We present a mechanism for wind-driven sea ice export from the Arctic Ocean through Fram Strait for the period 1979–2007, using the output of a high-resolution regional atmospheric climate model. By explicitly calculating the components of the atmospheric momentum budget, we show that not large scale synoptic forcing (LSC), but mainly thermal wind forcing (THW) causes the persistent northerly jet (the Greenland Sea Jet) over Fram Strait. The jet results from horizontal temperature gradients in the atmospheric boundary layer (ABL), set up between cold ABL-air over the sea ice covered western Greenland Sea and the relatively warmer ABL over the ice-free eastern Greenland Sea. From 1993 onwards we find a negative trend in THW, due to a stronger response to climate warming of the ABL over the sea ice covered ocean, compared to that over the ice free ocean. Although on average LSC is smaller than THW, year to year variations in LSC explain most of the interannual variability in the sea ice area flux through Fram Strait (R = 0.81). A small positive trend is found for LSC, partly compensating the decrease in THW in recent years.
 Sea ice forms an important component of the Arctic climate system. It reflects a significant amount of incoming solar radiation, limits the interaction between ocean and atmosphere and the relatively fresh meltwater impacts ocean circulation. Since the start of continuous observations in 1979 [Johannessen et al., 2004; Parkinson and Cavalieri, 2008], Arctic sea ice extent has strongly decreased, especially in summer with record low values in 2005 and 2007 [Comiso, 2006; Comiso et al., 2008]. Recent years also saw a decline in winter ice extent [Comiso, 2006], with a record low January 2011 extent of 13.55 × 106 km2, 8.5% less than the 1979–2000 average (http://nsidc.org).
 Here, we explicitly resolve the components of the lower atmospheric momentum budget over the GS region, to provide a theoretical framework for year to year variations in wind-driven sea ice export through FS. For this, we use output of the high resolution regional atmospheric climate model RACMO2.1/GR [van Meijgaard et al., 2008], for the period 1979–2007, and yearly averaged sea ice area flux data through FS (1979–2007) [Kwok, 2009]. The next section briefly describes the climate model output and sea ice data used. The calculation of the momentum budget components is also discussed. Section 3 gives results of the momentum budget components and the relation to sea ice export. Conclusions are given in section 4.
 Yearly averaged (September to October) sea ice area fluxes through FS for the 1979–2007 period were determined by combining sea ice motion and sea ice concentration retrievals from satellite data [Kwok, 2009]. The average area flux of sea ice through FS is 706 × 103 km2, with a standard deviation of 113 × 103 km2. The uncertainty in the annual estimates is 28 × 103 km2.
2.2. Momentum Budget
 Over a flat surface, the momentum budget in the x-direction can be approximated as follows:
with U and V the wind velocity components in the x- and y-direction, and θ0 the background potential temperature, which is calculated by linearly extrapolating the potential temperature in the free troposphere towards the surface. is defined as the vertically integrated temperature perturbation of the temperature deficit layer (TDL):
with Δθ(z) = θ(z) − θ0(z), the temperature perturbation. hf is chosen well above the TDL top in the free atmosphere. For this study we adopted the coordinate system as used in the RACMO2.1/GR model; U wind is in the x-direction as displayed in Figure 1a and not the easterly component. In this paper we present results for the lowest model layer at approximately 7 m above the surface.
Equation (1) contains two active forcing components, the thermal wind (THW) and the large scale (LSC) forcing. Near the surface, LSC depends on the strength and direction of the flow above the TDL (ULSC and VLSC) and on the vertical gradient in the background potential temperature (θ0) via the thermal wind balance. THW is driven by horizontal gradients in , i.e., horizontal gradients in the ‘cold content’ of the TDL. van Angelen et al.  provides a detailed discussion of LSC, THW, θ0 and over the larger Greenland region. The other terms (horizontal and vertical advection (HADV, VADV), the coriolis force (COR) and momentum flux divergence (FDIV)) are passive forces, i.e., they only become important once flow has been initiated.
3.1. Vertical Profiles
 Figure S1 of the auxiliary material shows average vertical profiles of θ and θ0 (Figure S1a), U and V (Figure S1b) and the momentum budget components (Figure S1c) over FS for the period October 1999 to September 2000. θ starts to deviate from θ0 approximately 2.5 km above the surface (Figure S1a), signifying the large depth of the TDL. This is also the level where U and V start to deviate from their large scale components (Figure S1b). Large scale winds in the free troposphere are from the southwest and become near-zero close to the surface. In the free troposphere, the momentum budget (Figure S1c) shows a simple geostrophic balance. Closer to the surface, however, the momentum budget is dominated by THW, which drives a persistent northerly jet over FS and the Greenland Sea. Figure 1a clearly identifies this boundary-layer jet (70–85°N), which from now on we will refer to as the Greenland Sea Jet (GSJ).
3.2. Momentum Budget
Figure 1b shows the average surface temperature perturbation (Δθ) over the region of interest. The strongest negative perturbations are found over the sea ice, with values of up to −7 K along the North and Northeast coast of Greenland; cold air from the ice sheet transported by katabatic winds [van Angelen et al., 2011] accumulates over the sea ice covered ocean, and can maintain its low temperature owing to the insulating properties of the sea ice. Δθ increases away from the coast, reaching positive values of up to 5 K over the eastern Greenland Sea, where the ocean surface is sea ice free throughout the year. Here, the air is colder than the relatively warm sea surface, leading to a positive temperature perturbation at the surface.
 The vertically integrated temperature perturbation (Figure 1c) shows the same features, with a pool of cold air over the Western Greenland Sea. The associated gradient in results in THW forcing of up to 2 ms−1 h−1. If balanced by COR, this forces a northerly geostrophic near-surface wind of up to 5 ms−1, representing the GSJ. The GSJ is strongest over FS and the central Greenland Sea, coinciding with the boundary between the sea ice covered and sea ice free ocean (Figure 1a).
Figure 1d displays LSC and the associated geostrophic wind. Although LSC is important in the higher atmosphere (Figure S1b), it is on average small near the surface in the FS region, less than 2 ms−1 h−1. LSC becomes large along the east coast further south, which explains the far southward extent of sea ice along the east coast of Greenland, and is in agreement with the findings during the Greenland Flow Distortion Experiment [Petersen et al., 2009].
Figure 2a shows that yearly averaged total forcing in the x-direction (THW + LSC) is a very robust predictor of the yearly averaged wind in the y-direction (V) over FS. V varies between 1 and 5 ms−1, and this interannual variability is fully explained by variations in total forcing (R = 0.98). The seasonal variability in V is pronounced (Figure 2b), with strongest southward winds in winter up to 5 ms−1 and a weak northward flow in summer (1 ms−1). This seasonality can be ascribed to both THW and LSC forcing, with a dominant contribution of THW (Figure 2b). In summer, a combination of less sea ice cover along the Greenland coast and weaker katabatic winds from the ice sheet, diminishes the accumulation of cold air in this region [van Angelen et al., 2011]. As a result the gradient of in the offshore direction is small, resulting in a weaker or absent GSJ.
3.3. Sea Ice Area Flux Through FS
Figure 2c shows the correlation and Figure 3a the correlation field of yearly averaged V with the sea ice area flux through FS. Correlation magnitude is largest over the eastern FS, close to Svalbard. Correlations are strong over the entire Greenland Sea, indicating a single forcing, in this case the GSJ. Figures 3b and 3c display correlation fields of yearly sea ice fluxes, LSC and THW. THW is the dominant forcing of the GSJ; it is also rather constant from year to year, in contrast to LSC. As a result, no significant correlation exists of THW with the sea ice flux (Figure 3b). Figure 3c shows that year to year variations in sea ice flux can be mainly ascribed to variations in LSC, with correlation in the FS > 0.8 (Figure 2d). By definition, LSC near the surface and in the upper air are strongly connected. As a result, correlations between LSC at ∼5000 m elevation and sea ice flux through FS is still strong (Figure 3d). Compared to the surface, the center of action has shifted to the west a result of the average westward tilt of the pressure field with height in response to the large scale SE-NW temperature gradient over Greenland. These results are not sensitive to the size or location of the FS averaging window.
3.4. Temporal Variability
Figure 4 shows 1979–2007 time series of annual total sea ice fluxes through FS and annual average THW and LSC: No significant trend in sea ice export is present; interannual variability is large (σ = 16%) and driven by LSC (red line);THW (blue line) is relatively constant from year to year. The pool of cold air that sets up THW is continuously replenished by katabatic winds flowing off the ice sheet, which is a rather steady phenomenon in NE Greenland [van Angelen et al., 2011]. Moreover, large scale winds that may disperse the cold air are weak in this region (Figure 1d).
 Between 1979 and 1993, THW is rather constant and significantly larger than LSC. From 1993 onwards, a downward trend in THW is visible, which is significant at the 99% significance level. This trend is the result of the different response of the ABL to upper air warming over a sea ice covered ocean, compared to a sea ice free ocean. Figure 4b compares the temperature trend over the west and east side of FS, close to the surface and at 500 hPa (for the definition of the east and west side of FS, see Figure S2). Upper air temperatures show a strong increase of +0.9 K/decade from 1993 onwards, both on the west and on the east side of FS. Close to the surface, temperature trends are much more variable. Over the almost continuously sea ice covered west side of FS we find a positive trend of +0.7 K/decade (similar to the upper air trend), compared to a negative trend of −0.4 K/decade on the east side. This negative trend is the result of a small increase in sea ice cover in recent years over the east side of FS, limiting the heating of the ABL by the ocean. As a result of this differential heating over FS, the gradient in is reduced, leading to a weakening of THW. Temperature trends for the entire region are shown in Figure S2. In general, temperature trends are strongest over sea ice covered regions (up to +1.5 K/decade), whereas over the ice free ocean trends are reduced due to the large heat capacity of the ocean water column. These findings are in agreement with the strong surface heating in the Arctic region in wintertime and reduced heating in summertime [Graversen et al., 2008], as a result of the seasonality in sea ice cover. The weakening of THW is partly compensated by a (less significant) positive trend in LSC.
 These results show that FS sea ice export is maintained by the GSJ, forced by the pooling of cold air, that originates from the Greenland ice sheet, over sea ice in the western GS, and the absence of this pooling further east over the open ocean. This implicates that in the absence of sea ice in FS, or when the FS was uniformly covered by sea ice, the southward transport by the GSJ would be less efficient, which is qualitatively supported by the absence of THW in summer. This represents a negative feedback of Arctic sea ice decline.
 The fact that the atmospheric boundary layer plays a pivotal role in FS sea ice transport implicates that thermal adjustment with the surface is important in explaining long term variability. In the case of the rapidly warming free atmosphere over the GS and Greenland, this has led to a weakening of the THW over FS, which represents another negative feedback on Arctic Sea ice loss.
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.