Bahía Culebra, located in the Gulf of Papagayo, North Pacific coast of Costa Rica (Fig. 1), is strongly influenced by the north easterly Papagayo winds. The strongest wind jets develop during the boreal winter (Amador et al., 2016) and are driven by large-scale variations of the trade winds (Chelton et al., 2000; Alfaro and Cortés, 2012). When Papagayo winds blow through the mountain gap between southern Nicaragua and northern Costa Rica, the resulting strong offshore winds on the Pacific side can lead to upwelling of cold and nutrient-enriched subsurface waters between December and April (McCreary et al., 1989; Brenes et al., 1990; Ballestero and Coen, 2004; Kessler, 2006). These cyclonic eddies also influence the magnitude and location of the Costa Rica Dome (CRD), which is located approx. 300 km off the Gulf of Papagayo (Fiedler, 2002). However, the CRD changes its distance to the Costa Rican coast throughout the year, as a result of differences in wind forcing (Wyrtki, 1964; Fiedler, 2002). During the dry season, particularly between February and April, offshore moving water masses strengthen upwelling at the coast and shoal the thermocline in the Gulf of Papagayo (Wyrtki, 1965, 1966; Fiedler, 2002). In May–June, during the onset of the rainy season, the CRD moves offshore (Fiedler, 2002; Fiedler and Talley, 2006) and the North Equatorial Countercurrent (NECC) can carry tropical water masses into Bahía Culebra until December, when upwelling sets in again (Wyrtki, 1965, 1966).

We measured in situ pH, pCO2 and seawater temperature (SWT) during two non-upwelling periods (15 days in June 2012 and 7 days in May–June 2013; Fig. 2). Measurements were undertaken with two Submersible Autonomous Moored Instruments (SAMI-pH and SAMI-CO2) (www.sunburstsensors.com, last access: 10 April 2017), in sampling intervals of 15 (June 2012) and 30 min (May–June 2013). SAMI sensors were deployed at the pier of Marina Papagayo (85∘39′21.41′′ W, 10∘32′32.89′′ N), on top of a carbonate sandy bottom in the inner part of Bahía Culebra (Fig. 1). The water depth varied approximately between 5 and 8 m depending on the tide, but sensors, hooked to the pier, moved up and down with the tide and were always at the same depth, 1.5 m below the surface. SAMI instruments measured pH (total hydrogen ion scale) and pCO2 spectrophotometrically by using a colorimetry reagent method (DeGrandpre et al., 1995, 1999; Seidel et al., 2008). Salinity from discrete samples was measured with a WTW probe (Cond3310) and was used for correction of pH values. Calculation of aragonite saturation state (Ωa) from parameters measured in situ with SAMI sensors is accurate (Cullison Gray et al., 2011; Gray et al., 2012), but discrete water samples were collected as often as possible to validate the instruments (Fig. 3). The 250 mL borosilicate bottles were filled with seawater at 30 cm below the surface and preserved with 200 µL of 50 % saturated HgCl2 solution to inhibit biological activity (Dickson et al., 2007). Samples were stored at 3–4 ∘C until analysis. Total alkalinity (TA) and dissolved inorganic carbon (DIC) were measured using a VINDTA 3C (Versatile INstrument for the Determination of Total inorganic carbon and titration Alkalinity; Marianda, Kiel, Germany) coupled with a UIC CO2 coulometer detector (UIC Inc., Joliet, USA). Both instruments were calibrated with Dickson Certified Reference Material (Batch 127) (Dickson et al., 2003). DIC concentrations as well as TA and Ωa were calculated with the CO2SYS program as a function of measured pH and pCO2, with dissociation constants of Mehrbach et al. (1973) for carbonic acid as refit by Dickson and Millero (1987) and Dickson (1990) for boric acid.

Wind speeds were obtained from a station of the Instituto Metereológico Nacional (National Meteorological Institute of Costa Rica), located at the nearby Liberia airport. The Módulo de Información Oceanográfica of the University of Costa Rica (www.miocimar.ucr.ac.cr, last access: 27 September 2016) supplied the tidal data. All coral growth values were taken from the literature; linear extension rates from Bahía Culebra were measured by Jiménez and Cortés (2003), whilst coral growth in Panama and Galápagos was measured by Manzello (2010a). For the correlation between coral growth and Ωa, we used the mean Ωa values from Panama and Galápagos previously reported by Manzello (2010b).

We compared our data with values measured during the upwelling season in 2009 (Rixen et al., 2012). In 2009, xCO2 was measured by an underway pCO2 system (SUNDANS) equipped with an infrared gas analyzer (LI-7000), and pH was measured using an Orion ROSS electrode (an Orion Star^™). Correlations between tidal cycles and physicochemical parameters (pH, pCO2, T, wind) during non-upwelling periods were tested via Pearson correlation in Python. Differences in parameters (temperature, pH, pCO2, TA, DIC and Ωa) between all periods (2009, 2012, 2013) were tested with a general linear model (GLM) in the statistical package R. The GLM was evaluated using graphical methods to identify violations of assumptions of homogeneity of variance and normality of residuals. All GLM assumptions were met. Additionally, we developed a simple model to improve our understanding of processes controlling the observed diel trends, as seen in the time series data of pH and pCO2 (Figs. 2, 4). The model simulates combined effects of metabolic processes (photosynthesis, respiration, calcification and dissolution) on the carbonate chemistry. Input parameters for starting the model were the calculated DIC (in 2012: 2037 µmol kg-1 at 07:00 UTC - 6 and 2019 µmol kg-1 at 15:00; in 2013: 1883 µmol kg-1 at 05:00 and 1805 µmol kg-1 at 15:00) and TA (in 2012: 2284 µmol kg-1 at 07:00; in 2013: 2193 µmol kg-1 at 05:00) values, corresponding to the highest and lowest measured pCO2 during the day. Calculation of TA and DIC from the pair pH and pCO2 is prone to errors (Millero, 2007; Cullison Gray et al., 2011); however, the values used as input parameters in the model are in range with those reported from other studies in tropical areas (Manzello, 2010b; Cyronak et al., 2013b). The difference between the two DIC concentrations (ΔDIC) was assumed to be caused by photosynthesis and respiration and the resulting formation and decomposition of particulate organic carbon (POC), as well as calcification and dissolution and the precipitation and dissolution of particulate inorganic carbon (PIC, Eq. 1). ROI describes the ratio between the production of organic carbon (POC) and precipitation of calcium carbonate carbon (PIC) and was used to link ΔPOC to ΔPIC (ROI = POC / PIC) (Eq. 2, 3). The ROI was further constrained by the determined change of TA (ΔTA). Therefore, it was considered that photosynthesis and respiration of one mole of carbon increases and reduces TA by 0.15 units, respectively (Broecker and Peng, 1982). Calcification and dissolution of one mole of carbon decreases and increases TA by two units (Eq. 4). To verify the results from the model, we used the output ΔDIC and ΔTA to calculate new pCO2 and pH values, which were further compared to the measured ones (Fig. 5). The best fit between modeled and measured values was achieved with a respective ROI of -2.6 for 2012 and 1.0 for 2013, whereas the assumption of calcium carbonate dissolution caused the negative sign. ΔDIC=ΔPOC+ΔPICΔPIC=ΔPOCROIΔPOC=ΔDIC/1+1ROIΔTA=ΔPOC⋅0.15-ΔPOCROI⋅2 This was calculated on hourly time steps, separately for 2012 and 2013, using the mean SWT (2012 = 29.61 ± 0.93 ∘C, 2013 = 30.08 ± 0.27 ∘C) and salinity (2012 = 32.5, 2013 = 32.5).

In June 2012, average SWT was 29.61 ± 0.93 (average ± standard deviation) ∘C and ranged from 27.13 to 31.37 ∘C. In May–June 2013, SWT ranged from 29.3 to 30.7 ∘C (average 30.08 ± 0.27 ∘C). During both periods, the salinity was 32.5 ± 0.8. During the study periods, the wind intensified during the afternoons, reaching speeds of up to 8.5 and 6.0 m s-1 in 2012 and 2013, respectively (Fig. 2). Average pH and pCO2 in June 2012 were 7.98 ± 0.04 and 456.38 ± 69.68 µatm, respectively; the corresponding averages for May–June 2013 were 8.02 ± 0.03 and 375.67 ± 24.25 µatm. Since the tidal cycle was not significantly correlated with the variability of pH, pCO2, T or wind (p>0.05) during the periods of observations (Table 2), it was excluded from further discussions. Mean Ωa values were 3.32 ± 0.46 in June 2012 and 3.50 ± 0.49 in May–June 2013 (Table 1).

Measured parameters showed significant differences between study periods (p<0.05). The SWT range differed among years (Table 1); 2013 was the warmest study period, followed by 2012 and 2009. Lowest measured pH was 7.81 in June 2012, 7.84 in April 2009 and 7.95 in May–June 2013. We also compared DIC and TA, in order to estimate to which extent the observed variations of pCO2 were caused by changes in temperature and/or DIC concentrations. Mean DIC values were 2098.71 ± 103.81 µmol kg-1 in April 2009, 1924.65 ± 195.07 µmol kg-1 in June 2012 and 1800.92 ± 142.78 µmol kg-1 in May–June 2013. Similarly, mean TA values were 2328.42 ± 118.45 µmol kg-1 in April 2009, 2204.54 ± 212.18 µmol kg-1 in June 2012 and 2102.66 ± 174.79 µmol kg-1 in May–June 2013. According to average values, April 2009 was the period with the most acidic water and greater CO2 enrichment, followed by June 2012 and May–June 2013 (Table 1). Mean Ωa values were 2.71 ± 0.29 during the upwelling season (April 2009) and 3.41 ± 0.13 during the non-upwelling season (June 2012, May–June 2013), resulting in annual average Ωa of 3.06 ± 0.49 at Bahía Culebra. Time series of pH and pCO2 in June 2012 and May–June 2013 showed a pronounced daily cycle (Fig. 4), which in addition to previously described data will be discussed in the following paragraphs.

The observed differences in pH and pCO2 between 2012 and 2013 suggest that the non-upwelling season exhibits a strong interannual variability (Table 1). In 2012, pH was lower and pCO2 higher than in 2013 (Fig. 2b, c). The June 2012 time series data showed that SWT decreased and pCO2 increased from 300 to 650 µatm in less than a week, after several days of strong afternoon winds (Fig. 2a). Similarly, this increase in pCO2 was accompanied by a drop in pH from 8.04 to 7.83 (Fig. 2a). This suggests that an enhanced wind-driven vertical mixing entrained cooler and CO2-enriched waters from greater water depth into the surface layer. The associated SWT drop from 31.4 to 27.1 ∘C was similar to that observed during the onset of the 2009 upwelling event (26.2 to 23.7 ∘C; Rixen et al., 2012). Nevertheless, the higher SWT during the 2012 non-upwelling season suggests that the entrained water originated from a shallower water depth, compared with the water upwelled in 2009. The pCO2 values with up to 650 µatm reached the same level during both events, which is partially caused by the higher SWT in 2012. However, DIC concentrations in 2012 (1924.65 ± 195.07 µmol kg-1) were lower than those in 2009 (2098.71 ± 103.81 µmol kg-1) but exceeded those in 2013 (1800.92 ± 142.78 µmol kg-1; Table 1). During the 7 days of the cold-water intrusion event in 2012 (10–17 June), the DIC concentrations dropped from 2355.39 µmol kg-1 down to 1715.30 µmol kg-1. This implies that in addition to high SWT, the entrainment of CO2-enriched subsurface water increased the pCO2 not only during the upwelling periods but also during the 2012 non-upwelling season.

Since in 2012 the pCO2 had already increased by 7 June and the SWT decreased 2 days later (10 June), the inflow of CO2-enriched waters seems to have increased the pCO2 already prior to the strengthening of local winds (Fig. 2b). Later, local wind-induced vertical mixing seems to have amplified the impact of the inflowing CO2-enriched water mass on the pCO2 in the surface water by increasing its input into surface layers. Accordingly, the CO2-enriched waters were apparently supplied from a different location than they are during the upwelling season. Since the NECC carries offshore waters towards the Costa Rican shore during the non-upwelling season (Wyrtki, 1965, 1966; Fiedler, 2002), it is assumed that the CO2-enriched subsurface water originated somewhere south of our study area in the open ETP. The absence of such a cold event during the non-upwelling season in 2013 suggests that the occurrence of this kind of event might be an irregular feature (Fig. 2c, d), and the driving forces are still elusive. Nevertheless, these types of events have the potential to affect the metabolic processes in the bay as will be discussed in the following section, which analyzes the daily cycles during the non-upwelling seasons in 2012 and 2013.

In 2012, the pH and the pCO2 values followed a pronounced diurnal cycle with highest pH and lowest pCO2 values during the late afternoon and lowest pH and highest pCO2 values around sunrise in the early morning (Fig. 4a). Such daily cycles are typical for tropical regions and are assumed to be caused by photosynthesis during the day and respiration of organic matter during the night (Shaw et al., 2012; Albright et al., 2013; Cyronak et al., 2013a). The aragonite saturation state as well as the DIC / TA ratio followed this pattern, with higher Ωa and lower DIC / TA ratio values during the day as well as lower Ωa and higher DIC / TA values at night (Fig. 4b). Although the pCO2 cycles in 2013 followed a similar pattern to 2012, pH cycles were less predictable (Fig. 4).

To characterize the relative importance of the processes responsible for the observed changes in pH and pCO2 (photosynthesis, respiration, calcification and dissolution), we used the model described earlier, which is based on the determined DIC concentrations during times when pH and pCO2 revealed their daily minima and maxima, respectively. For example, if photosynthesis of organic matter dominates the transition from early morning maxima of pCO2 to late afternoon minima of pCO2, it should be associated with a decline in DIC. Whether photosynthesis was accompanied with enhanced calcification can be detected by an associated decrease of TA. Since decreasing DIC raises the pH and a decrease in TA lowers the pH, such photosynthetic-enhanced calcification hardly affects the pH and could explain the weak daily cycle observed in 2013. Alternatively, if photosynthesis is accompanied by carbonate dissolution during the day, this would amplify the daily cycle of pH and pCO2 as seen during the cold-water intrusion event in 2012. Likewise, an increased photosynthesis resulting from higher nutrient concentrations (Pennington et al., 2006) could also be causing the observed large amplitude during the event in 2012. However, in our case, the determined TA and DIC concentrations constrain the impact of the formation of organic matter (POC is equivalent to photosynthesis minus respiration) and calcification (PIC is equivalent to calcification minus dissolution) on the carbonate system. This sets the boundaries within which the observed diurnal cycle of pH and pCO2 has to be explained (Fig. 5c, d). In order to reconstruct the diurnal cycle of pH and pCO2 within these boundaries, we assumed a photosynthetic-enhanced calcification during the day, and vice versa dissolution and respiration at night. Thereby, the best fit between pH and pCO2 measured in 2013 and the respective calculated values could be obtained by using a ROI of 1. This approach failed to explain the diurnal cycle of pH and pCO2 as observed during the 2012 cold-water intrusion event (10–17 June). The only solution we found to explain these pronounced diurnal cycles within the given DIC and TA boundaries was to assume that photosynthesis and dissolution prevailed during the day and respiration and calcification occurred at night. The ROI of -2.6 resulted in the best fit between the measured and calculated pH and pCO2 for the 2012 event, whereas the negative sign reflects the contrasting effects of calcification and dissolution on the DIC concentration.

Dissolution taking place during daytime is peculiar but not completely unusual, as it has been reported on tropical sandy bottoms under ambient (Yates and Halley, 2006a, b; Cyronak et al., 2013b) and high-CO2 conditions (Comeau et al., 2015). Similarly, dark calcification is not entirely uncommon and occurs in both sandy bottoms and coral reefs (Yates and Halley, 2006b; Albright et al., 2013). Accordingly, the entrainment of CO2-enriched water from the NECC seems to shift the carbonate chemistry of Bahía Culebra from a system where photosynthesis and calcification are the controlling processes during daylight hours to a system in which daytime is dominated by photosynthesis and dissolution. The net effect, as observed, is an enhanced pCO2 and lower Ωa during periods characterized by the inflow of CO2-enriched waters (Table 1). This has strong ecological implications for local coral reef ecosystems.

Coral reefs in Bahía Culebra were dominated by Pocillopora spp. and Pavona clavus (Jiménez, 2001; Jiménez et al., 2010), whereas Porites lobata is the main reef forming coral in the southern part of the Costa Rican Pacific coast (Cortés and Jiménez, 2003; Glynn et al., 2017). Although the reefs in the north are naturally exposed to periodic high-CO2 conditions during upwelling events (Rixen et al., 2012), as well as during cold-water intrusions in the non-upwelling season, the linear extension rates of Pocillopora spp. and P. clavus exceeded those of the same species in other regions (Fig. 6) (Glynn, 1977; Jiménez and Cortés, 2003; Manzello, 2010a; Rixen et al., 2012). This suggests that local corals are adapted and/or acclimatized to the upwelling of cold and acidic waters.

Aragonite saturation state (Ωa) is known as one of the main variables influencing coral growth and therefore reef distribution around the world (Kleypas et al., 1999). By integrating the data from the present study and values previously reported by Rixen et al. (2012), we estimated that the annual mean Ωa in Bahía Culebra is 3.06. Additionally, earlier studies in the ETP measured Ωa values and coral extension rates from locations that are under the influence of upwelling events (Manzello, 2010a), whilst extension rates from Bahía Culebra were measured by Jiménez and Cortés (2003). The correlation between our estimated Ωa with the available data from Bahía Culebra, Panama and Galápagos indicates that coral extension rates in each of those locations are predictable by their corresponding Ωa values (Fig. 6).

The dependency of coral growth on Ωa and the mean Ωa (2.71) during the upwelling season (Table 1) suggests that upwelling of acidic waters should reduce corals' relatively high annual mean growth rates in Bahía Culebra. The increased Ωa during the non-upwelling season in turn must enhance linear extension and explains corals' high annual mean growth rates. The Ωa values from this study suggest that most favorable conditions for coral growth occur during the non-upwelling season, the period that coincides with development of the rainy season. This implies that during the main growing season the eutrophication and siltation caused by human impacts on river discharges, as well as the development of harmful algal blooms, could also strongly affect the corals' annual mean growth rates (Cortés and Reyes-Bonilla, 2017).

Despite the corals' high annual mean linear extension rates, studies carried out in 1973 showed that the thickness of the reef framework within our study area was with 0.6 to 3 m (mean 1.8 m) among the lowest in the ETP, where Holocene framework accumulation in Pocillopora-dominated reefs could reach up to 9 m (Glynn et al., 1983; Toth et al., 2017). During the last decade, it further decreased (Alvarado et al., 2012), and during the period of our observation the reef frameworks of Pocillopora spp. in Bahía Culebra hardly exceeded a thickness of 0.5 m. This denotes that although Pocillopora spp. and P. clavus are adapted to the entrainment of acidic waters, these reefs are growing in an environment at the limit of reef-building corals' tolerance in terms of temperature, nutrient loads and pH (Manzello et al., 2017). Gaps in coral reef accretion at the ETP are known from the geological record (Toth et al., 2012, 2015, 2017). They have been linked to increased El Niño–Southern Oscillation (ENSO) variability (Toth et al., 2012, 2015) and stronger upwelling conditions (Glynn et al., 1983), favoring dissolution and erosion of reef frameworks while at the same time restricting coral growth.

The y intercept of the regression equation derived from the correlation between linear extension rates and Ωa furthermore implies that linear extension of P. damicornis and P. clavus should approach zero under a carbonate saturation state of Ωa < 2.5 (P. damicornis) and < 2.2 (P. clavus). According to climate predictions, the global Ωa will reach values < 2.0 by the end of this century (IPCC, 2014), and major upwelling systems such as those off California and South America will intensify (Wang et al., 2015). Combined effects of ocean acidification and impacts of stronger upwelling on Ωa in the ETP and on Ωa in Bahía Culebra are difficult to predict. Worldwide, OA is expected to reduce coral reefs' resilience by decreasing calcification and increasing dissolution and bioerosion (Kleypas et al., 1999; Yates and Halley, 2006a; Anthony et al., 2011). Coral reefs from the ETP are affected by chronic and acute disturbances, such as thermal stress and natural ocean acidification resulting from ENSO and upwelling events, respectively (Manzello et al., 2008; Manzello, 2010b). Historically, these reefs have shown a high resilience to both stressors separately, but their coupled interaction can cause coral reefs to be lost within the next decades. The ETP has the lowest Ωa of the tropics, near the threshold values for coral reef distribution; therefore, the reefs from this region may be the most affected by the increasing levels of anthropogenic CO2 and also show the first negative impacts of this human-induced OA (Manzello et al., 2017). This emphasizes the importance of the Paris Agreement and all the global efforts to reduce the CO2 emission into the atmosphere (Figueres et al., 2017).