Seasonal and short-term variations in atmospheric potential oxygen at Ny-Ålesund, Svalbard

Abstract Oxygen in the atmosphere undergoes variations and changes in response to biospheric activities, ocean–atmosphere exchange and fossil fuel combustion. Continuous in situ measurements of atmospheric δ(O2/N2) and CO2 mole fraction were started at Ny-Ålesund, Svalbard (78.93°N, 11.83°E, 40 m a.s.l.) in November 2012. Atmospheric potential oxygen (APO) calculated from the measured O2 and CO2 values during November 2012–January 2015 show a clear seasonal cycle with a peak-to-peak amplitude of approximately 50 per meg. The seasonal cycle of APO simulated using an atmospheric transport model, with prescribed oceanic O2, N2 and CO2 fluxes at monthly time intervals, is in excellent agreement with the observed APO. However, in spring and early summer, high values of APO are observed irregularly on a timescale of hours to days. By comparing backward trajectories of air parcels released from the site with distributions of marine net primary production (NPP), and tagged tracer experiments made using the atmospheric transport model for APO, it is found that these high APO fluctuations are primarily attributable to O2 emissions from the Greenland Sea, the Norwegian Sea and the Barents Sea, due to marine biological productivity. Marine net community production, estimated based on the sea-to-air O2 flux derived from observed APO fluctuations, agrees with NPP obtained from satellite observations within an order of magnitude. The results obtained in this study have still some uncertainties, but our continuous observations of atmospheric δ(O2/N2) and CO2 mole fraction at Ny-Ålesund can play an important role in detecting possible changes in the carbon cycle in the near future.


Introduction
Understanding the sources and sinks of O 2 and CO 2 is important for advancing our knowledge of the global carbon cycle. Atmospheric O 2 and CO 2 are closely related to each other through terrestrial biospheric photosynthesis and respiration, as well as the combustion of fossil fuel. However, the two gases are exchanged independently between the atmosphere and the ocean. Global oceanic and terrestrial biospheric carbon sinks can therefore be estimated separately by analysing long-term changes in atmospheric O 2 and CO 2 (e.g. Keeling and Shertz, 1992;Keeling and Manning, 2014).
Variations in atmospheric O 2 are usually expressed as a change in the ratio of O 2 to N 2 , as follows (Keeling and Shertz, 1992): where (O 2 /N 2 ) sample and (O 2 /N 2 ) reference refer to the mole ratios of O 2 /N 2 in the sample air and reference air, respectively. The value of δ(O 2 /N 2 ) defined by Equation (1) is commonly reported in the unit of 'per meg'. Change of 4.8 per meg in δ(O 2 /N 2 ) corresponds to 1 μmol mol −1 change in O 2 , when O 2 and CO 2 vary in the atmosphere at the -O 2 :CO 2 exchange ratio of 1.0. To eliminate the influence of terrestrial biospheric activity on δ(O 2 /N 2 ), Stephens et al. (1998) defined atmospheric potential oxygen (APO) in per meg unit as follows: where α B is 1.1 (Keeling and Manning, 2014), which is the -O 2 :CO 2 exchange ratio for terrestrial biospheric activity, X O 2 is the mole fraction of O 2 in the atmosphere, CO 2 is the mole at Ny-Ålesund. We discuss seasonal and short-term variations in the observed APO, and compare these variations with those simulated using an atmospheric transport model. We also estimate sea-to-air O 2 fluxes and marine net community production (NCP) in marginal sea areas, based on observed short-term variations in APO.

Observation site and measurement system
Ny-Ålesund (78.93°N, 11.83°E, 40 m a.s.l.) is located on the west coast of Spitsbergen Island, Svalbard ( Fig. 1), and is a base for international Arctic research. The ground surface is generally rocky, with some covering of moss. The annual mean temperature at the base is −4.5 °C, ranging between −12.0 °C in March and +5.8 °C in July. The prevailing surface winds are south-easterly (Maturilli et al., 2013). The ground is typically covered by snow from October to May. A continuous measurement system for atmospheric δ(O 2 /N 2 ) and CO 2 was installed and began operating in the Japanese observatory in November 2012, situated ~2 km west-northwest of the central part of Ny-Ålesund (Goto et al., 2013a).
A detailed description of our measurement system has been presented previously (Goto et al., 2013a(Goto et al., , 2013b. The system consists mainly of O 2 and CO 2 analysers, a gas handling system and a data acquisition system. We adopted a differential fuel cell analyser (Oxzilla-II) for O 2 measurements and a non-dispersive infrared analyser (Li-6252) for CO 2 measurements. An aspirated fraction of CO 2 in the atmosphere (ppm in mole fraction) and 355 is the CO 2 mole fraction (μmol mol −1 ) of Tohoku University O 2 primary standard. APO, thus defined, is mainly governed by air-sea O 2 exchange and has previously been used to evaluate air-sea O 2 flux (e.g. Tohjima et al., 2012). Although APO does not exclude the influence of fossil fuel combustion, its contribution is much reduced by defining it as above, since the -O 2 :CO 2 exchange ratios in burning fossil fuels are not so different from α B (Keeling and Manning, 2014).
Simultaneous observations of atmospheric δ(O 2 /N 2 ) and CO 2 mole fraction have been systematically conducted since the early 1990s using discrete flask sampling with subsequent laboratory analysis (e.g. Bender et al., 2005;Manning and Keeling, 2006;Ishidoya et al., 2012aIshidoya et al., , 2012bTohjima et al., 2015). These observations revealed spatio-temporal variations in atmospheric δ(O 2 /N 2 ) and APO. Measurements of atmospheric δ(O 2 /N 2 ) and CO 2 mole fraction have also been made using continuous in situ monitors (e.g. Lueker et al., 2003;Lueker, 2004;Yamagishi et al., 2008;van der Laan-Luijkx et al., 2010;Goto et al., 2013a;Ishidoya et al., 2013), revealing small and short-term fluctuations in atmospheric δ(O 2 /N 2 ) and APO, thereby providing new insight into the carbon cycle. Lueker et al. (2003) and Lueker (2004) observed low values of APO at the Trinidad Head coastal site in California (41.05°N, 124.15°E), lasting for several days or weeks. They attributed the uptake of atmospheric O 2 to O 2poor deep-sea water, which is transported from the deep ocean to the surface by strong coastal upwelling.
In this paper, we present the results of atmospheric δ(O 2 / N 2 ) and CO 2 observations obtained for the first two years at the first continuous monitoring site of atmospheric δ(O 2 /N 2 ) air-intake was mounted on the roof of the observatory and air samples are continuously fed into the measurement system using a diaphragm pump. To remove water vapor from the air sample, two water traps placed in parallel, equipped with a Stirling cooler, are incorporated into the gas handling system: one is cooled to −80 °C to dehumidify the sample, while the other is heated to 50 °C to dry its inside. The cooling and heating of the respective traps are switched every 6 h. Two standard gases with known values of δ(O 2 /N 2 ) and CO 2 mole fraction, in 48-L aluminium cylinders, are measured every one hour to calibrate the O 2 and CO 2 analysers. By examining a non-linear response of the CO 2 analyser using several standard gases with different CO 2 mole fractions, we confirmed that its influence on the present measured values is negligibly small, within our measurement reproducibility. The δ(O 2 /N 2 ) and CO 2 mole fractions of the two standard gases were determined against the Tohoku University O 2 primary standard (Ishidoya et al., 2003) and the CO 2 mole fraction scale prepared gravimetrically by Tohoku University in 2010 (TU-X10), respectively. These calibrations were conducted before and after their use, and confirmed there were no appreciable drifts in the respective mole fractions. The measurement reproducibility was estimated to be ±1.4-1.9 per meg for δ(O 2 /N 2 ) and ±0.03-0.05 ppm for CO 2 , based on analyses repeated 266 times for the same air sample over 24 h.

Atmospheric transport model
To interpret the APO variations observed at Ny-Ålesund, numerical simulations were performed using the CCSR/NIES/ FRCGC (Center for Climate System Research /National Institute for Environmental Studies/Frontier Research Center for Global Change) Atmospheric General Circulation Model (AG-CM)-based Chemistry Transport Model (ACTM) developed in Japan Agency for Marine-Earth Science and Technology (JAM-STEC). The ACTM uses a horizontal resolution of approximately 2.8° × 2.8° (T42 spectral truncation) and 1.125° × 1.125° (T106 spectral truncation), with 32 pressure-sigma vertical layers (Patra et al., 2009(Patra et al., , 2014). In the model simulations, ocean fluxes of O 2 and N 2 were taken from the TransCom experimental protocol (Garcia and Keeling, 2001;Blaine, 2005), and CO 2 from Takahashi et al. (2009). Emissions of anthropogenic CO 2 were taken from the Emission Database for Global Atmospheric Research (EDGAR, version 4.2) inventory scaled to Carbon Dioxide Information Analysis Center (CDIAC) global totals. Meteorological data used in the model were adjusted to the Japanese 25-year ReAnalysis data (JRA-25) (Onogi et al., 2007), and the atmospheric transport field of the model was validated by Patra et al. (2009). In addition, tagged tracer experiments were performed using the ACTM at T106 resolution to reveal the contributions of different geographical regions to the observed APO variations. The tag-tracer simulations are mass conserving since no chemical loss or gain is modelled in ACTM. In the experiments, five tag regions were assigned, as shown in Fig. 1.
The model-simulated APO values were derived in accordance with Nevison et al. (2008) as follows: where [O 2 ], [N 2 ] and [CO 2 ] are mole fractions of the respective gases calculated for Ny-Ålesund using the model. The superscripts 'OC' and 'FF' denote oceanic and fossil fuel origins, respectively. X O 2 and α B have the same meaning as in Equation (2), while X N 2 is the mole fraction of N 2 in the atmosphere and α F is the global average -O 2 :CO 2 exchange ratio for fossil fuel combustion. In this study, we adopted X O 2 = 0.2094, X N 2 = 0.7809 (Tohjima et al., 2005, α F = 1.4 (Keeling and Manning, 2014), and α B have the same meaning as Equation (2). Figure 2 shows hourly mean values of atmospheric δ(O 2 /N 2 ), CO 2 mole fraction and APO for the period from November 2012 to January 2015. The best-fit curves and long-term trends were obtained by applying a digital filtering technique ( Nakazawa et al., 1997) to the daily mean values. In this technique, signals with periods of longer than 24 months are regarded as the longterm trend, and the average seasonal cycle is approximated by the fundamental and its first harmonics. The δ(O 2 /N 2 ) and CO 2 mole fraction decrease and increase secularly, and vary seasonally in opposite phase to each other. The APO shows not only a secular decrease, but also a seasonal cycle that is almost in phase with that of δ(O 2 /N 2 ).

Seasonal cycle
The results reported by Ishidoya et al. (2012b) from mass spectrometry analyses of weekly flask air samples at Ny-Ålesund are also presented in Fig. 2, for the period from November 2012 to January 2015. Figure 3 compares the δ(O 2 /N 2 ) values reported by Ishidoya et al. (2012b) using the flask sampling method (δ(O 2 /N 2 ) flask ) with the daily mean values from our continuous measurements (δ(O 2 /N 2 ) daily ). Each daily mean value was calculated by averaging the measured values over each 24-h period, including the time when the pertinent flask sample was collected. Although δ(O 2 /N 2 ) flask is on average slightly lower than δ(O 2 /N 2 ) daily by 0.27 ± 14.48 per meg, the differences are within analytical precision, allowing a direct comparison of the values obtained from both measurements. The correlation coefficient between δ(O 2 /N 2 ) flask and δ(O 2 /N 2 ) daily is 0.95, indicating a strong linear relationship.
Seasonal components derived using the digital filtering technique from the observed and simulated APO values are shown for the period of 2001-2010 shows significant high values in autumn compared to their simulated APO seasonal cycle. Such a difference in autumn was also found in previous studies ( Battle et al., 2006;Tohjima et al., 2012). Ishidoya et al. (2016) attributed cause of this discrepancy to an intrusion of the shallow oxygen maximum layer, which exists in the subsurface ocean in summer, into the surface in autumn when the ocean mixed layer is deepened (e.g. Shulenberger and Reid, 1981). If the oceanic O 2 flux used in the model simulations underestimates this effect, then the calculated APO is lowered. On the other hand, the autumn difference between the observed and simulated seasonal APO cycles is not seen in our results. Considering our short-term observations of 2013-2014, the APO value may have not strongly affected by the oxygen maximum layer in those years. For a better understanding of interannual variations of the seasonal APO cycle, further model studies and longer continuous records of APO are required.

Short-term fluctuations in APO during spring and summer
The continuous measurements revealed short-term fluctuations in δ(O 2 /N 2 ), CO 2 and APO in both spring and summer, which were not detected by discrete flask sampling. As an example of such fluctuations, hourly mean values of δ(O 2 /N 2 ), CO 2 mole fraction and APO measured for May-June 2013 are shown in Fig. 5a and b. It should be noted that the values plotted for each variable in these figures are expressed as deviations from the best-fit curve obtained by applying the digital filtering technique to the measured daily means. The deviations are represented as 'Δ' in the figure. It is seen in Fig.  5a that values of δ(O 2 /N 2 ) are irregular and high during this period. The CO 2 mole fraction also varies temporally, but inversely with respect to δ(O 2 /N 2 ). If these fluctuations occur in association with terrestrial biospheric activity, the amplitudes of δ(O 2 /N 2 ) and CO 2 fluctuations are expected to be nearly equal, since the -O 2 :CO 2 exchange ratio is 1.1 (Keeling and Manning, 2014). However, the observational results indicate that the CO 2 mole fraction shows much smaller amplitudes than δ(O 2 /N 2 ), suggesting the exchange ratio is higher than 1.1. For example, the -O 2 :CO 2 exchange ratio for the δ(O 2 / N 2 ) and CO 2 variations observed between 30 May and 31 May 2013 is calculated to be 3.96 ± 0.03, showing that the fluctuations are more affected by oceanic process than by terrestrial biospheric process. In this regard, it is well known that the influence of air-sea exchange on atmospheric CO 2 is much smaller due to marine carbonate chemistry (Keeling et al., 1993), compared to O 2 . As seen in Fig. 8a to be appeared later, air arrives at Ny-Ålesund on 31 May 2013 after passing over its peripheral seas for 10 days. The CO 2 mole fractions calculated by the ACTM with only air-sea CO 2 fluxes also indicate low values during this period, as seen in Fig. 5a. Engardt et al. (1996) found similar CO 2 dips in spring and early summer in Fig. 4. The seasonal cycle of APO obtained from the continuous measurements reaches maximum in July, and minimum in late March, with a peak-to-peak amplitude of approximately 50 per meg. The observed seasonal APO cycle agree fairly well with both resolutions of the ACTM; using the T106 resolution the correlation coefficient is 0.99 (slope of 0.86), and using the T42 resolution the correlation coefficient is 0.98 (slope of 0.80). However, the observed seasonal maximum appears approximately one month earlier than the model simulations. As seen in Fig. 4, the observed APO often shows extremely high values in June and July, probably due to a phytoplankton bloom (Nakaoka et al., 2006). Such high values are not found in the model simulations, since the oceanic O 2 , N 2 and CO 2 fluxes are climatological data on a monthly basis. Ishidoya et al. (2016) reported that the average seasonal cycle of APO obtained by discrete flask sampling at Ny-Ålesund ratio for the ocean process is different from that for the terrestrial biosphere process.
To determine the origin of short-term APO variations observed at Ny-Ålesund, as well as to examine the relationship between air mass transport and APO variation, 10-day backward trajectories from the site were predicted using the CGER/ METEX trajectory model (Zeng et al., 2003). In this analysis, air parcels were released from the lowest level of the model (150 m above sea level) every 6 h during the observation period of 9 May-29 June 2013. We confirmed that raising the releasing level by several hundred metres does not appreciably change the results. The predicted trajectories are presented in Fig. 6. Red and blue lines show the trajectories for the respective times when high and low values of APO were observed. Here, we define 'high APO' as the time when ΔAPO exceeds 10 per meg (grey shading in Fig. 5), and all other times as 'low APO'. Figure 6 shows that the APO value varies with the transport path of air. In general, high APO values are observed when air masses arrive at Ny-Ålesund after passing over the seas on the south side of the Svalbard Islands, while most of low or stable APO values are observed when the air arrives from the ice-covered Arctic Ocean.
The APO values simulated using the ACTM are shown in Fig. 5b. Although the seasonal APO cycles for the two model resolutions (T42 and T106) are in fairly good agreement with each other, the observed short-term APO variations agree much better with the simulated APO values of T106 (correlation coefficient = 0.52, slope = 0.41) than those of T42 (correlation coefficient = 0.32, slope = 0.28). This indicates that higher resolution transport models would be better for interpreting shortterm APO variations. However, much finer model resolutions from continuous measurements at Ny-Ålesund, and suggested that CO 2 uptake in the North Atlantic Ocean and/or by the land biosphere on more southerly latitudes are responsible for such dips. However, it is difficult to separate these causes from only CO 2 variations. Our concurrent observations of δ(O 2 /N 2 ) and CO 2 would be useful for quantitatively interpreting short-term CO 2 fluctuations, since the -O 2 :CO 2 exchange  represents the uncertainty of the best-fit curve of seasonal APO components estimated by a bootstrap analysis (Press et al., 2007). The results of model simulations with two different resolutions (T42 and T106) for the same period with continuous measurements are shown by red and blue solid lines, respectively. cannot currently be achieved due to the unavailability of more detailed distributions of surface fluxes (currently 1° × 1° in horizontal and monthly time intervals). Tagged tracer experiments were performed using the ACTM with the T106 resolution, employing prescribed oceanic fluxes of O 2 , N 2 and CO 2 , to estimate the contributions of five different sea regions to the observed APO variations. The APO values predicted for the different regions during 9 May-29 June 2013 and the total are shown in Fig. 7a. The total APO values predicted for the five regions agree with the observed variations fairly well (Fig. 5b), although the model simulations show smaller short-term fluctuations than the observation, mainly due to prescribed oceanic O 2 , N 2 and CO 2 fluxes on a monthly basis and partly due to uncertainties in the atmospheric transport of the ACTM. The contributions of the regions to the total APO, derived from the results shown in Fig. 7a, are given in Fig. 7b. The tagged tracer experiments show that the APO value stabilizes temporally when the contributions from areas far from Ny-Ålesund take precedence over those from its peripheral areas. This result is likely due to the mixing of air during its transportation over long distances. For example, the . Each value plotted for δ(O 2 /N 2 ), CO 2 and APO was obtained as a deviation from the corresponding best-fit curve shown in Fig. 2. Grey shaded times represent the time intervals regarded as 'high APO', defined as over 10 per meg of ΔAPO, in the backward trajectory calculations (see Fig. 6). The vertical axis of APO is scaled 2 times compared to that of δ(O 2 /N 2 ). Fig. 6. Ten-day backward trajectories calculated for air parcels released from Ny-Ålesund (yellow star) every 6 h from 9 May to 29 June 2013. Red and blue lines indicate the trajectories for high and low APO, respectively. 7 ATMOSPHERIC POTENTIAL OXYGEN AT NY-ÅLESUND, SVALBARD with high NPP reached the Ny-Ålesund site. Air that arrived at Ny-Ålesund after moving over the Arctic Ocean, Greenland or other low-NPP areas resulted in low or stable APO values. Oceans with high NPP emit O 2 through marine biological activity, while oceans covered with sea-ice have a significantly reduced O 2 exchange between the atmosphere and ocean. The observed short-term fluctuations in APO at Ny-Ålesund in spring and early summer would be thus likely to be caused by the O 2 emissions from high marine biological production in the seas around the Svalbard Islands. In this regard, Yamagishi et al. (2008) found similar short-term APO fluctuations in the results of continuous measurements at Cape Ochi-ishi (43°10′N, 145°30′E), Japan. By comparing observed APO variations with distributions of marine NPP, using a back trajectory analysis, they reported that such fluctuations are associated with the spring phytoplankton bloom in the Okhotsk Sea and the Western North Pacific.

Estimates of oceanic O 2 flux and marine NCP
Here, we estimate sea-to-air fluxes of O 2 using the observed APO fluctuations and backward trajectory analysis for spring very small APO fluctuations between 6 and 18 June 2013 are produced predominantly by air masses transported from Region 5 (the rest of the world excluding Regions 1-4), while Regions 1 and 5 show near-constant APO values between 11 and 20 May 2013. The tagged tracer experiments also indicate that Regions 2 and 3, which correspond to the Greenland Sea, the Norwegian Sea and the Barents Sea, are the main contributors to the observed short-term APO fluctuations at Ny-Ålesund in spring and early summer. This is consistent with the comparison of backward trajectories.
To examine the observed short-term APO fluctuations in more detail, we selected some characteristic events, which are indicated by green arrows in Fig. 5b. The predicted trajectories for the respective events are presented in Fig. 8a for May 2013 and Fig. 8b for June 2013. Figure 8c and d shows the distribution of monthly average marine net primary production (NPP) for May and June, as obtained from the vertically generalized production model (VGPM; Behrenfeld and Falkowski, 1997, http://www.science.oregonstate.edu/ocean.productivity/standard.product.php). By comparing the trajectories with the NPP distributions, high APO values were found to occur when air from the Norwegian Sea, the Greenland Sea or the Barents Sea  Fig. 5b. During this period, APO increases rapidly by about 50 per meg and then declines to close to the former level. The backward trajectory for and early summer, assuming that the observed fluctuations are caused by O 2 emissions from marine biological production. As an example, we first focus on the event observed between 17  23 June 45 2.2 × 10 5 1.9 × 10 3~2 .0 × 10 3 CO 2 exchange is responsible for the observed CO 2 dips shown in Fig. 5a, then the APO value is also affected by the oceanic CO 2 flux. In this case, the sea-to-air O 2 flux derived from the observed APO is overestimated. However, the overestimation of the O 2 flux was calculated to be 13% at most for the case of b-5, on the assumption that all of the observed CO 2 dip is produced only by the ocean. In addition, since the observed APO fluctuations include the effect of atmospheric transport to some extent, the calculated NCP is possibly overestimated. Considering these uncertainties, it is difficult to strictly quantitatively compare NCP, derived in this study, with NPP obtained from satellite observations. However, the present results suggest that the continuous record of APO at Ny-Ålesund, especially for spring and early summer, is useful for detecting marine biological productivity signals in its peripheral seas.

Concluding remarks
To contribute to a better understanding of the global carbon cycle and air-sea O 2 exchange, continuous in situ measurements of atmospheric δ(O 2 /N 2 ) and CO 2 mole fraction were started at a remote station Ny-Ålesund in November 2012. This instrument set-up is controlled remotely, as a first trial in the Arctic region, serviced annually by the personnel from Tohoku University and NIPR. From the measurements, a continuous record of APO was derived. The results of the first two years of measurements show a clear seasonal cycle in δ(O 2 /N 2 ), CO 2 mole fractions and APO. We also performed numerical simulations of APO using the JAMSTEC's ACTM with prescribed oceanic O 2 , N 2 and CO 2 fluxes. Previous studies using a discrete flask sampling technique reported that the observed APO value is clearly higher than the model-simulated value in autumn, but such a difference was not found in this study for the short period of 2013-2014 (correlation coefficient = 0.99 and slope = 0.86 between the observed and simulated values). To better understand the interannual variability of the seasonal APO cycle, longer term measurements of APO and further model studies are required.
Irregular fluctuations in APO were observed, especially in spring and early summer, showing a rapid increase and decrease of up to 50 per meg on a timescale of hours to days. A comparison of the observed APO with model simulations and backward trajectories with marine NPP distributions suggests that such fluctuations are possibly caused by O 2 emissions associated with marine biological production in the Greenland Sea, the Norwegian Sea and the Barents Sea. The sea-to-air O 2 fluxes in the peripheral seas of Ny-Ålesund are estimated to be approximately 2.8 ± 1.5 × 10 5 μmol m −2 day −1 by analysing selected events of irregular APO fluctuations, using a well-mixed moving column model. The average marine NCP calculated from these sea-toair O 2 fluxes is about 2.4 ± 1.3 × 10 3 mgC m −2 day −1 , comparable with NPP estimated using the VGPM for seas with high biological production. These results indicate that our measurements the time when the highest APO value was observed is shown in Fig. 8b by the purple line labelled b-5.
To estimate the sea-to-air O 2 flux, we used a well-mixed moving column model (Jacob, 1999). In this simple model, it is assumed that the vertically well-mixed air column, with a finite height, arrives at the observation site while being supplied with O 2 from the ocean. The average O 2 flux (F) into the column during transport (i.e. the sea-to-air flux) is expressed by Jacob (1999) as follows: where ΔC is the increase of APO in the column, h is the vertical mixing height, L is the wind fetch, u is the wind speed and t is the e-folding lifetime for the dilution of APO in the column. To calculate the flux, we used 50 per meg, 280 m, 1000 km, 4.4 m s −1 and 12 h for ΔC, h, L, u and t, respectively. A ΔC value of 50 per meg was taken from the observations, and t of 12 h was employed in accordance with Lueker (2004) and Thompson et al. (2007). The remaining values were determined by the following procedures. First we estimated a location where an air parcel released from Ny-Ålesund moved outside of the boundary layer. Then, the average vertical mixing height and average wind speed were calculated. The boundary layer heights and wind speeds along the air parcel trajectory leading to that location were used, while the wind fetch was calculated by integrating the distance along the trajectory. Under the given conditions, the sea-to-air O 2 flux was calculated to be 2.6 × 10 5 μmol m −2 day −1 . By assuming an -O 2 :C exchange ratio of 1.4 for marine organic matter (Laws, 1991;Anderson, 1995), NCP was estimated to be 2.2 × 10 3 mgC m −2 day −1 from the calculated sea-to-air O 2 flux.
The same procedures as above were applied to the other events designated as a-2, a-3, a-5, b-1, b-2 and b-6 in Fig. 5. The results are summarized in Table 1, together with the relevant parameter values. The sea-to-air O 2 fluxes estimated from the events are between 1.3 × 10 5 and 5.0 × 10 5 μmol m −2 day −1 , with an average of 2.8 ± 1.5 × 10 5 μmol m −2 day −1 , showing a good agreement with each other. Similarly, a good agreement is found in the calculated NCP, which range between 1.1 × 10 3 and 4.3 × 10 3 mgC m −2 day −1 , with an average value of 2.4 ± 1.3 × 10 3 mgC m −2 day −1 . These values of NCP are fairly close to the NPP values of 0.5 × 10 3 -2.0 × 10 3 mgC m −2 day −1 derived using the VGPM for sea areas with high biological production. The values are also consistent, within an order of magnitude, with NPP in seas where the backward trajectories predicted air parcels passed before arriving at Ny-Ålesund ( Fig.  8a and b).
The sea-to-air O 2 fluxes and NCP estimated using the simple model in this study have more or less uncertainties. For example, in the case of b-5, change of 50 m in the vertical mixing height yields approximately 0.7 × 10 5 μmol m −2 day −1 for sea-toair O 2 flux and 0.6 × 10 3 mgC m −2 day −1 for NCP. If the air-sea (4) F = ΔCh t{1 − exp(−L∕ut)} of APO at Ny-Ålesund are useful for examining marine biological productivity in the surrounding seas.
It was recently reported that annual NPP in the Arctic Ocean increased by about 20% between 1998 and 2009, due to an increase in the extent of open water by the ongoing decrease of sea ice (Arigo and van Dijken, 2011). It is likely that the increase in NPP will enhance oceanic CO 2 uptake, changing the carbon cycle in the Arctic Ocean. However, it is also recognized that there is a large CO 2 variability in the surface layer of the Arctic Ocean, meaning it is not always a significant CO 2 sink (Cai et al., 2010). Our continuous observations of atmospheric δ(O 2 /N 2 ) and CO 2 mole fraction at Ny-Ålesund can play an important role in detecting possible changes in the carbon cycle in the Arctic region during the near future.