Portal de Periódicos da CAPES

Variability of the Norwegian Atlantic Current and associated eddy field from surface drifters

2011; American Geophysical Union; Volume: 116; Issue: C8 Linguagem: Inglês

10.1029/2011jc007078

2156-2202

Maria Andersson, Kjell Arild Orvik, J. H. LaCasce, Inga Monika Koszalka, Cecilie Mauritzen,

Ocean Waves and Remote Sensing

Journal of Geophysical Research: OceansVolume 116, Issue C8 Free Access Variability of the Norwegian Atlantic Current and associated eddy field from surface drifters M. Andersson, M. Andersson [email protected] Geophysical Institute, University of Bergen, Bergen, NorwaySearch for more papers by this authorK. A. Orvik, K. A. Orvik Geophysical Institute, University of Bergen, Bergen, NorwaySearch for more papers by this authorJ. H. LaCasce, J. H. LaCasce Institute of Geosciences, University of Oslo, Oslo, NorwaySearch for more papers by this authorI. Koszalka, I. Koszalka Institute of Geosciences, University of Oslo, Oslo, NorwaySearch for more papers by this authorC. Mauritzen, C. Mauritzen Norwegian Meteorological Institute, Oslo, NorwaySearch for more papers by this author M. Andersson, M. Andersson [email protected] Geophysical Institute, University of Bergen, Bergen, NorwaySearch for more papers by this authorK. A. Orvik, K. A. Orvik Geophysical Institute, University of Bergen, Bergen, NorwaySearch for more papers by this authorJ. H. LaCasce, J. H. LaCasce Institute of Geosciences, University of Oslo, Oslo, NorwaySearch for more papers by this authorI. Koszalka, I. Koszalka Institute of Geosciences, University of Oslo, Oslo, NorwaySearch for more papers by this authorC. Mauritzen, C. Mauritzen Norwegian Meteorological Institute, Oslo, NorwaySearch for more papers by this author First published: 26 August 2011 https://doi.org/10.1029/2011JC007078Citations: 46AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Abstract [1] The Norwegian Atlantic Current (NwAC) and its eddy field are examined using data from surface drifters. The data set used spans nearly 20 years, from June 1991 to December 2009. The results are largely consistent with previous estimates, which were based on data from the first decade only. With our new data set, statistical analysis of the mean fields can be calculated with larger confidence. The two branches of the NwAC, one over the continental slope and a second further offshore, are clearly captured. The Norwegian Coastal Current is also resolved. In addition, we observe a semipermanent anticylonic eddy in the Lofoten Basin, a feature seen previously in hydrography and in models. The eddy kinetic energy (EKE) is intensified along the path of the NwAC, with the largest values occurring in the Lofoten Basin. The strongest currents, exceeding 100 cm s−1, occur west of Lofoten. Lateral diffusivities were computed in five domains and ranged from 1–5 × 107 cm2 s−1. The Lagrangian integral time and space scales are 1–2 days and 7–23 km, respectively. The data set allows studies of seasonal and interannual variations as well. The strongest seasonal signal is in the NwAC itself, as the mean flow strengthens by approximately 20% in winter. The EKE and diffusivities on the other hand do not exhibit consistent seasonality in the sampled regions. There are no consistent indications of changes in either the mean or fluctuating surface velocities between the 1990s and 2000s. Key Points Lateral diffusivities in the Nordic Seas range from 1–5 × 107 cm2 s−1 Lagrangian integral time and space scales are 1–2 days and 7–23 km The mean flow of the NwAC strenghtens by approximately 20% in winter 1. Introduction [2] During the International Polar Year (IPY) a dedicated surface drifter project, the POLEWARD experiment, was conducted in the Nordic Seas during the period 2007–2009. In the experiment, 148 surface drifters, drogued at 15 m, were released at various locations west of the Norwegian coast, in the Barents Sea and south of Spitsbergen. The primary goal was to study the evolution of the warm waters entering the Nordic Seas as they flow north toward Spitsbergen, cool and become denser. The main feature of this flow is the Norwegian Atlantic Current (NwAC), the northern extension of the North Atlantic Current (NAC). This is the shallow branch of the thermohaline circulation, and the cooling and freshening which occurs in the Nordic Seas is fundamental in determining the density of the North Atlantic Deep Water which returns to the North Atlantic [Mauritzen, 1996]. [3] The NwAC commences at the Iceland-Scotland ridge, continuing northward in two branches [Dietrich et al., 1980; Orvik and Niiler, 2002]. It is strongly steered by topography and its path can accordingly be traced along the isobaths [Poulain et al., 1996; Orvik and Niiler, 2002]. The eastern branch of the NwAC lies over the continental slope and has a significant barotropic component. A portion of this branch flows into the Barents Sea and the Arctic Ocean, linking the North Atlantic, Arctic Ocean and the Barents Sea [Orvik and Skagseth, 2005]. The western branch is a baroclinic jet along the Polar front; it can be traced throughout the Nordic Seas toward the Fram Strait [Orvik et al., 2001; Orvik and Niiler, 2002]. [4] There is significant exchange between the two branches, as has been observed both with surface drifters [Poulain et al., 1996; Orvik and Niiler, 2002; Jakobsen et al., 2003] and with subsurface floats [Søiland et al., 2008]. [5] Using position and velocity data from surface drifters, Poulain et al. [1996] computed quasi-Eulerian statistics using drifter data from the 1990s. The authors found a vigorous eddy field near the principal currents and in the Lofoten Basin. The latter has a distinct hydrographic signature as well, as the isopycnals of the warm Atlantic water lay deeper in that basin than further north and south [Orvik, 2004]. The Lofoten Basin evidently acts as a reservoir for Atlantic water, facilitating heat exchange with the atmosphere. There are indications that warm water is transported into the basin from the eastern branch of the NwAC via eddies [e.g., Rossby et al., 2009]. The eddies also affect the circulation in the basin. In particular, model simulations suggest that anticyclonic eddies migrate into the interior where they sustain a quasi-permanent anticyclonic circulation Köhl [2007]. [6] Subsequent drifter studies in the Nordic Seas have provided greater detail of the flows [Orvik and Niiler, 2002; Jakobsen et al., 2003]. Recently, Koszalka et al. [2011] compared two methods of estimating mean velocities and diffusivities from surface drifter observations from the Nordic Seas. It was concluded that the clustering method, relying on a clustering algorithm, provides finer resolution in densely sampled regions compared to the more conventional binning method. The binning technique, on the other hand, is suitable for temporal variability analysis. In the present study, the POLEWARD data is used in combination with the existing data to obtain new estimates. As the combined data set is larger, the statistical estimates are more robust than previously. And as they span a longer period of time, they permit a study of the seasonal variability and of the contrast between the 1990s with the 2000s. For the first time, temporal variability analysis of diffusivities in the Nordic Seas using drifter data are performed. [7] Previously, Orvik and Skagseth [2003] suggested that the variability in the Nordic Seas is correlated with the wind stress curl over the North Atlantic. If so, the variability might also be correlated with the NAO index. The early 1990s correspond to strong positive NAO years, while the 2000s had more fluctuating values of NAO [Hurrell, 1995], and this would imply a difference in variability between the 2 decades. [8] The paper is organized as follows. In section 2, the drifters, the deployments and the methods of data processing are discussed. In section 3, the Eulerian mean circulation, Lagrangian statistics, and seasonal and interannual variability are presented. Section 4 concludes with a discussion of the results. 2. Data and Methods [9] The 148 POLEWARD surface drifters that were deployed in the Nordic Seas in the period from June 2007 to July 2009 yield 15,200 buoy days. The drifters were deployed under the Surface Velocity Program (SVP) of the Global Drifter Program. The historical data is from SVP drifters in the same region. The combined set comprises more than 77,000 buoy days of position from 487 drifters, from June 1991 to December 2009. [10] The drifter consists of a surface buoy, with a transmitter and a temperature sensor, and a subsurface drogue centered at 15 m depth. The drifter has a tether strain sensor to verify the presence of the drogue. Only data from drogued drifters was used in this study. The drifters follow the near-surface flow in the mixed layer, including the Ekman drift [Niiler et al., 1987; Lumpkin et al., 2001]. They are tracked by the Argos satellite system, yielding positions with 150–1000 m accuracy, with 16–20 satellite fixes per day. The AOML's (Atlantic Oceanographic and Meteorological Laboratory) drifter Data Assembly Center (DAC) assembles the raw data, applies quality control, and interpolates them to uniform 6 hourly intervals using a kriging technique [Lumpkin and Pazos, 2006]. This initial processing of the data, including quality control and interpolation of the positions to 6 h interval, is described in detail by Hansen and Poulain [1996]. [11] The drifters were released at 5 deployment sites along the path of the NwAC: the Svinøy Section, west of Lofoten Islands, Bjørnøya, the Barents Sea opening and in the Lofoten Basin. They were deployed during the June/July 2007 period, in October 2007, March 2008, October 2008, and in June/July 2009. The deployment positions, along with the major pathways of Atlantic water, are shown in Figure 1. Figure 1Open in figure viewerPowerPoint Major pathways of Atlantic water with bathymetry of the Nordic Seas and sites of drifter deployments by the POLEWARD project (red stars), historical deployments (green stars), and recent non-POLEWARD deployments (black stars). [12] The area of investigation stretches from 30°W to 30°E and 60°N to 80°N. It encompasses the Greenland, Lofoten, and Norwegian basins; the Iceland Plateau; the Barents Sea opening and the region where the Atlantic water enters the Nordic Seas. For the subsequent statistics, the area is limited to 15°W to 19°E and 60°N to 74°N, as this is the region with the greatest data availability. The different areas and the drifter trajectories are shown in Figure 2. Figure 2Open in figure viewerPowerPoint Drogued drifter trajectories from GDP (Global Drifter Program) database; POLEWARD trajectories in red and non-POLEWARD data in blue. The five domains where the Lagrangian statistics have been estimated are shown. IFF, Iceland-Faroe Front; IP, Iceland Plateau; NB, Norwegian Basin; LB, Lofoten Basin; NwAC, Norwegian Atlantic Current. [13] Drifter velocities were obtained by differencing their positions. To remove the high frequency current components, especially the tidal and inertial currents, the interpolated positions were low-pass filtered with a Butterworth filter with a cut-off period of 25 h. Other than this filtering the data were not averaged in time. Drifters with time gaps greater than 1 day were treated as separated drifters. [14] Drifters are not fully Lagrangian, as they only track horizontal velocities. Nevertheless, they can be used to infer transport properties of the surface flow [Davis, 1991]. Specifically, the evolution of a passive tracer can be described by the mean initial field from the statistics of single particles [Davis, 1983]. This involves both the mean velocity and the diffusivity, both of which are assumed to vary in space. The diffusivity parametrization assumes that the dispersion is due to small-scale eddies, i.e., that there is a scale separation between eddies and the mean flow. [15] To obtain pseudo-Eulerian averages, we grouped the drifter velocities into geographical bins and calculated means and variances [e.g., Poulain and Niiler, 1989; Poulain et al., 1996; Swenson and Niiler, 1996; Fratantoni, 2001]. The robustness of such maps is discussed by Lumpkin [2003]. The internal Rossby radius in the Nordic Seas is ∼10 km [Chelton et al., 1998] and the dominant eddy scale is about 50 km [Poulain et al., 1996]. As such, we grouped the observations into 2° longitude × 1° latitude bins, or roughly 110 km square, which provides a reasonable depiction of the major circulation features. To ensure that the bins contain sufficient data to form statistically reliable values only bins containing more than 50 observations and from at least 2 different drifters were used [Poulain et al., 1996; Poulain and Niiler, 1989]. Smaller bins does not provide enough data in bins to study seasonal and decadal variability. [16] In each bin, U(x, y) is computed as the ensemble average of all available velocity measurements u(x, y, t), where x and y denote the position and t the time. The residual velocity, u′(x, y, t), is defined as the deviation of u(x, y, t) from U(x, y). The eddy kinetic energy (EKE) is 1/2(〈u′2〉 + 〈v′2〉), where u and v are the zonal and meridional velocities, respectively, and where the brackets 〈〉 denote an average over the particle ensemble. All observations were accorded equal weight, as described by Davis [1991]. [17] Principal component analysis on the velocity fields was performed to determine the principal axes of current velocity variance (kinetic energy) [Emery and Thomson, 2001]. The trend and mean were removed from the data in each bin. [18] In calculating the Lagrangian averages, we assume the velocity statistics are stationary. The diffusivity is calculated following Taylor [1921], as prescribed by Davis [1991] where X(t) is the displacement and τ is the time lag. The diffusivity thus is the integral of the velocity autocorrelation and is calculated from the residual velocities. If the flow is stationary, then where u′2 is the velocity variance and R(τ) the time-lagged velocity autocorrelation, [Lumpkin et al., 2001; LaCasce, 2008]. This equation holds in locally homogeneous regions. The integral of the autocorrelation, normalized by the variance of the residual velocities, is the Lagrangian time scale.(For definitions of the Lagrangian time and length scales see Equations A1 and A2 in Appendix A.) Following Taylor [1921], the diffusivity obtains in the limit t → ∞. But as the record lengths are finite and noise dominates the autocorrelation function at large lags [Lumpkin et al., 2001], the integral is necessarily truncated at a finite lag. The most common approach is to integrate to the first zero crossing of R [e.g. Freeland et al., 1975; Krauss and Böning, 1987; Poulain and Niiler, 1989]. Integrating to the first zero crossing of R corresponds to the first maximum of the integral time scale and the resulting value is usually an upper bound to the actual value. However others integrate to a constant, prechosen lag (e.g., 20 days) [Speer et al., 1999]. In this case, the lag is chosen by visual inspection of the integral curves. [19] We adopt an approach similar to the latter above. Specifically, we estimate the diffusivities, and the corresponding time and length scales, by averaging the autocorrelation integral over the period from 6 to 10 days. In many cases, the integral levels off after several days, and using this period provides a reasonable average. Using a shorter time is problematic as the integral has not yet converged [e.g., Koszalka and LaCasce, 2010] and longer times are not desirable as the errors increase as the square root of time [Davis, 1991]. An example of the mean autocorrelation and its integral, from the Lofoten Basin region, is shown in Figure 3. Figure 3Open in figure viewerPowerPoint Time-lagged (a) velocity variance and (b) diffusivity versus time lag for the Lofoten Basin domain. [20] We calculated diffusivities and integral scales in five geographical domains. These are the Iceland-Faroe Front (IFF) region, the Iceland Plateau (IP), the Norwegian (NB) and Lofoten basins (LB), and the NwAC region (Figure 2). Using larger regions for the diffusivities improves their convergence, as the diffusivity is a more Lagrangian statistic, involving averages over particle paths. [21] Throughout the paper, errors were estimated using the independent number of observations, n*. The value of n* was computed as nΔt/2TL, where n is the total number of observations, Δt is the sampling interval and TL the Lagrangian timescale, here assumed to be 1 day. [22] To study how the statistics vary with season, we require that the observations are available for most of the year. The distribution of observations by month is close to uniform, as shown in Figure 4. We divide the data into two subsets, one representing summer (May to October) and the other winter (November to April). The data from the summer season constitutes 53% of the whole set and that from the winter season the remaining 47%. Figure 4Open in figure viewerPowerPoint Number of drifter days (a) per year and (b) per month. POLEWARD observations are marked in red. [23] There are nevertheless regional variations in the coverage, as expected due to spatially varying mean currents and eddy field and intermittent deployments; there are more observations in some regions in winter than in summer, and vice versa. To quantify this, we used the method of Lumpkin [2003], in which each observation is assigned a complex number, with unit amplitude and a phase determined by year day. The numbers are then averaged in the 2° × 1° bins. The results are shown in Figure 5. An amplitude of zero indicates homogeneous sampling through the seasons and an amplitude of one implies one season is sampled exclusively. The sampling is reasonably uniform, except in a few locations. It is large at the western periphery, where the sampling is sparse. But it is also elevated in some regions in the east (Figure 5a). Figure 5Open in figure viewerPowerPoint (a) Amplitude of the seasonal observational bias and (b) the ratio of summer to winter observations for bins where the amplitude exceeds 0.3. Blue areas show where winter data is over represented and red areas show where summer data is over represented, the more intense the color the greater the difference. [24] In Figure 5b, the ratio of summer to winter observations is plotted for regions where the amplitude exceeds 0.3. There are several regions along the Norwegian coast which have an excess of wintertime observations. Parts of the western Barents Sea on the other hand and the Lofoten Basin were sampled more in the summertime, due to summertime deployments. [25] As we have data spanning almost 20 years, we can also examine the change in statistics between the first and second decades. Specifically we will compare the mean velocities, the eddy kinetic energies and diffusivities for the two periods. Defining the first period to be from 1991 to 1998 and the second from 1999 to 2009 resulted in a nearly even distribution of observations. The first period corresponds to 56% of the whole set and the second the remaining 44%. Nevertheless, the regional sampling varies in the two periods, and this will be seen to be important. To reduce seasonal biasing, only bins with more than 25 observations from summer and 25 observations from winter were included. 3. Results Eulerian Mean Circulation 1991–2009 [26] The mean velocities for the entire sampled period are shown in Figure 6. Qualitatively the current structure is consistent with earlier studies: Both the eastern and the western branch of the NwAC are clearly seen, with strongest currents just west of the Lofoten and Vesterålen Islands. The strongest currents exceed 100 cm s−1, north of the Lofoten Basin and near the Lofoten Islands. The eastern branch follows the Norwegian shelf edge and its continuation towards the Fram Strait. There is clear exchange between the eastern branch and the Norwegian Coastal Current (NwCC), which lies near the coast. There is also a bifurcation north of Norway, with part of the flow entering the Barents Sea. The western branch follows the topographic slope of the Vøring Plateau toward Jan Mayen. Then it turns northeast, following Mohn's Ridge. West of Bjørnøya, it turns northward and continues along the Knipovich Ridge toward the Fram Strait. An anticyclonic recirculation, though barely resolved, is seen too in the western Lofoten Basin. Figure 6Open in figure viewerPowerPoint Mean velocity vectors computed from drogued observations in 2° longitude by 1° latitude bins. The results for bins with less than 50 six hourly observations and from less than two different drifters are not shown. [27] The eddy kinetic energy and principal axes of the variance are shown in Figures 7 and 8. The largest kinetic energies occur where the mean currents are strongest. There are both energetic regions (near the Iceland Faroe Front, the NwAC and in the Lofoten Basin) and quiescent regions (the Iceland Plateau and the Norwegian Basin). The largest eddy kinetic energies (>530 cm2 s−2) are found in the Lofoten Basin, northwest of the Lofoten and Vesterålen Islands. The minimum energies (<20 cm2 s−2) occur near the Iceland plateau. The axes suggest the variability is most anisotropic where the mean currents are strongest. Here the variability is also aligned with the topography. Offshore, the variability is more nearly isotropic. It has been suggested that there are jet-like structures extending westward from the strong currents [Poulain et al., 1996; Köhl, 2007; Rossby et al., 2009], but such structures are not apparent in the variance axes. Figure 7Open in figure viewerPowerPoint Distribution of the fluctuation or eddy kinetic energy calculated from drifter data in 2° longitude by 1° latitude bins. Figure 8Open in figure viewerPowerPoint Principal axes of variance and contours of eddy kinetic energy computed from drogued observation data in 2° longitude by 1° latitude bins. [28] Our variance contours are similar to those of Poulain et al. [1996, Figure 3], except that their values are slightly greater in the Lofoten region than ours. This may result from the fact that we only use drogued drifter data while Poulain et al. [1996] also include wind-corrected data, or simply that we have more data, with targeted deployments in the Lofoten Basin. Diffusivities 1991–2009 [29] We calculated diffusivities in five larger regions, denoted the Iceland Faroe front (IFF), the Iceland Plateau (IP), the Norwegian (NB) and Lofoten (LB) basins, and the NwAC region (Figure 2). The results are shown in Figure 9. Figure 9a shows the velocity variances in each of the regions, and Figure 9b shows the diffusivities, calculated using Equation 2. The corresponding values obtained by Poulain et al. [1996] are marked in green. The variances and diffusivities are also listed in Table A1 in Appendix A, along with the corresponding integral time and space scales. Figure 9Open in figure viewerPowerPoint (a) Velocity variance and (b) diffusivity as computed in the five domains in the Nordic Seas. The open circles and stars denote zonal and meridional directions, respectively. The error bars show the 95% confidence level based on a chi-square probability distribution with n* − 1 degrees of freedom for the variances and a Student's t distribution for the diffusivities. The green marks are the results from Poulain et al. [1996]. [30] Regarding the variances, we see that the zonal and meridional estimates are the same within the errors in all five regions. As could be anticipated from the previous results, the weakest variability is in the Iceland Plateau region (with variances of <40 cm2 s−2). The variances are over 6 times greater in the Lofoten Basin (near 260 cm2 s−2). The estimates in all five regions are consistent with those of Poulain et al. [1996]. [31] The diffusivities show similar variations, with the smallest values in the Iceland Plateau region and the largest in the Lofoten Basin. The diffusivities are near 1 × 107 cm2 s−1 in the former and 2.7–3.6 × 107 cm2 s−1 in the latter. In this case there are significant differences between our values and the values obtained by Poulain et al. [1996]. Theirs are generally larger than ours, particularly in the NwAC region. The result is in line with Koszalka et al. [2011] who found that the diffusivities were suppressed at the core of the NwAC, while they were elevated in the Lofoten Basin. [32] As discussed in section 2, the diffusivity is proportional to the variance for a stationary flow. As our variances are comparable to those of Poulain et al. [1996], the differences in diffusivity evidently stem from the Lagrangian time scales. Poulain et al. [1996] used the maximum values of the diffusivities in the 0–20 day range whereas we used the mean value between 6 and 10 days. To check this, we recalculated the diffusivities using data from the same period (1991–1995) and used the maximum value during the 0–20 day period. This yielded values comparable to those of Poulain et al. [1996], except in the NwAC region, where our diffusivities were still somewhat lower. We also calculate the residual velocities in a different manner than Poulain et al. [1996], but as our variances are comparable, this evidently does not affect the diffusivities. (We calculate the means in geographical bins and interpolate them onto individual trajectories to obtain the residuals. Poulain et al. [1996] calculated Lagrangian averages along the trajectories and subtracted those.) [33] The Lagrangian integral timescales vary from 1–1.5 days in most regions. The length scales are mostly between 10 and 20 km, with the smallest values occurring in the Iceland Plateau region. All values are listed in Table A1 in Appendix A. Seasonal and Decadal Variability [34] To examine seasonal variations, we divided the data into two sets, one for summer (May to October) and the other for winter (November to April). Then we calculated velocity statistics for each period and compared them. As noted, each season has approximately the same amount of data, and the regional coverage is also similar during the two periods. [35] The mean circulations for the two seasons are shown in Figure 10. The overall picture is very similar, indicating that there are no major changes in the structure of the currents. Both the eastern and the western branches of the NwAC are present, as is the NwCC. In the difference between the means, shown in Figure 11, the gray shaded bins represent means which differ at the 95% level, as determined by a vectorial t test [Garraffo et al., 2001]. (Using the t test assumes the data are normally distributed. We tested this by calculating the kurtosis of the residual velocities in each bin, and found that the values were indeed generally near three. A timescale of 2 days was used to determine the degrees of freedom in each bin.) While the means are different in a number of bins, the bins with significant differences are spread approximately uniformly throughout the domain. There is nevertheless evidence for winter intensification in several locations: near the inflow (0° and 63° N), at Svinøy, west of Lofoten Islands and in the continuation towards Spitsbergen. Furthermore, the anticyclonic recirculation in the Lofoten Basin is seen in summer but not in winter. Figure 10Open in figure viewerPowerPoint Mean velocity vectors computed from drogued observations in 2° longitude by 1° latitude bins for (a) summer and (b) winter seasons. Only bins that are present both seasons are shown. Figure 11Open in figure viewerPowerPoint Seasonal variation of the pseudo-Eulerian near-surface currents calculated as the difference between winter and summer. The shading indicates areas where the difference is statistically significant to the 95% level based on the vectorial t test described by Garraffo et al. [2001]. [36] Some of these differences stem from differences in sampling. Particularly in the Lofoten Basin, where there was a targeted deployment during a summer cruise in 2009. A number of these drifters were launched in or near the eddy. In contrast, the winter sampling comes from drifters deployed outside the basin which subsequently passed through. These drifters evidently were not entrained by the eddy, but simply skirted it. [37] The eddy kinetic energies for the summer and winter seasons are shown in Figure 12; Figure 13 shows the difference in EKE between winter and summer in bins where the variance is statistically different between the seasons at the 95% level. The results suggest the variability is greater in winter in particular regions, most noticeably in the Lofoten Basin. The difference moreover appear to be significant in many bins. Figure 12Open in figure viewerPowerPoint Distribution of the fluctuation or eddy kinetic energy calculated from drifter data for (a) summer and (b) winter seasons in 2° longitude by 1° latitude bins. Only bins that are present both seasons are shown. Figure 13Open in figure viewerPowerPoint Difference in EKE between winter and summer (winter minus summer), shown in bins where the difference in variance is statistically significant to the 95% level based on the F test. [38] Now consider the changes between the 1990s and the 2000s. The mean circulations for both periods are shown in Figure 14 and the difference field is shown in Figure 15. There are regional variations in the fields, but these are mostly insignificant or related to differences in sampling, for example west of Spitsbergen. The anticyclonic recirculation in the Lofo