www.biogeosciences.net/7/2509/2010/ doi:10.5194/bg-7-2509-2010 © Author(s) 2010. CC Attribution 3.0 License.

Abstract. Despite the potential impact of ocean acidification on ecosystems such as coral reefs, surprisingly, there is very limited field data on the relationships between calcification and seawater carbonate chemistry. In this study, contemporaneous in situ datasets of seawater carbonate chemistry and calcification rates from the high-latitude coral reef of Bermuda over annual timescales provide a framework for investigating the present and future potential impact of rising carbon dioxide (CO2) levels and ocean acidification on coral reef ecosystems in their natural environment. A strong correlation was found between the in situ rates of calcification for the major framework building coral species Diploria labyrinthiformis and the seasonal variability of [CO32-] and aragonite saturation state Ωaragonite, rather than other environmental factors such as light and temperature. These field observations provide sufficient data to hypothesize that there is a seasonal "Carbonate Chemistry Coral Reef Ecosystem Feedback" (CREF hypothesis) between the primary components of the reef ecosystem (i.e., scleractinian hard corals and macroalgae) and seawater carbonate chemistry. In early summer, strong net autotrophy from benthic components of the reef system enhance [CO32-] and Ωaragonite conditions, and rates of coral calcification due to the photosynthetic uptake of CO2. In late summer, rates of coral calcification are suppressed by release of CO2 from reef metabolism during a period of strong net heterotrophy. It is likely that this seasonal CREF mechanism is present in other tropical reefs although attenuated compared to high-latitude reefs such as Bermuda. Due to lower annual mean surface seawater [CO32-] and Ωaragonite in Bermuda compared to tropical regions, we anticipate that Bermuda corals will experience seasonal periods of zero net calcification within the next decade at [CO32-] and Ωaragonite thresholds of ~184 μmoles kg−1 and 2.65. However, net autotrophy of the reef during winter and spring (as part of the CREF hypothesis) may delay the onset of zero NEC or decalcification going forward by enhancing [CO32-] and Ωaragonite. The Bermuda coral reef is one of the first responders to the negative impacts of ocean acidification, and we estimate that calcification rates for D. labyrinthiformis have declined by >50% compared to pre-industrial times.

The widely ranging experimental response of scleractinian corals to elevated CO 2 conditions, decreasing seawater [CO 2− 3 ] and aragonite , likely reflects the complex interaction of factors that influence calcification such as light, temperature, coral host-endosymbiotic zooxanthellae interactions, species specific responses, life history, experimental design, and seawater carbonate chemistry.The influence of environmental factors on coral calcification is not clearly demonstrated and somewhat contradictory.In early studies, Goreau (1959) suggested that zooxanthellae photosynthesis would lower internal pCO 2 , enhancing CaCO 3 saturation and precipitation of CaCO 3 at internal sites of coral calcification.Field studies have subsequently indicated that rates of cal-cification are 3-5 times greater in the light than in the dark (Gattuso et al., 1999), with a coupling of photosynthesis and calcification.
Field studies of the seawater carbonate chemistry of coral reef ecosystems have focused mainly on CO 2 variability and air-sea CO 2 gas exchange (e.g., Broecker and Takahashi, 1966;Gattuso et al., 1993Gattuso et al., , 1995Gattuso et al., , 1996Gattuso et al., , 1997;;Kayanne et al., 1995Kayanne et al., , 1996Kayanne et al., , 2005;;Kawahata et al., 1997Kawahata et al., , 2000;;Bates et al., 2001;Bates, 2002), rather than relationships between coral calcification, [CO 2− 3 ], aragonite and other environmental factors.In a few studies, decreased rates of calcification have been observed on coral reef ecosystems associated with decreases in seawater [CO 2−  3 ] conditions (e.g., diurnal timescales, Suzuki et al., 1995;Yates andHalley, 2003, 2006; seasonal timescales, Silverman et al., 2007;Manzello, 2008).Under scenarios of future ocean acidification, it has been proposed that the combination of reduced rates of calcification and increased rates of CaCO 3 dissolution could result in coral reefs transitioning from net accumulation to a net loss in CaCO 3 material ("decalcification") during this century (e.g., Andersson et al., 2005Andersson et al., , 2006Andersson et al., , 2007Andersson et al., , 2009;;Hoegh-Guldberg et al., 2007;Manzello et al., 2008;Kleypas and Yates, 2009;Silverman et al., 2009).The balance of CaCO 3 production and dissolution can be defined as net ecosystem calcification (NEC).It is generally considered that CaCO 3 production occurs at saturation state values >1, while dissolution of a particular carbonate mineral phase occurs when with respect to this phase is <1.The transition from positive to negative net ecosystem calcification (NEC = calcificationdissolution) occurs at "critical threshold values" (Kleypas et al., 2001;Yates and Halley, 2006) of seawater pCO 2 , [CO 2− 3 ] and aragonite when NEC = 0.The transition is complicated due to the fact that individual coral species and other reef calcifiers may have different "critical threshold values" compared to the entire coral reef ecosystem that is influenced by a spectrum of hard coral and other marine calcifier responses as well as bioerosion and sediment dissolution.
As stated earlier, there is very limited field data on the relationships between calcification and seawater carbonate chemistry (Suzuki et al., 1995;Ohde and van Woesik, 1999), particularly over seasonal to annual timescales (Silverman et al., 2007;Manzello, 2008) and relevant reef spatial scales.The geographic distribution of coral reefs is generally dictated by light availability, sea surface temperature and by [CO 2− 3 ] and aragonite , with the high-latitude Bermuda coral reef at the geographic limit of this ecosystem (Kleypas et al., 1999a(Kleypas et al., , b, 2001;;Fig. 1).In this paper, we demonstrate seasonal relationships between in situ rates of coral calcification, seawater carbonate chemistry (i.e., [CO 2− 3 ] and aragonite ) and other environmental parameters at Hog Reef, a previously studied coral reef site within the Bermuda coral reef ecosystem (Bates et al., 2001;Bates, 2002).Furthermore, offshore data collected at the Bermuda Atlantic Time-series Study (BATS) site, ∼80 km SE of Bermuda (Steinberg et al., 2001;Bates, 2007;  of net ecosystem calcification (NEC) and net ecosystem production (NEP) in an improved method compared to previous studies (Bates, 2002).These contemporaneous datasets provide a framework for investigating the present and future potential impact of rising pCO 2 and ocean acidification on coral reef ecosystems in their natural environment.Furthermore, we evaluate the critical threshold values of [CO 2− 3 ] and aragonite at which chemical conditions may no longer be favourable for calcification on the Bermuda coral reef and the timing of these thresholds in response to future acidification of the oceans.In addition, we describe the evidence for a "Carbonate Chemistry Coral Reef Ecosystem Feedback" (CREF hypothesis), a case where there is a seasonal feedback between the primary components of the reef ecosystem (i.e., scleractinian hard corals and macroalgae) and CaCO 3 saturation states that enhance and suppress calcification rates at different times of the year.Diurnal enhancement and suppression of [CO 2− 3 ] and coral calcification by photosynthesis and respiration, respectively, have been modelled for the Shiraho Reef in the Ryukyu Islands by Suzuki et al., 1995 using short-term in situ observations (i.e., one daytime and one nightime collection of data).

Physiographic setting of the Bermuda coral reef
Bermuda has a geographically isolated subtropical coral reef ecosystem (∼1000 km 2 ), with a shallow central lagoon (i.e., North Lagoon) containing patch reefs, partly surrounded with a flank of outer rim and terrace reefs (Dodge and Vaisnys, 1977;Morris et al., 1977;Dodge et al., 1984Dodge et al., , 1985;;Logan et al., 1994) and the island of Bermuda (55 km 2 ) to the south (Fig. 1).The marine ecology of Bermuda is dominated by calcifying organisms, while the island's seamount is capped by Quaternary limestones and marine carbonate sediments.
Waters of the Bermuda coral reef continuously exchange with offshore waters of the North Atlantic Ocean surrounding Bermuda (Bates et al., 2001;Bates 2002).The typical residence time of water on the rim reef is approximately 1-4 days (Morris et al., 1977), while water residence times are longer in the North Lagoon (∼5-10 days) (R. J. Hard coral cover on the Bermuda rim and terrace regions of the reef system typically ranges between 15 and 70% (Fig. 1; CARICOMP, 1997aCARICOMP, , b, 2000) ) including the areally dominant calcareous sand and seagrass ecosystems of the North Lagoon.Over the last couple of decades, Bermuda's rim reefs have maintained long-term average of 21% coral cover varying between 18-23% year to year (MEP, 2006; R. J.Jones, unpublished data, http://www.bios-mep.info/NEW%20site/Sub Program 2c.htm) with macroalgae varying between 5 and 15%.The dominant coral reef taxa are Diploria labyrinthiformis and D. strigosa, with Montastrea franksii, M. cavernosa, Porites astreoides, and Millepora alcicornis being significant components of the reef ecosystem.D. labyrinthiformis and strigosa are arguably the dominant species and constitutes 25-35% of the reef hard coral cover.

Seawater carbonate chemistry considerations
The complete seawater carbonic acid system (i.e., CO 2 , H 2 CO 3 , HCO − 3 , CO 2− 3 , H + ) can be calculated from a combination of two carbonate system parameters, DIC, TA, pCO 2 and pH, along with temperature and salinity.Here, pCO 2 is the partial pressure of CO 2 in equilibrium with seawater, while pH is expressed on the total seawater scale.DIC is defined as (Zeebe and Wolf-Gladrow, 2001;Dickson et al., 2007): where [CO * 2 ] represents the concentration of all unionized carbon dioxide, whether present as H 2 CO 3 or as CO 2 .The total alkalinity of seawater (TA) is defined as:

where [HCO
4 are the principal components of seawater TA.
Calcium carbonate (CaCO 3 ) mineral production and dissolution is governed by the following chemical reaction: CaCO 3 production and dissolution rates vary as a function of saturation state ( ).For corals and other calcifying marine organisms whose carbonate mineralogy is aragonite, the seawater saturation state with respect to this mineral phase is defined as the ion concentration product of calcium and carbonate ions divided by the stoichiometric solubility product, K sp * (aragonite) , which is a function of temperature, salinity and pressure (Mucci, 1983), thus:

Seawater DIC, TA and pCO 2 observations
During 2002 and 2003, seawater samples were collected regularly at Hog Reef (∼2 m deep).Samples for DIC and TA were drawn from a Niskin sampler into clean 0.5 dm 3 size Pyrex glass reagent bottles, using established gas sampling protocols (Bates et al., 1996a).A headspace of <1% of the bottle volume was left to allow for water expansion and all samples were poisoned with 100 µl of saturated HgCl 2 solution to prevent biological alteration.Bottles were sealed with ground-glass stoppers and Apiezon silicon vacuum grease.Rubber bands were placed around the lip of the bottle and stopper to provide positive closure.Samples were returned to BIOS for analysis.DIC was measured by a gas extraction/coulometric technique (see Bates et al., 1996a, b for details), using a SOMMA (Single-Operator Multi-Metabolic Analyzer) to control the pipetting and extraction of seawater samples and a UIC CO 2 coulometer detector.The precision of DIC analyses of this system is typically better than 0.025% (∼0.4 µmoles kg −1 ) based on duplicate and triplicate analyses of >2000 seawater samples analyzed at BIOS from 1992 to present.Seawater certified reference materials (CRM's; prepared by A.G. Dickson, Scripps Institution of Oceanography) were analyzed to ensure that the accuracy of DIC was within 0.03% (∼0.5 µmoles kg −1 ).Salinity was determined analytically using a SeaBird SBE-9 conductivity sensor and calibrated against salinity collected at the ocean time-series BATS (Steinberg et al., 2001).In situ temperature was measured with a platinum thermistor (±0.05 • C) and temperature logger.TA was determined by potentiometric titration with HCl (see Bates et al., 1996a, b for details).CRM samples were also analyzed for TA and these values were within 0.15% (∼2-3 µmoles kg −1 ) of certified TA values reported by A.G. Dickson (http://andrew.ucsd.edu/co2qc/index.html).
A time-series of seawater pCO 2 was collected at Hog Reef using an autonomous CARIOCA (CARbon Interface OCean Atmosphere) buoy (Merlivat and Brault, 1998;Bates et al., 2000Bates et al., , 2001)).The CARIOCA buoy was deployed twice during the 2002-2003 period.Initially, the CARIOCA buoy was deployed on the 16 October 2002 (day 287 of the year) and recovered on the 20 January 2003 (day 20 of the year) after breaking its mooring line.Instrument repair and calibration delayed the subsequent deployment and the CARIOCA buoy was deployed on the 26 April 2003 (day 116 of the year).In anticipation of the passage of Hurricane Fabian over Bermuda (5 September 2003; day 247), the CARIOCA buoy was moved from Hog Reef to a protected inshore site (Ferry Reach) off BIOS's dock on the 28 August 2003 (day 239).
The CARIOCA buoy collected hourly measurements of seawater temperature, pCO 2 and fluorescence from an intake at 2 m depth.Seawater temperature data was measured using two Betatherm thermistors with an accuracy of 0.05 • C. Tri-butyl tin (TBT) tubing was used internally and a copper plate was mounted at the seawater intake of the CARIOCA buoy; both were used to reduce the possibility of biofouling affecting the pCO 2 sensor.Seawater pCO 2 measurements were conducted using an automated spectrophotometric technique (Hood et al., 1999;Bates et al., 2001; http:/www.lodyc.jussieu.fr/carioca/).CARIOCA buoy pCO 2 measurements were calibrated in the laboratory prior to deployment using a Licor infrared CO 2 analyzer (Model 6262) and CO 2 -in-air gas standards.Seawater was pumped in parallel through an equilibrator-Licor analyzer system and the CARIOCA exchanger cell.Linear regression curves of the spectrophotometric and Licor pCO 2 data were calculated and subsequently used to determine pCO 2 from spectrophotometric absorbance and temperature data.
In this study, pCO 2 , [CO 2− 3 ] and aragonite were calculated from in situ DIC and TA data sampled from Hog Reef.The carbonic acid dissociation constants of Mehrbach et al. (1973), as refit by Dickson and Millero (1997), were used to determine seawater pCO 2 and other carbonate parameters, using the equations of Zeebe and Wolf-Gladrow (2001).In addition, the CO 2 solubility equations of Weiss (1974), and dissociation constants for borate (Dickson, 1990), and phosphate (DOE, 1994) were used.DIC and TA data was also recalculated as salinity normalized DIC (i.e., nDIC) and alkalinity (i.e., nTA) using a salinity of 36.6.This correction accounts for the DIC changes imparted by local precipitation and evaporation (Bates et al., 1996a).
Meteorological data were collected each hour from the island of Bermuda by the Bermuda Weather Service.Wind speed data were corrected to 10 m using the equations of Smith (1988).Observations of net shortwave downward radiation were also used (Dutton, 2007).Net shortwave radiation, Q sw , was determined from observations of cloud cover, C f , and theoretical extraterrestrial solar radiation, E t , using a model of Beriland (1960) and Dobson and Smith (1980): where T r is the transmission coefficient and a is the cloud correction factor.The values for T r and a have been measured at 0.89 and 0.67 in the Sargasso Sea surrounding Bermuda (Johnson, 2003).The theoretical extraterrestrial solar radiation, E t , was determined using standard astronomical formulae for the solar constant, solar elevation and ephemera to account for seasonality and diurnality (equations from Payne, 1972;Partridge, 1976;Watt Engineering Ltd, 1978, Duffie andBeckman, 1991).Photosynthetically available radiation (PAR) at the ocean surface is ∼45% of estimated total insolation or Q sw (Baker and Frouin, 1987).

In situ coral colony calcification or skeletal growth rates
The buoyant weight technique (e.g., Jokiel et al., 1978, Davies 1989, 1990), a non-destructive method commonly used to determine calcification and growth of hermatypic corals (e.g., Dodge et al., 1984Dodge et al., , 1985;;Marubini et al., 2001Marubini et al., , 2003;;Abramovich-Gottlib et al., 2003), was used to determine in situ skeletal growth of D. labyrinthiformis at several sites across the Bermuda reef including Hog Reef, Twin Breakers and Crescent Reef (Fig. 1).At each site, coral colonies (n = 8) of D. labyrinthiformis were transplanted on racks and secured to the reef sites in a block design.Approximately every three months, colonies were transported to BIOS and weighed in water using the buoyant weight technique.The dry weight of the coral specimen in air is where W a and W w are the dry and wet (or buoyant) weights respectively, and ρ w and ρ s are the densities of seawater and specimen respectively (Jokiel, 1978;Langdon et al., 2010).With this method, the skeletal weight of the coral colony can be estimated from its buoyant weight in seawater whose density has been accurately determined, thereby providing a simple, non-destructive method for recording integrated coral skeletal growth (or calcification rate) over seasonal timescales.The calcification rate (G) or skeletal growth for D. labyrinthiformis is given by: where W a is the change in dry skeletal weight and t is the number of days between weighings.Thus, skeletal growth is expressed as weight increase per g weight (CaCO 3 plus very minor contributions from tissue) for each coral colony and expressed as mg CaCO 3 g −1 d −1 (Table 1).Skeletal growth rate per unit area was also calculated from weight changes and determination of individual coral colony surface area (determined at the end of deployment) expressed as mg CaCO 3 cm −2 d −1 (Table 1).

Results
The coral reefs of Bermuda experience large seasonal changes in physical conditions, such as light and temperature, seawater carbonate chemistry and calcification rates (Fig. 2).At the summer solstice, day and night length was ∼14 and ∼10 h, respectively, and reversed at the winter solstice (CARICOMP, 1997a(CARICOMP, , b, 2000)).Net shortwave radiation (Q sw ) and sea surface temperature showed distinct seasonality as observed previously (Bates, 2002).Light conditions were highly variable seasonally (Fig. 2a).For example, Q sw had a seasonal minima of ∼2000-3000 W m −2 in the December 2002 and January 2003 period, and a seasonal maxima of ∼6000-8000 W m −2 in the June-August 2003 period (Fig. 2a).The period of highest Q sw occurred around the June solstice period (Julian Day, JD ∼150-165).
Surface temperatures at Hog Reef decreased from midsummer maxima of ∼27 • C in 2002 to a winter minima of ∼20 • C in the January to March 2003 period (Fig. 2a).These seasonal changes are similar to those typically observed on the Bermuda reef (Fig. 1b).Subsequently, a mid-summer www.biogeosciences.net/7/2509/2010/Biogeosciences, 7, 2509-2530, 2010  (Bates, 2002(Bates, , 2007)).Inorganic nutrient concentrations across the Bermuda coral reef are low.For example, nitrate+nitrite concentrations are typically less than 0.1 µmoles kg −1 (MEP, 2006; http://www.bios-mep.info/;executive summary only; Fig. 1c) and similar to oligotrophic conditions observed in offshore waters at BATS (Steinberg et al., 2001).Freshwater inputs to the North Lagoon from the island of Bermuda are negligible and there is an absence of major sources of pollutants (e.g., anthropogenic nutrients).Bermuda reef surface salinity, typically has a seasonal range of ∼36.0 to 36.8, with slightly fresher conditions occuring during summertime (MEP, 2006;Fig. 1b) and similar to offshore conditions (Steinberg et al., 2001).
Wind speeds experienced by the Bermuda coral reef were also generally higher during the winter due to the regular passage of cold fronts originating from North America (Fig. 2b).Similar seasonal changes in windspeed have been observed at the BATS site offshore (Bates, 2007).The major event recorded in the windspeed data were sustained high winds of ∼120 mph (∼200 kph) during the passage of Hurricane Fabian over the island of Bermuda on the 5 September 2003 (JD 247; Fig. 2b).
Seawater carbonate chemistry observed at Hog Reef was also highly variable over seasonal timescales.Since the source of Bermuda coral reef waters is the surrounding Sargasso Sea, the variability of Hog Reef carbonate chemistry can be compared with contemporaneous carbonate chemistry data observed at the offshore BATS site.For the 2002-2003 period, surface seawater pCO 2 ranged from low wintertime values (∼300-360 µatm) to summertime values exceeding 550 µatm (Fig. 2b).In comparison, seawater pCO 2 values at the BATS site had a seasonal range of ∼300-420 µatm (Bates, 2007), with the major difference observed during the summertime, when seawater pCO 2 was significantly higher on the Bermuda coral reef.The continuous observations of seawater pCO 2 at Hog Reef also showed considerable diurnal variability of ∼20-100 µatm.In contrast, diurnal variability at the BATS site is significantly attenuated (∼5-25 µatm; Bates et al., 2000Bates et al., , 2001)).In other coral reef systems, diurnal to seasonal seawater pCO 2 ranged from as low as ∼100 µatm to as high as 1000 µatm, the largest amplitude in seawater pCO 2 typically observed in the shallower reefs.These previous studies have typically observed seawater CO 2 and associated variables over a few days only or with transects across reef systems (e.g., Smith, 1973;Smith and Key, 1975;Gattuso et al., 1993;Kayanne et al., 1995Kayanne et al., , 1996;;Frankignoulle et al., 1996;Kawahata et al., 1997Kawahata et al., , 2000;;Ohde and van Woesik, 1999;Suzuki and Kawahata, 2003).
Surface DIC at Hog Reef had a seasonal variability of ∼100 µmoles kg −1 , with a maxima of ∼2070 µmoles kg −1 and minima of ∼1970 µmoles kg −1 observed during the summer of 2003 (Fig. 2c).When compared to contemporaneous BATS DIC data, in general, Hog Reef DIC data generally follows (within ∼20 µmoles kg −1 ) seasonal changes of DIC observed at the BATS site (Fig. 2c).However, during the summer of 2003, Hog Reef DIC became depleted by as much as 30-40 µmoles kg −1 relative to DIC at the BATS site.
Total alkalinity at Hog Reef varied seasonally by ∼100 µmoles kg −1 (Fig. 2c), with considerable differences observed between Hog Reef and offshore at BATS.For example, Hog Reef TA was generally lower by ∼20-40 µmoles kg −1 compared to BATS TA for most of 2002 and 2003.However, during the summer of 2003, Hog Reef TA and DIC were depleted by ∼60-80 µmoles kg −1 and 30-40 µmoles kg −1 , respectively, compared to offshore concentrations at BATS (Fig. 2c).The depletion of Hog Reef TA and DIC had an approximate ratio of ∼2:1, similar to theoretical predictions that the formation of CaCO 3 decreases TA and DIC in a ratio of 2:1 due to the uptake of [Ca 2+ ] and [CO 2− 3 ] (Eqs.1-3).The seasonal values of [CO 2− 3 ] observed at Hog Reef ranged from 190 to 250 µmoles kg −1 , a smaller range than changes observed on other reefs (Table 1; Fig. 2d).Hog Reef [CO 2− 3 ] and aragonite values were generally lower by ∼30-70 µmoles kg −1 and ∼0.3 (not shown) relative to offshore [CO 2− 3 ] and aragonite values at BATS, with the exception of a few occurences during early summer 2003 (JD∼180-210) (Fig. 2d).
The annual range of skeletal growth rates (i.e,G diploria ) was ∼0.28-0.65 mg CaCO 3 g −1 d −1 for D. labyrinthiformis colonies (Table 2).Skeletal growth rates per unit area ranged from 0.40-0.96mg CaCO 3 cm −2 d −1 for the same D. labyrinthiformis colonies (Table 2).The highest rates were observed at Hog Reef for the period of July-August 2003 and lowest rates during the wintertime (Fig. 2e; Table 2).In situ skeletal growth rates for D. labyrinthiformis colonies deployed at Twin Breakers were also seasonally similar and included in Fig. 2d (with the period of in situ colony deployment denoted by the horizontal bars).Twin Breakers is assumed to have similar seasonal changes in carbonate chemistry to Hog Reef due to their close proximity.At both sites, in situ skeletal growth or calcification rates covaried with seasonal changes of [CO 2− 3 ] (Fig. 2e) and aragonite (not shown).

Estimates of annual rates of in situ coral calcification
Previous studies of Bermuda corals such as D. labyrinthiformis and Porites astreoides, have been shown to accrete narrow, high density bands of CaCO 3 during the summer, and wider low-density bands during the fall to spring (Logan and Tomascik, 1991; Cohen et al., 2004).If the in situ skeletal growth rates observed at Hog Reef are scaled up, we estimate that the calcification rate per unit area of the rim reef (i.e., G reef ) ranged from ∼1.3 to 3.2 g CaCO 3 m −2 d −1 , using the following equation: where G diploria is skeletal growth rate (expressed as mg CaCO 3 cm −2 d −1 ) scaled up to a m −2 area (i.e., 1 m 2 = 10 000 cm 2 ).α is a multiplier value that varies between 0 and 1 that is a function of the planar surface area of the reef.Here, α = 0.21 given that Bermuda's rim reefs have a long-term average of 21% coral cover.However, the actual surface area is larger due to the complex/hemispherical geometry of coral colony surface area.Thus in Eq. ( 8), β is a multiplier that accounts for the complex/hemispherical geometry of coral colony surface area, which is set at 1.57 assuming an ideal hemisphere for coral colony shape.Thus, for example, if a skeletal growth rate of 0.96 mg CaCO 3 cm −2 d −1 is used, and 21% coral cover assumed (i.e., α = 0.21), the G reef rate is 3.2 g CaCO 3 m −2 d −1 (i.e., 0.96 mg CaCO 3 cm −2 d −1 × 10 000 (cm 2 ) × 0.21 × 1.57).In  parts of the rim reef, coral cover can be up to 70%, for which the G reef rate would be ∼4.4 to 10.6 g CaCO 3 m −2 d −1 .In this calculation, we also assume that the skeletal growth rates for other coral species present at Hog Reef were similar to D. labyrinthiformis, and that other calcifying organisms such as coralline algae do not contribute substantively to this estimate of calcification rate.Calcification rates on other reefs can vary by a couple of orders of magnitude but the calcification rate estimated for the Bermuda coral reef is at the lower end of the typical observed range for other reefs (∼<2-40 g CaCO 3 m −2 d −1 ; e.g., Kinsey, 1985;Pichon 1997;Gattuso et al., 1993Gattuso et al., , 1996Gattuso et al., , 1999;;Barnes and Lazar, 1993;Yates and Halley, 2006;Silverman et al., 2007).The annual rate of calcification per unit area of the reef is estimated at Hog Reef to range between 0.5 and 1.2 kg CaCO 3 m −2 year −1 , slightly lower than the average calcification rate of 4 ± 0.7 kg CaCO 3 m −2 year −1 reported for other coral reefs (Kinsey, 1985).Benthic turf and fleshy macroalgae distributions were not directly measured at Hog Reef, but typically constitute <5-15% of the reef cover (MEP, 2006; http: //www.bios-mep.info/NEW%20site/SubProgram 2c.htm).
The highest macroalgal biomass is typically observed coincident with the period of highest solar irradiance in June (Smith, S. R., personal communication), a seasonal feature typically observed on other reefs (Gattuso et al., 1997).

Seasonal covariance of coral calcification and carbonate chemistry on the Bermuda coral reef
There are few datasets that can be used to test relationships between coral calcification and carbonate chemistry under natural conditions.Our results from the Bermuda coral reef indicate that calcification rates of D. labyrinthiformis at Hog Reef and Twin Breakers covaried seasonally with [CO 2− 3 ] and aragonite .Mean in situ skeletal growth rates had a range of ∼0.28-0.65 mg CaCO 3 g −1 d −1 while [CO 2− 3 ] and aragonite varied by ∼40 µmoles kg −1 and 0.4 respectively (Fig. 3).Despite a limited number of observations, in situ skeletal growth rates (either expressed as weight increase or per unit area) were well correlated with mean [CO 2− 3 ] and aragonite (Fig. 3a and b), with r 2 of ∼0.68.Similar findings have been shown in the natural environment (Silverman et al., 2007(Silverman et al., , 2009) ) and in vitro experiments with other coral species (Marubini et al., 2003;Schneider and Erez, 2006).
The correlation between in situ skeletal growth and other environmental factors were less statistically significant.For example, mean temperatures during each in situ skeletal growth measurement at Hog Reef were weakly correlated with rates of in situ skeletal growth (Fig. 3c).In the Eilat reef, coral community calcification is well correlated with temperature (Silverman et al., 2007), while in other reef systems, the highest seasonal rates of calcification have been observed a few degrees below the seasonal temperature maximum (e.g., Abramovitch-Gottlib et al., 2003;Marshall and Clode, 2004).In Sect.4.4, we show that net heterotrophy induced by other components of the reef ecosystem appears to suppress aragonite and rates of coral calcification during periods of the summertime.As a result, in situ skeletal growth rates are weakly correlated with temperature on the Bermuda reef.
There was a poor correlation between in situ skeletal growth of D. labyrinthiformis and mean shortwave radiation (i.e., Q sw ; Fig. 3c).This is perhaps surprising since other studies have shown a strong coupling between light and calcification (e.g., Gattuso and Jaubert, 1990;Marubini et al., 2003;Schneider and Erez, 2006).Short-term (<2 h) in vitro chamber experiments using D. labyrinthiformis colonies recovered from Hog Reef and acclimatized at BIOS, showed a strong coupling between light (∼200-1400 µE m −2 s −1 ) and zooxanthellae photosynthesis and respiration rates (as expressed as oxygen production or consumption).If Q sw is an appropriate proxy for coral photosynthesis, our in situ observations would suggest a weak coupling between coral photosynthesis and calcification.However, while Q sw is a good proxy for the seasonally integrated mean light conditions at Hog Reef, Q sw may not accurately reflect variability  (Bates, 2007). of in situ PAR over shorter time-scales.With limited data, we cannot statistically confirm either a strong coupling or uncoupling of light and calcification for corals at Hog Reef.Similar weak correlation between coral calcification and seasonal changes in shortwave radiation have been shown for other reefs primarily due to the seasonal lag of several months between peak solar input, and seawater temperatures and coral calcification (e.g., Silverman et al., 2007).

Potential mechanisms coupling seawater carbonate chemistry and coral calcification
The field data collected from the Bermuda coral reef indicates that the highest rates of calcification occurred when [CO 2− 3 ] in the external reef environment was at seasonally high concentrations (while [HCO − 3 ] was at seasonally low Furthermore, with lower external seawater pH and greater [H + ], increased amounts of energy may be required to pump H + against this gradient.Higher energetic demands required to remove H + as suggested by Cohen and Holcomb (2009), combined with lower energy production in winter due to reduced solar input, may make it more difficult for Bermuda corals to alkalinize calicoblastic fluids with lower calcification rates as a result.
If corals do actively take up both HCO − 3 and CO 2− 3 , changes in zooxanthellae photosynthesis could also enhance or suppress calcification.Since HCO − 3 (internally converted to CO 2 by CA) is the source of inorganic carbon for photosynthesis, increased demand for HCO − 3 by increased zooxanthellae photosynthesis (in response to enhanced light conditions) should shift the ratio of [HCO − 3 ]:[CO 2− 3 ] to lower values.This should further elevate pH, enhance alkalinization, and aragonite conditions in the calicoblastic layer.Thus, as evidenced by higher calcification rates during summertime for the Bermuda coral reef, photosynthesis and favorable carbonate chemistry changes may act synergistically to enhance rates of coral calcification.

The carbonate chemistry coral reef feedback (CREF) hypothesis
Seasonal changes in seawater carbonate chemistry of reef systems can be used to evaluate the net ecosystem metabolism (NEM) of the reef and the impact of benthic processes on water overlying the reef system (e.g., Chisholm and Barnes, 1991;Suzuki et al., 1995;Gattuso et al., 1996;Bates, 2002;Silverman et al., 2007;Langdon et al., 2010).Two processes dominate the net ecosystem metabolism of the reef, each with different influence on seawater pCO 2 and other components of the seawater carbonate system.The first process relates to the balance of coral calcification and dissolution or net ecosystem calcification (NEC).Positive NEC values represent net calcification, while negative NEC values represent net dissolution.In general, calcification release about 0.6 mole of CO 2 to the surrounding environment per mole of CaCO 3 precipitated in coral reef systems (Kinsey, 1985;Frankignoulle et al., 1994;Lerman and Mackenzie, 2005).When rates of calcification exceed dissolution (i.e., NEC is positive), the uptake of inorganic carbon into the coral skeleton as CaCO 3 decreases DIC and TA in a ratio of 1:2, with the net result of CO 2 production and increase in seawater pCO 2 .Thus, NEC on most coral reefs results in net production of CO 2 (Gattuso et al., 1999).In many coral reef systems, higher reef seawater pCO 2 values compared to offshore conditions have been observed (e.g., Kawahata et al., 2000;Suzuki and Kawahata, 2003), confirming that coral reef metabolism generally acts to increase seawater pCO 2 .Similar findings were reported from previous short-term observations at Hog Reef (Bates et al., 2001), and across the SE sector of the Bermuda platform (Bates, 2002), and in this paper (Fig. 4).
The second process relates to the balance of photosynthesis and respiration or net ecosystem production (NEP).On a typical coral reef, NEP is dominated by coral/zooxanthellae respiration/photosynthesis, and benthic macroalgal photosynthesis and respiration.In net autotrophic systems, where the rate of photosynthesis or gross primary production (P ) is greater than rate of respiration (R), NEP values are negative and the uptake of CO 2 decreases DIC only (and seawater pCO 2 also) and TA remains unchanged (minor changes do occur owing to the uptake of nutrients).In net heterotrophic systems, where P < R, NEP values are positive, CO 2 is produced and DIC and seawater pCO 2 increase over time.In many reef systems, net ecosystem production (NEP) is near zero despite high rates of gross primary production (e.g., Crossland et al., 1991;Gattuso et al., 1999;Ducklow and McAllister, 2004).
In a previous study, Bates (2002) used monthly differences of temperature corrected seawater pCO 2 between the Bermuda coral reef and offshore values to estimate net productivity (i.e., equivalent to NEP in this study) rates over the annual cycle.In the absence of contemporaneous in situ coral calcification rates, constant rates of calcification over the annual cycle were assumed and used to estimate net productivity.This previous analysis indicated that the Bermuda coral reef was net autotrophic over most of the year (i.e., NEP rates were negative while net heterotrophic conditions occurred in August and September (i.e., Fig. 3 in Bates, 2002).
In this study, NEC and NEP rates for the Bermuda reef were determined using mass balance methods following similar methods to other studies (e.g., Gattuso et al., 1996;Bates, 2002;Silverman et al., 2007).In this approach, observed differences between onshore and offshore seawater carbonate chemistry are used to quantify how reef processes (i.e., calcification, dissolution, photosynthesis and respiration) modify the TA and DIC content of waters overlying the reef, thereby determining the NEM of the reef system.Contemporaneous DIC and TA data from the BATS site (Bates, 2007) and Hog Reef were used for offshore and onshore seawater carbonate chemistry conditions, and both TA and DIC datasets were corrected to a constant salinity of 36.6 to account for local evaporation/precipitation differences between onshore and offshore.
In a mass balance sense, if the rate of NEC (i.e., NEC reef ) is positive and NEP of the reef (i.e., NEP reef ) is zero, waters modified by net reef metabolism will gain CO 2 (i.e., increase seawater pCO 2 ) compared to offshore conditions due to the production of CO 2 from calcification and formation of CaCO 3 (DIC and TA will decrease).If NEC reef is zero and NEP reef negative (i.e., net autotrophic), waters modified by net reef metabolism will lose CO 2 (i.e., decrease seawater pCO 2 ) compared to offshore conditions due to uptake of CO 2 from photosynthesis (i.e., photosynthesis > respiration; DIC will decrease while TA will increase marginally).

Calculation of reef NEC and NEP rates
The calculation of rate of NEC (i.e., NEC reef ) is based on the alkalinity anomaly-water residence time technique (Smith and Key, 1975;Kinsey, 1978;Chisholm and Gattuso, 1991;Langdon et al., 2010) that has been used previously for estimating in situ rates of calcification for reef systems (Gattuso et al., 1996;Silverman et al., 2007).In the method, differences between offshore and onshore nTA are assumed to result from the balance of reef calcification and dissolution (i.e., NEC) that changes the TA content of waters overlying the reef (i.e., nTA NEC ).Thus, seasonal values for nTA NEC are determined using the observed difference in salinity normalized TA (i.e., nTA) between offshore and onshore (nTA offshore -TA onshore ) using data from BATS and Hog Reef (Table 3).The NEC reef rate is then calculated by scaling the values of TA NEC to an appropriate water depth (Z) and water residence time (τ ) for the reef.Thus, following the method of Langdon et al. (2010): where ρ is the density of seawater.Here, an average water depth of 6 m and water residence time of 2 days is used in the calculations of NEC reef and NEP reef rates with scaling issues, caveats and uncertainties discussed further.Rates of NEC reef are expressed in units of mmoles CaCO 3 m −2 d −1 (or expressed as g CaCO 3 m −2 d −1 using a molecular weight of 100.09).Secondly, the change in DIC for waters overlying the reef due to NEC (i.e., nDIC NEC ) is calculated using a TA:DIC ratio of 2:1).Thus: The rate of NEP (i.e., NEP reef ) for the reef is calculated by mass balance given that NEP imparts a change in the DIC content of waters overlying the reef (with photosynthesis and respiration causing no change in TA).The rate of NEP reef is thus calculated by mass balance using the observed differences in nDIC between onshore and offshore (i.e., nDIC offshore−onshore ; Table 3) and nDIC NEC : NEP reef is then expressed in units of mmoles C m −2 d −1 (or expressed as g C m −2 d −1 using a molecular weight of 12).

Scaling of NEC and NEP rates
In the above method, the rates of NEC reef and NEP reef are scaled from observed seasonal changes in nTA offshore−onshore and nDIC offshore−onshore as a function of water depth and water residence time.Based on observations/models (Johnson, 2003; R. J. Johnson tide/wind mixing model of the Bermuda reef), an average water depth www.biogeosciences.net/7/2509/2010/Biogeosciences, 7, 2509-2530, 2010 Table 3. Hog Reef TA and DIC data compiled into a composite year.Julian day of sampling is shown along with original sampling data, sea surface temperature (SST), salinity (S), DIC and TA data (both µmoles kg −1 ).These data are used in Sect.4.4 to calculate rates of NEP reef and NEC reef using nTA offshore−onshore and nDIC offshore−onshore values.nTA offshore−onshore and nDIC offshore−onshore are calculated from nTA and nDIC (adjusted to a salinity of 36.6) using contemporaneous Hog Reef (onshore) and BATS data (offshore).
Julian of 6 m and water residence time of 2 days (i.e., Z and τ in Eq. 9) were used in the calculations of NEC reef and NEP reef rates (Fig. 5).It is important to recognize that this mass balance approach does not provide absolute values for NEC reef and NEP reef , but rather, provides a seasonal view of changes in the balance of calcification/dissolution, and net heterotrophy/net autotrophy.As a sensitivity test, the annual rates of NEC reef and NEP reef were plotted in Fig. 6 for a range of Z and τ values that are within observed ranges for the Bermuda rim reef (e.g., 4-8 m water depth and 1-4 day water residence time).If the Z term (i.e., water depth) in Eq. ( 9) is increased, rates of NEC reef and NEP reef also increase (Fig. 6) since reef rate processes (e.g., calcification) have to be higher for equivalency to observed nTA offshore−onshore and nDIC offshore−onshore data.In contrast, longer water residence times (i.e., τ in Eq. 9) reduce NEC reef and NEP reef rates (Fig. 6).The strong summertime net autotrophy and late summertime net heterotrophy shown in Fig. 5 and discussed later occurs for all proscribed values of Z and τ values shown in Fig. 6.

Further caveats and uncertainties for estimating rates of NEC reef and NEP reef
There are further caveats and uncertainties using the alkalinity anomaly-water residence time technique.Firstly, it should be noted that onshore and offshore seawater carbonate chemistry were not typically sampled on the same day, but, we have chosen data sampled as closely in time to estimate onshore-offshore differences.Secondly, seawater carbonate chemistry data were not corrected for long-term changes observed at the BATS due to the oceanic uptake of anthropogenic CO 2 (Bates, 2007;Bates and Peters, 2007) since the observations occurred over a 16 month period.These long-term changes are very minor compared to the observed changes in seawater carbonate chemistry over the timeframe of the study.It is also assumed that the uptake of nitrate by coral photosynthesis does not contribute significantly to changes in nTA.
In addition, as argued previously by Bates (2002), benthic coral calcification/dissolution, and coral/macroalgae photosynthesis/respiration are the dominant processes influencing NEM for the Bermuda reef, with air-sea CO 2 gas exchange, pelagic phytoplankton primary production  and vertical mixing processes having minor impact on the carbonate chemistry of waters resident for a short time (<2 days) on the rim reefs of Bermuda (Bates, 2002).The NEC for the reef (i.e., NEC reef ) includes contributions from other calcifiers such as coralline red algae, green algae, echinoderms, bryozoans, foraminifera and bivalves.In the absence of data for other calcifiers, we assume that their contribution is minor and that corals are the dominant calcifier on the Bermuda coral reef with NEC reef NEC coral .

Seasonal rates of NEC reef and NEP reef
The alkalinity anomaly-water residence time technique used here indicates that NEC reef seasonally ranged between −2.2 to 10.4 g CaCO 3 m −2 d −1 (Fig. 5a) with highest net calcification in winter (January-April) and mid-summer (July-August) and lower net calcification in late-summer to fall (September-December).For comparison, as shown in section 4.1, NEC rates scaled up to the Bermuda reef using observed skeletal growth rates for D. labyrinthiformis (i.e., G diploria ) and a 50% coral cover would be in the range of 4.5 to 11.1 g CaCO 3 m −2 d −1 .In addition, in situ observations (G diploria ) and mass balance approaches (i.e., NEC reef ), both determined independently of each other, show similar seasonal patterns (compare Fig. 5a and b).For most of the  3.The regression statistics for the line are: −0.244x + 0.700, r 2 = 0.607.Arrows indicates direction of net autotrophy (i.e., −NEP reef ), net heterotrophy (+NEP reef ), net calcification (i.e., +NEC reef ) and net dissolution (i.e., −NEC reef ).For example, the upper left quadrant denotes conditions on the reef with net heterotrophy and net dissolution.
The alkalinity anomaly-water residence time technique also reveals seasonal changes in net reef metabolism and shifts between net autotrophy and heterotrophy for the Bermuda coral reef ecosystem (Fig. 5a).Rates of NEP reef seasonally ranged between ∼ −2.8 to +1.5 g C m −2 d −1 .This compares to early estimates for NEP of other reefs that ranged from ∼−0.6 to +0.6 g C m −2 d −1 (Kinsey, 1985;Andersson et al., 2005).Over relatively shorttimescales, Gattuso et al. (1996) showed that the Moorea and Yonge reefs were net autotrophic with ranges of −0.4 to −5.8 g C m −2 d −1 .More recently, Silverman et al., 2007 showed that the Eilat reef was predominantly net autotrophic (up to −2.2 g C m −2 d −1 ) over the 1997-2002 period, but also occasionally net heterotrophic (+0.5 g C m −2 d −1 ).
In general, NEP reef rates were negative over most of the year indicating net autotrophic status of the reef, with rates of photosynthesis greater than respiration.In the summer (July/August) and fall (November-January) periods, NEP reef rates were negative indicative of net autotrophy.However, in late summer (September/October), NEP reef rates were strongly positive, indicative of net heterotrophic conditions that generate CO 2 , similar to previous findings of Bates (2002).These seasonal patterns suggest that CO 2 is taken up by the reef system in early summer and fall periods, while CO 2 is released from net reef metabolism to waters overlying the reef during the late summer.The seasonal changes in carbonate chemistry, NEC reef , and NEP reef are evidence for a feedback between seawater carbonate chemistry and reef metabolism that enhances or suppresses coral calcification.As shown in Fig. 5, the highest rates of net calcification (i.e., +NEC reef values) generally occur during periods when rates of net autotrophy are at their highest (i.e., −NEP reef ).We term this feedback as a seasonal carbonate chemistry coral reef ecosystem feedback (CREF).In this scenario, in early summer, when macroalgal biomass is at it's maxima on the Bermuda reef, strongly negative NEP reef indicates net uptake of CO 2 into the benthic biomass (i.e., macroalgae and coral zooxanthellae), which in turn increases the [CO 2− 3 ] and aragonite of waters resident on the reef.Thus, early summer net autotrophy enhances carbonate chemistry conditions favourable for calcification (evidenced by high rates in situ skeletal growth; Fig. 5c and high rates of net calcification, NEC reef , Fig. 5a).Similar seasonal enhancement of surface layer [CO 2− 3 ] and aragonite have been observed elsewhere as response to pelagic phytoplankton primary production and strongly net autotrophic conditions (Feely et al., 1988;Bates et al., 2009).In addition, a diurnal model of the enhancement and suppression of [CO 2− 3 ] and coral calcification by photosynthesis and respiration, respectively, has been shown for the Shiraho Reef by Suzuki et al. (1995) using one daytime and one nightime set of in situ observations for validation of the model.In our study, although there are caveats and uncertainties in using mass balance models, the NEP reef values for spring-summer net autotrophy suggest that in addition to coral metabolism, other components of the reef system (i.e., macroalgae photosynthesis) contributed to net autotrophy and enhancement of [CO 2− 3 ], aragonite , and NEC reef .In contrast to the early summer condition, NEP rates shift in late summer to positive values indicating a change from net autotrophy to net heterotrophic conditions.Release of CO 2 in late summer suppresses [CO 2− 3 ] and aragonite which in turn appears to suppress coral calcification rates (Fig. 5b  and c).During this period, benthic macroalgal biomass typically decreases from a seasonal maxima in early summer (S.R. Smith, unpublished data).Net heterotrophic conditions in late summer likely result from a combination of factors, such as reduction in zooxanthellae photosynthesis rates, and remineralization of organic matter produced from the earlier benthic macroalgal production in early summer.Thus, late summer net heterotrophy and release of CO 2 appears to depress carbonate chemistry conditions favourable for calcification (evidenced by low rates in situ skeletal growth; Fig. 5c and low rates of net calcification, NEC reef , Fig. 5a).It is likely that other components of the reef system (i.e., macroalgae respiration) contributed to net heterotrophy and suppression of [CO 2− 3 ] and aragonite .It may also be that late summer macroalgal respiration and entrainment of respiratory CO 2 from below the mixed layer due to the breakdown of the warm, shallow thermocline through mixing induced by cooling and storms act to increase seawater pCO 2 and decrease [CO 2− 3 ] and aragonite .The subsequent seasonal rebound in [CO 2− 3 ] and aragonite conditions and in situ skeletal growth rates during the fall is associated with a return to net autotrophic conditions.This perhaps reflects a combination of exhaustion of benthic macroalgal organic matter as a fuel for remineralization to CO 2 and dilution effects as mixing of reef and offshore waters become more vigorous in the fall due to higher windspeeds and weather frontal passages as observed at the BATS site (Bates, 2007).Since the Bermuda coral reef is a high-latitude reef that experiences strong seasonality in [CO 2− 3 ], aragonite and other environmental conditions (e.g., light, temperature) compared to tropical reef counterparts, we expect that the CREF mechanism would be attenuated in tropical reefs, and not as strongly manifested as shown for the Bermuda reef.

Ocean acidification, future seasonal decalcification and critical [CO 2− 3 ] and aragonite thresholds of the Bermuda coral reef
There is growing evidence from experimental and modeling studies that ocean acidification and decreasing [CO 2− 3 ] and aragonite will negatively affect marine calcifiers and ecosystems, but relatively little evidence exists from studies of the natural environment.For the Bermuda reef, we show that rates of calcification for D. labyrinthiformis were strongly correlated with [CO 2− 3 ] and aragonite .Ocean acidification and the gradual decline of [CO 2− 3 ] and aragonite should have impacted coral calcification in the past.Historical records of coral calcification on tropical reefs show a decline over the recent past (e.g., Wilkinson, 2000;Edmunds, 2007;Edmunds and Elahi, 2007;Cooper et al., 2008;De'ath et al., 2009).In Bermuda, calcification rates of mature colonies of D. labyrinthiformis sampled at Hog Reef have also been reconstructed using coral skeletal density analyses (A.Cohen and N. Jacowski, unpub. data;Cohen et al., 2004).Such historical records show that skeletal density for D. labyrinthiformis has declined from a high of 4.5 g cm −3 year −1 in 1959 to a low of 3 g cm −3 year −1 in 1999, a change of 1.5 g cm −3 year −1 , or decrease of ∼33%.At the BATS site offshore from the island of Bermuda, over the last 25 years, the observed annual rate of [CO 2− 3 ] decrease due to the oceanic uptake of anthropogenic CO 2 was 0.50 ± 0.03 µmoles kg −1 year −1 (Bates, 2007;Bates and Pe-ters, 2007).If the rate of [CO 2− 3 ] decrease is applied to the observed in situ correlation between skeletal growth of D. labyrinthiformis and [CO 2− 3 ] at Hog Reef (Fig. 3a), a ∼37% decrease in calcification would be predicted for the 1959-1999 period.Since these assessments are based on the same coral species, but using very different approaches, there seems to be strong evidence that ocean acidification has significantly decreased calcification rates on the Bermuda coral reef over the recent past.
We can also estimate the decrease in coral calcification due to ocean acidification from the pre-industrial period to present.At the BATS site, the observed [CO 2− 3 ] decrease of 0.50 ± 0.03 µmoles kg −1 year −1 is accompanied by an observed increase in salinity normalized DIC of 0.80 ± 0.06 µmoles kg −1 year −1 (Bates, 2007; see his Table 2).In the subtropical gyre of the North Atlantic, the increase in DIC due to uptake of anthropogenic CO 2 is estimated at ∼60 µmoles kg −1 (Sabine et al., 2004).Given the ratios of observed DIC/[CO 2− 3 ] change, we estimate that the mean [CO 2− 3 ] was ∼37.5 µmoles kg −1 higher in pre-industrial times compared to the 2002-2003 period of observations at Hog Reef.Since the mean annual skeletal growth of D. labyrinthiformis observed at Hog Reef was 0.47 mg CaCO 3 g −1 d −1 , the application of the skeletal growth/[CO 2− 3 ] correlation shown in Fig. 3 gives a hindcast estimate of mean annual skeletal growth of 0.97 mg CaCO 3 g −1 d −1 in the pre-industrial period (with a range of 0.78-1.15mg CaCO 3 g −1 d −1 ).Thus, our results suggest that coral calcification rates (for D. labyrinthiformis at least) at Hog Reef have declined by 52% compared to the pre-industrial period as a result of changes in seawater carbonate chemistry.
The future impact of ocean acidification on coral calcification on the Bermuda reef also appears to be negative.Based on a linear extrapolation, our in situ data suggests that the calcification rate of D. labyrinthiformis would reach zero at [CO 2− 3 ] and aragonite thresholds of ∼184 µmoles kg −1 and 2.65, respectively (for both skeletal growth nomalized to colony weight or colony surface area; Fig. 3).The aragonite threshold has a range of 2.22-3.08 at the 95% confidence level so some caution should be advised proscribing definitive thresholds.Furthermore, the dependence of community calcification on [CO 2− 3 ] and aragonite may not be linear, but rather based on a second or higher order relationship resulting in a weaker dependence closer to the critical threshold (Andersson et al., 2009).Nonetheless, with these caveats in mind, due to lower annual mean surface seawater [CO 2− 3 ] and aragonite in Bermuda compared to more tropical regions, the Bermuda reef should experience critical threshold values earlier than its tropical reef ecosystem counterparts in response to future acidification of the oceans.Given that the lowest observed [CO (i.e., 0.50 ± 0.03 µmoles kg −1 year −1 ) continues linearly in the near-future (Bindoff et al., 2007), we anticipate that the Bermuda coral reef should experience seasonal periods of zero calcification rates (i.e., NEC reef = 0) within the next decade.Silverman et al. (2007) suggest that decalcification of coral reefs occurs when the gross calcification rate is equal to or less than 20% of the pre-industrial calcification rate.Given our observations of skeletal growth rates of ∼0.28-0.65 mg CaCO 3 g −1 d −1 for D. labyrinthiformis colonies, the Bermuda reef is currently about 30% of the mean preindustrial calcification rate (i.e., 0.97 mg CaCO 3 g −1 d −1 ).
During wintertime, NEC is close to zero, a condition where dissolution and calcification are nearly in balance.Salinity normalized alkalinity data from Hog Reef (2002Reef ( -2003) ) also exhibits close to zero difference between onshore and offshore values in the wintertime and during the late summer when net heterotrophy on the reef suppresses and calcification.This suggests that the threshold for when NEC reef equals zero may already be occurring seasonally (2005)(2006) data from Hog Reef also shows NEC reef values close to zero or negative values indicating net dissolution; N. R. Bates and A. J. Andersson, unpublished data).Thus for the Bermuda coral reef, there are periods when the balance of calcification (from corals and other calcifiers such as coralline algae) and dissolution are equal, with the likelihood of net decalcification going forward in time as shown experimentally for reef mesocosms (Andersson et al., 2009).A potentially ameliorating process, as discussed earlier in Sect.4.4, may be that net autotrophy of the reef during winter and spring (as part of the carbonate chemistry coral reef ecosystem feedback) which enhances [CO 2− 3 ] and aragonite may delay the onset of zero NEC or decalcification going forward.
In the near-future, the above scenarios predict that the Bermuda coral reef will experience seasonal decalcification for increasing periods of the year.Given that the Bermuda coral reef experiences a maximum [CO 2− 3 ] seasonality of ∼60 µmoles kg −1 , we might expect that the reef system will experience seasonal decalcification for a further 100-140 years, if the long-term trend of [CO 2− 3 ] reduction continues under IPCC assessments of future anthopogenic CO 2 release.During this period, we anticipate that suitable conditions for corals and other organisms to calcify will decrease progressively going forward in time.In addition, seasonal decalcification will impact such processes as dissolution of the framework structure of the reef and settlement of juvenile corals.This impact is difficult to predict, but most likely negative.In the next century, carbonate saturation states will transition into conditions that no longer facilitate coral reef calcification.As discussed earlier, if anthropogenic CO 2 emissions continue to accelarate, this transition will occur earlier in time.Due to the seasonality of carbonate chemistry on the Bermuda coral reef, the critical thresholds for initiation of coral decalcification are not sharp transitions as suggested by Silverman et al. (2009), but relatively ex-tended transitions that potentially extend over a period of many years.Since, the Bermuda coral reef is a high-latitude reef that experience strong seasonality, we expect that the tropical reef counterparts (with reduced seasonality of temperature, light, NEP, and NEC) will have attenuated seasonality of carbonate chemistry.Thus, we anticipate that the period of seasonal decalcification on tropical reefs will be shorter compared to higher latitude reefs.

Conclusions
In our study, we show that rates of coral calcification were closely coupled with seawater carbonate chemistry [CO 2− 3 ] and aragonite , in the natural environment, rather than other environmental factors such as light and temperature.Our field observations provide sufficient data to hypothesize that there is a seasonal carbonate chemistry coral reef ecosystem feedback (i.e., CREF hypothesis) between the primary components of the reef ecosystem (scleractinian hard corals and macroalgae) and seawater carbonate chemistry.It is also likely that this seasonal phenomenon is present in other tropical reefs although attenuated compared to high-latitude reefs such as Bermuda.Furthermore, due to lower annual mean surface seawater [CO 2− 3 ] and aragonite in Bermuda compared to more tropical regions, the Bermuda coral reef will likely experience seasonal periods of zero NEC within a decade in response to future acidification of the oceans.It appears that the entire reef may already be experiencing periods of zero NEC during the wintertime, resulting in a transition to net decalcification (i.e., net dissolution over calcification).As such, the Bermuda coral reef appears to be one of the first responders to the negative impacts of ocean acidification among tropical and subtropical reefs.Furthermore, we anticipate that the Bermuda coral reef (as well as other high latitude reefs) will likely be subjected to "seasonal decalcification" with wintertime decalcification occuring many decades before summertime decalcification.However, net autotrophy of the reef during winter and spring, as part of the CREF feedback process may delay the onset of zero NEC or decalcification going forward.Thus, on societally relevant time-scales, we expect that the Bermuda reef will endure an extended transition to decalcified conditions over a period of decades rather than a short transition at sharply-defined critical thresholds expected for tropical coral reef counterparts.

Fig. 1 .
Fig.1.Location of rim and terrace reefs of Bermuda, the North Lagoon and island of Bermuda, and seasonal changes in temperature, salinity and nitrate+nitrite.(a) Two reef sites, Hog Reef (red symbol) and Twin Breakers (orange symbol), were chosen as representative of the broad rim reefs that enclose lagoonal waters of the North Lagoon.The North Lagoon contains patch coral reefs and extensive sand area, with two sites (Crescent 1 and 2; green symbol) representative of patch reefs.The track of weekly underway, shipboard sampling from the R/V Atlantic Explorer (green dashed line) and M/V Oleander (blue dashed line) are shown.The offshore Hydrostation S (blue symbol), Bermuda Atlantic Time-series Study (BATS; purple symbol) and Bermuda Testbed Mooring (BTM) sites are also shown(Bates, 2007).The CARIOCA pCO 2 buoy was deployed at Hog Reef from 2002 to 2003; (b) seasonal changes in temperature ( • C; open squares) and salinity (grey diamond) at the North Channel site in the North Lagoon from 2001 to 2006, and; (c) seasonal changes in nitrate+nitrite (µmoles kg −1 ) at the North Channel site in the North Lagoon from 2001 to 2006.North Channel WQMP data courtesy of Drs Richard Owen and Ross Jones (MEP, 2006; http://www.bios-mep.info/;executive summary).

Fig. 2 .
Fig. 2. Time-series of physical, chemical and biological variables from the coral reefs of Bermuda from August 2002 to October 2003.(a) Surface temperature ( • C) and short wave radiation (Q sw ; W m −2 ) from the coral reefs of Bermuda.Surface temperature was collected hourly at Hog Reef (∼15 km NW of the island of Bermuda) using a CARIOCA buoy (red line), and daily average from a temperature logger at 5 m deep (orange line).The red diamond symbols denote surface temperature collected during visits to Hog Reef.The daily short wave radiation (Q sw ) was calculated from meteorological measurements collected hourly from the island of Bermuda by the Bermuda Weather Service.(b) Wind speed (grey line; mph) and surface seawater pCO 2 (µatm; blue line).Wind speed was collected hourly from the island of Bermuda by the Bermuda Weather Service.The blue diamond symbols denote values of seawater pCO 2 determined from DIC and alkalinity measurements.(c) Time-series of DIC (black diamond; µmoles kg −1 ) and alkalinity (open circle; µmoles kg −1 ) from Hog Reef.The grey diamond and circle denote DIC and TA observed offshore at the BATS site.(d) Time-series of [CO 2− 3 ] (black square; µmoles kg −1 ) and aragonite (open triangle) from Hog Reef.The grey square denote [CO 2− 3 ] observed offshore at the BATS site.(e) Timeseries of [CO 2− 3 ] (black square; µmoles kg −1 ) and in situ skeletal growth rate (i.e., G diploria ; grey circle; Hog Reef and open diamond, Twin Breakers; mg CaCO 3 g −1 d −1 ) for the massive coral Diploria labyrinthiformis from Hog Reef and Twin Breakers.The horizontal bars denote length of time for each in situ skeletal growth determination.

Fig. 3 .
Fig. 3. Relationship between in situ skeletal growth rate of D. labyrinthiformis (i.e., G diploria ) at Hog Reef against mean [CO 2− 3 ], aragonite , temperature and light conditions observed at Hog Reef.Skeletal growth rates are expressed either as mg CaCO 3 g −1 d −1 (black square) or as skeletal growth rate per unit surface area (mg CaCO 3 cm −2 d −1 ; open diamond).(a) Relationship between in situ skeletal growth rate of D. labyrinthiformis and average [CO 2− 3 ] (observed at Hog Reef during the concurrent skeletal growth rate measurement time period.)Regression statistics were: 75.77x + 184.2, r 2 = 0.68 (skeletal growth rate per colony weight) and 49.30x + 183.3, r 2 = 0.69 (skeletal growth rate per unit surface area).(b) Relationship between in situ skeletal growth rate of D. labyrinthiformis and average aragonite observed at Hog Reef during the concurrent skeletal growth measurement time period.Regression statistics were: 0.976x + 2.65, r 2 = 0.68 (skeletal growth rate per colony weight) and 0.629x + 2.65, r 2 = 0.68 (skeletal growth rate per unit surface area).The 95% confidence levels for the zero skeletal growth intercept was 2.22-3.08 and 2.21-3.08,respectively.(c) Relationship between in situ skeletal growth rate of D. labyrinthiformis and temperature ( ˚C) observed at Hog Reef during the concurrent skeletal growth measurement time period.Regression statistics were: 13.44x + 17.36, r 2 = 0.28 (skeletal growth rate per colony weight) and 9.655x + 16.56, r 2 = 0.35 (skeletal growth rate per unit surface area).(d) Relationship between in situ skeletal growth of D. labyrinthiformis and average light (W m −2 ) observed at Hog Reef during the concurrent skeletal growth measurement time period.Regression statistics were: 6617.8x+ 2001.4,r 2 = 0.27 (skeletal growth rate per colony weight) and 3791.8x+ 2281.1, r 2 = 0.21 (skeletal growth rate per unit surface area).

Fig. 4 .
Fig. 4. Annual composite and comparison of surface seawater pCO 2 data (µatm) collected over the last twelve years from the coral reef of Bermuda and offshore in the North Atlantic Ocean at BATS and the Bermuda Testbed Mooring (BTM).All seawater pCO 2 datasets have been adjusted to the year 2006 using the longterm trend of +1.7 µatm year −1 observed at the BATS site in the North Atlantic Ocean from 1983-2006 (Bates, 2007).Coral reef seawater pCO 2 datasets include: (1) surface seawater pCO 2 from October 2002 to January 2003 collected hourly at Hog Reef using a CARIOCA buoy (red line); (2) surface seawater pCO 2 from April 2002 to September 2003 hourly at Hog Reef using a CAR-IOCA buoy (peach line); (3) surface seawater pCO 2 from October 1998 to November 1998 hourly at Hog Reef using a CARI-OCA buoy (orange line) (Bates et al., 2001); (4) surface seawater pCO 2 (brown closed circles) calculated from surface DIC and alkalinity samples collected at Hog Reef from July 2002 to November 2003; (5) daily mean surface seawater pCO 2 (purple closed circle) collected along the southeastern terrace and rim coral reefs of the North Lagoon, Bermuda, from the R/V Weatherbird II during ∼150 cruises between 1994 and 1998.Offshore seawater pCO 2 datasets include: (6) surface seawater pCO 2 (grey open diamond) from November 2005 to December 2006 collected every 3 h at the BTM site [C.L. Sabine and N. R. Bates, unpub.data] and; (7) daily mean surface seawater pCO 2 (black closed diamond) collected every 2 min at the BATS site from the R/V Weatherbird II during ∼150 cruises between 1994 and 1998(Bates, 2007).

Fig
Fig. 5. (a) Annual composite of rates of NEC reef (g CaCO 3 m −2 d −1 ; green line) and NEP reef (g C m −2 d −1 ; blue line) for the Bermuda reef using seawater TA and DIC data from Hog Reef and BATS, and the alkalinity anomaly-water mass residence technique.Positive NEC reef values represent net calcification, and negative values represent net dissolution, with the zero line denote by grey dashed line.Positive NEP reef values represent net heterotrophy, and negative values represent net autotrophy, with the zero line denote by grey dashed line.(b) Annual composite and comparison of surface seawater [CO 2− 3 ] data (µmoles kg −1 ; gray circles) and skeletal growth rates (i.e., G diploria ; mg CaCO 3 g −1 d −1 ).(c) Repeat of panel b showing the CREF hypothesis superimposed on Hog Reef data.In early summer, enhancement of [CO 2− 3 ] and calcification during June and July [green arrow] occurs due to negative NEP.In late summer, suppression during September and October on the Bermuda reef due to positive NEP [blue arrow].The dashed line illustrates the hypothesized [CO 2− 3 ] in absence of the feedback on carbonate chemistry due to seasonal changes in NEP.(d) Seasonal composite of alkalinity difference (i.e., nTA offshore−onshore ) between Hog Reef and the BATS site for the periods 2002-2003 (square symbol) and 2005-2006 (circle symbol).

Fig. 6 .
Fig. 6.Annual rates of NEC reef (g CaCO 3 m −2 d −1 ) plotted against water depth (m) using the alkalinity anomaly-water mass residence technique, and mass balance.Different water residence times (τ ) of 1 to 4 days are plotted with isolines in blue.NEP reef (g C m −2 d −1 ; green lines) are superimposed as isolines in green with negative values indicating net autotrophy.The most appropriate water depth of 6 m and residence time of 2 days for the Bermuda reef is shown by the square.

Fig. 7 .
Fig. 7. Rates of NEC reef (g CaCO 3 m −2 d −1 ) against NEP reef (g C m −2 d −1 ) using onshore and offshore seawater TA and DIC data from Hog Reef and BATS.NEC reef and NEP reef data are shown in Fig. 5a using seawater TA and DIC data, and nTA offshore−onshore and nDIC offshore−onshore data shown in Table3.The regression statistics for the line are: −0.244x + 0.700, r 2 = 0.607.Arrows indicates direction of net autotrophy (i.e., −NEP reef ), net heterotrophy (+NEP reef ), net calcification (i.e., +NEC reef ) and net dissolution (i.e., −NEC reef ).For example, the upper left quadrant denotes conditions on the reef with net heterotrophy and net dissolution.

Table 1 .
Silverman et al., 2007. 3 ] variability observed at coral reef sites.Diurnal and seasonal [CO 2− 3 ] values were calculated using average alkalinity and pH observed on the Eilat coral reef bySilverman et al., 2007.
• C was observed in August 2003 a couple of months after the seasonal solar input maxima.For context, winter temperatures on the Bermuda coral reef are typically 1-2 • C cooler than the surrounding offshore Sargasso Sea

Table 2 .
In situ rates of skeletal growth of Diploria labyrinthiformis from Hog Reef on the rim reef of Bermuda.
: standard deviation of skeletal growth rates, sea surface temperature (SST) and Q sw are also shown in the Table.aSeveral of the coral specimens had moderate signs of bleaching potentially suppressing coral skeletal growth; b surface temperatures only available for day of year 288-301; c in situ during Hurricane Fabian. Note 2− 3 ] on the Bermuda coral reef in 2002-2003 was ∼190 µmoles kg −1 during winter, and assuming that the rate of [CO 2− 3 ] decrease www.biogeosciences.net/7/2509/2010/Biogeosciences, 7, 2509-2530, 2010