Polynya area is derived from daily sea-ice concentration maps based on brightness temperatures of the Advanced Microwave Scanning Radiometer–Earth Observing System (AMSR-E; Spreen et al. 2008). AMSR-E sea-ice data have a resolution of 6.25 km, which is superior to SSM/I (12.5–25 km). All pixels with ice concentrations falling below a threshold of 70% (Massom et al. 1998) are treated as polynyas and subsequently an ice thickness of 0 or 10 cm is prescribed to COSMO for those areas. Adams et al. (2011) show that the 70% threshold yields polynya areas that agree well with those obtained by the polynya signature simulation method (Markus & Burns 1995). The ice thickness for sea-ice areas with ice concentrations exceeding 70% is set to 1 m. This value is in accordance with, for example, the value of 0.95 m used for the polar Weather Research and Forecasting model for ice concentration larger than 90% (Bromwich et al. 2001, 2005). Measurements of ice thickness in the Laptev Sea area by Rabenstein et al. (2013) show that ice is typically between 1.0 and 1.5 m thick in the vicinity of the polynya.

Atmospheric surface energy fluxes are the key variables when assessing polynya ice production. In our study, all fluxes were calculated by means of mesoscale model simulations. The non-hydrostatic, limited-area COSMO model was used. COSMO is the operational weather prediction model of the German Meteorological Service (Schättler et al. 2012). In addition to the operational application, the COSMO model provides a modelling framework for many scientific purposes, ranging from short case studies to climate simulations (e.g., Schröder et al. 2011; Gutjahr & Heinemann 2013). Turbulent fluxes at the surface are calculated using the Louis scheme (Louis 1979). Radiation fluxes are calculated according to Ritter & Geleyn (1992). A full description of the physical parameterizations is given by Doms et al. (2011).

Version 4.11 of the model was implemented with a horizontal resolution of 5 km for the Laptev Sea region (Fig. 1) by means of a two-step nesting. The initial data for the 15 km resolution domain was generated by interpolating from the German Meteorological Service's global model (Majewski et al. 2002) with 40 km horizontal resolution to the 15 km grid. Boundary atmospheric conditions were provided by GME every 6 h. In the second nesting step the initial and boundary data needed for 5 km model resolution were taken from 15 km model runs every hour. The thermodynamic sea-ice model of Schröder et al. (2011) was used to simulate the sea-ice surface temperature (T _s ) and energy fluxes over ice surfaces.

For comparison, we also used meteorological reanalysis data of the US National Centers for Environmental Prediction reanalysis data set (NCEP; Kanamitsu et al. 2002). NCEP data were chosen for a comparison with COSMO since many remote-sensing and sea-ice modelling studies use NCEP for forcing (e.g., Ernsdorf et al. 2011; Adams et al. 2013).

As mentioned in the introduction, polynya area was either assumed to be ice-free (COSMO 0 cm, or C00) or to be covered with 10 cm of ice (COSMO 10 cm, or C10). To assess seasonal ice production, all model runs were conducted in a forecast mode for 30 h; the first 6 h were omitted as a spin-up period. The merging of the remaining 24-h periods generates continuous atmospheric data sets for the winters 2007/08 and 2008/09 (November–April). While conventional satellite-based methods (Yu & Lindsay 2003; Tamura & Ohshima 2011; Willmes et al. 2011) apply simple empirical formulas to calculate atmospheric energy fluxes, COSMO includes state-of-the-art parameterizations for turbulent fluxes and a radiation transfer scheme. In addition, our COSMO simulations took into account the effect of the polynya on the atmospheric boundary layer. In all methods, the calculation of ice production was based on the energy balance equation. If the water is at the freezing point, it is assumed that the total energy loss of the polynya surface to the atmosphere is compensated by freezing (oceanic heat flux is neglected), and the potential ice production can be described by the following equation: ∂ h i / ∂ t = - Q A / ( ρ i g L f ) , where Q _A is the total atmospheric heat flux, h _i the ice thickness, ρ _i the density of sea ice (ρ=910 kg/m³) and the latent heat of freezing (L _f = 0.334·10⁶ J/kg). Q _A is calculated as

where Q ₀ is the net radiation balance, H ₀ is the sensible heat flux and E ₀ is the latent heat flux. Ice production will occur if the energy flux to the atmosphere Q _A is negative and the water is at its freezing point. The latter is generally observed throughout the winter period. Thin ice was kept constant during the model run, i.e., all newly formed ice was assumed to be transported away by advection, which is a realistic assumption for wind-driven flaw polynyas.

COSMO results are compared to Willmes et al. (2011), who used NCEP atmospheric data and the polynya signature simulation method for polynya area. Furthermore, the impact of polynyas on the atmospheric boundary layer is investigated by comparing NCEP-T2m and COSMO-T2m over polynyas and the entire Laptev Sea, respectively.

Verifying COSMO simulations in the vicinity of the Laptev polynyas is difficult due to the lack of observations. However, data are available from field measurements at the fast-ice edge in the Laptev Sea (Heinemann et al. 2010; Helbig et al. 2010) obtained from automatic weather stations as part of the German–Russian Laptev Sea System project during the Transdrift XIII/XV expeditions in April 2008 and April 2009. Mean biases and root mean square error (RMSE) between COSMO model variables and measurements of six automatic weather stations are −0.54 (1.54) m s⁻¹ for wind speed, −0.72 (1.46) hPa for surface pressure, −1.02 (30.5) W m⁻² for net radiation and −0.18 (2.74) K for T2m. This indicates a good quality of the simulations (see also Ernsdorf et al. 2011). Furthermore, Schröder et al. (2011) compared MODIS and COSMO surface temperatures and found that the sea-ice model implemented in COSMO performed well.

Time series of polynya area, ice production and COSMO-T2m and NCEP-T2m are shown in Figs. 2 and 3. It should first be noted that there are large differences for polynya area and the volume of ice produced between the two winters (see Table 1). Daily and seasonal ice production within the Laptev Sea polynyas are highly variable, with daily values reaching up to 3 km³ d⁻¹ and almost 6 km³ d⁻¹ on 10 January 2009 for C10 and C00, respectively (Fig. 3d). Monthly values range from 0.4 to 27 km³ for C10. The longest period without polynyas is 21 days (3–24 March 2009). While typical polynya events lasted for about six days, the longest event—30 consecutive polynya days—spanned December 2008 and January 2009.

Polynya openings are associated with temperature peaks and wind speed peaks (not shown), which can be expected for wind-driven flaw polynyas and which agrees well with the findings of Stringer & Groves (1991) for the Bering and Chukchi seas. Ice production values for COSMO runs with open-water polynyas exceed those with 10 cm of ice cover by 50–70% (Table 1). This is caused by higher absolute values for temperature gradients (T2m-T _s ) for C00 (−10.9 K), leading to higher heat loss into the atmosphere and therefore higher ice production compared to the thin-ice covered polynya in C10 (−5.8 K).

As shown in Figs. 2d and 3d, ice-production rates for C10 match quite well with those calculated by Willmes et al. (2011). This seems to be surprising considering that Willmes et al. used NCEP data—which have a resolution being too coarse to resolve polynyas—for the energy flux calculation. Consequently, T2m is too low in NCEP within the polynya areas, since polynyas and their warming effects on the atmosphere are not represented adequately. Hence, using NCEP, or any other coarse-grid global analysis data that cannot resolve the polynyas, for calculating heat fluxes should lead to systematic overestimation of ice production in polynyas. While the NCEP mean T2m over polynyas is 8K colder than for C00 in 2007/08 (Fig. 2a) and 6.7 K colder in 2008/09 (Fig. 3a), the differences between NCEP and C10 amount only to 2.4 K and 1.2 K, respectively.

The differences between mean values of T2m over the polynya and the entire Laptev Sea decline from 12.6°C for C00 to 7.2°C for C10 and to 3.5°C for NCEP (Table 2). This decline originates from the decrease in sensible heat fluxes between C00 and C10 and the different consideration of polynyas in NCEP. While in NCEP, polynyas are not resolved and hence are treated as sea ice, the open water in the C00 simulations leads to a temperature about 7 K higher over polynyas compared to NCEP (Table 2). However, when comparing average temperatures of NCEP and COSMO for the whole Laptev Sea, NCEP temperatures are generally 0.9 (2.8) K higher for 2007/08 (2008/09) than COSMO temperatures. This systematic warm bias of NCEP in relation to COSMO has the same effect as the atmospheric boundary layer warming by the polynyas. This explains why the results of ice production calculation for C10 and Willmes et al. (2011) are similar.

The difference between NCEP-T2m and COSMO-T2m is even greater when looking at daily values (Figs. 2a, 3a). Mean air temperature over the polynya is only shown if the polynya size is larger than 300 km² (corresponding to 10 model grid points). Mean T2m at polynyas from C00/C10 exceed the NCEP-T2m by 7.3/1.8°C on average and up to 20/8°C for single days. Polynya ice production would be higher when using NCEP-T2m instead of COSMO-T2m because the surface layer instability would be stronger (Table 2).

Figure 4 shows ice production rates and ice production of the polynyas of the Laptev Sea sub-regions—north-eastern Taimyr (NET), Taimyr (T), Anabar–Lena (AL) and western New Siberian (WNS)—for both winters. The production rates are calculated by dividing the seasonal sum of ice production in a sub-region by the seasonal sum of polynya area in the same sub-region. Generally, largest ice production rates can be found for C00 and the smallest by Willmes et al. (2011). The NET is the most productive polynya in terms of volume ice production in both years. While largest growth rates (m d⁻¹) occur in the WNS polynyas in 2008/09, the total contribution of this sub-region to ice production is small due to a generally smaller polynya size. Differences in ice production sums between that generated by Willmes et al. (2011) and C10 are also caused by slightly different polynya areas (Figs. 2c, 3c). Calculations of polynya area by Willmes et al. are larger for most small polynya openings, which results in modest differences for regional ice production compared to C10.

Figure 5 displays the regional distribution of the ice production in the Laptev Sea. It is generated by accumulating the ice production of one season for every pixel. The highest ice production rates occur in the western Laptev Sea, in the NET sub-region (Fig. 4), with up to 5 m per season for C00 and 3 m for C10, respectively. This kind of regional distribution differs from that of Tamura & Ohshima (2011), who show the maximum of ice production (5 m) in the area of AL. While the NET polynya is relatively broad in both years, ice production in the other sub-regions occurs in more narrow zones at the fast-ice edge. This highlights the importance of high-resolution atmospheric simulations in providing an appropriate description of air–sea interaction and ice production for Arctic flaw polynyas.

The frequency distribution of total daily ice production for both winters is shown in Fig. 6. The distribution shifts to the right from 2007/08 to 2008/09, which means that the percentage of days with high ice production is generally higher for winter 2008/09. This corresponds well with larger polynya openings in 2008/09 (Fig. 5). During about 20% of the observation period, no polynyas are observed at all for C00/C10 and consequently related polynya ice production is zero. Low ice production occurs less frequently than as shown by Willmes et al. (2011). In contrast, high production occurs more frequently for C00/10 than in simulations by Willmes et al. (2011). Differences between C10 and C00 are reflected by an increased number of days with high ice production for C00. Two main conclusions can be drawn from Fig. 6: (1) C00 leads to more extreme ice production events than C10; and (2) the variability of ice production is generally lower in simulations by Willmes et al. (2011) than for C00/C10, which is caused by differences in the methods for deducing polynya area and atmospheric data.