The sea ice surface, not open water, is the dominant source of sea salt to aerosol and ice cores in coastal Antarctica. Here, we show that it may also form the dominant source for central Antarctica. We can then explain higher concentrations in the winter and last glacial maximum (LGM) as being due to increased sea ice production. This suggests that ice core sea salt can indicate at least the timing of changes in Antarctic sea ice production. The pattern of sea salt in ice cores is consistent with marine evidence about sea ice changes in the Holocene and LGM. Sea salt shows no change at the initial CO2 increase out of the last glacial, making it unlikely this was primarily due to changing sea ice cover. The sea salt record should not be treated as an indicator of meridional transport.
 The extent of Antarctic sea ice affects global albedo, and ocean-atmosphere heat exchange. Some studies hold that, through its effect on deep water ventilation, it controlled glacial-interglacial changes in atmospheric CO2 concentrations [Stephens and Keeling, 2000]. Sea ice production leads to Antarctic bottom water formation; some authors believe that this effect is a dominant influence on the strength of glacial thermohaline circulation [Shin et al., 2003]. To understand and model Quaternary climate we must therefore obtain good records of changes in sea ice production and extent.
 Diatom records from marine cores show whether sea ice was present at a site, and for which times of year [Crosta et al., 1998; Shemesh et al., 2002]. From a network of sites, a picture of past sea ice extent may be established. The most promising proxy for sea ice from ice cores has been methanesulfonic acid (MSA), derived from marine biogenic emissions. It has been suggested that the concentration of MSA in coastal ice cores depends on the sea ice extent of the previous winter [Welch et al., 1993], and can be used as a record of sea ice extent over recent decades [Curran et al., 2003]. However, a good indicator of sea ice conditions over longer time periods is still needed.
2. The Open Water Source of Sea Salt
 Sea salt aerosol is one of the major impurities in polar ice cores. Until recently, it was assumed that the source was bubble bursting over open water. With this assumption alone, greater sea ice extent would lead to lower sea salt flux to Antarctica. It has therefore been difficult to explain why concentrations in aerosol and snow generally peak in the winter half-year [e.g., Bodhaine et al., 1986; Wagenbach et al., 1998], and were considerably higher in the LGM [Legrand et al., 1988; Watanabe et al., 1999].
 To explain this, most authors have suggested that storminess over the ocean, and transport inland, are enhanced in winter and in glacial periods, more than offsetting the greater distance [Legrand et al., 1988; Petit et al., 1999]. However, the spatial distribution of sea salt in snow when going inland from the coast [Mulvaney and Wolff, 1994], or across a flat ice shelf [Minikin et al., 1994], suggest that the depletion in sea salt load across 1000 km of (source-free) sea ice (typical for winter (Figure 1)) would be a factor 20.
 Wind speeds at Signy Island, near the edge of the sea ice zone, peak in spring and autumn, with only a modest seasonality (Figure 2). Using typical formulations for sea salt aerosol uplift, the seasonality in source strength would be less than a factor 2, which is also an upper limit to the seasonal variability of sea salt aerosol observed at Southern Ocean sites [Gong et al., 2002]. Meridional transport to central Antarctica from mid-latitudes is strongest in winter, but the seasonality is already very modest for tracers from Patagonian latitudes [Krinner and Genthon, 2003; Lunt and Valdes, 2001]; from the sea ice zone, the seasonality should be even smaller. In summary, there is little evidence that these factors can overcome the large reduction due to the increased distance to open water.
3. The Sea Ice Source of Sea Salt
 Recently, clear evidence has emerged that, for coastal Antarctic sites, the primary source of sea salt aerosol is not open water, but the sea ice surface [Rankin et al., 2002; Wagenbach et al., 1998]. Newly formed sea ice is covered in a highly saline brine and, in many cases, by fragile and equally salty frost flower crystals. These form a very effective source of sea salt in winter. Because mirabilite (sodium sulfate decahydrate) starts to precipitate below -8°C, the aerosol formed from them is strongly depleted in sulfate [Rankin et al., 2002], which explains the frequent observations at near-coastal sites of negative non-sea-salt (nss) SO42- in both aerosol and snow ([nss SO42-] = [SO42-] − a [Na], where a = [SO42-]/[Na] in sea water). Other data [Hall and Wolff, 1998; Rankin et al., 2002] confirm that air masses coming from fresh sea ice covered in frost flowers have high salt concentrations and negative nss SO42-. These episodes dominate the sea salt budget in coastal regions [Rankin et al., 2002]. Strong negative nss SO42-, diagnostic of an important sea ice surface source, has been observed in ice cores at sites some distance inland on the Ronne Ice Shelf [Minikin et al., 1994], at Berkner Island [Wagenbach et al., 1994], and at Siple Dome [Kreutz et al., 1998].
 It is harder to assess whether this source is also important for inland plateau sites; they have low sea salt concentrations, and a relatively high year-round background of nss sulfate from other sources, which masks the depletion in sea salt SO42-. Few winter aerosol measurements have been made at inland sites. However, there is a wide range of circumstantial evidence. (1). The few published data on Na and nss SO42- in South Pole aerosol show that the sea salt budget is dominated by a few events, mostly in the winter half year; in such events, the nss SO42- concentration falls sharply [Bodhaine et al., 1986]. A likely explanation is that the sea salt aerosol is depleted in nss SO42- at source. Analysis of one major salt event supports this [Harder et al., 2000]; the weight ratio in the increase of SO42- to that of Na is 0.11, similar to that in frost flowers, rather than 0.25 as in sea water. (2). It is difficult to explain why a sea ice source is important for near-coast sites but that a lower latitude open water source is more important inland. (3). The winter peak argues against an open water source, but is consistent with a sea ice source. (4). Similarly it seems unlikely that the open water source would be enhanced under glacial conditions. Measurements of the size distribution of dust in Antarctic snow [Delmonte et al., 2002] suggest that the meridional transport to Antarctica was not significantly greater at the LGM than the present, a view confirmed by aerosol transport models [Krinner and Genthon, 2003; Lunt and Valdes, 2001]. In one recent experiment, the modelled LGM concentration of sea salt at Vostok from an open water source alone was half that of the present-day [Reader and McFarlane, 2003]. This implies flux reduced by a factor 4, compared to an observed doubling.
 We therefore suggest that the surface of fresh sea ice, including frost flowers, is the major source of sea salt to the Antarctic plateau also. Production of frost flowers and brine should be determined by the amount of new sea ice production, in coastal and other leads. This peaks in winter and is stronger in the LGM. The amount of this material reaching inland sites will additionally be affected by: (1). the likelihood of it being covered by snow before it is blown away; (2). the location at which production takes place; (3). the likelihood of transport being inland: the evidence about tracer transport [Krinner and Genthon, 2003; Lunt and Valdes, 2001] suggests that this effect is small, both between winter and summer, and between present and LGM; (4). the amount that is lost by deposition during transport. Further work is needed to establish the importance of these factors, but at least for the Holocene, we expect them to be secondary to changes in the amount of sea ice production. Because of the very episodic nature of transport from a brine production area to an ice core site, it is unlikely that the proxy will work on an annual basis, but it should give an indication of sea ice production when averaged over decades or centuries.
 Additionally, since, at least in current conditions, the summer sea ice extent is very small, most sea ice forms anew each year, and the area of sea ice produced annually integrates to the winter sea ice extent. We therefore expect annual sea ice production, and hence sea salt flux to central Antarctica, to be positively related to sea ice extent over the region of influence.
 For the very different conditions of the last glacial period, other factors will play a major role. Although it is still discussed [Crosta et al., 1998; Gersonde and Zielinski, 2000], there may have been a significant increase in multi-year ice, loosening the link between production and extent. The extension of the large ice shelves over the continental shelf will have altered the location of some sea ice production. Because sea salt reaches central Antarctica in discrete episodes, with trajectory times from the coast that are only 1–2 days [Bodhaine et al., 1986], changes in the overall hydrological cycle should not affect the residence time significantly, but changes in the strength, frequency or location of these episodes could be important. As a result, sea salt in ice cores cannot yet be used quantitatively to indicate sea ice production in the glacial period, but nonetheless any major change in sea ice production should be seen as a simultaneous change in ice core sea salt flux.
4. The Sea Salt Record
 We now examine Antarctic ice core records, particularly from Dome C, Antarctica (Figure 1). Dome C is far from the coast, and probably influenced most strongly by the Ross Sea and Indian Ocean. We use Na to indicate sea salt, but make a small correction to remove the influence of Na from terrestrial sources (Figure 3) [Rothlisberger et al., 2002]. Because the snow accumulation rate (based on the glaciological and accumulation models used to date the ice core [Schwander et al., 2001]) is so low, dry deposition dominates the flux to the snow surface, and it is therefore appropriate to use the Na flux, rather than concentration, as the indicator for sea ice production at these sites.
 The main feature of the Holocene is an increase of Na flux by 30% during the last 5000 years. A similar increase has been noted for the Taylor Dome core, bordering the Ross Sea [Steig et al., 2000]. Our new interpretation is that winter sea ice production (and probably extent) has increased in both the Ross Sea and the wider region that feeds Dome C. Diatom assemblages in the Ross Sea [Steig et al., 1998] independently suggest that sea ice cover has increased over the last 6 kyr. Na data from Dome Fuji [Watanabe et al., 1999], in the Atlantic sector, suggest a Holocene increase there also. Again, marine evidence [Hodell et al., 2001] already suggests that sea ice in the Weddell Sea expanded around 5 kyr B.P. The ice core and marine evidence strongly suggest that sea ice production and extent have increased widely around Antarctica over the later half of the Holocene.
 The LGM sea salt flux at Dome C is roughly double that of the early Holocene. A slightly larger change is estimated for Vostok [Petit et al., 1999], and a similar estimate can be made for Dome Fuji [Watanabe et al., 1999]. Previously this was interpreted as being due to a greatly increased meridional transport, offsetting an assumed increase in sea ice extent. Although changes in transport play a role, we conclude instead that the main reason for increased sea salt in the ice cores is that winter sea ice production on both the Indian and Atlantic sides of Antarctica was significantly increased in the LGM. Diatom reconstructions [Crosta et al., 1998; Gersonde and Zielinski, 2000] suggest that maximum sea ice extent was between 5 and 10 degrees further north at the LGM than the present, giving approximately double the areal coverage. With this new interpretation, taken together with evidence that increased dust at the LGM was due to changes in South American climate [Rothlisberger et al., 2002], there is no longer any ice core evidence to support the often cited view that meridional transport to Antarctica was greatly increased at the LGM compared to the present. Model studies have found only limited or no [Krinner and Genthon, 2003; Lunt and Valdes, 2001] increases in such transport, so our conclusion removes a perceived discrepancy between modelling and observations.
 Finally, there are competing explanations for the difference in CO2 concentration of the atmosphere between the LGM and the Holocene; several of them involve changes in sea ice extent [Gildor et al., 2002; Stephens and Keeling, 2000]. Assuming that a decrease in extent was caused mainly by decreased annual production, we would expect to see a significant change in sea salt. From the detailed CO2 record (Figure 3) of the transition [Monnin et al., 2001], much of the CO2 increase was complete before there was any significant change in Na flux, implying that sea ice changes are very unlikely to be responsible for the initial rise from 185 to 240 ppmv, although they could play some role for the final increase to 270 ppmv.
 The previous paradigm, that sea salt in Antarctic snow derives from open water, is at odds with the evidence. The sea ice surface provides a plausible alternative source, although it remains difficult to find definitive proof at inland sites. Further high-resolution aerosol analyses of Na and SO42- at inland sites are certainly needed. Process studies in the sea ice zone could establish how we should interpret a sea ice indicator for periods with different climate.
 Similar estimates to those we have presented should now be made for cores facing each sector of Antarctica, so that a picture of sea ice production around Antarctica can be built up, with coastal cores providing a more regional view. These should be combined with the site-specific data available from marine cores to provide an integrated picture of sea ice extent at different time periods. Finally, the implications for Greenland sea salt records should also be considered.
 Part of this work contributes to the European Project for Ice Coring in Antarctica (EPICA), a joint European Science Foundation/European Commission (EC) scientific programme, funded by the EC and by national contributions from Belgium, Denmark, France, Germany, Italy, The Netherlands, Norway, Sweden, Switzerland, and the U.K. The Dome C CO2 data were supplied through WDC for Paleoclimatology at Boulder, Colorado.