The impacts of increased open water in the Beaufort Sea were investigated for a summer Arctic storm in 2008 using a coupled atmosphere-ice-ocean model. The storm originated in northern Siberia and slowly moved into the Beaufort Sea along the ice edge in late July. The maximum wind associated with the storm occurred when it was located over the open water near the Beaufort Sea coast, after it had moved over the Chukchi and Beaufort Seas. The coupled model system is shown to simulate the storm track, intensity, maximum wind speed and the ice cover well. The model simulations suggest that the lack of ice cover in the Beaufort Sea during the 2008 storm results in increased local surface wind and surface air temperature, compared to enhanced ice cover extents such as occurred in past decades. In addition, due to this increase of open water, the surface latent and sensible heat fluxes into the atmosphere are significantly increased. However, there were no significant impacts on the storm track. The expanded open water and the loss of the sea ice results in increases in the surface air temperature by as much as 8°C. Although the atmospheric warming mostly occurs in the boundary layer, there is increased atmospheric boundary turbulence and downward kinetic energy transport that reach to mid-levels of the troposphere and beyond. These changes result in enhanced surface winds, by as much as ∼4 m/s during the 2008 storm, compared to higher ice concentration conditions (typical of past decades). The dominant sea surface temperature response to the storm occurs over open water; storm-generated mixing in the upper ocean results in sea surface cooling of up to 2°C along the southern Beaufort Sea coastal waters. The Ekman divergence associated with the storm caused a decrease in the fresh water content in the central Beaufort Sea by about 11 cm.
 Loss of Arctic sea ice may have significant impacts on air-sea interactions during Arctic storms. The Arctic Ocean is generally a low energy ocean. When sea ice is present, even strong storms do not induce significant oceanic response. However, if the Arctic Ocean becomes mostly ice free, local storms can significantly impact the upper ocean [Holt and Martin, 2001; Yang et al., 2004; Rainville and Woodgate, 2009]. In addition, due to the increased momentum, moisture and heat exchanges between air and ocean, the loss of the Arctic sea ice coverage is accompanied by increases in the strength and size of Arctic storms [Simmonds and Keay, 2009].
 In this study, a coupled model system consisting of the Canadian Regional Climate Model (CRCM) and a coupled ice-ocean model (CIOM) is implemented for the Arctic basin (Figure 1) to simulate a storm that moved into the Beaufort Sea during the period from 29 July to 4 August 2008. The motivation for using CRCM is that in a related study, we investigate the Arctic storm climate and possible impacts of climate change. We focus on the air-sea interactions and their role on the life cycle of the storm. In particular, we try to understand the impacts of increased open water on storm intensity and development, and the surface fields.Section 2 describes the coupled model system, including the atmosphere, ice, and ocean components, and the experimental design. Section 3 gives an overview for the summer 2008 storm used as a case study. Section 4 discusses results of the study in terms of the impacts of increased open water on storm track and intensity, marine winds, atmospheric surface air temperature, specific humidity and related variables. Section 5 describes the oceanic responses to the storm, section 6 discusses the physical mechanism and ensemble results and section 7 gives the conclusions.
2. Model Description and Experiment Design
2.1. Model Description
 The coupled model system consists of three components, Canadian Regional Climate Model (CRCM), Princeton Ocean Model (POM) and Hibler ice model.
2.1.1. Atmospheric Model
 CRCM is based on the dynamical formulation of the Canadian Mesoscale Compressible Community (MC2) model and solves the fully elastic nonhydrostatic Euler equations using a semi-implicit semi-Lagrangian numerical scheme. The physical parameterization package of the second-generation Canadian Global Climate Model (CGCM2), followingMcFarlane et al. , is implemented to solve the subgrid-scale processes. The Kain-Fritsch scheme was chosen for deep convection while large-scale condensation is simulated using the CGCM2 physics formulation [Kain and Fritsch, 1990; Paquin and Caya, 2000]. In this study, the CRCM simulations were performed at a horizontal resolution of 25 km, with 29 levels in the vertical direction. A 15-min time step is employed. Initial and boundary conditions for CRCM are specified by CMC (Canadian Meteorological Centre) analyses data [Chouinard et al., 1994]. CMC analysis assimilates available observations into its operational forecast outputs. The atmospheric circulation, storm track and intensity in CMC analyses are very similar to those of NCEP reanalyses (figures not shown). These reanalysis fields are widely used in Canadian atmospheric and oceanographic community studies. CRCM was applied in previous Arctic studies [Gachon et al., 2003; Qian et al., 2008; Wyser et al., 2008; Joly et al., 2011]. In particular, Wyser et al.  implemented CRCM in the western Arctic to compare eight regional climate models, showing that CRCM has a reasonable simulation of Arctic climatology. A detailed description of CRCM is given by Caya and Laprise , Laprise et al.  and Caya and Biner .
2.1.2. Ocean Model
 POM is a three-dimensional, primitive equation model with complete thermohaline dynamics, a sigma (σ) vertical coordinate, and a free surface. A second-order turbulence closure scheme [Mellor and Yamada, 1982] is used to represent the mixed layer dynamics. To minimize pressure gradient errors, the bottom topography in the model was smoothed so that the difference in the depths of adjacent grid points divided by their means is less than 0.2 [Mellor et al., 1994]. In the experiments described here, 23 vertical sigma levels are used with higher resolution in the upper mixed layer and lower resolution in the deep ocean. In the central Beaufort Sea where the depth is about 3500 m, the vertical resolution decreases from about 6 m for the upper seven layers to as much as 764 m in the deep ocean. The model grids are distributed on a rotated spherical surface with the North Pole at 131.5°E, 8°N. The horizontal resolution is 0.29° × 0.25°, and the time step is 30 min.
 Along the open boundaries, we use radiation boundary conditions for the baroclinic current, and volume transports are specified. The inflow is prescribed as 0.8 Sv through Bering Strait, and the same amount is prescribed as outflow through the Canadian Archipelago. We also prescribe an inflow of 9 Sv Atlantic water into the Arctic Ocean via the Norwegian Sea, and an outflow of 9 Sv along the east coast of Greenland. Along the lateral boundary, the water temperature and salinity are relaxed to the Polar Hydrographic Climatology (PHC) [Ermold and Steele, 2005] within a 9-point buffer zone. The SSH gradient normal to the open boundary is zero.
2.1.3. Ice Model
 The sea ice component of CIOM was developed at Bedford Institute of Oceanography [Yao et al., 2000; Long et al., 2012] using a thermodynamic model based on a multicategory ice thickness distribution function [Thorndike et al., 1975; Hibler, 1980] and a viscous-plastic sea ice dynamics model [Hibler, 1979]. The model considers mechanical and thermodynamic redistributions of ice, ice ridging, and the formation of frazil ice. The ice thickness has seven categories (in units of m: 0.4, 0.8, 1.2, 2, 3, 5, and 7); heat and salt fluxes at the ice-ocean interface are governed by appropriate boundary processes as discussed byMellor and Kantha . The long-wave radiation is given by theSmith and Dobson formulation. Bulk formulations are used to estimate latent and sensible heat fluxes and the wind stress. The air-ice and air-water drag coefficients are set to 2.75 × 10−3 and 1.3 × 10−3, respectively, while the ice-water drag coefficient is 5.5 × 10−3. The sea-ice albedo scheme is based on the formula suggested byKøltzow . The CIOM formulation used here is based on an earlier version by Yao et al. .
2.2. Experiment Design
 The model simulation has been conducted for a storm from 29 July to 4 August 2008. For the experiments described here, the initial conditions for ice and ocean components came from an ice-ocean model simulation from 1970 to 2009, forced with daily NCEP reanalyses [Long et al., 2012]. The initial and boundary conditions for CRCM are specified by 6-hourly CMC analyses. CRCM provides the ice-ocean model with surface air temperature, precipitation rate, surface wind, sea level pressure, surface specific humidity, total cloud and short-wave radiation, while the ice-ocean model passes SST (sea surface temperature) and ice cover to CRCM. CRCM and the ice-ocean model exchange variables every 30 min. In the coupled system, the long-wave radiation, surface sensible and latent heat fluxes depend on the ice category, and they are calculated in CIOM. The integrations start at 0000UTC 29 July and last 7 days.
 Although the surface latent and sensible heat fluxes are calculated in CIOM, there is heat exchange between the coupled ice-ocean model and CRCM. The surface sensible heat flux is calculated using the difference between surface temperature and 2 m surface air temperature, using the bulk formula method. Here, surface temperature is SST for open water, ground surface temperature for land, and ice surface temperature for ice. In CIOM, the 2 m surface air temperature is explicitly passed from CRCM. Similarly, the surface latent heat flux in CIOM depends on the difference between surface humidity and 2 m air humidity, where the 2 m air humidity is calculated in CRCM and passed to CIOM. Indeed, CRCM uses the same approach to estimate the surface heat fluxes, as its SST fields are passed from CIOM, where they are calculated.
 Two experiments were conducted. In the first experiment (EP1), sea ice cover used by CRCM is prescribed by its climatology. The second experiment (EP2) is same as EP1 except that the sea ice used by CRCM is predicted by CIOM (Figure 2) for 2008. Climatologically, most of the area in the Chukchi and Beaufort Seas is covered with sea ice in late July, and the open water only exists near the coast. However, significant open water was present during the summer of 2008. Therefore, comparisons between the two experiments, EP1 and EP2, enable us to investigate the impacts of increased open water on the air-sea interactions during the Arctic storm used in this study. In the simulation, the coupled system is used to reproduce the open water in the Chukchi and Beaufort Seas (Figure 2). This result shows that the coupled system is able to reproduce the observed ice edge in the Beaufort Sea well.
3. The 2008 Summer Storm
 The storm originated in northern Siberia and slowly moved into the Chukchi and Beaufort Seas (Figure 3). On 0:00 UTC 29 July, it was still a weak low pressure system in the Chukchi Sea with a central pressure of 990 hPa. It intensified during the next 24 h as it slowly moved northeastward near the observed ice edge (see Figure 7a). It reached maximum intensity at 0:00 UTC 30 July with a central sea level pressure of 976 hPa (see Figure 7b). Thereafter, it lingered in Canada Basin and slowly weakened (see Figures 3 and 7). Consistent with the low pressure system near the surface, there was also a weak low pressure system at 500 hPa on 0:00 UTC 29 July, located near the Siberian coast. It intensified as it moved eastward along the coast of the Chukchi and Beaufort Seas (see Figure 5). The horizontal distribution of temperature at 500 hPa suggests that the storm is a cold core system (not shown).
 The CRCM-CIOM model system was able to simulate the observed spatial pattern of the 2008 storm and its temporal variation. At the atmospheric surface, the coupled system reproduced the observed low pressure system moving from the Siberian coast to the Beaufort Sea (Figure 4). During its strengthening phase, both the CRCM-CIOM simulation and the CMC analyses show that the storm lingered over the Chukchi Sea along a closed track. During its weakening phase, it moved into the Beaufort Sea along ice edge and eventually into the central Arctic where it dissipated (seeFigure 7a). Thus, the CRCM-CIOM simulation captured the strengthening and weakening processes of the storm life cycle. The model error for central pressure was less than 1 hPa during the first three days (seeFigure 7b). Consistent with the observations (Figure 5), the model simulations also show a low pressure center at 500 hPa (Figure 6). However, compared to the storm propagation in the CMC analyses, our simulation of the storm moved at a slightly lower speed (Figure 7a).
4. Impacts of Increased Open Water
4.1. Storm Track and Intensity
 In the Beaufort Sea, coastal erosion is often related to ocean processes such as waves and nearshore currents, driven by strong surface winds generated by storms [Reimnitz and Maurer, 1979; Atkinson, 2005]. Storm-induced waves can have significant geomorphological, infrastructural, and ecological impacts. For example, as much as 3 m of shoreline retreat occurred in the Tuktoyaktuk region in less than 3 days during each of two separate storms [Solomon et al., 1994]. The impact of storms is moderated by the presence of sea ice, because ice can reduce the effective fetch and limit the potential impact of waves and currents. Concomitantly, the loss of sea ice can potentially enhance the impacts of storms on the coastal region.
 Climatologically, most of the summer storms in the Beaufort Sea are baroclinic systems, with internal atmospheric dynamics playing an important role in their development [Serreze et al., 2001; Simmonds et al., 2008]. In this study, 6-hourly SLP (sea level pressure) fields are used to detect and track the storm. We first screen the SLP fields to identify a local minimum on our polar stereographic grid. We then use a simple nearest neighbor search to track the storm from the preceding (6 h) time step [Long et al., 2009]. Although the tracking of storms that move into the Arctic has a sensitivity to the techniques used [Mesquita et al., 2009], the center of the 2008 storm was well defined (Figure 3), and this approach can track the system well. Comparisons between experiments EP1 and EP2 show no significant impacts of the increased open water on the storm track (Figure 7a), and only rather slight modification of the minimum central sea level pressure (SLP) as the storm intensified (Figure 7b); atmospheric internal processes are the main factor driving the storm development [Murray and Simmonds, 1995]. In terms of storm intensity, both EP1 and EP2 reproduced the minimum central SLP close to that observed in CMC analyses, as the storm strengthened; the SLP difference between EP1 and EP2 is less than 1 hPa. Moreover, as the storm reached the Beaufort Sea, it weakened and the minimum central SLP in EP2 became about 1–2 hPa lower than that of EP1 (Figure 7b), suggesting that the increased open water slightly enhanced the depth of the minimum low pressure. However, compared to the observed storm track in CMC analyses, there is no significant difference between the storm tracks from EP1 and EP2 (Figure 7a).
 A similar pattern can be seen in the time series of maximum winds associated with the storm in Figure 7c, showing the wind speed time series, averaged over an area within 160 km around the maximum wind center, following the storm's development. When the storm moved into the Beaufort Sea, EP1 consistently underestimated the maximum wind by about 4 m/s. The wind speed in the EP1 results is also consistently 3–4 m/s weaker than in the CRCM-CIOM results (EP2), suggesting that the increased open water in the Beaufort has an important intensifying impact on the storm's surface winds.
 In estimating the maximum wind, we averaged over the maximum wind center rather than the minimum SLP in order to emphasize the impacts of sea ice loss on the maximum wind associated with the storm. A caveat of this approach is that the averaging areas may contain sub-areas of lower pressure gradients (e.g., relatively weak winds) and the size of the storm may change during its life cycle, leading to biases. However, if we calculate the wind average centered on the minimum SLP, it will include a relatively large area which has weak wind values (as shown inFigures 8–10).
4.2. Marine Winds
 Surface winds play a fundamental role in air-sea exchanges in general, including the Arctic. Winds not only affect the transfer of momentum from atmosphere to the ocean surface, but also influence heat and moisture fluxes. In the boundary layer, surface winds are often associated with the vertical momentum exchange, depending on the static stability and vertical shear [McFarlane et al., 1992; Edson, 2008].
 Simulated CRCM-CIOM 10 m wind fields in EP2 are compared with blended QSCAT-NCEP (http://dss.ucar.edu/datasets/ds744.4/) data in Figures 8–10. When the storm center moved into the Chukchi Sea at 0:00 UTC on 30 July 2008, the maximum wind speed was located south of the storm center (Figure 8a). It then intensified and slowly moved southeastward into the Beaufort Sea. At 0:00 UTC on 31 July 2008, the maximum wind speed reached about 16 m/s and was located along the Beaufort Sea coast (Figure 8b). Thereafter, the wind associated with the storm slowly decreased (Figure 8c). The winds suggested by the EP2 simulation (Figure 9) reproduced the temporal variation and magnitude of the maximum wind speed suggested by QSCAT-NCEP (Figure 8), with the maximum wind speed located over the open water near the coast. However, compared to results from EP2 and QSCAT-NCEP, results from EP1 underestimated the maximum wind speed by ∼30% (Figure 10). Moreover, in both QSCAT-NCEP data and EP2 results, the wind speed above 14 m/s dominated a relatively large area of the Beaufort Sea at 0:00 July 31, whereas maximum winds were underestimated by about 4 m/s in EP1, and the area of comparable high winds was much diminished, compared to EP2.
4.3. Surface Air Temperature
 In the coupled atmosphere-ice-ocean system, surfaceairtemperature is used to estimate long-wave radiation and surface sensible heat flux, and constitutes an important variable for air-sea interactions.Differences in surface air temperature between EP2 and EP1 are shown in Figures 11b–11d. Over the open water, the air temperature over the Beaufort Sea is higher than that over the Chukchi Sea (Figure 11a). Due to its high albedo, sea ice absorbs significantly less short-wave radiation than does open water, and therefore the open water has a higher surface temperature. As a result, the surface air temperature in EP2 is warmer in the Beaufort Sea than in EP1 results, by as much as 8°C (Figure 11). Overall, the surface air temperature over the open water in the Chukchi and Beaufort Seas increases by 2–8°C, with the maximum temperature changes occurring over open water. The related air-sea fluxes are much larger over open water, than over ice cover; they are enhanced by the strong winds associated with the storm's intensification, further enhancing the energy exchange between the open water and the atmospheric boundary layer. Thus, the strongest impacts of increased open water on surface air temperature in the Beaufort Sea occur when the storm center moves into the region.
4.4. Surface Heat Fluxes
 The loss of sea ice can increase the heat exchange between atmosphere and ocean, and high winds associated with the storm further enhance the exchange process [Perrie et al., 2004; Ren et al., 2004]. As shown in Figures 12 and 13, the loss of sea ice in the Beaufort Sea increases surface latent and sensible heat fluxes into the atmosphere. However, the impacts depend on the storm location. When the storm is located near the ice edge in the Chukchi Sea on July 29, the maximum increases in latent and sensible heat fluxes are located in the Chukchi Sea, while the increases in the Beaufort Sea are quite small, less than 20 Wm−2 (Figures 12a and 13a). Thereafter, the heat flux maximum slowly moves into the Beaufort Sea (Figures 12b and 13b), following the propagation of the storm and the area of maximum winds (Figure 9). The relatively high surface heat fluxes result from the large expanse of open water, combined with the large area of high winds associated with the storm, as it moved into the Beaufort Sea. The maximum differences of latent and sensible heat fluxes between the coupled simulation, EP2, and the uncoupled simulation, EP1, reached about 80 Wm−2, along the Beaufort Sea coast in the region of highest winds on July 31, relatively late in the life cycle of the storm (Figures 12c and 13c). Moreover, as this is a baroclinic storm, the storm track is largely determined by upper level circulation dynamics. Thus, there is little change on the storm development, which is consistent with earlier suggestions by Kuo et al.  that latent heating effects need to occur in the very early stages of the storm development, preceding rapid intensification, to significantly modify storm intensification. Similar results were found in studies of midlatitude storms [Perrie et al., 2005]. Here, the impacts of open water and sea ice loss on the differences in surface heat fluxes rapidly decrease when the storm moved out of the Beaufort Sea, in the decay phase of its life cycle (not shown).
4.5. Vertical Structure of Atmospheric Responses
Figure 14a shows the vertical profiles of the temperature differences between EP2 and EP1. In terms of temperature, these results suggest that the atmospheric responses to the loss of sea ice are mainly confined to the boundary layer due to the relatively static stable Arctic air (Figure 14d), below 925 hPa, with a maximum increase of about 5°C near the surface. However, there are increases of about 0.5∼1°C near 600 hPa which are located above the maximum in the boundary layer, showing that the effect is felt throughout the atmospheric column, to some extent. As a result of the smaller more moderate atmospheric temperature responses in the upper troposphere, and more intense warmer temperatures in the lower troposphere, there are increases in the atmospheric boundary turbulence and reduction of stability (Figure 14d), as well as the downward transport of atmospheric kinetic energy [McFarlane et al., 1992].
Figure 14b shows the vertical profile of horizontal wind speed differences along the Beaufort Sea coast at 0:00 July 31 when the maximum wind speed moved into the region. Consistent with the warming pattern seen in Figure 14a, the increases in wind speed associated with the increased open water mainly occur to the boundary layer. In addition, decreases in wind speed can be seen between 500 hPa and 250 hPa (Figure 14b), suggesting that increased momentum exchange is occurring between the boundary layer and the troposphere. As shown in section 4.3, the increased open water causes surface warming and weakens the boundary stratification. Moreover, the reduced static stability increases the vertical mixing and momentum exchanges during the 2008 storm. Therefore, the increases of surface wind speed related to the increased open water are mainly associated with the reduced stability in the boundary layer (Figure 14d), which strengthens the momentum exchange between the boundary layer and the troposphere.
 The responses of geopotential height are somewhat different, as shown in Figure 14c. Although the geopotential height is slightly lower in the boundary layer above the area with the maximum surface warming (Figure 14a), the maximum decreases in the geopotential height are located between 200 and 400 hPa. Moreover, the geopoential height increases to the east of maximum surface warming, with maximum increases occurring near the surface. Thus, the overall geopotential height responses in Figure 14c slightly tilt westward, suggesting the presence of baroclinic processes in the atmospheric responses.
 The loss of Arctic sea ice significantly increases solar radiation received by seawater, and the atmospheric warming due to the increased open water, is largely limited to the atmospheric boundary layer (Figure 14a). Although the atmosphere in EP1 is statically stable, the warming pattern in the coupled simulation EP2 reduces the stability within the boundary layer (Figure 14d), enhances the momentum exchange between the upper levels of the troposphere and the boundary layer, and results in increased surface wind. However, as mentioned in section 4.4, as this is a baroclinic storm and atmospheric responses to reduced sea ice are largely in the boundary layer, there is no notable change in storm development. As the increased vertical mixing in the atmospheric boundary layer has no significant impact on the mass balance, the changes in the central minimum SLP and the SLP gradient are relatively small. However, increased vertical momentum exchange (Figure 14b) can have significant impacts on the surface wind speed because stronger winds develop in the upper part of the boundary layer. This is the main cause for the changes in wind speed due to the loss of sea ice.
4.6. Time Series of Surface Variables
Figure 15 shows the time series of differences between EP2 and EP1 for surface wind, surface air temperature and surface specific humidity, averaged over the Beaufort Sea, as indicated by the box in Figure 11a. Generally, the loss of sea ice in the Beaufort Sea increases surface wind, surface air temperature and surface specific humidity due to the enhanced surface flux exchanges. Enhanced surface fluxes result in more energized storms. However, as mentioned in section 4.3, these impacts are sensitive to the magnitude of surface wind as well as the storm position. When the storm center is located over the Chukchi Sea and the wind speed in the Beaufort Sea is relatively weak (Figure 9a), the impacts of increased open water are also rather weak. Moreover, when the storm moves over open water in the Beaufort Sea and the surface winds are strong (Figures 9b and 9c), then the impacts of increased open water are also strong. In addition, Figure 15 shows strong diurnal variations in the time series of wind, air temperature and specific humidity, which peaks at 18:00 UTC. During daytime, the loss of sea ice increases the solar radiation received by the surface, decreases the atmospheric stability and increases the surface fluxes. Therefore, when incoming solar radiation is high during daytime, the surface air is relatively warm, humid and windy. At night, solar radiation is much reduced, surface air is relatively cool, and winds are diminished, as reflected by the diurnal pattern.
5. Ocean Responses to the 2008 Storm
5.1. Sea Surface Temperature
Figure 16 shows the changes of sea surface temperature (SST) at 0:00 UTC 30 July, 0:00 UTC 31 July and 0:00 UTC 1 August (when the storm moved into the Beaufort Sea), relative to the SST at 0:00 UTC on 29 July, simulated by EP2, showing the impacts of the storm on the sea surface temperature during its development. The SST near the Beaufort Sea coastal area decreased as the storm moved into the region; the maximum decrease was as much as 2°C in the coastal waters of the southern Beaufort Sea. However, no significant SST response can be seen in the area covered with sea ice. To further understand the process that is responsible for the cooling near the coast, Figure 17 shows the vertical profiles of water temperature and currents. When the storm moved into the Beaufort Sea, the vertical mixing was enhanced by high winds associated with the storm (Figure 17a). The increased vertical mixing brings cold water to the surface and causes the SST cooling seen in Figure 17b. Moreover, the increased vertical mixing due to the storm increases the mixed layer depth by about 5 m; the vertical mixing associated with the storm is mainly located above 14 m depth (Figure 17). Further discussion of interactions between the warm Mackenzie River plume and coastal Beaufort currents is given by Mulligan et al. .
5.2. Fresh Water Content
 In the Arctic Ocean, the largest fresh water storage is located in the Beaufort Gyre (BG), a dominant anticyclonic circulation in the Beaufort Sea [Aagaard and Carmack, 1989]. Studies suggest that changes in the Arctic fresh water balance play an important role in the North Atlantic circulation. A 25% increase in freshwater discharge through Fram Strait maintained for two years can account for the salinity deficit observed in the North Atlantic during the “Great Salinity Anomaly” of the 1970s [Dickson et al., 1988; Aagaard and Carmack, 1989; McPhee et al., 2009].
 In this study, fresh water content (FWC) is calculated using the formula: (34.8 − S)/34.8 dz, if S< 34.8, where S is salinity, z is water depth (m), and L is the uppermost level where S reaches 34.8. On decadal time-scales, the wind-driven circulation alternates between cyclonic and anticyclonic circulation regimes [Proshutinsky and Johnson, 1997]. The BG accumulates significant amounts of fresh water during the anticyclonic regime and releases it during the cyclonic regime [Proshutinsky et al., 2002; Häkkinen and Proshutinsky, 2004].
 During the storm considered in this study, the simulated FWC maximum is located in the central Beaufort Sea (Figure 18a), and decreases from about 20.87 m on July 29 to about 20.76 m on August 2 (Figure 18b). The FWC decrease in the central Beaufort Sea is caused by the Ekman divergence due to the cyclonic surface wind stress associated with the storm (Figure 19). Recent studies show that the changes in the cyclone frequency and intensity in the Beaufort Sea have significant impacts on the FWC in the region [Long et al., 2012]. For example, the strong anticyclonic wind stress dominant during 2004–2009 is associated with the occurrence of fewer cyclones, which caused the significant FWC increase in the central Beaufort Sea due to the enhanced Ekman pumping. An alternate view is given by Morison et al. who suggest that FWC increases in the Canada Basin are balanced by decreases in the Eurasian basin, and changes are due to a cyclonic shift in the ocean pathway of Eurasian runoff forced by strengthening of the west-to-east atmospheric circulation and characterized by an increased Arctic Oscillation (AO) index, rather than driven by the strength of the wind-driven Beaufort Gyre circulation.
 To understand the sensitivity of our results to different heat flux schemes, we conducted three experiments, EN1, EN2 and EN3. EN1 is same as the coupled CIOM-CRCM model system, EP2, as described insection 2. EN2 is the same as EN1 except that the short-wave radiation is calculated in CIOM, where an empirical formulation for clear-sky incoming short-wave radiation byShine and Crane , and the cloud correction of Reed are used to estimate short-wave radiation. EN3 is also the same as EN1, except that the long-wave radiation used in CIOM comes from CRCM. Our results show that there are no significant differences in the storm track (Figure 20a) or intensity (Figure 20b) among these 3 experiments. Figure 20c shows the average SST differences of the three experiments between 0:00 UTC 1 August and 0:00 UTC 29 July. The average SST responses shown in Figure 20c are very similar to the patterns shown in Figure 16c. Therefore, our results are not sensitive to the coupling schemes among the different ensemble members.
 The albedo for sea ice and seawater is about 0.5 and 0.1, respectively. Compared to sea ice, the open water receives about 40% more short-wave radiation. During the 2008 storm, the surface temperature of sea ice is near or just below the freezing point, but the sea surface temperature over the open water can reach 7°C (Figure 17b), suggesting that the loss of sea ice significantly increases the surface temperature. Moreover, the warm sea surface temperature increases the surface air temperature by as much as 8°C (Figures 11b–11d). However, the atmospheric warming is mainly limited to the boundary layer (Figure 14a), decreasing the atmospheric stability (Figure 14d) and enhancing heat, momentum and moisture exchanges in the boundary layer (Figure 15). By comparison, the storm caused about 2°C SST cooling in the Beaufort Sea (Figure 16), due to the enhanced mixing associated with the storm-generated wind (Figure 17), and the associated Ekman divergence due to storm-generated currents decreased the fresh water content in the Beaufort Sea by about 11 cm.
 In this study a coupled atmosphere-ice-ocean model system is used to simulate an intense Arctic storm occurring during 29 July–3 August 2008. The coupled system consists of POM, CRCM and Hibler ice model. The results of the coupled simulation (EP2) are compared to results of an uncoupled simulation (EP1) which assumes ice-covered conditions as prescribed by climatology. The 2008 Arctic storm originated in northern Siberia and slowly moved to open water areas of the Chukchi and Beaufort Seas. Compared to CMC analyses and QSCAT-NCEP data, the coupled model system is shown to simulate well the storm track, storm intensity and the maximum wind speed. In addition, the model simulates well the observed ice edge and open water expanse along the Chukchi and Beaufort Seas during this time period.
 For the 2008 storm, atmospheric dynamics dominate its strengthening and weakening processes. The increase of open water in Chukchi and Beaufort Seas gives a slight decrease in SLP by about 1∼2 hPa during its weakening phase while no significant impact can be seen during its strengthening phase, when the storm is largely over more ice-covered waters off Siberia. The increase of open water had no significant effect on the storm track. However, the loss of Arctic sea ice increased the maximum wind associated with the storm by about 4 m/s, compared to conditions when this area was largely ice-covered (which would be typical of possible ice conditions in past decades), mostly due to the enhanced momentum exchange between the atmospheric boundary layer and the troposphere (Figure 14).
 Due to the decrease in ice cover in the Beaufort, the air temperature over the sea increased by as much as 8°C (Figure 11). However, the atmospheric warming is mainly limited to the boundary layer, with no significant warming in the upper troposphere. Warmer temperatures in the lower troposphere increase the atmospheric boundary turbulence as well as the downward transport of atmospheric kinetic energy, because stability tends to be reduced. As a result, the increased open water in the southern Beaufort Sea enhances storm-generated surface winds, by as much as ∼4 m/s. The increased wind speeds associated with the increased open water are mainly limited to the boundary layer, although decreases in wind speed can be seen between 500 hPa and 250 hPa.
 Furthermore, the Ekman divergence associated with the storm-generated winds decreased the fresh water content in the central Beaufort Sea by about 11 cm. In addition, when the storm moved into the Beaufort Sea, the vertical mixing was enhanced by the high winds associated with the storm. The increased vertical mixing brought cold water to the surface and caused SST cooling. The maximum decrease of surface temperature is as much as 2°C in coastal waters of the southern Beaufort Sea. However, no significant SST changes were found in the region covered with sea ice.
 Support for this research comes from the Federal IPY (International Polar Year) Office of Canada. The authors thank three anonymous reviewers for numerous constructive comments that have helped improve this manuscript.