Observed APO Seasonal Cycle in the Pacific: Estimation of Autumn O₂ Oceanic Emissions

Abstract

In this work, we investigated the seasonal cycle of atmospheric potential oxygen (APO), a unique tracer of air-sea gas exchanges of molecular oxygen (O₂) and carbon dioxide (CO₂), expressed as APO = O₂ + 1.1 × CO₂. APO data were obtained from flask air samples collected since the late 1990s at three Japanese ground stations and on commercial cargo ships sailing between Japan and Australia/New Zealand, North America, and Southeast Asia. We also analyzed the APO spatial distribution and seasonal cycles with simulations from an atmospheric transport model using climatological oceanic O₂ fluxes from an empirical product that relate O₂ flux to ocean heat as input. Model simulations reproduced the observed APO seasonal cycles generally well, but with larger amplitudes and earlier occurrence of seasonal minima and maxima than in the observations. Moreover, the observed seasonal cycles exhibited larger APO enhancements than the simulations in autumn and early winter, especially in the North Pacific at 20°N–60°N. These enhancements remained when refining the comparison by adjusting the simulated APO peak-to-peak amplitudes and seasonal phases to the observations. This suggests additional O₂ emissions in the North Pacific, not well expressed in the air-sea O₂ fluxes used as input for our model simulations. The average autumn enhancement at 40°N–60°N was approximately twice that measured at 20°N–40°N. Confirming previous studies, our results indicate two distinct mechanisms possibly contributing to the additional oceanic O₂ emissions: outgassing from a subsurface shallow oxygen maximum at 20°N–40°N and autumn phytoplankton bloom at 40°N–60°N.

Key Points

• We analyzed atmospheric potential oxygen (APO), a tracer of air-sea exchanges of molecular oxygen (O₂) and carbon dioxide, in the Pacific

• The observed seasonal APO enhancement at 20°N–60°N in autumn, relative to model simulations, suggests additional oceanic O₂ emissions

• Outgassing from a shallow oxygen maximum (20–40°N) and an autumn phytoplankton bloom (40–60°N) likely contribute to this enhancement

Plain Language Summary

Seasonal variations of air-sea molecular oxygen (O₂) exchanges are mainly driven by spring/summer emission and autumn/winter absorption. Emission results mainly from O₂ production by phytoplankton at the ocean surface; absorption is related to O₂ uptake in deep oxygen-depleted water layers caused by oceanic ventilation. A phytoplankton bloom in autumn and outgassing from water rich in dissolved O₂ just below the mixed layer, known as the shallow oxygen maximum, have been suggested as possible mechanisms contributing to the O₂ seasonal cycles. Unfortunately, these oceanic autumn O₂ emissions are not well characterized. In this study, we analyzed air samples collected in the western and northern Pacific to evaluate the oceanic component of the O₂ seasonal cycle, known as “atmospheric potential oxygen” and calculated as the sum of atmospheric O₂ and carbon dioxide. By comparing these observations with model simulations that used previous air-sea O₂ flux estimates as input, we identified marked enhancements in autumn at latitudes between 20°N and 60°N in the Pacific. This corresponds to the geographic area where the autumn phytoplankton bloom and the shallow oxygen maximum occur. This result is a strong indication of additional autumn O₂ emissions in the extratropical North Pacific.

1 Introduction

Atmospheric molecular oxygen (O₂) is involved in most natural and anthropogenic processes at the Earth's surface. Increasingly precise techniques, developed to detect spatiotemporal variations of the atmospheric O₂ mole fraction (Bender et al., 1994; Keeling, 1988; Manning et al., 1999; Stephens et al., 2003, 2007; Tohjima, 2000), have demonstrated that measuring these variations is critical to understand O₂-related biogeochemical processes. In particular, joint observations of long-term atmospheric O₂ and carbon dioxide (CO₂) variations have been essential to assess the partitioning of fossil-fuel-derived CO₂ sequestration between the terrestrial biosphere and the ocean (Battle et al., 2000, 2006; Goto et al., 2017; Ishidoya, Morimoto, et al., 2012; Keeling & Shertz, 1992; Manning & Keeling, 2006; Tohjima et al., 2019). Since the early studies, numerous campaigns for concurrent measurements of atmospheric O₂ and CO₂ have been conducted from diverse platforms, including ground stations (e.g., Heimann et al., 2022; Manning & Keeling, 2006; Tohjima et al., 2003), ships (e.g., Ishidoya et al., 2016; Pickers et al., 2017; Tohjima, Mukai, et al., 2005), and aircraft (e.g., Ishidoya, Aoki, et al., 2012; Ishidoya et al., 2022; Morgan et al., 2019; Stephens et al., 2021).

In addition to global carbon budget estimations, atmospheric O₂ variations can be used to study air-sea gas exchanges, notably when introducing the atmospheric potential oxygen (APO), a tracer defined as APO = O₂ + α_B × CO₂ (Stephens et al., 1998). In this definition, α_B represents the −O₂:CO₂ exchange ratio for terrestrial biotic processes, including respiration and photosynthesis; its value is usually set to 1.1 (Severinghaus, 1995). By definition, APO is not sensitive to gas exchanges between the atmosphere and the terrestrial biosphere, but mostly reflects air-sea O₂ and CO₂ exchanges, which are influenced mainly by oceanic biological activity, oceanic circulation, and air-sea heat exchange. Therefore, APO spatiotemporal variations can be used to constrain such processes, as demonstrated in previous studies (e.g., Eddebbar et al., 2017; Hamme & Keeling, 2008; Nevison et al., 2012; Resplandy et al., 2016).

The primary APO temporal variation is a seasonal cycle with an increase in spring and summer and a decrease in autumn and winter, mainly driven by air-sea O₂ exchanges related to oceanic biological activity and dynamics (e.g., Keeling et al., 1993). In spring and summer, organic compounds and O₂ are generated from dissolved CO₂ and nutrients through photosynthesis at the ocean surface. The produced O₂ is almost immediately released into the atmosphere because of its limited solubility in seawater, while the organic compounds slowly sink to the deep ocean below 100 m where they are mineralized. In autumn and winter, surface cooling destabilizes the oceanic vertical stratification by enhancing the vertical mixing between deeper water layers, depleted in O₂ but rich in nutrients and CO₂, and the surface water layer. In turn, this enhanced vertical mixing reduces the surface O₂ partial pressure, resulting in the absorption of atmospheric O₂ into the ocean. This biological air-sea O₂ exchange is enhanced by seasonal sea-surface temperature (SST) variations: the high summer SST reduces O₂ solubility and enhances O₂ emissions into the atmosphere, whereas the low winter SST enhances O₂ solubility and reduces emissions. Conversely, seasonal variations of air-sea CO₂ exchanges are strongly limited by the chemical equilibrium of CO₂ in seawater, and biologically induced air-sea CO₂ fluxes, inversely proportional to biologically induced O₂ fluxes, are compensated by thermally induced CO₂ fluxes (Keeling et al., 1993).

Garcia and Keeling (2001) constructed a climatology of monthly air-sea O₂ fluxes (hereafter “GK2001 O₂ fluxes”) on the basis of the linear relationship between ocean heat flux anomalies and air-sea O₂ fluxes derived from sea-surface measurements of dissolved O₂ abundance, temperature, and salinity in global ocean basins. They also compared simulated APO distributions, calculated with the atmospheric transport model of Heiman (1995) using the GK2001 O₂ fluxes as input, with APO estimates derived from observations. Their results demonstrated that the APO seasonal cycle is generally well reproduced by the GK2001 O₂ fluxes, though slight differences in the peak-to-peak amplitudes and phases were noticed. Ishidoya et al. (2016) indicated discrepancies between the APO seasonal cycles observed in the Northern Hemisphere and those simulated from the GK2001 O₂ fluxes, with observed values often larger than the simulations in autumn and winter. Consequently, they postulated the existence of additional autumn O₂ oceanic emissions not fully accounted for in the GK2001 O₂ fluxes. Recently, Jin et al. (2023) evaluated seasonal variations in the global APO inventory using O₂ and CO₂ measurements from several airborne campaigns conducted between 2009 and 2018. They identified interhemispheric differences in the air-sea O₂ flux, with a longer enhancement period, extending into autumn, in the Northern Hemisphere than in the Southern Hemisphere. Their results also indicated additional O₂ emissions in autumn that the GK2001 O₂ fluxes do not express well.

A tentative mechanism for such emissions is O₂ outgassing from phytoplankton blooms. Phytoplankton bloom occurs in the midlatitude ocean in autumn, when ocean cooling induces a mixed layer depth (MLD) increase, thus replenishing the upper oligotrophic layer with nutrients before insolation decreases (e.g., Sapiano et al., 2012). Another tentative O₂ emission mechanism is outgassing from the oceanic subsurface layer, rich in dissolved O₂, associated with the increasing MLD in autumn. When O₂ is produced by the spring and summer surface-layer phytoplankton blooms, O₂ in the upper layer is rapidly released to the atmosphere; conversely, strong stratification of the seawater column in summer prevents mixing of the lower-layer O₂ with the upper layer, resulting in the formation of an O₂-rich subsurface layer known as the shallow oxygen maximum (SOM; Shulenberger & Reid, 1981). By allowing to discriminate between these hypotheses, the investigation of APO seasonal cycles derived from observations should improve our understanding of oceanic biogeochemical processes.

In this study, we derived APO from O₂ and CO₂ mole fractions measured in flask air samples collected from remote sites in Japan and from cargo ships sailing in the northern and western Pacific. Analyzing these measurements, we characterized the APO spatial distribution and temporal variations in the Pacific, in particular its average seasonal cycle. We also simulated the APO seasonal cycle with an atmospheric transport model and several sets of O₂ and CO₂ climatological fluxes, including the GK2001 O₂ fluxes. Then, by comparing the observed and simulated seasonal cycles, we confirmed the detection of APO enhancements in autumn, relative to the simulated values, and analyzed their spatial distribution in the Pacific. The manuscript is organized as follows: Section 2 describes the data and methods: description of the flask samples (Section 2.1), analytical method to determine the O₂ and CO₂ mole fractions (Section 2.2), numerical method to extract seasonal cycles from APO time series (Section 2.3), and APO model simulations (Section 2.4). Section 3 reports and discusses the following results: observed and simulated APO seasonal cycles (Section 3.1), the influence of a 2-week delay of the O₂ flux on the simulated APO seasonal cycle (Section 3.2), observational confirmation of autumn APO enhancements (Section 3.3) and possible enhancement mechanisms (Section 3.4). Finally, we characterize the relationship between additional autumn O₂ emissions and APO enhancement using model simulations (Section 3.5). Section 4 summarizes our results and concludes this study.

2 Data and Analysis

2.1 Flask Sampling From Ground Stations and Cargo Ships

We have been collecting air samples into glass flasks for atmospheric O₂ and CO₂ measurements in the Asia-Pacific region (Figure 1) since the late 1990s. Initially, we started routine flask sampling at two ground stations: Hateruma Island (HAT; 24.06°N, 123.81°E) in July 1997, and Cape Ochiishi (COI; 43.16°N, 145.50°E) in December 1999. To extend the spatial coverage, we started collecting air samples on commercial cargo ships sailing regularly on roundtrip cruises between Japan and North America (“North America route”) and between Japan and Australia/New Zealand (“Oceania route”) in December 2001. The flask sampling methodology and experimental setup for HAT, COI, and cargo ships sailing on the North America and Oceania routes were detailed in Tohjima et al. (2008, 2012). Briefly, for the ground station measurements, air samples, collected every 4 days, are drawn by a diaphragm pump through an intake placed at 47 and 94 m (above sea level) at HAT and COI, respectively, and introduced into a cold trap (−40°C) to remove water vapor. Then, each dried air sample is pressurized to 0.2 MPa (0.1 MPa above ambient pressure) into a 2-L glass flask. For the onboard measurements, air samples are drawn by a metal bellows pump through an intake placed at the top of the bridge (about 30 m above sea level) and dried by passing through a cold trap (−40°C). Then, each sample is pressurized to 0.2 MPa into a 2.5-L glass flask. During each roundtrip cruise, we usually collected 7 or 14 flask samples on the North America route and 21 flask samples on the Oceania route. The locations for the onboard flask sampling were usually determined by a predesignated set of longitudes on the North America route or that of latitudes on the Oceania route.

Figure 1

Spatiotemporal coverage of the flask air samples analyzed in this study. (a) Sampling locations at the Japanese ground stations (blue squares): Hateruma Island (HAT), Cape Ochiishi (COI), and Minamitorishima Island (MNM), and onboard commercial cargo ships (dots); (b) latitudinal and temporal coverage of the onboard samplings over the study period. Ship sampling locations in both panels are color-coded by longitude: gray (<130°E), red (130°E–180°E), and blue (<180°W). The pink-shaded grids in Panel (a) identify the 41 bins (10° × 10° in latitude/longitude) selected for seasonal cycle analyses.

We later expanded the geographic sampling coverage to a third Japanese ground station on Minamitorishima Island (MNM; 24.28°N, 153.98°E) in September 2011 after including a third maritime route between Japan and Southeast Asia (“Southeast Asia route”), in September 2007. Because these samples were not analyzed in our previous studies, we briefly describe the sampling setup. At MNM, the intake is placed at the top of the measurement tower (10 m above sea level). Each air sample is collected in serially connected duplicate 2.5-L glass flasks. The sampling frequency at MNM (twice a month) is lower than at HAT and COI, because its geographically remote location limits transportation capacity. On the Southeast Asian route, cargo ships usually call at several of the following ports: Batangas (Philippines), Kota Kinabalu and Kuching (Malaysia), Jakarta (Indonesia), Singapore, Port Klang (Malaysia), Leam Chabang (Thailand), and Hong Kong (China). Onboard air samples are collected in 2.5-L glass flasks. Usually, 7 or 14 flask samples are collected during each roundtrip cruise, lasting approximately 30 days, within 10°S–20°N in latitude and 100°E–120°E in longitude. However, the precise schedule and route are frequently modified because of the variable socioeconomic circumstances that frequently affect commodity distribution in Southeast Asia.

The locations of the ground stations and sampling points from cargo ships are illustrated in Figure 1a, with temporal and latitudinal coverage of the ship samples plotted in Figure 1b. In this study, we used all available flask data acquired from mid-1997 to late 2021. Although the sampling locations are well distributed in the South China Sea, western tropical Pacific, and North Pacific, the sampling frequencies are somewhat irregular, except along the Oceania route, mainly because of frequent modifications of ship service and cruise routes (Figure 1b).

2.2 Analysis of O₂ and CO₂ Mole Fractions From Flask Air Samples

Collected air samples were regularly sent back to our laboratory at the National Institute for Environmental Studies (NIES) to measure relative variations in the atmospheric O₂ and CO₂ mole fractions. We express the O₂ mole fraction variations as relative deviations of the O₂ to nitrogen (N₂) ratio from an arbitrary reference as follows:

$$$\delta \left({\mathrm{O}}_{2}/{\mathrm{N}}_{2}\right)=\left\{{\left({\mathrm{O}}_{2}/{\mathrm{N}}_{2}\right)}_{\text{sam}}/{\left({\mathrm{O}}_{2}/{\mathrm{N}}_{2}\right)}_{\text{ref}}-1\right\}\times {10}^{6},$$$ (1)

where subscripts “sam” and “ref” refer to sample and reference, respectively, and δ(O₂/N₂) is expressed in “per meg” (Keeling & Shertz, 1992). A variation of 1 μmol O₂ within 1 mol of dry air sample, equivalent to a 1 μmol/mol (1 part-per-million, hereafter ppm) variation in the abundance of minor constituents (e.g., CO₂), modifies the O₂/N₂ ratio by 4.77 per meg.

We measured O₂/N₂ ratios using a gas chromatograph equipped with a thermal conductivity detector (GC/TCD; Tohjima, 2000). By alternating O₂/N₂ ratio measurements in sample air and in reference air, we calculated the δ(O₂/N₂) values in accordance with Equation 1 using our original O₂/N₂ scale as the reference (Tohjima et al., 2008). We also determined CO₂ mole fractions in the air samples, using a non-dispersive infrared (NDIR) analyzer, against the “NIES 09” calibration scale established from a set of gravimetrically prepared CO₂-in-air standard gases (Machida et al., 2011). During the “Sixth CO₂ Round-Robin Standard Gas Comparison Experiment” of the World Meteorological Organization (https://www.esrl.noaa.gov/gmd/ccgg/wmorr/wmorr_results.php), differences between the NIES 09 scale and the reference scale established by the National Oceanic and Aeronautics Administration (United States of America) were consistently within ±0.15 ppm from 1993 to 2015.

Recently, Aoki et al. (2019) successfully applied a gravimetric method to prepare standard mixtures of purified N₂, O₂, argon (Ar), and CO₂ gases with sub-ppm precision for atmospheric O₂ measurements. Using these mixtures, they conducted an O₂/N₂ scale comparison experiment between several laboratories, including the NIES (Aoki et al., 2021). The results yielded a span sensitivity of the NIES O₂/N₂ scale that was 3.39 ± 0.13% lower than that of the gravimetric scale. Therefore, we multiplied the δ(O₂/N₂) values measured with the GC/TCD by 1.0339 to account for this negative bias.

Next, we calculated the APO value from the CO₂ mole fraction (X_CO2, in ppm) and δ(O₂/N₂) (in per meg) as follows:

$$$\delta \text{APO}=\delta \left({\mathrm{O}}_{2}/{\mathrm{N}}_{2}\right)+{{\alpha }_{\mathrm{B}}\times X}_{{\text{CO}}_{2}}/{S}_{{\mathrm{O}}_{2}}-1850,$$$ (2)

where $${S}_{{\mathrm{O}}_{2}}$$ is the O₂ dry-air mole fraction ($${S}_{{\mathrm{O}}_{2}}$$ = 0.2094; Tohjima, Machida, et al., 2005) and the subtracted constant, 1850 per meg, is the arbitrary reference for δAPO. In this study, we adopted α_B = 1.10 ± 0.05 in accordance with previous studies (e.g., Keeling & Manning, 2014; Tohjima et al., 2019) despite the ongoing debate on the true value of α_B (Keeling & Manning, 2014). The sensitivity test showed that there was no significant difference in the autumn APO enhancement between the α_B values of 1.1 and 1.0. Time series of the observed APO and resulting smooth-curve fits calculated with a digital filtering technique (Thoning et al., 1989) are plotted in Figure 2 for all air samples (ground stations and ship routes). For convenience, the ship data were binned into 10°-wide latitude bands (40°S–30°S to 50°N–60°N), since the amplitude of the APO seasonal cycle depends primarily on the observation latitude (e.g., Tohjima et al., 2012). The calculated O₂/N₂ ratios and CO₂ mole fractions are reported in the NIES Global Environmental Database (Tohjima et al., 2023a, 2023b, 2023c, 2023d).

Figure 2

Time series of atmospheric potential oxygen (APO) data (crosses, in per meg) for all air samples collected at ground stations (COI, HAT, and MNM) and onboard cargo ships, shifted vertically for visibility, and the associated smooth-curve fits (light blue curves). The ship sampling data points are binned into 10° latitude bands and displayed by decreasing latitude from top to bottom (50°N–60°N to 40°S–30°S).

Under our analytical conditions, the Ar mole fraction in the flask air samples could not be determined separately from O₂ with a GC/TCD. Therefore, we calculated the O₂ + Ar to N₂ peak area ratio instead and then converted it into the O₂/N₂ ratio, by assuming that the atmospheric Ar/N₂ ratio is constant through the entire period of the study (Tohjima, 2000). However, recent studies have established that the atmospheric Ar/N₂ ratio exhibits marked seasonal variations, approximately in phase with the APO seasonal cycle (Battle et al., 2003; Cassar et al., 2008; Ishidoya & Murayama, 2014; Keeling et al., 2004). For example, seasonal peak-to-peak amplitudes of δ(Ar/N₂) derived from flask measurements at globally distributed stations were within 6–26 per meg (see Table 1 of Ishidoya & Murayama, 2014), with the largest value at Macquarie Island (55°S, 159°E). Considering the Ar/O₂ abundance ratio (0.0093/0.2094), seasonal variations of δ(Ar/N₂) could enhance the APO seasonal cycle by up to 1 per meg. However, because we could not measure the atmospheric Ar/N₂ ratio, we did not consider Ar/N₂ seasonal variations in our analysis.

2.3 Extraction of the APO Seasonal Cycle From Flask Data

To refine the spatial distribution of APO seasonal variations determined in our previous studies (Tohjima, Mukai, et al., 2005; Tohjima et al., 2012), we grouped the ship flask sampling data into 10° × 10° latitude/longitude bins (Figure 1a). We excluded bins with fewer than 12 APO data points or less than 9 months of usable APO data from which a reliable seasonal cycle could not be determined. Finally, 41 bins meeting the data selection criteria were retained (Figure 1a, pink-shaded 10° × 10° bins).

To remove long-term trends from the APO data and extract the seasonal cycle, we proceeded as follows: first, the APO time series at each ground station or within each 10° × 10° bin was fitted with the sum of a quadratic polynomial and of the first-order and second-order harmonics of the APO annual cycle using a least-squares method. Since the seasonal cycles show temporally asymmetric shapes, as shown in Figure 2, we used the two harmonics, which can better express the seasonal cycles than one harmonic. Then, the long-term decreasing APO trend, represented by the fitting function, was subtracted from the time series. Next, we extracted the average seasonal APO variations at each location (ground station or 10° × 10° bin) from the detrended APO data using the following bootstrap method:

• -

  first, we prepared an APO data set from the individual detrended APO time series at each location by iterative random resampling with replacement (the numbers of points in the original and prepared data sets are identical).

• -

  then, we fitted each prepared data set with a function of the first-order and second-order annual cycle harmonics using a least-squares method.

• -

  we repeated this procedure 10,000 times at each measurement location to derive the corresponding average seasonal cycle properties: peak-to-peak amplitude, phase of the peaks (maximum and minimum), and associated standard deviations (1σ).

In this algorithm, we considered that the first-order and second-order annual cycle harmonics were sufficient to express APO seasonality.

2.4 Simulations of APO With an Atmospheric Transport Model

To further investigate the APO seasonal variations, we simulated the atmospheric CO₂, O₂, and N₂ variations using sets of surface CO₂, O₂, and N₂ flux estimates and an atmospheric transport model, the NIES Transport Model (NIES-TM) version 8.1i (Belikov et al., 2013; Maksyutov et al., 2008). The model has horizontal resolution of 2.5° × 2.5° and 32 vertical layers, driven by the Japanese 55-year Reanalysis of the Japanese Meteorological Agency (Harada et al., 2016; Kobayashi et al., 2015). We used the APO simulations at the lowest level in this study.

To simulate atmospheric O₂ and N₂ variations, we used the GK2001 O₂ fluxes and air-sea N₂ fluxes estimated by Blaine (2005), expressed as monthly means and annually balanced; their annually integrated value was zero in each grid cell. Oceanic annual-mean O₂ and N₂ fluxes were prescribed from the inversion studies of Gruber et al. (2001) and Gloor et al. (2001), respectively. For oceanic seasonal O₂ fluxes, we also applied two variations of GK2001 O₂ fluxes: one was shifted with its seasonal pattern (GK_delay), and the other was modified with seasonal emission strength (GK_mod). These additional seasonal O₂ fluxes are explained and discussed in Sections 3.2 and 3.5. To simulate atmospheric CO₂ variations, we used monthly mean sea-surface CO₂ fluxes from Takahashi et al. (2009) and monthly mean fossil-fuel-derived CO₂ fluxes taken from the Open-Data Inventory for Anthropogenic Carbon dioxide version 2016 (Oda & Maksyutov, 2011; Oda et al., 2018). To evaluate the influence of O₂ consumption from fossil-fuel combustion on the atmospheric O₂ abundance, we multiplied fossil-fuel-derived CO₂ abundances by the global mean −O₂:CO₂ exchange ratio (α_F), set to a value of 1.4, on the basis of an estimated global fossil-fuel mix (Friedlingstein et al., 2022) and of oxidative ratios of individual fossil-fuel types (Keeling, 1988). The bibliographic sources describing the fluxes used in our simulation are listed in Table 1. Each flux was processed as an individual tracer in the NIES-TM to simulate its spatiotemporal atmospheric variations. Using the output tracer mixing ratios (in ppm), we applied Equation 2 to calculate the APO (in per meg) as follows:

$$$\text{APO}=\left(\frac{{\mathrm{O}}_{2}^{\text{AM}}}{{S}_{{\mathrm{O}}_{2}}}-\frac{{\mathrm{N}}_{2}^{\text{AM}}}{{S}_{{\mathrm{N}}_{2}}}\right)+\left(\frac{{\mathrm{O}}_{2}^{\text{SA}}}{{S}_{{\mathrm{O}}_{2}}}-\frac{{\mathrm{N}}_{2}^{\text{SA}}}{{S}_{{\mathrm{N}}_{2}}}\right)+\left(\frac{\left({\alpha }_{B}{-\alpha }_{F}\right)\cdot \mathrm{C}{\mathrm{O}}_{2}^{\text{FF}}}{{S}_{{\mathrm{O}}_{2}}}\right)+\left(\frac{{\alpha }_{B}\cdot \mathrm{C}{\mathrm{O}}_{2}^{\text{OC}}}{{S}_{{\mathrm{O}}_{2}}}\right),$$$ (3)

where the superscripts “AM,” “SA,” “FF,” and “OC” represent the annual-mean, seasonally anomalous, fossil-fuel-derived, and oceanic CO₂ fluxes, respectively, and $${S}_{{\mathrm{N}}_{2}}$$ represents the atmospheric N₂ dry-air mole fraction ($${S}_{{\mathrm{N}}_{2}}$$ = 0.7809; Tohjima, Machida, et al., 2005).

Table 1. Input Fluxes in Our Atmospheric Transport Model Simulations

Flux	Reference
Oceanic seasonal O2
GK2001	Garcia and Keeling (2001)
GKdelay	GK2001 with 2-week phase delay (see Section 3.2)
GKmod	GK2001 with modified seasonality (see Section 3.5)
Oceanic seasonal N2	Blaine (2005)
Oceanic annual-mean O2	Gruber et al. (2001)
Oceanic annual-mean N2	Gloor et al. (2001)
Oceanic CO2 (seasonal + annual mean)	Takahashi et al. (2009)
Fossil-fuel CO2	Oda and Maksyutov (2011) and Oda et al. (2018)
Fossil-fuel O2	Friedlingstein et al. (2022) and Keeling (1988)

The model simulation ran for 16 years, from 2000 to 2016. To investigate only the APO seasonal cycle climatology, we also detrended the simulated APO results using the method previously applied to the observations (Section 2.3).

3 Results and Discussion

3.1 APO Seasonal Cycles in the Pacific

The detrended APO data and corresponding average seasonal cycles (determined as described in Section 2.3) with the associated 1σ standard deviations are plotted as a function of day of year in Figures 3 and 4 for the ground stations (COI, HAT, and MNM) and cargo ships (within the 41 selected 10° × 10° bins), respectively. Both figures also illustrate the simulated average seasonal cycles without and with adjustment to the observations (see Section 3.3 for the adjustment methodology) and the differences between the observed and adjusted simulated cycles. For each ground station (Figure 3), simulated data were interpolated from the values at the four nearest points on the 2.5° × 2.5° model grid. The associated standard deviations were less than 0.5 per meg (within the simulated curve thickness); thus, they are not visible in Figure 3. For the cargo ship bins (Figure 4), simulated seasonal cycles were calculated as the average (and associated 1σ standard deviation, shading in Figure 4) of the values at the 16 model grid points included within each 10° × 10° geographic bin.

Figure 3

APO seasonal cycle at the Japanese ground stations: COI (top), HAT (middle), and MNM (bottom). Dots: detrended observation-derived APO data with clear seasonal variations. Dark-blue curve: average APO seasonal cycle (least-squares fitting of the observed data by the sum of the first two harmonics of the APO annual cycle; Section 2.3). Orange curve: same as the dark blue curve but for an atmospheric transport model simulation using the original O₂ fluxes of Garcia and Keeling (2001; hereafter GK2001 O₂ fluxes). Light-blue dashed curve: same as the orange curve but adjusted to the observed seasonal cycle for peak-to-peak amplitude and seasonal phase of the maxima and minima (see Section 3.3). Red curve: differences between the observation and the adjusted simulation (dark-blue–light-blue).

Figure 4

Same as Figure 3, but for APO values derived from air samples collected onboard cargo ships in the 41 selected bins. The latitude/longitude labels (top left of each panel) represent the mid-bin values. Bins are arranged from top to bottom and from left to right in order of decreasing latitude and then decreasing longitude at each latitude. The blue, orange, and red shaded areas represent the standard deviation of the least-squares fit for the observation-derived, simulated, and (observed minus adjusted) APO data, respectively.

The observed APO cycles consistently exhibited a seasonal minimum in late winter and maximum in summer with peak-to-peak amplitudes increasing from the equator toward higher latitudes in both hemispheres, as established in our previous studies (Tohjima, Mukai, et al., 2005; Tohjima et al., 2012). The NIES-TM simulations reproduced the observed seasonality well. Spatial distributions of the peak-to-peak APO amplitudes and phase of the seasonal maxima and minima in the North and South Pacific are illustrated for the observations (Figure 5, left) and the simulations (Figure 5, right). Moreover, the observed peak-to-peak APO amplitudes exhibited not only the expected latitudinal gradients in the western Pacific but also a longitudinal gradient in the northern North Pacific around 40°N–50°N, with peak-to-peak amplitudes gradually decreasing eastward from approximately 180°E. These longitudinal gradients were well reproduced by the simulations.

Figure 5

Spatial distributions of observed (left) and simulated (right) APO seasonal cycle parameters at the Japanese ground stations (circles) and in 10° × 10° bins in the Pacific. Top row (a, b): peak-to-peak APO amplitude (in per meg); middle and bottom rows: phase (day of year) of the seasonal minimum (c, d) and maximum (e, f). For observations onboard ships, the 41 selected bins are illustrated.

For detailed comparison of seasonality between observations and simulations, we plotted together by latitude the observed and simulated peak-to-peak amplitudes (Figure 6a, left), the corresponding amplitude ratios (observation/simulation, Figure 6a, right), the phase of the minima/maxima (Figures 6b and 6c, left) and phase differences (observation minus simulation; Figures 6b and 6c, right). The simulations overestimated the peak-to-peak APO amplitude in 33 bins and at the HAT station, with an amplitude ratio and 1σ standard deviation of 0.92 ± 0.17 at all sites (number of “sites” n = 44, comprising the three ground stations and the 41 selected bins). Furthermore, the simulated APO maxima and minima occurred generally earlier (positive phase difference) than the observed peaks by a few weeks, except in the northern North Pacific where the simulated occurrence dates were in good agreement with the observed values. The average and standard deviation of the phase difference was 13.3 ± 16.5 days for the seasonal minima and 5.0 ± 14.0 days for the seasonal maxima. The seasonal cycles simulated in this study from the GK2001 O₂ fluxes were consistent with previous results (Garcia & Keeling, 2001; Tohjima et al., 2012). The complete results (peak-to-peak amplitude and days of occurrence of the seasonal minima and maxima) are reported in Table S1 in Supporting Information S1.

Figure 6

Comparison of APO seasonal cycle parameters between observations and simulations: (a) peak-to-peak amplitude (left) and amplitude ratio (observation/simulation, right); (b, c) phase of the seasonal minimum and maximum, respectively (left), and phase difference (observation minus simulation, right). Data within 10° latitude bands are ordered, from top to bottom, by decreasing latitude from 60°N to 40°S and from west to east in each band. The ground stations are indicated by colored squares inserted in the relevant latitude/longitude order. Labels indicate the center point coordinates. Data points within the same latitude band are connected by lines. Red and blue dots (left panels) represent the observations and simulations, respectively. Horizontal bars in all panels represent the standard deviations. The horizontal dashed lines represent the boundary lines of 20°S, 20°N, and 40°N.

In addition to the seasonal cycle differences noted above, the NIES-TM simulations frequently showed rapid drawdown from autumn to early winter (Figures 3 and 4) compared with the observations.

3.2 Influence of the O₂ Flux Delay on the APO Seasonal Cycle

To better analyze the discrepancies between the observed and simulated seasonal cycle patterns (shapes of the seasonal cycles), differences in the amplitude and phase of the seasonal cycle must be minimized. Assuming that the peak-to-peak amplitude of the APO seasonal cycle is proportional to that of the oceanic O₂ flux, the amplitude difference is minimized by normalizing the simulated amplitude to the observed value, that is, by scaling it with an empirically determined factor. The phase difference can also be adjusted if the delay in the observed seasonal cycle of the air-sea O₂ fluxes does not notably influence the APO seasonal cycle pattern, except for the phase shift. Therefore, to assess this influence, we conducted an additional simulation with the NIES-TM.

For this second simulation of atmospheric O₂ variations, we created “delayed fluxes” by applying a 2-week delay to the GK2001 O₂ fluxes (hereafter GK_delay O₂ fluxes) to reflect the average phase difference, of at most two weeks, calculated for the seasonal maxima and minima (Section 3.1). Garcia and Keeling (2001) had suggested shifting the GK2001 O₂ flux seasonal variations to an earlier occurrence by a few weeks relative to the true seasonal variations, because their flux estimation methodology did not consider the mixed-layer O₂ equilibration process. Because this process might similarly apply to the N₂ flux variations, we also simulated atmospheric N₂ using a seasonal N₂ flux delayed by 2 weeks. Subsequently, we calculated the APO seasonal cycles with the delayed fluxes and compared the results with those from the first simulation. Both simulated cycles exhibited nearly identical seasonal patterns. However, a more detailed comparison yielded slight but frequent peak-to-peak amplitude differences, from −2.5 per meg to +2.5 per meg depending on the location. The GK_delay O₂ fluxes caused a slight reduction of the peak-to-peak amplitude in the Southern Ocean and, conversely, a slight enhancement in the equatorial region in East Asia and in the northern North Pacific (Figure S1 in Supporting Information S1). Moreover, the phase differences between the seasonal cycles simulated with the GK2001 and GK_delay O₂ fluxes were consistent with the delay of two weeks between the original and delayed O₂ fluxes (Figure S1 in Supporting Information S1), with an average phase difference and associated standard deviation (all 2.5° × 2.5° model grid cells, delayed simulation minus original) of 14.5 ± 8.1 days for the seasonal minimum and 14.0 ± 8.1 days for the seasonal maximum.

Subsequently, we adjusted the original simulated seasonal cycles to the delayed cycles at each model grid point to minimize the differences in the peak-to-peak amplitudes and seasonal phases. Specifically, the original seasonal cycle was scaled by the amplitude ratio (delayed/original) to adjust the seasonal peak-to-peak amplitudes and time-shifted until the midpoints between the minimum and maximum coincided in the original and delayed seasonal cycles. Finally, we calculated the difference between the delayed seasonal cycle and the adjusted original (delayed minus adjusted). An example of this adjustment procedure is illustrated in Figure 7, which also demonstrates that evaluating differences between seasonal cycles without adjustment produces an erroneous result (see the red curve in Figure 7a).

Figure 7

Adjustment of the simulated data to the observations, illustrated in the 10° × 10° bin centered at 55°N and 175°W. (a) Simulated seasonal cycle using the original GK2001 O₂ fluxes (orange curve) and the GK2001 O₂ fluxes delayed by 2 weeks (hereafter GK_delay, blue curve), and the corresponding differences (red curve, delayed minus original). (b) Same as (a) but for the delayed simulation (blue curve), the original simulation after adjustment to the observed data for amplitude and phase (light-blue dashed curve, see Section 3.2), and the corresponding differences (red curve, delayed minus adjusted).

In general, the seasonal patterns simulated with the GK2001 and GK_delay O₂ fluxes were nearly identical: the average and standard deviation of the root-mean-square differences over all model grid cells were 1 ± 1 per meg. However, small but non-negligible consistent differences remained from autumn to early winter at most grid points (Figure 7b, red curve, positive values after Day 244). We refer hereafter to these persistent differences as the “apparent autumn peak.” Subsequently, we analyzed the amplitude and geographic distribution of the apparent autumn peak. Its amplitude is defined, at each model grid point, as the maximum absolute value of the differences between the delayed and adjusted simulated seasonal cycles (e.g., in Figure 7b, the largest absolute deviation of the red difference curve from the APO = 0 horizontal axis). Results yielded notably high amplitudes, of more than 3 per meg, in the northern North Pacific, northern North Atlantic, and Southern Oceans (Figure S2 in Supporting Information S1), in good geographic correspondence with large peak-to-peak amplitudes of simulated APO seasonal cycles (e.g., Tohjima et al., 2012). A possible partial explanation for this apparent autumn peak is a temporal covariation effect between atmospheric transport and oceanic O₂ fluxes, with lower mixing efficiency and higher O₂ flux induced by the seasonal cycle delay resulting in enhanced APO values in autumn-winter. Because the temporal covariation effect could cause the apparent autumn peak, these results must be considered with caution when analyzing tentative additional autumn oceanic O₂ emissions on the basis of an observed APO enhancement.

3.3 Comparison Between Observed and Simulated APO Seasonal Patterns

Applying the adjustment procedure described in Section 3.2, we calculated the differences between the observations and the adjusted simulation (observed minus adjusted). The adjusted simulation was in good agreement with the observed seasonal cycles, with differences generally close to zero (Figures 3 and 4) and especially small throughout the year in the equatorial region (20°S–20°N). However, as mentioned in Section 3.1 (and established in Section 3.2 for the simulated APO), marked differences persisted outside the equatorial region, with observed APO values higher than the simulations from autumn to early winter, representing the apparent autumn peak.

Figure 8 illustrates the properties (amplitude and phase) of the observation-based apparent autumn APO peaks (observation minus adjusted simulation) at the three ground stations and in the 41 selected 10° × 10° bins. Uncertainty on the apparent autumn peak amplitude was calculated as the root-sum-square of the standard deviation at the peak position for the compared seasonal cycles. The simulated apparent autumn peaks caused by the temporal covariation effect on the GK_delay O₂ flux (Section 3.2) are indicated for comparison. Although some APO peaks in Figure 8 did not occur in autumn or early winter, we still refer to them as “apparent autumn peaks” for simplicity. We also illustrate the spatial distribution of the observed apparent autumn peaks in Figure 9. In the equatorial region, the observed apparent autumn peaks were small (maximum 4 per meg) and consistent with those caused by temporal covariation. Conversely, they markedly exceeded the possible temporal covariation effect at higher latitudes in both hemispheres, with an amplitude range within 7–13 per meg (for an average and standard deviation of 9.2 ± 3.3 per meg) at 40°S–20°S, 2–19 per meg (10.0 ± 5.8 per meg) at 20°N–40°N, and 4–37 per meg (22.8 ± 10.5 per meg) at 40°N–60°N. Note that the rather small apparent autumn peak for the bin centered on 55°N and 175°W may be related to the insufficient data number in summer and early autumn, as shown in Figure 4. The observation-based apparent autumn peaks (Figure 8b) occurred consistently between Days 273–375 (365 + 10) in the Northern Hemisphere (20°N–60°N) and Days 150–170 in the Southern Hemisphere (40°S–20°S). These results suggest that a delay in the air-sea O₂ fluxes relative to the GK2001 O₂ fluxes cannot fully explain the observed apparent autumn peaks. Therefore, as also hypothesized in previous studies (Ishidoya et al., 2016; Jin et al., 2023), we postulate the existence of additional oceanic O₂ emissions in autumn, not expressed in the GK2001 O₂ fluxes, that enhance atmospheric APO in autumn-winter.

Figure 8

Same general representation as the left-hand side panels of Figure 6 but for the (a) amplitudes (horizontal bars represent the uncertainty) and (b) phase of the apparent autumn peaks. Red symbols: observation-based apparent autumn peaks estimated as the seasonal cycle differences (observation minus adjusted simulation) using the original GK2001 O₂ fluxes. Blue and gray symbols: simulated apparent autumn peaks estimated as the differences between the seasonal cycles simulated with the original GK2001 O₂ fluxes and with the GK_delay or modified GK2001 fluxes (GK_mod) (delayed or modified minus adjusted original), respectively (see Section 3.5). The horizontal dashed lines represent the boundary lines of 20°S, 20°N, and 40°N.

Figure 9

Same as Figure 5a, but for the amplitude of the observation-based apparent autumn peaks (red symbols in Figure 8a).

3.4 Possible Mechanisms for Autumn O₂ Emissions

In Section 1, we cited two possible mechanisms of additional oceanic O₂ emissions in autumn: outgassing from phytoplankton bloom and from the SOM. Sapiano et al. (2012) constructed a global climatology of marine phytoplankton phenology from a 10-year satellite observation data set of chlorophyll a concentrations. They established that a single phytoplankton bloom occurred over most of the global ocean in spring, but that a secondary bloom was sometimes observed in autumn at limited geographic locations that included the northern North Pacific at 40°N–60°N (see Figure 7b of Sapiano et al., 2012). This location corresponds to the larger apparent autumn peaks identified in this study (Figure 9). Previously, Shulenberger and Reid (1981) had reported SOM detections in the North Pacific at 10°N–40°N (see their Figure 1). By visualizing climatological depth-time cross-sections of dissolved oxygen data in the Pacific Ocean, extracted from the World Ocean Atlas 2018 (Garcia et al., 2019), we located separate SOM at 20°N–40°N and 40°S–20°S (Figure S3 in Supporting Information S1), corresponding to the moderate apparent autumn peaks determined in this study (Figure 9). Accordingly, we postulate that the larger apparent autumn peaks at 40°N–60°N might be caused by O₂ emissions associated with the autumn phytoplankton bloom, while the moderate apparent autumn peaks at 20°N–40°N and 20°S–40°S might be attributed to O₂ emissions from SOM. Finally, the SOM observed in the South Pacific is less marked than in the Northern Pacific (Figure S3 in Supporting Information S1), which is consistent with the small observation-based South Pacific apparent autumn peaks identified in our study.

Both mechanisms are strongly related to the deepening of the mixed oceanic layer, which replenishes the surface layer with nutrients from the deeper ocean and could enhance O₂ outgassing from the SOM layer into the atmosphere. We selected apparent autumn peaks with amplitudes larger than 10 per meg to calculate the corresponding average phase (Figure 8b): Day 318 ± 16 and Day 325 ± 17 at latitudes within 40°N–60°N and 20°N–40°N, respectively. From climatological MLD data established by de Boyer Montégut et al. (2004) and Hosoda et al. (2008), we evaluated MLDs in the 41 selected 10° × 10° bins at the calculated phase of the observed apparent autumn peaks. The apparent autumn peaks occurred during the deepening season of the oceanic mixed layer, when the MLD increased to 56 ± 10 m and 59 ± 10 m at 40°N–60°N and 20°N–40°N, respectively. These depths might be critical for initiating the autumn phytoplankton bloom at 40°N–60°N and the SOM outgassing at 20°N–40°N in the North Pacific. The estimated MLD at 20°N–40°N (59 ± 10 m) was consistent with the climatological SOM central depths at 20°N–40°N in the North Pacific (Figure S3 in Supporting Information S1).

3.5 Estimation of Autumn O₂ Emissions

Next, we evaluated the relationship between the autumn O₂ emission intensity and the apparent autumn peaks by conducting another model simulation in which we added a homogeneous oceanic O₂ outgassing flux of 0.3 μmol/m²/s to the original GK2001 O₂ fluxes in autumn: specifically in September–November between the equator and 60°N and in March-May between the equator and 60°S. Since the average amplitude of the observed apparent autumn peak at 40°N–50°N is about 25% of that of the simulated seasonal amplitude at 40°N–50°N, we adopted 25% of the peak-to-peak amplitude of the zonal-mean GK2001 O₂ flux variation between 40°N and 50°N as the additional flux. Before calculating the seasonal cycle of APO, we also subtracted O₂ fluxes of 0.075 μmol/m²/s uniformly from all the grids between 60°S and 60°N of the modified GK2001 O₂ fluxes (hereafter GK_mod O₂ fluxes) to balance the annual O₂ fluxes at all model grid points.

The seasonal cycles simulated by the NIES-TM with the GK_mod O₂ fluxes exhibited clear differences in seasonal amplitude and phase relative to the original simulation (Figure S4 in Supporting Information S1). Overall, peak-to-peak amplitudes in the GK_mod-based simulation decreased at latitudes higher than 20° in both hemispheres but increased between 20°S and 20°N. The modified seasonal minimum and maximum occurred approximately 10 days later than in the original simulation, except between 30°S and 30°N where the delay of the maximum frequently exceeded 50 days. Moreover, the GK_mod-based simulation reproduced the observed seasonal phases more consistently in the northern North Pacific (30°N–60°N) than the original simulation, with average phase differences improving from 11.0 ± 10.6 days to 4.4 ± 9.3 days for the APO seasonal minimum and from 6.4 ± 11.8 days to −3.4 ± 15.0 days for the maximum.

Applying the approach described in Section 3.2, we subsequently evaluated the apparent autumn peaks simulated with the GK_mod O₂ fluxes for the three ground stations and the 41 selected 10° × 10° bins, relative to the original simulation (GK2001 O₂ fluxes). The modified seasonal cycle nearly exactly reproduced the apparent autumn peaks, with minimum amplitude between 10°N and 20°N and poleward increase in both hemispheres (Figure 8). Furthermore, compared with the observation-based autumn peaks, the amplitudes of the modified apparent autumn peaks were higher between 20°S and 10°N but lower north of 30°N. These results indicate that the added 0.3 μmol/m²/s outgassing flux could not completely explain the observation-based apparent autumn peaks north of 30°N; on the contrary, it produced unrealistically high APO enhancements in the equatorial region between 20°S and 10°N.

The regular sinusoidal seasonal pattern of the GK2001 O₂ fluxes likely underestimates autumn O₂ emissions, especially in the North Pacific at 20°N–60°N. Although our observational coverage of the Southern Hemisphere is limited to the western Pacific, the resulting data suggest that rather strong additional autumn O₂ emissions (more than 0.3 μmol/m²/s) are not required to reproduce the observed oceanic APO seasonal cycle near the equator and in the Southern Hemisphere. Consequently, we conclude that oceanic O₂ fluxes exhibit a specific seasonal variation pattern in the mid- and high latitudes of the Northern Hemisphere, with a prolonged enhancement in autumn following the summer emission maximum. These results are partly consistent with the recent evaluation of hemispheric-scale oceanic O₂ fluxes by Jin et al. (2023).

4 Conclusions

We derived the seasonal cycle of APO = O₂ + 1.1 × CO₂ from flask air samples collected at three Japanese ground stations and on cargo ships sailing through the western Pacific and the northern North Pacific. We compared these observation-based APO seasonal variations with simulations from the NIES-TM atmospheric transport model that used the GK2001 seasonal air-sea O₂ fluxes as input. Despite very good overall consistency between the observations and the model simulations, the observed seasonal pattern exhibited moderate enhancements in autumn and early winter relative to the simulations, especially in the northern North Pacific. After adjusting the amplitudes and phases for precise comparison, the seasonal cycle differences (observation minus adjusted simulation) clearly exhibited apparent peaks in autumn-winter, except in the equatorial region between 20°S and 20°N. The average amplitude of these peaks was 9.2 ± 3.3 per meg, 10.0 ± 5.8 per meg, and 22.8 ± 10.5 per meg at 40°S–20°S, 20°N–40°N and 40°N–60°N, respectively, markedly exceeding a possible autumn enhancement caused by the temporal covariation effect, which appeared in the seasonal cycle simulated with the GK_delay O₂ fluxes. This suggests the existence of additional oceanic autumn O₂ emissions that are not well expressed in the GK2001 O₂ fluxes.

First, the formation and disappearance of a SOM at 20°N–40°N was identified in studies focusing on measured temporal variations of ocean-dissolved O₂ profiles (Shulenberger & Reid, 1981) and on an objective analysis of ocean-dissolved O₂ climatological data (Garcia et al., 2019). Second, a marine phytoplankton phenological study using satellite observations had previously observed a phytoplankton bloom in spring and autumn at 40°N–60°N in the North Pacific (Sapiano et al., 2012). Therefore, we conclude that the biogeochemical mechanisms most likely responsible for the enhanced autumn O₂ emissions in the North Pacific identified in this study were outgassing from the SOM between 20°N and 40°N and autumn oceanic phytoplankton blooms between 40°N and 60°N.

Although the addition of a homogeneous air-sea O₂ flux (+0.3 μmol/m²/s) to the GK2001 O₂ fluxes in autumn in both hemispheres allowed the simulation to better reproduce the observed seasonal cycle, the modified seasonal cycle still underestimated the observed autumn enhancement at 20°N–60°N. These results suggest that further analyses are required to better characterize the air-sea O₂ fluxes. Furthermore, atmospheric model dependency needs to be investigated to improve oceanic O₂ fluxes. The observed APO variations result from the covariation of atmospheric transport and air-sea O₂ fluxes. Therefore, analysis using multiple atmospheric models would help understand the relative importance of atmospheric transport on the apparent autumn APO enhancement and provide robustness and uncertainty in the O₂ flux estimates. Recent improvements in atmospheric inversions using APO data (e.g., Rödenbeck, Adcock, et al., 2023; Rödenbeck, Le Quéré, et al., 2008) should help refine the spatiotemporal distribution of observed air-sea O₂ fluxes, thereby contributing to a deeper understanding of oceanic biogeochemical and circulation processes.

Acknowledgments

This study was financially supported by the Global Environmental Research Coordination System of the Ministry of the Environment, Government of Japan (Grant E1451), and a Grant-in-Aid for Scientific Research (S) (KAKENHI) of the Japan Society for the Promotion of Science (Grant 22H05006). We gratefully acknowledge Fujio Shimano, Eiji Yoshida, and Noritsugu Tsuda of the Global Environmental Forum (GEF, Tokyo, Japan, https://www.gef.or.jp/en/) for conducting atmospheric samplings at the HAT and COI ground stations. Flask sampling and facility maintenance at the MNM ground station are under the responsibility of the Japan Meteorological Agency. We are also grateful to numerous staff members of the Center for Global Environmental Research, at NIES, for their assistance with the O₂/N₂ and CO₂ measurements from flask air samples. We further express our gratitude to GEF members Shigeru Kariya, Tomoyasu Yamada, Kosei Yumoto, and Nobukazu Oda, and to the owners, operators, and crew of the volunteer observation ships: the MOL Golden Wattle, MOL Glory, Fujitrans World, Trans Future 5, Pyxis, Skaubryn, and New Century 2 for their support in collecting flask air samples. Two anonymous reviewers made valuable comments that helped improve the manuscript. Finally, we thank Eric Dupuy, a freelance scientific editor, for editing a draft of this manuscript.

Open Research

Data Availability Statement

The O₂/N₂ ratios and CO₂ mole fractions for the flask samples used in this study are available through the NIES Global Environmental Database (GED, https://db.cger.nies.go.jp/ged/en/index.html) (Tohjima et al., 2023a, 2023b, 2023c, 2023d).