Importance of location for describing typical and extreme wind speed behavior

Authors


Abstract

[1] Several recent studies have considered the potential impact of climate change on regional wind intensity. However, previous wind speed studies in the Pacific Northwest (PNW) present conflicting results for wind speed trends in relation to climate drivers. This study analyzes the percentiles (50th, 75th, and 95th) of the strongly positively skewed distributions for PNW maximum daily wind speeds from 92 meteorological stations, and reveals different behaviors for average and extreme wind speeds. Considerably stronger winds are found at coastal locations compared with sites further inland. Extreme wind speeds at these coastal locations appear to follow an eight to nine-year cyclic pattern, while mainland sites have a small, linear downward wind speed trend. This finding of a behavioral dependence on location helps reconcile previous, apparently contradictory results and has important global implications for wind research and infrastructure planning, such as wind energy feasibility studies and air quality management activities.

1. Introduction

[2] Society's vulnerability to extreme weather events can change due to shifting weather patterns, especially when infrastructure development does not consider potential damages associated with these events. Non-monetary costs can be even greater with loss of life or damage to irreplaceable cultural or natural places [BC Hydro, 2007; Costanza and Farley, 2007; Vancouver Board of Parks and Recreation, 2007].

[3] Several studies have examined changes in wind speeds in the Pacific Northwest (PNW) and North America, but these studies come to differing conclusions regarding the trends and relationships to climate parameters with results appearing to depend on the type and amount of data used for analysis [Abeysirigunawardena et al., 2009; Enloe et al., 2004; Gower, 2002; Klink, 1999, 2002; Pryor et al., 2009; Pryor and Ledolter, 2010; Tuller, 2004; Wan et al. 2010]. Further, few studies have tried to conduct a comprehensive investigation examining all available observed wind speed data for a particular area.

[4] Here we explore the variability and trends of historical wind behavior in coastal and adjacent mainland areas of the PNW. Our approach differs from previous attempts to analyze changes in wind behavior in the PNW in that we (a) include all stations at which wind speed data have been recorded since 1950; (b) account for the asymmetric behavior of wind speed data by examining percentiles; and (c) use a hierarchical cluster analysis to tease apart regionally-dependent patterns in wind behavior.

2. Background

[5] Several studies assess the trends and variability of wind speeds for North America and specifically the PNW, but their findings are inconsistent. Studies examining wind behavior over the contiguous United States have demonstrated that wind speeds exhibit very different trends depending on the dataset used [Pryor et al., 2009; Pryor and Ledolter, 2010]. Observational data from these studies show wind speeds exhibit consistent negative temporal trends for the 50th and 90th percentiles and annual mean from the 1970's to early 2000's (time periods vary for each data set and type included in the study), across the entire U.S. Alternatively, data from re-analysis datasets and regional climate models (i.e., NARR, ERA-40, & RSM) demonstrate the opposite trend and do not agree with the observational data, a feature also reported for Australia [McVicar et al., 2008] and Tibet [You et al., 2010]. Two other studies of continental U.S. wind speeds support a declining trend in PNW observational data [Klink, 1999, 2002] as well as a Canadian national study [Wan et al., 2010]. This declining trend or ‘stilling phenomenon’ is also found in many other studies throughout the world, though primarily at inland locations [St. George and Wolfe, 2009; Roderick et al., 2007; Xu et al., 2006]. However, Pryor et al. [2009] point out that time series of in-situ wind speed measurements are typically highly discontinuous and exhibit large heterogeneities, making detection, quantification, and attribution of temporal wind speed trends difficult, especially as each of these national studies use relatively few monitoring station records for a given region.

[6] Regional studies have also suggested differing trends depending on the station. In the northeast Pacific Ocean, monthly mean wind speeds calculated for most of the 25 ocean buoys included in the work of Gower [2002] experience increasing trends from 1972 to 1999, but some buoys also indicate negative trends. Of the buoys with records longer than 20 years, three show positive and three indicate negative trends, though the positive trends are larger and more are statistically significant (at a p-value of 0.05). In urban areas of British Columbia, Tuller [2004] found a general, declining trend in mean annual and winter winds from the late 1940's and 1950's to mid-1990's for monitoring stations in south-coastal B.C. (Vancouver International Airport, Victoria International Airport, Comox Airport, and Cape St. James). However, Tuller [2004] also noted that over shorter temporal periods (i.e., approximately 10–15 years) stations have experienced distinct negative and positive trends that differ from each other and the long-term patterns (Pirazzoli and Tomasin [2003] found a similar pattern for Mediterranean coastal areas). Furthermore, the mean annual and winter wind speeds for one station, Comox Airport, deviate strongly from the decreasing trends and correlations determined for the other three stations.

[7] Instead of trends, some studies have found relationships between extreme wind events and climate oscillations such as the El-Niño Southern Oscillation (ENSO). For example, Enloe et al. [2004] found that the PNW experiences an overall increase in the magnitude of monthly mean peak wind gusts during the winter months of November to March in the cold, La Niña phase of ENSO, as compared to neutral phase years. The PNW experiences decreased gustiness during winter months in warm phase years, but fewer stations exhibit significant changes and the changes are smaller in magnitude. Similarly, Abeysirigunawardena et al. [2009] also found a significant relationship between extreme winds and ENSO, with extreme wind events occurring more frequently in Delta, B.C. during cold (i.e., La Niña) phases over the past 53 years. El Niño phases have also been associated with decreased surface wind speeds across the southern Canadian prairies [St. George and Wolfe, 2009].

[8] Collectively, these studies illustrate the difficulty in obtaining consistent results for wind speed trends in the PNW. Some studies suggest differing trends, while others suggest that winds experience oscillatory climate behavior, but an overview of wind behavior for the entire region is difficult because results vary depending on record length, location, scale, type, and statistical metric used.

3. Methods

[9] Available historical meteorological data from 179 meteorological stations for the years 1950 to 2008 were provided by Environment Canada and the U.S. National Climatic Data Center for the study area of the Pacific Northwest of North America (45–52°N, 129–122°W). To reduce sampling bias at each monitoring station, a day, month, or year of data was only included if it met the following completeness criteria:

[10] 1. Day – at least 90% of 3-hourly measurements per day (i.e., 7.2 measurements per day);

[11] 2. Month – at least 90% of days per month (e.g., January – 27.9 measurements); and

[12] 3. Year – at least 3 years with valid months and days.

[13] Stations were removed if clear discrepancies were detected through visual inspection.

[14] Additional standardizing was necessary due to varying monitoring frequencies over time, within and across stations. Many stations considered valid had either 1-hourly or 3-hourly sampling rates, which created large differences in the number of data points per station. Therefore, for each day of valid data, we calculated and retained only the maximum daily wind speed taken as the largest hourly value from a given 24-hour day regardless of sampling rate.

[15] In order to isolate interannual variability and to avoid spurious correlations induced across stations due to regional seasonality, we removed the seasonal component from each station time series using the deseasonalizing algorithm stl in R [R Development Core Team, 2009]. Since deseasonalizing requires a complete time series with no missing values, monthly averages were filled into any missing percentile values (e.g., the average across all March values for a given station time series filled any missing March values for that station). To ensure all months could be averaged, only time series with at least 10% of monthly values present were included for deseasonalizing. This data-augmentation method imparted as little extraneous information as possible to the non-seasonal components of the data while still accomplishing the deseasonalizing. Though we could not test the extent to which the added monthly average values altered the outcome of the deseasonalizing algorithm, we feel confident the algorithm is interpreting added-in data as seasonal-only information because the sections of time series with in-filled data show information mostly in the seasonal component and very little in trend or residual components (see Figure S1 of the auxiliary material).

[16] The completeness criteria for sampling bias reduced the initial 179 stations to 92 stations considered valid for our study. Many (58 of 92) stations do not have detailed secondary data, such as specific location relative to other structures or topographic features, type of anemometer, or height above ground. Consequently, no attempt was made to correct wind speed observations or standardize height (the World Meteorological Organization has set the standard height at 10 m above ground). Debate exists about the impact of correcting wind speed time series for measurement heights. Some authors suggest the effect is likely relatively small (i.e., <20%) for most stations [Pryor et al., 2009], while others seem to demonstrate that trend results can be considerably affected [Wan et al., 2010]. Since our goal was to capture regional wind patterns as comprehensively as possible, we relied on the large number of stations and observations to compensate for erroneous records.

[17] Wind speed frequency distributions display strong positive skew and exhibit heterogeneous variance across the magnitudes of observations. Thus, focusing solely on mean values may underestimate, overestimate, or fail to distinguish real nonzero changes in response-explanatory relationships [Cade and Noon, 2003]. As a result, we rely on summary percentiles rather than a single mean value to provide a more robust statistical description of wind speed distributions [Koenker, 2005]. Percentiles also help compensate for the shortcomings of meteorological station data, which often include short and intermittent time-series records, by smoothing the inherently spiky nature of wind speed data (i.e., sudden large changes in wind speed). Because the wind speed response-explanatory variable relationship may change across the range of observed values, we calculate the 50th, 75th, and 95th monthly percentile values from the maximum daily wind speeds. The 50th percentile (median values) represents typical wind speeds and can be expected to occur relatively commonly. Stronger wind speeds, represented by the 75th and 95th percentiles, will be less frequent but may have much more important impacts (e.g., cause extensive damage to infrastructure or ecosystem features).

[18] Finally, we conduct a hierarchical clustering analysis [Shumway and Stoffer, 2006; R Development Core Team, 2009] to determine commonalities among the wind speed time series, and then group them according to the shared features.

4. Results

[19] Hierarchical clustering of the 95th percentile wind speed time series produces five broad groups, which sort the stations into coast and mainland areas (Figure 1). Two groups (n = 37) generally represent coast monitoring locations that are exposed sites at low elevation. Two other groups (n = 53) characterize mainland locations which are generally further from the coastline. Hierarchical clustering for the 50th and 75th percentile wind speeds produces groups similar to those observed for the 95th percentile, but not as pronounced.

Figure 1.

Study area and sample locations – coastal (blue) and mainland (orange) groups do not appear to have any unidentified spatial patterns. Stations in each group are not correlated with record length or temporal period. Sartine and Solander Islands (green) represent a separate group from the hierarchical clustering and experience distinctly higher wind speeds than other stations.

[20] We have averaged the 50th, 75th, and 95th percentile maximum daily wind speeds across stations for the coast and mainland groups, respectively (Figure 2). Each of the percentiles show generally higher speeds and increased variability for coast relative to mainland sites. While mainland monitoring locations appear to have a downward wind speed trend (approximately −0.025 m/s/year for the 50th percentile to −0.034 m/s/year for the 95th percentile), it is unclear if the coast sites have an upward trend or are essentially stationary around a fixed mean. Percentiles for mean daily winds reveal similar patterns (Figure S2 of the auxiliary material).

Figure 2.

Variability of coast and mainland monitoring stations (1950–2008) – Within each panel, the coast (upper) and mainland (lower) group time series represent maximum daily wind speed percentiles averaged across all stations within the respective group. Average (solid lines) and one standard deviation (dashed lines) shown for each time series.

[21] Coast wind speeds appear to follow a cyclic pattern with an approximate period of eight to nine years, with a 3–5 m/s difference between peaks and troughs. This cyclicity is confirmed by a Fourier transform analysis [R Development Core Team, 2009] of the 95th percentile coast-time series, which shows peaks at 8 and 9.3 years (a 28-year period cycle may also be present, but total data record length is too short to put much confidence in this result). Peaks in wind speed averaged across hierarchical clustering groups can be seen at roughly 1960, 1969, 1977, 1986, and possibly 1992 and 1999, with troughs visible between each peak. A similar but less distinct pattern is apparent in the lower wind speed percentiles as well. This cyclic pattern is not well characterized by any of the indices representing Pacific Ocean climate influences on the PNW: Pacific Decadal Oscillation – period length of 15–25 & 50–70 years [Mantua and Hare, 2002], Arctic Oscillation – 1–2 years [Thompson and Wallace, 1998], Pacific/North American Pattern – 1–5 years [Wallace and Gutzler, 1981], El Niño-Southern Oscillation – 1–3 years [Rasmusson and Wallace, 1983], and Northern Oscillation Index – ∼14 years [Schwing et al., 2002]. The cyclic pattern in observed wind speeds in the PNW may be the result of an unidentified climate oscillation and/or an interaction which modifies behavior between current indices.

5. Discussion

[22] Wind speed behavior in the PNW appears to differ strongly by proximity to the coast (Figure 1). The differing relationships between wind and geographic location (coast or mainland) appear to reconcile conflicting results from previous wind studies. Several studies find declining trends for wind speeds over mainland areas throughout North America [Wan et al., 2010; Klink, 1999, 2002; Pryor et al., 2009; Pryor and Ledolter, 2010], while others indicate contradictory declining and increasing patterns near coastlines in the PNW and Mediterranean [Enloe et al., 2004; Tuller, 2004; Pirazzoli and Tomasin, 2003; Gower, 2002]. By separating wind regimes into coast-cyclic (both increasing and decreasing periods) and mainland-declining, discrepancies between studies can be resolved. For mainland sites, recent results for two mountainous regions (central China and Switzerland) have also shown that station elevation, as well as geography, influences wind speed trends [McVicar et al., 2010].

[23] Pryor et al. [2009] and Klink [1999] identify predominantly declining trends of 0.5–1.0% per year for the PNW. These results appear confined mostly to mainland monitoring locations. Our finding of a mainland declining trend of 0.3–0.5% per year is close to this range and agrees with the continental results. Links between declining wind speeds and warm El-Niño phases [Abeysirigunawardena et al., 2009; St. George and Wolfe, 2009] would not explain this steady, long-term trend occurring since 1950 for mainland stations.

[24] Interestingly, our results appear to reconcile a problem from Tuller's [2004] study focusing on the PNW, where Vancouver and Victoria International Airports (mainland locations) exhibit declining trends, while Comox Airport (a coastal site) seems to follow both increasing and declining trends. Representing Comox Airport with a cyclic pattern could explain that station's deviating behavior. Gower [2002] reports a similar pattern of contradictory trends for observation buoys off the Washington State coast; considered coastal, and therefore cyclic, locations in our study.

[25] All of the variations in wind speed distributions: decreasing trend for mainland locations and greater variability and cyclic pattern for coast sites, make planning infrastructure, ecosystem, and emergency response activities more difficult. For example, in urban settings, studies relating air quality measurements to wind trends may need to re-evaluate the monitoring locations used to represent regions because of differing patterns for coast and mainland areas. Vancouver and the Fraser Valley are an example of an urban/rural region where light winds can exacerbate the build-up of ground-level ozone and degrade air quality [Vingarzan and Thomson, 2004]. Future land use and public health decisions will need to consider weaker winds and geographic differences in wind behavior. Airborne pollutant studies should be careful when choosing representative locations for air quality studies.

[26] Cyclic and downward trending wind speed behavior is also of great importance to energy developers, who are increasingly considering wind farms as viable economic options that do not emit carbon dioxide. Wind power generation is increasing throughout the PNW, with many projects proposed for coastal areas [BC Hydro, 2007]. Fluctuations in wind speed, and therefore electricity generated, will impact power utilities. While the cyclic pattern is most apparent in the 95th percentile wind speeds at coastal sites, energy generated by wind turbines is proportional to the cube of wind speed, and strong winds are required to generate power. Furthermore, many wind turbines have a life span of only two to three decades, in which they must pay for themselves and generate a profit [Hau, 2006, pp. 754–759]. If power providers conduct feasibility studies during peak times of the wind speed cycle, the amount of power generated over the lifetime of a project may not meet the expectations of owners or utilities requiring electricity. For example, a 3 m/s decrease in the 95th percentile wind speed for coast stations (the change from peak to trough of the cycle) would result in a ∼49% decrease in power produced (Figure 2).

6. Conclusions

[27] Our finding of a cyclic pattern for coast areas and declining wind speeds for mainland locations in the PNW helps explain disparities between previous studies, and emphasizes the importance of regional heterogeneities and location when considering trends in typical and extreme wind behavior. While we have identified this pattern for the PNW, proximity to coastal areas could help explain studies in other climatic areas around the globe and is consistent with the ‘stilling phenomenon’ identified for many mainland locations. Future studies of wind speeds near coastal areas should include the relative locations of monitoring stations as an explanatory variable to help with interpretation of possibly conflicting results. Resource managers should explicitly consider their project locations and how wind patterns may change in the future as this may have strong impacts on sectors such as health and energy development.

Acknowledgments

[28] We thank two anonymous reviewers and P. Whitfield for helpful suggestions. This research was supported by Climate Change Impacts Research Consortium funded through the SFU Community Trust for Endowment Fund. Researchers were funded by the Michael Geller Foundation, the Planning Institute of BC, and the NSERC PGS program (to B.J.G.) and by the NSERC Discovery, CRC and CFI grant programs (to K.E.K.).