Analysis of seasonality and annual mean distribution of atmospheric potential oxygen (APO) in the Pacific region

Authors


Abstract

[1] We present a data set of atmospheric potential oxygen (APO = O2 + 1.1 × CO2) based on the atmospheric O2/N2 and CO2measurements of flask samples collected at two monitoring stations in Japan and on commercial cargo ships sailing between Japan and U.S./Canada and Australia/New Zealand. Since APO is invariant with respect to the terrestrial biotic exchange, its variation mainly reflects the spatiotemporal distribution in the air-to-sea gas exchange. From the observed APO for the years 2002–2008, we find: (1) elevated annual mean values near the equator, (2) elevated annual mean values and large seasonal amplitudes in the northwestern North Pacific, and (3) a deep trough of low annual mean values at latitudes 20–40°N in the Western Pacific. In addition, latitudinal distributions in the timing of the observed seasonal maximum and minimum show asymmetric patterns across the equator. Comparing these observations with a series of simulated APO generated in the NIES99 atmospheric transport model driven by a set of climatological oceanic O2 and CO2 flux fields, we find a good agreement except for the observed deep trough at the midlatitude. Simulations with different transport mechanisms and fluxes reveal that the seasonal covariation between oceanic O2 flux and atmospheric transport contribute significantly to the observed APO variations in the northern North Pacific; also the seasonal variation in the meridional transport affects the latitudinal difference in the seasonal cycle. The observed latitudinal gradient of the annual mean APO in the Southern Hemisphere is better reproduced by the model based on the recently revised ocean CO2 flux distribution than that based on the previous CO2 flux distribution. The observed APO trough at 36°N in the Western Pacific is about 10 per meg lower than the simulation with the more recent pCO2 data, suggesting the existence of additional APO sinks in that latitudinal region. Indeed, a model simulation performed with an additional ocean O2 sink flux of about 30 Tmol yr−1 within the region (30–50°N, 120–180°E) reproduces considerably well the observed APO trough.

1. Introduction

[2] Observation of atmospheric O2 concentration change, combined with atmospheric CO2observation, provides complementary information about the global carbon cycle and air-sea gas exchange. This is based on the fact that O2 and CO2fluxes are tightly coupled during terrestrial photosynthesis, respiration, and fossil fuel and biomass burning, but are essentially decoupled during air-sea gas exchange. For example, long-term O2 changes can be used to partition the sequestration of fossil fuel CO2 between the ocean and the terrestrial biosphere [Keeling and Shertz, 1992; Bender et al., 1996; Battle et al., 2000; Bender et al., 2005; Manning and Keeling, 2006; Tohjima et al., 2008]. The idea that a combination of atmospheric O2 and CO2observations can provide information about the air-sea gas exchange ledStephens et al. [1998] to introduce a new tracer called atmospheric potential oxygen (APO), which is defined as

display math

Because the factor of 1.1 in equation (1) represents the average stoichiometric –O2:C exchange ratio for land biotic processes [Severinghaus, 1995], APO is invariant with respect to the land biotic gas exchange. In addition, APO depends only slightly on fossil fuel combustion, which contributes to the global long-term decrease and the inter-hemispheric gradient due to slightly larger –O2:C stoichiometric ratios (1.4 on global average) and the localization of fuel consumption in the Northern Hemisphere. Consequently, the major contributor to the seasonal variation and the global distribution of APO is the air-sea O2 and CO2 exchange. As will be mentioned in section 2, O2/N2 ratio is used to express the atmospheric O2concentration. Therefore, the air-sea N2 exchange also slightly affects the APO variation.

[3] The oceanic O2 and CO2fluxes are closely related to ocean biological processes and various scales of ocean mixing processes. The spatiotemporal variation in the observed APO in the atmosphere is sensitively dependent on the time-space changes of these air-sea fluxes, as well as on atmospheric transport. Because of the significant influence the air-sea flux of O2 has on atmospheric APO, many studies have focused on this issue. It has been shown, for example, that the seasonal cycle in APO is predominantly governed by the summertime O2 emission associated with the primary production in the ocean surface and the wintertime O2 drawdown associated with ocean convection [Keeling et al., 1993; Bender et al., 1996]. Keeling et al. [1998] estimated oceanic O2 fluxes based on the global O2 saturation anomaly map compiled by Najjar and Keeling [1997]and used these fluxes in combination with APO data to optimize the air-sea gas exchange velocity.Balkanski et al. [1999] evaluated the ocean primary production from satellite ocean color data and then estimated the ocean O2fluxes by using a one-dimensional ocean model.Garcia and Keeling [2001] produced climatological monthly O2 fluxes for the global ocean based on monthly anomalies of dissolved O2 and ocean heat flux. Generally, the seasonal cycles in APO predicted by atmospheric transport models driven by estimated O2 fluxes have shown relatively good agreements with the observed APO from the sampling networks operated independently by Princeton University (PU) and the Scripps Institution of Oceanography (SIO).

[4] Annual mean net air-sea O2fluxes are related to large-scale ocean transport processes like equatorial upwelling and the thermohaline circulation.Stephens et al. [1998]used the observed annual mean APO to evaluate large-scale O2 fluxes from global ocean carbon cycle models. Gruber et al. [2001]developed an inversion technique to estimate annual mean net air-sea O2 fluxes on the basis of dissolved oxygen and related tracers in the oceans. These resultant fluxes showed that the O2 uptake occurs generally at mid and high latitudes in both hemispheres, while the O2 outgassing occurs in the tropics. Although these model simulated APO results showed elevated values over the equatorial region, reflecting net outgassing of O2 from the ocean, there were no observational measurements to confirm these model results at the time. Later, however, Tohjima et al. [2005b] and Battle et al. [2006] presented new observations over the equatorial Pacific to confirm the equatorial elevation of APO.

[5] Atmospheric transport is also definitely important in an accurate reconstruction of the seasonality and the annual mean distribution of APO. For example, even if seasonal O2 fluxes with no net flux on an annual basis at every grid are used to drive APO in an atmospheric transport model, the resultant annual mean APO often shows spatial gradients. Such APO gradient is caused by the seasonal covariance between the O2 fluxes and the atmospheric transport, and the effect is generally referred to as the rectifier effect [Denning et al., 1995]. Gruber et al. [2001] concluded that the noted discrepancies between the observed and the model simulated annual mean APO might be attributed mainly to the error in the seasonal O2 rectifier effect. Blaine [2005] employed APO as a tracer to evaluate various atmospheric transport models as part of the Atmospheric Tracer Transport Model Intercomparison Project (TransCom). She found that the predicted seasonality and rectifier effects show significant model dependency.

[6] Comparison between the observed and simulated APO is still one of the main methodologies to evaluate air-sea flux estimation and ocean biogeochemical models [Naegler et al., 2007; Nevison et al., 2008]. In the above studies, the observations have been mostly limited to those from the sampling networks operated by SIO and PU. Therefore, additional new observations would improve the reliability of air-sea flux estimates and the verification of atmospheric transport models.

[7] In this paper, we present an extended data set of the existing O2/N2 and CO2 measurements previously reported by Tohjima et al. [2005b] with a shipboard observation in the Pacific region. Using these shipboard APO data, together with the APO measurements from the two fixed sites in Japan, we investigate the spatiotemporal distribution of APO in the Pacific region. We also present results from an atmospheric transport model simulation of the global APO using climatological O2, N2 and CO2fluxes. First we investigate the global surface patterns of the modeled seasonality and annual mean APO, and discuss the relative contributions of the flux and the atmospheric transport, especially in the context of the seasonal covariation between air-sea flux and atmospheric transport. Then we compare the average seasonality and annual mean values of APO between the observation and the model simulation. Finally, we evaluate the validity of the air-sea fluxes and the atmospheric transport used in this study by comparing the simulated and the observed APO.

2. Flask Sampling and Analytical Methods for O2/N2 and CO2

[8] Air samples for O2/N2 and CO2measurements have been collected at two monitoring stations, Hateruma Island (HAT; lat. 24°03′N, long. 123°48′E) and Cape Ochi-ishi (COI; lat. 43°10′N, long 145°30′E) since July 1997 and December 1999, respectively [Tohjima et al., 2003, 2008]. Additionally, air samples in the northern and western Pacific regions have been collected onboard cargo ships on repeated round-trip cruises between Japan and Canada/the United States and between Japan and Australia/New Zealand since December 2001 [Tohjima et al., 2005b]. The distribution of the flask sampling locations is shown in Figure 1. The sampling locations in the northern North Pacific and eastern North Pacific regions are considerably scattered because ship tracks varied quite often due to ocean conditions and destination changes. Figure S1 in the auxiliary material shows the time and latitude locations of the shipboard flask sampling event. As can be noted, there are some gaps in the sampling time series, which correspond to the periods of untimely shift in the routine function of the cargo ships. The history of the cargo ships used in this study is summarized in Table S1. Flask samplings in the tropical and subtropical eastern North Pacific region, although sparse, were carried out on M/S Pyxis when its destination changed to the east coast of the United States during the period 2003 to 2007.

Figure 1.

Locations where flask samples were collected for O2/N2 and CO2measurements. Solid red and blue squares indicate the locations of Hateruma Island (HAT) and Cape Ochi-ishi (COI), respectively. Red dots represent locations at which flask samples were taken onboard cargo ships. The dots within each bin outlined by rectangle are used in time series analysis. Open circles represent the average positions of the binned data (bin centers). The thick solid and dashed lines show latitudinal and longitudinal transects, respectively, along which the observed and modeled APO values are compared.

[9] Here we briefly describe the sampling and analytical methods, since a detailed explanation has already been presented elsewhere [Tohjima et al., 2008]. Air samples were drawn by diaphragm pumps from the air intakes placed on top of the towers at the monitoring stations and on the bow mast or at the top of the bridge of the cargo ships. After passing through the cold trap (−40°C) to reduce water vapor, air samples were forced into Pyrex glass flasks equipped with two glass stopcocks sealed by Viton® O-rings to a pressure of 0.2 MPa (0.1 MPa above the ambient pressure). Before sample collection, each flask was flushed with dried sample air for at least 20 min at the sampling pressure maintained by a backpressure regulator downstream of the flask. The volume of the flask was 2 L and 2.5 L for the station sampling and onboard sampling, respectively. Note that 1 L flasks were partially used at HAT from July 1997 to March 2006.

[10] Collected air samples were sent back to our laboratory, where O2/N2 ratios were measured by a gas chromatograph equipped with a thermal conductivity detector (GC/TCD) [Tohjima, 2000], and CO2mole fractions were measured by a nondispersive infrared analyzer (Li Cor, LI-6252). Changes in O2/N2 ratio are reported in this study as relative deviations from a reference gas according to

display math

which are expressed in per meg units [Keeling and Shertz, 1992]. A value of 4.8 per meg corresponds to the mole fraction of 1 μmol mol−1 (ppm) in a trace gas abundance. δ(O2/N2) values and CO2 mole fractions were determined against the NIES original O2/N2 scale [Tohjima et al., 2008] and NIES 09 CO2 scale based on the CO2-in-air standard gases prepared by the gravimetric one-step dilution method [Machida et al., 2011]. Note that the O2/N2 ratios of our flask samples have been corrected for the effect of the preferential diffusion of O2through the Viton O-ring seals [Sturm et al., 2004] depending on flask storage period [Tohjima et al., 2008]. The analytical precision evaluated as the standard deviation of the flask replicate was about 5 per meg and 0.05 ppm [Tohjima et al., 2003]. In this paper, we calculated APO by using O2/N2 ratio in per meg, δ(O2/N2), and CO2 mole fraction in ppm, XCO2, according to

display math

where 1.1 represents the –O2:CO2 molar exchange ratio for the land biotic respiration and photosynthesis [Severinghaus, 1995], SO2 is the atmospheric O2 mole fraction in dry air (SO2 = 0.2094 [Tohjima et al., 2005a]), and 1850 is an arbitrary APO reference point.

3. Model Simulation of APO

3.1. Atmospheric Transport Model and APO Fluxes

[11] In order to deepen our understanding of the seasonality and spatial gradient of APO, we simulated the observed distribution of APO by using the NIES global atmospheric tracer transport model (NIES99 TM [Maksyutov and Inoue, 2000]) and a range of surface O2, N2 and CO2fluxes. The NIES99 TM model, running at 15 vertical levels in the sigma coordinates and at a 2.5° × 2.5° horizontal resolution, was forced by the 12-hourly reanalysis data provided by the Japan Meteorological Agency (JMA) Climate Data Assimilation System (JCDAS). Monthly mean climatological daily maximum planetary boundary layer (PBL) heights (from the Data Assimilation Office at NASA's Goddard Space Flight Center) were used in the calculation. The model was run from January 1996 to December 2008 for each simulation, and the simulated results for the 7-year period from January 2002 were used in this study. The 6-year period (1996–2001) was enough to achieve a steady state configuration in long-term transport mechanisms, including the inter-hemispheric exchange.

[12] For the APO simulation, we used air-sea fluxes of O2, N2 and CO2, and CO2 and O2 fluxes from fossil fuel burning, which are summarized in Table 1. As for the oceanic O2 and N2 fluxes, the climatological monthly anomalies of Garcia and Keeling [2001] and Blaine [2005] were used, respectively. These fluxes were computed to simply give the seasonal component, and their annual mean values at every grid point were zero. Additionally, we used the annual mean oceanic O2 and N2fluxes from the annual-mean ocean inversion studies ofGruber et al. [2001] and Gloor et al. [2001], respectively.

Table 1. Flux Data Sets and the Transport Model Used in the APO Simulations
Flux or Transport ModelReference
Fossil fuel CO2Andres et al. [2009]
Fossil fuel O2Andres et al. [2009] and Keeling [1988]
Oceanic CO2 flux (seasonal + annual mean)Takahashi et al. [2002, 2009]
Oceanic seasonal O2Garcia and Keeling [2001]
Oceanic seasonal N2Blaine [2005]
Oceanic annual mean O2Gruber et al. [2001]
Oceanic annual mean N2Gloor et al. [2001]
NIES transport model ver. 99 (NIES99-TM)Maksyutov and Inoue [2000]
Meteorological dataJCDAS25

[13] We used two sets of monthly sea surface CO2 flux climatology of Takahashi et al. [2002] and Takahashi et al. [2009]. Former and latter fluxes are based on about 0.94 and 3 million measurements of surface water pCO2 corrected to reference years of 1995 and 2000, respectively. In general, both data sets show similar oceanic CO2 flux distributions, with a net CO2 outgassing in the equatorial ocean and a net CO2 absorption in the midlatitude ocean. However, large differences exist in the subpolar to polar regions, especially in the Southern Hemisphere, where much less carbon uptake is depicted in the latter study. In this study, we examined how these differences in the oceanic fluxes affect the simulation of the annual mean APO distribution.

[14] Global fossil fuel CO2 emissions with a spatial resolution of 1° × 1° for the year 2006 obtained from the CDIAC database [Andres et al., 2009] were repeatedly used every year for the entire simulation period from 1996 to 2008. The O2 consumption associated with the burning of a fossil fuel type can be calculated from fossil fuel CO2 fluxes and the –O2:C exchange ratio associated with the burning. Using the –O2:C exchange ratio for each fuel type given by Keeling [1988], we calculated the –O2:C exchange ratio for any mix of fossil fuels and cement manufacturing. There are significant differences in the –O2:C exchange ratio for individual countries; for example, a ratio of ∼1.1 for China is lower than that of the global average (∼1.4). Although in previous studies [Stephens et al., 1998; Gruber et al.; 2001; Battle et al., 2006], a global mean ratio was used for APO simulations because the observations used were obtained at locations far from fossil CO2 emissions (and thus the influence of using different –O2:C ratios for individual countries on APO was negligible), many of our flask samples were obtained near the East Asian continental regions, close to fossil fuel emission sources. Thus, to calculate O2 consumption fluxes we used individual country's mean –O2:C exchange ratio for China, Korea, and Japan, and a single global mean ratio for rest of the world.

3.2. Definition of APO Components

[15] The above mentioned individual O2, N2 and CO2 fluxes were separately run in NIES99 TM, and changes in the mole fractions were computed as tracer components, where changes in the total moles of air were ignored. Changes in O2/N2 ratio expressed in per meg units were computed from the simulated O2 and N2 tracers in ppm according to the following equation:

display math

where SN2 represents the atmospheric N2 mole fractions in dry air (SN2 = 0.7809 [Tohjima et al., 2005a]). Thus, we computed APO in per meg units from the individual O2, N2 and CO2 tracers in ppm according to the following equation:

display math

where the superscripts AM, SA, FF, and OC denote the annual mean, seasonal anomaly, fossil fuel and oceanic CO2 fluxes, respectively. To investigate the spatial distribution of the annual mean APO in the following sections, we decomposed the APO values into four components corresponding to the first, second, third and fourth terms of the right hand side of equation (5). Consequently, we can rewrite equation (5) as:

display math

4. Time Series Analysis

[16] We binned the shipboard data into rectangular areas shown in Figure 1. Each bin is represented by the location of the bin center (latitude and longitude), which is taken as the average location of the binned data and is shown as an open circle in Figure 1. We applied a time series analysis technique based on a least squares fitting method and a low-pass digital filter [Thoning et al., 1989] to the binned shipboard data and to the fixed station data. Average seasonal cycles were estimated by the first 2 harmonics, and averaged distributions of the annual means were calculated from the smooth-curve fits. The details of the time series analysis are given in Text S1 in the auxiliary material.

[17] We also applied the same analysis technique to the model simulated APO data. We prepared two sets of simulated APO data by sampling the simulated APO field at the center of each bin with daily interval (denoted as the bin center simulated APO), and at the same locations and at the same times (denoted as the flask sampling simulated APO) as the individual flask samples. In the following, peak-to-peak seasonal amplitude, phasing of the seasonal cycles, and annual means are depicted along the series of bin centers connected with solid lines and broken lines inFigure 1, which are referred to as the ‘latitudinal transect’ and the ‘longitudinal transect’, respectively.

5. Results and Discussion

[18] Time series of APO for the two monitoring stations (HAT and COI) and the 14 bins in the Pacific region are shown in Figure 2a, together with the smooth-curve fits to the data. Similarly, time series of the bin center and flask sampling simulated APO are also plotted inFigure 2b. Note that the data for the 14 bins do not overlap each other. Figure 3shows the observed detrended seasonal variations, along with the average seasonal cycles taken from the least squares fits to the observed data (red lines). In the figure, the average seasonal cycles fitted to the bin center simulated APO (blue lines) and the flask sampling simulated APO (light blue lines) are also drawn. The average seasonal cycles for the two sets of simulated data agree fairly well with each other, indicating that the temporal and spatial variability in the simulated APO is relatively small within the individual binned data. In general, the model simulation reproduces well the spatial differences in the observed seasonal cycles except for bins 43°N-156°W and 24°N-136°W. We first investigated some global features of the model simulated seasonal cycle and annual mean distributions, to help us understand the causes of the APO variability. We then compared the average seasonality and annual mean gradients between the observed and simulated APO. Finally we have speculated on the difference between the observed and the simulated results. In the following model-observation comparison, we used only the bin center simulated APO because we found no large differences in the average seasonal cycle and annual mean distributions between the flask sampling and the bin center simulated APO. See Figures S3 and S4 for the full model-observation comparison, including the results of the flask sampling simulated APO.

Figure 2.

(a) Time series of observed APO for HAT, COI, and selected 14 bins shown in Figure 1. Each black dot represents flask data. Red solid lines indicate the smooth-curve fits to the data (see text). (b) Same as Figure 2a but for simulated APO. Each black dot represents the model-simulated APO at the sampling place and time, and each blue dot represents the daily model simulated APO at the bin center (average position of the binned data).

Figure 3.

Average seasonal cycles of APO from two fixed sites (HAT and COI) and 14 binned shipboard sites. Each dot represents detrended observation. The red curve is a least squares fit to the observed data [Thoning et al., 1989]. The blue and light blue curves are least squares fits to the bin center simulated APO and the flask sampling simulated daily APO, respectively. x axis labels correspond to the 15th day of the month.

5.1. Global Features of the Seasonality Based on the Simulated APO

[19] We extracted average seasonality of the simulated APO at every model grid by using the time series analysis described in section 4. A global distribution of the seasonal peak-to-peak amplitude and the dates of the average seasonal cycle maximum and minimum are shown inFigure 4. General geographical pattern of the seasonal amplitude shown in Figure 4a is similar to that of the previous work [e.g., Keeling et al., 1998; Blaine, 2005]. There are large amplitudes over the ocean within the latitude band between 40° and 70° in both hemispheres, while the amplitudes over the land and the equatorial region are small. Such geographical pattern is easily understandable from the climatological monthly anomaly map of air-sea O2 fluxes (see Plate 3 of Garcia and Keeling [2001], which shows air-sea O2 flux distributions in June and December). In our simulation, the APO seasonal amplitude over the northern North Pacific is the largest (especially northwestern North Pacific), followed by that over the Southern Ocean, while the amplitude over the northern North Atlantic is smaller than that over the northern North Pacific. Although the influence of seasonal oceanic O2fluxes on atmospheric APO seasonal cycle is significant, the role of the atmospheric PBL height and the transport is also important. In particular, the air-sea O2 fluxes in the northern North Pacific and northern North Atlantic show quantitatively similar seasonal cycle [Garcia and Keeling, 2001], suggesting that the atmospheric mixing could play an important role in determining the APO variations over these regions.

Figure 4.

Global distribution of (a) peak-to-peak amplitude, (b) time of maximum, and (c) time of minimum of the averaged seasonality of the simulated APO. Black circles with lines represent the latitudinal and longitudinal transects of the flask observation. The intervals of the black contour lines are 10 per meg (Figure 4a) and 1 month (Figures 4b and 4c).

[20] To investigate the influence of the seasonality in the PBL height on the APO seasonality, we performed an additional model simulation, in which NIES99 TM was run with oceanic seasonal O2 and N2 fluxes, but constant PBL height (no seasonal variation) at each grid point, following a similar strategy employed by Chan et al. [2008]. Monthly PBL heights were averaged over a year to obtain a constant height value at each grid point. APOSA was calculated from the resultant O2 and N2 concentrations according to equations (5) and (6). Global distributions of the seasonal amplitude of APOSA with seasonally varying and with constant PBL height distributions are shown in Figure 5. Note that the seasonal cycle of APOSA predominantly determines the seasonal cycle of APO and the slight difference in the distributions of the seasonal cycle between APO (shown in Figure 4a) and APOSA (shown in Figure 5a) is mostly attributed to the seasonality in APOOC. Compared to the seasonally varying PBL height case (Figure 5a), the seasonal amplitude for the constant PBL height case (Figure 5b) is reduced noticeably over the northern North Pacific (especially northwestern North Pacific) and the northern North Atlantic. However, except for the above two areas, the influence of the seasonal change in the PBL height is relatively small (see Figure 5c). Accordingly, the seasonal covariation between the PBL height and the oceanic O2 flux substantially enlarges the APO seasonal amplitude over the midlatitude ocean in the Northern Hemisphere, especially over the northwestern North Pacific around the Kurile Islands, where about 30% of the APO amplitude can be attributed to the effect of the seasonal covariation between the oceanic O2 flux and the PBL height. It should also be noted that there exists an apparent difference in the seasonal amplitude between the northern North Pacific and the northern North Atlantic shown in Figure 5b, in spite of the quantitatively similar O2 fluxes, suggesting that a relatively weaker atmospheric transport may enhance the seasonal amplitude in the northern North Pacific.

Figure 5.

Global distributions of peak-to-peak amplitude for average seasonal cycle of APOSAwith (a) the seasonally varying PBL height distributions (normal case) and (b) the constant PBL distribution. (c) The difference map for the peak-to-peak amplitude of APOSA shown in Figures 5a and 5b. Black circles with lines represent the latitudinal and longitudinal transects of the flask observation. The intervals of the black contour lines are 10 per meg (Figures 5a and 5b) and 5 per meg (Figure 5c).

[21] Spatial patterns of the phasing of the simulated APO seasonal cycles in terms of the timing of the occurrences of maximum and minimum are shown in Figures 4b and 4c, respectively. The model simulation predicts that the seasonal maximums appear mostly in July and August for the Northern Hemisphere and in January and February for the Southern Hemisphere. The timing of the modeled seasonal minimum, on the other hand, is less variable, occurring mostly in February for the Northern Hemisphere and in August for the Southern Hemisphere. Also, a change in the timing of the seasonal maximum occurs gradually across the equatorial region, while it is relatively discontinuous for the minimum.

[22] Figure 6ashows a time-latitude plot of the zonal mean seasonal variation of APOSA. In Figure 6, the timing of the seasonal maximum and minimum along the latitude corresponds to the ridge line (white lines in Figure 6) and the trough line (white broken line in Figure 6), respectively. As described above, the trough line of the seasonal minimum breaks at the equator while the ridge line of the seasonal maximum is connected at the equator. The oceanic O2fluxes at the mid-to-high latitude (40–70°) act as a predominant driver of the APO seasonal variations in both hemispheres [Garcia and Keeling, 2001]. To examine how the APO seasonal cycle propagates across the equator to the opposite hemisphere, we performed two model simulations of APOSAwith only hemispheric seasonal fluxes. Time-latitude plots of the zonal mean seasonal variation of APOSA driven separately by the Northern and Southern Hemispheric fluxes are shown in Figures 6b and 6c, respectively. The results of the simulations show two distinct features. First, the Southern Hemispheric seasonal variation is larger and, for that reason, shows a more noticeable propagation into the opposite hemisphere (Northern Hemisphere) than the Northern Hemispheric seasonal variation. This is attributed to the fact that the seasonal air-sea O2 exchange in the Southern Hemisphere is larger than in the Northern Hemisphere; the Northern and Southern Hemispheres contribute 36 and 64% of the global seasonal net outgassing (SNO) of O2, respectively [Garcia and Keeling, 2001]. Second, the phase lag of the summertime maximum propagation is larger compared to the wintertime minimum propagation. This different propagation speed between the summertime maximum and the wintertime minimum is consistent with the general feature of the seasonality in the meridional transport: the winter transport is about twice as strong as the summer transport because the equator-to-pole temperature gradient in the winter is on average twice as large as that in the summer [Peixoto and Oort, 1992]. Investigating the global-scale CO2 transport processes using an atmospheric transport model, Miyazaki et al. [2008]indicated that the mean-meridional advection effectively carries the air from the extratropics to the tropics in the lower troposphere of the winter hemisphere. Thus, these two features, namely the larger seasonal cycle in the Southern Hemisphere and the seasonal variation in the meridional transport speed (faster in the winter than in the summer), may explain the asymmetric distribution pattern of the timing of the seasonal maximum and minimum.

Figure 6.

Time-latitude plots of the average seasonal cycle for APOSA with (a) full oceanic fluxes, (b) Northern Hemisphere ocean fluxes, and (c) Southern Hemisphere ocean fluxes. To show the seasonal cycle clearly, the first 6 months of the seasonal cycle are repeated. The intervals of the black contour lines are 2 per meg. x axis labels correspond to the 15th day of the month.

5.2. Global Features of the Annual Mean Gradient Based on the Simulated APO

[23] A global distribution of the annual mean APO based on the 2002–2008 simulation is shown in Figure 7a, together with the four APO components, APOAM, APOSA, APOFF, and APOOC (Figures 7b–7e). Elevated APO values are located around the tropical Pacific and Atlantic regions. Other high values are also detected in the northern North Pacific, where the highest values are located to the southeast of the Kamchatka Peninsula, and in the southern South Atlantic. Low APO values are predicted over the Southern Ocean, especially over the Ross Sea and the Weddell Sea, over Western Europe and the east coast of the United States. Among these characteristic distributions, the equatorial peaks and the troughs over the Southern Ocean are mostly due to the annual mean distribution of APOAM (Figure 7b). The high APO value over the northern North Pacific is attributed to APOSA (Figure 7c). The low APO values centered on Western Europe and the east coast of the United States are caused by fossil fuel burning with relatively higher –O2:C exchange ratios than 1.1 shown in the global distribution of APOFF (Figure 7d). The APOOC distribution of the equatorial elevation centered on the Pacific and the poleward decrease in both hemispheres (Figure 7e) also contribute substantially to the equatorial APO elevation.

Figure 7.

Averaged global distribution of annual mean simulations: (a) total APO with the oceanic CO2 fluxes of Takahashi et al. [2009] (referred to as T09), (b) APOAM, (c) APOSA, (d) APOFF, (e) APOOC with the oceanic CO2 fluxes of Takahashi et al. [2009], (f) APOAM + APOSA, (g) APOFF with constant –O2:C exchange ratio of 1.4, (h) APOOC with the oceanic CO2 fluxes of Takahashi et al. [2002] (referred to as T02), and (i) total APO with the oceanic CO2 fluxes of Takahashi et al. [2002]. The units of the legend are in per meg and the intervals of the black contour lines are (except Figures 7e and 7h) 2 per meg and 1 per meg (Figures 7e and 7h).

[24] Although the fluxes adopted in the APOSA simulation were annually neutral at any one grid point, the APOSA map shows significant spatial gradients. These gradients in the annual mean APOSA are caused by the seasonal rectifier effect. In the oceanic subpolar region, the seasonal O2 outgassing, together with a relatively weak atmospheric transport and shallow PBL in the summer, results in a relatively high surface APO while the ingassing, together with a relatively strong transport and deep PBL depth in the winter, dilutes the influence of the seasonal O2 ingassing [Stephens et al., 1998; Gruber et al., 2001]. Consequently, these seasonal rectifier processes cause the elevation in the annual mean APO in the oceanic subpolar regions. In contrast, there is an APOSA trough in the equatorial areas. Similar APOSA troughs are generated by several different atmospheric tracer transport models driven by the same seasonal O2 and N2 fluxes as those used in this study using the TransCom APO protocol [Blaine, 2005]. This negative rectifier effect is not likely to be explained by the covariance between air-sea O2 flux and atmospheric transport because of the rather low seasonality in the tropical oceanic O2 flux. The Intertropical Convergence Zone (ITCZ) migrates seasonally across the equator, generally locating in the summer hemisphere. In the equatorial area, surface winds blow toward ITCZ. Therefore, the equatorial region tends to be influenced by the air in the winter hemisphere throughout the year. This phenomenon may be related to the faster meridional propagation of the APO seasonal minimum toward the opposite hemisphere, as discussed in section 5.1 (Figure 6). Such ITCZ rectifier effect may result in relatively low APO values around the equator.

[25] To separately evaluate the contributions of the atmospheric transport and the PBL height to the seasonal rectifier effect, we analyzed APOSA simulated with a constant PBL height distribution, as discussed in the previous section. Global distributions of the annual mean APOSA obtained with the seasonally varying PBL height and with the constant PBL height, along with the difference between these annual means are shown in Figures 8a–8c, respectively. Here, it should be noted that the annual mean APOSA with the constant PBL height (Figure 8b) represents the contribution of the rectifier effect caused by the seasonal atmospheric transport, and the difference in the annual mean APOSA (Figure 8c) represents the rectifier effect caused by the seasonally varying PBL height. The distribution pattern of the annual mean APOSA for the constant PBL height (Figure 8b) is basically similar to that of the varying PBL height case (Figure 8a), suggesting that the atmospheric transport is the dominant contributor to the rectifier effect. The equatorial trough is scarcely impacted by the seasonally varying PBL height, which is consistent with the above proposed mechanism of the ITCZ rectifier effect. Taking into account of the latitudinal distribution of zonal mean APOSA (not shown), we find that the rectifier effect related to the atmospheric transport and the PBL height accounts for about 70% and 30% of the elevated APOSAaround the mid-latitudinal area, respectively. However, in the northwestern North Pacific, the contribution of the rectifier effect related to the PBL height increases to about 50% of the total rectification, resulting in the most elevated observed annual mean APO already noted.

Figure 8.

Global distributions of annual mean APOSA with (a) the seasonally varying PBL height distributions (normal case) and (b) the constant PBL distribution. (c) The difference map for the annual mean APOSA shown in Figures 8a and 8b. Black circles with lines represent the latitudinal and longitudinal transects of the flask observation. The intervals of the black contour lines are 3 per meg.

[26] To understand the contribution of net oceanic O2 fluxes (annual and seasonal) to the APO distribution, we investigated the combined distribution of APOAM and APOSA components, which is shown in Figure 7f. In the distribution of APOAM + APOSA, the elevated APO value over the northern North Pacific region is remarkable. In addition, the equatorial peaks in the combined map of APOAM + APOSA are substantially suppressed because the equatorial peaks in APOAMare essentially counter-balanced by the equatorial APOSAtroughs. As a result, if the rectifier effect expressed in the NIES99 TM is valid, the contribution of the air-sea CO2flux to the equatorial peaks in the total APO distribution is comparable to that of the air-sea O2 flux in the NIES99 TM simulation.

[27] Figure 7g shows a global map of the annual mean APOFF, in which a single –O2:C ratio for the global average fossil fuel combustion (∼1.4) was used to calculate the O2 consumption fluxes. There is a region showing lower APOFF over East China in Figure 7g, compared to Figure 7d, because the –O2:C exchange ratio of about 1.1 was used for China to calculate APOFF in Figure 7d. The differences in APOFF between the two maps, reaching 10 per meg over East China, are less than 2 per meg at our flask sampling sites in the model simulation. The maps of the annual mean APOOCand APO based on the air-sea CO2 fluxes of Takahashi et al. [2002] are shown in Figures 7h and 7i, respectively. The oceanic CO2 fluxes of Takahashi et al. [2002] produce a more elevated equatorial APOOC in the Pacific and steeper poleward APOOC gradients in both hemispheres, in comparison with the results obtained by using the Takahashi et al. [2009] flux data (Figure 7e). These differences in the APOOC maps are also reflected in the total APO distribution maps shown in Figures 7a and 7i. In the following sections, we will compare averaged distributions of the seasonal cycle and the annual mean along the latitudinal and longitudinal transects (see Figure 1) between the observation and the model simulation.

5.3. Comparison of Average Seasonal Cycle Between Observation and Model

[28] The seasonal peak-to-peak amplitude in the observed APO along the latitudinal and longitudinal transects (Figure 1) is depicted in Figures 9a and 9b, respectively, as red symbols. The latitudinal distribution shows a poleward increase in both hemispheres, with a minima of about 20 per meg around the equator. The poleward gradient is steeper in the Northern Hemisphere than in the Southern Hemisphere. The seasonal amplitude along the longitudinal transect shows an eastward increase between 120°E and 160°E and an eastward decrease between 160°E and 100°W. The largest seasonal amplitude of about 110 per meg is observed at the north end of the latitudinal transect (50°N-163°E bin) in the northwestern North Pacific, corresponding to the peak in the longitudinal distribution. The seasonal amplitude of 110 per meg is larger than the seasonal amplitude of about 70 per meg observed at SIO Cold Bay, Alaska site (55.2°N, 162.72°W) [e.g.,Blaine, 2005; Battle et al., 2006].

Figure 9.

(a, b) Peak-to-peak amplitudes and (b, d) timing of the seasonal maximum and minimum along the latitudinal (Figures 9a and 9c) and longitudinal (Figures 9b and 9d) transects shown as solid and dashed thick lines inFigure 1, respectively. Colors represent: red = flask observations, blue = bin center simulated APO. Black lines in Figures 9c and 9d represent the time of the seasonal maximum and minimum for the bin center simulated APO with the seasonal oceanic O2flux delayed by one month. (Full observation-model comparison including the flask sampling simulated APO is shown inFigure S3.)

[29] The timing of the seasonal maximum and minimum along the latitudinal and longitudinal transects are depicted as red symbols in Figures 9c and 9d, respectively. The seasonal minimum occurs in March and September in the Northern and Southern Hemispheres, respectively, and the latitudinal distribution of the timing shows a discontinuous change at the equator. Contrary to the seasonal minimum, the time of occurrence of the seasonal maximum along the latitudinal transect changes smoothly across the equator from March at the south end to July at 15°N and levels off to the north of 15°N. The variations in the timing of the seasonal maximum and minimum along the longitudinal transect are small.

[30] The distribution of the seasonal amplitude and the timing of the seasonal maximum and minimum calculated from the bin center simulated APO (blue squares) along the latitudinal and longitudinal transects are also plotted in Figures 9a–9d. The model simulation, in general, reproduces the salient characteristics of the observed seasonal amplitude. Especially, good agreements on the large seasonal amplitudes in the northwestern North Pacific give some confidence to the ability of NIES99 TM to simulate the rectifier effect over the region. However, the model simulation overestimates the amplitude except at bins 5°S-153°E, 0°-151°E, and 14°N-101°W, and the average amplitude ratio of the simulation to the observation is 1.14 ± 0.22. The TransCom APO exercise also showed a general overestimation of the simulated seasonal amplitude, with an averaged amplitude ratio of 1.17 [Blaine, 2005]. Assuming the atmospheric transport model used in this study is accurate, these results suggest that the air-sea seasonal flux anomaly of O2 of Garcia and Keeling [2001] might be overestimated by about 15%.

[31] Although the predicted phases are shifted early by about one month, there are excellent agreements on the patterns of the latitudinal profiles for the maximum and minimum between the observation and the model (Figure 9c). These results give some confidence in the model representation of the processes responsible for the propagation of the seasonal APO signals toward the opposite hemispheres as described in section 5.1.

[32] The observation-minus-model differences in the timing of the seasonal maximum/minimum are about 4 weeks on average, except for the seasonal maximum in the Northern Hemisphere, where the observed and simulated seasonal maxima are almost in phase. Similar advanced phases of simulated seasonal cycles were reported in the TransCom APO exercise [Blaine, 2005]. Garcia and Keeling [2001]pointed out that the phase of their estimated air-sea seasonal O2 flux is expected to be advanced by a few weeks because their calculation did not include the mixed layer equilibration process for O2.

[33] Additional model simulations indicated that the time shift in the seasonal oceanic O2 flux causes a proportional time shift in the APO seasonal cycle, and that the seasonal variations in the PBL height do not significantly affect the phasing of APO. (Text S2 provides the details of the additional model simulations.) These results might indicate that the timing of the seasonal maximum of the oceanic O2 fluxes of Garcia and Keeling [2001]should not necessarily be changed in the western North Pacific where our model-to-observation comparison was made.

5.4. Comparison of Annual Mean APO Distribution Between Observation and Model

[34] Distributions of the annual mean APO along the latitudinal and longitudinal transects are plotted as red symbols in Figures 10a and 10b, respectively. The observed annual mean APO again confirms the equatorial bulge [Tohjima et al., 2005b; Battle et al., 2006] with gentle and steep poleward decreases in the Southern and Northern Hemispheres, respectively. The latitudinal profile shows a deep trough at 36°N and then an abrupt increase to the north of 40°N. The longitudinal profile shows an eastward increase between 120°E and 160°E and a decrease between 160°E and 100°W with a slight elevation around 135°W. Therefore, we now have an identifiable area with an elevated annual mean APO in the northwestern North Pacific, in addition to the equatorial region.

Figure 10.

Averaged distribution of the annual mean APO for the 7-year period from 2002 to 2008 along the (a) latitudinal and (b) longitudinal transects shown inFigure 1. Colors represent: red = flask observation, and green and blue = bin center simulated APO with the oceanic CO2 fluxes of Takahashi et al. [2002] and Takahashi et al. [2009], respectively. Light blue broken lines represent the bin center simulated APO with doubled annual mean air-to-sea O2flux in a rectangular region (30–50°N, 120–180°E) (see text). (Full observation-model comparison including the flask sampling simulated APO is shown in Figure S4.)

[35] Simulated annual mean APO values using the ocean CO2 emissions of Takahashi et al. [2002] (green) and Takahashi et al. [2009] (blue), and other fluxes as listed in Table 1 are also plotted along the latitudinal and longitudinal transects in Figures 10a and 10b, respectively. In the figures, the simulated APO distributions are shifted so that the values for the equatorial bin match the observation.

[36] It is obvious that the observed latitudinal profile in the Southern Hemisphere is better reproduced by the simulation using the Takahashi et al. [2009] flux data than the Takahashi et al. [2002] flux data. The difference between the equator and 34°S for the simulation based on Takahashi et al. [2009] and Takahashi et al. [2002] is about 4 per meg and about 8 per meg, respectively, with the former value agreeing with the observation. Additionally, the observed APO increase to the north of 40°N is also reproduced by the simulation. The simulated longitudinal profiles, showing higher APO values than the observation, trace fairly well the observed eastward increase and decrease. The most obvious disagreement in the latitudinal profiles occurs between 20°N and 40°N, where the observation shows more decrease than the model.

[37] To examine the contribution of each APO component to the net simulated APO distribution, the four components (APOAM, APOSA, APOFF, and APOOC) of the bin center simulated APO along the latitudinal and longitudinal transects are depicted in Figure 11. As described in section 5.2., the equatorial APOAM bulge is compensated by the equatorial trough of APOSA; the contributions of the oceanic O2 and CO2 components (APOAM + APOSA and APOOC) to the equatorial APO bulge are comparable in magnitude. On the other hand, the poleward increase in APOSA exceeds the poleward decreases in APOAM, APOFF, and APOOC to the north of 40°N, causing a net poleward increase in APO in the northwestern North Pacific. In addition, it is clear that the pattern of the longitudinal APO profile is mainly influenced by APOSA, which shows a “Λ-shaped” distribution with an elevated peak at 163°E.

Figure 11.

Components of the annual mean APO along (a) latitudinal transect and (b) longitudinal transect shown in Figure 1. Red squares are the net APO; blue circles = APOAM; purple circles = APOSA; light blue circles = APOAM + APOSA; orange circles = APOFF; and black circles = APOOC. The open symbols and closed symbols for APO and APOOC are based on the oceanic CO2 flux from Takahashi et al. [2002] (referred to as T02) and Takahashi et al. [2009] (referred to as T09), respectively. These APO components are based on the bin center simulated APO.

[38] Based on the above discussion, we can make the following conclusions. First, the ocean CO2 fluxes in the Southern Hemisphere of Takahashi et al. [2009] seem to be more realistic than those of Takahashi et al. [2002]. Second, the agreement between the model and observation on the steep increase to the north of 40°N and the “Λ-shaped” pattern of the longitudinal profiles seems to confirm the existence of the rectifier effect in the northern North Pacific, giving some confidence in the ability of NIES99 TM to mimic the rectifier effect, at least in the northern North Pacific.

[39] The decrease in the observed APO between 20°N and 40°N seems to suggest the existence of an additional APO sink in the latitude zone. The maps of the annual O2 and CO2 fluxes of Gruber et al. [2001] and Takahashi et al. [2009]indicate the existence of strong air-to-sea O2 and CO2 fluxes in the mid latitude of the western North Pacific, which corresponds to the western boundary current (Kuroshio Current) region of the North Pacific subtropical gyre where O2 and CO2 are absorbed due to rapid cooling and deeper mixing [e.g., McKinley et al., 2003]. A stronger absorption of O2 and/or CO2in the region could explain the discrepancy in the annual mean APO latitudinal profile between the observation and the simulation. In order to conduct a preliminary evaluation of the sensitivity of the APO sink flux to the magnitude of the APO trough, we doubled the annual mean air-to-sea O2 flux of Gruber et al. [2001] in a rectangular region (30–50°N, 120–180°E) and performed a model simulation with the modified annual mean O2 and other fluxes as listed in Table 1. The annual O2 flux for the rectangular region was changed from −33 Tmol yr−1 to −67 Tmol yr−1. The resultant latitudinal and longitudinal distributions of the bin center simulated APO are depicted as light blue broken lines in Figure 10. The latitudinal profile with modified flux shows a deepened trough around 40°N, which is fairly comparable to the observed trough. Accordingly, about −30 Tmol yr−1 of APO flux might be missing in the midlatitude western North Pacific.

6. Conclusions

[40] In this study, we have presented APO measurements in various regions of the Pacific Ocean obtained from flask samples taken at the monitoring stations HAT and COI and on the cargo ships sailing between Japan and U.S./Canada and Australia/New Zealand. We also conducted various model sensitivity experiments to investigate reasons for the observed spatiotemporal distribution of APO.

[41] The characteristic distribution patterns of the seasonality and the annual means of the observed APO are as follows: The seasonal amplitude along the latitudinal transect shows a poleward increase with a minimum of about 20 per meg around the equator. The seasonal amplitude along the longitudinal transect shows an eastward increase between 120°E and 160°E, and an eastward decrease between 160°E and 100°W. The largest seasonal amplitude of about 110 per meg is observed at the north end of the latitudinal transect (bin of 50°N-163°E) in the northwestern Northern Pacific. The seasonal minimum occurs in September in the Southern Hemisphere and in March in the Northern Hemisphere, with their latitudinal distribution showing a discontinuous change at the equator. On the other hand, the seasonal maximum along the latitudinal transect changes smoothly across the equator from March at the south end to July at the north end. The latitudinal profile of the annual mean APO re-confirms the equatorial bulge [Tohjima et al., 2005b; Battle et al., 2006], and shows a steep trough in value in the latitude zone of 20–40°N and an abrupt increase to the north of 40°N. This steep trough may be a longitude-specific feature tied to the Kuroshio Current. The longitudinal profile of the annual mean APO also shows an elevated value over the northwestern North Pacific.

[42] The model simulation generally reproduces well the observed distribution patterns of the seasonal amplitude along the latitudinal and longitudinal transects. There is specially a good agreement in the significantly elevated seasonal amplitude in the northwestern North Pacific. In the model simulation, the seasonal covariation between the PBL height and the oceanic O2 flux contributes about 30% of the seasonal amplitude in the northwestern North Pacific. The model also does well in simulating the discontinuous and continuous latitudinal distribution in the timing of the seasonal minimum and maximum, respectively. These asymmetric latitudinal distributions may be attributed to the seasonal variation in the meridional transport strength: stronger in the winter hemisphere than in the summer hemisphere, and the larger seasonal cycle in the Southern Hemisphere.

[43] In spite of the general agreement between the observed and simulated seasonality however, the simulated amplitudes are about 15% larger on average than the observation. The model overestimation of the seasonal amplitude was also reported in TransCom APO [Blaine, 2005]. The phasing of the simulated seasonality is advanced relative to the observation by about 4 weeks on average, except for the seasonal maximum in the Northern Hemisphere. Similar phase lag in the seasonality was also reported in previous studies [Garcia and Keeling, 2001; Blaine, 2005]. These results suggest that a modification in the timing of the O2 flux maximum and minimum, except for the O2 flux maximum in the North Pacific, reported by Garcia and Keeling [2001]would lead to an improvement in the model-observations comparison.

[44] The annual mean APO distributions in the Pacific region are also generally well reproduced by the model. An especially good agreement with the observation in the elevated annual mean over the northwestern North Pacific, combined with a good agreement in the enhanced seasonal amplitude, suggests that the rectifier effect (seasonal covariation between oceanic O2 flux and atmospheric transport) significantly influences accurate reconstruction of the APO spatiotemporal variations.

[45] The latitudinal gradient of the annual mean APO between the equator and the latitude of 34°S is found to be better reproduced by the model based on the oceanic CO2 flux of Takahashi et al. [2009] than that of Takahashi et al. [2002]. These results seem to support the validity of the revised net annual air-to-sea CO2 flux for the Southern Ocean as reported in Takahashi et al. [2009]. In the Northern Hemisphere between 20°N and 40°N, the observed APO shows a steeper decrease than the model. The discrepancy is about 10 per meg at 34°N. Such depressed APO annual mean is also seen in the longitudinal profile. These results suggest the existence of an APO sink (oceanic O2 and/or CO2 sinks) at the corresponding latitudes. A model simulation performed with the doubled annual mean oceanic O2 flux (corresponding to an additional O2 flux of −33 Tmol yr−1) in the rectangular region (30–50°N, 120–180°E) results in a better simulation of the magnitude of the observed low APO values in the latitude zone of 20–40°N. Therefore, an APO sink of about −30 Tmol yr−1 might exist around the midlatitude in the western North Pacific.

Acknowledgments

[46] We thank Shigeru Kariya, Tomoko Nojiri, Tomoyasu Yamada, Nobukazu Oda, Fujio Shimano and other staff of the Global Environmental Forum (GEF), and the owners, operators and crew of the voluntary ships, MOL Golden Wattle, Mol Glory, Fujitrans World, Trans Future 5, Pyxis, and Skaubryn, for their support in collecting air samples. Thanks are also expressed to Keiichi Katsumata, Hisayo Sandanbata, and Yoko Kajita for their assistance in the CO2measurements and flask preparations. We also grateful to Manuel Gloor and Shamil Maksyutov for providing us the climatological ocean fluxes for our model simulation, as well as to Ralph Keeling for valuable comments. We thank one anonymous reviewer and Andrew Manning, whose comments considerably helped to improve the manuscript. The data sets used for model simulation of this study are provided from the cooperative research project of the JRA-25 long-term reanalysis by the Japan Meteorological Agency (JMA) and the Central Research Institute of Electric Power Industry (CRIEPI). This work has been financially supported by the Ministry of the Environment through the Global Environment Research Account for National Institute (E0450 and E0955).

Ancillary