1. Introduction
a. Wind energy in New Zealand
New Zealand’s wind industry has grown rapidly since 1993 when the first single-turbine wind farm was employed. Today, 17 operational wind farms are spread throughout New Zealand with a combined installed capacity of 682 MW (NZWEA 2013). From the latest available data (MED 2013), the majority of electricity in New Zealand is generated by renewables (73%), yet, despite recent growth, generation from wind still composes a relatively modest 5% of total electricity generation. Of the other renewable sources, hydropower is dominant in contributing 53% of the total electricity generation, followed by geothermal at 14%. The remaining electricity (27%) is predominantly generated by fossil sources (natural gas and coal). A number of wind energy proponents have argued that New Zealand’s wind resources have been underexploited, despite being among the best in the world (e.g., Pretli 2003; Sovacool and Watts 2009; Schaefer et al. 2012). For similar reasons, other commentators have predicted future substantial growth in the industry over the coming decades (e.g., Mason et al. 2010; Kelly 2011; NZWEA 2013). For example, Mason et al. (2010) carried out a comprehensive examination of whether a 100% renewable electricity system for New Zealand would be feasible that was based on model simulations of certain criteria including meeting energy demand requirements at all times, maintaining hydro-lake levels, and minimizing surplus “spillage” generation. The authors concluded that such a system is feasible with wind supplying 22%–25% of generation. Further growth in the industry is also in line with the present government targets of achieving 90% of electricity generation from renewables by 2025 (Cullen et al. 2012).
b. Climatological setting
The quantity of New Zealand’s wind resources can be attributed to the nation’s midlatitudinal position within the Southern Hemispheric atmospheric circulation, known as the “roaring forties.” The atmospheric circulation of this region is characterized by a belt of westerly winds situated below semipermanent high pressure cells to the north. Seasonal variability in the latitudinal position of these features relates to seasonal variability in windiness, with strong westerly winds often most frequent in spring as pressure gradients tighten and become more constrained over much of the country (Sturman and Wanner 2001). In terms of spatial variability, New Zealand’s large latitudinal span means that there is often variability in wind fields between regions. This comes about as certain regions are often exposed to different features of the circulation associated with the general west-to-east progression of synoptic systems over daily time scales, yet the extent of this has not been formally examined in the literature.
Terrain complexity and proximity to the coast can also influence regional variability in wind fields. Notably, New Zealand’s long southwest–northeast-aligned mountain chains act to disrupt and modify various features of the prevailing westerly flow (McCauley and Sturman 1999; Kossmann and Sturman 2004). Other flow-channeling features of the New Zealand landscape (including Cook Strait and the Manawatu Gorge in the CKS and MWT regions that are defined and shown in Fig. 1) act to positively affect regional wind resources (Reid 2005). Operating at the mesoscale, sea breezes are experienced in all coastal regions of New Zealand (Sturman and Tapper 2006) and have been shown to produce regionally complicated wind fields in certain geographical settings (e.g., Sturman and Tyson 1981) with consequences for wind resources (e.g., Gibson and Cullen 2015).
Map showing regions (STH, CTY, CKS, MWT, CNI, NTH) where wind farms (15 in total) are located. For reference, STH includes Southland and Otago, CTY is Canterbury, CKS is Cook Strait, MWT is Manawatu and Wanganui, CNI is central North Island and Hawkes Bay, and NTH is for coastal parts of Waikato, Auckland, Coromandel, and Northland. [Map taken directly from Turner et al. (2011).]
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
c. Wind resource climatology
Despite the inherent link between wind resources and meteorology and climatology, research explicitly characterizing or assessing wind resources within this context remains in its infancy (Brayshaw et al. 2011). Many studies that have examined wind resource variability within a meteorological or climatological framework have done so in relation to how low-frequency climate variability or climate change might impact wind resources (Greene et al. 2010). Another growing area of interest within the wind energy research community relates to the idea that the geographic distribution of multiple electrically interconnected wind farms can help mitigate issues of intermittency (e.g., Archer and Jacobson 2003, 2007; Kempton et al. 2010; Fisher et al. 2013). Since an often-cited barrier to large-scale wind energy development relates to intermittency (Fisher et al. 2013), the idea that the stochastic nature of wind on multiple time scales acts as a risk to supply, this research is of direct practical relevance. The concept underlying geographically distributed wind farms is simple. Provided that the wind is not calm everywhere across the grid, power will be produced somewhere, with the benefits to supply reliability increasing as the correlations of wind speed between multiple wind farms decrease (Kempton et al. 2010). In this area of research, many studies have reported promising results with regard to reducing or eliminating wind speeds too low for generation (Archer and Jacobson 2003), increasing opportunities for baseload supply by wind energy (Archer and Jacobson 2007), and reducing the wind power standard deviation or fluctuations in capacity factor across the network (Archer and Jacobson 2007; Kempton et al. 2010). However, few studies have attempted to understand causal mechanisms of the benefits of interconnecting wind farms from a climatological or meteorological perspective. Those that have done so have focused on examining synoptic weather patterns or maps over selected periods of time to act as case studies (e.g., Oswald et al. 2008; Kempton et al. 2010; Leahy and Foley 2012), as opposed to giving consideration to the entire distribution of wind speed or power generation.
Interconnection through high-voltage direct current (HVDC) technology is increasingly being deployed and would be practical between wind farms spanning the length of New Zealand. However, to the knowledge of the authors, examination of the potential for electrically interconnecting wind farms within the context of generation reliability has not been carried out in New Zealand, which is the focus of this paper. Our examination is also unique in that it is carried out primarily within a climatological framework giving focus to how weather patterns influence the entire distribution of generation differently between wind farms spread between regions. The structure of the paper is as follows. First, we consider the similarity between wind farms in different geographic regions in terms of wind speed correlations. Second, through the use of a weather-typing procedure, we examine and contrast synoptic-scale links to power generation measures for different wind warms. Finally, we discuss the benefits of wind farm interconnection in terms of supply reliability.
New Zealand provides an excellent case study for such an examination because 1) the wind resource of New Zealand is considered among the best in the world with further industry growth expected over coming decades, 2) the latitudinal span of the country likely results in regions being exposed to different features of the synoptic-scale circulation at any given time, and 3) the mountainous and coastal terrain likely contribute to the spatial variability of wind climates between regions.
2. Data and methods
a. Data
The use of existing publicly available wind data is often not reliable for characterizing wind resources, largely because measurements are either made too close to the ground (10 m) or in population-dense areas atypical of where commercial wind farms are situated. Conversely, when wind data from wind farm locations have been made available by wind farm operators, often the record length is of shorter duration. As such, for many purposes, “synthetic” wind datasets can provide an attractive alternative. In this study, a model-derived synthetic wind speed dataset was used (Turner et al. 2011), comprising 5 yr of hourly data for 15 wind farms (actual or proposed) that collectively span the length of New Zealand (Fig. 1). The methodology used to construct the dataset was also designed in such a way to respect the commercial secrecy of wind farm operators, and, as such, the precise locations of wind farms were not indicated. The synthetic wind dataset was generated from a statistical–numerical weather prediction method, with the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) run at 12-km resolution. Statistical relationships between observational wind farm wind speed data and modeled wind speed data were developed at each wind farm location for the available overlapping period, and subsequently used to fill the complete 5-yr synthetic dataset. The statistical approach adopted was designed to include the influence of subgrid-scale processes (e.g., wind speed up from channeling), preserve temporal realism (e.g., the influence of synoptic-scale systems on wind speed variability), preserve spatial realism (e.g., the covariability of wind speeds between wind farms in different regions), and reproduce the observed diurnal cycle of wind speeds. Given these methodological considerations, the dataset is appropriate for the purpose of this study, which investigates the synoptic controls on spatial–temporal variability in the wind resource across New Zealand (R. Turner 2014, personal communication). A more detailed description of the methodology and validation of the synthetic dataset is provided by Turner et al. (2011).
A weather-typing approach previously developed for the New Zealand region (Kidson 2000) was employed in this study to characterize the synoptic-scale atmospheric circulation over the New Zealand region. The classification algorithm implements a combined principal component analysis (PCA) and k-means cluster analysis procedure using reanalysis data from the 1000-hPa geopotential height field of the NCEP–NCAR dataset (Kalnay et al. 1996) to produce a set of 12 weather types (“Kidson types”) (Fig. 2) in 12-hourly time series. Further details of the classification algorithm are given by Kidson (2000).
Maps of 1000-hPa geopotential height anomalies associated with each of the 12 weather types (Kidson types). Names of the types, as they are referred to in the text, are indicated in the top right of each panel and respectively correspond to trough, southwesterly (flow), trough-northwesterly, trough-southwesterly, high, high to the northwest, westerly, high to the southeast, high to the east, northeasterly, high to the west, and ridge. [Figure taken directly from Ackerley et al. (2011), as are the definitions.]
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
b. Statistical methods
As stated earlier, the benefit of aggregating wind farms relates to the degree of (or lack of) correlation between wind farms. Therefore, we first calculated correlation coefficients between wind farms from hourly wind speeds using the nonparametric Spearman’s rank correlation. This method uses ranks so that the correlation can be extended to provide a more robust measure of similarity between variables with nonlinear relationships. However, further sensitivity testing between Pearson and Spearman correlation methods revealed differences in the correlation coefficient of <0.02 at all wind farms. Wind farms were then ordered based on a measure of similarity introduced in Friendly (2002). This measure is calculated from the “angular order” of the largest two eigenvectors of the correlation matrix.
Descriptive statistics were calculated to initially explore how different weather types relate to wind resource variability, and how these relationships differ between regions. Specifically, for each weather type, the percentage of corresponding time (hours) spent in a particular wind speed “class” was compared between wind farms. The particular thresholds for the wind speed classes were chosen in consideration of actual wind turbine power curves (power as a function of wind speed for a particular turbine model). While there is some degree of variability in power curves between turbine models (Fig. 3), for the purpose of this analysis we use the thresholds shown in Table 1.
Individual wind turbine power curves from a database of 187 modern wind turbine models. [Figure taken directly from Carrillo et al. (2013).]
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
Thresholds for the wind speed classes used in this study.
For wind farm operators, the extremes in power density are often of most interest (i.e., the months in which generation is anomalously high or low). As such, a composite analysis was carried out to investigate the atmospheric circulation characteristic of anomalous high–low monthly power density at selected wind farm locations. In this analysis, atmospheric circulation over the New Zealand region was represented by monthly MSLP from the NCEP–NCAR dataset. Composites were defined from the distribution of monthly power density at each wind farm location as the top and bottom 10th percentiles of power density, with a two-tailed t test used to test statistical significance.
3. Results and discussion
a. Correlation analysis
Several studies have reported wind speed or wind power correlations between wind farms to be inversely related to the distance between wind farms (e.g., Kahn 1979; Sinden 2007; Kempton et al. 2010). Likewise, we observed wind speed similarity (calculated by an angular measure of similarity from the correlation matrix) to be largest for wind farms in the same region and smallest for wind farms between regions a greater distance apart (Figs. 1 and 4). In terms of the magnitude of correlation coefficients in Fig. 4, the lowest correlation pairs were observed for correlations against wind farms in the far north (NTH3) or south (STH1) of the country. Notably, the only negative correlation pair was also reported for the correlation between these two wind farms (NTH3 and STH1).
Corrgram displaying Spearman correlation coefficients between hourly wind speeds for all wind farm locations. Wind farms are ordered along the diagonal of the corrgram based on a measure of similarity described and introduced in Friendly (2002).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
These findings are of significance given that the intermittency of generation has been found to be smoothed by pairing weakly positively correlated or negatively correlated wind farms in other studies (Kempton et al. 2010). Strong negative correlations provide the greatest benefits to reducing overall intermittency (often one station will experience relatively strong winds when the other experiences relatively weak winds, and vice versa); however, such an occurrence is rarely reported. For example, Sinden (2007) reported one negative correlation pair (out of 2080 tested) in the United Kingdom, while Kempton et al. (2010) reported one very weak (essentially zero) negative correlation pair in the United States (tested between 11 stations). A lack of benefit has been reported for aggregating wind farms in the United Kingdom (e.g., Oswald et al. 2008). It has been suggested (e.g., Oswald et al. 2008; Kempton et al. 2010) that this might be related to the north–south orientation of the island with respect to the east–west passage of synoptic systems, such that the effect of these systems is experienced almost simultaneously between stations. However, in the present study for New Zealand (a nation in which both islands are also oriented roughly perpendicular to the west–east propagation of synoptic systems), our correlation results imply that geographical orientation is not itself detrimental to observing the benefits of aggregating wind farms. Instead, we propose that the north–south distance over which wind farms are correlated (with New Zealand spanning a relatively large latitudinal extent) is more important than orientation, which is explored further within the synoptic framework of the following sections. Another important factor (but beyond the scope of this study) likely relates to New Zealand’s terrain complexity (e.g., Gibson and Cullen 2015), whereby the influence of the associated local-scale flows may act to reduce wind speed correlations between stations, as shown by Santos-Alamillos et al. (2014) in Spain.
b. Descriptive statistics
The next step was to explore the extent to which certain weather types affect wind resource quantity differently between wind farms, through an examination of wind class frequencies (Figs. 5 and 6). The general spread in class frequency corresponding to a particular weather type indicates that weather types tended to affect wind resources differently between regions (colors). Wind farms in the STH region were generally found to experience winds too low for generation (class A) under HSE, HW, NE, R, and TSW weather types (see Fig. 2 for type definitions) more frequently than wind farms in the NTH region (see Fig. 1) under the same weather types (Fig. 5a). Such differences are to some extent explainable by the position of the high pressure center associated with these weather types relative to the wind farm location. This is particularly true for the HSE, HW, and R weather types, whereby pressure gradients are weak for regions to the south under these types (Fig. 2). The observed differences in terms of synoptic influences, particularly between wind farms in the NTH and STH regions, also reflect the earlier correlation-based findings for these regions.
Frequency occurrence of (a) wind class A (winds too weak) and (b) wind class D (winds too strong) under different Kidson types for the 15 wind farm locations.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
As in Fig. 5, but for (a) class B (generation at below rated wind speed) and (b) class C (generation at rated wind speed).
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
The frequency spread for class D wind speeds (too high for generation; see Fig. 5b) between regions was found to be often less than that of class A; however, regional differences were still apparent for certain weather types, particularly between the NTH and STH regions. These regional differences were most pronounced for the HE, HNW, and W types (Fig. 2), with a higher frequency of class D wind speeds for STH-situated wind farms than for those in the NTH region. These weather types are associated with tightly constrained isobars over the STH region and weak pressure gradients over the NTH region under close proximity to a high pressure center.
We now shift our attention to weather types positively related to generation (class B and class C wind speeds) (Figs. 6a,b). In comparing the frequency spread with respect to the regions, the frequency of class B (Fig. 6a) and class C wind speeds (Fig. 6b) generally appeared to be inversely related. That is to say, under a given weather type a region with a high frequency of class B events was likely to display a low frequency of class C events, and vice versa. Subsequently, and also given that class C wind speeds are most crucial to high generation, we examine only the class C frequencies here. As apparent in Fig. 6b, all weather types were associated with a high degree of frequency spread between wind farms; for brevity, we give focus to weather types with the highest degree of spread and where the degree of spread for wind farms within the same region is considerably less than the between-region spread. This was found to be particularly true for the H, HE, HNW, and W weather types, whereby a higher frequency of class C wind speeds was observed for wind farms in the more southern regions. With the exception of the H weather type (Fig. 2), these weather types were also found to be related to a higher frequency of class D wind speeds earlier in the STH region (Fig. 5b), suggesting that these weather types could frequently result in either rated power (class C) and wind speeds too high for generation in the STH region. The T and SW weather types (Fig. 2) were also of interest given the relatively high frequency of class C (Fig. 6b) wind speeds observed over most wind farm locations. The T and SW weather types were also unique in that these types were generally found to result in the highest frequency of class C wind speeds for wind farms in the center of the country, more so than for wind farms located in the far south or north.
These findings carry practical significance for various reasons. First, the results linking wind speed classes and weather types, and the associated regional variability in these linkages, appear physically reasonable in terms of the position and relative strength of pressure gradients under these weather types. This provides evidence that the utilization of the employed weather-typing procedure is useful within the context of characterizing the synoptic controls on wind resource variability. Second, given the finding that many weather types affect the wind resource in different ways between regions, the results suggest that interconnecting farms between regions could act to reduce supply variability and security, which has not been examined previously within a New Zealand context. For example, Mason et al. (2010) argued that strong correlation in “weather patterns” between wind farm locations in New Zealand (if observed) could lead to volatile output, which is undesirable for increasing the penetration of wind generation. In contrast, we provide evidence to support the idea that the synoptic forcing of wind speed variability is regionally different across New Zealand, particularly so between the far northern and southern regions, which was also supported by the earlier correlation results (Fig. 4).
c. Regression and composite analysis
Continuing along this line of enquiry, we developed predictive models examining the extent to which the monthly frequency of certain weather types can account for intermonthly variability in power density observed at each individual wind farm. The MLR 
Measures of model performance for the regression models developed for each of the 15 wind farms. Here, RSEa refers to the residual standard error for models trained under all data, and RSEb refers to the residual standard error under the LOOCV scheme.
The MLR models were developed in this study to untangle how the dominant synoptic controls on wind resource variability differ between regions, as opposed to being used to reconstruct the wind resource in time. However, the finding that the monthly frequency of a select number of weather types can account for a considerable proportion of the intermonthly variability in power density, even in New Zealand’s complex terrain and at multiple wind farm locations, suggests that such an approach might carry practical utility elsewhere for reconstructing wind resources over several decades. If suitably accurate, this approach is favorable since reanalysis products (which can be used in the classification of weather types) span several decades, whereas representative wind speed data for a specific location often only span a few years.
Consideration of the specific weather types included in the MLR models enabled further examination into what weather types exert the most influence on wind resource variability, and how these influences vary between regions. Such an approach built upon the descriptive methods earlier (Figs. 5 and 6), and examined wind resource variability over longer (monthly) time scales. Comparing colors (weather types) for a particular rank (the rank of regression coefficients in a particular model) between wind farms in Fig. 7 provides further evidence that regional differences exist in terms of synoptic influences on the wind resource. Most notable was the dominant influence of the W and HNW weather types for regions south of the MWT region, in contrast to the dominant influence of the T weather type for regions to the north. These findings broadly reflect those presented earlier (Fig. 6b) whereby the W and HNW weather types favored class C wind speeds for regions to the south. The T weather type was also completely absent from any of the STH region MLR models, supporting the idea that NTH and STH regions were most different with regard to the influence of synoptic circulation. Also of note is that the STH1 and STH3 wind farms did not share any of the same weather types included in their respective MLRs. This suggests that synoptic influences may impact wind resources differently in the STH region over relatively small spatial distances; in contrast, wind farms in other regions exhibit a higher degree of within-region similarity.
Order (rank) of coefficient magnitude (y axis) in the regression models developed at each of the 15 wind farms (x axis). For a particular regression model, “positive 1” indicates the highest positive regression coefficient (Kidson type) while “negative 1” indicates the highest negative regression coefficient (Kidson type); circle size is proportional to the statistical significance of each regression coefficient.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
While a number of weather types were found to result in winds often too low for generation for the STH region in the descriptive examination presented earlier (Fig. 5a), it is interesting to note that none of these weather types were included as a negative coefficient in the MLR models. Furthermore, MLR models with negative coefficients were observed at only two wind farm locations. This suggests that, in terms of the weather types most important for discerning a bad month from a good month, a low frequency of weather types associated with exceptionally strong winds appears more important than a high frequency of weather types associated with exceptionally weak winds. Finally, it is significant that the W weather type, found to influence the wind resource strongly and positively in many regions, actually exerted a negative influence on the wind resource for the NTH3 wind farm. This difference, in terms of the synoptic-scale forcing, has likely contributed to the previously found negative correlation between STH1 and NTH3 wind farms (Fig. 4).
In terms of the extremes in power density, composite analysis indicates that high and low monthly-mean power density anomalies were related to the monthly-mean position of monthly high pressure centers (Fig. 8). Differences in these relationships were found to exist between regions, particularly between NTH and STH regions (for brevity only wind farms in NTH and STH regions are considered in Fig. 8). Again, this is especially apparent when comparing NTH3 and STH1 wind farms, whereby the position of high pressure centers related to high and low power density were positioned farther north for NTH3 compared to STH1 while still maintaining the symmetry (general direction of contours) of the relationships between wind farms. Most importantly, the position of high pressure centers related to low power density at NTH3 closely resembled (albeit positioned slightly to the north of) the position of high pressure centers related to high power density at STH1. That is, in terms of the synoptic-scale forcing of monthly extremes in wind resource quantity, these wind farms appear (almost) completely out of phase: when STH1 is experiencing very low power density, NTH3 is likely to be experiencing considerably higher power density. This observation, over monthly time scales, was also consistent with the negative correlation between these wind farms over hourly time scales (Fig. 4).
Composite analysis maps displaying anomalous MSLP (hPa) for high-generation months (top 10th percentile of monthly mean power density) minus low-generation months (bottom 10th percentile). Maps are presented separately for six select wind farms (belonging to the NTH and STH regions). Red (blue) colors are for positive (negative) differences in space; light to dark shading indicates statistical significance at p value levels of 0.1, 0.05, and 0.01, respectively.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
d. Implications for generation reliability
The finding that wind farms across New Zealand, particularly in the far northern and southern regions, are affected in different ways by certain weather types suggests that interconnecting wind farms should reduce supply intermittency here. Figure 9 illustrates how the distribution of area-averaged wind “power” speed is affected by the number of regions added to the network (in order of latitude from south to north), about which several conclusions can be drawn. First, the shape of the distribution is found to narrow with an increase in the number of regions added to the network. Similarly, Archer and Jacobson (2003) found a reduction in the standard deviation with increasing wind farms added to the network. The authors concluded that the associated reduction in intermittency of generation from multiple wind farms (compared to a single wind farm) may imply a reduction in contingency reserve requirements. Second, we observe the probability of the wind being too low for generation to decrease as the number of regions added to the network increases (Fig. 9). Distributions of area-averaged wind power speed under increasing numbers of regions added to the network were also assessed separately under different seasons and times of day, yet revealed very similar findings (not shown). This suggests that the benefits of interconnecting wind farms in New Zealand is not strongly seasonally or diurnally dependent, which may carry further importance for meeting electricity demand.
(a) Probability density functions (PDFs) of area-averaged wind speed [wind power speed in Eq. (3)] across variable number of regions in the network. Regions were added to the network in order from lowest to highest latitude. (b) NCEP–NCAR daily MSLP (hPa) averaged over calm days. Calm days were defined when the area-averaged wind speed for the six-region network was <3 m s−1 for three or more hours that day. This occurred for 5 days over the 5-yr period.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
Since the occurrence of hourly winds too low for generation was not found to be completely eliminated within a network of wind farms in six interconnected regions, of interest were the synoptic conditions that could still result in such occurrences. As might be expected, such conditions were found to occur only under intense high pressure systems that extend beyond the length of the country (Fig. 9). These findings support our earlier claims that supply reliability is largely dependent on the latitudinal span of the country relative to the typical extent and passage of synoptic systems in the region, which is also in general agreement with the findings of Kempton et al. (2010). In this study, synoptic events of the scale shown in Fig. 9 likely occurred infrequently enough so that a network of wind farms that spans the entire length of the country can achieve almost uninterrupted supply.
Finally, we examine all possible combinations of regions added to the network (Fig. 10) in terms of the frequency of zero-generation hours. The decrease in median hours of zero generation with increasing number of regions (Fig. 10) suggests that earlier findings appear robust in terms of the order in which regions are added to the network. However, it is also apparent that the six-region network was not necessarily optimal. That is, certain combinations of the two-, three-, four-, or five-regions network were shown in Fig. 10 to produce lower frequencies of zero generation than did the six-region case (as indicated by the lower tails of these distributions in Fig. 10). In particular, one of the combinations for a four-region network produced 0.23% whereas a two-region network produced 0.36%, in comparison with the six-region network with 0.56%. In terms of the specific regions included in these optimal networks, the two-region network combination consisted of STH and NTH regions, while the four-region network consisted of STH, CKS, MWT, and NTH regions. These results indicate that the inclusion of both STH and NTH regions in a network is required to optimize supply reliability, and reflects earlier findings that the synoptic linkages with wind speed variability are most different between these regions.
Boxplot of zero-generation hours as a function of the number of regions in the network from all possible combinations of regions. The bottom, middle, and top of each box display the 25th percentile, median, and 75th percentile, respectively, and the whiskers display outliers. The minimum zero-generation frequency for a particular group (number of regions) is indicated below the lower whiskers.
Citation: Journal of Applied Meteorology and Climatology 54, 5; 10.1175/JAMC-D-14-0273.1
In this study the geographic distribution of wind data included for analysis was subject to the locations of existing or proposed wind farms. The location of these wind farms is not random (site selection is strongly determined by the quantity of the wind resource in that particular region in combination with various other practical issues) and therefore we have not considered a network of wind farms from all possible wind farm locations in our analysis. With this in mind, it is possible that intermittency and supply reliability issues could be further reduced by an extended meteorologically based selection of wind data from a greater number of locations. A robust methodological approach for examining relationships between all potential wind farm locations within a region is outlined by Santos-Alamillos et al. (2014) and may be of relevance for future research in New Zealand.
Kempton et al. (2010) suggested that wind energy developers should take a broader approach to site selection. For example, rather than simply selecting a location based on its individual “windiness” (as is done today), site selection should be extended to give recognition to the benefits associated with a meteorologically oriented network of interconnected wind farms (within certain geographical contexts). Within a New Zealand context, our results stand to support the suggestions of Kempton et al. (2010); in particular, we would advocate for further examination of wind resources in the far northern and southern regions of New Zealand. Future research within the New Zealand context could also be expanded to include the sensitivity (and covariability) of multiple sources of electricity generation (e.g., hydro, solar, wind) and electricity demand to weather and climate.
4. Conclusions
This study has investigated linkages between synoptic-scale atmospheric circulation and wind resource quantity at 15 New Zealand wind farms, and the extent to which these linkages are regionally different. Subsequently, within the context of the synoptic-scale circulation, we explored the benefits to the supply reliability through interconnecting wind farms between six different regions. The main conclusions of this paper are as follows:
Correlations between wind farms were generally found to be strongest between wind farms in the same region, while correlations were found to be weakest, including one negative correlation, for correlation pairs in the far north and south of the country.
Regionality was found to exist in the linkages between wind resource quantity and synoptic-scale atmospheric circulation. These linkages were broadly consistent across a range of statistical methods of examination (descriptive statistics, multiple linear region models, and composite analysis). Differences, in terms of the synoptic-scale linkages, were often found to be most pronounced between regions in the far north and south of the country, with these regions exposed to different features of the synoptic situation at a given time. Weather types associated with stagnant high pressure to the far north of New Zealand with approximately westerly flow to the south were most strongly related to regional variability in wind resources. Intense high pressure systems that spanned the entire length of the country could result in very calm winds across all wind farm locations, but the occurrence of such events was found to be rare.
The benefit of interconnecting wind farms, in terms of supply reliability, was found to be optimized in a network that included regions in both the far north and south of the country, reflecting the disparity in the synoptic-scale influence between these regions. These findings carry implications for the future planning and site selection of wind farm projects in New Zealand.
Acknowledgments
The updated Kidson-type data were kindly provided by Dr. James Renwick. The authors acknowledge Dr. Richard Turner and coauthors (Turner et al. 2011) and the NZ Electricity Authority for making the synthetic wind dataset freely available (https://data.govt.nz/dataset/show/616) and acknowledge NOAA/OAR/ESRL PSD for the NCEP–NCAR reanalysis data (http://www.esrl.noaa.gov/psd/).
REFERENCES
Ackerley, D., and Coauthors, 2011: Using synoptic type analysis to understand New Zealand climate during the Mid-Holocene. Climate Past, 7, 1189–1207, doi:10.5194/cp-7-1189-2011.
Archer, C. L., and M. Z. Jacobson, 2003: Spatial and temporal distributions of U.S. winds and wind power at 80 m derived from measurements. J. Geophys. Res.,108, 4289, doi:10.1029/2002JD002076.
Archer, C. L., and M. Z. Jacobson, 2007: Supplying baseload power and reducing transmission requirements by interconnecting wind farms. J. Appl. Meteor. Climatol., 46, 1701–1717, doi:10.1175/2007JAMC1538.1.
Brayshaw, D. J., A. Troccoli, R. Fordham, and J. Methven, 2011: The impact of large scale atmospheric circulation patterns on wind power generation and its potential predictability: A case study over the UK. Renewable Energy, 36, 2087–2096, doi:10.1016/j.renene.2011.01.025.
Carrillo, C., A. Obando Montaño, J. Cidrás, and E. Díaz-Dorado, 2013: Review of power curve modelling for wind turbines. Renewable Sustainable Energy Rev., 21, 572–581, doi:10.1016/j.rser.2013.01.012.
Cullen, N. J., A. Bowden, and R. Spronken‐Smith, 2012: Partnering between a geography department and a community initiative to provide a wind resource assessment for the Blueskin Bay region, Otago, New Zealand. N. Z. Geogr., 68, 49–61, doi:10.1111/j.1745-7939.2012.01220.x.
Fisher, S. M., J. T. Schoof, C. L. Lant, and M. D. Therrell, 2013: The effects of geographical distribution on the reliability of wind energy. Appl. Geogr., 40, 83–89, doi:10.1016/j.apgeog.2013.01.010.
Friendly, M., 2002: Corrgrams: Exploratory displays for correlation matrices. Amer. Stat., 56, 316–324, doi:10.1198/000313002533.
Gibson, P. B., and N. J. Cullen, 2015: Synoptic and sub-synoptic circulation effects on wind resource variability—A case study from a coastal terrain setting in New Zealand. Renewable Energy, 78, 253–263, doi:10.1016/j.renene.2015.01.004.
Greene, S., M. Morrissey, and S. E. Johnson, 2010: Wind climatology, climate change, and wind energy. Geogr. Compass, 4, 1592–1605, doi:10.1111/j.1749-8198.2010.00396.x.
Kahn, E., 1979: The reliability of distributed wind generators. Electr. Power Syst. Res., 2, 1–14, doi:10.1016/0378-7796(79)90021-X.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kelly, G., 2011: History and potential of renewable energy development in New Zealand. Renewable Sustainable Energy Rev., 15, 2501–2509, doi:10.1016/j.rser.2011.01.021.
Kempton, W., F. M. Pimenta, D. E. Veron, and B. A. Colle, 2010: Electric power from offshore wind via synoptic-scale interconnection. Proc. Natl. Acad. Sci. USA, 107, 7240–7245, doi:10.1073/pnas.0909075107.
Kidson, J. W., 2000: An analysis of New Zealand synoptic types and their use in defining weather regimes. Int. J. Climatol., 20, 299–316, doi:10.1002/(SICI)1097-0088(20000315)20:3<299::AID-JOC474>3.0.CO;2-B.
Kossmann, M., and A. Sturman, 2004: The surface wind field during winter smog nights in Christchurch and coastal Canterbury, New Zealand. Int. J. Climatol., 24, 93–108, doi:10.1002/joc.981.
Leahy, P. G., and A. M. Foley, 2012: Wind generation output during cold weather-driven electricity demand peaks in Ireland. Energy, 39, 48–53, doi:10.1016/j.energy.2011.07.013.
Mason, I., S. Page, and A. Williamson, 2010: A 100% renewable electricity generation system for New Zealand utilising hydro, wind, geothermal and biomass resources. Energy Policy, 38, 3973–3984, doi:10.1016/j.enpol.2010.03.022.
McCauley, M. P., and A. P. Sturman, 1999: A study of orographic blocking and barrier wind development upstream of the Southern Alps, New Zealand. Meteor. Atmos. Phys., 70, 121–131, doi:10.1007/s007030050029.
MED, 2013: Energy in New Zealand. Ministry of Business, Innovation and Employment Rep., 76 pp. [Available online at http://www.med.govt.nz/sectors-industries/energy/energy-modelling/publications/energy-in-new-zealand/previous-energy-in-new-zealand-editions/Energy-in-New-Zealand-2013.pdf/view.]
NZWEA, 2013: Wind farm development in New Zealand. New Zealand Wind Energy Association Pamphlet, 4 pp. [Available online at http://www.windenergy.org.nz/store/doc/WindFarmDevelopmentinNZ.pdf.]
Oswald, J., M. Raine, and H. Ashraf-Ball, 2008: Will British weather provide reliable electricity? Energy Policy, 36, 3212–3225, doi:10.1016/j.enpol.2008.04.033.
Pretli, D. G., 2003: Winds of change: Exploring New Zealand’s phenomenal wind resource and options to drive renewable energy development. Greenpeace New Zealand Rep., 34 pp. [Available online at http://www.greenpeace.org/new-zealand/Global/new-zealand/report/2006/12/winds-of-change-report.pdf.]
Reid, S. J., 2005: Wind and wind energy. Water & Atmosphere, Vol. 13, No. 4, NIWA, Auckland, New Zealand, 12–13. [Available online at https://www.niwa.co.nz/sites/niwa.co.nz/files/import/attachments/wind.pdf.]
Santos-Alamillos, F. J., D. Pozo-Vázquez, J. A. Ruiz-Arias, V. Lara-Fanego, and J. Tovar-Pescador, 2014: A methodology for evaluating the spatial variability of wind energy resources: Application to assess the potential contribution of wind energy to baseload power. Renewable Energy, 69, 147–156, doi:10.1016/j.renene.2014.03.006.
Schaefer, M. S., B. Lloyd, and J. R. Stephenson, 2012: The suitability of a feed-in tariff for wind energy in New Zealand—A study based on stakeholders’ perspectives. Energy Policy, 43, 80–91, doi:10.1016/j.enpol.2011.12.032.
Sinden, G., 2007: Characteristics of the UK wind resource: Long-term patterns and relationship to electricity demand. Energy Policy, 35, 112–127, doi:10.1016/j.enpol.2005.10.003.
Sovacool, B. K., and C. Watts, 2009: Going completely renewable: Is it possible (let alone desirable)? Electr. J., 22, 95–111, doi:10.1016/j.tej.2009.03.011.
Sturman, A. P., and P. Tyson, 1981: Sea breezes along the Canterbury coast in the vicinity of Christchurch, New Zealand. J. Climatol., 1, 203–219, doi:10.1002/joc.3370010302.
Sturman, A. P., and H. Wanner, 2001: A comparative review of the weather and climate of the Southern Alps of New Zealand and the European Alps. Mt. Res. Dev., 21, 359–369, doi:10.1659/0276-4741(2001)021[0359:ACROTW]2.0.CO;2.
Sturman, A. P., and N. J. Tapper, 2006: The Weather and Climate of Australia and New Zealand. 2nd ed. Oxford University Press, 541 pp.
Turner, R., X. Zheng, N. Gordon, M. Uddstrom, G. Pearson, R. de Vos, and S. Moore, 2011: Creating synthetic wind speed time series for 15 New Zealand wind farms. J. Appl. Meteor. Climatol., 50, 2394–2409, doi:10.1175/2011JAMC2668.1.
Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences. 2nd ed. Academic Press, 627 pp.
