Using present-day observations to detect when anthropogenic change forces surface ocean carbonate chemistry outside pre-industrial bounds

One of the major challenges to assessing the impact of ocean acidification on marine life is detecting and interpreting long-term change in the context of natural variability. This study addresses this need through a global synthesis of monthly pH and aragonite saturation state (Ωarag) climatologies for 12 open ocean, coastal, and coral reef locations using 3-hourly moored 20 observations of surface seawater partial pressure of CO2 and pH collected together since as early as 2010. Mooring observations suggest open ocean subtropical and subarctic sites experience present-day surface pH and Ωarag conditions outside the bounds of pre-industrial variability throughout most, if not all, of the year. In general, coastal mooring sites experience more natural variability and thus, more overlap with pre-industrial conditions; however, present day Ωarag conditions surpass biologically relevant thresholds associated with ocean acidification impacts on Mytilus californianus (Ωarag<1.8) and Crassostrea gigas (Ωarag 25 <2.0) larvae in the California Current Ecosystem (CCE) and Mya arenaria larvae in the Gulf of Maine (Ωarag<1.6). At the most variable mooring locations in coastal systems of the CCE, subseasonal conditions approached Ωarag=1. Global and regional models and data syntheses of ship-based observations tended to underestimate seasonal variability compared to mooring observations. Efforts such as this to characterize all patterns of pH and Ωarag variability and change at key locations are fundamental to assessing present-day biological impacts of ocean acidification, further improving experimental design to interrogate organism response 30 under real-world conditions, and improving predictive models and vulnerability assessments seeking to quantify the broader impacts of ocean acidification.


Introduction
The global ocean has absorbed ∼ 30 % of anthropogenic carbon dioxide (CO 2 ) emissions released since the beginning of the industrial revolution (Khatiwala et al., 2013;Le Quéré et al., 2015).While ocean uptake of CO 2 has played a role in mitigating the atmospheric-associated impacts of anthro-Published by Copernicus Publications on behalf of the European Geosciences Union.
pogenic CO 2 , it has also resulted in changes to seawater chemistry.Seawater pH has decreased globally by 0.1 since the preindustrial era (Feely et al., 2004;Orr et al., 2005) and is predicted to decrease by another 0.3 by 2100 under Intergovernmental Panel on Climate Change (IPCC) business-asusual emission scenarios (Caldeira and Wickett, 2005;Orr et al., 2005).Ocean acidification also results in reductions in carbonate ion concentrations and the saturation states of calcium carbonate minerals utilized by calcifying marine organisms to make their shells or skeletons.Globally, average surface seawater pH change is currently about −0.002 yr −1 and the saturation state of aragonite, a common form of calcium carbonate mineral utilized by marine organisms, is approximately −0.008 yr −1 (Bates et al., 2014).
These assessments of ocean-wide change inform globalscale predictions of ocean acidification impacts and provide boundaries on carbonate chemistry in designing biological experiments.However, a major challenge to assessing local-to regional-level ecological and economic consequences of ocean acidification is the lack of understanding of how global anthropogenic change manifests relative to natural variability, especially in dynamic coastal regions (Bauer et al., 2013).Coastal systems are sites of large variability, where terrestrial, atmospheric, and marine nutrient and carbon cycles all interact and where ocean acidification conditions can be more extreme than in open ocean environments (Cai et al., 2011;Feely et al, 2010).Important processes that affect ocean acidification in the coastal ocean include upwelling, riverine/estuarine input, air-sea gas exchange, production and respiration, calcification, dissolution, sediment burial, and sea-ice dynamics.Despite the ecological, biological, and economic importance of coastal regions, the magnitude and variability of these key biogeochemical processes are poorly quantified.
Earth system models provide some insights into carbon system variability; however, they do not often capture the full magnitude of variability, especially at seasonal to subseasonal timescales (Pilcher et al., 2015;Sasse et al., 2015).For example, Sasse et al. (2015) estimated that earth system models underpredict the seasonal cycle of seawater partial pressure of CO 2 , i.e., p(CO 2 ), by 30 %.Despite these biases, some studies have made progress by utilizing current estimates of arag variability and change in assessing ecosystem impact and vulnerability to ocean acidification.Using predicted aragonite saturation state ( arag ) change between 2005 and 2050 and an estimate of annual variability from the Community Climate System Model (CCSM) 3.1, Cooley et al. (2009) identified that ocean acidification will exceed natural conditions throughout the global ocean by 2050, especially in low-latitude regions of the Atlantic, Indian, and western Pacific.Applying similar approaches to three other earth system models, Friedrich et al. (2012) also concluded that these regions are most vulnerable, but also found that present-day ocean acidification already exceeds preindustrial variability by a factor of 5 in shallow water tropical Pacific and Atlantic coral reef ecosystems.Both studies pointed out that these global models do not currently resolve coastal processes and therefore lack important sources of natural variability, which bias these results when extrapolating to coastal and coral systems.In a vulnerability assessment of US shellfisheries, this lack of coastal information was addressed by using earth system model output as a baseline for arag conditions but also by adding a term in the final assessment for amplification of ocean acidification in coastal systems that experience eutrophication, upwelling, and river inputs of low arag water (Ekstrom et al., 2015).Although this approach did not resolve all coastal carbonate variability and change, it allowed for some of the first regional-level assessments of ocean acidification impact.Regional models in coastal systems are also showing promise at eddy-resolving spatial scales and monthly to seasonal temporal scales.Applied to the California Current Ecosystem (CCE), regional models predict that surface ocean arag has already moved outside the bounds of preindustrial variability (Hauri et al., 2013).Direct measurements of the full range of variability will help parameterize and evaluate these models.
Defining present-day seawater carbonate conditions can also contribute to studies addressing how marine life responds to a myriad of stressors.Biological impacts could manifest, for example, through exposure to conditions beyond the baseline, preindustrial conditions to which marine life is adapted.In a field experiment within a natural coral reef system, Albright et al. (2016) found that net community calcification increased when seawater chemistry was modified to preindustrial conditions.Preindustrial conditions likely varied among different ecosystems, so designing biological impact studies like this depends on defining the natural range of variability and identifying when and where other organisms may be exposed to conditions outside of this range.Biological impacts may also manifest once seawater carbonate conditions surpass a certain physiological threshold.Many experiments have focused on measuring biological response to undersaturated arag conditions; however, recent studies have identified physiological thresholds for shellfish larvae at arag > 1 (Barton et al., 2012;Gaylord et al., 2011;Hettinger et al., 2013;Salisbury et al., 2008;Waldbusser et al., 2015a, b).Knowledge about when and where these corrosive conditions occur and how the timing of such conditions relate to key life stages is critical to assessing vulnerability to ocean acidification.
Understanding what dominant patterns of variability define the natural range of carbonate chemistry is also critical to future predictions of impact.Whether the variability is dominated by stochastic events (e.g., storms), the seasonal cycle, interannual variability (e.g., El Niño and La Niña events), or decadal climactic oscillations (e.g., Pacific Decadal Oscillation) will control how climate change impacts these natural variations in ocean carbonate chemistry.By resolving these scales of variability, researchers can start to refine biological experiments, ecosystem models, and economic vulnera-bility assessments in the context of full ecosystem variability and change.As evidence emerges that some shelled marine organisms are already being impacted by corrosive seawater chemistry conditions (Barton et al., 2012;Bednaršek et al., 2012Bednaršek et al., , 2014a, b;, b;Reum et al., 2015) and stakeholders are seeking locally relevant solutions (Kelly and Caldwell, 2013;Kelly et al., 2011), these analyses are becoming increasingly important.
High-frequency moored observations can be highly effective in capturing the full range of variability at key locations (Cullison Gray et al., 2011;Harris et al., 2013;Hofmann et al., 2011;Shadwick et al., 2015;Sutton et al., 2014a;Xue et al., 2016).Here we present the first global-scale ocean acidification mooring synthesis of 3-hourly surface ocean pH observations and arag calculated from direct measurements of p(CO 2 ) and pH collected together on 12 open ocean, coastal, and coral reef moorings since as early as 2010.We used these observations to define present-day and preindustrial subseasonal to interannual pH and arag variability at each location, and compared these observations to past modeling estimates of seasonal to interannual variability of arag .This assessment identifies the patterns in which present-day conditions have exceeded preindustrial bounds and biologically relevant thresholds at these mooring locations, and documents when and where marine life currently encounters ocean acidification conditions that may impact growth and survival.

Methods
In 2003, the National Oceanographic and Atmospheric Administration's (NOAA) Pacific Marine Environmental Laboratory (PMEL) began to establish a global moored CO 2 network that presently includes 38 mooring locations (www.pmel.noaa.gov/co2/).With the recent development of autonomous pH sensors, PMEL has been able to expand this network to include pH at 21 of the mooring locations.These moored CO 2 and ocean acidification time series are part of a long-term, sustained effort intended to advance our scientific understanding of the ocean carbon cycle and how it is changing over time.This network leverages other observing system efforts including, but not limited to, the National Data Buoy Center, the Research Moored Array for African-Asian-Australian Monsoon Analysis and Prediction, and the OceanSITES network.This study focuses on 12 of these moorings with paired p(CO 2 ) and pH measurements that allow for estimates of subseasonal to interannual variability of pH and arag (Fig. 1, Table 1).This set of moorings includes sites in each major ocean basin and in a variety of open ocean, coastal, and coral reef environments.The time series used in this study include observations primarily between 2010 and 2015, with the exception of the WHOTS time series, which goes back to 2004."Present-day" observations used here refer to these time series outlined in Table 1.
Each mooring time series summarized in Table 1 included a Moored Autonomous p(CO 2 ) (MAPCO 2 ) system and a Sunburst SAMI-pH sensor deployed on the surface buoy.For a detailed description of the MAPCO 2 system and data processing, refer to Sutton et al. (2014b).In brief, the MAPCO 2 system utilizes an automated equilibrator-based gas collection system to measure surface seawater x(CO 2 ) (the mole fraction of CO 2 in air in equilibrium with surface seawater) every 3 h in addition to sample temperature, pressure, and relative humidity.The x(CO 2 ) measurement is made by a nondispersive infrared gas analyzer (LI-820, LI-COR) calibrated before, during, and after field deployment with reference gases traceable to World Meteorological Organization standards.A Sea-Bird Electronics (SBE) 16plus V2 Sea-CAT was also deployed and integrated with the MAPCO 2 system to collect sea surface temperature (SST) and salinity (SSS) measurements used to calculate p(CO 2 ) consistent with ocean CO 2 standard operating procedures (Dickson, 2007;Weiss, 1974).Overall uncertainty of the MAPCO 2 is < 2 µatm for seawater p(CO 2 ).The SAMI-pH sensor utilizes the spectrophotometric method for measuring seawater pH with a laboratory-based accuracy of ±0.003 and precision of < 0.001 (Seidel et al., 2008).
The SAMI-pH and SBE SeaCAT sensors collected 3hourly measurements during the MAPCO 2 10 min seawater equilibration time.All sensors were factory calibrated between each 1 year deployment.Data quality control for the SAMI-pH data involved utilizing the sensor software to identify and flag outliers and sensor failures such as bad blanks.We also used the relationship between total alkalinity (A T ) and SSS (Cullison Gray et al., 2011;Fassbender, 2014;Lee et al., 2006;Xue et al., 2016) to calculate pH from p(CO 2 ) and A T (using the CO2SYS program) in order to identify sensor drift within a deployment or offsets between deployments, which only occurred in 8 % of the pH data sets.All pH data are on the total pH scale (Dickson et al., 2007).Data quality control for the SBE SeaCAT was limited to flagging and eliminating outliers.All measurements are archived at the relevant data centers: the Carbon Dioxide Information Analysis Center (http://cdiac.ornl.gov/oceans/Moorings/) and the National Centers for Environmental Information (https://www.ncei.noaa.gov/).
In all calculations of arag we used the MATLAB version (v1.1) of the CO2SYS program (Lewis and Wallace, 1998;van Heuven et al., 2011) with the carbonic acid dissociation constants of Lueker et al. (2000), sulfate dissociation constants of Dickson (1990), and borate-to-salinity ratio of Lee et al. (2010) according to recommended best practices (Dickson, 2007;Orr et al., 2015).We also used average surface ocean phosphate and silicate concentrations from the World Ocean Atlas 2009 for each mooring location (Garcia et al., 2010).We used two pairs of carbonate system parameters to calculate arag : (1) p(CO 2 ) and pH observations and (2) p(CO 2 ) observations and A T estimated from SSS (Cullison Gray et al., 2011 for CCE2 and Chá bȃ;Fassbender, 2014 Lee et al., 2006 for the remaining sites).We then averaged the two arag data sets for the final present-day arag values presented here.The only exception to the arag calculations was the WHOTS mooring time series where we have good pH data over multiple years but only during the months of June to November.For this site we used only p(CO 2 ) and A T estimated according to Lee et al. (2006) to calculate arag .Similarly, all pH data presented are direct autonomous measurements of pH except in the case of the WHOTS mooring where pH was calculated from p(CO 2 ) observations and estimated A T using CO2SYS.
We used this approach of averaging calculated arag from p(CO 2 ) and pH observations ( arag: pCO 2 -pH ) and calculated arag from p(CO 2 ) observations and estimated A T ( arag: pCO 2 -AT ) to minimize the following errors: (1) the covariance of p(CO 2 ) and pH can lead to small errors in predicting arag , which can enhance arag variability (Cullison Gray et al., 2011;Millero, 2007), and (2) the A T proxies only account for dilution and evaporation processes, which can result in underestimation of A T variability and, when combined with other carbon parameters, can reduce surface arag variability.We tested the A T proxies by comparing A T measured from discrete bottle samples collected at the surface at each mooring site to A T estimated from the proxies listed above, which were developed using A T and SSS measurements throughout the mixed layer.A T estimated from 3-hourly moored SSS was generally within stated errors (3 to 20 µmol kg −1 ) of discrete A T (e.g., Table 2).However, at a few sites where discrete A T was measured at the highest frequency (weekly to seasonally; sample size [n] > 100), discrete A T was more variable than A T estimated from the A T -SSS proxies (Table 2).At the La Parguera (coral reef) and WHOTS (open ocean) mooring sites, variability of discrete A T (as measured by 1 standard deviation (SD) of the mean) was greater than estimated A T by 21 and 52 %, respectively.
In addition to the discrepancies in discrete and proxy A T data sets, the SD of both arag calculated from discrete measurements ( arag: discrete ) and arag: pCO 2 -pH were greater than arag calculated from the p(CO 2 )-A T and pH-A T pairs ( arag: pCO 2 -AT and arag: pH-AT ; Table 2).One limitation of the WHOTS data set is that high-frequency, moored pH observations only exist during the months of June to November; thus, the mean and SD of the arag: pCO 2 -pH and arag: pH-AT data sets may be biased.Therefore, we used La Parguera as a guide and assumed that actual variability (i.e., from high-quality discrete bot-  4), and the size of symbols represents arag variability as measured by 1 standard deviation (SD) of the annual mean.
tle samples) of arag ( arag: discrete SD = 0.16) lies between arag: pCO 2 -pH (SD = 0.19) and arag calculated from estimated A T ( arag: pCO 2 -AT SD = 0.12; arag: pH-AT SD of 0.13; Table 2); we used the approach of averaging mooring ocean acidification data sets to reflect this.Since we are confident in the uncertainty of the moored autonomous p(CO 2 ) measurements (Sutton et al., 2014b), for arag calculated from estimated A T we selected arag: pCO 2 -AT to average with arag: pCO 2 -pH .While there were not enough discrete measurements to do this same comparison at the coastal sites, we found that the arag: pCO 2 -pH , arag: pCO 2 -AT , and arag: pH-AT data sets were not significantly different (p < 0.05) at the coastal sites, likely due to high natural variability in the coastal carbon system.This suggests that averaging arag: pCO 2 -pH and arag: pCO 2 -AT vs. using one arag data set will not have a significant impact on results.
The ideal method for calculating arag may differ across sites dependent on access to and analysis of high-quality ship-based measurements.However, averaging arag: pCO 2 -pH and arag: pCO 2 -AT provides a conservative estimate of monthly arag conditions allowing for a broad-scale comparison with consistent methodology across 12 buoys in different marine ecosystems.Continued high-frequency (ideally, weekly to monthly where feasible) discrete sampling, development and analysis of regional A T proxies, and devel-opment of autonomous sensors capable of measuring carbonate parameters other than p(CO 2 ) and pH will all help to further refine these methods for calculating arag from moored autonomous observations.
To estimate preindustrial pH and arag , we used p(CO 2 ) and A T with the following adjustments to present-day observed p(CO 2 ) and SST: (1) a decrease of 105 ppm in surface ocean p(CO 2 ) assuming the delta between atmospheric and surface seawater p(CO 2 ) has remained constant from the preindustrial era to a reference year of 2010 (from Pieter Tans, NOAA/ESRL, www.esrl.noaa.gov/gmd/ccgg/trends/),and (2) changes in SST that vary regionally from 0.5 to 1.5 • C consistent with the IPCC Fifth Assessment Report (Rhein et al., 2013).Although magnitudes of atmospheric CO 2 uptake can also vary regionally (Bauer et al., 2013), for the purposes of this synthesis we assumed that the change in air-sea CO 2 differences was consistent globally but that SST changes varied regionally as presented by Rhein et al. (2013).We also assumed small changes in SSS, A T , phosphate, or silicate since the preindustrial era would not have a significant impact on calculated pH and arag .We applied these changes to monthly mean and range of variability of presentday observations, and used CO2SYS to calculate preindustrial pH and arag .The range of monthly variability in preindustrial and present-day conditions was defined by descrip- tive statistics used in box and whisker plots excluding outliers: where Q1 is the 25th percentile (or first quartile), Q3 is the 75th percentile (or third quartile), and IQR is the interquartile range (Q3-Q1) of pH and arag values.These limits equate to approximately ±2.7 SD of the mean for a normally distributed data set.
The variables used to estimate total uncertainty of presentday calculated arag are shown in Table 3.This uncertainty estimate included a preliminary assessment of in situ validation similar to p(CO 2 ) validation by Sutton et al. (2014b) to estimate pH error in the field.The average difference between SAMI-pH measurements and pH calculated from discrete measurements of dissolved inorganic carbon and A T was ±0.018, which was larger than laboratory-based assessments of pH measurement error (Seidel et al., 2008).While this estimate included error caused by slight mismatches in space (< 1 km) and time (< 1.5 h) between the moored and discrete measurements, we used it here to develop a conservative estimate of total estimated uncertainty for calculated arag: pCO 2 -pH from moored observations, which is 0.37 for seawater at SST of 25 • C, SSS of 35, p(CO 2 ) of 370 µatm, and a pH of 8.1 (Table 3).This estimate meets the target relative uncertainty for arag of 10 % needed to identify relative spatial patterns and short-term variation in ocean acidifica-  2015).c Combined effect of thermodynamic constants (Mucci et al., 1983).d Range of error from AT proxies described in methods.
tion (Newton et al., 2015).A more detailed assessment of pH sensor error is planned as more discrete and autonomous pH data become available.When the MAPCO 2 system is paired with estimated A T , arag: pCO 2 -AT also meets this uncertainty target (Table 3); however, natural variability of surface ocean arag: pCO 2 -AT may be underestimated as discussed previously.This estimated uncertainty is likely higher during the months of May through July at the Gulf of Maine mooring site, where SSS < 30 25 % of the time due to freshwater inputs; at low salinity, the A T -SSS relationship likely deteriorates.Total relative uncertainty of arag for all mooring data sets, averaged from the arag: pCO 2 -pH and arag: pCO 2 -AT data sets, likely falls between 5 and 10 % (at the seawater conditions described in Table 3) and within the target relative uncertainty for describing short-term ocean acidification variability.From this point on, arag reported here is arag: pCO 2 -AT for the WHOTS mooring time series and averaged arag: pCO 2 -pH and arag: pCO 2 -AT for the other mooring time series.

Observations of variability and change
Direct observations of p(CO 2 ) and pH revealed present-day conditions of surface ocean carbonate chemistry in 12 different oceanic and coastal systems.The open ocean moor-ing time series sites are located in subtropical oligotrophic regions (WHOTS, Stratus), biologically productive subtropical regions that experience seasonal monsoons (BOBOA) and tropical cyclones (KEO), and subarctic regions with pronounced seasonality of physical and biological conditions (Papa, Iceland).Annual mean arag at these sites ranged from 1.83 to 3.56; annual mean pH ranged from 7.99 to 8.12 (Figs.1-4, Table 4).High biological productivity is a feature at each of the four coastal mooring time series sites on the continental shelves of the US east (Gulf of Maine, Gray's Reef) and west (CCE2, Chá bȃ) coasts.Summer upwelling is another important driver of conditions at Chá bȃ, located mid-shelf at 100 m bottom depth offshore of La Push, Washington (Alford et al., 2012).While upwelling can also impact the CCE2 site located mid-shelf at 800 m bottom depth farther south in the CCE, this subregion in particular has shown sensitivity to climatic drivers, such as the El Niño-Southern Oscillation (ENSO; Nam et al., 2011).Seasonal temperature and freshwater inputs impact natural variability at the two coastal moorings in the Atlantic with the Gulf of Maine site located 10 km from shore at 65 m bottom depth (Salisbury et al., 2009) and Gray's Reef 70 km from shore at 20 m bottom depth (Xue et al., 2016).The Chuuk and La Parguera sites are located in coral reef ecosystems within a semi-closed atoll lagoon in Micronesia at 23 m bottom depth and a patch reef in the Caribbean Sea southwest of Puerto Rico at 5 m bottom depth, respectively.Despite the more variable coastal and coral reef conditions, the range of annual mean arag at these sites was 1.97 to 3.37, less than the range observed at the more diverse set of ocean regimes represented by the open ocean sites (Figs. 1, 5-7, Table 4).However, the range of annual mean pH was approximately the same as the open ocean sites from 8.01 to 8.15 (Figs.5-7).
Of the open ocean time series, the moorings located in subtropical oligotrophic regions, WHOTS and Stratus, experienced lower seasonal to subseasonal variability in surface pH and arag (Figs.2-4).Consistent trade winds and shallow mixed layer depth throughout the year along with the lack of deep winter convection likely contribute to this relatively low open ocean variability.Temporal variability was higher at the other four open ocean mooring locations, which was likely driven by (1) more prevalent seasonal changes in SST (on average 2 times more variable than WHOTS and Stratus) and productivity, and (2) stochastic events such as storms and typhoons.In general, the range of variability tended to be consistent throughout the annual cycle at each of the open ocean sites with exceptions of increased variability at the Iceland location in late summer and early fall and at Papa during winter (Fig. 4).Present-day arag values were mostly > 3 yearround at the subtropical open ocean sites except at Stratus, where arag values mostly fell between 2.5 and 3.0 (Figs. 2  and 3).Surface arag conditions were further reduced at Papa and Iceland, the subarctic sites, which range from 1.5 to 2.5 (Fig. 4).Present-day pH observations were > 8 throughout the average year at these mooring sites except at Stratus, where pH fell below 8 half the year from December through May (Fig. 2).Moored observations were consistent with seasonal means from ship-based time series observations at the WHOTS and Iceland sites (Bates et al., 2014).
The seasonal cycle of surface ocean arag and pH were not always consistent with one another.Seawater arag is largely determined by variations in the concentration of the carbonate ion (CO 2− 3 ); pH is influenced by gas exchange of CO 2 , physical conditions, and biological activity.Observations of surface ocean pH were consistent with a seasonal thermodynamic response, i.e., pH decrease (increase) with SST increase (decrease), at the four subtropical open ocean sites and at the Papa mooring (Figs.2-4).However, this strong relationship was not consistent at the subarctic Iceland site.At this site, pH and SST were positively correlated (Fig. 4), suggesting that the seasonality of surface ocean pH was dominated by biological activity in the summer and/or winter mixing of upwelled deep water low in temperature and pH (Takahashi et al., 1993;Chen et al., 2007).At all open ocean sites, arag was highest during summer months, which led to the timing of low arag and low pH conditions to be anticorrelated over the annual cycle at all open ocean sites except Iceland (Figs. 2-4).This pattern at the Iceland mooring was consistent with seasonality of surface arag and pH derived from quarterly ship-based observations at the same site (Bates et al., 2014;Olafsson et al., 2009).
Comparisons to preindustrial bounds of variability also revealed differences between open ocean sites.All open ocean sites experienced surface arag outside the bounds of preindustrial variability year-round with the exception of BOBOA and Iceland.Present-day surface arag conditions still partially overlapped with preindustrial conditions at BOBOA during the monsoon season from June through August (Fig. 3), at Iceland during the summer to fall transition in August and September (Fig. 4), and slightly at KEO during November and December (Fig. 3).However, presentday surface pH observations fall completely outside preindustrial pH conditions at all open ocean sites year-round, except at BOBOA where there was a slight overlap of 4 % in August (Figs.2-4).
The coastal mooring sites experienced higher subseasonal to seasonal variability in surface pH and arag compared to the open ocean sites (Figs. 5 and 6).Each coastal time series exhibited clear seasonal patterns with annual amplitudes of arag ranging from 0.66 to 1.32 (Table 4).Gray's Reef and Chá bȃ experienced the highest subseasonal to seasonal variability in surface pH and arag , likely driven by upwelling/relaxation/downwelling dynamics that can change rapidly at Chá bȃ (Alford et al., 2012;Hickey and Banas, 2003) and high productivity and freshwater inputs in the spring and fall at Gray's Reef (Salisbury et al., 2009;Xue et al., 2016).We also observed the lowest pH values (7.8) and surface arag values close to undersaturation ( arag < 1) primarily in the winter at Chá bȃ and in the spring at CCE2 (Fig. 5).These observations of near-undersaturated Table 4. Descriptive statistics of arag : annual mean, annual amplitude, and 1 SD of annual anomalies from the CO 2 and pH mooring observations, from a global data synthesis of ship-based observations (Takahashi et al., 2014), and a biogeochemical model (Friedrich et al., 2012) conditions are consistent with other observations and models within the northern CCE where the Chá bȃ mooring resides (Harris et al., 2013;Hauri et al., 2013) and may indicate respiration in the absence of photosynthetic uptake typical of winter/non-bloom periods.
Unlike the subtropical open ocean mooring sites, seasonality of surface ocean pH at these coastal sites showed strong influence of factors other than SST and were not always correlated with arag values.At the moorings in the CCE, these parameters (i.e., SST, pH, and arag ) generally followed similar seasonal patterns, suggesting factors other than seasonal thermodynamic response influenced surface ocean pH (Fig. 5).However, surface ocean pH and arag did not always follow the same seasonal pattern at the Gray's Reef and Gulf of Maine mooring sites (Fig. 6).While SST likely influenced some of the seasonal variation in pH at these sites, biological activity and freshwater input also influenced seasonality of the carbonate system at these US east coast locations (Salisbury et al., 2009;Xue et al., 2016).
In general, the coastal sites experienced arag outside of the preindustrial range mainly during winter.One exception was Chá bȃ, the coastal site with the highest subseasonal variability (Fig. 5).This high subseasonal variability during spring through fall caused high month-to-month variability in the overlap with preindustrial conditions, suggesting this system may be on the threshold of a shift outside preindustrial conditions during this time of the year.Observations of pH at Chá bȃ followed this same pattern.In general, presentday observations of pH fell outside preindustrial conditions more so than arag at all coastal sites (Figs. 5 and 6).
Finally, similar to the coastal moorings, the coral reef mooring sites also experienced subseasonal to seasonal variability but not as large as within the subtropical coastal systems (Fig. 7).Mean annual arag at the Caribbean (La Parguera) and Pacific (Chuuk) moorings was 3.62 and 3.42, respectively, while mean pH was 8.02 and 8.01, respectively (Fig. 7, Table 4).With the exception of low arag outliers at Chuuk, most arag conditions were > 3 throughout the year at both sites, and surface pH observations were > 7.9.The seasonal cycle of pH and arag was more pronounced at La Parguera with relatively consistent monthly range in variability, but the Chuuk site experienced greater variability December  1.
through April, likely driven by local mixing during the trade winds season.Even with small seasonal fluctuations in tropical ocean temperature, both coral mooring sites did show patterns of pH and arag seasonality associated with SST, with lower pH and arag values coinciding with slightly warmer summer months and higher pH and arag values during winter (Fig. 7).Present-day variability at these sites did not cause extensive overlap with preindustrial conditions.Present-day surface pH observations fell completely outside preindustrial conditions year-round at both coral reef sites (Fig. 7).Present-day arag conditions at La Parguera were largely outside of preindustrial bounds year-round, while this mainly occurred during the season of lowest variability from May to November at Chuuk (Fig. 7).
The results from these 12 mooring time series highlight the different patterns of variability of surface ocean arag and pH in both space and time.Figure 8 compares the relative influence of subseasonal, seasonal, and interannual variability at the mooring locations.Since the mooring observations were well distributed throughout the year, we are confident in the subseasonal and seasonal estimates of variability.How-ever, considering that most of the time series were only 2 to 5 years long, we expect to refine the estimates of interannual variability as we obtain more observations over the coming years.For example, ENSO is a driver of ocean conditions, including biogeochemistry, at CCE2 (Nam et al., 2011).While there were weak El Niño-like conditions that developed in the tropical Pacific in 2014 (McPhaden, 2015), there were no major La Niña or El Niño anomalies during the CCE2 time series used in this analysis (March 2012(March -2015)).Hence, the estimate of interannual variability presented here is likely an underestimate of the true interannual signal at this location.In addition, this was a period of anomalously rapid warming in the Gulf of Maine, which may have caused arag to trend higher due the reduced solubility of arag in warmer waters (Mills et al., 2013;Pershing et al., 2015).Potential variations in warming trends over time would also impact interannual variability of arag observations in the Gulf of Maine as the time series continues.
The coastal sites generally experienced higher subseasonal to interannual arag variability than the open ocean and coral reef sites.Relative to other patterns of variability, interanwww.biogeosciences.net/13/5065/2016/Biogeosciences, 13, 5065-5083, 2016 nual arag variability tended to be low at all sites except for at Chá bȃ, Gray's Reef, and CCE2 (Fig. 8a).The other sites tended to be equally influenced by subseasonal and seasonal variability with the exception of the Iceland mooring site, which was controlled more by seasonal variability over the annual cycle (Fig. 8a); however, subseasonal variability played a large role in August through October (Fig. 4).For pH, most mooring sites exhibited similar patterns of variability with low interannual variability and approximately equal influence from seasonal and subseasonal variability (Fig. 8b).Similar to arag , Chá bȃ, Gray's Reef, and CCE2 were the clear outliers with the highest values of interannual pH variability.

Biologically relevant arag thresholds
Research on response of shellfish larvae living in nearshore environments of the CCE and Gulf of Maine to changes in carbonate chemistry allowed us to investigate when observations at the Chá bȃ, CCE2, and Gulf of Maine moorings exceeded biological thresholds.Crassostrea gigas, the Pacific oyster whose larvae are raised in hatcheries in coastal Washington and Oregon, has shown sublethal impacts on larval development, such as shell development and growth, when exposed to levels of arag < 2.0 (Barton et al., 2012) and acute impacts when arag < 1.5 (Waldbusser et al., 2015a, b).Other studies suggest that chronic exposure thresholds for the larvae of Ostrea lurida, the Olympia oyster, and Mytilus californianus, the California mussel, occur at arag < 1.4 (Hettinger et al., 2013) and arag < 1.8 (Gaylord et al., 2011), respectively.All of these shellfish larvae, whether naturally occurring or hatchery raised, are found in coastal environments in the region of the Chá bȃ mooring and M. californianus also exist farther south in the nearshore region of the CCE2 mooring.In addition, larvae of Mya arenaria, the soft-shell clam commercially harvested on tidal mudflats of the western Gulf of Maine, has shown a lack of shell formation and growth in laboratory experiments at arag < 1.6 (Salisbury et al., 2008).Monthly climatology of arag developed from the mooring observations at Chá bȃ suggest that present-day arag conditions reached chronic exposure levels for C. gigas larvae ( arag < 2.0) over 50 % of the time from November to March, with nearly the entire months of December through March at arag values less than 2.0 (Fig. 9b).These presentday conditions prevailed over more of the year compared to preindustrial times, when the most extensive chronic exposure occurred only up to 64 % during March (Fig. 9a).Conditions that cause acute responses in C. gigas larvae ( arag < 1.5) were minimal year-round at Chá bȃ except for March, when these conditions persisted in the present day during 37 % of the month (Fig. 9b) and only 14 % of the month during preindustrial times (Fig. 9a).A similar sea- sonal pattern also existed for O. lurida larvae ( arag < 1.4), when chronic exposure levels in March exceeded 27 % during the present (Fig. 9b) compared to only 11% during preindustrial times (Fig. 9a).For M. californianus larvae, presentday chronic exposure levels ( arag < 1.8) prevailed over 40 % of the time in January through March at Chá bȃ, while there was less chronic exposure at CCE2, at 11 to 38 % of time in March through July (Fig. 9b).In both cases, present-day exceedance of these thresholds prevailed over fewer months and at a smaller percentage of the time during those months (Fig. 9a).For M. arenaria, present-day arag conditions exceeded chronic exposure levels at the Gulf of Maine mooring between 11 and 31 % of the time during December through April, with peak exposure levels in February and March (Fig. 9b).In contrast to the CCE, which experienced corrosive arag conditions before ocean acidification, Gulf of Maine surface water conditions did not exceed biological thresholds for M. arenaria at any point during the year in preindustrial times (Fig. 9a).
These observations suggest that present-day coastal arag conditions exceeded thresholds for sublethal effects on shellfish larvae in the Gulf of Maine and during both presentday and preindustrial times at Chá bȃ and CCE2.However, present-day coastal conditions surpass these thresholds more often than preindustrial times (Fig. 9).In some cases, un-favorable surface ocean arag conditions overlap with the spawning season.Coastal conditions of arag < 1.4 at Chá bȃ do not currently occur during the May to August O. lurida larvae spawning season.M. californianus tends to spawn year-round, and while natural populations of C. gigas do exist in Washington coastal waters and tend to spawn in the late summer, hatcheries raise C. gigas larvae year-round.Mooring observations suggest that present-day chronic exposure effects on M. californianus larvae may be more common in the winter in the northern CCE and in the spring in the southern CCE (Fig. 9).The summer spawning season of natural populations of C. gigas avoids chronic and acute exposure levels during winter months; however, hatcheries may encounter these conditions if raising larvae during this time.In the Gulf of Maine, M. arenaria spawns when seawater temperatures reach 10 • C, which during the moored time series occurred in May through November (Fig. 6).According to the Gulf of Maine mooring observations through 2013, corrosive conditions of arag < 1.6 did not occur during this spawning season (Fig. 9b).However, maximum SST observations in April of 9.7 • C were at the verge of this spawning threshold, and rapid warming in the Gulf of Maine of 0.23 • C yr −1 since 2004 suggests that SST as of April 2015 may have exceeded 10 • C at the mooring site (Mills et al., 2013;Pershing et al., 2015).If this warming causes www.biogeosciences.net/13/5065/2016/Biogeosciences, 13, 5065-5083, 2016 M. arenaria to begin spawning in April, larvae may become exposed to arag conditions that limit shell formation and growth (Fig. 9b).While these observations on the continental shelf were offshore from the inshore habitats where natural populations of shellfish and oyster hatcheries exist, these results provide valuable information on endmember coastal conditions that affect the nearshore regions.These monthly climatologies suggest surface water conditions corrosive to shellfish larvae presently exist year-round in the CCE (primarily during winter/spring) and during winter/spring in the Gulf of Maine.For shellfish hatcheries that utilize real-time coastal ocean acidification data and monitor conditions within their facilities, managing the impacts of these corrosive conditions on larvae may be possible.These climatologies may also inform the development of experiments testing the vulnerability of shelled organisms in other coastal regions.For example, target species may include ecologically or economically important species that undergo critical life stages when low arag conditions persist during spring in the region around Gray's Reef (Fig. 6).However, the coastal mooring climatologies also illustrate that low arag and low pH conditions do not always coincide in the natural environment, and experiments testing how arag , pH, and other stressors independently affect marine organisms are necessary for understanding ocean acidification impacts under the diversity of present-day conditions (Breitburg et al., 2015).

Comparison to models and ship-based data syntheses
Since high-frequency autonomous ocean acidification time series are relatively new, much of our current knowledge about ocean carbonate variability comes from ship-based observations.Due to the limitations of ship-based oceanography, these observations can have a seasonal measurement bias leading to errors in seasonal climatology estimates and only capture opportunistic stochastic events, which has resulted in limited knowledge about the influence of subseasonal processes on ocean carbonate variability.In general, we found fairly good agreement between annual mean mooring arag observations and annual mean ship-based data syntheses, which primarily used repeat hydrographic cruise data from the Global Data Analysis Project (Key et al., 2004).Both the Jiang et al. (2015) and Takahashi et al. (2014) data syntheses overestimated surface ocean arag at the Stratus mooring in the South Pacific by 0.31 (Fig. 1; Table 4).Undersampling likely contributed to this discrepancy.Moored observations revealed the lowest arag conditions during August through October; however, ship-based observations were lacking in this region of the Southern Hemisphere dur-  4).The shipbased data syntheses also slightly overestimated annual mean arag at WHOTS, KEO, BOBOA, Iceland, and Chá bȃ; however, these overestimations were roughly within the change in arag expected between the time of the mooring observations (typically 2010-2015) and the ship-based observations (adjusted to a reference year of 2000 by Jiang et al., 2015and 2005by Takahashi et al., 2014).Assuming a global average rate of change of surface ocean arag of −0.008 yr −1 (Bates et al., 2014), the change over this 5-15 year period would be 0.04-0.12.
Of the 10 mooring locations with observations presented in the Takahashi et al. (2014) data synthesis, seasonal variability was overestimated by the ship-based observations at all open ocean sites except Iceland, and underestimated at Iceland and the coastal and coral reef sites (Table 4).These differences could be driven by sparse ship-based data in space and time used to estimate climatological seasonal vari-ability in the Takahashi et al. (2014) synthesis.This analysis demonstrates that in addition to new information about subseasonal variability that is not captured by ship-based observations, moored observations can also be used to improve ship-based data synthesis estimates of seasonal to annual arag conditions in undersampled regions such as the Southern Hemisphere, Iceland Sea, and coastal systems.
Overall, earth system models tend to underestimate natural variability of the carbonate system.The series of earth system models used by Friedrich et al. (2012) underestimated both seasonal and interannual variability of surface arag at all mooring locations except for WHOTS and Stratus, which were the sites with the lowest variability (Table 4).These underestimations are expected at the coastal and coral sites since the models do not capture small-scale biogeochemical processes occurring in these environments.When Friedrich et al. (2012) extrapolated an average annual arag amplitude of ∼ 0.1 in subtropical oligotrophic open ocean regions to coral locations, they concluded that present-day coral conditions fell 5 times outside the preindustrial range of variability.However, we found that actual seasonal variability was 2 to 3 times higher than 0.1 at the Chuuk and La Parguera mooring locations (Table 4), and present-day arag conditions were only 1 to 2 times below the preindustrial range of variability (Fig. 7).On the other hand, we found CCSM3-  2009) to be conservative in some regions.They found that by 2050 all regions will experience surface arag conditions outside preindustrial bounds of variability with emphasis in low-latitude regions.Our present-day mooring observations suggest that not only has the shift outside of preindustrial conditions already occurred year-round at the low-latitude coral reef sites, but also at subtropical and subarctic open ocean sites.
Some state-of-the-art earth system models have improved the characterization of background natural variability in the open ocean.A recent study found that global mean surface ocean pH conditions ( arag not assessed) moved outside preindustrial bounds of variability by 2008 (Mora et al, 2013), which is more consistent with the open ocean moored observations (Figs.2-4) compared to the CCSM3-based estimates (Cooley et al., 2009).Newer earth systems models may still underestimate the full magnitude of variability; however, they can illustrate the relative variability signal between different open ocean regions (Rodgers et al., 2015).Similar to the moored observations, Rodgers et al. (2015) showed that arag variability is higher in the North Pacific and North Atlantic regions (KEO, Papa, Iceland) compared to subtropical and tropical regions (i.e., WHOTS, Stratus, BOBOA).Unlike global earth system models, some regional models are able to resolve small-scale coastal processes and may provide better estimates of natural variability in these dynamic systems.We found the highest levels of natural variability at the Chá bȃ mooring location with an annual range of surface ocean arag of 3.3, from arag values of 1.06 to 4.36 (Fig. 5).Estimates of this range for the northern CCE produced by a regional model were only 0.2, which led to the conclusion that surface ocean arag conditions in 2005 were already outside the bounds of preindustrial conditions (Hauri et al., 2013).Observations at Chá bȃ from 2010 to 2014 suggest conditions only fell partially outside preindustrial variability, primarily during the lower variability season from October to February (Fig. 5).

Conclusions
Direct, high-resolution observations of seawater p(CO 2 ) and pH reveal that marine life is currently exposed to surface ocean pH and arag values outside the envelope of preindustrial variability that they have adapted to at all 12 study locations.Marine life at several study locations is also exposed to conditions exceeding thresholds that may impact the growth and survival of shellfish and conditions approaching undersaturation ( arag < 1).These ocean acidification im- pacts are occurring at the same time that marine organisms are also experiencing different patterns of temporal variability and other anthropogenic stressors, which can be unique to distinct locations and seasons.This high-resolution mooring work provides a new perspective on variability, since earth system models and ship-based observations generally underestimate the temporal variability of surface ocean arag conditions, especially in coastal regions.These results highlight the need to further interrogate these biases, which are often the basis for predictions of future ocean acidification impact.In most cases, such as the WHOTS mooring time series, ocean carbonate observations are also paired with additional autonomous physical and biogeochemical measurements at the surface and at depth, as well as long-term ship-based time series measurements, which are not as temporally resolved as the moored measurements but often include biogeochemical parameters that cannot be measured autonomously.Further synthesis of these data sets from multiple platforms will contribute to improving understanding of the biogeochemical processes controlling carbonate chemistry at these time se- ries locations and to developing parameterizations for global and regional models.Here we focused on assessing ocean carbonate variability and change as a step in advancing these efforts; however, future research at these and other ocean acidification mooring sites should also include assessments of additional biogeochemical parameters, such as dissolved oxygen and optical properties, as these new observational data sets become available.
Sustained, autonomous observations resolving subseasonal conditions are likely to match timescales relevant to biological processes in the natural environment such as energy availability, biological threshold exceedance, seasonal spawning, and recruitment.This characterization of temporal variability of ocean carbonate is one of the major challenges to understanding how anthropogenic change will impact marine life.Impacts to marine life could manifest through one or a combination of the following environmental stressors: the slow, steady change over time as pH and arag conditions respond to gradual ocean uptake of anthropogenic CO 2 , the point in time when pH and arag conditions leave the preindustrial envelope of variability to www.biogeosciences.net/13/5065/2016/Biogeosciences, 13, 5065-5083, 2016 which organisms have adapted, when average or seasonal pH and arag conditions pass a certain threshold, or when episodic corrosive conditions surpass a tipping point in terms of frequency and duration.Our mooring time series exhibit periods of time when surface ocean pH and arag conditions fall outside preindustrial bounds of variability along with surpassing biologically relevant thresholds, but also time periods where none or only one of these stressors is present.
A broad understanding of how this myriad of environmental stressors impact marine life will require a range of approaches including continued and expanded biogeochemical observing, research interrogating the fundamental processes underlying organism response to ocean acidification under laboratory and field conditions, experiments designed to address how different patterns of variability and change impact organisms, and paired chemical and biological observations in the field to assess potential present-day impacts.
Characterizing natural variability and biological impact of ocean acidification conditions at key locations will also be fundamental to improving vulnerability assessments seeking to quantify the economic impact of ocean acidification at local to global scales.

Figure 1 .
Figure 1.Locations of moored p(CO 2 ) and pH observations.Base map is adapted from Jiang et al. (2015) and shows annual climatological distribution of surface arag throughout the global oceans.Added to this base map are moorings from this study where symbol color denotes annual mean arag (values also listed in Table4), and the size of symbols represents arag variability as measured by 1 standard deviation (SD) of the annual mean.

Figure 2 .
Figure 2. Box and whisker plots of present-day monthly surface seawater arag (top) and pH (bottom) and monthly mean SST (orange lines) at the open ocean mooring locations in subtropical oligotrophic regions (WHOTS, Stratus).Boxes represent data between Q1 and Q3, with the line between representing Q2 (i.e., the median).Whiskers represent 1.5 IQR, or ∼ 2.7 SD; Eqs. 1 and 2), of the upper and lower quartiles with data outside that range shown as outliers (open circles).Outliers here represent natural deviations in ocean chemistry, not measurement outliers, which were removed in the data quality control process.Estimated monthly preindustrial arag and pH variability (1.5 IQR or ∼ 2.7 SD) is shown in gray (top) and blue (bottom) shaded areas, respectively.Shaded portions of the pie charts indicate the percent of present-day arag and pH values falling outside the bounds of preindustrial variability for each month.For mooring location see Fig. 1 andTable 1.

Figure 3 .
Figure 3. Box and whisker plots of present-day monthly surface seawater arag (top) and pH (bottom) and monthly mean SST at the open ocean mooring locations in biologically productive subtropical regions that experience seasonal monsoons (BOBOA) and tropical cyclones (KEO).See detailed description of figure components in Fig. 2 caption.

Figure 4 .
Figure 4. Box and whisker plots of present-day monthly surface seawater arag (top) and pH (bottom) and monthly mean SST at the open ocean mooring locations in subarctic regions with pronounced seasonality of physical and biological conditions (Papa, Iceland).See detailed description of figure components in Fig. 2 caption.

Figure 5 .
Figure 5. Box and whisker plots of present-day monthly surface seawater arag (top) and pH (bottom) and monthly mean SST at the coastal mooring locations on the continental shelves of the US west coast (CCE2, Chá bȃ).See detailed description of figure components in Fig. 2 caption.

Figure 6 .
Figure 6.Box and whisker plots of present-day monthly surface seawater arag (top) and pH (bottom) and monthly mean SST at the coastal mooring locations on the continental shelves of the US east coast (Gulf of Maine, Gray's Reef).See detailed description of figure components in Fig. 2 caption.

Figure 7 .
Figure 7. Box and whisker plots of present-day monthly surface seawater arag (top) and pH (bottom) and monthly mean SST in coral reef ecosystems within a semi-closed atoll lagoon in Micronesia (Chuuk) and a patch reef in the Caribbean Sea southwest of Puerto Rico (La Parguera).See detailed description of figure components in Fig. 2 caption.

Figure 8 .
Figure 8. Relational plot of different modes of (a) arag and (b) pH variability for each ocean acidification mooring location.Statistics describing variability include 1 SD of monthly anomalies (monthly mean -monthly observations), annual amplitude (maximum monthly climatological mean -minimum monthly climatological mean), and 1 SD of annual anomalies (annual meanmean observations).Circles represent open ocean mooring locations, squares denote coastal sites, and triangles denote coral reefs.

Figure 9 .
Figure 9. Percent time that (a) preindustrial and (b) present-day surface seawater arag conditions fall below biological thresholds: chronic exposure for Ostrea lurida larvae at arag < 1.4 in red, acute effect of Crassostrea gigas larvae at arag < 1.5 in gray, chronic exposure for Mya arenaria larvae at arag < 1.6 in gold, chronic exposure for Mytilus californianus larvae at arag < 1.8 in blue, and chronic exposure for C. gigas larvae at arag < 2.0 in black.Thresholds at the Chá bȃ mooring are shown as circles; thresholds at the CCE2 mooring (only for M. californianus larvae) are shown as triangles; thresholds at the Gulf of Maine mooring (only for M. arenaria larvae) are shown as squares.The one acute threshold is indicated by a dashed line.

Table 1 .
Details on each surface seawater p(CO 2 ) and pH mooring time series including abbreviation, name, coordinates, and dates of time series.n is the total number of 3-hourly samples for each pH and arag time series.Finalized data sets can be found organized by mooring location by navigating through the clickable CO 2 data portal map at http://cdiac.ornl.gov/oceans/Moorings/.

Table 2 .
Lee et al. (2006)ity, as measured by 1 SD, for A T and arag data sets at the WHOTS and La Parguera mooring sites.Discrete measurements for WHOTS are from the Hawaii Ocean Timeseries program (http://hahana.soest.hawaii.edu/hot/)andforLa Parguera are from the Atlantic Ocean Acidification Test-Bed project.Proxy A T measurements are estimated from moored SST and SSS usingLee et al. (2006)A T -SSS relationships.p(CO 2 ) and pH used to calculate arag are direct observations from the mooring time series; A T used to calculate arag are the A T proxy measurements.
Notes: * while other observations listed here span the full annual cycle, the WHOTS mooring pH observations only cover the months of June to November.

Table 3 .
Sources of error to the calculation of arag at SST of 25 • C, SSS of 35, p(CO 2 ) of 370 µatm, a pH of 8.1, A T = 2350, and arag of 3.7.Total estimated absolute (relative) uncertainty was calculated using the root sum of squares (RSS) method: RSS = e 2 1/2 .
Notes: a from Sutton et al. (2014b).b Error estimates of thermodynamic constants from McLaughlin et al. ( . Bold values represent values larger than observed values; italicized values represent values lower than observed values.ND signifies no data.

5
Data availabilityPMEL ocean carbon mooring data are archived at the Carbon Dioxide Information Analysis Center (CDIAC) and the National Centers for Environmental Information (NCEI).Finalized data sets can be found organized by mooring location by navigating through the clickable CO 2 data portal map at http://cdiac.ornl.gov/oceans/Moorings/or through the table of PMEL moorings at http://www.nodc.noaa.gov/oceanacidification/stewardship/mooring_table.html.