The mooring array from the physical oceanography group of the Alfred Wegener Institute (AWI) (F1–F10 and F15–F16; Beszczynska-Möller et al., 2012) and the Norwegian Polar Institute (NPI) (F11–F14; Steur et al., 2009) has been located in the Fram Strait since 1997 (Table 1). The mooring data (14 moorings before 2002, and 16 thereafter) contain year-round velocity, temperature, and salinity, and are available from the data publisher PANGAEA (Beszczynska-Möller et al., 2015). The data accuracy and treatment are described in some details by Fahrbach et al. (2001) and Schauer et al. (2004). It deployed along a zonal section along 78°50′N across the whole Fram Strait from western Svalbard Island shelf through the deep part to the shelf east of Greenland Island (Fig. 1), and are instrumented at five standard levels: ~50 m, ~250 m, ~750 m, ~1 500 m where available; ~10 m above the bottom) (Beszczynska-Möller, et al., 2012). The temporal sampling frequency is ranged from 20 min to 2 h. However, the mooring data are discontinuous in some years. F1, F3 and F10 are not available from August 22, 2006 to September 11, 2007, from August 1, 2000 to August 1, 2002, and from August 30, 2005 to September 8, 2006, respectively.

The ERA-Interim product from the European Center for Medium-Range Weather Forecasts (ECMWF) (http://apps.ecmwf. int/datasets/) is a reanalysis set of the global atmosphere with a spatial resolution of 1°×1° on 60 standard levels in the vertical direction from 1979 to the present (Berrisford et al., 2011). It has four rounds of analysis per day at 00, 06, 12, and 18 UTC. The ERA-Interim product has multiple variables, such as sea level pressure (SLP), wind speed, and 2 meters temperature and so on. We use SLP to identify and track polar lows and to composite SLP based on the intensity of the Atlantic inflow in the Fram Strait.

Following Zhang et al. (2004) and Wernli and Schwierz (2006), we used 6-h snapshots of SLP in the ERA-Interim product and developed the following modified procedure to identify and track polar lows: (1) there exist closed SLP contours, and if the SLP difference between its center and its outermost contour is greater than 6 hpa, a polar low candidate is identified; (2) if polar lows candidates appear with a radius less than 1 200 km at the same time, they are considered to be a polar low; (3) if a polar low candidate’s location is within a radius of 600 km of a polar low’s location during the previous 6-h time, this location is considered to be a new location of the existing polar low; (4) if the lifetime of a polar low candidate is shorter than 12 h, it is removed from the polar lows candidates. The modified procedure is almost the same as the method described by Zhang et al. (2004), except the first threshold for choosing the polar lows we need.

Figure 2a gives that current vectors distribution of F1–F10 at different depths from 50 to 1 500 m. It shows that current vectors of F1–F5 and F8–F10 are almost in the same direction from the upper layer to the bottom, indicating the barotropic feature of the flow field, however, it is almost oppositely directed for the current vectors of F6–F7 due to the baroclinic feature (Schauer et al., 2008). Current vectors of F6–F10 are mainly southward, westward or eastward, while the ones of F1–F5 have an obvious northward component which is also shown in Fig. 2b in the vertical direction. The northward component of F1–F5 can reach 2 500 m deep and the maximum value of the northward component has exceeded 15 cm/s. Because we only consider the meridional component of the current vectors in the study of the Atlantic inflow from the Nordic Sea to the Arctic Ocean, the meridional component of F1–F5 is chosen to represent the Atlantic inflow candidates in the Fram Strait. The Atlantic water is defined as warmer than 2°C (Beszczynska-Möller, et al., 2012). Figure 2c shows that the Atlantic inflow is mainly distributed in the upper 500 m and extends west to about 0.5°E. The maximum temperature of the Atlantic water in the Fram Strait can reach 4°C and the temperature decreases from the coast to the inside, which is consistent with the results of Beszczynska-Möller et al. (2012).

Due to the temporary discontinuity of the mooring observation in some years, we calculate the sampling probabilities distribution of F1–F5 (Fig. 3). Figure 3 shows that sampling probabilities are high at most observation points, except in the 750 m layer. The high value is often in the upper layer, the maximum value can reach 0.91 in the 250 m layer of F4, and the minimum value is 0.35 which is located in the 750 m layer of F5. In order to maintain continuity of the Atlantic inflow observation, we choose the observation points whose sampling probabilities are greater than 70% and mean temperature is greater than 2°C as the Atlantic inflow observation points, which are marked by the blue dots and are located in the 50 m and 250 m layers of F1–F2 and F4–F5, respectively in Fig. 3.

Figure 4a gives the time series of daily Atlantic inflow anomaly from the year 1998 to 2010, showing that the daily Atlantic inflow anomaly is generally between –20 cm/s and 20 cm/s, with obvious high-frequency variability and seasonal variability. The maximum and minimum values of the seasonal variability are in winter and in summer, respectively. After removal of the seasonal signal, we apply the residual time series to the wavelet analysis. As shown in Fig. 4b, there are multiple time scale periods, for example, synoptic scale periods, one month period, two months period and so on. Among these periods, synoptic scale periods have a remarkable rule and mainly exist in winter and spring (from January to April), which is emphasized by the black straight full lines in Fig. 4b. Figure 5 shows that, on the synoptic scale, the 3–16 days period is subject to the significance test (as shown in Fig. 5a), and the occurrence days number of the 3 days period reaches the maximum (as shown in Fig. 5b).

Polar lows mainly occur in winter and spring, and their periods are based on the synoptic scale (Rojo et al., 2015). The occurrence time and periods of the polar low are almost the same as the synoptic scale variability of Atlantic inflow in the Fram Strait, which indicates that the temporal coherence of polar low and synoptic scale variability of Atlantic inflow in the Fram Strait is significant. And previous studies have demonstrated that polar lows could impact variability of the Atlantic inflow in the Fram Strait in climate state (Lien et al., 2013). So we deduce that synoptic scale variability of the Atlantic inflow in the Fram Strait is related with polar lows. The polar lows are identified and tracked based on the method described in Section 2. Using the time series of daily Atlantic inflow anomaly of each winter and spring as the Atlantic inflow intensity index (AIII) (Fig. 6a), we composite polar lows locations and SLP anomaly (Figs 6b and c) based on the days when the positive and negative intensity index values are more than one standard deviations away from the mean. A seesaw pattern of SLP anomaly and number of polar lows locations between in the Nordic Sea and in the Barents Sea are shown in Figs 6b and c. As there are less (more) polar lows in the Barents Sea, especially west of the Svalbard Island, and more (less) polar lows in the Nordic Sea, the SLP anomaly is positive (negative) in the Barents Sea and negative (positive) in the Nordic Sea, and further the Atlantic inflow in the Fram Strait weakens (strengthens). The dynamic mechanism will be discussed in Section 3.3.

What is the relative importance of polar lows in the Nordic Sea and in the Barents Sea? To solve this, we need to construct an index to reflect the relationship between occurrence number of polar low and AIII in the Nordic Sea and the Barents Sea. Because the days number of positive and negative AIII is different, it is irrational to use the sum of polar low occurrence number when the AIII is positive or negative to construct the index. It is necessary to construct a relative sense index, so we define an index named polar lows occurrence rate, which means the occurrence number of polar lows divided by occurrence days of polar lows in a spatial region. Figure 7 gives that polar lows occurrence rate anomaly for the days when the positive and negative intensity index values are more than one standard deviations away from the mean in the Nordic Sea and the Barents Sea, respectively. Figure 7 shows that the polar lows occurrence rate anomaly in the Barents Sea is markedly bigger than the one in the Nordic Sea. When the positive and negative AIII values are more than one standard deviations away from the mean, the polar lows occurrence rate anomaly in the Barents Sea can reach –0.56 and 1.1, however, the polar lows occurrence rate anomaly in the Nordic Sea can only reach 0.1 and –0.21. Based on the above conclusion that the synoptic scale variability of the Atlantic inflow in the Fram Strait is related with polar lows, the polar lows occurrence rate anomaly in the Barents Sea is five times more than the one in the Nordic Sea, which means that, under the condition that the strength and distance of polar lows in the Barents Sea are the same as those in the Nordic Sea, the contribution of polar lows in the Barents Sea to AIII variation is five times more than those in the Nordic Sea. So polar lows in the Barents Sea play a more important role in the synoptic scale variability of the Atlantic inflow in the Fram Strait. Note that the conclusions above are not sensitive to the choice of the spatial scope of the Barents Sea and the Nordic Sea. We run a series of tests with different spatial scopes of the Barents Sea and the Nordic Sea, and the conclusions are consistent.

Based on the conclusion above that the synoptic scale variability of the Atlantic inflow in the Fram Strait is related with polar lows in the Barents Sea, we composite the meridional current anomaly in the Fram Strait based on the days when there aren’t or are polar lows in the Barents Sea (Figs. 8a and b). Figure 8 shows that, the meridional current anomaly is positive and negative in the most region of Fig. 8a and Fig. 8b, respectively. As there are no polar lows in the Barents Sea, Atlantic inflow strengthens and the maximum value can reach 0.96 cm/s. However, when polar lows occur in the Barents Sea, Atlantic inflow weakens and southward velocity anomaly can reach 8.3 cm/s. It indicates that polar lows in the Barents Sea can markedly impact the strength of the Atlantic inflow in the Fram Strait. The abnormal values in Figs 8a and b both gradually decrease from east to west, which implies that the impact of polar lows on Atlantic inflow gradually weakens from east to west. Note that the abnormal value in Fig. 8a is obviously smaller than the one in Fig. 8b. This is because that, the number of occurrence days of polar lows in the Barents Sea is 78 in all and only accounts for 10.5% of the total number of days, and the abnormal value in Fig. 8b is more away from the average than the one in Fig. 8a.

Figure 9 shows that the contour of SLP anomaly is dense and closed around the Barents Sea, which indicates that there are strong negative and positive wind stress curls around the Barents Sea in Figs 9a and b, respectively. Lien et al. (2013) have come to the conclusion below by using the numerical model and observation data. Wind-induced Ekman-transport causes sea surface height to reduce off the northern Barents Sea shelf, and the resulting decrease in sea surface height can induce a cyclonic circulation anomaly along the slope encircling the northern Barents Sea shelf area, further inducing the weakening of the Atlantic inflow in the Fram Strait. So according to Lien et al. (2013), the dynamic mechanism by which polar lows in the Barents Sea impact Atlantic inflow in the Fram Strait can be described as follow. Wind stress curl induced by polar lows in the Barents Sea causes Ekman-transport, leads to decrease of sea surface height in the Barents Sea, further induces a cyclonic circulation anomaly around the Barents Sea, and causes the weakening of the Atlantic inflow in the Fram Strait.

At the same time, the direct atmosphere forcing (wind) can affect the upper ocean. The question is: what role does the wind play in the synoptic scale variability of the Atlantic inflow in the Fram Strait. When wind acts on the ocean surface, the upper ocean will produce the Ekman drift, and according to the computational formula of the Ekman depth (Stewart, 2009), the Ekman depth is between 3.7 m and 37.2 m. However, the Atlantic inflow is deeper than 250 m according to Fig. 3, and the Atlantic inflow is barotropic, so we think that wind doesn’t play an important role in the synoptic variation of the Atlantic inflow in the Fram strait.