The impacts of Arctic summer wind on ice-ocean heat processes in the sea ice reduction zone were investigated using a pan-Arctic ice-ocean model. The interannual simulation demonstrated event-like sea ice melting because of enhanced solar radiation input and upward ocean heat transfer in the Canada Basin area. Both factors were derived from mechanical ice divergence under cyclonic wind stress. Atmospheric heat penetration into the newly formed open leads was estimated to be a primary contributor to sea ice lateral/bottom melting. The vertical ocean heat flux associated with wind-driven upwelling/mixing was the secondary factor on a monthly time scale. The modeled interannual variability indicated that the relative role of internal ocean dynamics in sea ice variability was greater under cyclonic wind patterns compared with anticyclonic wind conditions.
 Arctic summer wind has remarkable impacts on sea ice distribution through various processes. For example, the anticyclonic anomaly of atmospheric circulation in the western Arctic has been reported to keep much sea ice inside the Beaufort Gyre region and to support its survival against summertime thermal melting [Proshutinsky and Johnson, 1997]. The pan-Arctic-scale seesaw pattern, which is composed of high sea level pressure (SLP) over the Canadian Arctic and low SLP covering the Siberian side, produces an intense meridional wind anomaly and increased sea ice export from the Arctic Ocean to the Greenland Sea through the Fram Strait [Wu et al., 2006]. Previous studies have investigated the idea that the 2007 record low of Arctic summer sea ice extent was substantially triggered by this Dipole Anomaly pattern [e.g., Wang et al., 2009]. The summertime sea ice extent has also been discussed using a statistical index for the Northern Annular Mode proposed by Ogi et al. . Recent summer sea ice retreat showed a significant correlation with a negative trend of the summer Northern Annular Mode [Ogi and Yamazaki, 2010]. However, the zonal-mean signal does not always linearly explain the summer sea ice conditions, because regional patterns of SLP, sea ice, and even ocean fields change dramatically every year. Atmospheric low pressure systems entering the central Arctic from the Siberian and North American coasts temporarily increase the sea ice extent [e.g., Ogi and Wallace, 2007]. Consequently, the seasonal ice decline slowed in July 2010, and the minimum ice extent was more than that in 2007 [Stroeve et al., 2012]. On the other hand, cyclonic wind associated with low SLP favors ice divergence, which can accelerate thermal melting via shortwave absorption in open leads exposed between ice floes during summer [Serreze et al., 2003]. Thus, ice-albedo feedback is also expected to proceed under the Arctic summer cyclones.
 The impacts of ocean dynamics on the observed sea ice retreat remain uncertain. Recent field campaigns have captured a great amount of ocean heat stored in the Canada Basin interior. The near-surface temperature maximum (NSTM) was originally formed by enough solar heat absorption in open water area [Jackson et al., 2010]. The obtained heat was eventually isolated below the stratified surface layer that receives sea ice melt water. A modeling analysis suggested a NSTM deepening process via Ekman downwelling in the central Beaufort Gyre region [Steele et al., 2011]. The winter survival of the NSTM resulted in a warming trend and northward expansion in the 2000s [Jackson et al., 2010; Mizobata and Shimada, 2012]. In addition, ocean heat transport from the Alaskan northern coast toward the western Canada Basin produced subsurface temperature maximum layers along the pathway of Pacific summer water for several recent years [Shimada et al., 2006]. Although the ocean heat accumulated in the near-surface and subsurface layers has the potential to melt overlying ice and/or to delay winter ice growth, the mechanism and timing for heat release in sea ice reduction zone are still unclear. Wind-driven upwelling and mixing events can be regarded as possible opportunities [Yang et al., 2004].
 To characterize summertime wind pattern, we calculated wind stress curl in the Canada Basin area from June to August in each year. The data source is described in section 2. In our analysis, the Canada Basin area was defined to be enclosed by a 3000 m isobath (Figure 1a). This region covers a wide area of the western Arctic basin, where the oceanic Beaufort Gyre circulates, but excludes the Makarov Basin and the Chukchi Borderland. The total area is 8.8 × 105 km2. Cyclonic wind circulation was detected in the summers of 2002 and 2003, whereas an anticyclonic wind circulation appeared from 2006 to recent years (Figure 1b). In 2012, the wind pattern changed from anticyclonic to cyclonic in August. In this study, an interannual simulation was performed using a pan-Arctic ice-ocean model to compare ice-ocean heat processes under different wind patterns, especially in the Canada Basin area.
2 Experimental Design and Model Performance
 The coupled ice-ocean model used in the present work is Center for Climate System Research Ocean Component Model (COCO) version 4.9 [Hasumi, 2006]. The sea ice part includes a one-layer thermodynamic formulation [Bitz and Lipscomb, 1999], elastic-viscous-plastic rheology [Hunke and Dukowicz, 1997], and a seven-thickness-category configuration based on that of Bitz et al. . The ocean component is a free-surface ocean general circulation model formulated with the turbulence closure scheme of Noh and Kim  for the surface mixed layer. The model domain contains the entire Arctic Ocean, the Greenland-Iceland-Norwegian seas, and the northern part of the North Atlantic (Figure 1a). The horizontal resolution is about 25 km, and there are 28 hybrid sigma-z vertical levels. The model was integrated from 1979 to 2012 after a 10 year spin-up (see the experimental details in Watanabe ). The atmospheric forcing components were constructed from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis daily data set [Kalnay et al., 1996]. The wind stress was calculated from the SLP following a formulation adopted in the Arctic Ocean Model Intercomparison Project (AOMIP) (http://www.whoi.edu/page.do?pid=30565). At the Bering Strait, Pacific water inflow with a seasonal cycle was provided. For comparison, the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) sea ice concentration products were obtained from the web page of the Global Change Observation Mission 1st - Water, Japan Aerospace Exploration Agency (https://gcom-w1.jaxa.jp/auth.html).
 The sea ice and ocean properties demonstrated by the COCO model captured well-known features, as described in Watanabe . The simulated Arctic sea ice gradually decreased after the mid-1980s, as reported in many studies [e.g., Rothrock et al., 2008]. The decline trend of annual mean sea ice volume in the entire model domain was 2.7 × 103 km3 decade–1. This rate is within the range of previous AOMIP experiments [Gerdes and Koberle, 2007; Schweiger et al., 2011]. The sea ice export through the Fram Strait fluctuated around 3.0 × 103 km3 yr–1, which is close to the estimate of Vinje . The spatial distribution of sea surface height represented that the model reproduced the major basin-scale ocean circulation, such as the Beaufort Gyre and the Transpolar Drift (Figure 1a). The trans-Arctic hydrographic structure in August 2012 captured the deepening of halocline layers in the Canada Basin relative to the salinity data of Polar Science Center Hydrographic Climatology version 3.0 [Steele et al., 2001] (Figure 1c). This feature was accompanied by a remarkable surface freshening in the 2000s [Rabe et al., 2010]. Whereas the boundary between the NSTM and underlying Pacific-origin warm waters was somewhat vague, a subsurface temperature maximum above 100 m depth in the southern Canada Basin was also produced.
3 Wind-Driven Ice-Ocean Processes
 The simulated ice-ocean fields in the Canada Basin area were compared between the summers of 2003 and 2007. First, the mechanical divergence of daily mean sea ice velocity was calculated in each model grid and then averaged in the Canada Basin area (Figure 2a). As indicated by a positive anomaly of wind stress curl (Figure 1b), the sea ice divergence was associated with cyclonic ice drift under low SLP in 2003. Several positive peaks appeared when distinct cyclone events occurred (e.g., Figure 3). The highest peak of 1.5 × 10–2 d–1 was recorded on 27 July 2003. The ice divergence directly produced open water areas between ice packs and enhanced atmospheric heat input into the ocean surface. Conversely, anticyclonic wind circulation around high SLP caused sea ice convergence throughout July and August in 2007. Hence, we regarded 2003 and 2007 as cyclonic and anticyclonic cases, respectively.
 The area total sea ice volume showed a similar seasonal transition until June of both years (Figure 2b). The sea ice reduction continued until the volume reached a minimum at the beginning of September in 2003. In 2007, the sea ice volume was retained in the Canada Basin. Note that the 2007 record low extent reflected an anomalous sea ice retreat in the Siberian Arctic rather than in the Canadian side. The factors for sea ice volume change in the specified area can be divided into net ice export to surrounding seas via advection, and local thermodynamic melting (Figures 2c–2d). We found that cyclonic wind stress drove sea ice export from the Canada Basin interior toward the channels of the Canadian Arctic Archipelago and north of Greenland in 2003 (not shown). Simultaneously an influx of thinner ice from the East Siberian side occurred. Cyclonic wind is certainly thought to expand sea ice extent (i.e., total area including open water fraction inside sea ice margin) [Stroeve et al., 2012]. The modeled total area in the ice-covered region of the Canada Basin area, where the sea ice concentration in each model grid was above zero, reached a September minimum of 8.7 × 106 km2 (7.5 × 106 km2) in 2003 (2007). The AMSR-E products also indicated that the minimum area in 2003 (7.5 × 106 km2) was larger than that in 2007 (6.7 × 106 km2). On the other hand, the mean sea ice thickness in the fixed region should rather decrease under cyclonic (divergent) wind, and this situation becomes a precondition of rapid sea ice melting. In contrast, the Beaufort High emerging in 2007 [Moore, 2012] worked on sea ice accumulation inside the Canada Basin area. The thermal melting rate of sea ice also had a positive anomaly in 2003 relative to 2007 (Figure 2d). Sea ice melting explicitly intends the reduction of sea ice volume without any increase in other regions. Besides, the decrease in sea ice concentration causes a positive feedback loop via solar radiation input into open water areas fronting sea ice floes.
 In the present model setting, where downward light attenuation in the water column was parameterized by Rosati and Miyakoda , approximately 60% of the absorbed solar heat is immediately assigned for sea ice lateral/bottom melting in ice-covered grids. The residual 40% of the shortwave flux penetrates into deeper layers (below 2 m) that are located at the bottom of the uppermost model layer. Wind-driven momentum input into the ocean surface can work on heat exchange between cold surface layers and underlying subsurface warm layers via Ekman upwelling and turbulent mixing processes. The ocean heat provided to the uppermost layer is also consumed for ice lateral/bottom melting. The COCO model calculates the ice surface melting and lateral/bottom melting, separately. The simulated melting rate on ice surfaces in the Canada Basin area had a different period of its seasonal maximum (Figure 2e) but a similar amount on average in the summers of 2003 (899 km3 from June to August) and 2007 (1068 km3). Hence, atmospheric thermal conditions such as air temperature and radiative fluxes just above the ice surface would not account for the essential parts of the ice volume difference at the end of melting season between two years. The daily rate of sea ice lateral/bottom melting showed a lag of the seasonal maximum behind the ice surface melting by about one month, as indicated by previous findings [Steele et al., 2010], and several sharp peaks in 2003 (Figure 2f). These peaks were recorded in phase with the enhanced ice divergence (Figure 2a). The summertime total from June to August was 684 km3 (282 km3) in 2003 (2007).
 To address the heat source, atmospheric heat input into the open water fraction is shown in Figure 2g. The average area was confined to the ice-covered model grids of the Canada Basin area. In accordance with the reduced sea ice concentration (i.e., many open leads inside sea ice margin), the heat absorption was larger in 2003 than in 2007. The monthly heat input of 92 MJ m–2 in August 2003 was equivalent to a sea ice melting rate of 30 cm, where the sea ice density and heat of fusion are assumed to be 9.0 × 102 kg m–3 and 3.4 × 105 J kg–1, respectively. The ocean heat loss, which was exactly the reduction in ocean heat content due to sea ice melting, was then checked (Figure 2h). The model simulation represented the event-like ocean heat loss in summer 2003. In 2007, the signal was relatively weak. It is generally expected that sea ice divergence associated with cyclonic wind stress favors upwelling flow under the sea ice cover. The Ekman upwelling calculated from the modeled ocean surface stress (i.e., the combination of wind and ice-ocean stresses) also had remarkable peaks in late July and early August 2003 (Figure 2i). However, the Ekman contribution for sea ice bottom melting was limited to ocean heat release just in surface several meters, because the upwelling velocity was smaller than 5 m mon–1 even in the model grid where the maximum value was recorded. Another candidate contributing to the upward heat transfer is wind-driven turbulent mixing. Synoptic cyclones produced a robust momentum input on the ocean surface (Figure 2j). The area-mean ocean surface stress increased up to 0.2 Pa during a strong cyclone event in early August 2003, whereas anticyclonic wind resulted in smaller ocean surface stress magnitude in 2007.
 In the COCO model, vertical mixing coefficients in the water column are diagnosed following the closure scheme of Noh and Kim . In this scheme, the temporal evolution of turbulent kinetic energy is solved using a function of the vertical shear of horizontal velocity and the vertical density gradient. In August 2003, a synoptic cyclone passed over the Canada Basin (Figure 3a) and the enhanced ocean surface stress became a crucial turbulent kinetic energy source. The model demonstrated that after the maximum of momentum input, the intensified turbulent mixing reached a depth of 20 m through strong density stratification (Figures 3b–3c). The simulated vertical diffusivity exceeded 0.9 m2 s–1 under the storm activity. This value was an order of magnitude larger than the eddy-induced diffusivity [Watanabe et al., 2012]. Part of the ocean heat stored in the near-surface layers was then released in this event.
 Interannual variability in the linkages among monthly sea ice divergence, atmospheric heat input, and ocean heat loss is plotted in Figure 4. It was confirmed that the situation in 2002 (2008) was very similar to that in 2003 (2007). Therefore, the COCO model result supports the idea that cyclonic wind circulation promoted sea ice lateral/bottom melting via both atmospheric heat input to open leads and ocean heat release due to upwelling/mixing. In this regard, August 2012 can be categorized as a cyclonic case. In 2012, an extreme summer storm event is considered to be a supplementary factor for drastic sea ice reduction especially in the western Arctic and the consequent record minimum extent of Arctic sea ice [Zhang et al., 2013]. In our simulation, the monthly ocean heat loss for sea ice melting was 23 MJ m–2 in August 2012, which did not exceed that in 2003 (28 MJ m–2) but was larger than that in anticyclonic years (Figure 4b). This amount corresponded to 23% of the atmospheric heat input into open leads. Therefore, we should pay attention to both the surface heat budget and the internal ocean dynamics in analyses of sea ice variability.
4 Summary and Discussion
 The impacts of Arctic summer wind on ice-ocean heat processes were investigated using the pan-Arctic ice-ocean model. The interannual simulation demonstrated event-like sea ice melting due to enhanced absorption of solar radiation and upward ocean heat transfer in the Canada Basin area. Both factors were linked to mechanical ice divergence under cyclonic wind stress. The heat loss for sea ice lateral/bottom melting estimated in our model simulation revealed that atmospheric heat penetration into newly formed open leads had larger contribution than vertical ocean heat flux associated with wind-driven upwelling/mixing on a monthly time scale. An important finding of the present modeling analyses is that the relative role of ocean processes in sea ice variability becomes greater under cyclonic wind patterns. Toole et al.  suggested that strong density stratification below the surface mixed layer had limited heat flux from deeper warm layers to the sea ice bottom in the Beaufort Gyre region. However, extreme storm activities would induce ocean heat release, even on hourly to daily time scales, in contrast to the Ekman effect. Although precipitation during cyclone events might sometimes cause ocean surface stratification, the meteorological water flux (precipitation minus evaporation in the NCEP reanalysis) was less than the amount of sea ice melt water by one or two orders during summer. Whereas the emerged summer Beaufort High was accompanied by a negative anomaly of cloud cover in 2007 [Schweiger et al., 2008], it can also be interpreted that stronger solar radiation was compensated by higher albedo owing to wind-driven sea ice accumulation. For further accurate estimates, possible biases in the model products partly derived from atmospheric forcing data and unresolved dynamics should also be considered. For example, the multiple reanalysis comparison revealed the excessive downward shortwave radiation in NCEP data over the central Arctic Ocean [Serreze and Hurst, 2000]. Sensitivity experiments using other turbulence closure schemes [e.g., Nakanishi and Niino, 2009] are a practical way to assess uncertainties in mixed-layer processes. These concerns will be addressed in the next steps.
 E. W. was funded by a Grant-in-Aid for Scientific Research (S) of Japan Society for the Promotion of Science (JSPS) (KAKENHI, 22221003). M. O. is funded by a Grant-in-Aid for Young Scientists (B) (22740317) and a Grant-in-Aid for Scientific Research on Innovative Areas (22106010). Modeling experiments were executed using Earth Simulator version 2 of Japan Agency for Marine-Science and Technology (JAMSTEC). AMSR-E data were supplied by GCOM-W1 data providing service of JAXA. The constructive comments of Dr. Jiayan Yang and an anonymous reviewer were beneficial for the presented product.