A high-frequency atmospheric and seawater p CO 2 data set from 14 open-ocean sites using a moored autonomous system

In an intensifying effort to track ocean change and distinguish between natural and anthropogenic drivers, sustained ocean time series measurements are becoming increasingly important. Advancements in the ocean carbon observation network over the last decade, such as the development and deployment of Moored AutonomouspCO2 (MAPCO2) systems, have dramatically improved our ability to characterize ocean climate, sea–air gas exchange, and biogeochemical processes. The MAPCO 2 system provides high-resolution data that can measure interannual, seasonal, and sub-seasonal dynamics and constrain the impact of shortterm biogeochemical variability on carbon dioxide (CO 2) flux. Overall uncertainty of the MAPCO 2 using in situ calibrations with certified gas standards and post-deployment standard operating procedures is < 2 μatm for seawater partial pressure of CO 2 (pCO2) and< 1 μatm for airpCO2. The MAPCO2 maintains this level of uncertainty for over 400 days of autonomous operation. MAPCO 2 measurements are consistent with shipboard seawater pCO2 measurements and GLOBALVIEW-CO2 boundary layer atmospheric values. Here we provide an open-ocean MAPCO 2 data set including over 100 000 individual atmospheric and seawater pCO2 measurements on 14 surface buoys from 2004 through 2011 and a description of the methods and data quality control involved. The climate-quality data provided by the MAPCO 2 have allowed for the establishment of open-ocean observatories to track surface ocean pCO2 changes around the globe. Data are available at doi:10.3334/CDIAC/OTG.TSM_NDP092 andhttp://cdiac.ornl.gov/oceans/Moorings/ndp092 .


Introduction
The global ocean as well as its interactions with the atmosphere, climate, and marine ecosystem is undergoing a rapid and dramatic transition as it responds to multiple drivers on timescales from days to decades.Sustained observations guide our understanding of this ever-evolving earth system, which, in turn, informs the development of solutions for human societies to cope with global change.The iconic Mauna Loa atmospheric carbon dioxide (CO 2 ) time series, or "Keeling curve", is an example of how observations gain importance with time, as they provide the basis for understanding future changes to the earth system in the context of current and historical observations (Keeling et al., 1976;Thoning et al., 1989;Hofmann et al., 2009).Similar "ocean observatories" must be sustained in order to track ocean carbon uptake and ocean acidification in the midst of the large natural temporal and spatial variability in the marine environment.These observations will provide a record of past and current behavior of the ocean carbon system and are central to predicting its future.
While high-quality ocean carbon measurements collected on global hydrographic surveys have been carried out approximately once a decade since the 1980s, the scientific community identified that constraining ocean biogeochemical models would require much greater temporal and spatial resolution of field data (Sabine et al., 2010).Autonomous technology to measure surface ocean carbon was developed to address this need and has undergone rapid advancement in the last three decades (Takahashi, 1961;Weiss et al., 1982;Wanninkhof and Thoning, 1993;Feely et al., 1998;Pierrot et al., 2009).Autonomous underway systems that can measure the partial pressure of CO 2 (pCO 2 ) on ships were the first major breakthrough in our ability to collect high-frequency observations in the global ocean.These systems are designed to produce climate-quality data sets with measurements accurate to within 1 µatm for atmospheric CO 2 and 2 µatm for surface seawater pCO 2 .This level of accuracy has allowed the scientific community to constrain regional sea-air CO 2 fluxes to 0.2 Pg C yr −1 , a level of resolution necessary to test process-based models and predict the future behavior of the carbon cycle (Bender et al., 2002;Pierrot et al., 2009).
While underway pCO 2 observations have greatly enhanced our understanding of the spatial variability in seaair CO 2 fluxes (Takahashi et al., 2009;Wanninkhof et al., 2013), they have not solved the problem of quantifying temporal variability at a given point in space.In highly variable regions such as the equatorial Pacific and coastal systems, fixed, high-frequency observations can improve our understanding of how short-term variability impacts CO 2 flux.Episodic phenomena are important drivers of biogeochemical variability, and mooring time series of pCO 2 and related properties provide the ability to assess the controls and impacts at these short timescales.Seawater pCO 2 observations that fully capture diurnal variations at a fixed site can also be used to test parameterizations of carbon cycle processes used in ocean biogeochemical models.The Moored Autonomous pCO 2 (MAPCO 2 ) system was developed to address this need by autonomously measuring surface ocean pCO 2 and marine boundary layer (MBL) atmospheric CO 2 every 3 h on surface buoys at approximately the same level of accuracy as underway pCO 2 systems.With this recent development of mooring autonomous pCO 2 technology, the combination of all three monitoring approaches (i.e., hydrographic surveys, underway, and buoy measurements) has improved our understanding of the spatial and temporal variability of ocean carbon at the sea surface.For the first time, ocean pCO 2 observations from multiple platforms have been incorporated into the most recent update (v2.0) of the Surface Ocean CO 2 Atlas (SOCAT), a data synthesis effort aimed at bringing together all available CO 2 data in the surface ocean in a common format (Bakker et al., 2014).The data presented here are identical to those in SOCATv2.0.
Here we describe the methods, data quality control (QC), and data access for an open-ocean MAPCO 2 data set collected on 14 surface buoys from 2004 through 2011.These surface ocean pCO 2 observatories are critical for characterizing the natural variability of the ocean carbon cycle, contributing to our understanding of secular trends in ocean chemistry, validating and interpreting modeling results, and developing more sophisticated global carbon models.

Methods and data quality control
In 2004, the National Oceanographic and Atmospheric Administration's (NOAA) Pacific Marine Environmental Laboratory (PMEL) began to work with the Monterey Bay Aquarium Research Institute to improve the accuracy, reliability, and ease of use of an early moored pCO 2 system developed for buoys in the equatorial Pacific.Like the wellestablished underway pCO 2 method (Wanninkhof and Thoning, 1993;Feely et al., 1998;Pierrot et al., 2009), this early moored system described by Friederich et al. (1995) and the MAPCO 2 system described in Sect.2.1 combine air-water equilibrators with an infrared (IR) analyzer for CO 2 gas detection.In 2009, the MAPCO 2 technology was transferred to Battelle Memorial Institute and is commercially available as the Sealogy ® pCO 2 monitoring system.This system is now accessible to the larger scientific community and deployed at over 50 locations in open-ocean, coastal, and coral reef environments, including on NOAA's global moored CO 2 network (www.pmel.noaa.gov/co2/story/Buoys+and+Autonomous+Systems) and Australia's Integrated Marine Observing System (http://imos.org.au).

Description of MAPCO 2 system
The MAPCO 2 system includes four separate watertight cases that house the electronics, battery, transmitter, and a reference gas cylinder.The reference gases used on all the PMEL systems are traceable to World Meteorological Organization (WMO) standards and are provided by NOAA's Earth System Research Laboratory (ESRL).In the electronics case are the controls for the system, a memory flash card for data storage, a LI-COR LI-820 CO 2 gas analyzer, and a Sensirion SHT71 relative humidity and temperature sensor.The MAPCO 2 also includes an oxygen sensor for internal diagnostic purposes.The LI-820 determines the CO 2 gas concentration by measuring the absorption of IR energy as a sample gas flows through an optical path.The CO 2 concentration is based on the difference ratio in the IR absorption between a reference and a sample optical path.The MAPCO 2 uses temperature and relative humidity (RH) to calculate the mole fraction of CO 2 (xCO 2 ) in air in equilibrium with surface seawater.The LI-820 is calibrated before every measurement using a zero-CO 2 reference and an ESRL standard gas that spans the ocean pCO 2 values where the system is deployed.The system also includes a GPS for accurate position and time, an iridium satellite communication link, an airblock deployed approximately 1 m above the ocean surface for atmospheric sampling, and an "h"-shaped bubble equilibrator assembly described by Friederich et al. (1995) (Figs. 1, 2).The equilibrator is the only part of the system in seawater and is made of copper-nickel alloy to prevent bio-fouling.
A schematic diagram of the main components and sampling paths in the MAPCO 2 system is shown in Fig. 1.A typical measurement cycle, including in situ calibration and the atmospheric and seawater measurements, takes approximately 20 min.At the beginning of each cycle, the system generates a zero standard by cycling a closed loop of air through a soda lime tube to remove all of the CO 2 .This scrubbed air establishes the zero calibration.Next, the system is calibrated with a high standard reference gas, or "span" gas.The value of this gas is set in the MAPCO 2 system before deployment (typically ∼ 500 µmol mol −1 ).The gas flows through the detector for CO 2 analysis and is vented to the atmosphere through the airblock.Once the detector is fully flushed, the flow is stopped and the system returns to atmospheric pressure.Using a two-point calibration from the zero and span values, the LI-820 is optimized for making surface ocean CO 2 measurements.
To make the seawater xCO 2 measurement, the MAPCO 2 system equilibrates a closed loop of air with surface seawater in the h-shaped equilibrator, which is mounted in a float designed by PMEL to ensure the optimum depth for equilibration (Fig. 2).The air cycles through the system by pumping air out of flexible polytetrafluoroethylene (PTFE) tubing Air is pumped from the MAPCO 2 through a PTFE tube and bubbled into the equilibrator.As the bubbles rise through the water, the air comes into equilibrium with the dissolved gases in the surface seawater.The rising air bubbles in the equilibrator also create circulation by pushing water up and over the horizontal leg of the hshaped equilibrator and out the short leg of the equilibrator.Image is not to scale.
to approximately 14 cm beneath the surface of the seawater.While the air bubbles through the column of water, the air comes into equilibrium with the dissolved gases in the surface seawater.This air then returns to the system, passing through a silica gel drying agent and the relative humidity sensor.The drying agent is used to prevent condensation in the LI-820 detector and is replaced after each deployment.The air then circulates through the equilibrator again.The closed loop of air repeats this cycle for 10 min.The rising air bubbles in the equilibrator create seawater circulation in the equilibrator by pushing the water up and over the horizontal leg of the equilibrator and out the short leg of the equilibrator (Fig. 2).This draws new water into the long leg of the equilibrator, ensuring that the recirculated air is always in contact with new seawater.After 10 min of equilibration, the pump is stopped and the LI-820 values are read on the air sample at 2 Hz for 30 s and averaged to give the seawater xCO 2 measurement.This is a measurement of integrated seawater CO 2 levels during the 10 min equilibration time.
After the equilibrator reading, a MBL air reading is made by drawing air in through the airblock, partially drying it, and Inner circle color illustrates the mean pCO 2 of the finalized data at that location.Inner circle size is relative to the environmental variability in the time series defined here as the standard deviation of seawater pCO 2 values.The outer ring shows the proportion of environmental variability in seawater pCO 2 due to the seasonal cycle (black) and interannual variability (gray).Seasonal variability is defined as the mean seasonal peak amplitude, and interannual variability is the mean of annual mean values.Seasonal and interannual variability cannot be quantified at JKEO with a time series of < 1 year and is represented here by an outer ring with no color.
passing it through the LI-820.Once the LI-820 path has been flushed, the flow is stopped and a 30 s average reading is collected.All measurements and calibrations are made at atmospheric pressure.The seawater CO 2 measurement occurs approximately 17 min after the start of the measurement cycle followed by the air CO 2 measurement 2 min later.Response time of the MAPCO 2 is dictated by the length of the full 20 min measurement cycle, and in fast mode the MAPCO 2 system can measure atmospheric and seawater CO 2 once every 30 min.
Different types of sensors are used throughout the system for analytical, troubleshooting, and data quality control purposes.Additional parameters measured in each cycle (i.e., zero, span, equilibrator, and air) include temperature, pressure, relative humidity, and oxygen.Other sensors can also be integrated into the MAPCO 2 , including CTD (conductivity, temperature, and depth) instruments with auxiliary sensors attached (e.g., dissolved oxygen, fluorescence, turbidity) and pH sensors.The raw data collected by the MAPCO 2 and integrated sensors are stored on a memory flash card, and averaged data from each 3-hourly cycle are telemetered from the buoy via the iridium satellite communications system.This communications system also enables the user to control the MAPCO 2 remotely.The user can determine the sampling frequency and other variables, but the MAPCO 2 is nominally designed to make CO 2 measurements every 3 h with daily data transmissions for at least 400 days.
PMEL's MAPCO 2 systems have been deployed on openocean buoys starting in 2004 with the establishment of NOAA's global moored CO 2 network and the efforts of numerous partners (see Acknowledgements).Table 1 lists the mooring coordinates and dates of CO 2 time series operation; Fig. 3 illustrates the locations, number of measurements, and average pCO 2 (sea-air) from the 14 surface CO 2 buoys included in this data set.These mooring pCO 2 observations are consistent with results of a synthesis of underway observations reported by Takahashi et al. (2009).Other than the Bermuda Testbed Mooring (BTM) and Japanese Kuroshio Extension Observatory (JKEO) time series, which have been discontinued, and the Multi-disciplinary Ocean Sensors for Environmental Analyses and Networks (MOSEAN) buoy, which was moved approximately 20 km to the new Woods Hole Oceanographic Institution (WHOI) Hawaii Ocean Time-Series Station (WHOTS) location, the MAPCO 2 time series shown in Fig. 3 and Table 1 continue to be maintained.Seven of the 14 CO 2 buoys are located in the equatorial Pacific on the Tropical Atmosphere Ocean (TAO) array.Additional open-ocean MAPCO 2 sites maintained by PMEL now exist in the North Atlantic, northern Indian, and Southern oceans (see http://www.pmel.noaa.gov/co2/story/Buoys+and+Autonomous+Systems); however, they have been deployed since 2011 and are not included in the finalized data set presented here.

Data reduction and processing
The IR analyzer has a nonlinear response to CO 2 , but that response is very well characterized by the manufacturer.LI-COR has a function built into their firmware that accounts for the nonlinear response and linearizes the output data.The linear function is calibrated prior to each atmospheric and seawater measurement with the zero-(intercept) and high-CO 2 standard reference gas (slope).The accuracy of the linearized, calibrated output is confirmed prior to deployment by analyzing a range of intermediate-CO 2 standards in our laboratory.
The primary check of accuracy before and after deployment is a comparison to ESRL CO 2 standards traceable to WMO standards, typically six standards that range from 0 to < 800 µmol mol −1 .Systems are not certified for deployment until values are within the expected range of the standards that span the typical seawater CO 2 values at the mooring location (typically within 2 µmol mol −1 ).A comparison to the underway pCO 2 system in the lab is then done to assess stability of the measurements over at least 1 week.During this test, each MAPCO 2 is tested in a seawater tank in the lab against another MAPCO 2 system and a General Oceanics 8050 underway pCO 2 system that are permanently mounted for continuous sampling in the seawater tank.The standard MAPCO 2 is regularly compared to the underway system, which is calibrated every 8 h using four standard reference gases from approximately 0 to 1000 µmol mol −1 .Laboratory testing of the MAPCO 2 systems suggests instrument precision is < 0.6 µmol mol −1 for xCO 2 values between 100 and 600 µmol mol −1 .
When the MAPCO 2 is recovered from the field, the system is compared against six gas standards to verify accuracy, and the high-frequency raw data stored on the internal memory flash card are downloaded to a local database.The highfrequency raw data from each 3-hourly cycle are then used for final processing of each data set.Averaged xCO 2 (wet) seawater and atmospheric measurements (defined in Table 2) from each cycle are calculated starting with the raw detector counts using the published LI-COR function.The span gas coefficients used in the function during post-processing are derived from the linear regression between the calibration coefficients and the corresponding LI-820 temperature measurements acquired during the span cycle over the course of the deployment.This post-deployment reprocessing facilitates the accurate calculation of xCO 2 (wet) values from the raw detector counts when rare miscalibrations occur, resulting in erroneous coefficients during the deployment.Since the LI-820 is calibrated prior to each cycle of xCO 2 (wet) measurements using the zero-and high-CO 2 standard reference gas, detector drift is negligible.This is confirmed by a mean difference between corrected and original raw data of −0.02 µmol mol −1 .
Data are quality-controlled and flagged according to the SOCAT guidelines (Pfeil et al., 2013).For pCO 2 mooring purposes, we use three quality flags (QFs): a flag value of 2 represents an acceptable measurement, 3 is a questionable measurement, and 4 is a bad measurement.A measurement can be questionable for a variety of reasons often revealed by MAPCO 2 system diagnostic information (e.g., low equilibrator pressure causing incomplete seawater equilibration), and the reasoning for each flag is included in the metadata QC log so the end user can decide whether or not to use questionable data.Prior to a data QC software update in June 2013, xCO 2 values flagged as bad (QF = 4) were still included in the published data sets, but after the software update bad values are replaced with −999.Other parameters published in the data sets that do not have an associated flag, such as sea surface temperature (SST) and sea surface salinity (SSS), are given a value of −999 or −9.999 when the measurement is missing or bad.Oxygen measured in the MAPCO 2 system is exposed to air and likely modified within the system prior to measurement.Rapid changes in oxygen are not properly captured using this method.This data should not be used as a quantitative measure of oxygen.c Usually measured by other academic partners at each site.See metadata for each deployment for details on SST and SSS measurements.d pCO2 only presented in data sets submitted to CDIAC after June 2013 when QC software was upgraded to include this calculation.Data users of earlier data sets can calculate pCO2 as defined in Eq. ( 4).
As a final check of the data QC process, atmospheric xCO 2 (dry) data are compared to MBL data from the GLOBALVIEW-CO2 product and the MAPCO 2 systems deployed before and after the deployment of interest (GLOBALVIEW-CO2, 2013).When a MAPCO 2 system is recovered and a new system deployed, there is typically some overlap in measurements at each location.In cases when there is an offset in air xCO 2 values between systems at the same location, which is often corroborated by an offset from the GLOBALVIEW-CO2 MBL time series as well, a correction (typically ≤ 3 µmol mol −1 ) is applied to the atmospheric and seawater xCO 2 (wet) values.This correction is noted in the metadata and can be removed by the data user if desired.The GLOBALVIEW-CO2 MBL data set serves as a useful and unifying comparison data set, especially since other in situ comparison data are often lacking.As we build MAPCO 2 time series at each of these locations, we start to build an understanding of how the MAPCO 2 observations typically compare to the MBL data set.For example, winter atmospheric xCO 2 values measured by our MAPCO 2 systems at Papa are consistently lower than MBL values (Fig. 4a).
Post-QC calculation of pCO 2 and f CO 2 (fugacity of CO 2 ) are made according to recommendations of the underway pCO 2 community (Pierrot et al., 2009).However, MAPCO 2 measurements of xCO 2 vary from the underway pCO 2 method.The MAPCO 2 system uses the LI-820 and the RH to report the mole fraction of CO 2 in air in equilibrium with surface seawater, called xCO 2 (wet).This "partially wet" measurement typically has a RH of ∼ 75 % (seawater and atmospheric samples), which is not completely dried as in the underway pCO 2 method, due to lack of drying methods available for extended autonomous operation.However, since we measure RH and temperature of the sample air stream exiting the LI-820, we can calculate xCO 2 (dry) using Eqs.(1)-(3).First, xCO 2 in dry air is calculated by where xCO 2 (wet) is the LI-820 measured concentration (µmol mol −1 ), P Licor is the pressure of the atmospheric and seawater samples measured in the LI-820 (kPa) and considered atmospheric pressure, and VP Licor is the vapor pressure in the LI-820 (kPa).RH measurements of the air samples exiting the LI-820 are used to calculate VP Licor in Eq. (1) using the following as defined by Buck (1981) and LI-COR for the IR analyzers: (2) where VP sat is the saturation vapor pressure of the RH sensor cell (kPa); T RH is the temperature of the RH sensor (  RH sample is the RH of the air sample (%); and RH span is the RH of the span (%), i.e., the background RH level for the system.Equation ( 2) is a calculation of vapor pressure optimized for the temperature interval of −20 to 50 • C as defined by Buck (1981).This equation includes coefficients for calculating VP sat with an enhancement factor (a correction for dealing with moist air as a function of temperature and pressure) of 1.004 for 20 • C and 1000 mb (Buck, 1981).
VP sat and RH of the air sample are then used to calculate VP Licor .Once the VP Licor is known, the dilution effect can then be removed from the partially wet xCO 2 measurement using Eq. ( 1) to calculate xCO 2 (dry).Since the MAPCO 2 equilibration occurs directly in the ocean, it does not require the warming correction necessary for underway pCO 2 systems.Therefore, pCO 2 in wet air (100 % saturation) in equilibrium with the surface seawater is calculated by where P Licor is atmospheric pressure for the atmospheric and surface seawater samples (atm) and pH 2 O is the water vapor pressure (atm) at equilibrator temperature as defined by Weiss and Price (1980).f CO 2 in wet air (100 % saturation) in equilibrium with the surface seawater is calculated by where the ideal gas constant R = 82.0578cm 3 atm mol −1 K −1 , T is SST (K) from the CTD, and the B 11 virial coefficient and δ 12 cross-virial coefficient for CO 2 are as defined by Weiss (1974).The raw CO 2 data, temperature, salinity, and pressures are included in all published MAPCO 2 data sets so other data users can recalculate xCO 2 , f CO 2 , and pCO 2 .Additional parameters included with the pCO 2 mooring data set are listed and described in Table 2.

Uncertainty of pCO 2 measurements
Precision and accuracy of the MAPCO 2 measurements have been assessed in both laboratory and field settings.As stated in Sect.2.2, the precision of the MAPCO 2 system in a laboratory setting is 0.6 µmol mol −1 .Standard deviation of the high-frequency raw data (∼ 58 repeated measurements over 30 s) in the field is a good assessment of the in situ precision of the MAPCO 2 system.Mean standard deviation of the raw data from the 14 buoy time series presented here is 0.7 µmol mol −1 for seawater xCO 2 and 0.6 µmol mol −1 for air xCO 2 , which is similar to precision measured in the laboratory.While estimating accuracy in a laboratory setting is feasible, the more-desired estimate of in situ accuracy is difficult to obtain due to the limited availability of validation samples for comparison and the mismatch in space and time of these validation samples compared to the MAPCO 2 measurements.These issues related to accuracy will be discussed in more detail below.In this section, we present MAPCO 2estimated in situ precision, accuracy, and uncertainty, which we define as the overall error of the measurement encompassing instrument precision and accuracy as well as propagation of error.
Propagation of error must be considered when calculations are based on variables with individual uncertainties.These types of errors that impact the calculated pCO 2 and f CO 2 values have been assessed for underway pCO 2 systems and are typically small (< 0.1 µatm) with minimal impact to the overall uncertainty when combined with the larger uncertainty (< 2 µatm) in the actual xCO 2 measurement (Feely et al., 1998;Wanninkhof and Thoning, 1993;Pierrot et al., 2009).However, we utilize a different method to calculate xCO 2 (dry) for the MAPCO 2 system, as discussed in Sect.2.1, so it is important to address the potential error in this new method.The RH measurements used to calculate xCO 2 (dry) have separate precisions and accuracies that can propagate through Eqs. ( 1)-(3) (Table 3).The total estimated precision and accuracy of xCO 2 (dry) are calculated by summing each variable's precision and accuracy using the root-sum-of-squares method.As presented in Table 3, propagation of all the errors from the separate variables does not cause the precision of calculated xCO 2 (dry) to differ from measured xCO 2 (wet) and results in a small impact to the accuracy (0.1 µmol mol −1 ).
In addition to the propagation of error, an estimate of in situ accuracy is key to determining the overall uncertainty of the MAPCO 2 system.The GLOBALVIEW-CO2 data product maintained by NOAA ESRL can be used as one data set for comparison to the MAPCO 2 air xCO 2 (dry) measurements (GLOBALVIEW-CO2, 2013).Figure 4 shows 3-hourly atmospheric MAPCO 2 measurements and biweekly atmospheric CO 2 values from the MBL layer of GLOBALVIEW-CO2 at the latitude closest to each MAPCO 2 location.Atmospheric MAPCO 2 data presented here are in the finalized, processed form as described in Sect.2.2.Both MBL and MAPCO 2 data capture seasonal variability and long-term trends, but, as expected, highfrequency MAPCO 2 measurements show short-term variability typically deviating from the smoothed MBL data product by < 5 µmol mol −1 (Fig. 4).The Mauna Loa atmospheric CO 2 record is also shown in Fig. 4c and provides a reference for illustrating the larger seasonal variability in the lower atmosphere directly influenced by the presence of the ocean's surface.For the time series longer than 2 years, growth rates of the 3-hourly MAPCO 2 and biweekly MBL atmospheric CO 2 are presented in Table 4. Atmospheric CO 2 growth rates observed by five of the seven mooring time series differ from the MBL data by ≤ 0.1 µmol mol −1 yr −1 , suggesting that the finalized MAPCO 2 observations are consistent with other atmospheric data products generated using different methods.
MAPCO 2 and MBL data are compared in more detail in Table 5.This includes descriptive statistics of the finalized, processed atmospheric data in addition to prefinalized data prior to any adjustments or offsets.The 3hourly MAPCO 2 measurement that is closest in time to the biweekly MBL estimate is used to calculate the (MAPCO 2 -MBL).Pre-QC MAPCO 2 data show a slight negative bias (−1.5 ± 2.4 µmol mol −1 ) to MBL values (Table 5).The mean difference between finalized MAPCO 2 data and MBL values is smaller (−0.3 ± 1.7 µmol mol −1 ) due to the application of occasional offsets during data QC described in Sect.2.2.Standard deviations likely reflect the natural variability in atmospheric CO 2 at the sea surface illustrated in Fig. 4. Low standard error of the mean and low confidence Earth Syst.Sci.Data, 6, 353-366, 2014 www.earth-syst-sci-data.net/6/353/2014/ level values reported for the atmospheric comparison in Table 5 suggest strong statistical significance in the mean MAPCO 2 -MBL values.
While environmental variability may introduce some error to the MAPCO 2 and MBL air comparison, the resulting mean differences in the atmospheric data are likely due primarily to uncertainty in the measurements, which in this case we associate with the MAPCO 2 system.However, surface ocean pCO 2 exhibits large temporal and spatial variability.For example, it is common to observe variability in underway pCO 2 measurements from the R/V Atlantic Explorer of approximately 10 µatm within 10 km of BTM over a period of 3 h (Fig. 5).We observe even larger variability in the eastern equatorial Pacific, with changes up to 50 µatm over a period of 3 h and > 100 µatm over the course of a day (Fig. 6a).This patchiness can create errors in comparing MAPCO 2 measurements to ship-based measurements made at safe distance from the surface buoy.In Fig. 6b, for example, the difference between the TAO125W MAPCO 2 and underway measurements from the R/V Ka'imimoana (made within 10 km and 10 min of the MAPCO 2 measurement) start at ±2 µatm on 31 January 2006 at 14:00:00, but as the ship begins to leave the surface buoy 6 h later the measurements diverge as the MAPCO 2 starts to detect a decreasing trend in surface seawater pCO 2 values at the buoy location that persists for the next 8 days.In another example shown in Fig. 6c, the 15 µatm difference between the MAPCO 2 and underway system observed on 10 November 2008 is similar to the daily variability observed at the buoy in the 4 days prior to arrival of the Ka'imimoana and could reflect true differences observed by the underway and MAPCO 2 systems located 1-7 km apart.These examples highlight the difficulty of separating environmental variability and instrument uncertainty in these types of comparison exercises.
In order to minimize environmental variability while maximizing sample size for descriptive statistics, we use discrete measurements made within 10 km and 1.5 h and averaged underway pCO 2 measurements made within 10 km and 10 min of the MAPCO 2 system measurements for the seawater pCO 2 comparison analysis.While underway and MAPCO 2 systems utilize similar methodology, discrete pCO 2 presented in Table 5 is calculated from measurements of dissolved inorganic carbon (DIC) and total alkalinity (TA) using the program CO2SYS developed by Lewis and Wallace (1998) with the constants of Lueker et al. (2000).Typical error in calculated pCO 2 using this method is < 5 %.Only finalized seawater MAPCO 2 data are used for the descriptive Table 5. Descriptive statistics of (MAPCO 2 measurement -comparison measurement).The MAPCO 2 measurements (both pre-and post-offset if applied during data QC) are compared to biweekly GLOBALVIEW-CO2 MBL values from the latitude nearest to average buoy location, single discrete measurements made within 10 km and 1.5 h, and averaged underway pCO 2 measurements made within 10 km and 10 min of the MAPCO 2 system measurement.Standard error is the standard error of the mean, and confidence intervals illustrate that with a 95 % probability the actual population mean = sample mean ± confidence interval.statistics presented in Table 5.Unlike the descriptive statistics for the MAPCO 2 air comparisons, the statistics that result from using MAPCO 2 seawater measurements pre-MBL offset are not statistically different than the finalized, post-MBL offset statistics presented in Table 5.This could be due to the large natural variability in seawater pCO 2 compared to atmospheric CO 2 .Agreement between discrete and mooring surface ocean pCO 2 measurements is within 1.3 µatm (mean in Table 5; BTM example in Fig. 5).Although more discrete measurements have been made at these and other mooring locations, this comparison is based on discrete samples restricted to within 10 km and 1.5 h of the MAPCO 2 system measurements with n > 5.Even with these restrictions, it is likely that environmental variability is not completely removed and is reflected in the mean standard deviations of 3.7-6.2µatm (Table 5).The small sample sizes (≤ 10 at each site) also resulting from these restrictions create large uncertainty in mean values, with standard error and confidence levels exceeding mean values.This analysis shows promising results with a close agreement between discrete and MAPCO 2 measurements; however, more discrete samples will need to be collected within 10 km and 1.5 h of MAPCO 2 system measurements in order to improve the statistical significance of the seawater pCO 2 comparison.Sample sizes are larger (13 ≤ n ≤ 76) for the comparison between underway and MAPCO 2 measurements at the BTM, TAO125W, and TAO140W locations.While underway measurements exist at other equatorial Pacific mooring locations, comparisons within 10 km and 10 min are restricted to TAO125W and TAO140W due to the large gaps in pCO 2 mooring data, the infrequent mooring-servicing ship visits to each site (∼ once every 1-1.5 years), and the necessity for the mooring-servicing ship to leave for the next station before the MAPCO 2 system has gone through a few cycles and measurements have stabilized.Even with these challenges, there are 76 comparison samples at BTM during the two buoy deployments in 2006deployments in -2007 (Fig. 5 (Fig. 5).These measurements show a mean of 1.8 ± 4.8 µatm with a low confidence interval of 1.1, indicating strong statistical significance (p < 0.05) that the actual mean is between 0.7 and 2.9 µatm (Table 5).Standard deviations of the difference between the BTM versus discrete (5.6) and underway (4.8) measurements are similar, which may be reflective of the environmental variability in this region of the surface ocean.Mean in the equatorial Pacific is higher (−3.3 ± 15.2 µatm at TAO125W and 2.1 ± 8.3 µatm at TAO140W), but statistical significance of these values is low due to the lower sample sizes and higher environmental variability in this region (Fig. 3).The largest standard deviation in mean of 15.2 is at TAO125W, which is the site that exhibits the largest natural variability (i.e., total range of ∼ 200 µatm, Fig. 6a) in surface seawater pCO 2 of the open-ocean mooring data sets compared in Table 5.

Confidence
The MAPCO 2 system has also been involved in two independent ocean pCO 2 instrument intercomparisons.During an Alliance for Coastal Technologies demonstration project, the difference between the MAPCO 2 system and an underway pCO 2 system was −9 ± 8 µatm in coastal Washington, USA waters and −3 ± 9 µatm in coral reef waters of Kaneohe Bay, Hawaii, USA (Schar et al., 2010).Separating environmental variability from instrument uncertainty in this case is challenging.Small-scale environmental variability (i.e., meters) due to natural spatial patchiness of pCO 2 was determined to be 10-15 µatm at the coastal site and < 2 µatm at the coral site and may account for much of the difference observed between the MAPCO 2 and reference measurements.An intercomparison between buoy and underway pCO 2 systems held at the National Research Institute of Fishery Engineering in Hasaki, Kamisu city, Ibaraki, Japan, was done in the more controlled environment of an indoor seawater pool (UNESCO, 2010).In this intercomparison, the MAPCO 2 was within 1 µatm compared to the underway pCO 2 reference system in conditions within the calibration gas range.
In summary, the MAPCO 2 system performs very well in laboratory and field settings in comparison to a variety of other methods.Considering the precision estimate of the MAPCO 2 measurements in the field (< ±0.7 µmol mol −1 ), the statistically strong (p < 0.05) mean differences in MAPCO 2 versus comparison measurements in Table 5 (< ±1.8 µatm), and the small propagation of error resulting from the xCO 2 (dry) calculation (< ±0.1 µmol mol −1 ), we estimate in situ MAPCO 2 precision at < ±0.7 µmol mol −1 and accuracy at < ±2.0 µmol mol −1 for xCO 2 (dry) measurements.Overall uncertainty of pCO 2 and f CO 2 observations from the MAPCO 2 system is estimated to be < 2.0 µatm for values between 100 and 600 µatm for over 400 days of autonomous operation.However, the uncertainty of finalized, quality-controlled data is likely better for atmospheric pCO 2 and f CO 2 observations at < 1.0 µatm when following the post-deployment standard operating procedures described in Sect.2.2.

Data description and access
Finalized MAPCO 2 data are reported to the Carbon Dioxide Information Analysis Center (CDIAC; http://cdiac.ornl.gov/oceans/Moorings) and archived at additional data centers such as the National Oceanographic Data Center (http://www.nodc.noaa.gov).The archived data are organized by site and deployment date.The numeric data package (NDP) associated with this publication includes the 56 deployments listed in Table 6 and is available at doi:10.3334/CDIAC/OTG.TSM_NDP092 or http://cdiac.ornl.gov/oceans/Moorings/ndp092.The methods described here are associated with the mooring pCO 2 data included in this NDP.These data are made freely available to the public and the scientific community in the belief that their wide dissemination will lead to greater understanding and new scientific insights.Users of the data are requested to cite this publication when using the entire open-ocean mooring data set or cite according to the CDIAC data archive when using individual mooring data sets.When preparing manuscripts using these data, users are asked to invite lead pCO 2 mooring investigators to coauthor or to send draft manuscripts using these data to the lead investigators to ensure that the quality and limitations of the data are accurately represented.
The mooring data set includes 3-hourly seawater and atmospheric CO 2 observations from 14 moorings since 2004, encompassing over 100 000 individual measurements.As presented in Fig. 3, climatological means of surface ocean pCO 2 measured on moorings are consistent with observations from other platforms (Bakker et al., 2014;Takahashi et al., 2009); however, much of the value in high-frequency mooring observations is demonstrated at shorter timescales.Figure 3 shows that short-term (≤ 2 years) variability at the subtropical sites tends to be dominated by the seasonal cycle, and tropical sites tend to be dominated by interannual variability.At the subtropical sites, seawater CO 2 is typically highest in the summer and lowest in the winter.The Papa site is the highest-latitude mooring in this data set and exhibits approximately equal short-term variation driven by the seasonal cycle and interannual variability caused by strong weather events in this region of the North Pacific.The highest interannual variability is observed in the equatorial Pacific driven by El Niño and La Niña events (Fig. 3) and dominates any small seasonal signal that may exist in this region (Sutton et al., 2014).In the most extreme conditions, seawater pCO 2 values can vary over 100 µatm within 24 h at 0 • , 125 • W (Fig. 6a).Variability of 100-150 µatm is also common in the equatorial Pacific during the extension of the warm water pool during El Niño events on timescales of months and the passing of tropical instability waves on timescales of weeks (e.g., Fig. 4 in Sutton et al., 2014).Sustained, long-term mooring time series also provide the opportunity to identify and remove the short-term variability from the time series and investigate long-term trends.For example, in a synthesis of equatorial Pacific mooring data, Sutton et al. (2014) found that the uptake of anthropogenic CO 2 and an acceleration in equatorial upwelling since the shift in the Pacific Decadal Oscillation in 1998 has led to high rates of pCO 2 change of +2.3 to +3.3 µatm yr −1 in this region.This decadal shift in CO 2 outgassing is consistent with underway pCO 2 observations made in this region since 1982 (Feely et al., 2014).
Mooring data from most of the deployments through 2010 listed in Table 6 are also included in the most recent version of SOCAT (Bakker et al., 2014).This SOCATv2.0 synthesis involves a standardized, second-level quality control of 10.1 million surface seawater f CO 2 measurements from many different sources, including underway and mooring systems.SOCAT also produces a gridded surface ocean f CO 2 data product in a uniform format available at http: //www.socat.info.Rödenbeck et al. (2013) compared the previous version of SOCAT (v1.5), which did not include mooring data, to some of the open-ocean MAPCO 2 time series in Table 6.In a comparison between seawater pCO 2 data from the TAO170W MAPCO 2 and data-driven model estimates based on SOCATv1.5, Rödenbeck et al. (2013) find that seawater pCO 2 estimates in the tropics are unrelated, or even opposite, to the mooring observations.This discrepancy arises because that particular location is not well constrained by the SOCATv1.5 data set.We expect the recent mooring additions to SOCATv2.0 and the open-ocean MAPCO 2 data set presented here to make a large impact on our efforts to model and understand the global carbon cycle in the coming years.

Conclusion
Mooring observations can play a critical role in improving our ability to model, understand, and describe the ocean carbon cycle on all timescales.In particular, time series from remote, data-sparse areas of the ocean collected on moorings fulfill a unique niche by providing the high-resolution data necessary to explore questions about short-term variability at fixed locations.Here we provide a data set of 3hourly surface seawater and marine boundary layer atmospheric pCO 2 observations on 14 open-ocean moorings in the Pacific and Atlantic from 2004 to 2011.When using the in situ and post-calibration methods described here, overall uncertainty for the MAPCO 2 data is < 2 µatm for seawater pCO 2 and < 1 µatm for air pCO 2 , making the MAPCO 2 system a climate-quality method for tracking surface ocean pCO 2 .These types of sustained, temporally resolved observations allow us to improve our understanding of the role of shorter-term variability and key biogeochemical processes on the global carbon system.Potential uses of these data to inform our understanding of a changing ocean include investigating high-frequency variability in surface ocean biogeochemistry, developing seasonal CO 2 flux maps for the global oceans (e.g., Takahashi climatology and SOCAT), studying ocean acidification, and evaluating regional and global carbon models.

Figure 1 .
Figure 1.Schematic diagram of main components and sampling paths within the MAPCO 2 system.The floating air-water equilibrator is shown in more detail in Fig. 2.

Figure 2 .
Figure 2. Schematic diagram of the floating air-water equilibrator assembly in the MAPCO 2 system during the seawater equilibration cycle.Air is pumped from the MAPCO 2 through a PTFE tube and bubbled into the equilibrator.As the bubbles rise through the water, the air comes into equilibrium with the dissolved gases in the surface seawater.The rising air bubbles in the equilibrator also create circulation by pushing water up and over the horizontal leg of the hshaped equilibrator and out the short leg of the equilibrator.Image is not to scale.

Figure 3 .
Figure3.Location of open-ocean moorings in the MAPCO 2 data set.Inner circle color illustrates the mean pCO 2 of the finalized data at that location.Inner circle size is relative to the environmental variability in the time series defined here as the standard deviation of seawater pCO 2 values.The outer ring shows the proportion of environmental variability in seawater pCO 2 due to the seasonal cycle (black) and interannual variability (gray).Seasonal variability is defined as the mean seasonal peak amplitude, and interannual variability is the mean of annual mean values.Seasonal and interannual variability cannot be quantified at JKEO with a time series of < 1 year and is represented here by an outer ring with no color.

Figure 6 .
Figure 6.(a) TAO125W surface seawater MAPCO 2 observations (gray points) for the entire time series at this location with average R/V Ka'imimoana underway pCO 2 data within 10 km and 10 min of the MAPCO 2 measurements (black open circles).Two examples of comparison data over 1-week time series are shown in panels (b) and (c), with MAPCO 2 measurements corresponding to the average underway observations illustrated in gray open circles.Selection boxes in (a) are not to scale of actual axes in (b) and (c) panels.Ka'imimoana data from NOAA PMEL, http: //cdiac.ornl.gov/oceans/VOS_Program/kaimimoana.html.

Table 1 .
Details of each CO 2 mooring time series including name, coordinates (decimal degrees), and dates of operational CO 2 measurements.

Table 2 .
Final data variable names and descriptions.
Notes: a SOCAT flags used in this data set: 2 = acceptable measurement; 3 = questionable measurement; 4 = bad measurement (note: bad data values are reported in the final data file submitted to CDIAC prior to QC software upgrade in June 2013 but reported as −999 in files submitted after the upgrade).b

Table 3 .
Sources of error for the calculation of xCO 2 (dry) at atmospheric pressure = 101 kPa, RH sample = 75 %, RH span = 30 %, SST = 25 • C, SSS = 35, and xCO 2 (wet) = 375 µmol mol −1 .Total estimated precision and accuracy are calculated using the root-sum- Notes: a Error calculated using manufacturer-estimated error for T RH of ±0.1 • C precision and ±0.3 • C accuracy (see Eq. 2).b Error reported by manufacturer.c Precision estimate based on standard deviation of the high-frequency raw data (∼ 58 repeated measurements over 30 s) in the field; accuracy estimate based on pre-deployment testing in the laboratory.Negligible indicates value < significant digits of variable.

Table 4 .
Growth rate of GLOBALVIEW-CO2 MBL and MAPCO 2 atmospheric xCO 2 time series over the time period of the data sets listed in Table 5(GLOBALVIEW-CO2, 2013).For mooring time series locations see Fig.3.
Figure 5. Seawater pCO 2 values from BTM MAPCO 2 (gray points), Bermuda Atlantic Time-series Study (BATS) discrete (plus signs), and R/V Atlantic Explorer underway (open circles) used in the Table 5 statistics.BATS data from Bermuda Institute of Ocean Sciences, bats.bios.edu.Atlantic Explorer data from Bermuda Institute of Ocean Sciences, http://cdiac.ornl.gov/oceans/CARINA/.

Table 6 .
List of open-ocean mooring deployments in the open-ocean MAPCO 2 data set.n is the total number of measurements collected at each mooring location during these deployments.