Is the NAO winter index a reliable proxy for wind climate and storminess in northwest Europe?

Authors


H. Burningham, Coastal and Estuarine Research Unit, UCL Department of Geography, Pearson Building, University College London, Gower Street, London WC1E 6BT, UK. E-mail: h.burningham@ucl.ac.uk

Abstract

Climate measures synthesized from the instrumental wind record at 53 stations in northwest Europe are compared with various derivations of the North Atlantic Oscillation (NAO) winter index. The NAO is highly correlated with wind direction frequency, with positive phases in the NAO winter index associated with increased frequency of winds from the southwest. These correlations are sensitive to the choice of winter period and diminish with longer period, especially the inclusion of October. Evidence for connections between the NAO and wind speed or storminess measures is far less convincing, particularly in terms of the frequency of extreme wind speed events. These results have important implications for the use the NAO winter index as a proxy for storminess, whether in palaeoenvironmental reconstructions or as a basis for storm forecasting. Copyright © 2012 Royal Meteorological Society

1. Introduction

The North Atlantic Oscillation (NAO) is the primary mode of climate variability across the North Atlantic region, and is associated with an oscillation in atmospheric mass between the Arctic and subtropical North Atlantic. Empirically, the NAO is expressed as the difference in normalized sea-level pressures across the mid-latitude Atlantic (Hurrell, 1995; Jones et al., 1997), between stations with suitably long records in the south (Lisbon; Ponta Delgada in The Azores; Gibraltar) and the north (southwest Iceland). Quantification of the NAO can also be achieved through an Empirical Orthogonal Function (EOF) analysis of a spatially distributed climate variable (i.e. sea level pressure), which has the advantage of accounting for the full spatial pattern of the NAO (Hurrell and Deser, 2009).

The NAO is most pronounced in winter, when the atmosphere is most active dynamically (Hurrell and Deser, 2009). Accordingly, a NAO winter index is normally computed, from regional differences in normalized pressure anomalies or the time series of the principal component of an EOF analysis of spatial sea level pressure, for various combinations of winter months. Positive phases in the index, describing a north-eastward orientation in the NAO, correspond to wetter, stormier weather in northwest Europe, dominated by strong mid-latitude westerlies; negative phases, describing an east to west alignment of the NAO, are associated with drier, calmer weather in northwest Europe, and storms tracking into Mediterranean Europe (Hurrell, 1995; Rogers, 1997).

The NAO can be reliably calculated from observations back to the early 19th century (Jones et al., 1997), and has been extended back to the 1500s using palaeoenvironmental data (Luterbacher et al., 2002). As an integrated measure of weather, the NAO is increasingly employed as a regionally coherent proxy for a range of weather and climate factors and their variability at interannual to interdecadal scales. The NAO winter index has proved to be a good predictor of a wide range of ecosystem behaviours, including organism growth, population dynamics, and prey–predator interactions (Ottersen et al., 2001). As Straile and Stenseth (2007) observe, the fact that the index is, by implication, associated with multiple aspects of climate (air temperature, wind, and precipitation) that are integrated over space and time means that it is possibly a better predictor of ecological variability than local meteorological measurements (see also Hallet et al., 2004). Beyond ecology, however, more direct linkages between the NAO, specific climate variables, and environmental system responses have been proposed (for a recent review, see Vincente-Serrano and Trigo, 2011). Specific linkages include those between regional precipitation and river flow (Trigo et al., 2004; Mares et al., 2009), wind speed and the vertical mixing of lakes (Spears and Jones, 2010) and wind speed and ocean wave height (Bouws et al., 1996). In coastal studies, the NAO winter index is used as a proxy for storminess, manifest not only in wave heights (e.g. Esteves et al., 2011) and directions (Bruneau et al., 2011) but also in the timing and magnitude of extreme wind events and associated episodes of coastal erosion or sediment deposition (Bartholody et al., 2004; Dawson et al., 2004; Burningham, 2005; Clarke and Rendell, 2006). In studies of this kind, a further attraction of NAO indices as a winter wind climate proxy is that whilst regional atmospheric pressure data often extend back to the 18th century, wind records are largely limited to the 20th century. Changes in extreme wind climate are also of societal, economic and political interest on account of the potential for more frequent and more widespread damage to infrastructure and forest resources (Kaas et al., 1996). Accurate forecasting of winter storminess to reduce these risks is increasingly important (Qian and Saunders, 2003) and improved predictability of the NAO (Keenlyside et al., 2008) is a potential basis for this. Brayshaw et al. (2011) also highlight the possibility of using large-scale atmospheric patterns such as the NAO as a basis for predicting wind statistics and thereby the power generation potential of wind turbines, especially at decadal scales.

Despite several studies of the NAO in the context of North Atlantic wind and wave storms (Qian and Saunders, 2003; Atkinson, 2005; Wolf and Woolf, 2006), there has been virtually no systematic analysis of the associations between NAO indices and instrumental wind records, or the extent to which the NAO winter index constitutes a robust measure of local storminess over longer time scales. This is surprising, given that coastal oceanographic and geomorphological studies in particular have been very much concerned with the occurrence of high-energy events capable of effecting significant geomorphic and ecological change (Davis et al., 2004; Knight and Burningham, 2011; Jiménez et al., 2012).

In this article, we examine the robustness of the association between wind climate, storminess and the NAO winter index, with reference to data for northwest Europe. The analysis focuses on the UK and Ireland, where any correlation between NAO and wind climate should be especially strong given that this region lies in the path of the majority of Atlantic storm tracks (Lozano et al., 2004; Pinto et al., 2009). Our objectives are, first, to determine the correlations between various forms of the winter NAO index and a range of wind forcing and storminess measures derived from a large sample of meteorological records and, second, to determine the extent to which these correlations exhibit systematic geographic variation.

2. Data and methods

Monthly NAO station-based index data were acquired from the Climate Research Unit, University of East Anglia, UK. These relate to sea-level pressure difference between Gibraltar and Reykjavik, Iceland (Jones et al., 1997). The period used to define the winter index varies [e.g. DJF (Weisse et al., 2005); DJFM (Lehmann et al., 2011); NDJF (Kaas et al., 1996); NDJFM (Lozano et al., 2004); ONDJF (Hurrell and Deser, 2010); and ONDJFM (Donat et al., 2010)], but always includes December and January. The rationale for different winter durations is largely based on whether a broad winter average (i.e. the full boreal winter season) or a mid-winter interval best represents the climate characteristics of interest. Here, winter indices were calculated for six different winter periods (DJF, DJFM, NDJF, NDJFM, ONDJF, and ONDJFM) each referenced to the December year. Winter indices derived from a principal component analysis of sea level-pressure anomalies over the North Atlantic (within the region defined by 20°–80°N and 90°W–40°E) were also acquired from the Climate Data Guide at the National Center for Atmospheric Research. These were available as DJF and DJFM PC-based NAO indices, referenced to the January year, which were then adjusted to conform to the December year convention adopted here.

Hourly wind data (hourly 10-min averages of wind speed and direction) were acquired from Met Éireann (Irish sites) and the British Atmospheric Data Centre, Met Office–MIDAS land surface stations (UK). Data span, with variable duration and completeness, the period from 1949 to 2011. Only sites with at least the equivalent of 30 years of data were selected for the final analysis. The resulting dataset comprised 53 sites that encompass a broad selection of coastal, inland, North Atlantic-margin and North Sea-margin sites (Figure 1; Table 1).

Figure 1.

Location of wind record stations (Table 1) used in this analysis. Far northwest and southeast stations, which are considered separately in the later analyses, are highlighted

Table 1. Land surface meteorological stations: summary of station details and wind records
IDNameLat (deg)Lon (deg)Alt (m)Data recordCompletenessRecord (yrs)Region
Met Éireann
1Belmullet54.2278− 10.006911October 1956December 2011100%55.2NW
6Malin Head55.3722− 7.338922January 1956December 2011100%56.0NW
10Valentia51.9397− 10.244411January 1956December 2011100%56.0 
Met Office–MIDAS land surface stations
9Lerwick60.1395− 1.1829910January 1957February 2012100%55.2NW
23Kirkwall58.9539− 2.8998869January 1957February 201280%44.1NW
32Wick58.4541− 3.08846January 1968February 201299%43.5NW
54Stornoway58.2138− 6.3177210January 1957February 2012100%54.9NW
132Kinloss57.6457− 3.5620272January 1959February 201299%52.8 
161Dyce57.2051− 2.20375January 1957February 201282%45.4 
235Leuchars56.3774− 2.8605121January 1957February 201283%45.8 
246Turnhouse55.9508− 3.34692103January 1957November 1999100%42.9 
315Boulmer55.4208− 1.5996644October 1975February 2012100%36.3 
384Waddington53.1751− 0.5217368August 1950February 201299%61.3 
386Cranwell53.0309− 0.5019433January 1968February 201267%29.6 
393Coningsby53.0935− 0.1711910April 1978February 201298%33.2 
409Marham52.6510.567729January 1957February 201278%43.2SE
429Coltishall52.75631.3532157November 1962April 200669%30.1 
440Wattisham52.12340.9591101November 1961February 201270%35.2SE
484Stansted51.88050.22456262January 1959April 199779%30.3SE
556Nottingham53.0053− 1.24969101January 1957February 201283%45.5 
583Wittering52.611− 0.4596335January 1957February 201264%35.5 
593Elmdon52.4524− 1.7409910Junuary 1949April 1999100%49.6 
605Brize Norton51.758− 1.5764940January 1968February 201291%40.0 
613Benson51.6199− 1.0971236October 1974February 201286%32.0 
643Shawbury52.7943− 2.663296January 1957February 201283%45.8 
657Pershore52.148− 2.0397963November 1957February 201258%31.3 
708Heathrow51.4787− 0.4490463January 1949February 2012100%63.1SE
725Gatwick51.151− 0.1911184June 1958June 199879%31.7SE
775Manston51.3461.3371615January 1961February 201299%50.9SE
779Thorney Island50.8142− 0.9209820August 1958February 201268%36.3SE
842Hurn50.7789− 1.8348343January 1957February 201282%45.5SE
869South Farnborough51.2794− 0.7710789January 1957February 201256%31.0SE
886Lyneham51.5028− 1.9909465January 1968February 201294%41.6 
889Boscombe Down51.1613− 1.7531717January 1968February 201270%30.9 
971Abbotsinch55.8695− 4.42948145May 1966May 1999100%33.0NW
1006Prestwick55.5014− 4.5825549January 1957April 199888%36.2NW
1023Eskdalemuir55.3118− 3.2054582January 1957February 2012100%55.1 
1046Ronaldsway54.0851− 4.630716January 1957February 201272%39.8 
1070Carlisle54.9342− 2.9622382January 1961May 200871%33.5 
1135Ringway53.3559− 2.27949236January 1949November 2004100%55.8 
1145Valley53.2524− 4.5352426January 1957February 2012100%55.1 
1198Aberporth52.1391− 4.5699916January 1957February 201299%54.5 
1267Rhoose51.4001− 3.34284January 1957February 1998100%41.1 
1302Yeovilton51.0059− 2.6414859September 1964February 201278%37.0 
1336Plymouth50.3544− 4.1198650January 1949February 201298%62.2 
1346Chivenor51.0886− 4.147435January 1957February 201258%31.9 
1393Culdrose50.0838− 5.25609126January 1977February 12100%35.1 
1405St Mawgan50.438− 4.9965596January 1957December 200882%42.6 
1450Aldergrove54.6636− 6.2243625January 1949February 2012100%63.1NW
1572Guernsey49.432− 2.5989January 1960February 201260%31.3 
17314Leeming54.2968− 1.53145133October 1965February 2012100%46.3 
18974Tiree56.5− 6.8796117January 1957February 201299%54.8NW
19144London Weather Centre51.521− 0.1108852October 1974November 200990%31.6SE

The definition of a storm varies between disciplines, but where a measure of extreme wind-forcing is needed, a storm is usually defined as the exceedance of a specific wind speed for a specific duration (e.g. MacClenahan et al., 2001). For example, storm event frequency may be calculated as exceedance of a threshold wind speed for a specified duration, such as ≥ 20 kts (10.3 ms−1) for at least 5 hours (Carter and Stone, 1989). Alternatively, the winter gale day frequency (Dawson et al., 2002, 2004) expresses the number of days during a winter on which gale force wind speeds [≥34 knots (kts); 17.5 ms−1] are recorded.

In this study, winter wind climate was synthesized using a range of parameters to represent time-averaged winter forcing, extremes and ‘storminess’. First, a standard wind rose (directional frequency distribution) was calculated for specific wind direction quadrants (northeast, southeast, southwest, and northwest) for each winter. Second, the 50th and 99th percentile wind speed within each quadrant during each winter was calculated to provide a measure of average and extreme wind strength associated with direction. Third, the 50th and 99th percentile wind speed (irrespective of direction) was calculated to provide a measure of winter averages and extremes (Pirazzoli et al., 2010). Finally, event-scale storminess was explored using the standard winter gale day frequency (Donat et al., 2010) in addition to frequencies of a series of extreme wind speed events. From the possible speed-duration storm definitions (Carter and Stone, 1989; MacClenahan et al., 2001; Qian and Saunders, 2003; Atkinson, 2005; Weisse et al., 2005), we use a Beaufort scale-based scheme (cf. Sotillo et al., 2008). The occurrence of four types of event was investigated: (1) breeze sustained for at least a day (Force 4 = 11 kts = 5.7 ms−1, over at least 24 h); (2) strong breeze sustained for 12 h (Force 6 = 21 kts = 10.8 ms−1, over at least 12 h); (3) sustained high wind (Force 7 = 27 kts = 13.9 ms−1, over at least 6 h); and (4) sustained gale (Force 8 = 34 kts = 17.5 ms−1, over at least 3 h). Event frequency and 50th/99th percentile measures were also considered in terms of their westerly (vector mean wind direction from 180° to 360°N) and easterly (vector mean wind direction from 0° to 180°N) components (W/E). The final wind climate dataset thus comprised 33 annual winter measures [wind direction frequency (NE/SE/SW/NW); 50th and 99th percentile wind speed within each wind direction quadrant (NE50/SE50/SW50/NW50 and NE99/SE99/SW99/NW99); 50th and 99th percentile wind speed (P50 and P99); gale day frequency (GALE); storminess (E11k24h, E21k12h, E27k6h, and E34k3h)]. All wind climate measures were calculated for the same winter periods used in the NAO index calculations (six station-based indices, and two PC-based indices), giving 264 paired arrays for each station. The analysis calculated correlation coefficients (R, supported by p-values between all paired NAO index and wind climate measures. The correlations were examined to identify strong (|R | > 0.71) and significant (p < 0.01) associations, and the nature of these correlations was compared across different winter periods and wind climate measures. Correlations were also explored for spatial associations. Time series were summarized as winter-medians, the raw data for which were first normalized (standardized to a mean of 0 and standard deviation of 1 using the z-score) to allow synthesis across all sites.

3. Results

The definition of the NAO winter index strongly influences its correlation with locally observed wind climate variables. Overall, use of the DJFM station-based index period generated the greatest number of strong and significant (|R | > 0.71, p < 0.01) correlations, with the DJF period a close second (Table 2). Including October in the winter period significantly reduced the likelihood of a strong correlation, whereas including March increased the number of strong correlations (Figure 2(a); Table 2). The PC-based indices generated fewer strong correlations than the equivalent station-based winter indices, with the DJF period achieving more strong correlations than DJFM.

Figure 2.

(a) Synthesis of correlation coefficients between NAO winter index and wind climate (direction frequency and the 50th and 99th percentile wind speeds across the NE-SE-SW-NW) quadrants, for all sites, for each winter period. In-plot numbers show how many of the 53 stations exhibit strong and significant (|R | > 0.71, p < 0.01) correlations for each wind climate—NAO index pairing. (b) Geographical distribution of sites exhibiting strong and significant correlations: concentric circles reflect the different winter periods (small to large circles—DJF, DJFM, NDJF, NDJFM, ONDJF, ONDJFM station-based NAO; inner and outer squares—DJF, DJFM PC-based NAO).

Table 2. Abbreviations and descriptions of the wind climate measures used here, and a review of correlations with NAO indices
VariableDescriptionNumber of stations exhibiting a significant correlation
  Station-based indexaPCb
  DJFDJFMNDFJNDJFMONDJFONDJFMDJFDJFM
  1. a

    Data source: Climate Research Unit, University of East Anglia (www.cru.uea.ac.uk).

  2. b

    Data source: Climate and Global Dynamics, National Center for Atmospheric Research (www.cgd.ucar.edu).

NEWind frequency from 0 to 90°N4046414621281819
SEWind frequency from 90 to 180°N00000000
SWWind frequency from 180 to 270°N5050505038384141
NWWind frequency from 270 to 360°N00000000
NE5050th percentile wind speed from 0 to 90°N10000001
SE5050th percentile wind speed from 90 to 180°N00000000
SW5050th percentile wind speed from 180 to 270°N65430298
NW5050th percentile wind speed from 270 to 360°N00000000
NE9999th percentile wind speed from 0 to 90°N00000000
SE9999th percentile wind speed from 90 to 180°N00000000
SW9999th percentile wind speed from 180 to 270°N10000030
NW9999th percentile wind speed from 270 to 360°N00000010
GalesGale day frequency11000031
GalesWWesterly gale day frequency12220172
GalesEEasterly gale day frequency00000000
p5050th percentile wind speed22111032
p50WWesterly 50th percentile wind speed30100041
p50EEasterly 50th percentile wind speed00000000
p9999th percentile wind speed23110000
p99WWesterly 99th percentile wind speed66552174
p99EEasterly 99th percentile wind speed00000000
E11k24hFrequency of breeze, sustained wind events [≥11 kts (5.6 ms−1) for ≥ 24 h]10000000
E11k24hWFrequency of breeze, sustained wind events [≥11 kts (5.6 ms−1) for ≥ 24 h] from the west (180–360°)12115723139
E11k24hEFrequency of breeze, sustained wind events [≥11 kts (5.6 ms−1) for ≥ 24 h] from the east (−0–180°)12110020
E21k12hFrequency of moderate, sustained wind events [≥21 kts (10.8 ms−1) for ≥ 12 h]32120011
E21k12hWFrequency of moderate, sustained wind events [≥21 kts (10.8 ms−1) for ≥ 12 h] from the west (180–360°)65331155
E21k12hEFrequency of moderate, sustained wind events [≥21 kts (10.8 ms−1) for ≥ 12 h] from the east (−0–180°)00000000
E27k6hFrequency of strong, sustained wind events [≥27 kts (13.9 ms−1) for ≥ 6 h)22220044
E27k6hWFrequency of strong, sustained wind events [≥27 kts (13.9 ms−1) for ≥ 6 h] from the west (180–360°)34130255
E27k6hEFrequency of strong, sustained wind events [≥27 kts (13.9 ms−1) for ≥ 6 h] from the east (−0–180°)00000000
E34k3hFrequency of very strong, sustained wind events [≥34 kts (17.2 ms−1] for ≥ 3 h)00000011
E34k3hWFrequency of very strong, sustained wind events [≥34 kts (17.2 ms−1) for ≥ 3 h] from the west (180–360°)01000011
E34k3hEFrequency of very strong, sustained wind events [≥34 kts (17.2 ms−1) for ≥ 3 h] from the east (−0–180°)00000000
Total number of strong and significant correlations1411421181266576128105

The variable most strongly associated with the NAO winter index is wind frequency from the SW, and to a lesser extent frequency from the NE [winter wind rose (Figure 2)]. Frequencies of winds from the NW or SE exhibit no strong correlations with the NAO indices. Only three sites (Kinloss, Guernsey Airport and London Weather Centre) show no strong correlations with wind frequency from the SW: correlations at Kinloss are significant though when the PC-based NAO winter indices are used. Correlations between the NAO indices and SW frequency are positive, and are negative with NE frequency, consistent with the expectation that positive expressions of the NAO winter index are linked to increased occurrence of winds from the SW, and decreased occurrence of winds from the NE. The reduced correlation associated with October winter periods is most evident in sites to the north and west (Figure 2(b)), and this spatial divide is also apparent in the fact that the stronger correlations with PC-based NAO indices are primarily in the north and west, particularly in the case of NE frequency.

In comparison, wind strength within the four quadrants is not well correlated with the NAO winter indices. Nine sites show significant correlations between the median (50th percentile) wind speed in the SW quadrant and the PC-based DJF NAO index, and there are six stations with strong correlations with the station-based DJF NAO index. These sites cluster toward the north of the study region (Figure 2(b)). No sites show a strong correlation between the NAO and median wind speed in the SE and NW quadrants, and only Manston (southeast England) and Turnhouse (southeast Scotland) exhibit correlations in the NE quadrant (with DJFMpc and DJFi, respectively). Correlations with the 99th percentiles are weaker again, with significant correlations only evident at three sites, and only within the DJF winter period. These sites again cluster toward the north of the study region.

Examination of the time series for the collated wind rose confirms the strength of the association between NAO winter index and wind direction, as illustrated in the results for the DJFM period (Figure 3). The extreme negative NAO index in the winter of 1995 is matched with significant reduction in the frequency of winds from the SW and increase in those from the NE. The extended positive phase of winter NAO through the late 1980s and early 1990s, coincides with peaks in frequency of winds from the SW. There is no evidence of a secular trend in wind direction frequency within any quadrants, and the small fluctuations in frequency of winds from the NW and SE do not obviously correspond to variations in the NAO winter index. The strongest correlations between NAO winter index and frequency are associated with winds from the SW. This applies at 87% of sites, and the other 13% of sites are more strongly correlated with NE wind frequency. Median wind speeds in each quadrant are, in general, less well correlated with the NAO (Figure 3). Correlations with the NAO winter index are most evident with the SW median wind speed, which increases during the positive phases of the NAO (e.g. late 1980s/early 1990s) and decreases during the distinctly negative winters of 1968, 1995 and 2009. These three winters are characterized by peaks in the NE median wind speed, but several other winters exhibit stronger median wind speeds in this quadrant, such that the long-term correlation is reduced. Sites in the far northwest feature the strongest associations between NAO winter index and the SW median wind speed (0.51≤ | R | ≤0.8), and sites in the southeast are best associated with the NE median wind speed (0.34≤ | R | ≤0.66). In the entire dataset, only five sites show stronger correlations with the NW-SE quadrants, but there is no specific pattern in the geography of these sites.

Figure 3.

Time series of winter (DJFM) NAO station- (solid line) and PC- (dashed line) based indices, wind direction frequency (collated to NE-SE-SW-NW quadrants) and 50th percentile wind speeds (in each direction quadrant). Wind direction frequency across all stations is summarized as a collated winter distribution. Percentile wind speed data (measured in knots) from all stations are normalized (z-scores), and summarized here as a median with associated inter-quartile range. Ten sites from the far northwest and ten from the far southeast are also summarized with their respective medians.

Correlations between NAO winter index and extreme- and event-scale wind speed statistics are notably weaker than those for wind direction frequency (Figure 4(a)). Correlation coefficients are largely positive except for the easterly component of each measure (e.g. E11k24hE). This implies that positive phases of the NAO are associated with increased intensity and frequency of these extreme- and event-scale measures, and decreased intensity and frequency of easterly measures. But no strong correlations exist for the easterly component of the wind measures (p50E, p99E, galesE, E21k21E, E27k6hE, and E34k3E), except for the low magnitude, long duration event frequency (E11k24hE). Correlations improve in 69% of cases when only the westerly components are considered. The DJF and DJFM winter periods generate most of the significant correlations (in 70% of cases): the PC-based NAO indices are better correlated with frequency of gale days and the higher magnitude (E27k6h and E34k6h) events, whilst the station-based NAO indices are best correlated with the median wind speed and lower magnitude (E11k24h) event frequency. Most of the strong/significant correlations are associated with more northerly sites (Figure 4(b)), and strong correlations only exist in the south for the lower magnitude, longer duration (E11k24h and E21k12h) event frequencies. The higher speed threshold clearly separates the far north and west sites, but it is notable how few sites show any correlation at the lower speed thresholds. The 50th percentile wind speed is more strongly associated with the winter NAO index than the 99th percentile in at least 10% of cases; this increases to 23% when considering only the westerly component. Gale day frequency and frequency of extreme events (E34k3h) are the most weakly correlated ‘event’ types. Correlations improve significantly as magnitude (of the event measure) decreases and duration increases, indicating that the NAO is more closely associated with a time-averaged measure of wind climate than achievement of an extreme wind speed.

Figure 4.

(a) Synthesis of correlation coefficients between NAO winter index and wind climate measures across all sites, for each winter period. In-plot numbers show how many sites exhibit strong and significant (|R| > 0.71, p < 0.01) correlations for each wind climate—NAO index pairing. (b) Sites exhibiting strong and significant correlations: concentric circles reflect the different winter periods (small to large circles—DJF, DJFM, NDJF, NDJFM, ONDJF, ONDJFM station-based NAO; inner and outer squares—DJF, DJFM PC-based NAO). NB: no strong correlations exist for the wind measures p50E, p99E, galesE, E11k24h, E21k21E, E27k6hE and E34k3E, so these are not shown.

Weak event-scale correlations are evident in time series of westerly winter climate (Figure 5). For example, median westerly wind speed peaks in the mid-1960s, the late-1980s, the early-1900s and the late-2000s during positive peaks in the NAO index, but other peaks in the NAO do not stand out particularly in the wind speed record. The association is far less evident in the westerly 99th percentile wind speeds. Here, the 1995 NAO low does coincide with a dip in 99th percentile speed, but equivalent and more extreme minima in 99th percentile speed occur also in the winters of 1963 and 1984, which is not reflected in the NAO indices. The far northwest of the region tends to exhibit stronger correlations between wind speed and the DJFM indices than the southeast (westerly 99th percentile 0.18≤ | Rnorthwest | ≤0.76 vs 0.04≤ | Rsoutheast | ≤0.49; westerly 50th percentile 0.5≤ | Rnorthwest | ≤0.82 vs 0.02≤ | Rsoutheast | ≤0.65).

Figure 5.

Time series of winter (DJFM) NAO station- (solid line) and PC- (dashed line) based indices and selected wind climate measures. The data from all stations are normalized (z-scores) from the raw wind speed (in knots) and frequency (occurrences per winter) measures, and summarized here as a median with associated inter-quartile range. Ten sites from the far northwest and ten from the far southeast are also summarized with their respective medians.

The wind speed and event-scale measures appear to improve in their correlation with the DJFM indices post-2000 (Figure 5). This is primarily due to the strong positive peak in the NAO winter index 2006 and 2007 followed by the ‘record-breaking’ low (Osborn, 2011) in 2009, which are closely reflected in high wind speeds and frequencies, followed by significantly reduced wind speeds and frequencies. The previous strong shifts in NAO index (1966–1972 and 1994–1999) are represented by comparable shifts in the median wind speed and low magnitude event frequency, but not in the higher wind speed threshold measures.

Comparison of Belmullet, west Ireland (Figure 6) with Plymouth, southwest UK (Figure 7) provides a site-specific expression of the variations in correlation between the NAO winter index and recorded wind climate. Overall, the station-based index generates stronger correlations, but this is more evident in the Belmullet dataset (91%) than Plymouth (70%). Both sites exhibit strong and significant correlations between wind frequency from the SW (positive) and NE (negative) and the NAO winter index, and weaker correlations with median and extreme wind speeds in these quadrants (Tables 3 and 4). Belmullet exhibits strong correlations across a number of other wind climate measures, including gale day frequency, median wind speed, and event frequency at all put the highest speed threshold event type (E34k3h). Contrasting with this, correlations within the Plymouth wind record are far weaker, and no strong/significant correlations exist beyond the SW/NE wind direction frequencies. This lack of significant correlations is not a consequence of insufficient occurrence (Figure 7). Some of the main characteristics of the NAO winter index, such as the 1995 low, are evident in the Plymouth wind record (which shows a notable dip in frequencies and intensities), but there is far more variation here than can be explained by the NAO alone.

Figure 6.

Time series of winter (DJFM) NAO station- (solid line) and PC- (dashed line) based indices and wind climate for Belmullet, west Ireland. Wind speed percentiles are measured in knots, and event frequency is given as days per winter (gale days) or number of occurrences per winter (all other event types). Correlations are reported in Table 3.

Figure 7.

Time series of winter (DJFM) NAO station- (solid line) and PC- (dashed line) based indices and wind climate for Plymouth, southwest UK. Wind speed percentiles are measured in knots, and event frequency is given as days per winter (gale days) or number of occurrences per winter (all other event types). Correlations are reported in Table 4.

Table 3. Correlation coefficients describing the association between winter (DJFM) wind climate and NAO (station-based (i) and PC-based (pc) indices) at Belmullet, west Ireland (** p < 0.001; * p < 0.01; bold R2 > 0.5), as shown in Figure 6
 NAOiNAOpc NAOiNAOpc NAOiNAOpc
NE0.87**0.78**Gales0.71**0.68**E11k24h0.7**0.56**
SE0.47**0.38*GalesW0.73**0.7**E11k24hW0.83**0.7**
SW0.87**0.85**GalesE0.270.21E11k24hE− 0.08− 0.15
NW− 0.63**− 0.62**p990.64**0.6**E21k12h0.74**0.7**
NE50− 0.09− 0.09p99W0.64**0.61**E21k12hW0.76**0.73**
SE500.56**0.48**p99E0.42*0.35*E21k12hE0.150.09
SW500.69**0.64**p500.75**0.66**E27k6h0.78**0.76**
NW500.22− 0.03p50W0.79**0.67**E27k6hW0.81**0.79**
NE99− 0.07− 0.04p50E0.58**0.51**E27k6hE0.240.18
SE990.41*0.33   E34k3h0.68**0.7**
SW990.45**0.44*   E34k3hW0.66**0.69**
NW990.140.04   E34k3hE0.190.15
Table 4. Correlation coefficients describing the association between winter (DJFM) wind climate and NAO (station-based (i) and PC-based (pc) indices) at Plymouth, southwest UK (** p < 0.001; * p < 0.01; bold R2 > 0.5), as shown in Figure 7
 NAOiNAOpc NAOiNAOpc NAOiNAOpc
NE0.72**− 0.51**Gales0.42**0.26E11k24h0.30.17
SE− 0.34*− 0.38*GalesW0.45**0.29E11k24hW0.62**0.5**
SW0.74**0.6**GalesE− 0.02− 0.04E11k24hE− 0.61**− 0.6**
NW0.250.16p990.42**0.28E21k12h0.42**0.27
NE50− 0.45**− 0.51**p99W0.280.2E21k12hW0.48**0.35*
SE50− 0.19− 0.36*p99E0.06− 0.08E21k12hE− 0.23− 0.36*
SW500.210.06p500.280.06E27k6h0.48**0.37*
NW500.41**0.3p50W0.51**0.34*E27k6hW0.51**0.41*
NE99− 0.48**− 0.59**p50E− 0.34*− 0.41*E27k6hE− 0.16− 0.22
SE990.05− 0.02   E34k3h0.43**0.34*
SW990.250.15   E34k3hW0.45**0.36*
NW990.170.06   E34k3hE0.030.05

4. Discussion

The results presented here confirm that the NAO winter index is a robust measure of average winter climate, as expressed in the wind direction frequency distribution, and to a less extent, average wind speed. In positive phases of the NAO, the increased pressure gradient across the northeast Atlantic strengthens the eastward movement of weather systems into mid-latitude northwest Europe. This is manifest in the wind record as an increase in the proportion and winter-average intensity of southwesterly and westerly winds. This correspondence can be partly attributed to the fact that both the NAO index, wind direction frequency distribution and median wind speed are integrated measures. However, the analysis also shows that use of the NAO winter index as a proxy for ‘storminess’ or extremes in wind climate is problematic. Although, the increased pressure gradient is ordinarily linked to the northeast Atlantic storm-track, and hence an increase in storminess, this is not regionally evident in the wind record in northwest Europe. The weaker link between NAO and storminess may again be linked to the scale and calculation of the measures used. As Clarke and Rendell (2009) note, it is inherently difficult to relate synoptic scale wind climate characteristics to an average pressure gradient index covering several months. The results presented here may similarly reflect the extent to which the datasets are matched in terms of their temporal scale and resolution.

There is strong evidence from previous studies that the latitudinal geography of North Atlantic storm tracks is well expressed by the NAO (Rogers, 1997; Lozano et al., 2004), which would imply a further association with the wind climate characteristics of these storms and, in particular, the occurrence of higher than average wind speeds. This reasoning then invites the interpretation of NAO indices as proxies for aspects of wind climate known to force other geophysical processes (Clarke and Rendell, 2009). Several studies have made connections between the NAO winter index and storm frequency, and the increase in storminess that occurred over the 1980s to 1990s was regularly attributed to a sustained positive phase in the NAO winter index (Rogers, 1997; Dickson et al., 2000). The association is also apparent across the first half of the 20th century, when the NAO winter index experienced two sustained (cf. 15 years) positive phases and wind records suggested stronger and more frequent high wind speed events (e.g. Pye and Neal, 1994). Burningham (2005) showed how the timing of storms impacting northwest Ireland during the 19th and early 20th centuries was clustered during periods of sustained positive NAO winter index.

Whilst linkages do exist between the winter NAO and wind climate in northwest Europe, the nature of this association is clearly more complicated than is generally assumed. Dawson et al. (2004) noted that the positive correlation between the NAO and storminess in the 1980s and 1990s could not be reproduced in the late 19th century. In their analysis of extreme wind speeds (99th percentile) across northwest Europe between 1949 and 2008, Pirazzoli et al. (2010) found that considerable decadal variability caused significant shifts in the association between wind speed, direction and the NAO index. Their findings are repeated here, in that temporal and spatial variations in correlation between the NAO and wind climate are evident across the UK and Ireland. Whilst the frequency of winds from the SW and NE consistently exhibit strong correlations with the NAO index, the association with extremes and discrete events in the wind climate is only apparent at northwest sites, and largely missing from the south and east of the region (Figures 2(b) and 4(b)). This does not appear to be simply a reflection of the reduced frequency of such events in the south and east, and detailed time series from such sites shows that sufficient data exist (Figures 5 and 7).

The winter period used to derive the NAO index has an important effect on the correlations that emerge between it and wind climate. Selection of a longer winter period reduces the number and strength of strong and significant correlations. In our analysis, the DJFM station-based index yielded the largest number of strong and significant correlations, and the highest mean correlation overall, and the DJF period a very close second (Table 2). The DJF PC-based index performs better than DJFM, and better than all other station-based index winter periods. Inclusion of October in the winter index period is particularly problematic. The 5-month NDJFM period generated over twice as many strong and significant correlations than the same duration ONDJF period. This may also reflect the importance of including March: there were 20% more strong and significant correlations in the DJFM winter than the NDJF period. All of the preceding results appear to be regionally coherent, except that the PC-based index is more strongly associated with wind climate at north and west sites (Figures 2(b) and 4(b)). It is notable that most studies consider only a single winter definition, often with little by way of justification [e.g. Lehmann et al., 2011)]. Where provided, the rationale varies. For example Pirazzoli et al. (2010) used the full ONDJFM winter due to an appreciation that peak winter wind speeds often occur in October, whilst Lozano et al. (2004) show that when relating monthly wind speeds to the NAO, January to March often provide the stronger correlation. Our study suggests that the results reported by previous authors might be significantly altered by the use of a different winter period.

Whilst our work is primarily concerned with interactions between winter wind climate, storminess and the NAO, the analyses presented also invite consideration of trends in these linkages. As previously noted, there is no clear evidence of any long-term trend in wind direction frequency. However, the aggregated time series suggest stronger associations with the NAO winter index may exist over the shorter term (for example the last decade), but the reduction in data associated with these smaller time frames presents difficulties when statistically identifying significant strong correlations. Pirazzoli et al. (2010) document a stronger correlation between NAO index and wind speed between 1976 and 1992 for sites north of 54°N. Associations between wind climate and the NAO are thus both spatially and temporally variable, and further analysis in this area is warranted.

The analysis presented here has followed a relatively conventional approach for identifying event-scale ‘storms’ in the wind record that requires exceedance of a given wind speed threshold for a specified duration. Various speed and duration thresholds are used within the literature, and further constraints on the analysis may be applied when particular physical processes are of interest. For example, Carter and Stone (1989) examine strong onshore winds (>25 kts) associated with extreme spring tides. The criteria used are important since, as we show, stronger associations with the NAO index are generally only achieved with longer duration and lower speed thresholds, and by separately resolving the westerly events. Further work on the wind speed and duration thresholds associated with specific physical processes that force the dynamics and longer term behaviour of physical systems would be desirable. This could lead to the derivation of more formal criteria for the identification of storms or high-energy wind events within the wind record. This would permit more robust analysis of changing storminess in the northeast Atlantic and its relation to regional climate teleconnections.

5. Conclusions

  1. The NAO winter index is strongly associated with observed wind direction frequency across a broad region of the northeast Atlantic represented by 53 stations in the UK and Ireland (strong significant correlations at 96% of sites). Wind frequency from the SW is positively correlated with the NAO winter index, which is then negatively correlated with frequency of winds from the NE. Wind direction frequency from the NW and SE are not correlated with the winter NAO.

  2. Average winter wind speed is weakly correlated with the NAO winter index, and the few strong correlations that exist (26% of sites) only apply to winds from the SW or those with a westerly component. Extreme wind speed is less well correlated (11% of sites).

  3. Evidence for any association between the NAO winter index and the frequency of event-scale storm measures derived from the wind record is very weak. Those positive correlations that do exist are confined to the far north and west of the study region, and are associated with longer duration and lower speed event thresholds. This finding is not simply a consequence of a lack of storm events elsewhere, but reflects significant temporal variations in storminess that cannot be explained by the NAO index.

  4. The winter period used has a major influence on the strength and significance of correlations between wind climate and NAO winter index. The inclusion of October in the winter period significantly weakens these correlations, whilst the inclusion of March improves them.

Acknowledgements

We are grateful to the British Atmospheric Data Centre (www.badc.ac.uk), Climate Research Unit, University of East Anglia, UK (www.cru.uea.ac.uk), Climate Data Guide at the National Center for Atmospheric Research (www.cgd.ucar.edu) and Met Éireann, the Irish National Meteorological Service (www.met.ie) for access to data. We are also grateful to Ricardo Trigo and an anonymous reviewer for their constructive comments and suggestions on an earlier draft of this manuscript.

Ancillary