An examination of storm activity in the northeast Atlantic region over the 1851–2003 period using the EMULATE gridded MSLP data series

Authors


Abstract

[1] The gridded mean sea level pressure (MSLP) series developed by the European and North Atlantic Daily to Multidecadal Climate Variability (EMULATE) project is used in this paper to analyze changes in storm activity in the northeast Atlantic region over the period 1851–2003. Zonal and meridional geostrophic wind speed components were calculated for each grid square over the domain 65°W–45°E, 30°N–65°N, and seasonal percentiles from these data were used as a measure of storm activity in the region. Despite the relatively coarse temporal (daily) and spatial (5° × 5°) resolution of the data, the results indicate that the series are able to provide useful information about storm activity across the northeast Atlantic domain back to 1881 and in the data-rich area of the North Sea back to 1851. The variability of winter storm activity observed across the North Sea over the last 153 years, and particularly the increase to high values in the 1990s, is closely associated with the variability of the North Atlantic storm track. In contrast, high summer values in the late nineteenth century appear to have been related to a Greenland-blocking system, leading to high meridional flow in the region. This latter conclusion is limited, however, by poor data coverage in the northwest Atlantic region.

1. Introduction

[2] Several recent studies have indicated that a significant increase in winter storm activity occurred in the North Atlantic Region during the latter half of the twentieth century following a relatively quiescent period in the mid-twentieth century [e.g., Chang and Fu, 2002; Wang et al., 2006, 2009a; Chang, 2007]. Attempts have been made to assess the influence of anthropogenic forcing on this feature of the storm climate of the region using reanalysis data [Wang et al., 2009a], although these data only allow analysis back to the 1940s. Other studies have used station-based mean sea level pressure (MSLP) data to study this subject, and due to the extended length of station series they have been able to place the variability over the last 50 years in the context of the last 150 years or longer [Hanna et al., 2008; Bärring and Fortuniak, 2009].

[3] A useful way of extracting storm information from station MSLP data is to calculate geostrophic wind energy indices from station triangles. These data provide a more robust indication of extratropical storm activity than the more directly related wind speed series, which are notoriously susceptible to inhomogeneities and are generally short in length [Lamb and Weiss, 1979; The WASA Group, 1998; Weisse and Storch, 2009]. Early studies using this pressure triangle technique concluded that a peak in annual storm activity observed during the late twentieth century in the northeast Atlantic region was of equal magnitude to a previous maximum in the late nineteenth century [Schmidt and von Storch, 1993; Alexandersson et al., 1998, 2000; The WASA Group, 1998; Trenberth et al., 2007]. A recent study by Wang et al. [2009b, hereinafter WANG09] marks a significant development from these previous studies by showing that large seasonal and regional differences have occurred in storm activity in the northeast Atlantic over the last 130 years. Using the pressure triangle technique, WANG09 have shown that winter storm activity in the North Sea over the period 1874–2007 reached a peak in the 1990s whereas summer storm activity peaked in the 1880s. The earlier pressure triangle studies failed to appreciate this important feature because they only analyzed storm activity at the annual resolution.

[4] Midlatitude cyclones are important for the transport of heat from equatorial regions to higher latitudes and therefore provide an indication of the general state of the global climate system [Wang et al., 2006; Bärring and Fortuniak, 2009]. Wind storms arising from deep cyclones have a direct influence on several sectors of the economy and changes in the frequency and severity of storms in the northeast Atlantic region have serious consequences for the offshore oil and gas industry, particularly in the North Sea area [Bijl et al., 1999]. The results from WANG09 are therefore of particular importance not least because their results suggest that storm activity in the North Sea area may be somewhat different to other areas of the northeast Atlantic area. However, WANG09's analysis was restricted to a limited area of the northeast Atlantic (∼10°W–10°E; 50°N–60°N) for results that extend back to the nineteenth century. There is need to ascertain if the changes that WANG09 observe are peculiar to the analyzed region or if they are connected to wider changes across the North Atlantic, particularly in relation to the storm track. This requires the use of gridded data series.

[5] The longest gridded data series currently available that is of a sufficient resolution (at least daily) to provide information on storm activity over the last 150 years is the MSLP product developed by the European and North Atlantic Daily to Multidecadal Climate Variability (EMULATE) project [Ansell et al., 2006]. This data series has not been used extensively to study storm activity, which is probably due to the low temporal (daily) and spatial (5° × 5°) resolution of the data compared to reanalysis products. Further, only the surface parameter of MSLP is gridded, and there are known homogeneity problems with the data in the period before 1881, particularly at the extremities of the spatial grid where few pressure series exist [Bhend, 2005; Ansell et al., 2006]. Nonetheless, synoptic reconstructions of certain storm events have proved useful [Ansell et al., 2006] and storm tracking procedures have been applied to the data [Bhend, 2005]. This latter study indicated that while the EMULATE series will fail to detect the smallest and fastest moving cyclones, it is possible to use these data to analyze storm activity in a relative sense. However, the relatively low spatiotemporal resolution of the EMULATE data set means that feature-tracking techniques, such as the automatic procedure of Murray and Simmonds [1991] used by Bhend [2005], could have difficulty in identifying synoptic disturbances in the data beyond the background pressure field.

[6] The aim of this paper is to explore the potential for the EMULATE data to analyze storm activity across the northeast Atlantic region during the period 1851–2003. Given that the EMULATE series covers the majority of the North Atlantic region (70°W–50°E, 25°N–70°N), these data have the potential to show how storm activity in the northeast Atlantic sector over the last 150 years relates to the wider region. Simple pressure gradient indices are used in this study, which allow comparison with the results of pressure triangle analyses and are considered more suitable than feature-tracking techniques given the spatiotemporal resolution of the data.

2. Data Preprocessing

[7] As has been noted by Chang and Fu [2002], amongst many others, there are several ways to define storm activity using gridded data products. The EMULATE data series with its only parameter being MSLP restricts the use of many of these definitions. In addition, the spatiotemporal resolution of the data further limits the methods that can be used. The EMULATE data represent a 24 h average of MSLP and not instantaneous values, which results in a smoother field than would be found with synoptic or subdaily reanalysis data [Ansell et al., 2006]. The smoothness of the MSLP field is compounded by the relatively coarse spatial resolution of the data and the interpolation technique used. Feature-tracking techniques would therefore have difficulty in distinguishing cyclonic systems from the background pressure field. This has been demonstrated in detail by Bhend [2005] who applied the automated technique developed by Murray and Simmonds [1991] to both the EMULATE data and the 6-hourly NCEP/NCAR 2.5° × 2.5° gridded MSLP data. His results indicate that storm track length and lifetime is overestimated in the EMULATE data, whereas the number of events tracked is underestimated.

[8] In this study, squared MSLP gradients are used as a measure of storm activity. The EMULATE MSLP data were obtained from http://hadobs.metoffice.com/emslp/ and zonal (ug) and meridional (vg) gradients were calculated for each day from 1 January 1850 to 31 December 2003 as described in Appendix A. Due to gradients of MSLP being used, the resulting latitude/longitude grid is reduced by 1 × 5° grid square on each border and hence by 10° on each axis compared to the original field (70°W–50°E, 25°N–70°N). Seasonal percentiles (95th and 99th) were calculated from the daily geostrophic wind speed values but only the results for winter (DJF) and summer (JJA) are considered in this paper. The results from the 95th and 99th percentiles are broadly similar and the results presented in the main part of this paper are from the 95th percentiles. Results for the 99th percentiles are provided in the auxiliary material. The winter season is defined as the values in December, January and February and hence the final percentile series runs from 1851 to 2003.

[9] In previous studies, geostrophic wind series derived from squared MSLP gradients of gridded data series have been shown to be a good indicator of the true wind speed and hence storminess [Jenkinson and Collison, 1977; Hulme and Jones, 1991; Jones et al., 1999; Chen, 2000; Wang et al., 2009a]. These studies used the sum of squared zonal (ug) and meridional (vg) wind components to calculate the total flow associated within a given region or grid box (see Appendix A). The two components (ug and vg) are considered separately in this paper due to differences in the results from the components, which point to changes over time in the relative importance of different dynamical mechanisms affecting storm activity in the northeast Atlantic region. The MSLP gradient series developed in this study differ slightly from previous studies in two other ways: Wang et al. [2009a] calculated the geostrophic wind using seasonal averages of MSLP whereas Hulme and Jones [1991], Jones et al. [1999], and Chen [2000] incorporated a vorticity term into their calculations.

[10] In a study by The WASA Group [1998], a qq plot was used to demonstrate the similarity in the probability distribution of wind speed data and geostrophic wind speed data derived from a pressure triangle. Wind speed series provide a direct measure of storm activity, but as mentioned in the introduction to this paper such series are notoriously inhomogeneous and short in length. With these problems in mind, the example of The WASA Group [1998] was repeated and the probability distributions of the daily EMULATE zonal and meridional gradients were compared against the distributions of daily wind speed data from the UK. The wind speed data were obtained from the British Atmospheric Data Centre (BADC) repository [Met Office, 2008] and the homogeneity of the data was checked by comparing the annual means against the UK Department of Energy and Climate Change's Windspeed Database [Brocklehurst, 1997]. Based on this comparison the data are considered homogeneous. In Figure 1, qq plots are shown for each of the 5° EMULATE grid squares that cover the UK, in comparison with station wind speed series that lie within each grid. The WASA Group [1998] showed that the distributions of wind speed for a single Danish station and pressure triangle derived geostrophic wind speed were very similar, with a near-45° slope achieved in their qq plot. The results shown in Figure 1 also indicate a similar distribution in both the geostrophic (ug and vg components) and the station wind speed series, although the wind speed series have slightly longer tails at the lowest quantiles. This results from the wind speed series being more sensitive to light winds than the geostrophic wind series, which have more in common with background geostrophic flow than storm activity associated with individual cyclonic disturbances. Similar results are achieved regardless of the station wind series used, and if the results are calculated separately for winter and summer. The results in Figure 1 show that there are no major inhomogeneities residing in the wind speed series, and in general the similarity of the geostrophic wind speed and observed wind speed distributions indicate that changes in percentiles of the ug and vg components are able to provide a good approximation of changes in the distribution of the wind speed data.

Figure 1.

QQ plots of daily station wind speed and geostrophic wind components for the five EMULATE grid squares that cover the United Kingdom. The names of the wind speed stations are provided with the respective grid square, and the 95% and 99% percentiles are indicated by the dashed lines for reference purposes. The diagonal line joins the 25% and 75% percentiles. The data cover the period 1961–1980, and details about the wind speed series are provided in Table S1 in the auxiliary material.

3. Results

3.1. Geostrophic Wind Flow in the North Sea and Northeast Atlantic Sectors

[11] Geostrophic wind speed percentiles in the North Sea sector were initially considered by averaging the seasonal values from the four grid boxes 0°–5°E, 55°N–60°N. Prior to averaging the individual grid series had been converted to anomalies from the 1851–2003 mean and were weighted by the cosine of latitude, due to the differences in the areas of the grids at each latitude.

[12] The zonal component of the geostrophic wind during the winter (Figure 2a) bears a close relationship to the WANG09 results at both the interannual and decadal time scales, but the relationship is much weaker in summer (Figure 2c). In contrast, the results for the meridional component (Figures 2b and 2d) show a similar association in both the winter and summer. The trend in the zonal component while generally indicating a decline over the 1851–2003 period in both winter and summer, shows the important seasonal difference highlighted by WANG09 of an increase during the late twentieth century during the winter but not the summer. There is a clear division, however, between the results for the zonal component and those for the meridional component. In the winter much of the increase at the end of the twentieth century is attributable to a rise in the zonal component. Conversely the high values at the end of the nineteenth century seem to be associated with high values in the meridional component. This indicates that, alongside the seasonal differentiation, there may be different dynamical mechanisms responsible for the two periods of high storm activity in the North Sea sector. This potentially important feature is lost in the results of WANG09 because their geostrophic wind series is formed from the root sum of squared zonal and meridional components. However, a difficulty arises in that the results for the meridional component appear less reliable than those for the zonal component. Anomalously high values are evident in Figure 2 during the 1940s in the meridional component. High values are indicated in both winter and summer, and while the WANG09 results show high values in winter the values during the summer are quite low. This may point toward problematic data in the EMULATE series during World War II.

Figure 2.

The winter and summer variability of geostrophic wind speed percentiles (95%) in the North Sea sector (0°–5°E, 55°N–60°N) derived from the EMULATE and WANG09 data. Both series have been normalized to allow the series to be plotted together. The thick lines show the 11-point Gaussian filtered values. Also indicated is the linear regression line for the EMULATE data calculated over the period 1851–2003 and the least-squares linear trend and ±2 SE. The correlations between the series (1875–2003) are shown; the filtered correlations are shown in brackets. Correlations significant at p < 0.01 (double asterisks) are indicated. Autocorrelation has been taken into account for the significance of the filtered values (following Woollings et al. [2008, p. 621]). The dashed lines highlight the periods used for the calculation of temporal averages in Figures 4 and 5.

[13] The results for the northeast Atlantic sector (15°W–20°W, 55°N–60°N) are shown in Figure 3. This sector slightly overlaps the Valentia-Stykkisholmur-Torshavn pressure triangle of WANG09 but is mostly outside of the area considered in their study. Indeed the sector was deliberately selected for this reason but also because a relatively high number of ship observations were available for this region in the EMULATE data set [see Ansell et al., 2006; Bhend, 2005]. However, it should be noted that while there are relatively many ship observations in this sector after 1881 there are no direct observations before 1881. During the 1850–1880 period the MSLP data in this region are interpolated from land-based station observations and more southerly ship observations [Bhend, 2005; Ansell et al., 2006].

Figure 3.

The winter and summer variability of geostrophic wind speed percentiles (95%) in the northeast Atlantic sector (15°W–20°W, 55°N–60°N). The thick lines show the 11-point Gaussian filtered values. Also indicated are the linear regression line, the linear trend, and ±2 SE, all calculated over the period 1851–2003. As with Figure 2, the dashed lines highlight the periods used for the calculation of temporal averages in Figures 4 and 5.

[14] The results in Figure 3a for the zonal component of the geostrophic flow during the winter show pronounced decadal variability, superimposed upon a gradual decline in values from the second half of the nineteenth century until the mid twentieth century. A recovery of values is apparent after the 1960s to high values in the 1990s, although the overall trend is one of decline. This is similar to the results for the North Sea sector (Figure 2). The values during the latter half of the nineteenth century and the first decade of the twentieth century are comparable, if not higher, than the values during the last decade of the twentieth century. Interestingly the values during the 1851–1880 period do not appear grossly erroneous given the lack of direct observations, but do contribute to the general downward trend. WANG09 also showed a downward trend in their Valentia-Stykkisholmur-Torshavn pressure triangle for the winter season, although their series only extended back to 1892. This feature was not evident in their results for the North Sea, which led WANG09 to suggest that a zonal difference in storminess was evident across the region during the winter. In the results here from the EMULATE data, a zonal difference is only apparent during the summer, when a significant downward trend is achieved for the North Sea (Figure 2c) but not the northeast Atlantic (Figure 3c); the trends during the winter are similar in both the North Sea and northeast Atlantic sectors.

[15] The results for the meridional component of geostrophic wind in Figures 3b and 3d indicate that there are potential inhomogeneities in the EMULATE data for the mid-Atlantic region. A clear break is evident in circa 1950 in both seasons and an examination of various grid box values (not shown) indicates that this breakpoint exists in varying degrees of severity across much of the region west of 5°W, but appears to reach a maximum in the region covered in Figure 3 (15°W–20°W, 55°N–60°N). The cause of this breakpoint is uncertain but is probably related to a change in data input in the gridded series of Jackson [1986], which is a major component of the EMULATE data series in the region. After 1949 different charts were used to construct the gridded data and some uncertainty exists regarding the exact time stamp of the data [Ansell et al., 2006].

3.2. Spatial Patterns of Geostrophic Wind Flow

[16] In an attempt to place the results from Figures 2 and 3 in the context of storm activity in the wider North Atlantic region, the averages of geostrophic wind percentile anomalies for the decades of high values are plotted in Figures 4 and 5. The choice of these decades (1878–1883 and 1989–1998) is somewhat arbitrary, but the periods were selected to encompass the highest values in the results of WANG09 for the North Sea area.

Figure 4.

Averages of the zonal percentile anomalies during certain decades. The data are in the unit of hPa. The anomalies are calculated as differences from the long-term (1851–2003) mean at each grid point. The areas highlighted indicate the grid squares used to compute the spatial averages in Figures 2 and 3 above.

Figure 5.

Same as Figure 4 but for the meridional component of the geostrophic wind.

[17] The average winter zonal anomalies in the 1989–1998 decade (Figure 4b) are highly positive across much of the central North Atlantic basin, the North Sea and Scandinavia areas. Toward the south and north of this area the values are close to zero, and even slightly negative in places. During the 1874–1883 decade (Figure 4a), high positive anomalies are evident over the North Sea, extending toward the south of Iceland in the winter; during the summer (Figure 4c) weak positive values are evident across much of the domain.

[18] The results for the meridional component (Figure 5) indicate a similar bimodal pattern in all four plots, although with different polarity depending on the decade. In the winter during the 1874–1883 period (Figure 5a) high positive anomalies are evident across the southeast Atlantic stretching across the North Sea; high negative anomalies are apparent in the northwest Atlantic region reaching a maximum to the south of Greenland. A similar pattern is evident during the summer (Figure 5c). For the period 1989–1998 a broadly similar bimodal pattern can be observed although the polarity is reversed in both seasons (Figures 5b and 5d), with highly negative anomalies in the northeast Atlantic and highly positive values to the south of Greenland. In this later decade there are also slight shifts in the locations of the maximum and minimum anomalies.

[19] The results from the two geostrophic components shown in Figures 4 and 5 are complementary and the variations appear to be linked to differences in the position of the north Atlantic storm track and atmospheric blocking. In the decade 1874–1883 the pattern of winter anomalies seems to be associated with a Greenland blocking regime [see Woollings et al., 2010], with highly negative anomalies of meridional flow extending southeast from Greenland. Weakly negative anomalies in meridional flow can also be seen across western Russia. As a result of this configuration there appears to have been a tendency for positive zonal anomalies across the North Sea. In summer a similar Greenland blocking pattern can be observed, although the storm track seems to have been more elongated and tended to extend further into Scandinavia, giving high positive anomalies in the meridional flow across the North Sea and much of Scandinavia. This pattern corresponds to weak anomalies in the zonal flow. In summer 1989–1998 the anomalies in both the meridional and zonal components are relatively weak. However, during the winter in that decade strong positive zonal anomalies are evident in the region of the north Atlantic storm track, and high positive meridional anomalies seem to have been a feature only across the northwest extremity of the storm track, across south Greenland.

3.3. The Dominant Modes of Interannual Variability

[20] The results shown in Figures 4 and 5 suggest that the high rate of winter storm activity in the 1989–1998 period was closely associated with a more general intensification of the North Atlantic storm track. To study this further an Empirical Orthogonal Function (EOF) analysis was conducted on the seasonal geostrophic wind speed percentiles to isolate the dominant modes of interannual variability. The seasonal percentile anomalies were first weighted by the square root of the cosine of latitude and then normalized. It was discovered that no clear pattern was achieved in the results for the summer season, which is probably due to a known problem with the data in that the high-pressure systems appear too low and the low-pressure systems appear too high [Ansell et al., 2006]. This appears to be more of a problem during the summer when gradients are naturally lower. An EOF analysis applied to the winter meridional index also failed to produce reliable results. That analysis placed the highest loadings in the northwest of the domain, toward the south of Greenland. This result may be an indication of the importance of the Greenland blocking regime in the interannual variability of meridional flow across the North Atlantic domain [see Woollings et al., 2010], but the EOF pattern was not robust over time, with different patterns evident when the EOF analysis was applied to the data for various different time periods. This finding, and the general low quality of the data in that region [see Bhend, 2005; Ansell et al., 2006], led us to the conclusion that the results could not be relied upon. Conversely, the EOF pattern for the zonal index in winter is similar regardless of the time period considered, which allows more confidence in those results.

[21] During the winter, the leading EOF of the zonal index (Figure 6) represents the North Atlantic storm track very well, with the area of dominant activity stretching across the North Atlantic basin from Newfoundland to Scandinavia [cf. Hoskins and Hodges, 2002, Figure 2; Chang and Fu, 2002, Figure 3]. The leading pattern depicts the common feature of synoptic disturbances that occur in this region in association with the subpolar jet stream. However, in contrast to the depictions of the North Atlantic storm track in previous studies, the area of maximum intensity is shifted farther to the northeast in these results. This is probably due to the more restricted spatial domain of the data rather than the introduction of biases from poor quality data in the early period. This was confirmed by an EOF analysis that was conducted for the period 1950–2003 (not shown) which showed a similar pattern to that achieved for the period 1851–2003. Given this information, PC 1 may be viewed as a proxy for the North Atlantic winter storm track. Much of the success of the results probably lies with the fact that the area of high loadings correspond to the areas of highest data availability in the EMULATE series, according to the analyses by Bhend [2005] and Ansell et al. [2006].

Figure 6.

The leading EOF of the zonal geostrophic wind component during the winter season. This pattern was constructed following Chang and Fu [2002] by regressing the geostrophic wind speed percentile anomaly field onto the standardized PC time series (1851–2003). Hence, the units of this diagram are hPa per unit standard deviation of the PC time series. The percentage of variance explained by the EOF is indicated. As with Figures 4 and 5, the areas highlighted indicate the grid squares used to compute the spatial averages in Figures 2 and 3.

[22] The results shown in Figure 6 are indicative of the two regimes (zonal and blocked) of the North Atlantic Oscillation (NAO) described by Woollings et al. [2010], with positive NAO conditions associated with the subpolar jet stream oriented southwest-northeast across the domain, and being distinct from the subtropical jet; negative NAO conditions are associated with a joining of the subpolar and subtropical jet. Following Woollings et al.'s regime view of the NAO, negative phases of PC1 should be associated with an increased frequency of blocking over Greenland, although this cannot be directly inferred from these results.

[23] A close association is evident between the leading winter PC time series and the results from WANG09 for the North Sea area (Figure 7). Given that PC 1 represents the North Atlantic storm track, the results in Figure 7 suggest that the variability of geostrophic wind percentiles in the winter shown by WANG09, and particularly the increase in the latter half of the twentieth century to a peak in the 1990s particularly in the North Sea region, is linked to a more general intensification of the North Atlantic storm track. High values of the PC time series during the 1989–1998 decade reflects the close correspondence shown in Figure 2a between the time series of zonal flow depicted in the EMULATE data across the North Sea and the results of WANG09, as well as the strong zonal conditions shown in Figure 4b across the mid–North Atlantic during the decade 1989–1998. However, whereas a significant long-term decline is evident in the zonal flow across the North Sea in the winter (Figure 2a), the decline is much weaker, and not significant at the 95% level, in the PC time series (Figure 7).

Figure 7.

The winter variability of the PC time series derived from the zonal geostrophic wind speed percentiles (95%). The structure of this plot follows that used in Figure 2. Correlations significant at p < 0.1 (asterisks) and p < 0.01 (double asterisks) are indicated.

4. Discussion

[24] A major problem with using the EMULATE MSLP series to analyze storm activity is the spatial and temporal variation in the number of data used to construct the gridded series. Analyses by Bhend [2005] and Ansell et al. [2006] have demonstrated this problem. In the North Sea area a high number of land and marine observations were included in the EMULATE data series after 1881, but ship observations alongside land stations surrounding the North Sea were also available for the 1850–1880 period. Hence the data in that region can be considered more reliable throughout the 1851–2003 period compared to other areas. This reliability is reflected in Figure 2, which shows a close relationship between the EMULATE results and those from WANG09's study during the winter for the dominant zonal component of geostrophic wind. However, in comparing the results derived from the EMULATE data and those of WANG09 it must be remembered that the two data series are not strictly independent. All of the station data used by WANG09, with the exception of Jan Mayen, were used in the construction of the EMULATE series [Ansell et al., 2006]. An alternative way of viewing these results, given the risk of circular relationships, is to consider that daily MSLP data gridded at the 5° resolution are able to provide useful information about winter storm activity in areas with a high density of data.

[25] During the summer the relationship between the WANG09 and EMULATE results for the North Sea area is not as strong, and the meridional component is almost as equally important to the interannual/decadal variability of WANG09's geostrophic wind series as the zonal component. Part of the weaker correspondence between the EMULATE and WANG09 results during the summer may be due to a known problem with the EMULATE series in that the data are generally “dampened,” with high-pressure systems not “high” enough and low-pressure systems not “deep” enough [Ansell et al., 2006]. As this appears throughout the series, the authors concluded that this was due to the infilling and smoothing of the marine observations or the use of the Reduced Space Optimal Interpolation (RSOI) technique, which was used to complete the spatial field. The effects of this “flattening” of the data on geostrophic wind percentiles would probably be most profound during the summer months when pressure gradients are generally lower. This may explain why the correlations observed in the summer between the EMULATE and WANG09 series are generally lower compared to winter, although a lower coherence of the summer storm climate may also have an influence on those statistical relationships, as well as the relatively coarse temporal and spatial resolution of the EMULATE data.

[26] An examination of the variability of storminess in the North Atlantic region during the latter half of the nineteenth century, outside of the data-rich North Sea, is problematic. Prior to 1881 the EMULATE data series is composed of a limited number of ship observations, which are restricted to the main shipping routes, and land-based station data. After 1881 many more ship observations were included and combined with the gridded MSLP data of Jackson [1986], which [Ansell et al., 2006, p.2718] “implicitly contain many thousands of station observations.” However, even after 1881 the number of ship observations vary with very few observations available during the two world wars [Bhend, 2005; Ansell et al., 2006]. The winter North Atlantic Storm track lies in an area that is relatively rich in ship observations over the 1881–2003 period [see Bhend, 2005], and is also suitably positioned for station data from Newfoundland and the Faroe Islands to be of use. In contrast, the northwest Atlantic is an area that is generally poor in data, and there are particular problems with the data around Greenland [Ansell et al., 2006], especially in the period prior to 1881 [Bhend, 2005]. This leads to considerable uncertainty in the conclusion that the high storm activity in the northeast Atlantic during the summers of the 1870–1880s arose from high meridional flow that resulted from increased blocking around Greenland. The Greenland blocking pattern evident in these results, but also in other studies [e.g., Woollings et al., 2010], lies across the area that has large errors related to the interpolation procedure [Bhend, 2005]. However, the coherence of the pattern from the zonal and meridional components when viewed across the entire North Atlantic adds credence to this Greenland blocking hypothesis. Support for this assertion is also provided by Moses et al. [1987] who showed that there was an increased frequency of Greenland blocking in the winters during the period 1873–1900. Folland et al. [2009] in their Summer North Atlantic Oscillation (SNAO) index, showed strongly negative values in the 1870–1880s, which relate to reduced pressure anomalies across the North Sea and increased anomalies over Greenland during high summer (July and August).

[27] A comparison with independent subdaily station data would be required to adequately assess the ability of the EMULATE data to capture storms, and specifically to establish the influence of the RSOI and infilling techniques on the results. Given that all available station MSLP series across the northeast Atlantic/western European domain were used to construct the gridded data this cannot be achieved without opening further questions about the degree to which any observed relationships are circular. However, wind speed series are entirely independent of the gridded data and may be used. As has been demonstrated in Figure 1, the EMULATE geostrophic wind data percentiles are indicative of the “true” wind speed data, although the wind series appear to be influenced more by small-scale/short-lived events.

5. Conclusions

[28] Indices of storm activity developed from station-based pressure triangles have been used in several studies to provide a proxy for wind storms in the northeast Atlantic region. Most notably, WANG09 have shown appreciable changes in storm activity in the northeast Atlantic since 1874, but have also indicated that seasonal and regional differences exist over that time period. The gridded MSLP data series developed by the EMULATE project [Ansell et al., 2006] was used in this paper to provide spatial and temporal extension to the results of WANG09. The usefulness of these data in analyzing storm activity is limited across much of the North Atlantic region by variations in the availability of data, and is severely compromised in the period 1851–1880 when few ship observations are available. However, the data covering the North Sea region are generally reliable to due a high number of available MSLP data. It has been shown that the results of WANG09 for the North Sea can be replicated, and indeed reliably extended by some 25 years in that area. This indicates that gridded MSLP data of a relatively coarse temporal and spatial resolution can be used to provide useful information on storm activity in areas where a high number of observations have been used to construct the series.

[29] In the data-rich North Sea sector, the results presented in this study broadly support the findings of WANG09, who showed that high values in the late nineteenth century mostly occurred during the summer season whereas high values at the end of the twentieth mostly occurred during the winter. The results indicate that the increase in winter values toward the end of the twentieth century was associated with increased zonal flow connected with a general intensification of the North Atlantic storm track. The high summer storm activity at the end of the nineteenth century, and particularly the 1880s, appears to have been largely related to increased meridional flow. This pattern is evident across the northeast Atlantic region, and seems to have been connected to a strengthened Greenland-blocking action. Unfortunately, while this conclusion is plausible and fits in with the results across the relatively data-rich North Sea/European area, as well as the findings of other studies, a definite conclusion relies on data from the relatively data-poor northwest Atlantic. Problems appear in the data as late as 1950 in the region west of ∼15°W that seem to relate to a change of input data. In addition, the results across the gridded domain for the summer season must be viewed with caution due to the construction of the gridded data series. The interpolation and infilling techniques used to construct the gridded data limits the usefulness of the data in detecting summer storm activity, when pressure gradients tend to be low. In the winter when gradients are higher the EMULATE data perform better.

[30] The decadal variations in geostrophic wind values in the North Sea sector are superimposed on long-term downward trends. These trends are evident for both the zonal and meridional components of the geostrophic wind, although certain clearly defined inhomogeneities are apparent in the results for the meridional components, which reduces our confidence in those results. WANG09 showed similar downward trends for the summer season, but not the winter season. In the results from the EMULATE data shown in this paper, the trend in the zonal component is much stronger during the summer, although it is still evident during the winter. The reason for these divergent results is unclear but may be related to differences in the spatial-temporal resolution of the data used. Due to the coarse temporal and spatial resolution of the EMULATE data, the results will fail to capture severe storm activity arising from smaller-scale weather systems. If such systems contributed more to the total winter storm activity at the end of the twentieth century compared to the late nineteenth century then this may explain the different results.

[31] The analysis of severe storms using subdaily data is an important area of current research [Allan et al., 2009]. The recently released Twentieth Century Reanalysis Project data series (version 2, http://www.esrl.noaa.gov/psd/data/20thC_Rean/) provides subdaily gridded MSLP back to 1871 and should capture more of these events than the EMULATE data. This reanalysis product also contains zonal and meridional wind components and geopotential heights, and a more detailed analysis of the temporal variation of the storm climate of the North Atlantic region should be possible using this data set [Compo et al., 2011]. Most importantly, many more data for the northwest Atlantic have been incorporated into the data set, which should allow a more reliable assessment of the role of Greenland blocking in storm activity during the late nineteenth century. An in-depth study concerning storm activity in the North Atlantic area using this data set should be a research priority.

Appendix A:: Calculation of Geostrophic Wind Components

[32] Taking the EMULATE gridded MSLP data (p), zonal (ug) and meridional (vg) gradients were calculated as:

equation image
equation image

for each latitude (j ∈ {65, 60, …, 30°N}), longitude (i ∈ {−65, −60,…, 45°E}) and time step (t ∈ {1, 2,…, N}). The calculation of these geostrophic components follows the example of Wang et al. [2009a], although they used the total geostrophic flow (G) by calculating the sum of the squared components, i.e., G = ug + vg. The weighting according to latitude (cos(ϕj)) in the vg calculation accounts for the difference in angular distance between neighboring grid points [Wang et al., 2009a]. The square roots of the components (ug and vg) were used in our analysis and where the components are referred to in the main part of the paper, it is implied that ug = equation image and vg = equation image. This technique follows the example used by Jones et al. [1999]Chen [2000], but not Wang et al. [2009a]. The final ug and vg components are in the units of hPa per 5° spherical distance.

Acknowledgments

[33] We are grateful to Xiaolan Wang (Climate Research Division, Environment Canada) for providing the pressure triangle data for the North Sea area and to Colin Harpham (CRU) for checking the homogeneity of the wind speed data. Figure 1 was constructed using the Lattice package for R [Sarkar, 2008]. This research was funded by a U.S. Department of Energy research grant (DE-FG02-98ER62601). This paper has been significantly improved by the incorporation of comments made by several anonymous reviewers.

Ancillary