A QuikSCAT climatology of ocean surface winds in the Nordic seas: Identification of features and comparison with the NCEP/NCAR reanalysis



[1] High-resolution satellite-derived QuikSCAT ocean surface wind data are used to provide an 8-year climatology of average as well as below- and above-average winds in the Nordic seas. A number of localized wind speed maxima are identified, with average wintertime wind speeds exceeding 14 m s−1 in the Denmark Strait. Five distinct wind speed nadirs are also discovered, of which the two most pronounced are located in the Greenland Sea. In the second part of the paper, two data sets derived from the National Centers for Environmental Protection/National Center for Atmospheric Research (NCEP/NCAR) reanalysis are found to concur well with QuikSCAT in terms of both low-frequency (monthly means) and high-frequency (daily means) variability. Correlation coefficients between QuikSCAT and reanalysis daily and monthly mean wind speed are around 0.9, but notable and systematic differences are also identified. Root mean square differences between QuikSCAT and reanalysis of daily mean wind speed averaged over four separate regions range from 1.11 to 1.81 m s−1, and from 0.75 to 1.00 m s−1 for monthly means. Furthermore, a correspondence between the ocean surface wind speed and five teleconnection indices, of which the North Atlantic Oscillation (NAO) and the Scandinavian pattern are the most important, is found.

1. Introduction

[2] In this paper, an 8-year wind climatology of the Nordic seas (the northeast Atlantic Ocean north of Iceland, see Figure 1 for a geographical reference) is compiled using high-resolution satellite-derived QuikSCAT wind data, which are available on a sufficiently high horizontal resolution to fully capture small-scale systems. Because of the limited period with QuikSCAT data, it is also investigated whether the coarser National Centers for Environmental Protection/National Center for Atmospheric Research (NCEP/NCAR) reanalysis (NNR hereafter) [Kalnay et al., 1996] adequately reproduces high-frequency (daily means) and low-frequency (monthly means) wind variability.

Figure 1.

The Nordic seas region. Topographic height and bathymetric depth are shown with colored contours. The four regions marked with lines are used to produce area-averaged wind speeds in this paper. The meaning of the abbreviations are as follows: IS, the Irminger Sea; GS, the Greenland Sea; NS, the Norwegian Sea; and BS, the Barents Sea.

[3] The high spatial resolution of the gridded QuikSCAT data (0.25 degrees) is particularly relevant in the Nordic seas region. Although the local wind field is largely governed by synoptic systems, this is a region where mesoscale weather phenomena and the topography have a significant impact on both the average and extreme surface winds, of which there exists no systematic study to date. Because there is human activity in the region, knowledge about the nature of the local winds are of intrinsic interest. For instance, Grønås and Skeie [1999] estimated that 342 people were killed in ship accidents in Norwegian waters in the last century alone.

[4] As discussed by Chelton et al. [2004], an inaccurate surface wind field can lead to errors in ocean models because of its influence on heat and momentum fluxes between the ocean and the atmosphere. Orvik and Skagseth [2003] found a high level of lag correlation between the zonally integrated NNR wind stress curl at 55°N and the volume transport in the Norwegian Sea at 62°N. Atmospheric forcing contributes to the transport of warm water masses into the Nordic seas [Furevik and Nilsen, 2005; Sorteberg et al., 2005], most prominently through the North Atlantic Oscillation (NAO) system [e.g., Rogers, 1997]. As surface wind stress is roughly proportional to the square of the wind speed, errors in the surface wind field can potentially induce large errors in ocean models, and in coupled models these errors will be returned to the atmospheric component. Surface heat fluxes are approximately proportional to the surface wind speed and are also prone to errors. However, it must be emphasized that it is not only the bulk fields that are used in numerical weather prediction products (like the reanalysis) that determine how well the energy and momentum fluxes are represented, but also the assumptions that go into the models' surface layer (see discussion by Renfrew et al. [2002]).

[5] This is not a study of synoptic activity in the North Atlantic, although the synoptic forcing on the surface wind field is strong. The large-scale storm track has been widely examined in the literature. In an analysis of the North Atlantic westerly storm track, [Hoskins and Hodges, 2002] tracked cyclones over a 22-year period and found primary genesis regions just east of the Rocky mountains and to the east of the North American continent. From the characteristics of the sea level pressure field, a secondary genesis region with high growth rates was identified over the Nordic seas. These findings are supported by other studies; distinct maxima of cyclone counts were found over the Norwegian and Barents Seas by both Zhang et al. [2004] and Wernli and Schwierz [2006]. One of the reasons for this is that this region during winter is characterized by exceptionally low static stability in the lower atmosphere, especially during “outbreaks” of cold air from the sea ice over the ocean [Kolstad, 2006].

[6] Besides providing a favorable environment for the reintensification of synoptic lows, low atmospheric stability and strong, shallow baroclinicity are conducive to the formation of polar lows (see Rasmussen and Turner [2003] for a comprehensive review). These mesoscale cyclones are formed as cold air masses are advected out over a much warmer ocean. Corresponding well with previous climatological work [Businger, 1985; Ese et al., 1988], two recent studies of favorable conditions for polar low formation both found a regional maximum centered over the Norwegian Sea [Kolstad, 2006; Bracegirdle and Gray, 2008], mainly due to the weak low-level static stability. The Nordic seas in general are known for a high frequency of polar lows.

[7] Although the NAO system is the dominant teleconnection mechanism for storminess in the northern North Atlantic as a whole, it has been shown that other patterns can also be important [e.g., Mailier et al., 2006]. In particular, Seierstad et al. [2007] found that the Scandinavian pattern (SCAN hereafter) had a strong influence on storminess in the Norwegian Sea, and that the east Atlantic pattern was important in the Labrador Sea region.

[8] There exist few climatological studies of small-scale cyclones in the Nordic seas. Harold et al. [1999a, 1999b] inspected satellite images over a 2-year period manually, and found that wintertime cyclones with a diameter of 200–400 km occurred most frequently in the Norwegian Sea and to the south of the Denmark Strait. The slightly larger cyclones (400–600 km) were primarily found over the Greenland Sea. In a study of mesoscale cyclones that were detectable in the ERA-40 data set in the same period, Condron et al. [2006] found centers of action west of Spitsbergen (at the southern entrance to the Fram Strait), over the northern Norwegian Sea, over the Greenland Sea near Jan Mayen, and just south of Iceland. The strongest cyclones were found over the Greenland Sea and to the west of Spitsbergen.

[9] Numerous orography-induced wind features have also been noted in the Nordic seas. Two wind maxima at either end of the Denmark Strait, both caused by orographic forcing from Greenland and Iceland, were identified by Moore and Renfrew [2005]. Near the east coast of Greenland, katabatic winds have been analyzed by Klein and Heinemann [2002]. Topographic jets have also been found in easterly flow over both the southern [Skeie and Grønås, 2000] and northern [Sandvik and Furevik, 2002] tips of Spitsbergen and along the west coast of Norway in southwesterly flow [Barstad and Grønås, 2005, 2006].

[10] Wakes and other wind speed nadirs are also important for the interaction with the ocean, but have not received much attention in the literature. Although wake regions have been discussed in a few case studies [e.g., Skeie and Grønås, 2000; Olafsson and Shapiro, 2002; Reeve, 2007], to our knowledge there exists no climatological study of such minima in the Nordic seas. Here we identify five distinct locations with low wind speed relative to their surroundings.

[11] The purpose of this study is twofold. First, the QuikSCAT data are used to investigate the average surface wind field in the Nordic seas. The upper and lower quartiles of wind speed are also examined. The second objective is motivated by the unfortunate fact that the QuikSCAT data are only available from 1999, currently spanning an insufficient interval for trend analysis. The NNR data set, on the other hand, provides decades of data. However, the high level of detail in the QuikSCAT data is not available in reanalysis data, which are a compromise between assimilation of observations and the underlying numerical model's requirements for dynamical consistency. They may be regarded as the synoptic background forcing field, and the difference between reanalyses and observations is due to mesoscale processes, topographic resolution or averaging issues associated with the differing grid cell sizes. Other studies have concluded that the NNR surface wind speed has a considerable bias in different regions [e.g., Smith et al., 2001; Goswami and Sengupta, 2003]. On the other hand, Renfrew et al. [2002] found a good agreement between in situ measurements and reanalysis wind speed. The second aim of this paper is therefore to evaluate to which degree NNR is able to adequately represent surface winds in the Nordic seas.

2. Data and Method

[12] The QuikSCAT surface wind field is derived from the microwave scatterometer SeaWinds, which was launched on the QuikBird satellite in June 1999. In short, a scatterometer is an active radar which measures the sea surface roughness through measurements of the radar backscatter cross section. From this the 10-m wind speed and direction near the sea surface can be inferred, using a “model function,” in which it is assumed that the atmospheric boundary layer is neutrally stratified. A thorough introduction to the technical aspects of QuikSCAT data retrieval is given by Chelton and Freilich [2005] and Hoffman and Leidner [2005].

[13] The QuikSCAT data used here are supplied by Remote Sensing Systems (REMSS; http://www.remss.com/) and sponsored by the NASA Ocean Vector Winds Science Team. The gridded data set is available on a twice-daily basis (morning and evening) over the entire globe, starting in July 1999. The resolution is 0.25 by 0.25 degrees, which is equivalent to 28 km in the meridional direction and 7 km in the zonal direction at latitude 75N. Data from August 1999 through May 2007 were used.

[14] The QuikSCAT data were calculated by REMSS in October 2006, using version 3a of the retrieval algorithm, in which version 6 SSMI (Special Sensor Microwave/Imager) rain rates and sea ice distribution, calculated in September 2006, were used as input. The data used in this study have all been marked with binary (0/1) rain flags.

[15] The NNR data are available throughout the QuikSCAT period and were compared to the QuikSCAT data. The NNR 10-m surface wind field was used. It is available on a grid with a resolution of 1.9 by 1.9 degrees.

[16] The recording time of the QuikSCAT winds at each grid point varies on a daily basis. This makes it difficult to compare the QuikSCAT data directly to NNR data. However, the QuikSCAT recording time of the morning samples is usually between 0400 and 0800 UTC in the northeast Atlantic, while the evening samples are generally recorded between 1900 UTC and midnight. In order to obtain two approximately comparable wind fields, we calculated daily mean NNR winds from the 0600 and 1800 UTC samples, as well as daily mean QuikSCAT winds from the morning and evening samples. On the rare (22 of 2861) days when one or both QuikSCAT observations were missing the daily mean was excluded from the analysis.

[17] Another factor that makes comparison between NNR and QuikSCAT challenging is that a large number of grid points are undefined in the QuikSCAT data set, either because of potential rain contamination or erroneous sea ice cover (see discussion below). We addressed this by interpolating the nearest-neighbor NNR grid point onto the QuikSCAT grid. The invalid QuikSCAT grid points were ignored in the resulting NNR field, yielding two data sets with identical spatial and temporal resolution. However, this approach is only viable when QuikSCAT observations are available (from 1999). It is desirable to assess whether it is possible to use NNR independently of QuikSCAT. We therefore compiled a new data set from the reanalysis wind field, simply by ignoring all grid points that are either on land or covered by sea ice. For consistency, the same time samples were used (0600 and 1800 UTC). This data set is independent of QuikSCAT and is referred to as NNRI (NNR “Independent”).

[18] One of the objectives of this paper is to assess the ability of NNR and NNRI to represent high- and low-frequency variability. We therefore calculated daily and monthly mean time series, averaged and area-weighted over all the grid points inside the four nonoverlapping regions shown in Figure 1. A number of diagnostic parameters, such as the means, standard deviations and root mean square (RMS) differences, were calculated for the different time series using standard methods. The correlation coefficients (hereafter referred to as CCs) were calculated for wind speed anomalies from climatology. Nonparametric Spearman rank CCs were used because they do not require any assumptions about the frequency distribution of the variables or that they have a linear relationship. When calculating the statistical significance of the CCs, the reduction of the number of degrees of freedom due to autocorrelation was accounted for as described by von Storch and Navarra [1999]. For monthly means, the long-term mean (over 8 years) was removed. The long-term daily mean was calculated as follows: first, the long-term mean for each date was calculated in a vector with 365 elements. Then a 31-day running mean filter was applied to this vector. This procedure corresponds roughly to interpolating long-term monthly means onto individual dates.

[19] As a measure of the persistence of a given wind direction, we use the scalar “directional constancy” as used in several other studies [e.g., Moore, 2003]. This quantity is the absolute value of the average wind vector divided by the average wind speed.

[20] The monthly time series of teleconnection indices used here were downloaded from the Climate Projection Center at NOAA (http://www.cpc.ncep.noaa.gov/). They were calculated using the Rotated Principal Component Analysis (RPCA) used by Barnston and Livezey [1987]. The RPCA technique was applied to monthly standardized 500-mbar height anomalies from the NCAP/NCAP reanalysis data set in the analysis region 20°N–90°N between January 1950 and December 2000.

3. On the Validity of QuikSCAT Data

[21] Before presenting the wind climatology, it is appropriate to discuss possible errors in the QuikSCAT wind data. It has been shown that the presence of rain may contaminate the wind retrieval process [Weissman et al., 2002; Chelton et al., 2006]. The data are marked with rain flags, and the integration of SSMI rain rates in version 3a of the retrieval algorithm probably leads to an improvement on earlier versions in this respect.

[22] There is also a concern that pixels in the marginal ice zone may be incorrectly interpreted as open water by the QuikSCAT algorithm. Because of the significantly different roughness of sea ice and open water, undetected sea ice has the potential to introduce errors in the derived surface wind field. This is addressed by marking possible sea ice locations with a rain flag. The following sentence is taken from the REMSS web site: “Using the SSMI daily observations of sea ice, the scatterometer observations can be properly flagged so that reliable wind vectors can be obtained immediately next to the marginal ice zone.”

[23] Pickett et al. [2003] found that when excluding rain-flagged data and observations with low wind speed (<3 m s−1), QuikSCAT observations agreed more closely with station and buoy observations. In a study of the effect of the vorticity of surface winds on tropical cyclogenesis, while acknowledging that it might introduce noise, Wang et al. [2007] chose not to exclude rain-flagged data. Sea ice was not a concern in those studies. In a climatological study such as this, where it is not feasible to assess every single case manually, and where sea ice is an issue, the balance of evidence led us to exclude all rain-flagged data. The overall results remained qualitatively unchanged by this decision.

[24] It has also been suggested that errors in both wind speed and wind direction may vary with the wind speed. However, according to Yuan [2004] the “Ku-2001” model function, which is used in version 3 (and 3a) of the QuikSCAT retrieval algorithm, is an improvement of the previous “Ku-2000” model function in that it provides higher accuracy when the wind speed exceeds 20 m s−1. Other studies [Hoffman and Leidner, 2005; Leslie and Buckley, 2006; Satheesan et al., 2007], in which recent versions of the model function were used, indicate that QuikSCAT errors actually decrease as the wind speed increases, although a slight positive bias (<1 m s−1) is generally noted. To our knowledge, no validation studies of version 3a QuikSCAT data has been published yet.

[25] We are not aware of any QuikSCAT validation studies at high latitudes. The near-surface atmospheric stability in the northeast Atlantic is highly variable. During cold-air outbreaks the stability is sometimes significantly higher than the one found at lower latitudes, but when the heat fluxes from a warm ocean surface are high, the stability can be very low. However, Ebuchi et al. [2002], in a comparison study between QuikSCAT winds (using the Ku-2000 model function) and ocean buoy data, found that the dependency of the derived wind field on the sea surface temperature and the stability was not physically significant.

4. Results

4.1. QuikSCAT Wind Climatology

[26] In Figure 2 (left), the average QuikSCAT wind speed for each season from winter (December–February; DJF) to autumn (September–November; SON) is shown. As wind data are only available over open water, the wind field in locations which are frequently covered by sea ice will not have the same temporal resolution as open water locations. Here, only grid points that are ice-free at least 10% of the time are shown. Pixels that are covered by sea ice 50% of the time are marked with a thick line.

Figure 2.

(left) Average QuikSCAT wind speed in m s−1 for the four seasons of the year from August 1999 through May 2007. (right) As Figure 2 (left) but average wind vectors with arrows (the longest arrow is marked with a circle and translates to 11 m s−1) and directional constancy with filled contours. The thick, unmarked contours indicate pixels that are covered by sea ice 50% of the time.

[27] The wind field follows a distinct seasonal cycle. During winter, at the height of the synoptic activity, the wind speed to the south of Iceland exceeds 12 m s−1. Values above 10 m s−1 are sustained from autumn to spring. Further north, the wind speed decreases over the open ocean, with the exception of a tongue of strong winds stretching north from the southwestern coast of Norway. The barrier flow off the Norwegian coastline has been studied by Barstad and Grønås [2005, 2006].

[28] Near the edge of the sea ice in the Greenland Sea and near the southern tip of Spitsbergen the average wind speed again surpasses 12 m s−1 during winter, but values over 10 m s−1 are only found near the western tip of Spitsbergen in the other seasons.

[29] The highest average wintertime wind speed is found in the Denmark Strait, where it is greater than 14 m s−1 in some locations. Values well over 10 m s−1 are found in autumn and spring, reaching almost 12 m s−1 in the vicinity of the east Greenland topography. This is the imprint of the barrier flow described by Moore and Renfrew [2005].

[30] The average wind speed in the Barents Sea is lower than in the other regions, never reaching 12 m s−1. The strongest winds in that region during winter are found along the east coast of Novaya Zemlya.

[31] There are five distinct wind speed nadirs, all of which are most pronounced during winter and spring. Between Iceland to the southwest and Jan Mayen to the northeast, as well as between Jan Mayen and Spitsbergen, the average wind speed is below 10 m s−1 during winter and around 8 m s−1 in the spring. Two further minima are found between Hopen and Bjørnøya (Bear Island) to the southeast of Spitsbergen and in the eastern part of the Barents Sea, respectively. To the southwest of Iceland, a final nadir is found, with wind speeds below 12 m s−1 during winter and below 10 m s−1 in the spring. Moore and Renfrew [2005] reasonably attributed this minimum to wake effects downstream of the Greenland topography in westerly flow.

[32] The average wind vectors and the directional constancy are shown in Figure 2 (right). Along the east coast of Greenland, the directional constancy is high, revealing a predominance of northerly winds. This is especially evident through the Denmark Strait, where the directional constancy is high even in the summer. Other dominant wind patterns include southwesterlies to the south of Iceland during winter, northeasterly winds to the south of Svalbard from autumn to spring, southwesterlies along the southwest coast of Norway in autumn and winter, and southerly winds in the eastern Barents Sea during winter.

[33] The average wind speed in the upper quartile (the samples for which the wind speed exceeds the 75th percentile of each grid point) is shown in Figure 3 (left). (Note that pth percentile of a time series with N samples is found by sorting the data in descending order and then selecting the value with rank n = pequation image + equation image rounded to the nearest integer.) Values exceeding 26 m s−1 during winter and 22 m s−1 in the autumn are found in two locations in the Denmark Strait. These wind speeds are substantially higher than in the region to the south of Iceland, where values around 20 m s−1 and below 18 m s−1 are found in winter and autumn, respectively. Further north near the sea ice edge, values in excess of 18 m s−1 are found in the winter. Here the spring and autumn winds are not especially high compared to their surroundings. An exception is the location to the northwest of Jan Mayen (at approximately 73°N), where the wintertime value is higher than 20 m s−1.

Figure 3.

As Figure 2 but only for the winds in the upper quartile of wind speed for each grid point. The longest arrow in Figure 3 (right) is marked with a circle and translates to 22 m s−1.

[34] High wind speeds are also found near the southern tip of Spitsbergen (>18 m s−1 during winter and spring), in a tongue stretching northward from the southwest coast of Norway (between 18 and 20 m s−1 during winter), near the southeast coast of Iceland (>20 m s−1 during winter and >18 m s−1 in the autumn), and along the west coast of Novaya Zemlya (>16 m s−1 during winter and >14 m s−1 in the autumn). The wind speed nadirs discussed above are also distinguishable in Figure 3.

[35] Figure 3 (right) shows the average wind vectors and the directional constancy. The predominance of northerly winds during strong wind episodes along the east coast of Greenland is striking, with directional constancy values approaching 1. This shows that the strongest winds in this region almost only take place in a particular wind regime. It is also clear that strong winds usually take place in southwesterly flow along the southwest coast of Norway (during winter and autumn), in northeasterly flow near the southern tip of Spitsbergen (from autumn to spring), and in southerly or southeasterly flow along the west coast of Novaya Zemlya (only during winter). It is interesting that the strong winds near the southeast coast of Iceland are not associated with high levels of directional constancy.

[36] In Figure 4 the daily mean wind speed for the regions defined in Figure 1 are shown (a 31-day running mean filter was applied for readability prior to plotting). A seasonal cycle is evident in all the regions, with the strongest winds normally occurring in the winter. The maximum values of daily mean (nonsmoothed) wind speed for the Irminger Sea (IS), the Barents Sea (BS), the Norwegian Sea (NS) and the Greenland Sea (GS) are 23.6, 22.5, 21.7 and 21.6 m s−1, respectively. The corresponding dates are 10 December 2006, 8 January 2000, 24 February 2004, and 23 December 2004. The time series are too short for a meaningful discussion of trends.

Figure 4.

The 31-day running mean area-averaged wind speed for each of the regions defined in Figure 1. Values are shown every day from 1 August 1999 through 31 May 2007, and every 1 January is marked with a vertical, dotted line and marked with the corresponding year.

[37] Long-term monthly mean wind speeds are shown for each region in Figure 7 in section 4.2.3. The Irminger Sea is the windiest region, with QuikSCAT winds in excess of 12 m s−1 in January and February. By contrast, the average wind speed in the Barents Sea is greater than 10 m s−1 only in January.

4.2. Comparison With NCEP/NCAR Reanalysis

4.2.1. Spatial Distribution

[38] The average difference between the NNR and QuikSCAT wind speed is shown in Figure 5. Note that the lowest value shown is −4 m s−1, although lower values exist.

Figure 5.

The average difference between NNR and QuikSCAT wind speed in m s−1 for the four seasons of the year from August 1999 through May 2007. The thick, unmarked contours indicate pixels that are covered by sea ice 50% of the time. Values below −4 m s−1 are truncated.

[39] In general, the NNR winds are lower than QuikSCAT over open water. The differences are especially pronounced near coastlines and in the vicinity of the sea ice edge. In the Denmark Strait, the negative wind speed difference amounts to 4 m s−1 or more during winter. Differences of more than 2 m s−1 are found to the west of Spitsbergen from autumn to spring, near the southern tip of Spitsbergen during winter and spring, near the Norwegian coastline during winter, along the east Greenland sea ice edge during winter, and along the southeast coast of Greenland during winter and spring.

[40] The locations where the NNR wind speed is higher than QuikSCAT are also of interest. The largest positive differences are found at the two QuikSCAT Greenland Sea wind speed nadirs that were identified in Figure 2. There is no indication of these wind speed minima in NNR. The QuikSCAT wind speed nadirs near Spitsbergen and in the eastern Barents Sea are also absent in NNR. In other words, four of the five regions with local minima in the QuikSCAT wind speed field noted earlier have no counterpart in NNR.

4.2.2. High-Frequency Wind Speed

[41] In Table 1 we present a number of statistical parameters for time series of daily mean wind speed averaged over the four regions defined in Figure 1. The numbers are shown both for the entire time series and for the season with the largest variance, December to February (DJF). Note again that NNR has the exact same spatial and temporal resolution as QuikSCAT, whereas NNRI (NNR “Independent”) is calculated independently of QuikSCAT by removing all grid points which are either on land or covered by sea ice. The rank correlation coefficients are calculated for daily wind speed anomalies with respect to the 8-year long-term mean for each date.

Table 1. A Number of Statistical Measures Illustrating the Differences Between the Three Time Series of Daily Mean Wind Speeda
StatisticData SetsGreenland SeaBarents SeaNorwegian SeaIrminger Sea
  • a

    Unit is m s−1, but the numbers in italics are given in percent of the corresponding QuikSCAT values. The total (DJF) sample size is 2861 (720). All the Spearman rank correlation coefficients are significant at the 0.05 level. Further details are given in the text.

95th percentileQSCAT14.
95th percentileNNR90.990.792.391.394.291.388.588.4
95th percentileNNRI88.388.588.787.591.289.586.586.0
Standard deviationQSCAT3.022.512.752.592.952.763.353.15
Standard deviationNNR91.390.292.292.593.996.586.983.1
Standard deviationNNRI88.586.986.786.891.193.486.282.1
Root mean square differenceQSCAT-NNR1.111.371.181.421.251.661.351.81
Root mean square differenceQSCAT-NNRI1.241.511.341.651.371.831.662.22
Root mean square differenceNNR-NNRI0.320.340.400.460.310.450.600.69
Correlation coefficientQSCAT, NNR0.910.890.890.900.900.890.930.92
Correlation coefficientQSCAT, NNRI0.910.890.870.890.890.880.900.89
Correlation coefficientNNR, NNRI0.990.990.990.990.990.990.970.97

[42] The highest wind speed is found in the Irminger Sea, with an annual and DJF QuikSCAT average wind speed of 9.9 and 12.2 m s−1, respectively. The corresponding 95th percentiles are 15.9 and 17.8 m s−1. The Norwegian Sea is the second windiest region, followed by the Greenland and Barents Seas. NNR reproduces roughly 90% of the magnitude of the mean and the 95th percentile, although the percentages are slightly lower in IS. NNRI consistently lies 2–4 percentage points lower than NNR.

[43] The largest variance (expressed by the standard deviation in Table 1) is also found in IS, followed by GS, NS and BS (NS, BS and GS during winter). The relative standard deviation with respect to QuikSCAT is shown for NNR and NNRI. The numbers for NNR range from 83.1% in the IS winter to 96.5% in the NS winter. Again, the QuikSCAT numbers are uniformly higher than the ones for NNR (and NNR always higher than NNRI).

[44] The root mean square (RMS) differences are also shown in Table 1. The highest numbers for the difference between QuikSCAT and NNRI are found in the Irminger Sea (1.81 m s−1 during winter and 1.35 m s−1 overall). The lowest values occur in the Greenland Sea (1.37 and 1.11 m s−1, respectively). The RMS difference between QuikSCAT and NNRI is consistently higher. Furthermore, the RMS difference between the two NNR-derived time series is nonnegligible, ranging from 0.31 m s−1 in NS for the whole year to 0.70 m s−1 in IS during winter.

[45] The CCs between the three time series of daily wind speed anomalies (from the long-term daily mean) are shown in the bottom section of Table 1. Overall, NNR must be said to capture the daily variability quite well, with coefficients ranging from 0.89 in the NS winter to 0.93 in the Irminger Sea. The QuikSCAT/NNRI values are slightly lower, but never below 0.89. NNR/NNRI values range from 0.97 to 0.99.

[46] To further illustrate the differences between the high-frequency wind speed in the data sets, scatterplots of the day of the year versus the difference between QuikSCAT and NNR daily mean wind speed for each region are shown in Figure 6. The largest individual difference is found in the Barents Sea on 12 January 2003, when the daily mean area-averaged QuikSCAT wind speed was 15.0 m s−1, while the corresponding NNR wind speed was 8.4 m s−1. Generally, in keeping with the seasonal cycle and the RMS differences in Table 1, the largest differences are found during winter.

Figure 6.

Scatterplots of the day of the year versus the difference between daily mean QuikSCAT and NNR area-averaged wind speed for each of the regions defined in Figure 1. The period shown includes every day from 1 August 1999 through 31 May 2007.

4.2.3. Low-Frequency Wind Speed

[47] Table 2 is similar to Table 1, only for monthly mean wind speed in each region. Note that it would have been redundant to list the overall mean (it is listed in Table 1), and that the sampling size is too small for a meaningful 95th percentile. The CCs were calculated for anomalies from the long-term monthly mean.

Table 2. A Number of Statistical Measures Illustrating the Differences Between the Three Time Series of Monthly Mean Wind Speed From August 1999 Through May 2007a
StatisticData SetsGreenland SeaBarents SeaNorwegian SeaIrminger Sea
  • a

    Unit is m s−1, but the numbers in italics are given in percent of the corresponding QuikSCAT values. The sample size is 94. All the Spearman rank correlation coefficients are significant at the 0.05 level. Further details are given in the text.

Standard deviationQSCAT2.121.812.022.31
Standard deviationNNR89.387.886.682.1
Standard deviationNNRI88.184.083.981.2
Root mean square differenceQSCAT-NNR0.750.750.811.00
Root mean square differenceQSCAT-NNRI0.910.950.961.31
Root mean square differenceNNR-NNRI0.
Correlation coefficientQSCAT, NNR0.870.900.930.96
Correlation coefficientQSCAT, NNRI0.870.890.910.92
Correlation coefficientNNR, NNRI0.990.991.000.97

[48] We note again that the NNR time series is closer than NNRI to QuikSCAT in all the regions by all the measures listed in Table 2. Moreover, the relative low-frequency standard deviation is lower than the relative high-frequency standard deviation listed in Table 1 in all the regions for both NNR and NNRI. The RMS difference between NNR and QuikSCAT is never higher than 1 m s−1, but reaches 1.31 m s−1 in IS for NNRI. The CCs are all high; the lowest value is found in the Greenland Sea (0.87 for both NNR and NNRI with respect to QuikSCAT), while the highest QuikSCAT/NNR correlation coefficient is found in the Irminger Sea (0.96).

[49] In Figure 7, the long-term monthly average wind speed is shown for QuikSCAT, NNR and NNRI. While the average January and February wind speed in the Irminger Sea is nearly 13 m s−1, the corresponding values for NNR are just over 11 m s−1 and even less for NNRI. Similar but smaller differences are found during winter in the other regions as well. The NNR wind speed is lower than QuikSCAT throughout the year in all the regions, and NNRI has uniformly lower values than NNR. On the positive side, the QuikSCAT seasonal cycle (the ranking of each month in the year) is reproduced by both NNR and NNRI with a few minor exceptions.

Figure 7.

Long-term monthly mean area-averaged wind speed for the four regions shown in Figure 1. The data sets are (from left to right) QuikSCAT (blue), NNR (green), and NNRI (red). The averaging period is from 1 August 1999 through 31 May 2007.

5. Summary and Discussion

[50] High-resolution satellite-derived QuikSCAT wind data were used in this study to provide a climatology of both average and above-average winds in the Nordic seas. A number of localized wind speed maxima, with average wind speeds exceeding 12 m s−1 and even 14 m s−1 in isolated locations, were identified, such as the persistent strong northerly winds in the Greenland Sea, barrier winds in the Denmark Strait, a southwesterly jet along the coast of western Norway, as well as a northeasterly jet near the southern tip of Spitsbergen. Average upper quartile wind speeds of more than 26 m s−1 were found during winter in the Denmark Strait. In addition to the wind speed maxima, five wind speed nadirs were found. The two most pronounced minima were located in the Greenland Sea.

[51] When averaging the wind speed over the four regions in Figure 1, a maximum daily wind speed of 23.6 m s−1 was found in the Irminger Sea on 10 December 2006. The maximum long-term monthly mean wind speed was found in the Irminger Sea in January and February (12.5 m s−1).

[52] The QuikSCAT wind speed climatology was compared to two data sets which were derived from the NCEP/NCAR reanalysis (NNR). In the first data set, the NNR winds were interpolated onto the valid QuikSCAT grid at each time step. The second data set (NNRI) was compiled independently of QuikSCAT, by ignoring all grid points that were either on land or covered by sea ice. Broadly speaking, both reanalysis-derived data sets were found to concur well with QuikSCAT in terms of both low-frequency (monthly means) and high-frequency (daily means) variability. Correlation coefficients between QuikSCAT and NNR daily and monthly mean wind speed were around 0.9, and similar values were found between QuikSCAT and NNRI. However, notable and systematic differences were identified. The largest difference between the regional daily mean wind speeds of QuikSCAT and NNR was almost 7 m s−1, and root mean square (RMS) differences of daily mean wind speed ranged from 1.11 to 1.81 m s−1. RMS differences for monthly mean wind speed varied between 0.75 to 1.00 m s−1.

[53] The wind speed near the east coast of Greenland is remarkably high during winter. Moore and Renfrew [2005] studied the QuikSCAT wind field in the Denmark Strait, and found two regions with exceptionally strong winds. The high wind speed was attributed to barrier winds. In support of that claim, it is shown here that the strong winds are also present in the autumn in the vicinity of the east Greenland topography. Furthermore, it is shown that the winds along the ice edge upstream of the topography are also strong, and particularly during winter. We propose that these high wind speeds are a winter phenomenon connected to strong baroclinicity near the sea ice edge. There is a high level of directional constancy along the ice edge. Such a configuration, with warm water and cold stable air over the ice, may lead to an “ice breeze” and a strengthening of the winds along the sea ice edge [Langland et al., 1989; Brümmer, 1996; Økland, 1998], in extreme cases leading to hurricane-force surface winds [Grønås and Skeie, 1999]. It is not unlikely that these processes accelerate the winds upstream of the topography that creates the barrier winds identified by Moore and Renfrew [2005], thereby acting as a preconditioner or preamplifier for the strong orography-induced winds. An interesting prospect for numerical simulation would be to artificially modify the sea ice distribution upstream of 70°N to quantify the effect it has on the barrier winds. We also put forward that the strong winds through the Denmark Strait are likely to create a streamer of shear vorticity which may perhaps be transported southward and influence the properties of the Icelandic Low, another prospect for future studies.

[54] One of the most interesting findings in this paper is the identification of the wind speed nadirs, especially as they are not discernible in the reanalysis wind field. The average wind speed, wind vectors and the directional constancy of the wind field in the lower wind speed quartile are shown in Figure 8.

Figure 8.

As Figures 2 and 3 but only for the winds in the lower quartile of wind speed for each grid point. Note that the color scale in Figure 8 (right) is from 0 to 0.5, not to 1.0 as in Figures 2 and 3. The longest arrow in Figure 8 (right) is marked with a circle and translates to 2.7 m s−1.

[55] As in Figures 2 and 3, the region between Iceland and Spitsbergen is characterized by lower wind speed than its surroundings, but the directional constancy in Figure 8 is lower than for the winds in the upper quartile in Figure 3. Even so, it seems that the lowest wind speeds to the immediate northeast of Iceland take place in southerly or southwesterly flow during winter, while the upper quartile winds occurred in northerly or northwesterly flow. This suggests that Iceland acts as a topographical barrier, and that the nadir to the northeast is located in a wake region downstream of Iceland in westerly or southerly flow. Indeed, Hoskins and Hodges [2002] found that the feature density, the track density and the mean intensity of cyclones all had substantially higher values to the south of Iceland than further north. As far as the author is aware, the only previous investigation of wakes to the east of Iceland was performed by Olafsson and Shapiro [2002], who investigated “the Great Saltstorm” of 9–10 November 2001, during which the strong westerly wind field was split into two jets to the northeast and southeast of Iceland, with a wake region between them. That study supports the conjecture that Iceland can modify the southerly or westerly flow passing over it.

[56] Another possible partial explanation for the low wind speed at this location and to the northeast of Iceland in general is the influence of the Greenland topography. Kristjansson and McInnes [1999] found that lows traveling through the Denmark Strait or over Iceland from the southwest can be strongly modified by the topography of Greenland. In westerly flow, a secondary, quasi-stationary lee low tends to form to the south of the Denmark Strait, absorbing some of the energy and momentum from the original low. The result can be either that a weakened low advances into the Iceland and Greenland Seas, or that the northward progress is stopped altogether, as shown in a case study by Petersen et al. [2003].

[57] In spring, there is a band of weak lower-quartile winds in southeasterly flow from Iceland to the coast of Norway. It is possible that these winds occur in high-pressure situations over the Norwegian Sea. In the autumn, the wind direction is also from the southeast, although there is no clear wind speed nadir in this season. A composite analysis of low-wind-speed events might shed more light on the nadir to the north of Iceland.

[58] The low wind speeds in the region between Jan Mayen and Spitsbergen are surprising given the results presented by Harold et al. [1999a, 1999b] and Condron et al. [2006]. Both of these studies found high cyclone counts over the Greenland Sea [Condron et al., 2006, Figure 6]. We do not presently put forward an explanation for this wind speed minimum, apart from noting that long wakes in the lee of Spitsbergen in northeasterly flow seem to be common [Skeie and Grønås, 2000; Reeve, 2007]. Such wakes may be partly responsible for the low wind speed in this region, especially since the average flow direction is from the north-northeast. In spring, the average wind direction in the lower quartile is from the northeast (Figure 8), although this is not noticeable during winter and autumn.

[59] A possible explanation for the low wind speed in the eastern Barents Sea is that synoptic lows are retarded as they move into the region from the southwest. Hoskins and Hodges [2002] found a local lysis maximum over the Norwegian Sea (their Figure 6d). As the average wintertime flow is from the southwest in the eastern Barents Sea, it is also likely that wake effects downstream of northern Scandinavia and the Kola peninsula contribute to the low wind speed found there.

[60] An additional point of interest is that the NNR winds are stronger than QuikSCAT over the shallow water of the Svalbard Bank (to the southeast of Spitsbergen, between Bjørnøya (Bear Island) and Hopen). The water in this region is very shallow, and rapid cooling of surface water and formation of sea ice is known to occur [Nghiem et al., 2005]. It is possible that the low wind speeds are an artifact of undetected sea ice in QuikSCAT, or that cooling and freezing has an influence on the wind field, although the physical explanation for the latter mechanism is not clear.

[61] In addition to the wind speed extrema that have been identified, it is interesting to note that there seems to be no obvious features in the spatial distribution of the surface wind speeds in Figures 2 and 3 that can be attributed directly to mesoscale activity such as polar lows. Numerous climatological studies have shown a high frequency of polar low occurrence in the Norwegian Sea near 70°N [Businger, 1985; Ese et al., 1988; Kolstad, 2006; Bracegirdle and Gray, 2008], and there is a tongue of strong winds stretching from the southwest coast of Norway up to around 70°N (Figures 2 and 3), but this may be associated with strong winds in the cold air on the northwestern flanks of synoptic lows. These flanks are also known to provide favorable conditions for an important subclass (the reverse-shear type) of polar lows [Bond and Shapiro, 1991; Kolstad, 2006]. A composite study of high wind speed events near 70°N might yield more information about the climatological importance of polar lows.

[62] One of the purposes of this paper is to assess the NCEP/NCAR reanalysis data's ability to reproduce the variability found in the QuikSCAT data set. If this capability were found to be adequate, one could confidently use the reanalysis surface winds in analyses which demand longer time series than the currently available 8 years of QuikSCAT data. The statistics presented in Tables 1 and 2 are promising, but it must be stressed that this study does not claim to provide the final answer to whether the NCEP/NCAR reanalysis is capable of realistically reproducing the actual wind speed variability. This can only be done when a longer QuikSCAT period is available (or if other data sets are used as reference, as it has been shown that QuikSCAT may also be biased). Nevertheless, we include a brief comparison between NNRI and the teleconnection indices that were found to have an important influence on the northeast Atlantic cyclone climatology [Mailier et al., 2006; Seierstad et al., 2007].

[63] In Table 3 we present correlation coefficients between QuikSCAT, NNR and NNRI with respect to the monthly time series of these teleconnection indices: NAO, SCAN, east Atlantic (EA), east Atlantic/western Russia (EAWR) and Polar Eurasian (POL). The study period is from August 1999 through May 2007. Note that we only show indices that are significantly correlated with all three wind data sets. In the Barents Sea region no such persistent links are found in this period. In the Greenland and Norwegian Seas the low-frequency wind speed is associated with both the NAO and SCAN indices. The SCAN index CCs are negative because the positive phase of the SCAN index is usually linked to high-pressure systems over Scandinavia, while the negative phase can be associated with an anomalous low stretching into the Nordic seas [Rogers, 1990]. In the Norwegian Sea, the east Atlantic pattern is also influential. In the Irminger Sea, which is collocated with the center of synoptic activity in positive NAO phases, the correlation with the NAO index is high. All in all, these results suggest that the correspondence between the low-frequency wind speed variability of QuikSCAT and teleconnection indices is roughly equivalent to the one in NNRI even for the short time period used here (8 years).

Table 3. Spearman Rank Correlation Coefficients Between the Monthly Teleconnection Indices and Wind Speed Anomalies Averaged Over Each Regiona
Data SetGreenland SeaNorwegian SeaIrminger Sea: NAO
  • a

    The sample size is 94. Only teleconnection patterns that are significantly correlated at the 0.05 level with all three data sets are shown. Further details are given in the text.


[64] The seasonal CCs between NNRI and the five teleconnection indices mentioned above from June 1958 through May 2007 are presented in Table 4. The sample size is 49. The NAO index is very important for the wintertime wind speed in the Greenland, Norwegian and Irminger Seas, with CCs ranging from 0.57 in NS to 0.69 in GS. In the Irminger Sea the NAO is also highly influential in the spring, but it is surprising that there is no significant correlation during autumn. In the Norwegian Sea the impact of the NAO is evident throughout the year. We also note that there is no significant correlation between the NAO and the wind speed in the Greenland Sea during autumn and spring.

Table 4. Spearman Rank Correlation Coefficients Between the Seasonal Teleconnection Indices and Wind Speed Anomalies From Autumn to Spring Averaged Over Each Regiona
  • a

    The sample size is 49. Coefficients that are not significant at the 0.05 level are not shown. Further details are given in the text.

Greenland SeaNAODJF0.69
Greenland SeaSCANSON−0.32
Greenland SeaSCANDJF−0.41
Greenland SeaSCANMAM−0.30
Greenland SeaEAWRSON−0.36
Greenland SeaEAWRDJF0.36
Barents SeaSCANMAM−0.35
Barents SeaPOLSON0.32
Norwegian SeaNAOSON0.29
Norwegian SeaNAODJF0.57
Norwegian SeaNAOMAM0.29
Norwegian SeaSCANSON−0.47
Norwegian SeaSCANDJF−0.41
Norwegian SeaSCANMAM−0.40
Norwegian SeaEAWRDJF0.48
Norwegian SeaPOLSON0.29
Irminger SeaNAODJF0.66
Irminger SeaNAOMAM0.62
Irminger SeaEASON0.40
Irminger SeaEAWRSON−0.36

[65] The Scandinavian pattern is important from autumn to spring in the Norwegian and Greenland Seas, as well as in the spring in the Barents Sea. As before, the CCs are negative because the negative phase of this pattern is associated with low pressure over Scandinavia and the Nordic seas.

[66] The other indices seem to have a less pronounced influence on the wind speed. The east Atlantic/western Russia pattern is characterized by low pressure outside Newfoundland and over western Russia and high pressure centered over the North Sea in its positive phase. It is interesting that this pattern seems to be tied to the wind speed in the rather remote Greenland Sea in both autumn (negative CC) and winter (positive CC), in the Norwegian Sea during winter (positive CC), and in the Irminger Sea in the autumn (negative CC). In the latter region, there is also a positive correlation between the wind speed and the east Atlantic pattern, which is typified by anomalously low pressure to the south and southeast of Iceland. In the positive phase of the Polar Eurasian pattern, a region of low pressure is centered on the Siberian coastline toward the Arctic Ocean. This should normally lead to westerly or northwesterly flow over the Barents Sea. It is positively correlated with the wind speed both there and in the Norwegian Sea in the autumn.

[67] As mentioned in the introduction, there are many incentives for providing a surface wind speed climatology for the Nordic seas. The marine human activity in the region is projected to increase with receding sea ice and expanding oil and gas exploration. Moreover, the northeast Atlantic is a crucial region for understanding the atmospheric and oceanic circulation on a global scale. This study shows that there are significant similarities, but also important differences, between the spatial distribution of QuikSCAT and NCEP/NCAR reanalysis surface winds. This information is important because ocean models are routinely forced with atmospheric reanalysis products [e.g., Haak et al., 2003]. Also, coupled climate models are often run with a spatial resolution which is similar to or lower than the one used by NCEP/NCAR, but without the significant benefit of data assimilation. It is therefore to be expected that features which are not resolved by the reanalysis will almost certainly not appear in climate models.

[68] The possible effects of the wind field on sea ice transport are also important. We have shown that the differences between the NCEP/NCAP reanalysis and QuikSCAT are large near the marginal ice zone. In a modeling experiment, Harder et al. [1998] found that the sea ice export through the Fram Strait had a roughly linear dependence on the wind speed over the ice. The effect of ocean currents was found to be much weaker. On a more local scale, Brümmer et al. [2003] investigated the impact on the sea ice movement in the marginal ice zone of a weakening cyclone moving northward into the Fram Strait. They found that the passage of the cyclone (and thus of the winds) had a significant influence on both the speed and direction of the sea ice drift.

[69] The discovery of the Greenland tip jet [Doyle and Shapiro, 1999; Moore and Renfrew, 2005] led Pickart et al. [2003] to force an ocean model with a number of tip jet events. They found that deep convection took place, and claimed that the tip jet was “the most likely cause of the convection in the Irminger Sea.” It is not unlikely that a model forced with QuikSCAT winds in the Greenland Sea would have provided new insight into the properties of the cold East Greenland Current, the deep water formation over the Greenland Basin and the transport of freshened water and sea ice.

6. Conclusions

[70] The first purpose of this paper was to investigate the surface wind field in the Nordic seas. Wind speed maxima were identified at other end of the Denmark Strait, along the ice edge in the Greenland Sea, close to the southern tip of Spitsbergen and along the southwest coast of Norway. These maxima are associated with topography and/or sea ice distribution. Wind speed minima, mostly also due to orographic forcing, were found to the southwest of Iceland, to the immediate north of Iceland, between Jan Mayen and Spitsbergen, over the Svalbard Bank and in the eastern Barents Sea.

[71] The QuikSCAT wind speed spatial and temporal distributions were compared to the NCEP/NCAR reanalysis. Important differences were found. Features that are due to orographic forcing are generally not captured by the reanalysis. The wind speed minima are also absent. As far as it is possible to judge from only 8 years of data, however, the low-frequency temporal distribution is captured well enough for the reanalysis to be reasonably compatible with the QuikSCAT winds.

[72] The ocean surface wind speed has been shown to be closely tied to planetary-scale, low-frequency measures of variability. The seasonal NAO index during winter is intimately correlated with the wind speed in the Greenland, Norwegian and Irminger Seas. A negative correlation between the wind speed and the phase of the Scandinavian pattern is found from autumn to spring in the Greenland and Norwegian Seas.


[73] The author would like to thank Kent Moore and one anonymous reviewer for thorough and useful comments. Burghard Brümmer and Stephen Mobbs are thanked for fruitful discussions. This is publication A 191 from the Bjerknes Centre for Climate Research. This work was supported by the Norwegian Research Council through its International Polar Year programme and the project IPY-THORPEX (grant 175992/S30).