Thin-ice dynamics and ice production in the Storfjorden polynya for winter seasons 2002/2003–2013/2014 using MODIS thermal infrared imagery

. Spatial and temporal characteristics of the Storfjorden polynya, which forms regularly in the proximity of the islands Spitsbergen, Barentsøya and Edgeøya in the Svalbard archipelago under the inﬂuence of strong northeasterly winds, have been investigated for the period of 2002/2003 to 2013/2014 using thermal infrared satellite imagery. Thin-ice thicknesses were calculated from MODIS ice-surface temperatures combined with ECMWF ERA-Interim atmospheric reanalysis data in an energy-balance model. Associated quantities like polynya area and total ice production were derived and compared to previous remote sensing and modeling studies. A basic coverage-correction scheme was applied to account for cloud gaps in the daily composites. On average, both polynya area and ice production are thereby increased by about 30 %. The sea ice in the Storfjorden area experiences a late fall freeze-up in several years over the 12-winter period, which becomes most apparent through an increasing frequency of large thin-ice areas until the end of December. In the course of an average winter season, ice thicknesses below 10 cm are dominating within the Storfjorden basin. During the regarded period, the mean polynya area is 4555.7 ± 1542.9 km 2 . Maximum daily ice production

. Map of the Svalbard archipelago in western Barents Sea. The study region is located in Storfjorden, which is surrounded by the main island Spitsbergen in the west and Barentsøya/Edgeøya in the east. The applied polynya mask is marked in red, enclosing the Storfjorden basin north of the sill at 77 • N. Bathymetric data by Jakobsson et al. (2012) (IBCAO v3.0). physical processes in the Arctic shelf regions and the Arctic sea-ice cover in general. Ice production in thin-ice regions can either be estimated by the use of remote sensing methods (Willmes et al., 2010) or by using model approaches (Ebner et al., 2011;Krumpen et al., 2011;Bauer et al., 2013). The use of thermal infrared and microwave remote sensing data offers the opportunity for daily monitoring of thin-ice thicknesses and their distribution on large spatial scales (Yu and Rothrock, 1996;Willmes et al., 2010;Adams et al., 2013).
Storfjorden (see Fig. 1), situated in the southeastern part of the Svalbard archipelago between the main island of Spitsbergen in the west and Barentsøya/Edgeøya in the east, is a region of high coastal polynya activity during winter, as reported by a number of earlier studies (e.g., Haarpaintner et al., 2001;Skogseth et al., 2004). Two shallow straits, Heleysundet and Freemansundet, connect the northern part of Storfjorden with the Barents Sea and show tidally induced currents as well as exchange processes between water masses inside the fjord and the Barents Sea (Skogseth et al., 2013).
Persistent and strong northerly to northeasterly winds are the main cause for larger polynya openings inside Storfjorden. These opening events may result in large volumes of brine-enriched shelf water, which can be significant regarding the total amount of bottom water in the Arctic Ocean by supplying between 5 and 20 % of all the newly formed dense water (Skogseth et al., 2004;Hendricks et al., 2011). Besides thermodynamically induced ice growth in Storfjorden during wintertime, the dynamical part due to rafting and deformation processes may play a substantial role in the overall ice-thickness distribution, as southerly (onshore) winds can potentially advect sea ice from the Barents Sea into the fjord ). An estimation of thermodynamic IP within the open water and thin-ice area (i.e., Storfjorden polynya) has been conducted in previous studies using mainly microwave remote sensing (both passive and active systems) in combination with different model approaches. Therein, values are ranging between 20 and 35-45 km 3 for single winter seasons (November-May) between 1997/1998and 2000/2001(Haarpaintner et al., 2001Skogseth et al., 2004Skogseth et al., , 2005 and 47 km 3 when averaged over the winter seasons (October/November-June/July) from 2002/2012(Jardon et al., 2014. Iwamoto et al. (2014) state that the Storfjorden polynya contributes with around 4 % to the average total ice volume produced in Arctic polynyas between 2002and 2010.
While these estimates make the Storfjorden area appear to be of minor importance regarding overall ice production in Arctic polynyas, we chose this region due to several distinct requirements. These included the availability of a number of related polynya studies in order to set our estimates into a proper context as well as the comparatively small spatial extent of the Storfjorden polynya to profit from the enhanced resolution imagery.
Based on daily empirically derived ice-thickness distributions in thin-ice areas at a comparatively high spatial resolution, we perform a broad investigation on the associated quantities polynya area (POLA) and ice production for the period between 2002/2003 and 2013/2014 using highresolution Moderate Resolution Imaging Spectroradiometer (MODIS) thermal infrared data. The presented results will develop a more comprehensive understanding of the polynya dynamics in Storfjorden as well as associated processes over both annual and interannual timescales (in the course of this paper, the term "annual" refers to one winter season from November to March). Furthermore, results will contribute to pan-arctic studies of polynya dynamics.

MODIS ice-surface temperatures
Thin-ice thickness are computed using the MOD/MYD29 sea-ice product (Hall et al., 2004;Riggs et al., 2006) derived from MODIS satellite data. We used data from both MODIS instruments on board the Terra and Aqua polar-orbiting satellite platforms. They are available from the National Snow and Ice Data Center (NSIDC) located in Boulder, USA (ftp: //n5eil01u.ecs.nsidc.org/SAN/). The product contains swath data of ice-surface temperatures (ISTs). Swath-based ISTs have a spatial resolution of 1 km × 1 km at nadir. The accuracy of the MOD/MYD29 ISTs is given with 1-3 K (Hall et al., 2004).
For our analysis, swaths covering the Storfjorden area were extracted using metadata information for each MODIS swath, resulting in 27 individual swaths per day on average for later composite generation. These composites solely consist of nighttime scenes to exclude the influence of incident shortwave radiation (Yu and Lindsay, 2003;Adams et al., 2013; cf. Sect. 2.3 and 2.3.1). The total number of incorporated MODIS swaths is illustrated in Table 1. As MODIS swath data presumably suffer from inherent panoramic distortion effects, all IST swaths were mapped onto a 2 km × 2 km (0.018 • ) equirectangular grid covering the southern part of Svalbard (75-80 • N, 10-30 • E). A polynya mask (red area in Fig. 1) is applied in order to cover an area comparable to previous studies (Skogseth et al., 2004(Skogseth et al., , 2005Jardon et al., 2014), as well as to limit the probability of remaining hidden cloud artifacts. The mask comprises a total area of 12 594 km 2 and has its southern limit at 77 • N (following Skogseth et al., 2004), where a 120 m deep sill serves as kind of a boundary for the accumulation of brine-enriched shelf water in the Storfjorden basin (see Fig. 1).

ERA-Interim reanalysis
For the calculation of thin-ice thicknesses, atmospheric variables from the ERA-Interim reanalysis product (Dee et al., 2011) are used to provide 2 m temperature, 2 m dew point temperature, 10 m wind speed components (u and v) and the mean sea-level pressure at a temporal resolution of 6 h. The data set is provided by the European Center for Medium-Range Weather Forecasts (ECMWF) in a horizontal resolution of 0.75 • (∼ 79 km). As the satellite data set is by far spatially higher resolved than the atmospheric data set, a linear interpolation of the ERA-Interim data is performed to match the MODIS data. In the course of the TIT calculations, each MODIS swath is linked to the closest time step of the atmospheric fields.

Thin-ice thickness retrieval using a 1-D thermodynamic surface energy model
In order to derive daily TIT distributions, we use an approach that follows the work of Yu and Rothrock (1996), Yu and Lindsay (2003) and Drucker et al. (2003), with most recent improvements and modifications made by Willmes et al. (2010Willmes et al. ( , 2011 and Adams et al. (2013). The applied method utilizes the distinct relationship between IST and thickness of thin ice depending on atmospheric energy fluxes. The IST of very thin ice (with no snow cover) is close to the freezing point of the upper ocean layer and decreasing IST can be observed for thicker ice (Drucker et al., 2003;Kwok et al., 2007). However, IST is governed by atmospheric radiation fluxes and turbulent fluxes of heat, which does not allow for a simple relationship between IST and ice thickness. The ice thickness h ice (likewise abbreviated "TIT" for thin ice in the course of this paper) is therefore calculated using a 1-D thermodynamic sea-ice model. Several assumptions have to be made to apply this model. First, the temperature profile through the ice is assumed to be linear. This approximation is valid for h ice ≤ 0.5 m (Drucker et al., 2003). Second, the water temperature at the boundary between ice and ocean is constant and at its freezing point (T f = 271.35 K). Third, the ice is assumed to be free of snow. The ice thickness is calculated by using MODIS IST and ERA-Interim reanalysis data. The method is based on the condition, that the conductive heat flux through the ice (Q ice , Eq. 1) equals the total heat flux to the atmosphere (Q atm , Eq. 2). This implies that all energy loss at the ice-surface is compensated by the conductive heat flux Q ice .
where Q 0 is the net radiation balance, H 0 and E 0 are the turbulent fluxes of sensible heat and latent heat, respectively, κ ice is the thermal conductivity of sea ice, T surf is the surface temperature, T f is the freezing point of sea water and h ice is the ice thickness. Because of difficulties with the parametrization of shortwave radiation terms (Adams et al., 2013), only nighttime scenes were incorporated into the calculations (by checking the solar incidence angle on a pixelby-pixel basis), so that the radiation balance Q 0 simplifies to the long-wave radiation terms L ↓ and L ↑. Upwelling long-wave radiation L ↑ is calculated using the IST from MOD/MYD29 with the surface emissivity s set to 1. L ↓ can be obtained using ERA-Interim 2 m temperatures and the atmospheric emissivity atm , calculated using the formulation of Jin et al. (2006). Turbulent fluxes of sensible (H 0 ) and latent (E 0 ) heat in Eq.
(2) are calculated using an iterative bulk approach by Launiainen and Vihma (1990), where nearsurface stratification is considered using Monin-Obukhov similarity theory and associated universal functions for the www.the-cryosphere.net/9/1063/2015/ The Cryosphere, 9, 1063-1073, 2015 heat transfer coefficients C H and C E . The 10 m wind speed is interpolated to the 2 m level. A constant roughness length for momentum z 0 = 1 × 10 −3 m is used. As could be shown by Adams et al. (2013), the standard approach used in previous studies using constant heat transfer coefficients tends to underestimate the ice thickness by approximately 7 cm on average, when compared to an iteratively calculated heat transfer coefficient. To obtain the ice thickness (h ice ), the total atmospheric flux Q atm (Eq. 2) is set equal to the conductive heat flux through the ice (Q ice ; Eq. 1), and Eq. (1) is solved for h ice using a value for the thermal conductivity of κ ice = 2.03 Wm −2 K −1 (Drucker et al., 2003). Calculated ice thicknesses are assumed to be uniform at a given pixel location and therefore hypothetical (Tamura and Ohshima, 2011). The described method is only applicable to clear sky conditions, as clouds and fog strongly influence the accuracy of the recorded ISTs (Riggs et al., 2006).
In a sensitivity analysis of the above described method, Adams et al. (2013) state an uncertainty for the ice-thickness retrieval of ±1.0, ±2.1 and ±5.3 cm for TIT classes 0-5, 5-10 and 10-20 cm, respectively. Therefore, we constrain our analysis to ice thicknesses ≤ 0.2 m, as this range is regarded sufficient to get reliable results for ice production (Yu and Rothrock, 1996;Adams et al., 2013).

Calculation of daily thin-ice thickness composites
From all available MODIS swaths covering the Storfjorden polynya region, daily composites of ISTs and thin-ice thicknesses were computed. The TIT is first calculated from each swath on its own with the procedure described in Sect. 2.3. Subsequently, the mean TIT is computed pixel-wise and stored with its corresponding IST value.
By using daily composites, the daily IST coverage (percentage of pixels within the polynya mask that feature at least one valid IST value, based on all MOD/MYD29 swaths covering the Storfjorden region on a given day) is significantly increased compared to single swaths, reaching an average value of 82 % for the winter seasons 2002/2003 to 2013/2014. Figure 2 shows the seasonal and interannual variability of IST coverage. Noteworthy is a very low coverage in November to early January 2009/2010, as well as generally less frequent coverage at the beginning (November) and end (March) of each winter season. The latter effect originates from the previously stated restriction to nighttime scenes and consequently a reduction in the amount of available MODIS swaths. Hence, it is also the main reason why the months of October and April were left out in our investigations.

Derivation of ice production and polynya area
Following Tamura et al. (2007Tamura et al. ( , 2008 and Willmes et al. (2011), daily ice production rates are calculated from the heat loss at the ice surface by assuming that the entire heat loss to the overlying atmosphere contributes to new ice formation.
Therein, ∂h ∂t stands for the ice production rate,Q ice is the daily mean conductive heat flux through the ice, ρ ice is the density of sea ice (taken as ρ ice = 910 kg m −3 ) and L f is the latent heat of fusion of ice (L f = 0.334 MJ kg −1 ; e.g., Tamura and Ohshima, 2011). Note that the negative sign in the right side of Eq. (3) handles the convention that Q atm is considered positive when the surface gains energy, as well as considering that ice production only takes place when there is a net energy loss from the surface. Multiplying ∂h ∂t with the areal extent of each pixel in the regarded region yields the volume IP rate ∂V ∂t . IP rates are calculated for each pixel with an ice thickness ≤ 0.2 m and afterwards extrapolated to daily rates for later accumulation and/or averaging. Considering the total polynya area, a threshold of h ice ≤ 0.2 m is applied to identify pixels containing thin ice in the daily composites as part of the POLA. The total area of those pixels is then accumulated to obtain the total daily POLA in square kilometers.

IST coverage correction
Cloud-induced gaps in our daily IST and TIT composites are a serious problem when we want to compare our initially calculated POLA and IP to estimates from other studies in which the use of active and passive microwave remote sensing systems largely reduces the effect of atmospheric disturbances.
To overcome these difficulties, a simple scaling approach is applied that works under the assumption that pixels in the uncovered part of the masked (Fig. 1) daily composites also contribute to the total POLA by the same proportion as those areas that are covered by a IST signal.
The initially calculated daily POLA is therefore scaled by a factor F: The coverage in Eq. (4) could range from 0 to 1, but we apply the correction only for coverages exceeding 0.5. Similarly, estimated IP values are also scaled by the same factor yielding a corrected ice production IP CC . Application of this correction scheme is limited to cases where the daily IST coverage (compare Fig. 2) within the fjord (i.e., the predefined polynya mask) surpasses a threshold of 50 %. By using this threshold, unrealistically high POLA CC and IP CC values are avoided. Otherwise, both POLA CC and IP CC are linearly interpolated using bounding days with a coverage fulfilling this precondition.
In order to test the performance of the cloud-correction method, we selected a total of six case studies from Jan- The coverage correction has been applied for each case study after randomly removing 45 % of the pixels within the polynya mask for 1000 realizations. The resulting deviation (in percent) from the "true" polynya area from all case studies combined is shown in Fig. 3. POLA CC deviations almost perfectly follow a typical normal (Gaussian) distribution, with a mean value of 0.77 % and a standard error of 5-6 % of the daily POLA. Because the correction of IP values uses the same scaling factor F , the retrieved error margins can also be regarded as the uncertainty of our IP CC estimates.

Results
The bars in Fig. 4 show the mean relative contribution of each ice-thickness class from the total number of pixels with a TIT ≤ 0.2 m between the winter seasons 2002/2003 and 2013/2014. A temporal differentiation between the beginning of the freezing season (November-December) and the end of the freezing season (January-March) as well as a combined distribution for the whole winter period is given. Error bars refer to ±1 standard deviation (SD) of the interannual averaged distribution frequencies. Between November and March, ice thinner than 0.02 m contributes around 15 % to the total polynya area in an average winter season. Due to a higher sensitivity of open water and very thin ice for sea smoke effects and consequently an exclusion of thinice pixels by the cloud mask, this thickness class is potentially biased . The average ice thickness within the polynya mask is 10 ± 1.8 cm. Thicker ice classes (> 0.1 m) cover around half of the entire Storfjorden polynya in January to March and show overall lower contributions at the beginning of the freezing season. This follows from the For each case study, the coverage correction was repeated several times (n = 1000 × 6), after randomly removing 45 % of the pixels within the applied polynya mask.The red curve illustrates a fitted normal distribution.
high contribution of very thin ice in November to December, where over 60 % of the total POLA consists of ice thinner than 0.1 m. SD is highest in the lowest thickness classes (< 0.04 m) and appears to be lower towards the end of the freezing season, possibly indicating a stronger seasonal variability of ice-thickness distributions within the polynya in November and December over the 12-year record (also apparent in relative TIT frequencies for November-December;   (November-December, Fig. 5) and the end of each freezing season (January-March, Fig. 6). Therein, a pixel value of, e.g., 0.25 means that on 25 % of all days, a TIT threshold of 0.2 m was not exceeded. Following the previously stated definition of a polynya, these pixels represent the number of polynya occurrences in the given period. Figure 6 shows that the main regions for polynya development towards the end of the freezing season are at the lee sides of Barentsøya and Edgeøya at the eastern side of the fjord, extending more (2006-2008, 2010, 2012-2014) or less (2003-2005, 2009, 2011) further southwestwards. At the beginning of the freezing season (Fig. 5), annual distributions show a more variable pattern of TIT frequencies. In some years (e.g., 2007, 2009, 2012), frequencies exceeding 0.3 cover large areas within the masked area in Storfjorden, which might be an effect of a late fall freeze-up and subsequently longer periods with thin ice and open water in the fjord. In other years (e.g., 2003, 2004), the observed pattern of high TIT frequencies is similar to typical polynya  (Fig. 6). Low values (< 0.05) at the (north) western side of Storfjorden towards the end of a freezing season (Fig. 6) further indicate typical fast-ice areas, which are most pronounced in the years 2003, 2004, 2005, 2007, 2009 and 2011. The  Table 2. Accumulated ice production (IP) in cubic kilometers per winter and average polynya area (POLA) in square kilometers for each winter season (November to March) in the Storfjorden polynya, together with the interannual average (mean) and its standard deviation (SD). CC denotes values after coverage correction, with relative increase compared to uncorrected values in brackets. All values are calculated within the predefined polynya mask (Fig. 1). well into January with a series of large POLA CC events (> 5000 km 2 ) developing in February and March. The mean (November to March) POLA for the entire regarded period amounts to 3511.8 ± 999.6 km 2 , with an increase of 29 % for POLA CC when the presented coverage correction is applied. Around 60 % of this area consists of ice thinner than 10 cm. A complete overview on wintertime mean POLA and POLA CC values is given in Table 2. A spatial overview of the resulting accumulated daily ice production rates (in m winter −1 ), averaged between 2002/2003 and 2013/2014 is presented in Fig. 8. Largest values occur in regions which also show high TIT frequencies (Fig. 6). The average ice production in the western part of Storfjorden is generally lower, as thin ice in these areas is mainly present at the beginning of the freezing season.
Maximum daily ice production rates in the fjord can reach as high as 26 cm d −1 (winter season 2006/2007), with the highest rates occurring on average in the northern part of Storfjorden (north of 77.5 • N). Figure 9 shows values of accumulated ice production per winter season from November to March, both with and without applied coverage correction. The impact of seasonal differences is highlighted in yellowish and blueish colored bars. A detailed overview for each winter season is additionally given in Table 2. On average, the correction scheme increases the wintertime ice production by approximately 30 %. While the average IP value is 28.3 ± 8.5 km 3 (CC; 21.7 ± 5.9 km 3 for no CC), a high interannual variability of ice production is found, with values as low as 16.

Discussion
In Skogseth et al. (2004Skogseth et al. ( , 2005, a manual classification of several ERS-2 synthetic aperture radar (SAR) images from 1998 to 2002 was performed to derive parameters for a winddriven polynya width model (Haarpaintner et al., 2001), which gives estimates of ice production and brine-enriched shelf water based on satellite imagery, wind data from nearby weather stations and surface hydrography. Ice production was calculated from surface heat balance and includes contributions from open water (frazil ice), thin ice as well as fast and pack ice inside the Storfjorden basin north of 77 • N. This setup was later extended to a 33-year record of modeled total ice production covering the winters 1970 to 2002 using model parameters derived from the examined 5-winter period (Skogseth et al., 2005). While the average total ice production for 1998 to 2002 was estimated to 43.6 ± 9.7 km 3 , it was slightly lower for the 33-year time series with only 39.9 ± 11.7 km 3 . When comparing these numbers to our average IP CC estimate of 28.3 ± 8.5 km 3 for 2002/2003 to 2013/2014, one has to keep in mind that Skogseth et al. (2004Skogseth et al. ( , 2005 also include ice production under fast and pack ice inside the fjord, while our study concentrates on open water and thin-ice areas only. Assuming an average contribution of 25 % from fast-ice and pack-ice areas ( Table 3 in Skogseth et al., 2005) and after reducing annually accumulated total ice production accordingly, the average ice production (1970 to 2002) inside open water and thin-ice areas of Skogseth et al. (2005) is lowered to approximately 30 ± 9 km 3 which is within the standard deviation presented here. Skogseth et al. (2005) give some further information on the interannual variability of ice production within the Storfjorden polynya, although it is stated that the uncertainty for the 33year time series is increased. The presented 33-year time series shows a slightly negative and non-significant trend of approximately −2 km 3 decade −1 , while the last 12 winter seasons presented here show a positive trend of 2.02 km 3 yr −1 during the analyzed period. However, the long-term model estimates by Skogseth et al. (2005) indicate the presence of multi-decadal fluctuations. Jardon et al. (2014) developed an ice production model which uses Advanced Microwave Scanning Radiometer -Earth Observing System (AMSR-E) sea-ice concentration data to derive daily open water fractions. Heat flux calculations over water were performed using ERA-Interim reanalysis data in a bulk algorithm. In contrast to our results, their estimates of POLA and IP incorporate both frazil ice growth as well as ice growth under ice thicker than 20 cm. In comparison, our results should profit from the enhanced resolution of MODIS and daily derived TIT distributions. Nevertheless, derived POLA from TIT ≤ 0.2 m in our study compares well with POLA derived from AMSR-E sea-ice concentration data. Both wintertime mean POLA and POLA CC are within the 25th and 75th percentile of the estimated open water area by Jardon et al. (2014) in almost every winter season except for 2004/2005 and 2010/2011, during which both values derived by MODIS exceed the 75th percentile. Discrepancies in derived total ice production are obvious but can be partly explained by differences in the regarded time frame of each winter season. While Jardon et al. (2014) analyze the full period with sea ice in the fjord (roughly October/November to June/July), we focus on the November to March period. Even when we reduce the presented IP val-ues by the amount of frazil ice that is formed in the months April-July, which in 2006(example shown in Jardon et al., 2014 contributed around 15 % to the total volume of frazil ice in this particular winter season, the time series differ on average by approx. 40 %. Therefore, Jardon et al. (2014) presents higher IP values in all winter seasons except 2004/2005 and 2010/2011, but overall the interannual variability is comparable.
The studies by Tamura and Ohshima (2011) and Iwamoto et al. (2014) face similar restrictions concerning the spatial resolution of the applied passive microwave data from Special Sensor Microwave Imager (SSM/I) and AMSR-E, which is most apparent in smaller polynyas such as in Storfjorden. In Tamura and Ohshima (2011), an average ice production of 137 ± 35 km 3 for the period 1992-2007 (September-May) is presented. This value largely exceeds our estimates as well as those by Skogseth et al. (2005) and Jardon et al. (2014), most probably due to the discrepancies in spatial resolution of the input data. More recent numbers by Iwamoto et al. (2014), who use an updated algorithm based on MODIS and AMSR-E satellite data as well as ERA-Interim reanalysis, are also lower than those in Tamura and Ohshima (2011), with the mean annual (also September-May) ice production between 2002/2003 and 2010/2011 estimated as 47 ± 5 km 3 (roughly 40 km 3 for November-March; based on presented monthly mean values). A positive trend of 3 km 3 per decade is presented which is much smaller than our value of 18.7 km 3 per decade based on the same period from 2002/2003 to 2010/2011. Although Iwamoto et al. (2014) provide enhancements in terms of spatial resolution and the inclusion of land fast-ice detection in comparison to Tamura and Ohshima (2011), our estimates are still exceeded in every winter season. The presented interannual variability is similar to our study, with 2005/2006 and 2008/2009 being the only winter seasons with a strongly contrasting development. Both Tamura and Ohshima (2011) and Iwamoto et al. (2014) apply noticeably larger polynya masks to derive ice production values in the Storfjorden area which extend well into the Greenland and Barents Sea regions and far beyond Storfjorden itself. This omission of local characteristics in the Storfjorden basin in addition to a differing base period, which spans the months from September to May, could therefore explain part of the difference. However, the use of coarse-resolution passive microwave data certainly leads to a higher uncertainty of IP estimates, while the sign of the bias cannot be determined here. It will depend on the distribution of thin ice within the sensor's footprint and potential land spillover effects.

Summary and conclusions
In this study, we presented a comprehensive overview of daily thin-ice thickness distributions and resulting thermodynamic ice production rates for the Storfjorden polynya, located in the southern part of the Svalbard archipelago. The www.the-cryosphere.net/9/1063/2015/ The Cryosphere, 9, 1063-1073, 2015