Coastal polynyas are areas in an ice-covered ocean where the ice cover is exported, mostly by off-shore winds. The resulting reduction of sea ice enables an enhanced ocean-atmosphere heat transfer. Once the water temperatures are at the freezing point, further heat loss induces sea ice production. The heat exchange and ice production in coastal polynyas in the southwestern Weddell Sea is addressed using the Finite-Element Sea-ice Ocean Model, a primitive-equation, hydrostatic ocean circulation model coupled with a dynamic-thermodynamic sea-ice model, which allows to quantify the amount of heat associated with cooling of the water column. Three important polynya regions are identified: at Brunt Ice Shelf, at Ronne Ice Shelf and along the southern part of the Antarctic Peninsula. Multiyear winter means (May–September 1990–2009) give an upward heat flux to the atmosphere of 311 W/m2 in the Brunt polynyas, 511 W/m2 in Ronne Polynya and 364 W/m2 in the Antarctic Peninsula polynyas, whereof 57 W/m2, 49 W/m2 and 48 W/m2, respectively, are supplied as oceanic heat flux from deeper layers. The mean winter sea ice production is 7.2 cm/d in the Brunt polynyas corresponding to an ice volume of 1.3 ×1010 m3/winter, 13.2 cm/d at Ronne polynya (4.4 ×1010 m3/winter), and 9.2 cm/d in the Antarctic Peninsula polynyas (2.1 ×1010 m3/winter). The heat flux to the atmosphere inside polynyas is 7 to 9 times higher than the heat flux in the adjacent area; polynya ice production per unit area exceeds adjacent values by a factor of 9 to 14.
 For the deep circulation of the world ocean, the production of dense water masses in the polar regions is a driving force of great importance. In the Southern Ocean, these dense water masses are formed on the continental shelves by cooling and salinification of the water. The salt enrichment is due to the salt rejection of freezing sea water. The highest freezing rates are encountered at coastal polynyas, areas where the sea ice is removed mechanically (usually by winds) while freezing conditions prevail. Here, the ocean-atmosphere interaction is hardly obstructed, and the heat flux is strongly enhanced compared to the ambient ice-covered ocean. Once the water is at freezing point, ice production is very high. As a consequence, the high brine rejection leads to the formation of a very dense water mass.
 The southwestern Weddell Sea is considered a major contributor to Antarctic Bottom Water (AABW) production [Orsi and Bullister, 1999]. On the wide continental shelves, water velocities are relatively small at 5–10 cm/s [Fahrbach et al., 1992; Foldvik et al., 2001; Kottmeier and Sellmann, 1996], and the salt rejected by ice formation accumulates in the water column. Coastal polynyas are the areas of the highest ice production during the winter months and thus contribute most to the dense water formation [Smith et al., 1990; Morales Maqueda et al., 2004]. It is therefore of great interest to understand the processes and quantify the fluxes at coastal polynyas in the southwestern Weddell Sea.
 In recent studies on polynyas in the Weddell Sea, Comiso and Gordon  found that strong meridional winds are related to large polynya areas and that years featuring large polynyas coincide with years of large sea ice extent in the Atlantic sector. They considered this a confirmation of the importance of polynyas in ice formation, although the large ice extent may simply be caused by the strong meridional winds leading to a faster northward transport, without directly affecting the ice volume. Markus et al.  identified polynya areas from SSM/I data and estimated heat flux and ice production using European Centre for Medium-Range Weather Forecasts (ECMWF) air temperatures and wind data while assuming constant relative humidity and an ocean temperature at the freezing point. Focusing on the Ronne Polynya, Renfrew et al.  established a more complete surface energy budget and derived ice production based on SSM/I data, automated weather station observations, and NCEP/NCAR Reanalysis data. Tamura et al.  estimated thin ice thickness from SSM/I data and calculated ice production using a heat flux model and ECMWF data for all major Antarctic polynyas including Ronne Polynya. For polynyas in the Weddell and Ross Seas, Drucker et al.  computed sea ice production with a heat balance algorithm from ECMWF air temperature and wind data, and ice motion and thin ice thickness derived from AMSR-E data.
 While all of the previous studies neglected the heat content of the ocean, the work presented in this paper uses a full three-dimensional ocean model and thus includes the oceanic heat fluxes. It aims to quantify polynya heat flux to the atmosphere, the heat flux due to ocean cooling and the resulting sea ice production during the winter months in the southwestern Weddell Sea, addressing multiyear mean as well as interannual variability. Fluxes within polynyas will be set into relation with fluxes in the ambient pack ice. In the following section, we will introduce the model and the data sets used, in sections 3–6, we present and discuss our results and conclude with a summary in section 7.
2 Model and Data
 The Weddell Sea coastal regions are difficult to access, especially in winter, when sea ice production is at its highest. Direct measurements are scarce and hard to obtain. Models provide a possibility to gain knowledge on a wide range of spatial and temporal scales. Their drawbacks and limits have to be kept in mind, however.
2.1 The Sea Ice-Ocean Model
 We used the coupled Finite Element Sea-ice Ocean Model (FESOM) [Danilov et al., 2004; Timmermann et al., 2009] to study the processes of polynya development, heat flux and sea ice formation at the coastal polynyas of the southwestern Weddell Sea. FESOM combines a hydrostatic, primitive-equation ocean model with a dynamic-thermodynamic sea-ice model.
 The ocean component of the model solves the horizontal momentum equation using the hydrostatic and Boussinesq approximations. The vertical mixing scheme follows Pacanowski and Philander  in combination with additional vertical mixing near the surface as proposed by Timmermann and Beckmann . Temperature and salinity are determined by the traditional tracer evolution equations.
 The sea ice component applies thermodynamics following Parkinson and Washington  and the elastic-viscous-plastic rheology suggested by Hunke and Dukowicz . A snow layer evolution depending on precipitation, air temperature, and melting processes is included in the model [Owens and Lemke, 1997]. Heat storage within the ice or snow layer is not considered. The ice drift is determined by wind stress, ocean surface velocity, sea surface slope and the internal forces of the ice, which are dependent on ice thickness and concentration following Hibler III . For the polynya in front of Ronne Ice Shelf, modelled ice drift has been shown to be in good agreement with remote sensing data [Hollands et al., 2013].
 The two model components communicate after each time step and exchange heat, salt, and momentum fluxes. Both share the same global unstructured horizontal grid with 3–5 km resolution close to the southwestern Weddell Sea coastline (Figure 1). The ocean model features 37 z-layers with increased resolution toward the surface (6 layers within the top 100 m).
 The atmospheric forcing in our study is derived from the NCEP/NCAR Reanalysis [Kalnay et al., 1996]. Daily data sets of 10 m wind velocity, 2 m temperature, 2 m specific humidity, precipitation rate, relative cloud cover and latent heat flux were used, from these evaporation and incoming longwave and shortwave radiation were calculated. The latent heat flux of the NCEP/NCAR data set was only used to calculate mass exchange by evaporation; the latent heat exchange between ocean and atmosphere was calculated independently as part of the sea ice/ocean surface energy balance (see below).
 As most data sets, also the NCEP/NCAR Reanalysis comes with uncertainties, especially in the Antarctic region where only sparse measurements could be used for data assimilation. The coarse resolution of their model grid also complicates the representation of small scale features like the high but narrow mountain range of the Antarctic Peninsula. The even smaller coastal polynyas and consequently the local effect of polynyas on the atmosphere are not represented in the data set. However, we assume that the steady offshore winds and the small width of the polynyas keep the magnitude of the effect small so that we consider the NCEP/NCAR Reanalysis to be a valid first approach to force the sea ice-ocean model. The forcing data were spatially interpolated from the 1.875° NCEP grid to the model grid points; also, the data were interpolated between the subsequent daily fields to each model time step.
 The model was initialized on 1 January 1980 with data from the Polar Science Center Hydrographic Climatology [Steele et al., 2001]. The time step was 3 min, and the results were recorded as daily mean values. For analysis, only data from the 20 year time period from 1990 to 2009 was used. The focus was put on the winter months May–September.
2.2 Heat Flux Components
 Following Parkinson and Washington , we split the heat flux to the atmosphere Qa into several components: The radiative heat flux is a combination of the shortwave radiative heat flux Qsw and the longwave radiative heat flux Qlw; the turbulent heat flux can be split into sensible heat flux Qs and latent heat flux Ql.
 Here, the radiative fluxes are calculated as net upward heat fluxes (i.e., upward flux minus downward flux), so that heat flux to the atmosphere is positive. The downward shortwave radiation is dependent on angular zenith distance of the sun ζ and inhibition by relative cloud cover C; part of it is reflected at the surface depending on its albedo α, so that
with the solar constant , the cloud factor [Laevastu, 1960] and the vapor pressure in the air ev,a in Pa. In the winter months at high latitudes, the shortwave radiation, if at all, is a very small contribution to the atmospheric heat flux.
 The longwave radiative heat flux is a function of the 2 m air temperature Ta, the surface temperature Ts and relative cloud cover:
with the emissivities of the ice/ocean surface ϵs=0.97 and of the atmosphere ϵa=0.765+0.22·C3[König-Langlo and Augstein, 1994] and the Stefan-Boltzmann constant σ.
 The sensible heat flux is determined by the ocean-atmosphere temperature difference and the 10 m wind speed u10following
with the specific heat of air cp, the density of air ρa, the heat transfer coefficient for sensible heat Cs=1.75×10−3over ice, snow and water [Maykut, 1977; Parkinson and Washington, 1979], and the surface temperature Ts. For open water, Ts is the temperature of the ocean model surface layer, for ice it is obtained as part of the diagnostic computation of the sea ice surface energy budget.
 The latent heat flux is the heat flux linked with evaporation, sublimation and their reversed processes. However, while the mass flux associated with evaporation is determined by the forcing data, the latent heat flux is re-calculated using wind speed and the difference between specific humidity at the surface Qs (where saturation is assumed) and at 2 m height Qa
with the heat of evaporation Le, the density of air ρa and the heat transfer coefficient for latent heat Cl=1.75×10−3 over ice, snow and water [Maykut, 1977; Parkinson and Washington, 1979]. Thus, we obtain a self-consistent energy budget while at the same time precipitation P and evaporation E are based on one data set, providing a good estimate for net precipitation P−E, which is important for the ocean surface freshwater budget in this global model.
 The heat flux to the atmosphere is partly compensated by eroding the heat content of the ocean and partly by the latent heat released in the process of ice formation. The former contribution results in an upward oceanic heat flux and causes a decrease in temperature of the water column. It is usually considered negligible, and the water is assumed to be at freezing point. However, intrusions of warmer water onto the continental shelf occur perennially [Foster et al., 1987] and are evident in the wintertime measurements on the southern continental shelf obtained from a CTD-equipped Weddell Seal in 2007 [Nicholls et al., 2008]. In contrast to satellite-data-based energy budget models, in this study, the ocean model provides us with data about the heat gained by ocean cooling.
2.3 Sea Ice Concentration Data
 To validate simulated sea ice concentration, we used the AMSR-E 89 GHz sea ice concentration data set [Spreen et al., 2008] without land mask, which we obtained from the Center for Marine and Atmospheric Sciences (ZMAW) in Hamburg, Germany. Since in the Weddell Sea, much of the coastline is determined by the ice shelf fronts and is thus highly variable, any prescribed land mask would very soon be outdated. In our comparisons, we therefore only mask the satellite data in areas of solid land (using RTopo-1 [Timmermann et al., 2010] as a reference) and mark the model's coastline in the figures for orientation.
3 Polynya Activity
3.1 Long-Term Mean
 A map of simulated mean sea ice concentrations in the southwestern Weddell Sea for the winter months May–September 1990–2009 (Figure 2) reveals that most of the southwestern Weddell Sea coastline features small but recurring polynyas. Three regions were defined: The Brunt region off Brunt Ice Shelf in the east, the Ronne region in front of Ronne Ice Shelf in the south and the Antarctic Peninsula region along the southern part of the Antarctic Peninsula in the west. These regions were chosen due to their enhanced polynya activity. The polynyas east of our Brunt region are located close to the continental shelf break or over the deep ocean. Thus, they are of little significance to the production of dense water masses [Fahrbach et al., 1994]. The area in front of Filchner Ice Shelf is not specifically considered, since the grounded iceberg A-23A (since 1986) and various smaller icebergs at Berkner Bank often lead to the formation of a fast ice bridge that prevents polynya formation in front of Filchner Ice Shelf [Grosfeld and Gerdes, 1998]. Markus  found that polynyas in this situation tend to open in the lee of the fast ice bridge. Since the iceberg is not included in the model, significant differences between model and reality can be expected in this area.
 The northern part of the Antarctic Peninsula was not included, since increasing air temperatures toward the north reduce the importance of polynyas there. In our simulation, their extent is likely to be overestimated, since the NCEP/NCAR Reanalysis model strongly smooths the topography of the Antarctic Peninsula. Thus, the westerly winds are too strong in the forcing [Windmüller, 1997; Stössel et al., 2011], causing excess off-shore sea ice drift in the simulation.
 The simulated sea ice thickness (Figure 3) features thin ice at the main polynya sites. Compared to the ice concentration, polynya signatures are visible over larger areas since the newly formed ice, while drifting away from the polynya, only slowly grows thicker over time. While at the northward and eastward facing fronts of Brunt Ice Shelf, the ice banks up and ice thickness has a local maximum, at the westward borders, we find very thin ice with thickness increasing in southwesterly direction. Along the Coats Land coast and especially in front of Filchner Ice Shelf, Berkner Island and reaching as far west as 54°W, the ice accumulates against the coastline; here, we find the maximum thickness. Farther west in front of Ronne Ice Shelf, thin ice is found which leaves a visible track on its northward drift. Another thickness minimum is found at the coastline of the Antarctic Peninsula between 72.5°S and 69°S, which again represents a polynya formation area.
3.2 A Case Study
 A polynya event (29 May–03 Jun 2008) is presented in Figure 4 to illustrate the performance of the model in producing coastal polynyas. In general, the model's ability to reproduce polynyas is very good in time and in space. Polynyas open at the same locations, during similar time intervals and to a similar width as observed.
 For day 150, the two polynyas at Brunt Ice Shelf exhibit similar size and ice concentration values in the simulation and in satellite observations. In front of Ronne Ice Shelf, the satellite data shows an opening of the Ronne polynya, which in the simulation is only visible as a trace of reduced ice concentration along the coastline. The polynyas along the Antarctic Peninsula agree well with observations, showing strong signatures in the northern part and very weak in the south. The satellite data additionally shows the signature of the fast ice bridge north of Filchner Ice Shelf and a flaw polynya on its western side.
 On day 152, the polynyas at the peninsula have closed. Ronne polynya and the Brunt polynyas have opened wider and simulation and satellite data agree well. Still, the satellite observations show a flaw polynya at the fast ice bridge, although smaller than on day 150.
 Three days later, on day 155, the satellite data shows Ronne polynya, the Brunt polynyas and the flaw polynya in the closing stage with rising ice concentrations. The simulation also features a weakened signature at the Brunt polynyas, while Ronne polynya is reduced in size, but still shows very low sea ice concentrations.
3.3 Integrated Polynya Area
 In this study, we define polynyas as the area where sea ice concentration A<70% or ice thickness hi<20 cm, thus polynya area is the corresponding area for all model nodes, which meet the criterion. We find a polynya size of 1013 km2 in the Brunt region in the 20 year mean over all winter (May–September) days. In the Ronne region, coastal polynyas on average cover an area of 1998 km2, which is twice the size of the Brunt polynyas; in the Antarctic Peninsula region, we find a mean winter polynya size of 1712 km2. Over the years, 2003–2009 (chosen due to the availability of the data), a comparison of the simulated polynya area to the polynya area derived with the criterion of 70% ice concentration from SSM/I data shows that the simulated daily mean polynya size is smaller in Brunt region by 40% and in Ronne region by 30%, while in the Antarctic Peninsula region, the simulated polynyas exceed the satellite-derived area by 68%. On a basin-wide scale, the various regional differences between polynyas from the SSM/I data and the simulation compensate each other very well, and the simulation underestimates the observation-derived value by only 10%. The overestimation of polynya size at the peninsula can largely be attributed to the fact that the westerly winds in the forcing data are hardly affected by the Antarctic Peninsula due to its coarse representation. Thus, the NCEP forcing has winds with an overestimated westerly component [Windmüller, 1997; Stössel et al., 2011], while observations indicate that the wind field east of the peninsula is often dominated by barrier winds [Schwerdtfeger, 1975; Parish, 1983].
 Interannual variability of polynya area is very pronounced, as indicated by a compilation of winter mean polynya areas (Figure 5). Seasonal means of half or twice the long-term mean are not uncommon. The range of polynya area in Brunt region spans from 440 to 2000 km2, which is rather moderate compared to the range in Ronne region (600–5200 km2) and at the Antarctic Peninsula (390–4100 km2).
4 Atmospheric Heat Flux
4.1 Multiyear Mean Within Polynyas
 Mean polynya values, here and in the following sections, are calculated from area-weighted daily averages over model nodes classified as polynya, i.e., all model nodes with an ice concentration <70% or ice thickness <20 cm. Days when no polynya is present are not considered (contrary to the calculation of the mean polynya area in the section above). During winter (May–September), we find a 20 year mean heat flux to the atmosphere of 311 W/m2in the polynyas of the Brunt region. The Ronne polynyas feature a mean of 511 W/m2; in the Antarctic Peninsula region the mean winter heat flux is 364 W/m2. The annual mean values and the standard deviations from the multiyear mean are found in Table 1. If all polynyas in the Southwestern Weddell Sea (Figure 2) are considered, the mean winter heat flux to the atmosphere is 368 W/m2. These mean values are calculated considering only the days when a polynya is present, thus they represent the mean flux that can be expected over a polynya in the corresponding region.
Table 1. Annual and 20 Year Winter Mean of Atmospheric Heat Flux and Ice Production of the Polynyas in Brunt, Ronne and Antarctic Peninsula (Ant. P.) Regiona
Heat Flux [W/m2]
Sea Ice Prod. [cm/d]
Accum. Sea Ice Prod. [km3]
Minimum and maximum values for each column is printed in italic font.
 While at the Brunt polynyas the highest heat flux occurs during the months July and August (≈ 345 W/m2), at Ronne polynya (≈ 550 W/m2) and at the peninsula coastline (≈ 420 W/m2), June must also be listed among the months with a particularly high polynya heat flux. These values are monthly means averaged over 20 years and thus are well below peak values that are possible under favorable circumstances.
 The annual mean heat fluxes to the atmosphere and their components at polynyas during the winter months May–September for the three most active polynya regions are presented in Figure 6. As might be expected, in winter, the shortwave radiation is the smallest contributor. Given that we define upward fluxes to have a positive sign and the reflected upward shortwave radiation is a fraction of the downward shortwave radiation, the net shortwave radiation must always be a negative number.
 The shortwave heat flux has an average of −4.7 W/m2at the Brunt polynyas, −4.5 W/m2at Ronne polynya and −7.7 W/m2at the polynyas along the Antarctic Peninsula (which are located a bit further north on average). The longwave radiation features a winter mean of 76 W/m2 over the Brunt polynyas, 116 W/m2 over Ronne polynya and 92 W/m2over the Antarctic Peninsula polynyas. The latent heat flux provides a slightly smaller contribution: 56 W/m2 in Brunt region, 71 W/m2 in Ronne region and 59 W/m2in Antarctic Peninsula region. In all polynyas, winter heat flux and its variability are dominated by the sensible heat transfer, which contributes 59–64% of the total heat flux. This fraction, as well as the relative contributions of latent and radiative heat fluxes, is in very good agreement with the findings of Renfrew et al.  during high-winter (June–July 1998). The simulated 20 year winter mean of the sensible heat flux over polynyas is 184 W/m2 in Brunt region, 328 W/m2 in Ronne region and 220 W/m2in the Antarctic Peninsula region. An overview over the atmospheric heat flux components within polynyas including annual minimum and maximum values and standard deviations is presented in Table 2.
Table 2. Twenty Year Winter Mean (Bold Font) of the Atmospheric Heat Flux Components and the Oceanic Heat Flux in the Polynyas of Brunt, Ronne and Antarctic Peninsula Region With the Annual Mean Minimum and Maximum Value as Well as the Standard Deviationa
All values are given in W/m2.
 The 20 year mean of heat gained by the atmosphere over the polynyas during the months May–September is 4.7±2.0×1018 J in the Brunt region, 14.7±9.0×1018 J in Ronne region and 7.7±4.8×1018 J in Antarctic Peninsula region. Renfrew et al.  estimated the mean heat gain by the atmosphere at Ronne polynya during the freezing season for the period 1992–1998 to be 3.48±0.98×1019 J. In their study, Renfrew et al.  defined the duration of the freezing period individually for every year. For the same time frame, our simulation yields an atmospheric heat gain of 3.76±1.14×1019 J, which implies (1) a good agreement between the two estimates and (2) the importance of a careful consideration of the time frame any estimate represents. Further comparisons of our results with independent studies are found in section 6 about sea ice production.
4.2 Multiyear Mean Outside Polynyas
 To assess the importance of polynyas in terms of heat transferred to the atmosphere compared to the ambient pack ice, we calculated the heat flux outside the polynyas in the three regions (also depicted in Figure 6). For the area with ice concentrations higher than 70%, the mean winter heat flux is 30 W/m2 in Brunt region, 70 W/m2 in Ronne region and 46 W/m2in the Antarctic Peninsula region, which is only 10–14% of the heat flux within polynyas, but due to the much larger area amounts to an atmospheric heat gain of 8.7±1.2×1020 J/season in the Southwestern Weddell Sea.
 In the Brunt region, most heat flux components are negligible outside polynyas and the longwave radiation is responsible for almost all the heat transferred to the atmosphere with an average of 31 W/m2. The latent heat flux over ice gives a small negative contribution (mean: −4.0 W/m2), which corresponds to resublimation of atmospheric humidity. In the Ronne region, due to the very low air temperatures, the sensible heat flux (mean: 26 W/m2) is of similar magnitude as the net longwave radiation (mean: 39 W/m2). Both other heat flux components are negligible; latent heat flux is positive (but very small) here. At the Antarctic Peninsula the heat flux over high ice concentration areas is mainly due to the longwave radiation (mean: 37 W/m2) although in years with low temperatures and strong winds the sensible heat flux can substantially add to this (mean: 9.1 W/m2). The modelled latent heat flux transfers heat out of the atmosphere at a rate of −1.8 W/m2.
4.3 Interannual Variability Within Polynyas
 Compared to the strong interannual variability found for polynya area (Figure 5), interannual variability of atmospheric heat flux (averaged over polynya days) (Figure 6) is much smaller and so is the variability of the key parameters air temperature, wind speed and specific humidity (Figure 7). A compilation of annual maximum and minimum values, multiyear mean and standard deviation of these three forcing parameters is found in Table 3.
Table 3. Twenty Year Winter Mean (Bold Font) of Wind Speed, Air Temperature and Specific Humidity in Brunt, Ronne and Antarctic Peninsula (Ant. P.) Region With the Annual Mean Minimum and Maximum Value as Well as the Standard Deviation
Wind Speed [m/s]
Air Temperature [°C]
Specific Humidity [g/kg]
 For the Brunt polynyas, we find the highest winter heat flux to the atmosphere in 2004 (385 W/m2) and the lowest atmospheric heat flux in 2009 (237 W/m2). Air temperature, wind speed and specific humidity feature 20 year winter means of −20.7°C, 6.5 m/s and 0.83 g/kg, respectively. For 2004, neither of these exhibit extraordinary values of the winter mean ( −20.0°C, 6.2 m/s and 0.87 g/kg), but the maximum in the turbulent heat flux components is easily explained when the mean of the daily product of air temperature and wind speed (Figure 7, bottom row) is considered, which represents the main driving parameter of the sensible heat flux. Short periods of strong winds and low temperatures leave only little trace in the means of the individual parameters, but when strong anomalies coincide, as, for example, in the case of cold air outbreaks, they produce a heat flux maximum that persists in the seasonal mean. Similarly, the heat flux minimum in 2009 is hard to explain by the seasonal average of air temperature (−19.5°C) and wind speed (6.4 m/s), but shows clearly in the mean product of air temperature and wind speed. Again, short-term events dominate the seasonal mean of the atmospheric heat flux.
 Of the three different regions, Ronne polynya features the coldest air temperatures in winter with a 20 year mean of −33.3°C, the highest wind speeds with a mean of 7.4 m/s and, mainly due to the cold temperatures, the lowest specific humidity of 0.29 g/kg. At Ronne polynya the minimum heat flux in 2008 is accompanied by the second-warmest air temperatures (−30.1°C), a close-to-average wind speed (7.6 m/s) and the second-highest specific humidity (0.41 g/kg). Therefore, sensible and latent heat flux are low and together cause the atmospheric heat flux minimum. The maximum in 1992 (590 W/m2), again, is not explained by looking at the seasonal means of the individual forcing parameters. It coincides with below-average air temperatures (−34.4°C), below-average wind speed (6.8 m/s) and below-average specific humidity (0.28 g/kg). However, a look at the mean product of air temperature and wind speed reveals a peak in 1992 that triggers the heat flux maximum.
 At the polynyas in the Antarctic Peninsula region, we find the 20 year winter mean of the air temperature to be −23.3°C, the mean wind speed is 7.0 m/s and the mean specific humidity is 0.73 g/kg. The by far largest atmospheric heat flux (509 W/m2) is in 1992. It coincides with the maximum in Ronne region, but is more easily explained by the coincidence of minimum air temperature ( −26.1°C), highest wind speed (8.7 m/s) and the minimum specific humidity (0.51 g/kg). The minimum winter heat flux of 278 W/m2in 2008 is just as obviously caused by the warmest air temperature (−18°C), the third-lowest wind speed (5.9 m/s) and the highest specific humidity (1.07 g/kg).
 Since the sensible heat flux is the main contributor to the atmospheric heat flux and the surface temperature in polynyas can be assumed to vary little, the mean of the daily product of air temperature and wind speed features a strong relation with the mean total atmospheric heat flux. Of course, not all details correlate, but it gives a very good first approach to the variability and individual seasonal means of the atmospheric heat flux.
5 Oceanic Heat Flux
 Sea ice production per unit area is strongly dependent on the heat flux to the atmosphere. If no further energy is supplied, a direct proportionality is expected. However, if the ocean is not at the freezing point, part of the heat loss to the atmosphere is compensated by the ocean's heat content, thereby cooling the water column. This part is expected to be highest in Brunt region, since in the east, the warm water of the Weddell Gyre enters upon the continental shelf as a coastal current, but also farther west intrusions of Modified Warm Deep Water occur [Foster and Carmack, 1976, Nicholls et al., 2008, 2009]. Originating from Warm Deep Water, its temperatures of up to 0.5°C are higher than the freezing point, but it experiences fast cooling on the shelf.
 In our simulation, the average oceanic heat flux due to the erosion of oceanic heat content is 57 W/m2 in the Brunt polynyas, 49 W/m2 in Ronne polynya and 48 W/m2in the Antarctic Peninsula region (all averaged over polynya days only) (Table 2). As expected, the oceanic heat flux is highest in the east, but does only slightly decrease and is still substantial in the western polynyas. The small difference in oceanic heat flux between the regions may partly be due to the warm water experiencing thorough cooling even before reaching the Brunt polynyas, but also attests the intrusion of above-freezing point waters to the farthest corners of the continental shelf. The contribution of the oceanic heat flux to the atmospheric heat flux is 19% in the Brunt polynyas, 10% in Ronne polynya and 13% in the Antarctic Peninsula polynyas in the long-term winter mean. So, although the absolute values of oceanic heat flux are similar for the three regions, in the Brunt region the relative contribution to atmospheric heat flux is higher by a factor of two.
 Outside polynyas, only 16 W/m2in Brunt region, 21 W/m2 in Ronne region and 20 W/m2in Antarctic Peninsula region are due to oceanic cooling, which is 30–57% of the total heat flux to the atmosphere outside polynyas. The fact that the oceanic heat flux is higher within polynyas than outside is easily explained by the increased convection under polynyas due to the higher salt enrichment. The seasonal means of the oceanic heat flux of the individual years are marked with a black line in Figure 6. As already explained, the difference between atmospheric and oceanic heat flux is compensated by latent heat gained from sea ice production.
6 Sea Ice Production
 The ice production per unit area in our simulation has a 20 year mean of 7.24 cm/d ( 11.1 m/winter) in the Brunt polynyas. In the Ronne polynya, the mean ice production is 13.23 cm/d ( 20.2 m/winter) and in the polynyas of the Antarctic Peninsula region, we found 9.21 cm/d ( 14.1 m/winter) (all averages over polynya days only). Since the oceanic heat supply within polynyas is a non-negligible, but small contribution to the atmospheric heat flux, minima and maxima of ice production per unit area (Figure 8) coincide with the minima and maxima of winter-mean atmospheric heat flux in most cases. At the Brunt polynyas, the highest mean winter ice production per unit area is in 2004 (9.6 cm/d 14.7 m/winter), while 1992 features the highest ice production per unit area in Ronne region (15.6 cm/d 23.9 m/winter) and the Antarctic Peninsula region (12.7 cm/d 19.43 m/winter). The lowest ice production in the Brunt polynyas is found in 2009 with 4.9 cm/d 7.5 m/winter. At the Ronne polynya, ice production minima are found for 1994, 2003, and 2008. The absolute minimum (2003) does not coincide with the smallest atmospheric heat flux: Atmospheric heat flux for 2003 is slightly higher than for 2008. Instead, 2003 stands out as the year with the highest oceanic heat flux (82 W/m2, Figure 5), which, combined with a small heat loss to the atmosphere, limits ice production to 10.94 cm/d ( 16.7 m/winter). The Antarctic Peninsula polynyas have their lowest ice production in 2008 (6.4 cm/d 9.8 m/winter), which again coincides with the absolute minimum of atmospheric heat flux.
 Outside the polynyas the regions feature ice production rates per unit area of 0.53 cm/d ( 0.8 m/winter) for Brunt region, 1.39 cm/d ( 2.1 m/winter) for Ronne region and 0.86 cm/d ( 1.3 m/winter) for the Antarctic Peninsula region. The ice production per unit area outside the polynyas is only 7% of the ice production within polynyas in the Brunt region, 11% in Ronne region and 9% in the Antarctic Peninsula region.
 If accumulated, the multiyear mean of ice production per winter season amounts to 12.9 km3 in the Brunt polynyas, 43.8 km3 in the Ronne polynyas and 21.0 km3in the polynyas along the southern Antarctic Peninsula. For the entire Southwestern Weddell Sea (Figure 2), polynya ice production in winter features a mean of 105 km3, which is 11% of the total ice production of 993 km3, but originates from only 0.6% of the area.
 The highest and lowest production of ice volume (Figure 9) depends very much on polynya area so that only in the Brunt region does the lowest (highest) annual polynya ice volume production coincide with the lowest (highest) annual ice production per unit area. In 2009, the Brunt polynyas form only 4.5 km3 of ice on an area of 479 km2, while their highest production is 22.8 km3 in 2004 on an area of 1403 km2. Both times, the coincidence with the minimum (maximum) ice production per unit area is amplified by a relatively small (large) polynya area.
 For Ronne polynya, the highest accumulated ice production over the winter months occurs in 1998 (88.8 km3), when the third highest ice production per unit area coincides with the second largest polynya area (4052 km2). The minimum total ice production is found in 1996 (12.6 km3), induced by a relatively low production per area and the smallest polynya area (603 km2).
 In the polynyas along the Antarctic Peninsula, the smallest amount of sea ice was produced in 2007 (3.1 km3), by the second lowest sea ice production per unit area and the smallest mean polynya area of only 390 km2. The maximum ice production occurred in 1998. Although the ice production per area was below average, the largest annual polynya extent (4079 km2), by far, more than compensated for it.
 Previous studies on the polynya sea ice formation in the Weddell Sea usually based their calculations on satellite observations of sea ice concentration (mostly SSM/I) and coarse global atmospheric data sets (often ECMWF or NCEP/NCAR). The heat flux to the atmosphere that resulted from the energy budget was converted into sea ice production using the assumption that the ocean surface is permanently at freezing temperature and that the oceanic heat flux can be neglected. Our study, although still dependent on a coarse-scale atmospheric data set, is independent of the satellite observations and includes the heat flux provided by ocean cooling, which turned out to be 10–20% of the heat flux to the atmosphere. We thus expect less ice production in our simulation than in previous studies that did not consider the ocean's heat content. Ice production rates based only on the heat flux to the atmosphere are prone to overestimate the true rates and must be regarded as an upper limit.
Markus et al.  determined the seasonal mean ice production in their southern region including Ronne polynya in the years 1992–1994 to be 87 km3. For the same time period, our simulation yields 77 km3in the Ronne polynya. The numbers agree well, although study areas do not match exactly, and the region of Markus et al.  includes the coastline of Filchner Ice Shelf.
 For the period 1992–1998, Renfrew et al.  calculated a mean ice production in Ronne Polynya of 24±5.1 m per unit area and a total of 111±31km3 during the full freezing season, which they individually identified for every year. Using the same intervals, our simulation gives 19.3±6.2 m per unit area and a total of 104±36km3. Tamura et al.  found the ice production from March to October in Ronne polynya to accumulate up to 85 km3as a mean over the years 1992–2001. For the same period, our model gives 89 km3. The good agreement in both cases is facilitated by Renfrew et al.  and Tamura et al.  using a study area containing a very similar part of the coastline as our Ronne region; it also indicates the robustness of our (and their) results.
 For the period of April–October 2003–2008, Drucker et al.  calculated a mean accumulated ice production of 112 km3/season for the Brunt polynyas and 99 km3/season for Ronne polynya. The corresponding values from our simulation are 89 km3/season in the Brunt region and only 50 km3in Ronne region. The differences can be explained by the negligence of the oceanic heat flux and the different extent of the study areas. Drucker et al.  include no locations with water depth over 1000 m and their eastern region stretches farther east and includes part of the polynya forming off Riiser-Larsen Ice Shelf. Also, their southern region extends farther west including a part of the polynyas, which we include in the Antarctic Peninsula region.
 In general, the comparisons show good agreement, and as expected, our values are slightly lower than the ice production calculated from satellite data. Only the study by Tamura et al.  gives a lower sea ice production than our simulation. However, the values are very close and the uncertainties in the data sets used and parameterization on both sides can easily explain this outcome.
 Using a coupled sea-ice ocean model forced with data from the NCEP/NCAR Reanalysis, we investigated the importance of coastal polynyas in the southwestern Weddell Sea in terms of heat transfer to the atmosphere, the heat flux supplied by ocean cooling and resulting sea ice production. We found the Ronne Ice Shelf front to be the region where the largest polynyas form and the highest atmospheric heat flux and sea ice production occur. Two other important regions were identified located at the Brunt Ice Shelf and along the southern part of the Antarctic Peninsula.
 The atmospheric heat flux over coastal polynyas during the months May–September has a 20 year mean of 313 W/m2 in Brunt region, 515 W/m2 in Ronne region and 374 W/m2in Antarctic Peninsula region. The interannual variability of the atmospheric heat flux is high and usually dominated by the variability of the sensible heat flux, which is the dominant contributor also in the long-term mean. Outside polynyas, the atmospheric heat flux is mostly determined by the longwave radiation budget, but still variability is ruled by the sensible heat flux component.
 The oceanic heat flux was found to be a non-negligible component of the ocean surface heat budget even at these high latitudes. Maximum values were found in the easternmost region next to Brunt Ice Shelf with 57 W/m2 in the 20 year mean due to the warm water entering upon the shelf, but is almost as high in both regions farther west with a little less than 50 W/m2, indicating that surface cooling does not prevent above-freezing point water from reaching far onto the continental shelf, which is consistent with observations presented by Nicholls et al. . Therefore, 10–20% of the atmospheric heat flux during winter at the polynyas in the southwestern Weddell Sea does not result in sea ice production, but is compensated by an erosion of the ocean's heat content.
 In the 20 year mean, we find a sea ice production of 7.2 cm/d at the Brunt polynyas leading to 12.9 km3 per winter, 13.2 cm/d at Ronne polynya creating 43.8 km3per winter and 9.21 cm/d in the polynyas along the Antarctic Peninsula giving 21.0 km3per winter. Keeping in mind the reduction of ice formation by the oceanic heat flux, our results compare very well with previous studies about atmospheric heat flux and sea ice production in Weddell Sea polynyas by Markus et al. , Renfrew et al. , Tamura et al.  and Drucker et al. .
 Within the investigated regions, areas with a sea ice concentration higher than 70% feature an atmospheric heat flux of only 11–14% and a sea ice production per unit area of 7–11% of the corresponding value within polynyas. Due to the small area of coastal polynyas in the Weddell Sea, the contribution of the much larger ice-covered ocean to heat exchange and sea ice production is prevailing on a large scale, primarily because the leads within the pack ice add up to a considerable area. Locally, however, the coastal polynyas are of paramount importance since here the ocean-atmosphere exchange is enhanced by an order of magnitude. Only the exceptionally high freezing rates and the local stability of coastal polynyas enable the salinity enrichment necessary for the production of dense shelf water, an indispensable ingredient for bottom water formation.
 The high interannual variability of the atmospheric heat flux and the sea ice formation in polynyas is dominated by the interannual variability of the atmosphere. Neither air temperature (T) nor wind speed (u) anomalies alone are sufficient to explain the interannual variability of the sensible heat flux; only seasonal averages of the product u' ·T' yield time series that correspond to those of heat flux and ice production anomalies. While this is a finding well familiar from turbulence theory, it still reminds us that looking at mean fields of atmospheric variables alone may be misleading when trying to assess potential impacts of future climate change.
 As a follow-up, two subsequent studies are underway: a detailed assessment of the consequences of the brine rejection entailed by sea ice production on the on-shelf water masses, and an investigation on how atmospheric forcing data sets with different resolutions influence polynya formation, heat flux and sea ice formation.
 The work for this study was funded by Deutsche Forschungsgemeinschaft in SPP 1158 under grant number TI 296/5. Special thanks go to our project partners G. Heinemann and L. Ebner from the Department of Environmental Meteorology, University of Trier, Germany, for inspiration and collaboration. The NCEP/NCAR Reanalysis atmospheric forcing data was obtained from NOAA Climate Diagnostics Center, Boulder, USA via the website http://www.cdc.noaa.gov.