Rapid sea-ice retreat over the Arctic Ocean has a leading role in Arctic amplification. The sea-ice extent dramatically recovers during every freezing season, so despite the recent summer sea-ice retreat, there must be extraordinary heat exchange between the lower atmosphere and upper ocean. However, the underlying mechanisms for this remain uncertain. Here we show that autumn frontal cyclogenesis is a crucial event in the Arctic air-sea coupled system. Our shipboard Doppler radar and intensive radiosonde observations at the marginal ice zone detected an explosive frontal cyclogenesis, with coupling between upper and lower tropospheric vortices. The thermal contrast between ocean and ice surfaces is likely favorable to cyclogenesis with an identical life-cycle to that at mid-latitudes. This suggests a northward shift of meridional heat transport. The 1.5 K temperature decrease in the upper ocean after the cold front has passed reveals that a large amount of heat is transported into the atmosphere. This is an invaluable example of the fact that sea ice retreat contributes to polar amplification of surface air temperature increase.
 The Arctic lower troposphere has been warming more than twice as fast as the global average [Serreze and Francis, 2006] due to the decrease in Arctic sea ice and the feedback system, i.e., the Arctic temperature amplification [Graversen et al., 2008; Screen and Simmonds, 2010a]. The Arctic temperature amplification has a strong seasonal dependency reaching the strongest in winter and the weakest in summer. One direct trigger is the decrease in sea-ice extent during summer, which results in the anomalous absorption of a large amount of solar radiation into the upper ocean, enhancing subsurface ocean warming [Perovich et al., 2008; Steele et al., 2008]. During autumn, this excessive oceanic heat is returned to the atmosphere before the onset of freezing [Screen and Simmonds, 2010b]. As a result, the lower troposphere is warmed through strong heat exchange between the warm upper ocean and cold lower atmosphere [Serreze et al., 2009]. This warming delays the refreezing and reduces ice growth until the end of spring [Markus et al., 2009]. In a sea-ice-thinning situation, the sea-ice drift speed is increased with deformation and shrinking [Rampal et al., 2009], which accelerates the sea-ice melt during summer. This feedback loop leads to further warming.
 The diminishing trend of sea-ice extent has been partly caused by extratropical cyclones traveling from Eurasia during summer [Inoue and Kikuchi, 2007; Simmonds et al., 2008; Asplin et al., 2009; Simmonds and Keay, 2009]. Active peaks of cyclones over the Arctic Ocean are occur in July and August, when the temperature rises on the continent but stays close to freezing point over ice/ocean surfaces. During autumn, in contrast, the meridional heat contrast between the warm ice-free ocean and the cold ice surface becomes large, likely enhancing a baroclinity [Overland and Wang, 2010]. However, to date, there is little observational evidence of cyclogenesis along the marginal ice zone. Over the Pacific sector, the recent rapid sea-ice retreat has increased observational targets from the sea ice itself to the changes in the ocean and atmosphere. It is now possible for ships that are not ice-breakers to reach and observe the Arctic Ocean. For example, the Japanese R/V Mirai, an ice-strengthened ship, has frequently entered the Arctic Ocean, particularly the ice-free region, and continues to break northernmost records (76.4°N in 2002, 76.6°N in 2004, 78.9°N in 2008, 79.0°N in 2009, and 79.1°N in 2010). Using its up-to-date observation systems, it has investigated unique meteorological events [Inoue et al., 2010]. In September 2010, we encountered a developing extratropical cyclone near the marginal ice zone. Using the shipboard radiosonde and Doppler radar systems, we studied the fine structure of the cyclone, its generation and development mechanisms, and its impact on the upper ocean.
2. Observations Made by the R/V Mirai
 The R/V Mirai conducted an Arctic cruise of the Chukchi and Beaufort Seas from 2 September to 16 October 2010. During the cruise, intensive meteorological observations near a marginal ice zone were made from 23 to 27 September 2010. In this period, 3-hourly GPS radiosonde observations (Vaisala RS92-SGPD) were made. Expendable conductivity-temperature-depth (XCTD) sensors and processing equipment (Tsurumi-Seiki, Yokohama, Japan) were used to measure temperature and conductivity (and hence salinity) near the radiosonde stations. Ancillary datasets included shipboard meteorological surface observations.
 Shipboard C-band Doppler radar observation (RC-52B, Mitsubishi Electric Co. Ltd., Japan) was simultaneously performed. During the observation, a volume scan consisting of 21 plan position indicators (PPI) was made every 10 minutes (detection range: 160 km). Meanwhile, a surveillance PPI scan at an elevation angle of 0.5 was performed every 30 minutes (detection range: 300 km). The velocity-azimuth display (VAD) method [Browning and Wexler, 1968] was used to estimate horizontal wind fields from the Doppler velocity. It was validated that the relationship between the radiosonde and VAD winds was highly correlated in the Arctic cruise [Inoue et al., 2010], suggesting that the VAD winds at 10-minute temporal resolution were valuable for investigating cloudy boundary layers.
 All of our radiosonde data obtained by R/V Mirai was sent to the Global Telecommunication System (GTS), improving the accuracy of reanalyses over the Arctic Ocean compared with other years. Therefore, both upper and surface analyses are useful for understanding Arctic frontal cyclogenesis. Because the treatment of the sea-ice concentration in reanalysis data is a potential source of uncertainty (particularly for surface-heat budgets), we used the ERA-Interim reanalysis dataset [Dee and Uppala, 2009], which treats this explicitly, and thus surface heat fluxes and resultant air temperature must be reproduced well [Inoue et al., 2011]. The time and horizontal resolutions of this reanalysis were 6 hours and 1.5° × 1.5°, respectively.
3. Lifecycle of the Arctic Cyclone
Figure 1 shows the 6-hourly centers defined by the maximum potential vorticity (PV) at 925 hPa and 500 hPa, along with the air temperature at 925 hPa detected by ERA-Interim reanalysis data and the underlying sea-ice concentration measured by the AMSR-E satellite sensor [Comiso et al., 2003]. The temperature gradient at 925 hPa exceeded 2 K per 100 km between the cold sea-ice surface (−12°C) and warm ice-free ocean (−6°C), resulting in a broad baroclinic region for the occurrence of frontogensis. In the following subsections, a lifecycle of the cyclone at each stage is described as by Neiman and Shapiro .
3.1. The Incipient Surface Cyclone
 At 0000 UTC 24 September, an incipient surface cyclone with weak circulation defined by positive PV was situated in the east of the marginal ice zone. During the incipient phase of cyclogenesis, a ‘baroclinic leaf’ [Weldon, 1979] cloud extended eastward from the marginal ice zone (Figure 2a). Southwesterly surface winds transported warm and moist air over the northeastern ice edge (Figure 2e). Below the leaf cloud, our Doppler radar detected cold and warm fronts characterized by intense convection (Figure 3c). The structure was very similar to those that developed in mid-latitude regions [Wakimoto et al., 1992]. The heaviest precipitation along the cold front was organized in narrow rainbands with regular 30–50 km spacing. After passing this surface cold front, the center of the surface cyclone appeared with a central pressure of 1005 hPa (Figures 3b and 3d). In the northwest region of the incipient surface cyclone, another cold front induced by a pre-existing upper cyclone was clearly seen in a cloud image (Figure 2a). This upper cyclone traveled southward in association with advection of a positive PV anomaly (approximately 1.7 PVU; see the trajectory of the positive PV at 500 hPa in Figure 1), anticipating explosive development with coupling between the upper and surface vortices [Takayabu, 1991].
3.2. The Frontal Fracture
 Eighteen hours later (1800 UTC 24 September), the cloud masses associated with cyclones at the surface and upper level greatly expanded and sharpened up into a comma cloud (Figure 2b). These developing cyclones accompanied two cold fronts. The primary cold front originated from the incipient surface cyclone (along the −7°C contour in Figure 2f). Behind the front, the sea surface was covered by stratus clouds (Figure 2b). During the passage of this front, the surface temperature decreased from −2 to −6°C (Figure 3b). The warm and primary cold fronts distinctly separated south of the cyclone. This is a typical feature of extratropical cyclones at mid-latitudes, i.e., frontal fracture [Neiman and Shapiro, 1993]. The secondary cold front consisted of stratocumulus cloud streets (approximately 1.2 km deep: around 1800 UTC 24 September in Figure 3a) in the west of the cyclone parallel to northwesterly cold winds (along the −11°C contour in Figure 2f). The surface temperature dropped to −7.5°C (Figure 3b). The potential vorticity at near the surface (925 hPa) and in the mid-troposphere (500 hPa) increased in 12 hours (Figure 1), suggesting that the positive PV anomaly initiated a strong cyclogenetic response (type B cyclogenesis [Petterssen and Smebye, 1971; Deveson et al., 2002].
3.3. The T-Bone Structure
 At 0000 UTC 25 September, the cyclone developed into its T-bone phase [Neiman and Shapiro, 1993] (Figures 2c and 2g). The warm front seemed to extend southwestward through the cyclone center and into the westerly flow as a secondary cold front (bent-back warm front [Neiman and Shapiro, 1993]. The radar image shows a cumulus cloud street (less than 2 km deep) within the cold air rotating cyclonically around the east side of the cyclone center (Figure 3e). The upper level features gradually became more vertically aligned with the low-level cyclone (Figure 1). Due to the advection of the positive PV anomaly, the tropopause dropped to the 5-km level, resulting in a cold dome (e.g., −36°C in Figure 3a). This is the mature phase of the cyclone. The low deepened to 995 hPa (Figure 3b).
3.4. The Seclusion
 Finally, the cyclone evolved into its warm-core seclusion phase around 1200 UTC 25 September. Due to the advection of a strong PV anomaly on 25 September (greater than 3 PVU: red circle in Figure 1), the developed surface cyclone moved relatively fast following the upper PV anomaly. Although the cold-core induced by the upper PV dominated, a warm core was also evident over Chukchi Sea (Figure 2h). This is a typical feature of mid-latitude extratropical cyclones [Neiman and Shapiro, 1993]. The R/V Mirai was situated between a central vortex and a spiral band (Figure 3f). The increase in surface air temperature was limited up to −6°C because the air originated within the post-cold-frontal airstream. Until 28 September, stratocumulus cloud streets induced by this cyclone covered an ice-free area over the Chukchi Sea (not shown). Overall, we concluded that the lifecycle of this cyclone was very similar to those frequently observed at mid-latitudes.
4. Impact on Ocean Cooling and Seasonal March for the Arctic Warming
 On 14 and 28 September (i.e., before and after this cyclogenesis), we performed intensive oceanic observations by XCTDs along 162°W from 71°N to 75°N in 0.5° latitudinal intervals (orange in Figure 1). This repeat section enabled us to understand how quickly the upper ocean was cooled by the cyclone. Along this section on 14 September, we deployed three Surface Velocity Program (SVP) Drifters (MetOcean, Nova Scotia, Canada) at 72°N, 73°N, and 74°N to estimate the upper ocean current. The distance traveled from 14 to 28 September was 53 km, 12 km, and 142 km, respectively. Therefore, the water mass around at least 73°N stayed at approximately the same place for two weeks, which allowed us to calculate the change in heat content within the water column. Figure 4 shows the latitude-depth cross-sections of water temperature and salinity on 14 and 28 September, respectively. On 14 September, the temperature to 20-m depth exceeded 4°C south of 73°N. Even north of 73°N, it remained warm, up to 2.5°C (Figure 4a). On 28 September, when the cold air mass continued to be advected after rapid cyclogenesis, the temperature decreased dramatically. The difference in temperature of the 20-m depth water column over two weeks reached approximately 1.5 K, corresponding to 123 MJ m−2 (= 102 W m−2 for two weeks: positive upward). Considering that most of the coldness prevailed after cyclogenesis during 25–28 September, the main sea-surface cooling likely amounted 356 W m−2 over 4 days. Even on 28 September, the Doppler radar still detected stratocumulus streets along the 162°W line, suggesting that a large amount of turbulent heat flux was continuously supplied to the atmosphere. This event became the freeze onset of 2010.
 Our observations demonstrated two heat-exchange processes during the transitional season from autumn to winter. First is the frontal cyclogenesis along the marginal ice zone, and the other is the air mass modification and resultant ocean cooling under a cold air outbreak behind the cyclone. The occurrence of cyclogenesis at high latitudes is consistent with a projected northward shift of cold-air outbreaks associated with sea-ice retreat [Kolstad and Bracegirdle, 2008]. It seems that frontal cyclogenesis during autumn works as an accelerator of heat exchanges between the atmosphere and the ocean. In other words, the Arctic warming during autumn and winter is a result of a series of air-sea heat transfers from summer to winter. Therefore, high temporal resolution which captures synoptic characteristics is essential in determining surface fluxes [Simmonds and Dix, 1989]. The temperature gradient between sea and ice surfaces during summer is strengthened due to the absorption of anomalous solar radiation into the ice-free ocean [Perovich et al., 2008; Steele et al., 2008]. Meridional atmospheric heat exchange by frontal cyclogenesis eventually takes place during autumn to dissolve the temperature contrast, and atmospheric warming induced by air-mass modification occurs during early winter. Therefore, the final heat storage as a result of sea-ice reduction occurs in the Arctic atmosphere during winter. This excess of heat likely reduces ice growth until spring, resulting in thinner and more fragile ice during summer. Our invaluable in situ observations provide evidence that frontal cyclogenesis during autumn near the marginal ice zone is a crucial phenomena for the seasonal march of Arctic temperature warming.
 We are greatly indebted to K. Sato, S. Okumura, S. Sueyoshi, N. Nagahama, A. Doi, and W. Tokunaga for conducting radiosonde observations. The authors would like to thank crews of the R/V Mirai. Radar analysis was done by using the Draft tool developed by JMA/MRI. We also thank two anonymous reviewers for their useful comments.
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.