Coral reefs are predicted to be one of the ecosystems most sensitive to ocean acidification. To improve predictions of coral reef response to acidification, we need to better characterize the natural range of variability of pH, partial pressure of carbon dioxide (pCO2) and calcium carbonate saturation states (Ω). In this study, autonomous sensors for pH and pCO2 were deployed on Media Luna reef, Puerto Rico over three seasons from 2007 to 2008. High temporal resolution CaCO3 saturation states were calculated from the in situ data, giving a much more detailed characterization of reef saturation states than previously possible. Reef pH, pCO2 and aragonite saturation (ΩAr) ranged from 7.89 to 8.17 pH units, 176–613 μatm and 2.7–4.7, respectively, in the range characteristic of most other previously studied reef ecosystems. The diel pH, pCO2 and Ω cycles were also large, encompassing about half of the seasonal range of variability. Warming explained about 50% of the seasonal supersaturation in mean pCO2, with the remaining supersaturation primarily due to net heterotrophy and net CaCO3 production. Net heterotrophy was likely driven by remineralization of mangrove derived organic carbon which continued into the fall, sustaining high pCO2 levels until early winter when the pCO2 returned to offshore values. As a consequence, the reef was a source of CO2 to the atmosphere during the summer and fall and a sink during winter, resulting in a net annual source of 0.73 ± 1.7 mol m−2 year−1. These results show that reefs are exposed to a wide range of saturation states in their natural environment. Mean ΩAr levels will drop to 3.0 when atmospheric CO2 increases to 500 μatm and ΩAr will be less than 3.0 for greater than 70% of the time in the summer. Long duration exposure to these low ΩAr levels are expected to significantly decrease calcification rates on the reef.
 The oceans absorb approximately one third of the anthropogenic CO2 emitted to the atmosphere [Sabine et al., 2011]. While this CO2uptake helps ameliorate human-caused greenhouse warming, the amount of absorbed CO2 is so massive that it is significantly changing the chemistry of the oceans [Feely et al., 2004]. When CO2 reacts with seawater it forms carbonic acid, lowering pH, carbonate ion concentration ([CO32−]) and CaCO3 saturation states [e.g., Cao et al., 2007; Fabry et al., 2008; Orr et al., 2005]. This CO2-driven “ocean acidification” has already decreased sea surface pH by more than 0.1 pH units since the beginning of the industrial revolution [Caldeira and Wickett, 2003]. Once anthropogenic CO2 enters the oceans there is no practical way to remove it and the oceans will require thousands of years to naturally return to a higher pH state [The Royal Society, 2005; Solomon et al., 2009].
 Increasing dissolved CO2, described as the partial pressure of CO2 (pCO2), and decreasing pH will likely affect many marine organisms and alter ecosystem community structure [Fabry et al., 2008; The Royal Society, 2005]. Corals and other calcifying organisms are particularly at risk due to their dependence on CO32− concentration and CaCO3 saturation states (Ω) [Gledhill et al., 2008]. Laboratory tests have shown that as the pCO2 of seawater is increased, coral CaCO3 production begins to decline [Jokiel et al., 2008; Langdon et al., 2003; Langdon and Atkinson, 2005; Leclercq et al., 2000]. These experiments provide insights into how corals may react to changing CO2 levels, but it can be difficult to extrapolate the data to natural systems, in part because the natural range of the CO2 system that corals are exposed to is not well characterized. This lack of information is largely due to the difficulty of making CO2 system measurements with sufficient duration and temporal resolution. Two of the inorganic carbon parameters, such as pCO2, pH, total alkalinity (AT) or total dissolved inorganic carbon (DIC), must be measured to calculate the seawater CO2 system, including CaCO3saturation states, restricting most past studies to ship-based sampling. To obtain more continuous inorganic carbon system data for coral reefs, studies have used in situ potentiometric pH sensors [e.g.,Frankignoulle et al., 1996; Gattuso et al., 1999; Yates and Halley, 2006; Santos et al., 2011] and autonomous pCO2 systems [e.g., Kayanne et al., 2005; Bates et al., 2001]. However, with a few exceptions [e.g., Bates et al., 2001; Drupp et al., 2011], these studies have been short duration (days to weeks). Consequently, there is a lack of long-term high temporal resolution pH,pCO2, and Ω data for coral reefs.
 There now exists improved measurement technology that can be used for coral reef biogeochemical studies. The revitalization and improvement of the indicator-based pH method [Byrne and Breland, 1989] have made precise and accurate pH data available for CO2 system calculations compared to past studies that used electrochemical pH measurements. Several in situ pH systems have now been developed based on the spectrophotometric pH methodology [Liu et al., 2006; Nakano et al., 2006; Seidel et al., 2008]. The SAMI-pH, or Submersible Autonomous Moored Instrument for pH, is an indicator-based sensor capable of long-term, in situ pH measurements [Martz et al., 2003; Seidel et al., 2008; Cullison Gray et al., 2011; Emerson et al., 2011]. Recent studies have shown that [CO32−] and CaCO3 saturation states can be accurately calculated from in situ pH and pCO2 time series data [Cullison Gray et al., 2011]. In this study, we deployed the SAMI-pH with the autonomouspCO2sensor SAMI-CO2 [DeGrandpre et al., 1995] in three separate seasons off the southern coast of Puerto Rico spanning the years 2007–2008. Our goals here are to report the observed diel to interseasonal range of reef pH, air-sea CO2 fluxes, and CaCO3 saturation states, evaluate the controlling processes, and compare our results with other studies.
2.1. Site Description
 Measurements were focused on Media Luna Reef in the La Parguera shelf reef system (Figure 1). The reef is a 1.5 km long, emergent shelf reef 3.3 km south of La Parguera, Puerto Rico and is adjacent to an extensive mangrove system [Acevedo et al., 1989]. Water on the reef has primarily an oceanic source, with a general east to west current flow. The location was selected to take advantage of measurements being made at the NOAA Integrated Coral Reef Observing Network (ICON) buoy (site B, Figure 1). Media Luna met NOAA's requirements that the site be within a U.S. territory, have a bottom surface hard enough to drill into, be accessible for cleaning every ten days to two weeks and be located on the lee side from prevailing winds (J. Hendee, personal communication, 2009). The University of Puerto Rico, Mayagüez field station on Magueyes Island in La Parguera, Puerto Rico provides maintenance and research support for the buoy.
 Two sites were selected for this study, one at the head (upwind) and one at the tail of the reef, in order to characterize spatial variability and to determine if net calcification could be quantified along the reef. Three deployments spanning June 19–August 21, 2007, January 7–March 14, 2008, and September 16–November 21, 2008, are referred to as summer, winter and fall deployments, respectively. After examining current and salinity data from the second deployment, we determined that water from site A was not traveling directly to site B. For the third deployment, a new reef tail site was chosen (site C, Figure 1) so that the site would be directly in the mean flow path of water coming from site A.
2.2. Time Series Data
 Air and water temperature, salinity, photosynthetically active radiation (PAR), and wind velocity at 6.5 m were measured hourly on the NOAA ICON mooring. Water temperature, salinity and PAR sensors were located near the bottom at ∼5 m depth. An ADCP (RD Instruments 1200 KHz Workhorse Monitor) was deployed at the reef head during the winter and fall periods. The SAMI-pH and SAMI-CO2 sensors were deployed on the bottom at sites A and C and approximately 1 m from the bottom at the ICON site B (Figure 1). The sensors were programmed to make measurements every half hour. The SAMI-pH and SAMI-CO2are indicator-based sensors (Sunburst Sensors). The SAMI-pH combines metacresol purple (mCP) indicator with a seawater sample to determine pH [Martz et al., 2003; Seidel et al., 2008]. The SAMI-pH accuracy was checked prior to deployment by comparison with UV-VIS measurements of seawater samples or with tris buffer in synthetic seawater [DelValls and Dickson, 1998]. These measurements found an accuracy ranging from 0.001 to 0.005 pH units and precision of ±0.0006 pH units. In the SAMI-CO2 sensor, CO2 equilibrates across a silicone rubber membrane filled with bromothymol blue (BTB) indicator [DeGrandpre et al., 1995]. The SAMI-CO2s were calibrated with CO2 gas mixtures, verified with an infrared CO2analyzer (LI-COR, LI-840A), and then checked for accuracy in a 200 L water tank. A membrane contactor (Membrana Liqui-Cel MiniModule) was used to equilibrate a flowing air stream with the CO2 in the tank. The equilibrated air passed through a gas drier (Perma Pure), then into an infrared CO2analyzer (LI-COR, LI-840A) for detection. A bubbling stone, attached to a room air supply and a soda lime CO2 trap, was used to drive down the CO2in the tank, when necessary. Based on these tests, accuracy of the SAMI-CO2 is ∼5 μatm with a precision of ±1 μatm. Dissolved oxygen (Aanderaa 4175 Optode) was measured every half hour during the last two deployments. Oxygen sensors were calibrated in saturated and zero oxygen solutions before deployment. The O2 sensors have an accuracy of ∼8 μM based on the factory calibration with a resolution of <1 μM. Chlorophyll-a (chl-a) fluorescence was measured using an in situ fluorometer (Chelsea Instruments Minitracka) with chl-a concentration calculated from the factory calibration.
 Atmospheric pCO2 was calculated from hourly dry mole fraction CO2 at Mauna Loa atmospheric CO2 time series station which is the closest in latitude (19°32′N) to the Media Luna sites at 17°56′N. Mole fraction was converted to pCO2using local reef barometric pressure and sea surface temperature (water vapor). Oceanic temperature and salinity data for the Caribbean Time Series (CaTS) location (17°36′N, 67°W) were used for comparison with the coastal data. Near daily resolution temperature data were derived from satellite data at CaTS using the NOAA Comprehensive Large Array-data Stewardship System (CLASS) for 2007. The salinity data from CaTS were measured using a conductivity-temperature-depth sensor (Sea-Bird Electronics SBE19 CTD) on a nearly monthly basis from 1993 to 1999 [Corredor and Morell, 2001].
2.3. Discrete Measurements
 Discrete samples were collected from the reef head (A) and reef tail (B or C) once or twice a day during the first 1–2 weeks of each deployment, as well as several other times throughout the deployment to verify sensor measurement data quality and provide additional data for interpretation of the in situ time series. Samples were collected using a Van Dorn horizontal water sampler (Wildco) and were stored in the dark at room temperature until analyzed. ATsamples were analyzed within 24 h by open-cell potentiometric titration following DOE procedure SOP3b [Dickson et al., 2007] using a custom-built automated Gran titration system [Langdon et al., 2000]. Alkalinity certified reference materials (CRMs) [Dickson et al., 2003] were used to standardize the HCl titrant. Accuracy and precision of field samples was ±1.5 μmol kg−1. pH was measured spectrophotometrically on the total pH scale within 8 h of collection on a UV-VIS spectrophotometer (Shimadzu UV-1601) following DOE procedures [Clayton and Byrne, 1993; Dickson et al., 2007]. Precision was ±0.003 pH units. During the third (fall) deployment, tris seawater buffers [DelValls and Dickson, 1998] became available to assess the accuracy of the UV-VIS pH measurements. The accuracy was 0.0061 ± 0.0023 pH units. DiscretepCO2, calculated from AT and pH data, was used to quality control the in situ pCO2. Oxygen samples were analyzed using the Winkler titration method [Culberson and Huang, 1987]. The comparison between sample and in situ measurements is presented in the Results.
2.4. CO2 Equilibrium and Temperature Calculations
 All CO2 system calculations were performed using CO2SYS [Pierrot et al., 2006] with K1 and K2 from Mehrbach et al.  refit by Dickson and Millero , KSO4 from Dickson  and pH on the total scale. Temperature effects were examined using CO2SYS by inputting pH and pCO2 using the in situ temperature and salinity, with the program set to output the pH and pCO2 at the mean annual temperature (28°C).
2.5. Calcium Carbonate Saturation States
 Saturation states were calculated in CO2SYS using Ω = [Ca2+][CO32−]/Ksp where Ksp is the temperature, pressure and salinity dependent solubility product constant. [Ca2+] was determined using its known relationship with salinity, whereas [CO32−] was calculated from two measured inorganic carbon parameters. In Cullison Gray et al. , we showed that in situ pH and pCO2 data can be used to accurately calculate [CO32−] and Ω. We use the high-resolution pH andpCO2 time series presented here to characterize reef aragonite (ΩAr) and calcite (ΩCa) saturation states with unprecedented detail.
2.6. Air-Sea Flux Calculations
 The air-sea CO2 flux was estimated using FGAS = kSΔpCO2, where k is the gas transfer velocity, S is the gas solubility and ΔpCO2 is the difference in pCO2 between the surface ocean and the atmosphere. A negative FGAS represents a flux from the atmosphere to the ocean. The gas transfer velocity was estimated using the wind speed relationship of Ho et al. . Tidal currents can affect gas transfer rates in shallow coastal and estuarine locations [Borges et al., 2004] but was not likely to be important in this area because of the small tidal range (max of ∼0.4 m). The net annual flux was calculated from pCO2 interpolated between seasons with k from hourly winds at the NOAA ICON site corrected to 10 m [Large et al., 1995].
3.1. Reef Hydrography
 Hydrographic data from the open ocean CaTS site, located 52 km south of Puerto Rico (17°36′N, 67°W), and Media Luna reef are compared in Figure 2. The reef temperature and salinity records are in general very similar to the CaTS site, indicating that local processes, with exceptions noted below, do not significantly alter the T and S of the open ocean source water. These data also show that the period of our study is, at most times, representative of typical conditions on the reef, based on comparison with the ICON Media Luna data from other years. CTD casts conducted throughout the study indicated the entire water column over the reef was always well mixed. Current meter measurements showed that the mean water flow across the reef was toward the west-southwest (from sites A to C,Figure 1). Tides were diurnal with water depth changes between 0.05 and 0.4 m.
3.2. Data Overview
 Data for all three deployments are shown in Figure 3. Gaps in the fall pH, pCO2, O2and temperature data correspond to when the instruments were removed from the reef preceding severe weather. For an unknown reason, the winter sample pH data did not closely match the in situ pH values. Because the SAMI-pH and discrete AT combination produced reasonable pCO2values compared to the SAMI-CO2 (Figure 3b), the winter sample pH values were not used. The SAMI-pH and SAMI-CO2 and the discrete samples, excluding the winter pH, matched to within +0.0006 ± 0.0082 pH units (n = 86) and −1 ± 14 μatm (n = 86), respectively, reported as the mean difference ± standard deviation of the difference. No offsets were applied to the SAMI pH data; however, pCO2 data were corrected to the discrete (calculated) pCO2 when offsets were present. Only constant offsets were applied and there was no detectable drift. The difference between the O2 optodes and discrete Winkler O2 measurements was −0.6 ± 3.7% saturation (n = 28); no offsets were applied to the O2data. The pH-pCO2 derived CaCO3 saturation states (Figure 3h) compared well with those calculated from discrete pH and AT with differences of −0.002 ± 0.14 (n = 75) for aragonite and −0.002 ± 0.20 (n = 75) for calcite. Random spatial and temporal mismatches between sampling and in situ measurements likely contributed to the relatively large standard deviations between in situ and discrete biogeochemical data sets.
 The mean data ranges in Figure 3 are given in Table 1. Over the three deployments, temperature and salinity ranged from 25.5 to 30.7°C and 31.4–36.3, respectively. Oxygen saturation varied from 62 to 138%. Reef pH and pCO2 were between 7.89 and 8.17 pH units and 176–613 μatm, respectively, with the lowest pH and highest recorded pCO2 in the fall. The ranges of pH and pCO2 were dominated by the seasonal cycle but the diel ranges were also large, at times encompassing the seasonal pH range (Table 1) and extending the pH minimum to ∼7.89 for a short period in October (Figure 3). The diel ranges were largest in the fall. The ΩAr and ΩCa ranged from 3.0–4.4 and 4.3–6.8, respectively and changed by 0.64–0.83 and 0.95–1.08, respectively, over a diel cycle.
Table 1. Average Water Column Value of Each Time Series Parameter by Season at the Reef Head (Site A)a
The mean diel range is shown below each average. AT is from all available discrete samples and DIC is calculated from pH and AT from all available discrete samples (reef head and tail). ND = no data.
29.3 ± 0.2 0.41
26.3 ± 0.4 0.45
29.1 ± 0.5 0.53
35.8 ± 0.22 0.08
35.2 ± 0.35 0.04
33.8 ± 0.50 0.10
8.01 ± 0.02 0.079
8.09 ± 0.02 0.067
8.00 ± 0.03 0.093
460 ± 33 77
356 ± 43 130
437 ± 44 130
AT (μmol kg−1)
2315 ± 6 ND
2295 ± 39 ND
2223 ± 30 ND
DIC (μmol kg−1)
1996 ± 10 ND
1974 ± 32 ND
1921 ± 21 ND
O2 saturation (%)
97.5 ± 9.5 28.6
101.6 ± 22 60.6
CO32− (μmol kg−1)
238 ± 17 41
250 ± 16 53
209 ± 16 43
3.94 ±0.24 0.64
3.94 ± 0.25 0.83
3.42 ± 0.26 0.72
5.90 ± 0.36 0.95
5.95 ± 0.38 1.25
5.13 ± 0.39 1.08
 Because of the interest in understanding how ocean pH relates to other measured parameters, correlations with pH, by season, are shown in Table 2. The highest pH correlations were found with pCO2 and ΩAr. The lower correlation between pH and pCO2 for the summer season is likely due to the use of pH from the reef tail site B and pCO2 from reef head site A, which was necessary due to missing data at the corresponding sites (Figure 3). Winter pH and ΩAr do not correlate indicating that pH is not always a good predictor of saturation states. The pH, pCO2 and ΩAr relationships are discussed in more detail in Section 4.5. There was no evidence of a biogeochemical signal associated with the tidal cycles (depth in Table 2).
Table 2. R2 Correlations Between pH and Other Parametersa
For summer, pH, temperature (T), salinity (S), and depth (D) are from site B and pCO2 is from site A (Figure 1). For winter and fall, all parameters are from site A except salinity and depth, which are from site B. ND = no data.
 Small biogeochemical and temperature gradients existed between the reef head and tail (Figure 3). Spatial differences are more clearly shown by comparing the monthly means (Figure 4) and seasonal averages (Table 3). There were larger differences during the first two deployments (summer, winter) when the sensors were located on opposite sides of the reef (Figure 1). At times large differences between sites A and C (fall) were observed, with the largest deviations in pH and pCO2 occurring soon after tropical storms Kate (September 21–23) and Omar (October 14–17) (Figure 3). Site A, which is not sheltered by nearby emergent reefs like the tail sites (Figure 1), may be more directly influenced by terrestrial inputs during these storm periods (also see section 4.1 below). While our hope was to quantify biogeochemical rates of change across the reef using the pH, pCO2, O2and current (ADCP) data (essentially the pH-AT technique) [e.g., Gattuso et al., 1999], the small differences during the winter relative to the large error in calculated AT [Cullison Gray et al., 2011] and failure of the ADCP in the fall (the ADCP was not deployed in the summer), prevented this. The spatial data do show, however, that the local water is not strongly influenced by a single reef and that the observed biogeochemical changes are, in general, representative of the broader reef-shelf system. To simplify the discussion that follows, the data collected at the reef head (site A,Figure 1) are primarily used and data at the reef tail are only discussed when data at the reef head were not available (see black and red traces in Figure 3).
Table 3. Average Difference Between the Media Luna Reef Head and Tail Sites (Head-Tail) for Each Measured Parameter by Seasona
Lines indicate that data was not available at one or both sites.
pH (pH units)
0.011 ± 0.022
0.012 ± 0.021
−0.008 ± 0.018
24.0 ± 26.2
6.0 ± 28.4
O2 Saturation (%)
6.9 ± 26.5
0.2 ± 0.1
0.1 ± 0.1
−0.04 ± 0.08
4.1. Comparison of pH With Other Reef Systems
 The large range of short-term and seasonal pH andpCO2 variability on Media Luna reef (Table 1) appears to be typical of coral reefs [Kayanne et al., 2005; Yates and Halley, 2006; Manzello, 2010; Santos et al., 2011]. Here we highlight data specifically for pH because it is one of the most commonly measured inorganic carbon parameters and because of the pH connection to ocean acidification. Saturation states are discussed below. Kayanne et al.  measured pH over an entire year with a pH sonde on a fringing reef (1.5–2.5 m depth) off Ishigaki Island, Japan and reported an average diel change of ∼0.5 pH units (7.9–8.4) and a seasonal pH change of ∼0.1–0.2 pH units. Manzello found a pH range of 7.65–8.26 for upwelling-impacted reefs located in the eastern Pacific Ocean. Reef pH was recorded over 24–48 h periods, approximately monthly, from 2000 to 2002 bySilverman et al. . Over the two-year period, the monthly pH average in the Red Sea reef lagoon (1.5–1.8 m water depth) varied by 0.1 pH units, from 8.2 to 8.3 pH units. The higher pH range found in this reef ecosystem is controlled by the high AT in the Red Sea. In Santos et al. , the diel pH ranged from 7.7 to 8.4 at Heron Island on the Great Barrier Reef. The large diel pH range found in reefs is dominated by primary production that is supported by the reef community [Kleypas et al., 2011]. In La Parguera, the productivity of the reef has been affected by frequent coral bleaching since the mid-1980s, with an increase of 0.7°C in the maximum summer temperature from 1966 to 1995 [Winter et al., 1998; García et al., 1998]. Moreover, reefs on the southern coast of Puerto Rico have shown decreased species diversity and coral cover due to terrestrial sediment inputs [Acevedo et al., 1989]. These impacts have likely resulted in decreased pH and pCO2 variability compared to healthy reefs [Kayanne et al., 2005].
4.2. Sources of Biogeochemical Variability
4.2.1. Physical Processes
 Here we evaluate processes that control pH and pCO2 variability. Saturation state variability is evaluated separately in section 4.5. As shown in Table 2, the pH was not correlated with temperature, salinity or depth (tides). Although heating and cooling are typically important contributors to pH and pCO2 variability, and on Media Luna reef there is a significant increase in temperature from winter to summer (Figure 3d), the temperature correlation is weak because other sources of variability muddle the relationship (discussed below in this section). To separate out the temperature forcing, the pH and pCO2 time series data were recalculated at a constant temperature (28°C), as described in section 2.4. The temperature adjusted data indicate that 50%, or 0.04 pH units, of the 0.08 pH unit seasonal change (Table 1) and 46%, or 48 μatm, of the 104 μatm seasonal pCO2 change (Table 1) were due to warming from winter to summer, with a similar magnitude but in the opposite direction from fall to winter.
 Variations in source water inputs, net community production (NCP = gross primary production – net community respiration), CaCO3production/dissolution and air-sea gas exchange could also contribute to the observed variability. We can estimate the oceanic source waterpCO2using an empirical SST-pCO2 relationship developed by Olsen et al. for the oligotrophic Caribbean Sea. Their temperature, latitude and longitude-dependent relationships were developed using shipboardpCO2 and remotely sensed sea surface temperature measured in the Caribbean Sea for the year 2002. This relationship was used with the local SST and location of the reef head to estimate the oceanic source water pCO2. We added the difference in the annual mean atmospheric pCO2 at Mauna Loa between 2002 and 2007 (10.6 μatm) to the calculated pCO2 to estimate the 2007 open ocean values, assuming the surface ocean pCO2 tracks the atmospheric CO2 increase. Mean summertime offshore pCO2 was 390 ± 2.5 μatm, or about 70 μatm below the mean summer reef pCO2 (Figure 3b and Table 1). During the winter, the estimated mean offshore pCO2 was 359 ± 4 μatm, very close to the reef mean of 356 μatm, while the fall offshore pCO2 (388 ± 4.6 μatm) was again well below the reef mean (437 μatm). The seasonal temperature range on the reef and open ocean are very similar, as shown in Figure 2 but, using CO2SYS, the seasonal range of the offshore water pCO2 shown in Figure 3 is mostly explained by temperature (i.e., the Olsen et al.  algorithm is primarily driven by the thermodynamic pCO2 variability), in contrast to the reef where, as shown above, only ∼50% of the observed seasonal range is due to seasonal heating and cooling. It is clear that the biogeochemical properties of offshore waters traversing the insular shelf are strongly imprinted by local processes. These observations are similar to those of Bates et al. , where the local reef profoundly alters the open ocean biogeochemical composition.
 The monthly mean values are compared to the temperature-corrected data inFigure 5. The constant temperature pCO2 is near atmospheric saturation in the winter (Figure 5b) whereas in the summer and fall it is significantly above saturation. As described by Takahashi et al. , this “residual” constant temperature pCO2represents contributions from water mass changes, air-sea gas exchange and biological processes. In the summer, when the reef water strongly resembles offshore water (based on temperature and salinity;Figure 2), only biological processes could generate the observed supersaturation; whereas in the fall there is significant freshwater input that could alter the inorganic carbon system. Straight dilution has negligible effects on pH and pCO2, estimated by decreasing the seasonal DIC and AT (Table 1) in direct proportion to salinity and recalculating in COSYS. Therefore, the fall supersaturation also requires some other explanation. These processes are discussed in the following section.
4.2.2. Biological Processes
 In this discussion biological variability is assumed to originate from NCP and calcification within the sediments, water column and corals. Calcification can be estimated by quantifying changes in AT over time [e.g., Gattuso et al., 1999]. We compared the balance between the open ocean and reef AT measured on discrete samples. Open ocean ATwas calculated from the temperature-salinity relationship ofLee et al.  using T and S at the CaTS site shown in Figure 2. Because our AT measurements are not for a single year, these calculations assume that changes in AT are consistent from year to year.
 The offshore calculated AT is primarily dependent upon salinity, i.e., the Lee et al.  relationship has a weak temperature dependence, with increasing AT throughout the winter, leveling off in the spring, decreasing in the late summer and increasing again in the late fall (Figure 6). These changes in AT (or salinity) are in response to seasonal rainfall, evaporation and freshwater inputs from the Amazon and Orinoco Rivers [Corredor and Morell, 2001]. The reef ATwas similar to offshore levels in the winter and spring but dropped below them in the summer. The large drop in late September corresponds to a period of intense rain (∼20 cm in 12 h) associated with tropical storm Kate. The differences between the salinity-normalized AT ( = 35xAT/salinity) for the reef and offshore waters can indicate CaCO3formation or dissolution on the reef. From January to June, the salinity-normalized AT at the offshore site decreased 3 μmol kg−1, due to the temperature dependence of the Lee et al. relationship. The salinity-normalized reef AT decreased 31 μmol kg−1, or a net calcification-driven AT decrease of −28 μmol kg−1. During the fall, salinity-normalized reef AT for late September was significantly higher than offshore values (maximum mean difference was +57 ± 9 μmol kg−1) (Figure 6). The high salinity-normalized AT could either be generated by CaCO3 dissolution or terrestrial (karst) inputs of AT as a result of the September rainstorm. Any freshwater inputs with nonzero ATwill increase salinity-normalized AT, as observed in other coastal areas [e.g., Kawahata et al., 2000].
 These seasonal AT changes were used to calculate the pH and pCO2 changes due to calcification/dissolution, by assuming that the change in DIC:AT stoichiometry is 0.5. The change in AT and DIC were added to the initial AT and DIC for each season, with DIC values estimated using CO2SYS and the pH, AT, T and S from discrete sample data at that time. The winter to summer pH and pCO2 changes that resulted from the 28 μmol kg−1 decrease of AT are −0.018 pH units and +15 μatm. In the fall, if the increase in salinity-normalized AT was solely due to CaCO3 dissolution, the pCO2 would decrease by ∼30 μatm. Alternatively, if the source of AT was runoff accompanied by DIC, pCO2 would increase by ∼20 μatm, assuming that the DIC increased the same amount as AT, i.e., AT was 100% bicarbonate alkalinity. The lower mean pCO2 observed in September (Figure 5) suggests that CaCO3 dissolution was the dominant process perhaps brought on by the large freshwater input and net respiration (see in this section below).
 Of the winter to summer pH and pCO2 changes shown in Table 1 and Figure 5, −0.02 pH units and 41 μatm could not be accounted for by heating and calcification. The remaining unexplained summer and fall pCO2 supersaturation likely originated from net respiration of organic carbon. The southwestern coast of Puerto Rico contains 995 ha of mangroves, as well as several offshore mangrove colonies on emergent portions of the reef [Cintrón et al., 1978; García et al., 1998]. Mangrove-derived organic carbon can subsequently be remineralized leading to highpCO2 values as observed in other coastal areas [Borges et al., 2003, 2005; Bouillon et al., 2007; Chen and Borges, 2009; Koné and Borges, 2008]. At Media Luna reef, organic carbon inputs from adjacent mangroves and seagrass beds would need to be high in the summer and fall and low in the winter to account for the difference between pCO2 levels. Koné and Borges  found higher pCO2 during the summer to fall rainy season and lower pCO2during the winter to spring dry season in waters with surrounding mangrove forests. At Media Luna reef, average January rainfall is <10% of that during the summer and fall months, supporting the seasonality of mangrove-derived organic matter input.
 To summarize the observed seasonal changes, the summer pCO2 supersaturation can be broken down to ∼46% heating/cooling, 14% net calcification, and 40% net respiration. In the fall, pCO2 and pH likely return to winter levels through net dissolution of CaCO3, net gas exchange to the atmosphere, along with a decrease in mangrove-derived organic carbon fluxes. Net production could also play an important role in the fall to winter decline. However, net production could not be discerned from water mass movement as the mean CO2 characteristics of the reef water begins to more closely resemble the offshore values during this transition (see 2008 salinity record in Figure 2 and winter pCO2 in Figure 3b).
4.3. Short-Term Processes
 While seasonal changes were large, short-term (primarily diel) processes were also important in establishing the range of biogeochemical variability on the reef. We looked more closely at the short-term variability before and after tropical storm Kate (September 21–23) both to further show the effects of organic matter input discussed above and to examine the dynamics of the diel cycle on the reef (Figure 7). First, and this is consistent throughout the time series data, pH and O2 concentration increased and pCO2 decreased during the day, showing that calcification and heating/cooling (for pH and pCO2) were secondary to primary production for controlling the diel cycles of pH and pCO2 on the reef [Kleypas et al., 2011]. It can be seen that both temperature and salinity dropped sharply during and after the storm (Figure 7e) and that low light levels (Figure 7g) initially led to a dramatic decrease in the diel variability of pH, pCO2 and O2. Approximately one week after this rain event a large algal bloom occurred which initially increased pH and drew down the pCO2, similar to post-storm blooms observed on other reefs [Drupp et al., 2011]. After September 30 the salinity stabilized and the mean and diel amplitude of pCO2 increased rapidly (and mean O2 decreased) suggesting that over the next few days net respiration primarily drove these changes. During this period, the pH reached its lowest recorded level (7.89 pH units) and had a diel range as large as 0.167 pH units. For comparison, the offshore source water pCO2 is estimated to have very small diel variability with pCO2 increasing by an average of ∼4 ± 1.5 μatm from heating during the day [Olsen et al., 2004]. The observed diel pCO2 cycle on the reef is typically larger than the source water diel cycle by more than an order of magnitude and, because of the biological signal, is in the opposite direction. The effects on ΩAr are also shown (Figure 7c) with ΩAr dropping to ∼3 during the periods of high freshwater input and high rates of respiration (low pH, high pCO2). The seasonal sources of variability on ΩAr are discussed in section 4.5.
4.4. CO2 Gas Flux
 The mean seasonal and annual CO2 gas fluxes were calculated to determine if Media Luna reef was a net source or a sink of CO2 to the atmosphere. As shown in Table 4 and Figure 5d, the reef was a source of CO2 to the atmosphere during summer and fall and a sink in the winter, controlled by the processes discussed above. It was a net source of CO2 to the atmosphere over an annual period, with a flux of +0.73 ± 1.7 mol m−2 year−1. By comparison, the annual flux using the Olsen et al. pCO2 was −0.04 ± 0.43 mol m−2 year−1 (Table 4). While the offshore source water has a negligible CO2 flux, seasonal heating, organic matter remineralization and net calcification make the reef a significant CO2 source. A compilation of five coral reef systems by Fagan and Mackenzie  showed net release of CO2 to the atmosphere between +1.2 and +1.8 mol m−2 year−1 with the exception of one reef near equilibrium (+0.1 mol m−2 year−1).
Negative values are a flux from the atmosphere into the ocean. Seasonal fluxes are in mmol m−2 day−1. Olsen et al.  fluxes were calculated using pCO2 predicted from their relationships, using the temperature at Media Luna reef, adjusted for changes in atmospheric pCO2 (see text). The annual Olsen et al.  value is calculated from the continuous temperature data whereas the Media Luna flux is calculated using all of the time series data with data interpolated between seasons.
 One of the major seasonal features in the Media Luna data is that ΩAr was very similar from winter to summer but dropped significantly in the fall – a surprising difference because the pH and pCO2 are more similar in the summer and fall. While the same processes that control pH and pCO2 on the reef will regulate ΩAr and ΩCa, there are some important differences. The summer Ω saturation states are close to the winter values (Figure 8 and Table 1) despite the summer's significantly lower pH and higher pCO2 (Figure 3 and Table 1). Summer saturation states could be expected to be lower because of terrestrial organic matter remineralization that reduces pH and elevates pCO2 as discussed above; however, the mean [CO32−] is similar for both seasons (Table 1). Seasonal variability in the reef inorganic carbon levels is strongly influenced by seasonal variability in offshore source water, as indicated by AT in Figure 6. During the summer “acidic” conditions, AT was near its highest level (Table 1), offsetting the pH effect on [CO32−] (Figure 3g). For example, if the mean winter AT is combined with the mean summer pH (Table 1), [CO32−] is near 210 μmol kg−1; whereas, the mean summer AT and pH estimate that [CO32−] is ∼230 μmol kg−1, approximately a 10% difference (T and S were held constant in this calculation). [CO32−] would be even higher in the summer except that summer net calcification, as discussed above, reduces the open ocean source water AT significantly. Solubility changes (changes in Ksp) were of comparatively lesser importance between winter and summer, with the higher mean summer temperature (lower solubility) and higher mean summer salinity (higher solubility) changing ΩAr by +1.5% and −2.1%, respectively, nearly offsetting each other.
 In the fall, mean temperature, pCO2 and pH are comparable to summer levels (Table 1) indicating similar contributions from calcification and respiration; however, saturation states are lower than in either winter or summer (Figure 3 and Table 1). The [CO32−] dropped from summer to fall due to a decrease in AT and DIC driven by rain and local freshwater runoff (Figure 3g). The change in [CO32−] alone dropped ΩAr by 12.2% with an additional 5.6% decrease due to a decrease in [Ca2+] (lower salinity, Figure 2b). The decrease in salinity also increased ΩAr by 7.4% through a decrease in Ksp. Consequently, ΩAris lower in the fall largely due to salinity-driven decreases in [CO32−] and [Ca2+].
Gledhill et al.  mapped ΩAr for Caribbean waters using empirical relationships for pCO2 and AT. They concluded that changes in temperature and freshwater input are the most important processes controlling seasonal ΩAr variability in the southern Caribbean. Their estimated ΩAr had a higher mean and smaller seasonal range (∼3.95–4.10 for 2006) compared to our results (Table 1). Because their ΩAr values were derived from open ocean variability, they do not capture the large contributions from organic matter respiration and other local events, such as freshwater runoff and calcification that tend to decrease ΩAr. In the summer and winter there are times that our saturation states are close to the mean values in Gledhill et al. ; however, the diel cycle, particularly in the winter, brings the aragonite saturation state close to 3 for short periods (Figures 3 and 7).
5. Future Implications
 Atmospheric CO2 values are expected to continue to rise from current values of ∼390 μatm, possibly up to 500 μatm or higher by 2035–2065 [Meehl et al., 2007], which will intensify ocean acidification. To estimate the effect of this increase on the pH and saturation state of the reef, we increased the measured pCO2 from each season and calculated the expected pH and saturation states under these conditions. The calculations used a constant AT of 2250 μmol kg−1 at the measured reef temperature and salinity. A 120 μatm increase in pCO2 (∼500 μatm total) decreases pH by 0.11, 0.12 and 0.08 for the summer, winter and fall seasons, respectively, relative to contemporary values. The ΩAr decreases by 0.99, 0.89 and 0.40 for the three seasons (Figure 8). Kleypas and Langdon  show that the response to acidification varies widely among different coral species, but that on average a 1.0 unit change in ΩAr results in a ∼20% decrease in calcification rate. Calcification rates on Media Luna reef are therefore likely to be significantly reduced by ocean acidification within the next 30–50 years. Long duration exposure to low ΩAr levels could intensify these effects. ΩAr is projected to be less than 3.0 from 50 to 70% of the time if pCO2 increases to 500 μatm (Table 5). In contrast, there was no exposure to <3.0 levels for preindustrial waters (Table 5). Mean ΩAr will be at the lowest recorded levels for all reefs (∼2.5) when atmospheric pCO2 reaches 655 μatm.
Table 5. Total and Longest Continuous Exposure to Saturation States ≤3.0 in Hours for Preindustrial, Current and Future pCO2 Levelsa
Mean pCO2 Level
The percentage of total time is shown in parentheses.
 Our in situ sensors have made it possible to accurately characterize pH and saturation states on coral reefs and have shown that the reef currently experiences large seasonal and diel variability. Similar studies are needed to characterize saturation state variability on other reefs. These studies should be combined with field measurements of calcification rates to verify laboratory and mesocosm data that show declining rates of calcification with decreasing ΩAr [Kleypas and Langdon, 2006]. Lastly, pH and pCO2 data can be used to develop regional geochemical models, such as the Caribbean saturation state model of Gledhill et al. , which often do not contain enough data for coastal and reef ecosystems to make accurate predictions in these areas.
 We thank Cory Beatty from the University of Montana and Helena Antoun, Valentine Hensley and Belitza Brocco from the University of Puerto Rico, Mayagüez for their assistance. We greatly appreciate the in-depth comments provided by two anonymous reviewers. We also thank the NOAA ICON program for data and mooring support. Funding for this research was provided by the National Science Foundation (grants OCE-0836807 and OCE-0628569) and a NASA Montana Space Grant Consortium Fellowship to S.E.C.G.