Understanding CO2 exchange between atmosphere, ocean, and land is important, because marine and terrestrial environments are currently absorbing about half of the CO2 emitted by fossil-fuel combustion. Where and how land and ocean vary in their uptake of CO2 from year to year has been the subject of much scientific research and debate [Ciais et al., 1995; Everett et al., 1998; Feely et al., 2001; Prentice et al., 2001].
 The Arctic Ocean's role in determining regional CO2 balance has been ignored because continuous sea-ice cover is considered to impede gaseous exchange with the atmosphere so efficiently that no global climate models include CO2 exchange over sea-ice [Tison et al., 2002]. However, pioneering measurements by Gosink and Kelley in the 1960–70s [Gosink et al., 1976; Kelley and Gosink, 1979] showed one year sea-ice was highly permeable to CO2 at temperatures above −15°C through numerous tiny channels. Mechanism involved in this process has been roughly discussed. In this paper we present new data concerning the carbon flux across the sea ice and dynamics of the carbonate system in the Amerasian Arctic seas.
 The leads, polynyas, and melt ponds of the Arctic Ocean can be significant sinks of air CO2 in summer and sources in winter. Flight data [Bolin and Keeling, 1963; Kelley and Gosink, 1979] support this conclusion. Shipboard data show that low summer CO2 partial pressure (pCO2) in surface waters is rapidly replaced with high pCO2 in fall and winter [Kelley, 1970; Pipko et al., 2002; Semiletov, 1999a], concurrent with rising atmospheric CO2 levels at Barrow, Alaska. In synchrony with variations in atmospheric CO2, phytoplankton blooms in perennial pack ice and ice algae density over the Arctic Basin and continental shelves [Gosselin et al., 1997; Melnikov, 1997] also show large variation in primary and secondary production.
 To determine the impact of CO2 transport through sea-ice of the North American-Siberian Arctic Ocean shelf zones, we conducted observations in two representative areas: 1) fast ice near Barrow, influenced strongly by marine input of organic material, and 2) the eastern Laptev Sea and western East-Siberian Sea, an area with high coastal erosion and fluvial input rates.
 Initial field investigations of CO2 exchange through melting sea-ice were conducted for the first time on fast ice near Point Barrow, June 6–24, 2002 using previously described eddy correlation [Oechel et al., 1998] and chamber techniques [Zimov et al., 1993] which were applied early only for land-based studies. Figures 1a and 1b show data from tower-based turbulent CO2 flux measurements. Flux magnitude was typically 0.01–0.02 mg m−2 s−1, close to values previously measured in transects above Alaskan North Slope tundra [Oechel et al., 1998]. The turbulent flux of CO2 (FCO2) measured above fast ice during onshore winds (Figure 1a) becomes more negative with time, indicating increasing CO2 absorption by the sea-ice surface from air masses originating over the Arctic Ocean, which contain higher CO2 concentrations than “terrestrial” air masses. Absolute values are similar for offshore winds, but the FCO2 trend is reversed (Figure 1b). The positive trend of FCO2 in Figure 1b may result from a decreasing CO2 concentration in air masses coming across Alaska to the sea-ice surface, as burgeoning photosynthesis depletes tundra air CO2 in the second half of June (Y. Harazono, personal communication, 2002). In general, flux observations show significant absorption of air CO2 by the melting sea-ice surface. Note, that in fall the situation can be reversed and “terrestrial” air masses may contain higher CO2 concentrations than “marine” air masses because soil respiration becomes dominate over the land.
 Measurements of pCO2 in sea-ice brines and in under-ice water (Figure 1c) demonstrated significant CO2 undersaturation of 220–280 μatm and 130–150 μatm, respectively, compared to air concentrations of 365–375 μatm. The chamber data show a drastic decrease of equilibrium CO2 concentration (down to 300 μatm) in the headspace above growing melt ponds, especially during the last four measurement days, when daily mean temperature rose above 0°C and melt pond depth increased dramatically from a few cm to 20 cm (Figure 1d). Almost all chamber CO2 flux measurements show atmospheric CO2 uptake by the sea-ice surface. Chamber CO2 influx ranged between 0.002 and 0.003 mg CO2 mg m−2 s−1 at sites with water depth = 0–2 cm, and between 0.003 and 0.026 mg m−2 s−1 at sites with water depth = 2–5 cm.
 These results confirm the turbulent flux measurements and reflect increased incoming solar radiation absorption by melt ponds and beneath sea-ice. We suggest that increased photosynthetically active radiation (PAR) may cause decreased pCO2 in and beneath sea-ice (Figure 1c) by enhancing photosynthesis. CO2 transport rates across sea-ice likely also increase because sea-ice becomes thinner and more gas-permeable when ice temperature increases [Gosink et al., 1976]. Some evidence of such processes can be seen by comparing drastically decreased air head-space CO2 concentrations in the chamber situated on dry ice near and also directly above a melt pond (curves 2 and 3, Figure 1d).
 The effects of summer temperature and PAR increase on the sea-ice/sub-ice water ecosystem is expected to enhance Arctic Ocean-scale sinks. We evaluated the significance of this Arctic sea-ice air CO2 sink, assuming that mean downward turbulent flux (−0.005 mg CO2 m−2 s−1) obtained over fast ice in June is representative of the entire Arctic Ocean ice during melting (>50 days). It is very a rough estimate though wide distribution of deep melt ponds and increased brine drainage over multi-year ice and flight data give us a clue that summertime multi-year ice can become gas permeable. Estimated total air CO2 absorption by sea-ice would be ≈150 Mt CO2 (40 Mt C) per summer. Updated global carbon budgets in the IPCC assessment [Prentice et al., 2001] show that the net terrestrial biosphere sink of atmospheric CO2 was ≈200 Mt C/yr in the 1980s, only five times more than mean uptake by sea-ice. We suggest that sea-ice plays a significant but hitherto unappreciated role in the spring/summer invasion of atmospheric CO2.
 Data obtained in the Buor-Khaya Gulf of the Laptev Sea, an area with high rates of coastal erosion and fluvial input [Semiletov, 1999b], show dramatic pCO2 changes in bottom and upper sea layers from summer, when the sea surface was ice-free (Figure 2a), to winter (Figure 2b), when fast ice thickness exceeded 2.0 m. In winter and early spring, when organic materials decomposition dominates, pCO2 exceeded 4,000 μatm near the bottom and 1,500 μatm near the surface. Both present and literature data [Kelley and Gosink, 1979; Semiletov, 1999a] show that CO2 supersaturation beneath sea-ice is a principal feature of carbon cycling in Arctic Siberian seas. Assuming an evolution of CO2 from porous one year sea-ice in the range 0.01–0.1g C m−2 day−1 (recalculated from Gosink et al.  and Kelley and Gosink ) and ≈2 × 106 km2 of Arctic Siberian sea-ice coverage, we estimate possible winter (November–May, ≈200 days) efflux from sea-ice over Siberian shelves to the atmosphere of ≈4–40 Mt C.
 In contrast, in summer and early autumn, due to dominant photosynthetic processes above a strong pycnocline, surface water over the Laptev sea study area was undersaturated by CO2 (≅250 μatm) compared to the atmosphere. Similar data were obtained over other Siberian and Alaskan shelves [Kelley, 1970; Pipko et al., 2002; Semiletov, 1999a]. Note that bottom waters were significantly oversaturated (up to ∼10 times) by CO2 year-round due to high influx of old eroded terrestrial carbon, available for fast biodegradation [Semiletov, 1999b]. The highest pCO2 values in the marine environment may exist over the East Siberian shallow shelf (I. Semiletov et al., Air-sea CO2 exchange in the Arctic, manuscript in preparation, 2004) where biogeochemical consequences of coastal erosion are most pronounced.
 New information on pCO2 dynamics beneath fast ice was obtained February 17–March 23, 2003, when mooring-based pCO2 observations were executed with the autonomous device SAMI-CO2 designed by De Grandpre et al. . The device was installed 2.5 m below the sea-ice surface and measured pCO2 at half-hour intervals in stable water temperature. In late February, pCO2 values increased to 400–410 μatm (that may be an effect of biological respiration), while in March a drastic pCO2 decrease from 410 μatm to 288 μatm was recorded. The decrease in pCO2 may reflect increased photosynthetic activity beneath sea-ice shortly after polar sunrise.
 We compared temporal variations in amplitude of CO2 seasonal cycles and summarized amplitudes of sea-ice extent in the central Arctic and the Laptev, East Siberian, Chukchi, and Beaufort seas for 1974–2001. The region for comparison was chosen after Parkinson et al. . Using available sea-ice data from http://polynya.gsfc.nasa.gov/seaice_datasets.html, we calculated the amplitude of Arctic sea-ice extent as the yearly difference between May and September areas. We found similar behavior of CO2 and sea-ice seasonal amplitudes (Figure 3a) and inverse behavior of annual mean CO2 concentration and sea-ice extent (Figure 3b). The calculated correlation coefficients, however, are rather small, especially if we calculate them after removing linear trends in the time series. The maximum correlation coefficient (0.61) was found between air CO2 concentration and sea-ice extent in September. Such low correlation is not surprising in view of the complicated behavior of gas exchange processes, partly described above, and the difficulties of comparing the data obtained at a point with data averaged over thousands of square kilometers. But the relation between changes in sea ice area, determined by opening or closing leads, especially, by seasonal variations of ice cover in the marginal seas, is evident.
 Decreased sea-ice extent and lengthening ice-free periods, longer photosynthetic activity, and absorption of air CO2 through leads and melt ponds in late spring and summer could be a significant unappreciated determinant of Arctic CO2 balance. Our results show that the spring/summer CO2 sink formed by sea-ice melt ponds and open brine channels must be included in any complete Arctic regional CO2 budget. Large, rapid changes in seawater pCO2 and vertical CO2 fluxes above/through sea-ice were found in two representative areas of the Arctic Ocean, indicating that pCO2 beneath sea-ice can be highly variable in near-shore environments. Extremely high marine pCO2 values indicating wintertime accumulation of CO2 beneath sea-ice exist over the shallow East Siberian shelf, where biogeochemical consequences of coastal erosion are most pronounced. In winter the shelf waters may be a source of CO2 through leads and open brine channels. The Arctic Ocean's contribution to the formation of the largest seasonal variations of air CO2 on Earth cannot be ignored.