Introduction

Coral reefs are one of the most vulnerable ecosystems to ocean acidification (Kleypas and Yates 2009), which is the product of increasing anthropogenic carbon dioxide (CO2) emissions, and the dissolution of CO2 in seawater. The dissolution of CO2 into the surface ocean elevates the aqueous partial pressure of CO2 (pCO2), thereby decreasing seawater pH and the saturation state (Ω) of calcium carbonate [CaCO3 as aragonite (Ωar) and calcite (Ωca)] without affecting total alkalinity (AT) (Doney et al. 2009). It is widely recognized that these effects will decrease the rates at which many coral reef calcifiers deposit CaCO3 (Kroeker et al. 2013; Comeau et al. 2014a), and increase the CaCO3 dissolution of reef communities (Andersson and Gledhill 2013; Eyre et al. 2014; Comeau et al. 2016). Combined, these trends are expected to cause community net calcification (Gnet, gross calcification minus dissolution) to decline on coral reefs, and potentially degrade structural habitat complexity that is formed through biogenic calcification (Hoegh-Guldberg et al. 2007; Comeau et al. 2016).

In situ measurements of coral reef community Gnet have shown that it is affected by natural diel variations in seawater carbonate chemistry (Shaw et al. 2012). For example,

Gnet was associated significantly (r2 = 0.33) with diurnal variation in Ωar (with a median diurnal variation of 3.25) at Lady Elliot Island, Australia (Shaw et al. 2012). Based on field measurements of the relationships between Ωar and Gnet, Silverman et al. (2009) predicted that coral reefs would transition from net calcification to net dissolution at 560 ppm atmospheric pCO2, which potentially could occur within 50 years (IPCC, 2014). However, for scleractinian corals that are the ecosystem engineers of coral reefs (Jones et al. 1994; Wild et al. 2011) and a major contributor to their CaCO3 production (Perry et al. 2013), there is interspecific variation in the sensitivity of calcification to increases in pCO2 (Chan and Connolly 2013; Comeau et al. 2014a). The calcification rates of many species are affected strongly and negatively by high pCO2 (Chan and Connolly 2013; Comeau et al. 2014a), but in some species, they are affected less severely (Comeau et al. 2014a). The sensitivity of calcification to pCO2 can be both species and life history stage specific and related to the extent to which organisms are able to biologically control the calcification process (Comeau et al. 2014a). Generally, net calcification of individual coral species appears to be less sensitive to elevated pCO2 than it is for the coral reef communities they build (Pandolfi et al. 2011; Comeau et al. 2015; 2016).

It is unknown, however, whether the greater sensitivity of coral reef community Gnet to ocean acidification, relative to coral colonies, reflects methodological artifacts, or is the result of emergent properties of the communities, for example, the propensity of some of their inorganic components for dissolution (Andersson and Gledhill 2013; Eyre et al. 2014). Unlike the laboratory experiments that typically are employed to study the response of corals to elevated pCO2, when in situ approaches are used to evaluate the relationship between community Gnet for coral reefs and pCO2, it is difficult methodologically to control seawater pCO2 or pH precisely to simulate future conditions (although see Albright et al. 2016). The other conditions that are likely to covary with pCO2 include light availability, seawater temperature, nutrient concentrations, and flow regimes, all of which could affect the response of community Gnet to seawater carbonate chemistry (Comeau et al. 2014b). The quality and quantity of light, for example, affects community net photosynthesis (Pnet) on coral reefs, which is a driver of community Gnet (Kinsey 1985; Falter et al. 2008), but community Pnet also affects seawater carbonate chemistry through the uptake of dissolved inorganic carbon [(DIC), Anthony et al. 2008b; Shaw et al. 2012]. Together, these effects cause community Gnet to be associated with seawater carbonate chemistry, light availability, and community Pnet (Falter et al. 2012).

Rates of primary production on coral reefs primarily are driven by light availability, but they can be modified by additional factors such as temperature, nutrient availability, and water flow (Dennison and Barnes 1988; Atkinson et al. 1999; Langdon and Atkinson 2005). Pnet and Gnet typically are associated strongly for both organisms and communities on minute-to-hourly time scales, in part because photosynthesis provides fixed carbon that can be used to supply metabolic energy supporting the costs of calcification (Allemand et al. 2011; Frieder et al. 2016), and it can also create a pH environment favoring calcium carbonate deposition (Allemand et al. 2011). For coral reefs, evidence that ocean acidification affects Pnet at any level of biological organization is equivocal (Anthony et al. 2008a; Crawley et al. 2010; Dove et al. 2013; Comeau et al. 2016); however, the relationship between community Pnet and Gnet can be altered by ocean acidification (DeCarlo et al. 2017), reflecting a disruption of the coupling between organic and inorganic carbon fixation.

In situ analyses of coral reef community metabolism (i.e., Gnet and Pnet) typically use short experiments in which measurements employing the alkalinity anomaly techniques (Kinsey 1978; Smith and Kinsey 1978) are completed over minutes-to-days (Gattuso et al. 1996; Shaw et al. 2015; DeCarlo et al. 2017). This resolution reflects the need to constrain some experimental work to periods with reduced seawater flow (e.g., slack water), but it also is important to detect rapid changes in community Gnet, such as how it occurs in response to diel variation in carbonate chemistry (Ohde and van Woesik 1999; Shaw et al. 2012). Other approaches (e.g., Lagrangian transects) quantify metabolic rates over one to several days (e.g., Kinsey 1985; Gattuso et al. 1996). To date, however, there is limited evidence of the extent to which short-term experiments and measurements are indicative of responses occurring over months-to-years (Kroeker et al. 2013). The present study is part of a larger project to measure the metabolic response of coral reef communities to elevated pCO2 over ecologically relevant time scales, and here we present the results from the first 4 months of this experiment. Working with a benthic community from the back reef of Moorea, French Polynesia, we present the results of an experiment in which we measured community Gnet and Pnet under four pCO2 regimes from late Austral spring to early Austral autumn. Using community Gnet and Pnet as dependent variables, we tested the hypothesis that the effects of pCO2 are consistent over times-scales ranging from 1 day to 4 months. As a means to test our hypothesis, we also explored the effects of pCO2 on diel variation in community Gnet and Pnet as well as the degree to which these variables were associated.

Methods

Coral reef communities were assembled in four, outdoor flumes in Moorea, French Polynesia. Logistical constraints of constructing, instrumenting, and maintaining additional flumes precluded an experimental design with replicate flumes for each pCO2 treatment. Instead of using a factorial experimental design, we employed a regression approach to explore the community level responses to increasing pCO2, and to evaluate whether those responses varied over time. Each flume was assigned randomly seawater carbonate chemistry conditions created by four pCO2 regimes, targeted at ambient (400 μatm), 700, 1000 and 1300 μatm pCO2, and treatments were maintained for 4 months from late Austral spring (November 2015) to early Austral autumn (March 2016); actual pCO2 treatments differed slightly from target values (described below). The working section of each flume was filled to 30-cm depth with seawater, and the communities were assembled in each of the 5.0 m × 0.3 m working sections. The flumes (described in Comeau et al. 2015) contained ~ 500 L of seawater that was circulated continually and augmented with fresh seawater at ~ 5 L min−1. Seawater was pumped from 14-m depth in Cook’s Bay, and filtered through sand (nominal pore size 450–550 µm) prior to flowing into the flumes. This effectively allowed smaller particles to enter the flume and serve as a potential food source for heterotrophic organisms.

Preparation of back reef communities in flumes

Coral reef communities were assembled to correspond to the mean percent cover of the major benthic space holders on the back reef of Moorea in 2013 (Fig. S1) (Carpenter 2016a; Edmunds 2016) and the communities were similar to those we have tested previously for response to elevated pCO2 over shorter periods (≤ 56 days) (Comeau et al. 2014b; 2015). Each flume contained ~ 25% coral cover, comprised of 11% cover of massive Porites spp., 7% Porites rus, 4% Montipora spp. and 3% Pocillopora spp. There was ~ 7% cover of crustose coralline algae (CCA), with 4% Porolithon onkodes and 3% Lithophyllum kotschyanum, and ~ 5% cover of small pieces (i.e., ~ 1-cm diameter) of coral rubble (Fig. S2).

The working section of each flume included a 2.4-m long sediment chamber that extended the width of the flume, and contained 30-cm depth of sand collected from the back reef. The sediment chamber was in the center of the working section of the flume, and was flanked on either end by 1.3 m of the fiberglass floor of the flume. In addition to the live coral, sand, CCA, and rubble, the flumes also included three small holothurians (~ 8-cm long, Holothuria leucospilota), and thalli of the macroalgae Turbinaria ornata and Halimeda minima to approximate the cover of these algae in the back reef in 2013. The cover of these algae, therefore, were targeted at 11 and 1% (respectively), but the necessity of replacing them consistently throughout the incubation resulted in periodic and substantial downward variability in macroalgal cover in each flume. As the objective was to create ecologically relevant back reef communities, and expose them for multiple months to treatment conditions, the communities were not standardized for the abundances of small (i.e., ≤ 1-cm diameter) taxa associated with coral reef communities. We rationalized this approach by the large size of the flume communities, the strong likelihood that effectively they sampled the same community from which they were collected, and the difficulty of ensuring that 1.2 m2 areas (and sediments to 30-cm depth) of experimental reef were identical biologically.

Corals, CCA, and rubble were collected from < 2-m depth in the back reef of Moorea. After collection, corals and CCA were epoxied to plastic bases, and placed in a seawater table for 3 days before being installed in the flumes under ambient seawater carbonate chemistry conditions. Sediments were collected from the back reef at ~ 2-m depth, and were placed into plastic sediment boxes (0.4 × 0.3 × 0.3 m) that were left in situ (i.e., buried in sand with their upper surface at the natural sand level) for 3 days to promote natural sediment stratification (after Comeau et al. 2015). The sediment boxes then were inserted into the flume in the same vertical orientation to maintain sediment stratification. Reef communities were assembled in the flumes on 12 November 2015, and were maintained under ambient conditions for 5 days (until 17 November), when treatment pCO2 levels were applied to three randomly selected flumes (the fourth remained at ambient pCO2). Stable treatment levels were established within a 24-h period.

Physical and chemical parameters

The flow of seawater in each flume was maintained at ~ 0.1 m s−1 using a pump (W. Lim Wave II 373 J s−1). Flow speeds were measured across the working sections of the flumes using a Nortek Vectrino Acoustic Doppler Velocimeter, and the flow speed was selected to match representative flow speeds across the back reef of Moorea (Hench et al. 2008). The flumes received natural sunlight that was reduced with neutral density screening to light levels similar to those experienced at 2-m depth in the back reef (Carpenter 2016b). Maximum daily photosynthetically active radiation (PAR, 400–700 nm) typically ranged between 1000 and 2000 μmol quanta m−2 s−1 (Fig. 1), with strong variation attributed to seasonal and daily weather. Temperatures in each flume were controlled individually with dedicated chillers, and were maintained to approximate the mean monthly seawater temperature in the back reef of Moorea. They were increased from ~ 27 °C in November to ~ 30 °C in March (Edmunds et al. 2010). Temperatures were ramped slowly (1 °C/month) between seasons.

Fig. 1
figure 1

Daily integrated PAR in each flume from 14 November 2015 to 15 March 2016 (n = 123 days). PAR was measured with a cosine-corrected sensor recording at 0.6 mHz, and daily values were obtained by integrating records for each 24-h period. a Ambient flume (344 µatm pCO2), b flume maintained at ~ 633 µatm pCO2, c flume maintained at ~ 870 µatm pCO2, and d flume maintained at ~ 1146 µatm pCO2. Vertical dashed lines indicate dates on which metabolism measurements were made

Full size image

Seawater carbonate chemistry was uncontrolled in one flume (ambient, ~ 344 μatm pCO2) and controlled in the other three to simulate conditions arising from targeted pCO2 values of 400, 1000 and 1300 μatm. The mean pCO2 values achieved in these treatments were 344, 633, 870, and 1146, respectively. Treatments were selected to: (a) span the range of atmospheric pCO2 values expected by the middle of the next century assuming a pessimistic scenario for anthropogenic effects (IPCC 2014), and (b) to facilitate an ANCOVA statistical approach, in which pCO2 was as a covariate, Gnet and Pnet were response variables, and time was a fixed effect. Seawater pH was not altered in the ambient flume, but was controlled in the treatment flumes (as a means to manipulate seawater pCO2), with the bubbling pure CO2 into the seawater. The desired pH set point (and corresponding pCO2 set point) was maintained with the use of a pH–stat (Aquacontroller, Neptune systems, USA), which controlled the rate of bubbling of CO2 through a solenoid which was opened when the attached pH electrode measured an increase in pH above the set point. A diurnal upward adjustment of ~ 0.1 pH unit was applied to the three treatment flumes to simulate current levels of diurnal variability in the back reef of Moorea (Hofmann et al. 2011); the ambient flume (which was not regulated for pCO2) maintained a similar diel cycle with nighttime pH ~ 0.1 lower than daytime values (Table 1). Ambient air was bubbled continuously into all flumes (except during dissolved oxygen measurements) and there was no downward regulating of pCO2 at night. We reasoned that the effects of nocturnal respiration in generating pCO2 was a component of natural reef metabolism, and that extreme swings in pCO2 due to this effect would be minimized by the constant in-flow of fresh seawater. Periodic measurements of nocturnal pCO2 in the flumes confirmed that nocturnal pCO2 exceeded daytime target values.

Table 1 Mean (± SE) of physical and chemical parameters measured/calculated in the flumes from 17 November 2015 to 15 March 2016
Full size table

Throughout the experiment and in all flumes, logging sensors (sampling every 30 min) recorded PAR (using cosine-corrected PAR loggers, Dataflow Systems Ltd, Christchurch, New Zealand), and temperature [Hobo Pro v2 (± 0.2 °C), Onset Computer Corp., Bourne, MA]. pH, measured on the total hydrogen ion scale (pHT), was monitored daily at the downstream end of the flumes with a handheld electrode (described below). Based on the values relative to target values, the thermostat and pH–stat set points were adjusted as necessary to maintain stable treatments. Seawater carbonate chemistry and salinity were measured weekly, both in the day (02:00) and night (20:00); salinity was measured using a bench-top conductivity meter (Thermo Scientific, Orionstar A212, Waltham, MA, USA). The parameters of the seawater carbonate system were calculated from measurements of temperature, salinity, pHT, and AT, using the R package seacarb (Lavigne and Gattuso 2013). Calculations were made using the carbonic acid dissociation constants of Lueker et al. (2000), the KSO4 concentration for the bisulfate ion from Dickson (1990), and the Kf constant of Perez and Fraga (1987).

pHT was measured using a DG 115-SC electrode (Mettler-Toledo) that was calibrated with a TRIS buffer (SOP 6a Dickson et al. 2007). Total alkalinity was measured potentiometrically using the open-cell method of acidimetric titration (SOP 3b, Dickson et al. 2007) using certified acid titrant (from the Dickson lab) with a T50 Mettler-Toledo automatic titrator fitted with a DG 115-SC electrode (Mettler-Toledo). The accuracy and precision of these measurements were determined through analysis of certified reference materials (CRMs; from A. Dickson Laboratory, Scripps Institution of Oceanography). The mean (± SE) difference between measured and certified values (i.e., accuracy) was 1.7 ± 0.3 μmol kg−1 (n = 15) and the mean (± SE) precision of AT analyses based on duplicate samples was 1.8 ± 0.1 μmol kg−1 (n = 475).

Community metabolism

Community Gnet and Pnet for the reef communities in the flumes initially were measured under ambient seawater carbonate chemistry conditions on two occasions on 14 and 16 November, 2015. The pCO2 treatments then were initiated (on 17 November 2015), and community Gnet and Pnet were measured on 17–18 November 2015. Thereafter, community Gnet and Pnet were measured biweekly or monthly until March 2016.

Community Gnet was measured using the alkalinity anomaly method (Eq. 1, after Smith (1973)), and community Pnet was measured using changes in dissolved oxygen (DO) (Eq. 2):

$$ G_{\text{net}} = \frac{{ - 0.5\Delta A_{\text{T}} V\rho }}{{\Delta t {\text{SA}}}} $$
(1)
$$ P_{\text{net}} = \frac{{\Delta {\text{DO}}V\rho }}{{\Delta t{\text{SA}}}}, $$
(2)

where ΔAT is the change in total alkalinity (μmol kg−1), V is the volume of seawater in the flumes (L), ρ is the density of the seawater (kg L−1), Δt is the time interval of the incubations (h), SA is the planar area of the floor of the working area of the flumes (m2), and ΔDO is the change in dissolved oxygen concentration over the incubation time (μmol kg−1 h−1).

The addition of fresh seawater into the flumes was halted during each measurement of community Pnet and Gnet, so that changes in AT and dissolved oxygen (DO) for the fixed volume of water within the flumes could be used to calculate Pnet and Gnet. Circulation of seawater within the flumes was maintained at a constant velocity (0.1 m s−1) throughout the measurement period. The exchange of O2 across the air–seawater interface at the surface of the flumes was assumed to be negligible during the incubations based on the rationale that UV-transparent acrylic covers that were placed over the flumes to prevent access by rain, and also greatly reduced wind movement at the seawater surface in the flumes. Incubations were designed to be implemented with four consecutive measurements of Pnet and Gnet in each flume during the day, with each incubation lasting ~ 3 h, and two determinations of respiration (R) and Gnet at night, with each incubation lasting ~ 6 h. This protocol was consistent with our previous work (Comeau et al. Comeau et al. 2014a, b; 2015), but during the 4-month experiment, logistical constraints resulted in slight departures from this design, in the number of replicates during the day and night, incubation duration, and the start and end of incubations relative to sunrise and sunset. Between incubations within a measurement day, flumes were flushed with seawater for ~ 30 min at 5 L min−1, which ensured that 25–30% of the seawater within each flume was replaced before the next incubation began. AT was measured at the start and end of each incubation as described above, and DO was recorded each minute with MiniDOT DO loggers (Precision Measurement Engineering, Inc., Vista, CA, USA). DO sensors were calibrated by the manufacturer.

Measurements of community Gnet and Pnet over 3-h periods for daytime (n = 4) and 6-h periods for nighttime (n = 2) were averaged, and used to estimate daily community Gnet (over 24 h) and daytime community Pnet (over ~ 12 h). As Gnet and Pnet predictably vary through the day, missing values from the set of six possible daily incubations biased the daily averages. To adjust for this effect, missing values were replaced with values calculated by interpolation using empirical, best-fit relationships of mean hourly Gnet and Pnet against time for each treatment. Third-order polynomials were used to describe the relationships between Gnet and time, and fourth-order polynomials were used to describe the relationships between Pnet and time (described in results). Daytime community Gnet and Pnet were calculated from the mean of the four, ~ 3-h incubations during the day, and scaled to the mean local day length by month: 12.93 h in November 2015, 13.15 h in December 2015, 12.03 h in January 2016, 12.63 h in February 2016, and 12.17 h in March 2016. Nighttime community Gnet was calculated from the mean of the two, 6-h night incubations scaled to the length of the night. Daily (24 h) values of Gnet were derived from the sum of the day and night rates, and Pnet was calculated only for daytime.

Statistical analyses

To test the hypothesis that daily community Gnet and Pnet varied among treatments over 4 months, ANCOVA was used where pCO2 was the covariate, and time (sampling periods) was a fixed effect. The assumption of linearity for each treatment was tested using linear regressions, and an effect of time was detected as differences in slopes or elevations (intercepts) among incubations. Post hoc tests of levels within main effects were conducted using Tukey’s HSD. To evaluate the role of Pnet in modulating Gnet on a scale of hours, the association between hourly Pnet and Gnet for each treatment was tested with Pearson correlations. Where these associations were significant, best-fit linear regressions were fit by Model I techniques, because the purpose was prediction of Gnet from Pnet (Sokal and Rohlf 2012). Pearson correlation also was used to test for an association between Gnet and Ωar, and the predictive relationship between the two was described with Model I linear regression. Statistical analyses were performed using Systat 13 software, and statistical assumptions of ANCOVA (normality and equal variances) were tested using graphical analyses of residuals.

Results

Physical and chemical conditions

Seawater temperature, salinity, and carbonate chemistry for each flume are reported in Table 1, and are separated by day, night, and each 24 h day. The temperature logger in one flume (870 µatm) failed and continuous records from this flume were lost. Spot measurements with a handheld meter showed that the temperature of this flume closely tracked the temperature of the other three flumes. Seawater temperature was similar among the three flumes with continuous records, and in all cases the mean daily temperatures were ~ 28.4 °C (Table 1). Seawater temperature in the flumes varied slightly throughout each day, with solar warming and nocturnal cooling, so that mean daytime temperatures were up to 0.3 °C higher than mean nighttime temperatures (Table 1). Temperature in the flumes tracked seasonal warming throughout the austral summer, and increased from ~ 27.5 °C in November to ~ 29.0 °C in March at ~ 0.014 °C day−1 (r2 = 0.790).

Salinity was consistent at 35.8 ± 0.1 psu among flumes throughout the experiment (Table 1). PAR varied among days in response to local weather (Fig. 1), and while it did not vary strikingly throughout the summer, it was reduced by clouds and rain in February and March 2016. Mean daily integrated PAR was 17.1 ± 0.6 mol quanta m−2 day−1 in the 344 µatm pCO2 flume, 20.1 ± 0.6 mol quanta m−2 s−1 in the 633 µatm pCO2 flume, 15.8 ± 0.5 mol quanta m−2 s−1 in the 870 µatm pCO2 flume, and 16.9 ± 0.6 mol quanta m−2 day−1 in the 1146 µatm pCO2 flume (± SE, n = 120). PAR differed among flumes (F = 10.990, df = 3, 476, P < 0.001) and was higher in the 633 µatm flume compared to the other flumes (Scheffé post hoc, P ≤ 0.001). Variation in PAR between flumes arose from variable shading from the structure supporting the shade cloth above the flumes. AT of seawater was similar among flumes, with mean daily values ranging from 2315 to 2362 μmol kg−1 (Table 1), and mean daytime values ≤ 0.3% lower than mean nighttime values (Table 1). Actual pCO2 treatments departed from targeted values, with discrepancies strongest during the day and smallest at night (Table 1). pCO2 differed among flumes for daily, daytime, or nighttime values (F ≥ 12.879, df = 1.66, P ≤ 0.001); actual treatments contrasted mean pCO2 values of 344, 633, 870 and 1146 µatm, which corresponded to mean Ωar values of 4.26, 2.92, 2.31, and 1.90, respectively (Table 1).

Community metabolism prior to pCO2 treatments

Prior to establishing pCO2 treatments in the flumes (i.e., ~ 344 µatm pCO2), mean (± SD) daily Gnet (n = 2 days) was similar among flumes, ranging from 85 ± 18 to 92 ± 5 mmol CaCO3 m−2 day−1 (Fig. 2). Mean (± SD) daytime Pnet was more variable among flumes, ranging from 42 ± 3 (344 µatm pCO2 flume) to 60 ± 3 mmol O2 m−2 day−1 (1146 µatm pCO2 flume) (Fig. 2).

Fig. 2
figure 2

Community net calcification (Gnet) and primary productivity (Pnet) of reef communities assembled in four flumes and incubated for 4 months at pCO2 values of ~ 344 µatm (ambient), ~ 633 µatm pCO2, ~ 870 µatm pCO2, and ~ 1146 µatm pCO2 (Table 1). a Values before pCO2 treatments were initiated (n = 2, ± SD), b 17 November, c 28 November, d 15 December, e 3 January 2016, f 15 January 2016, g 15 February 2016, and h 15 March 2016. Logistical constraints resulted in some lost incubations (Gnet in flume 3 between 15:00 and 18:00 h on 15 December, Gnet in all flumes between 15:00 and 18:00 on 15 January; Pnet in all flumes between 09:00 and 12:00 h on 3 January, and between 15:00 and 18:00 h on 15 January), and these values were interpolated using the relationships shown in Fig. 3; there was only one night time incubation on 28 November, but this lasted ~ 11 h. Lines are best-fit Model I regressions with slopes (b) and proportion of variance explained (r2) reported

Full size image

Diel patterns of G net and P net

Hourly community Gnet and Pnet showed consistent diel responses, increasing during the morning to maxima near the solar zenith and decreasing throughout the afternoon (Fig. 3). To describe the shape of these relationships and develop predictive capacity for Gnet and Pnet as a function of time of day, the data were fitted with third- or fourth-order polynomials. The use of these relationships is not intended to imply any explicit type of functional relationship. For Gnet, the relationships with median incubation time were well fit by third-order polynomials (r2 ≥ 0.900), and while the shapes of the responses were similar across treatments, maximum values differed among flumes, with the highest value under ambient pCO2, and lowest under the highest pCO2 (Fig. 3). For Pnet, the relationships with median incubation times were well fit with fourth-order polynomials (r2 ≥ 0.976), but for this dependent variable, the shapes of the relationships and the maximum values of Pnet were very similar across treatments. The overall shape of the diel response curve for community Gnet was broader than for community Pnet, with Gnet increasing earlier in the day and maintaining positive values later in the day, while Pnet increased and decreased steeply on either side of the solar zenith and was negative both early and late in the day, reflecting respiration.

Fig. 3
figure 3

Hourly Gnet, Pnet and R for the four flumes operated with pCO2 treatments from 17 November 2015 to 15 March 2016. Values are mean (± SE) of all determinations and are plotted against the mid point of the intervals the incubations targeted for sampling (00:00–06:00, 06:00–09:00, 09:00–12:00, 12:00–15:00, 15:00–18:00, and 18:00–24:00). Curves are third-order polynomials for Gnet (r2 ≥ 0.899), and fourth-order polynomials for Pnet and R (r2 ≥ 0.976). n = 6–7 for Gnet and 5–7 for Pnet and R

Full size image

Effect of treatment duration on net community calcification

Following initiation of the pCO2 treatments, the responses of the communities were measured on 17/18 November (time = 1 day), 28 November (time = 12 days), 15/16 December (time = 29 days), 3/4 January (time = 48 days), 15/16 January (time = 60 days), 15/16 February (time = 91 days), and 15/16 March (time = 120 days). During these periods, in the 344 µatm flume, hourly Gnet ranged from − 0.47 to 8.37 mmol m−2 h−1, and hourly Pnet from − 10.67 to 13.97 mmol m−2 h−1. In the 633 µatm flume, hourly Gnet ranged from − 1.63 to 7.36 mmol m−2 h−1, and hourly Pnet from − 11.76 to 16.66 mmol m−2 h−1. In the 870 µatm flume, hourly Gnet ranged from − 0.86 to 7.45 mmol m−2 h−1, and hourly Pnet from − 7.07 to 12.75 mmol m−2 h−1. In the 1146 µatm flume, hourly Gnet ranged from − 0.53 to 7.31 mmol m−2 h−1, and hourly Pnet from − 8.79 to 16.40 mmol m−2 h−1. Hourly Gnet and Pnet calculated from all of the incubations conducted in the day and the night were positively correlated in all four flumes (Fig. 4) (r ≥ 0.689, df = 48–48, P < 0.001). Results from factorial ANCOVA indicated that the slopes of these relationships were similar between treatments (b = 0.247–0.279), but they differed in elevation and were depressed by high pCO2.

Fig. 4
figure 4

Relationships between hourly coral reef community Gnet and Pnet for incubations completed in the day (filled circles) and the night (open circles) over the 4-month experiment. a 344 µatm pCO2 flume, b 633 µatm pCO2 flume, c 870 µatm pCO2 flume, and d 1146 µatm pCO2 flume. Lines are best-fit Model I regressions with their slopes (b), the proportion of variance explained (r2) and sample size (n) shown

Full size image

Over the 4-month period, community metabolism in the flumes was measured seven times under treatment conditions. During these measurements, logistical constraints resulted in five missing Gnet values (18% of planned measurements), and one of the paired nighttime values was replaced with a single determination over ~ 12 h when the intermediate 6 h measured was omitted. Likewise, there were eight missing Pnet values (29% of planned measurements). Missing values were interpolated using best-fit polynomial relationships for each flume (described above and in Fig. 3), and the complete set of measurements (empirical plus interpolated values) were used to calculate daily community Gnet and Pnet (Fig. 2b–h). Daily community Gnet was negatively correlated with pCO2 on four of seven sampling periods (with strong trends for two additional periods, Fig. 2, Table 2). On 17 November, Gnet was unaffected by pCO2, on 28 November, there was a trend (P = 0.069) for Gnet to decline with increasing pCO2, Gnet declined (P ≤ 0.036) with increasing pCO2 on 15 December, 3 January, 15 January, and 15 February, and again there was a trend (P = 0.053) for Gnet to decline with pCO2 on 15 March (Table 2). For the periods when Gnet declined with increasing pCO2, the slopes of these relationships ranged from − 0.020 ± 0.004 mmol m−2 day−1 µatm pCO −12 to − 0.095 ± 0.001 mmol m−2 day−1 µatm pCO −12 . In contrast to community Gnet, community Pnet was unaffected by pCO2 for any of the seven sampling times when it was measured (Fig. 2, Table 2).

Table 2 Results of least squares linear regression analysis (Model I) of daily community Gnet and Pnet against flume treatments (pCO2) for seven periods (Time) over which treatments were sustained (Fig. 3)
Full size table

For the relationships between community Gnet and pCO2, a contrast of the four significant relationships demonstrated that the slopes were heterogeneous (F = 15.262, df = 3.8, P = 0.001), with the relationships on 15 January relatively steeper than 15 December (P = 0.022), and 3 January (P = 0.001), and the relationship on 3 January relatively steeper than on 15 February (P = 0.039); no other contrasts were significant (P ≥ 0.057). Although a contrast of all seven regression lines of community Gnet against pCO2 must be treated with caution, since three of the lines are not significant (Table 2), the analysis nevertheless suggests that the slopes differ among times (F = 5.037, df = 6.14, P = 0.006). Post hoc contrasts indicate that the slope of the relationship on 3 January is steeper than the slope on 18 November (P = 0.034), but less steep than the slope on 15 January (P = 0.010); no other contrasts were significant (P ≥ 0.141).

When averaged over all measurement periods (17 November 2015 to 15 March 2016), daily community Gnet was associated positively with Ωar (Fig. 5) (F = 376.448, df = 1.2, P = 0.003), such that Gnet declined with decreasing Ωar. The proportional effect differed depending on the initial value of Ωar, but, for example, this relationship suggests that a unit decline in Ωar from 4.0 is associated with a 24% reduction in Gnet, but a 45% reduction from a Ωar value of 3.0.

Fig. 5
figure 5

Relationship between daily integrated community Gnet and aragonite saturation state (Ωar) for each of the four experimental flumes maintained at a different pCO2. Values shown are mean ± SE (n = 15 for Ωar 4.25, all others, n = 7). Gnet and Ωar were associated significantly (r = 0.998, df = 2, P = 0.002), and the line shows the Model I least squares linear regression that is used to describe the functional relationship between the variables

Full size image

Discussion

The major questions addressed by this study were: (1) how are Gnet and Pnet of reef communities affected by a range of potential ocean acidification conditions (700–1300 uatm), (2) Does OA affect the relationships between community Gnet and Pnet, and (3) are community responses of Gnet and Pnet to ocean acidification consistent over an extended period of time (4 months)? Before the pCO2 treatments were established, Gnet and Pnet were broadly similar among flumes (zero slopes across treatments), indicating that the communities that were assembled within them to be similar to one another, and to the back reef community of Moorea in 2013, also were similar to one another in terms of their community metabolism. This outcome was striking even though the flumes differed in light intensity and likely, in the composition of the infauna residing within the sediment. After treatments began, Gnet responded quickly and negatively to elevated pCO2, with an approximate reduction of 70% in the highest pCO2 treatment compared to the ambient pCO2 at the conclusion of this experiment (a contrast between the extremes of our pCO2 treatments). While comparisons among experiments might result from variation in the sensitivity of organismal physiology and/or variation in actively calcifying area, this reduction in Gnet is consistent with results of previous experiments in which back reef communities similar to those studied here were exposed to ~ 1000–1300 µatm pCO2. Working with a nearly identical community in Moorea, Comeau et al. (2015), for example, started with community Gnet under ambient pCO2 conditions (ca. 76 mmol m−2 day−1) that was within the range of values recorded here for similar pCO2 conditions (~ 70–110 mmol m−2 day−1), but they described a 59% decrease in community Gnet at 1318 versus 456 µatm pCO2 over 8 weeks. Working with patch reef communities from Heron Island, Australia, Dove et al. (2013) reported a greater decrease in Gnet for experimental reef communities maintained at different pCO2 levels over 3 months, with net calcification declining from ~ + 3 g day−1 at 405 µatm pCO2 and 26.1 °C, to ~ − 1.4 g day−1 (i.e., dissolution occurred) at 1009 µatm pCO2 and 30.2 °C. The stronger negative effect of high pCO2 in Dove et al. (2013) (versus our study, Fig. 2), most likely was due to their implementation of a broader test of future environmental conditions in which both pCO2 and temperature were elevated. In our experiment, the effects of pCO2 treatments on Gnet also were evident on diurnal time scales, where Gnet was reduced throughout the day with a consistent monotonic, negative relationship with pCO2. Gnet in all treatments increased rapidly in the morning, peaked at midday, and then gradually decreased in the afternoon and evening. One reason for this strong diurnal response is that it might reflect the requirements for, and the capacity to supply, the metabolic energy necessary to support higher rates of Gnet, with the parsimonious source of this energy being community Pnet earlier in the day. Gnet was positive throughout the day and became negative during the night in all treatments.

There was no evidence that Pnet was affected by high pCO2. It appeared that Pnet was more variable over time than between treatments, likely due to variation among measurement days in PAR. Several previous studies also have reported no effect of OA on Pnet for coral reef communities (Leclerq et al. 2002; Langdon and Atkinson 2005; Dove et al. 2013; Comeau et al. 2015). These results suggest that photosynthesis of coral reef primary producers is not “fertilized” by the additional CO2 and HCO3 available under reduced seawater pH conditions, at least over the time scales of our measurements (but see e.g., Noonan and Fabricius 2016). There was a consistent and positive relationship between Pnet and Gnet expressed on an hourly basis. Generally, it is presumed that photosynthesis stimulates Gnet in at least two ways. First, photosynthesis produces energy (ATP) and compounds that can be used to support calcification, either to fuel calcium uptake, proton pumping (Comeau et al. 2017), and/or the construction of an organic matrix on which calcium carbonate is laid down (Allemand et al. 2011; Tambutté et al. 2011). Additionally, photosynthesis elevates the local pH, which can create chemical conditions that facilitate calcification (Comeau et al. 2017). As a result, on hourly time scales, Pnet may drive Gnet directly. Since Pnet is closely coupled with PAR, Gnet too will be driven (indirectly) by PAR, although the effect of PAR on Pnet and Gnet is evident only on short time scales (hours), and is unlikely to appear over daily periods (i.e., involving the integrated effects of PAR). This suggest that the coupling between Pnet and Gnet is related to processes occurring over short time scales (e.g., energy production, pH alteration), rather than the overall balance between the long-term demands for metabolic energy and the capacity to supply it through the catabolism of short- and long- term food reserves, all of which create time lags in the hysteresis pattern between Gnet and PAR.

The relationships between Gnet and Pnet were not affected by the pCO2 treatments, as the slopes of these relationships did not differ. The slope of Gnet regressed on Pnet for coral reefs typically are related to the composition of the benthic community (Page et al. 2016), with higher values for communities dominated by calcifying taxa, and lower values for communities dominated by fleshy macroalgae (Lantz et al. 2014). The values of the Gnet/Pnet slopes [unit-less and often given as NEC/NEP (e.g., Shaw et al. 2015)] in the present study (0.247–0.279) are within the range of those reported for other coral reefs having a community structure of scleractinians and algae similar to that employed herein (Suzuki and Kawahata 2003). The elevations of these relationships, however, were affected by pCO2 and were increasingly lower, as pCO2 increased from 344 to 1146 µatm.

Because our experimental reefs contained a combination of calcifying and non-calcifying organisms together with reef sediments, the responses of community Gnet to pCO2 reflects the combined effect of elevated pCO2 on organismal calcification, dissolution of organism carbonate skeletons, and abiotic calcification/dissolution of inorganic carbonate sediments (Eyre et al. 2014). Results from mesocosm experiments on the effects of OA on organismal Gnet for several coral and calcified algal species suggest that the effects of elevated pCO2 are variable, but generally negative (Chan and Connolly 2013; Kroeker et al. 2013; Comeau et al. 2014a, b), and this reduction in organism Gnet is a combination of reduced calcification, enhanced bioerosion, and skeletal dissolution. In a very similar experiment to that reported here, ~ 50% of the reduction in community Gnet under elevated pCO2 was due to dissolution of sediments, mostly at night (Comeau et al. 2015). For this reason, extrapolation of OA effects from organism-scale experiments to the community scale are problematic, and the important role of sediments needs to be incorporated when attempting to predict responses at the reef scale (Eyre et al. 2014; Edmunds et al. 2016). This may be less problematic on coral reefs with a relatively low diversity of benthic calcifiers where net calcification of component taxa and functional groups can be scaled with a high degree of accuracy to estimate net community calcification (Courtney et al. 2016).

To date, little is known about the extended effects (i.e., over months or years) of OA on organism or community primary production or calcification (Kroeker et al. 2013). To test hypotheses regarding these effects, we established a multi-month OA experiment to examine whether OA effects on Gnet and Pnet are time-dependent. With our results to date, and based on previous experiments, we can evaluate the support for three alternative hypotheses with respect to how OA effects are manifested through time. The first hypothesis (the null) is that the slopes of community Gnet and Pnet as a function of pCO2 are negative and do not change over time. This would indicate that OA effects on reef metabolism are negative (based on prior results) and constant. A second hypothesis is that the slopes of community Gnet and Pnet with pCO2 increase over time, indicating that OA effects are cumulative and have additive (or multiplicative) effects. A third alternative hypothesis is that slopes of community Gnet and Pnet with pCO2 are negative, but the slope of the relationships became less steep over time. This could be construed as evidence of acclimatization of organisms in the community to OA, and/or amelioration of negative effects by longer-term community processes (e.g., sediment dissolution and local increases in buffering capacity).

For the first hypothesis, the community Pnet was unaffected by pCO2, and thus, the slope of Pnet on pCO2 was zero initially, and did not change over 4 months. This result is consistent with previous studies, based on which a recent meta-analysis concluded that photosynthesis in a variety of tropical marine taxa (corals, coccolithophores, fleshy macroalgae, and seagrasses) was unaffected by OA over any of the time scales tested (Kroeker et al. 2013). In the test of the second hypothesis, our results for the response of Gnet to pCO2 were very different, however, for while Gnet did not vary among flumes at the start of the experiment (i.e., before treatments were applied), it was strongly affected by pCO2 throughout the study, with Gnet-pCO2 slopes differing among the months. The steepest slopes were recorded on 15 January and 15 March, the shallowest slopes on 28 November and 15 February, and the seven samplings created a ranking of sensitivities of Gnet to pCO2 of 16 January > 15 March > 15 December > 3 January > 17 November > 28 November > 15 February. This ranking does not conform well to any of our three a priori hypotheses (described above), although the responses consistently were negative (an outcome encapsulated in our first alternative hypothesis). Critically, our results support neither an intensifying (Ha 2 above) nor an attenuating (our Ha 3 above) model of coral community sensitivity of Gnet and Pnet to pCO2 over time scales as long as 4 months. However, the labile response of the Gnet–pCO2 relationship of Moorea back reef communities suggests their net calcification is sensitive to seasonally varying environmental conditions. In this regard, the day-to-day, monthly, and seasonal variation in integrated PAR (Fig. 1) in the flumes is high, and it appears that the lowest sensitivity of Gnet to pCO2 was coincident with weather patterns that reduced PAR through clouds and rain (i.e., in late November and middle February). Likewise, the greatest sensitivity of Gnet to pCO2 was coincident with high PAR in the middle of January, December, and March. These patterns of variation in the sensitivity of Gnet to pCO2, and the inferred association with PAR, are consistent with past results at the organismal level for tropical scleractinian corals and calcified algae. Critically, our previous work suggest that calcifying organisms that are calcifying at high rates, for example, as occurs under high PAR, are the most sensitive to OA (Comeau et al. 2014a), and the present study suggests a similar relationship may extend to reef communities and affect Gnet.

In summary, the results of this experiment begin to provide temporal rigor to the emerging understanding of the effects of OA on coral reef organisms and the communities they build, specifically by extending the duration of exposure to predicted future seawater conditions to 4 months. Relative to the effects of OA on coral reefs that already have been measured, our results provide more nuanced answers to the three questions at the core of this study. First, we demonstrate that Gnet of an experimental back reef community was related negatively to pCO2 over a four-month period, while community Pnet was unaffected by the same treatments (Fig. 2). Second, we show that the relationships between community Gnet and Pnet were not affected by treatment, although the elevations of these relationships were reduced progressively with increases in pCO2 (Fig. 4). Finally, we show that the effect of elevated pCO2 on community Gnet consistently was negative over seven consecutive measures extending over the 4 months, with no evidence of increases or decreases in sensitivity (slope) with time (Fig. 4). However, the sensitivity of Gnet to OA appeared to be modulated, at least in part, by variation in light. Overall, our results highlight the utility of using experimental reef communities under ecologically relevant environmental conditions to test hypotheses about the immediate and time-dependent effects of OA on coral reef function. Further work over longer time scales is warranted to explore the possibilities that longer-term effects of OA on reef communities are less striking than those arising from short-term, acute exposures, and moreover, if reef communities exhibit the ability to acclimate to OA conditions.