf cover in the previous year had a slight positive association with corals in the following year, suggesting that corals outgrew algal turfs to occupy that space (i.e. contrary to predictions of algal turf being a net cause of coral decline in inshore systems).

Table 3.

Estimates of driver impacts on coral cover from best-fit models with 95% confidence intervals. Positive and negative estimates (C) imply a positive and negative effect on coral cover, respectively. Significant estimates (lower and upper CIs not straddling zero) are italicized. Each driver estimate is presented for individual reefs (in parentheses). Full reef names are presented in figure 1.

drivers (2 m)	estimates	low CI	upper CI	drivers (5 m)	estimates	low CI	upper CI
Daintree
storm (SIN)	0.070	0.028	0.111	macroalgae (SIN)	0.022	−0.010	0.051
storm (SIS)	−0.005	−0.041	0.027	macroalgae (SIS)	−0.039	−0.059	−0.021
				disease (SIN)	0.066	0.038	0.092
Johnstone
disease (FIW)	0.057	0.041	0.074	disease (FIW)	0.039	0.024	0.054
storm (FGW)	0.017	0.001	0.033	storm (FGW)	0.075	0.050	0.102
Tully
turf (DIN)	0.073	0.042	0.102	storm (DIN)	0.151	0.079	0.212
storm (DIN)	0.222	0.173	0.267
Burdekin
base		n.a.		storm (PAN)	0.038	0.019	0.060
				storm (GEO)	0.012	−0.006	0.027
Mackay
clay/silt (observation)	−0.025	−0.040	−0.011	flood (PIN)	0.044	0.022	0.065
				JuvHC (DCI)	0.017	−0.008	0.041
				JuvHC (DAY)	0.054	0.031	0.077
				JuvHC (PIN)	−0.032	−0.061	−0.004
Fitzroy
flood (HHI)	0.028	−0.004	0.404	flood (PEL)	0.016	−0.002	0.033
flood (PEL)	0.084	0.050	0.119

4. Discussion

Inshore coral reefs tend to be subjected to persistent, chronic stresses associated with relatively poor water quality [67]. It might be expected, therefore, that chronic stressors would tend to dominate the dynamics of such systems. However, our findings suggest that inshore reefs share similarities with offshore reefs, in that episodic events can exert striking impacts on their state [19]. Overall, storms were identified as the key driver of the underlying trends of coral dynamics across hundreds of kilometres in the inshore GBR. Apart from storms, the relative exposure to acute disturbances varies across the shelf with inshore reefs having greater vulnerability to floods and disease, mid- and offshore reefs having high risks of coral bleaching, and multiple areas being vulnerable to crown-of-thorns starfish outbreaks [19]. We stress, however, that relative exposure to acute events is dependent on the observation period. For instance, bleaching has been shown to reduce inshore coral cover substantially [21,68], but the relative unimportance here reflects a period of relatively mild water temperatures or potentially a high tolerance to bleaching from corals in turbid environments as demonstrated by Morgan et al. [69] and van Woesik et al. [70].

Reefs of the inshore GBR can be categorized into those that exhibit either fluctuating or stable trends of coral cover post 2005, hereinafter termed fluctuating and stable reefs, respectively. Of the stable systems, reefs of the Burdekin catchment generally resembled a degraded state with low cover and limited recovery, likely owing to their low resilience and historical disturbance regime [71,72]. Some Burdekin reefs have suffered early Acropora collapses in the 1920–1950s [33], while others have experienced relatively recent declines [36]. Cover in Mackay was also stable but unlike Burdekin, cover was relatively high (40–50%), likely due to the paucity of acute disturbances during, or preceding, the survey period. Identifying drivers in these relatively stable reefs is challenging and best-fit models either lacked covariates or had little difference between base and covariate models. The reason for this statistical difficulty or ambiguity is that analyses are predicated on detecting associations between the changes in response and covariates, yet the response was almost constant, exemplifying the difficulty in identifying processes that maintain systems in relatively stable states (albeit low or high).

Reefs with fluctuating coral cover (Daintree, Johnstone, Tully, and Fitzroy) were mainly driven by changes in Acropora, as documented in both offshore [73,74] and inshore reefs [75,76]. Fluctuations can be attributed to the rapid growth, high recruitment rates [77], and high sensitivity of Acropora to disturbance and disease [78]. Many reefs in Daintree and Johnstone exhibited stable or recovering trends until disease, COTS, flooding, and storms drove a decline after 2011. The recovery of Tully is evident between two major cyclones in 2006 and 2011. Identifying drivers in the Fitzroy catchment was difficult despite considerable changes in coral cover. Multiple disturbances impacted simultaneously and reefs exhibited variable responses to the same stressors. Bleaching caused extensive coral decline at Barren Island (BAR) but not at Humpy and Halfway Islands (HHI) in 2006, whereas coral cover remained constant at Pelican Island but declined at BAR and HHI during high prevalence of disease. These examples highlight the high spatio-temporal variability among reefs, where drivers often have site-specific impacts that are dependent on additional factors such as geomorphology, reef histories [79], fish assemblages [80,81], and other physical environmental conditions like aspect and wave exposure that were not accounted for in this study [82].

Storms are a major driver of coral dynamics [83,84] and vary in their frequency and temporal clustering along the GBR [85,86]. The catchments identified to have a strong association between inshore coral dynamics and storms all fall within the cyclone belt (i.e. Daintree, Johnstone, Tully, and Burdekin). Daintree was impacted by Cyclone Ita in 2014 and has a history of major storm disturbance, such as Cyclone Rona in 1999, where average coral cover declined from 66% to 17% in Snapper Island North [87]. Storms cause mechanical damage to both shallow and deeper reefs [88]. Johnstone and Tully were impacted by cyclones Larry in 2006 [89] and Yasi in 2011 [79,90]. Located in the central GBR, Burdekin also falls within the region of highest cyclone activity [85].

At first glance, the study results appear to be inconsistent with the wider literature demonstrating the negative impacts of declining water quality on inshore reefs [47,67]. Rather, we found that acute impacts have the greatest explanatory power and that impacts from intense chronic stress are difficult to identify. The apparent inconsistency can be reconciled through two mechanisms, which may act in concert. First, while we might classify flooding as an acute event, its mechanism may act through intensification of chronic stressors that are difficult to identify, particularly if data are aggregated over say annual timescales. For example, flooding was strongly associated with coral losses in the southern catchments and may involve mechanisms associated with water quality such as freshwater kill in Fitzroy [91] and high sediment load in Mackay [92]. Moreover, turbidity was found to be more important than the effect of floods in the shallow reefs of Mackay, which corresponds with observations of high loads of sediment physically accumulated on corals during surveys in 2009. The inability to detect impacts from water quality highlights the complexity of mechanisms involved, which vary over time and space and are complicated by variable exposure and sensitivity among locations.

A second reason to find little evidence of chronic drivers reflects the difficulty of drawing inference from systems that have undergone major historical alteration. Compelling evidence suggests that GBR anthropogenic pollution is linked to European settlement [34,93]. A likely scenario is that water quality stress impacted reefs decades ago and profoundly shaped the types of communities observed, but these chronic effects are not discernible over and above the recent acute events considered here.

A third, related possibility is that historical changes in the environment caused inshore reefs to transition to an alternative stable state that is now reinforced by ecological feedbacks. Such states are difficult to reverse and exhibit little fluctuation [37,94,95], as mooted by Johns et al. [96]. This scenario parallels what has been demonstrated in other ecosystems, where recovery dynamics show time lags in responses due to shifting baselines and hysteresis [97]. For example, seagrass can exhibit slow declines over a decade in Tampa Bay because of chronic changes in water quality [98] and time frames for recovery can reach 100 years once the stressor—trawling—has ceased [99]. Similarly, Lefcheck et al. [100] required a 30-year dataset to demonstrate the impact of nutrients and water quality on submerged aquatic vegetation in Chesapeake Bay. Whether communities of inshore GBR reefs exhibit alternate attractors remains to be seen, but is important to evaluate.

The emerging picture of coral reef dynamics is that the drivers of coral cover on inshore and offshore systems have more in common than previously thought. Although flooding was important in the southern catchments, storm impacts were found to be a major driver of inshore reef state as they are offshore [19]. A recent analysis of offshore reefs of the GBR found that water quality had the strongest (and negative) association with long-term recovery rate [3]. While our study could not tease out chronic influences on the dynamics of inshore coral cover, there is further work to be done in unpacking the cumulative impacts of acute and chronic pressures on coral reefs.

Data accessibility

Data supporting this article have been uploaded as the electronic supplementary material.

Authors' contributions

P.J.M., M.C., and A.T. conceived the study. A.T. collected the data. M.C., P.J.M., and C.D. provided advice on study design and statistical analysis. Data were provided by A.T. V.Y.Y.L. prepared the data, created models, analysed, interpreted the results, and wrote the manuscript. All authors discussed the results, edited the manuscript, and gave their final approval for publication.

Competing interests

The authors declare that they have no competing interests.

Funding

Funding for this research was provided by an ARC Linkage grant and ARC Centre of Excellence for Coral Reef Science grant to P.J.M. and an IPRS to V.Y.Y.L. Benthic data were sourced from the Marine Monitoring Programme undertaken by the Australian Institute of Marine Science, funded by the Australian Government Reef Programme and managed by the Great Barrier Reef Marine Park Authority.