Applied Energy 264 (2020) 114731

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Impact of weather regimes on wind power variability in western Europe T

a,b,⁎ a b a,b
Jose M. Garrido-Perez , Carlos Ordóñez , David Barriopedro , Ricardo García-Herrera ,
Daniel Paredesc
a
Departamento de Física de la Tierra y Astrofísica, Facultad de Ciencias Físicas, Universidad Complutense de Madrid, Madrid, Spain
b
Instituto de Geociencias (IGEO, CSIC-UCM), Madrid, Spain
c
Department of Energy Resource, Iberdrola, Madrid, Spain

HIGHLIGHTS

• New set of weather regimes that improve the capability to reproduce wind power.
• Monthly correlations of wind power with weather regime occurrence up to ~0.8.
• High reproducibility of wind farms’ capacity factors in Iberia and the United Kingdom.
• Weather regimes explain day-to-day evolution of anomalous regional wind episodes.

ARTICLE INFO ABSTRACT

Keywords: We have assessed the dependency of wind power resources in Western Europe on the atmospheric circulation as
Wind power represented by a new set of 8 tailored weather regimes (WRs). For this purpose, we have derived wind capacity
Weather regimes factors (CFs) from a meteorological reanalysis dataset and from high-resolution data simulated by the Weather
Atmospheric circulation Research and Forecasting (WRF) model. We first show that WRs capture effectively year-round onshore wind
Wind resources
power production variability across Europe, especially over northwestern/central Europe and Iberia. Since the
Wind energy
influence of the large-scale circulation on wind energy production is regionally dependent, we have then ex-
amined the high-resolution CF data interpolated to the location of more than 100 wind farms in two regions with
different orography and climatological features, the United Kingdom and the Iberian Peninsula. The use of the
monthly frequencies of occurrence of WRs as predictors in a multi-linear regression model allows explaining up
to two thirds of the month-to-month CF variability for most seasons and sub-regions. These results outperform
those previously reported based on Euro-Atlantic modes of atmospheric circulation, indicating that the use of
WRs customized to the region of study is preferred to reproduce the evolution of wind energy resources. Finally,
we have applied these WRs to understand the day-to-day evolution of specific episodes with anomalous regional
wind power production. In particular, the wind energy deficit of summer 2018 in the United Kingdom and the
surplus of March 2018 in Iberia stemmed from the combination of WRs associated with low and high CFs,
respectively. These findings are relevant for the forecast of wind energy resources as the large-scale features of
the atmospheric circulation captured by WRs can be modelled with considerably less uncertainty than wind
speeds at wind farm sites.

1. Introduction electricity capacity [1]. The future development and implementation of

Over the last years, wind power production has become key in the
European electricity market. From the beginning of the century, the
deployment of wind installed capacity has increased more than tenfold,
with the largest contributions coming from Germany, Spain and the
United Kingdom (UK). At present, it generates the largest share of re-
newable electricity with more than one third of the total installed
this renewable energy source in Europe and other regions of the globe
will help meet some sustainable development goals, including afford-
able and clean energy, sustainable cities and communities, responsible
consumption and production, and climate action [2,3]. However, a
drawback of this power source is its dependency on meteorological
drivers and therefore its intermittency. This issue has grown in im-
portance, with recent studies analyzing the impact of wind power

⁎
Corresponding author at: Departamento de Física de la Tierra y Astrofísica, Facultad de Ciencias Físicas, Universidad Complutense de Madrid, Madrid, Spain.
E-mail address: josgarri@ucm.es (J.M. Garrido-Perez).

https://doi.org/10.1016/j.apenergy.2020.114731
Received 8 August 2019; Received in revised form 20 February 2020; Accepted 22 February 2020
0306-2619/ © 2020 Elsevier Ltd. All rights reserved.
J.M. Garrido-Perez, et al.
variability on the reliability of the operating power systems and the
associated economic costs [4–7].
The main meteorological driver of wind energy is wind speed,
which mainly depends on the atmospheric circulation and the local
orography. Therefore, improving our understanding of wind speed
variability is required for the evaluation and further implementation of
wind energy resources. This can be achieved through the use of me-
teorological data derived from numerical weather prediction models,
but these suffer from different sources of uncertainty, including im-
perfect physical parameterizations that limit their ability to simulate
the dynamics of the lower atmosphere (e.g. [8–11]). In addition to
improving the representation of physical processes in numerical
weather models, it is essential to develop a good understanding of the
relationships between the large-scale atmospheric circulation and wind
power production at the regional scale [12–14].
This study focuses on wind power output over the western European
façade and its response to large-scale circulation patterns. Because of
the importance of wind power resources for the renewable energy
sector, this topic has recently attracted considerable attention [15–21].
Previous studies used the main teleconnection patterns as wind power
predictors in the region: the North Atlantic Oscillation (NAO), and the
East Atlantic (EA) and Scandinavian (SCAND) modes (e.g. [22]). Most
of those analyses focused on wind generation during winter, when these
large-scale circulation patterns have their greatest influence. A more
accurate classification of the large-scale atmospheric circulation can be
obtained through the use of weather regimes (WRs), which are defined
as recurring large-scale weather patterns [23–25]. However, most of
these classifications are seasonally-dependent and have mainly been
applied to the winter season [26–30], therefore posing limitations to
address their influence in wind power variability all year-round.
Recent studies have extended these WR classifications to other
seasons. The results by Grams et al. [31] or Cortesi et al. [32] have
evidenced the usefulness of Euro-Atlantic WRs to capture year-round
wind power production on European countrywide scales. The reported
influence of the large-scale circulation on wind resources was found to
be seasonally and geographically dependent, explaining more varia-
bility in winter than in summer as well as in northern than in southern
countries of the continent. This contrasting behavior is also observed in
the dominant wind regimes. As an illustration, we have used the daily
wind speed at a height of 124 m from the ERA-Interim reanalysis [33]
as a proxy of wind power output in the UK and the Iberian Peninsula
(Fig. S1). To identify regional differences, these data have been fitted
with Weibull distributions separately over each grid cell in those re-
gions (see Section S1 in Supplement). Overall, wind speeds in the UK
are not strongly dependent on the reanalysis grid cell considered, while
in Iberia the distribution varies with the geographical location (Fig. S1,
top). Regional differences in the distribution of the wind speed and the
expected wind production can also be inferred by depicting the range of
values of the scale and shape parameters of the corresponding Weibull
fits (Fig. S1, bottom). Both parameters are lower for Iberia (Fig. S1d),
indicating lower wind speeds and a steeper lower tail than in the UK
(Fig. S1c). In addition, the larger spread of the shape parameter in
Iberia can be attributed to the higher spatial variability in this region.
These results evidence differences in wind regimes, and hence wind
power, on sub-country spatial scales, stressing the need for regional
assessments for some areas like Iberia. To the best of our knowledge,
the skill of WRs as predictors of wind power generation has only been
assessed countrywide, therefore disregarding regional aspects that are
of paramount importance for the development of planning strategies
and decision-making on deployment of energy resources.
This study makes a step forward by introducing novelty in the as-
sessment of WRs as predictors of wind power variability. First, to better
quantify and exploit the skill of WRs in predicting wind power pro-
duction, we have used customized and year-round WRs that amplify the
circulation signal on wind power output for the region of study. Second,
unlike previous studies, we have evaluated the ability of WRs in
2
Applied Energy 264 (2020) 114731
reproducing wind power production on finer (i.e. sub-country) spatial
scales. Addressing such scales is as important as improving the re-
presentation of physical processes in numerical weather models, as both
are needed for the refinement of wind production forecasts. Finally, the
observed relationships between the large-scale flow and wind power
have been further exploited to reproduce episodes of anomalous re-
gional wind power production, which represent a major concern for
capacity planning in energy generation. Consequently, the three main
objectives of the paper are: (i) to identify robust relationships between
Euro-Atlantic WRs and wind resources in the UK and Iberia; (ii) to as-
sess the skill of WRs as predictors of wind power production on monthly
time-scales all year-round, and (iii) to evaluate the potential of WRs to
reproduce the day-to-day variability of wind generation during periods
of anomalously low and high production.
The paper is structured as follows. Section 2 describes the datasets
and methodology used to estimate wind power production as well as to
provide a classification of WRs. Section 3 identifies the relationships
among wind resources and WRs for the UK and Iberia. Section 4 in-
troduces a statistical model to reproduce monthly wind production as a
function of the occurrence of WRs. Section 5 analyzes the day-to-day
evolution of wind resources and WRs during specific case studies. Fi-
nally, Sections 6 and 7 discuss and summarize the main findings of this
work.
2. Data and methods
2.1. Wind energy resources
We have used two different data sources to characterize near-sur-
face wind at the large and local scales: output from a meteorological
reanalysis and data provided by the electric company Iberdrola at the
location of some of their wind farms in the Iberian Peninsula and the
UK.
First, we have extracted meteorological fields from the ERA-Interim
reanalysis at 0.75° × 0.75° horizontal resolution for the 1979–2018
period. Daily data have been used to estimate normalized wind speeds
(vn ) at hub height (assumed to be 93 m in this study), and then fed
through the power curve of a Siemens-Gamesa G114-2.5 MW wind
turbine to estimate capacity factors (CFs) (see section S2 in
Supplement). We have also used time series of CFs at the locations of
105 wind farms operated by Iberdrola in the UK and Iberia. The UK
sites (26 wind farms) are mainly concentrated in Scotland, while the
Iberian ones (79) cover most of the Spanish territory and part of
northern Portugal (not shown). These data were provided by Iberdrola
following the post-processing (spatial interpolation and bias correction)
of high-resolution data simulated by the Weather Research and
Forecasting (WRF) model for the 1991–2018 period. More details are
provided in Section S3 of the Supplement.
For the sake of clarity, in the following, the CFs computed from the
reanalysis and from high resolution data interpolated to the locations of
the wind farms will be referred to as CFR and CFWF, respectively. Note
that these data are available for different periods (CFR during
1979–2018 and CFWF during 1991–2018).
2.2. Weather regimes
For the identification of WRs, the methodology applied in this work
is similar to that of previous studies that have derived WRs over the
North Atlantic-European sector [23,31,34–36]. We have used daily
500 hPa geopotential height (Z500) fields from the ERA-Interim re-
analysis. Other meteorological variables from the same dataset such as
SLP and surface wind speed at 10 m have also been employed to better
understand the low-level flow and wind speed patterns for each of the
WRs defined. Traditionally, in the North Atlantic-European sector, WRs
have been defined for winter or summer with most studies considering
four WRs per season [26–30]. Grams et al. [31] took a step forward and
J.M. Garrido-Perez, et al. Applied Energy 264 (2020) 114731

Fig. 1. Cluster centroids of the weather regimes (WRs) derived from the ERA-Interim reanalysis for 1981–2010. Colors show their 500 hPa geopotential height
(Z500) anomalies (in m). The average Z500 field is plotted for each WR as black contour lines (in m). The percentage of days a given WR occurs is shown on the top
right corner of each panel. The full name and a brief description of each regime are provided in Table 1.

utilized seven WRs to capture the large-scale flow year-round. These
WRs were defined over a large domain (80°W – 40°E, 30°N – 90°N).
However, the CFs derived by that study present little inter-WR varia-
bility for some countries such as Spain in most seasons. This problem
hinders the connection between the large-scale flow and regional wind
power production, which can partially be solved by reducing the size of
the domain used to define WRs. This approach has been shown to in-
crease the regional surface responses to the atmospheric circulation
(e.g. [37]). After testing different regions, we have finally used a do-
main (30°W – 25°E, 30°N – 65°N) that maximizes the signal of the WRs
on near-surface wind both over Iberia and the UK. This domain is
substantially smaller than those traditionally used to derive WRs for the

3
J.M. Garrido-Perez, et al.
Fig. 2. As Fig. 1 but for surface wind speed at 10 m (in m s−1).
North Atlantic-European sector as it only includes the eastern half of
the North Atlantic (see Fig. 1).
Firstly, we have extracted 8 recurrent patterns of the atmospheric
circulation from the period 1981–2010 (see section 4 in Supplement for
more details). In a second stage, each day from January 1979 to
December 2018 has been classified into one of these 8 WRs, based on
the minimum Euclidean distance of the daily Z500 anomaly fields to the
eight centroids. Figures 1 and S2 respectively display the average Z500
4
Applied Energy 264 (2020) 114731
and SLP anomalies associated with each WR for the 1981–2010 cli-
matological period. Overall, there is close resemblance between the
Z500 and SLP mean anomaly patterns, although this might not always
be the case for individual days. Fig. 2 shows the 10-m wind speed
anomaly composites for each WR. We will refer to these WRs, from left
to right and from top to bottom of Figs. 1, 2 and S2 as Climatology (CL),
European blocking (EB), Atlantic Low (AL), Atlantic High (AH), Zonal
Regime (ZR), Scandinavian Blocking (SB), Scandinavian Low (SL) and
J.M. Garrido-Perez, et al. Applied Energy 264 (2020) 114731

Table 1
Brief overview of the main features of each weather regime (WR).
WR Frequency of Main meteorological features
occurrence (%)

Climatology CL 11.4 Small anomalies of Z500 associated with weak zonal flow.
European Blocking EB 12.7 Anticyclonic anomaly over northern Europe, covering the British Isles and Scandinavia. This blocking situation leads
to strong negative wind speed anomalies over the former region.
Atlantic Low AL 11.2 Low with its center to the SW of the British Isles, affecting western Europe. Positive wind speed anomalies over part of
the European Atlantic façade, particularly affecting Iberia.
Atlantic High AH 14.8 High affecting western Europe and strong SLP gradients over the British Isles, leading to a wind speed dipole: positive
anomalies extending from the North-East Atlantic to Scotland, and negative anomalies around the Gulf of Biscay.
Zonal Regime ZR 13.8 Extratropical cyclone located over the North Atlantic, with center to the NW of the British Isles. Associated with
strong zonal flow and positive wind speed anomalies over the British Isles.
Scandinavian Blocking SB 13.8 Z500 dipole anomaly with a low over the North Atlantic and a high over Scandinavia. The latter blocks the circulation
leading to negative wind speed anomalies over central and northern Europe, with maximum anomalies around the
British Channel.
Scandinavian Low SL 11.2 Deep low over northern Europe and anticyclone west of Iberia. The strong SLP gradients over France and the British
Isles yield positive wind speed anomalies over a large part of northwestern/central Europe, with maxima over land
areas mainly in Germany but also affecting the British Isles.
East Atlantic EA 11.2 Atlantic ridge west of the British Isles and negative Z500 anomalies with center to the east of Scandinavia. Associated
with negative wind speed anomalies west of the British Isles and overall moderate positive anomalies over some parts
of the continent.

East Atlantic (EA). Table 1 provides a brief overview of the main me-
teorological features of each WR. Three regimes are predominantly
cyclonic (AL, ZR, SL) and four represent blocked flow over different
regions of the domain (EB, AH, SB, EA). The remaining regime (CL)
represents days with weak Z500 anomalies, i.e. with undefined atmo-
spheric flow patterns as compared with the other WRs. As the area used
in this work is relatively small, most of our WRs resemble but do not
necessarily coincide with those from other studies (e.g. [31]).
The frequency of occurrence of the 8 WRs ranges from 11.2% to
14.8%, as seen in Table 1 and on the upper right corners of the panels in
Fig. 1. Nevertheless, the occurrence of WRs varies depending on the
season (Fig. S3). In particular, CL is less often found in winter and
shows a clear predominance in summer, since this is the time of the
year when atmospheric circulation is least organized. The rest of the
WRs are more evenly distributed during the year. We have also com-
puted the WRs from the NCEP/NCAR reanalysis [38] and the results
evidence that the WR classification is not very sensitive to the choice of
the reanalysis product (not shown).
3. Relationships between WRs and wind production
We have first examined how wind resources change with the oc-
currence of different atmospheric patterns, as inferred from the WRs
previously defined. Daily vn data, averaged separately over all re-
analysis grid cells in the UK and Iberia, have been pooled into 8 classes
according to the WRs and then fitted to Weibull distributions (Fig. 3).
The figure also shows the medians of vn under each WR (vertical co-
lored lines) as well as the cut-in and nominal wind speeds of the Sie-
mens-Gamesa G114-2.5 MW turbine (vertical dashed lines). The de-
pendence of vn on the WRs is higher in the UK, where the medians of
wind speed for the different WRs range from 5 to 10 m s−1. In Iberia,
the medians are around 4 m s−1 for most of the WRs, with the exception
of AL (6 m s−1). The peaks of the Weibull curves for the UK are clearly
above the cut-in wind speed regardless of the WR, while in the case of
Iberia most curves maximize around that value. Hence, situations with
no wind power production are more frequent in Iberia than in the UK,
although they are also more likely to occur under specific WRs.
To improve our understanding of the relationship between WRs and
wind generation, we have examined the impact of the former on the CF
distributions of the UK and Iberia. Daily CFR are averaged separately
over all reanalysis grid cells in each of these two regions. In the case of
the wind farms, we adopt a simple approach: we assume a single tur-
bine for every site and average the daily CFWF data across all farms in
each region. Note that although British wind farms are mainly located
in Scotland, the average CFWF time series fairly represent the national
behavior due to the low spatial variability of CF across the UK (as
shown previously in Figure S1). Figs. 4, and S4 respectively display the
distributions of the daily CFs (ranging from 0 to 1), obtained from wind
farms and reanalysis, and averaged over the UK (left) and Iberia (right),
when the different WRs occur.
Irrespectively of the WR considered, these figures confirm that wind
power production is more efficient in the UK than in Iberia, as the CFs
are considerably higher for the former region. This difference is sub-
stantially higher for the reanalysis than for the wind farms. Note that
the median CFs are lower for the wind farms than for the reanalysis in
the UK, and vice versa in Iberia. Similar systematic spatial biases have
previously been found for meteorological reanalysis datasets. As an
example, Staffell and Pfenninger [39] reported significant over-
estimation of wind output by 50% in northwestern Europe and under-
estimation by 30% in the Mediterranean for the MERRA and MERRA-2
reanalyses [40,41].
In spite of the mentioned differences, the dependence of wind power
production on the WRs obtained at the location of the wind farms
(Fig. 4) is consistent with that provided by the reanalysis (Fig. S4). In
the case of the UK, ZR, SL and, to a lesser extent, AH stand out over the
rest, as their median CFs exceed ~ 0.5 considering reanalysis data and
0.4 for the wind farms. This result was expected based on the WR
characteristics described previously, because the three regimes lead to
enhanced zonal flow over the UK. On the other hand, EB shows the
lowest CF values, with median below 0.3 and around 0.2 for the re-
analysis and wind farms, respectively. The stagnant flow over northern
Europe associated with EB explains the low wind power production. In
the case of Iberia, AL and, to a lesser extent, SL and EA present rela-
tively high CF values. The highest median CFs are found for AL (~0.5
for the wind farms and ~ 0.3 for reanalysis), because of the cyclonic
flow affecting Iberia (Figs. 1 and S2). The median CF values for the rest
of WRs (EB, AH, ZR and SB) are around 0.3 for the wind farms, and
considerably lower for the reanalysis. It is noteworthy the fact that SL is
the only pattern associated with relatively high wind power production
for both Iberia and the UK. This is due to a generalized enhanced zonal
flow over the Atlantic, induced by the strong SLP gradient between the
low pressure system over northern Europe and the high pressure system
west of Iberia that characterizes this regime. On the other hand, CL
yields low production in both regions. This is not very surprising, given
that it is most frequent in summer, when the wind speed is on average
lower than in other seasons.
Table S1 presents the mean CFWF values for the different WRs in
each season and region. Overall, we find the highest CFWF values in

5
J.M. Garrido-Perez, et al. Applied Energy 264 (2020) 114731

Fig. 3. Weibull distributions of vn at 93 m height for the UK (left) and the Iberian Peninsula (right). vn data were derived from the ERA-Interim reanalysis for the
period 1979–2018. Vertical dashed lines show the cut-in (3 m s−1) and nominal (14 m s−1) wind speeds of the Siemens-Gamesa G114-2.5 MW turbine. Each colored
line represents the fit for a different WR and its corresponding median is indicated by a vertical line. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)

winter and the lowest ones in summer. CFs are somewhat higher during
spring than during autumn in Iberia, and vice versa in the UK.
Furthermore, the inter-WR variability of the CFs found previously for
the whole year is also present for the different seasons, although it
maximizes in winter and minimizes in summer. We have assessed the
likelihood that the mean values of CFWF are significantly different
among the 8 WRs. For that purpose we have used a methodology to
address multiple independent tests [42]. We have first applied two-
sided t-tests to the CFWF time series of each pair of WRs (28 possible
combinations in each season and region) to test for differences of
means. This yields N significantly different means at the 95% con-
fidence level out of the 28 possible combinations tested in each case.
The results indicate that N is dependent on the region and season. The
lowest values are found in spring, with 21 and 25 significantly different
means for Iberia and the UK, respectively. Moreover, the UK presents
slightly higher N values than Iberia for all seasons (on average 25.75 for
the former and 23 for the latter, considering all seasons), indicating a
higher sensitivity of CFWF to WRs in the former region. Despite these
seasonal and regional differences, N is above 20 for all seasons and
regions. Furthermore, the likelihood of obtaining this value by random
chance is well below 1%, as inferred from the binomial probability
distribution function. Therefore, the use of WRs is effective to sig-
nificantly discriminate situations with different wind power
production.
As mentioned, in the UK, the largest CFWF difference among WRs is
found in winter, when the mean CFWF increases from 0.29 for EB to
0.71 for ZR (i.e. a 0.42 difference), while the smallest differences often
occur in summer (e.g. 0.24 for the mentioned WRs, but still significant,
and considerably lower for others). The same seasonal behavior is
found in Iberia, although with inter-WR differences in CFWF almost half
of those in the UK for all seasons. The largest differences in the case of
Iberia occur for AH and AL in winter (0.27), while they are lower in
summer (maximum difference of 0.10 for AL versus ZR or SB). These
results can be compared to those by Grams et al. [31], who used na-
tional aggregate CFs from the wind-power dataset Renewables.ninja
[39] (see their Supplementary database). Their inter-WR differences in
the mean CFs are low compared to those reported herein in all seasons,
especially in summer, when they do not exceed 0.16 in the UK and 0.04
in Spain. This confirms that the use of a relatively small domain to
define WRs (as done in the present study) maximizes the regional re-
sponses of wind power production to the atmospheric circulation.
4. Modeling wind power production with WRs
In this section we explore the skill of the WRs in reproducing wind
power generation. For this purpose, we have modelled the monthly

Fig. 4. Dependence of the daily average distributions of CFWF on the WRs, considering data from 26 wind farms in the UK (left) and 79 wind farms in Iberia (right).
The boxes extend from the lower (Q1) to the upper (Q3) quartile values of the data, with a horizontal line indicating the position of the median (Q2). The whiskers
extend from the boxes to show the range of the data between the 10th and 90th percentiles. Period of analysis: 1991–2018.

6
J.M. Garrido-Perez, et al.
mean CFs, separately for each grid cell of the ERA-Interim reanalysis
and for each wind farm location, using a multi-linear regression model
(MLRM) on the monthly frequencies of occurrence of each weather
regime (WRi):
N
CF = c0 +
i=1
ci f(WRi ) (1)
where:
N: number of WRs
c0: intercept
ci: monthly regression coefficients
f(WRi): monthly frequency of WRi
Note that this regression is separately done for each month instead
of each season of the year, because both the frequency of occurrence
and the signal of each WR on wind power production present some
intra-seasonal variability. This way the length of the CF time series
equals the number of years with data (40 for reanalysis and 28 for the
wind farms). Then we re-compute the monthly CF time series with the
estimated coefficients and the observed monthly frequencies of WRs for
the same period. Finally, the coefficient of determination (R2) of the
regression between the modelled and observed monthly CF time series
is used to evaluate the skill of the model. This is computed by separately
grouping the 12 months of the year and the 3 months of each season
during the whole period (40 years for reanalysis and 28 for wind
farms).
Fig. 5 shows the spatial distribution of R2 in Europe considering
monthly reanalysis data. These R2 values are distributed with con-
siderable spatial variability across Europe. The MLRM explains the
largest fraction of inter-monthly onshore CFR variance over Iberia and
northwestern/central Europe, with R2 values up to 0.7–0.8 over large
parts of Spain, Germany, Benelux, the British Isles and some regions of
France. Even higher values, up to ~0.9, are found over some parts of
the Mediterranean basin. On the other hand, there are regions with very
complex topography (e.g. the Alps and the Atlas Mountains) or with
local wind regimes (e.g. the Strait of Gibraltar) where the influence of
the large-scale flow is reduced, resulting in R2 values below 0.4.
The application of the MLRM to the reanalysis has provided a first
view of the ability of the WRs to capture the month-to-month variability
of wind power production, indicating considerable spatial hetero-
geneity not only across Europe but also within our regions of interest,
especially Iberia. However, reanalysis datasets do not resolve local
Fig. 5. Coefficient of determination (R2) between the original and the modelled
monthly CFR time series, considering ERA-Interim reanalysis data from 1979 to
2018 (12 months × 40 years = 480 data points).
7
Applied Energy 264 (2020) 114731
features that might determine the variability of wind resources at some
specific wind farms. Hence, from now on we use CFWF data to test the
ability of WRs to reconstruct monthly wind power production. For that
purpose, we pool the wind farms within 6 sub-regions (see black boxes
in Fig. 5), attending to climatological features, to the presence of re-
gional winds, which are often related to the orography, and to the
spatial patterns observed so far. The sub-regions are: Scotland, and
Northwestern (NW), Northeastern (NE), Eastern (E), Southwestern
(SW) and Southeastern (SE) Iberia. Over the UK we have focused on
Scotland, simply because most of the British wind farms are located
there. This region is characterized by temperate and oceanic climate,
like the rest of the UK, but slightly windier, especially in the north.
Iberia has been divided into 5 sub-regions, considering that Iberian
wind farms tend to be placed within specific geographical clusters (not
shown). Note also the large spatial variability of wind speed in this
region compared to the UK (Fig. S1), besides the mentioned uneven
performance of the MLRM in the region (Fig. 5). NW Iberia, with
oceanic climate and frequent frontal activity, can be considered as a
transition region between the rest of Iberia and the UK. NE Iberia covers
the Ebro Valley, which frequently channels northwestern winds under
specific large-scale patterns. E Iberia presents a warm semi-arid climate,
with extreme precipitation events associated with the occurrence of
mesoscale convective systems especially in autumn. SE Iberia has a
warm semi-arid Mediterranean climate affected by local winds char-
acteristic of the Strait of Gibraltar. SW Iberia, although characterized by
a warm-summer Mediterranean climate, is influenced by maritime air
masses from the Atlantic Ocean, consequently receiving more pre-
cipitation than E and SE Iberia.
Table 2 presents the seasonal (columns 3–6) and yearly (last
column) R2 values between the original and modelled monthly time
series of CFWF averaged for the wind farms in the 6 sub-regions. In
addition, Fig. 6 displays the original (black) and reconstructed (red)
monthly CFWF values for the wind farms of Scotland (top), NW Iberia
(middle) and SE Iberia (bottom), during summer (left) and winter
(right). Note that, for the sake of brevity, the results are not provided
separately for each month but for all months in each season. The MLRM
is slightly more skillful over Scotland than for most Iberian sub-regions
in all seasons. Moreover, the poorest performance is often found in
summer, with the exception of the southernmost sub-regions (SE and
SW Iberia). This seems to be consistent with the regional and seasonal
sensitivities of CFWF to the atmospheric circulation (as summarized by
the WRs) reported above. The results found for summer could be at
least partly due to the low CFWF values and therefore low signal-to-
noise ratio in that season compared to winter, as shown in Fig. 6. NW,
NE and E are the Iberian sub-regions with the highest skill. These re-
gions display R2 values close to 0.80 when using data from all seasons,
only slightly below that of Scotland (see last column of Table 2). In
particular, NW behaves like Scotland, as both of them have R2 values of
Table 2
Coefficients of determination (R2) between the original and the modelled
monthly CFWF time series averaged over six sub-regions in the UK and Iberia for
each season of 1991–2018. The location of the sub-regions is shown in Fig. 5.
The number of wind farms considered for each region is shown in column 2.
The number of data points is 84 for each season (28 years × 3 months; columns
3–6) and 336 when combining data from all seasons (28 years × 12 months;
last column).
Region No. Wind Farms R2
spring summer autumn winter year
Scotland 22 0.70 0.53 0.80 0.73 0.85
NW Iberia 12 0.70 0.40 0.71 0.71 0.80
SW Iberia 8 0.66 0.63 0.48 0.63 0.66
SE Iberia 8 0.34 0.44 0.16 0.40 0.42
NE Iberia 28 0.69 0.43 0.74 0.60 0.78
E Iberia 23 0.68 0.52 0.81 0.59 0.80
J.M. Garrido-Perez, et al. Applied Energy 264 (2020) 114731

Fig. 6. Time series of monthly CFWF averaged for the wind farms in Scotland (top), Northwestern Iberia (middle) and Southeastern Iberia (bottom) for the summer
(left) and winter (right) months of 1991–2018 (3 months × 28 years = 84 data points in each panel). Black lines depict the original CFWF time series and red lines the
time series generated by the MLRM based on the occurrence of WRs (Eq. (1)). R2 values are shown in brackets on the top of each panel. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

at least 0.70 for all seasons except summer. This is arguably due to the
proximity of these regions to the Atlantic Ocean. Nevertheless, the skill
of WRs to capture inter-monthly variations of CFWF is not limited to
regions under Atlantic influence, as seen from the mentioned good
performance of the MLRM in NE and E. On the other hand, SE presents
the weakest relationship between wind power production and large-
scale circulation. The highest R2 in this sub-region occurs in summer,
coinciding with the lowest seasonal skill in most of the other sub-re-
gions. In spite of this, summer R2 values in SE are not higher than those
reported elsewhere during this season, indicating an overall poorer
performance of WRs in SE. This is likely associated with the prevalence
of local winds and Mediterranean-related mesoscale circulation over
the large scale.
Summarizing, although the performance of the MLRM differs across
the sub-regions and seasons, overall the use of WRs can explain more
than half of the variance in sub-regional wind power production, with
the eventual exception of summer (for some regions) and SE Iberia.
Recall that the MLRM is based on the observed monthly frequencies of
WRs and ignores the proximity of each daily WR to its respective cen-
troid. Bearing this in mind, the high explained variances found for most
regions and seasons show that changes in the frequency of occurrence
of WRs are more relevant for wind power output than the so-called
“within-type” changes, which account for the intra-WR spread of
weather patterns [43]. However, understanding such differences would
require more analyses that are beyond the scope of this study.
5. Case studies
In the previous section we have shown the ability of WRs as pre-
dictors of wind power production on monthly timescales. Although a
MLRM based on the monthly frequencies of WRs is able to reproduce
the time series of CF reasonably well for most regions and seasons, the
results indicate that wind power production is not always well pre-
dicted by the large-scale atmospheric circulation. In this section we
examine two recent periods to assess whether the use of WRs can help
understand the day-to-day variability of wind power production during
specific episodes of anomalous wind. The first case study is a summer
episode with wind power generation in Scotland well below the cli-
matology, while the second one presents an event with outstanding
production over Iberia. These case studies allow testing the WR per-
formance for extreme meteorological / climatological events, which are
expected to cause the largest deviations in power production with re-
spect to climatology.

8
J.M. Garrido-Perez, et al. Applied Energy 264 (2020) 114731

Fig. 7. Top panels: Daily evolution of WRs (color shading) and accumulated energy normalized by the climatological median at the end of the period (solid blue line)
during (a) June–July 2018 over Scotland and (b) March 2018 over Iberia. Grey dashed lines indicate the 25th and 75th percentiles and black line represents the
median of the winter 1991–2018 climatology of the normalized accumulated production. Bottom panels: Frequency distribution of days in each WR for the (c)
Jun–July and (d) March 1981–2010 climatology (boxes). The boxes extend from the lower (Q1) to the upper (Q3) quartile values of the data, with a horizontal line
indicating the position of the median (Q2). The whiskers extend from the boxes to show the range of the data between the 10th and 90th percentiles. Red circles
correspond to (c) the period Jun – July 2018 and (d) March 2018. The two boxes and red circles on the right represent the combined frequencies of WRs with high
and low production in the region of interest. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

5.1. The 2018 summer in Scotland
Summer 2018, in particular the months of June and July, was a
period of unusually hot and dry weather in the British Isles and other
parts of northern Europe, as reported by different monitoring centers
(e.g. Met [44]). Fig. 7a illustrates the daily evolution of WRs (color
shading), together with the accumulated energy for Scottish wind farms
from 1st June to 31st July 2018 (solid blue line) and for the Jun–Jul
1991–2018 climatology (median as solid black line and 75th/25th
percentiles as dashed grey lines). The accumulated energy has been
computed from the CFWF data under the simplistic assumption of an
equal installed capacity of 2.5 MW at each wind farm, i.e. the daily
CFWF values are multiplied by 2.5 MW to get daily production at each
farm and then integrated across all farms and days in the period. By
using an equal power across all farms (regardless of their actual in-
stalled capacity) we pretend to avoid local and regional biases in the
cumulative power production and therefore facilitate establishing links
to large-scale atmospheric circulation patterns. Finally, the estimated
curves have been normalized by the value of the climatological median
at the end of the period.
Overall, there was a large deficit of energy with respect to the cli-
matology during the two months. It is noteworthy the fact that the
steepest slopes of the accumulated energy (blue line) coincide with the
occurrence of WRs associated with high CFWF in the region (e.g. ZR and
AH between days 13 and 23), whereas the flattest slopes occur under
WRs associated with low CFWF (e.g. CL and EB for days 1 to 12 and 24
to 50) (see also Fig. 4 left). Consequently, the accumulated wind energy
production was relatively close to normal only around the 20th of June
and gradually deviated from the climatological median afterwards,
being the lowest ever recorded towards the end of July.
Fig. 7c shows the climatological frequencies of WRs during the June
- July months of the 1981–2010 period (boxplots), together with the

9
J.M. Garrido-Perez, et al.
frequencies during June - July 2018 (red circles). It proves that the
atmospheric circulation during the period of study was exceptional,
because the occurrence of EB was sevenfold its climatological median.
The two boxplots and red circles on the right also show the combined
frequencies for the WRs associated with high (AH + ZR + SL) and low
(CL + EB) CFWF in the UK (as determined from Fig. 4 left). The fre-
quency of occurrence was between the 10th and 25th percentile of the
climatology for the former group of WRs and clearly above the 90th
percentile for the latter, indicating conditions unfavorable to wind
power production. As climatological conditions were relatively in-
frequent, the very low wind power production during that summer can
be largely attributed to the high frequency of atmospheric blocking in
northern Europe.
5.2. March 2018 in Iberia
A recent study [26] has reported extraordinarily rainy and windy
conditions in Iberia during March 2018 that ended the most severe
drought at continental scale since at least 1979 [45]. Strong positive
anomalies of wind speed were present across Iberia. Therefore, we have
considered all wind farms in the region for the analysis of this case
study. Fig. 7b shows the daily evolution of accumulated energy during
that month (solid blue line) compared to the climatology (black and
grey lines), and the evolution of the WRs. Wind power in March 2018
was outstanding, well above the 75th percentile of the climatology of
the month. This difference becomes evident from the beginning of the
month and increases day after day, consistent with the outstanding
recurrence of AL (e.g. 1st–18th of March, blue shading), which is as-
sociated with high CFWF in the region (see also Fig. 4 right). This, to-
gether with the near-normal occurrence of other favorable WRs (SL and
EA), resulted in record-breaking accumulated energy in March.
Fig. 7d displays the frequencies of occurrence of each WR during
March 2018 compared to their March 1981–2010 climatology. It con-
firms the prevalence of WRs associated with high production
(AL + SL + EA), which occurred almost 90% of the days in March
2018, representing around 2.5 times the climatological median of that
month. In particular, the occurrence of the WR with the highest CFWF in
Iberia, i.e. AL, was by far over the 90th percentile of the climatology.
Ayarzagüena et al. [26] reported that at the end of February a negative
North Atlantic Oscillation (NAO-) pattern settled down and persisted
throughout most of March. This prevented subtropical air masses to
reach European mid-latitudes, resulting in enhanced wind speed and
precipitation in the south of the continent. Although the WRs used in
that work differ from ours, both NAO- and AL are characterized by si-
milar meteorological anomalies over Iberia, suggesting good con-
sistency between the large-scale patterns responsible for the extreme
precipitation and wind in March 2018.
As seen from these two case studies, the catalogue of WRs developed
in this work can easily be applied to understand the day-to-day evo-
lution of wind energy production during specific periods of interest, in
particular for extreme events. The combination of high frequency of
unfavorable WRs and low frequency of favorable WRs caused the wind
energy deficit of summer 2018 in Scotland, and vice versa in the case of
the March 2018 surplus in Iberia. In these case studies, the anomalies in
power output can largely be traced back to the anomalous occurrence of
one specific WR. Still, it is the balance between favorable and un-
favorable WRs that really explains the departures in accumulated en-
ergy. This stresses the need for accounting for the daily succession of
WRs.
6. Discussion
The customized WRs derived in this study have allowed dis-
criminating situations with different power production over the western
European façade. This classification of weather patterns has two main
advantages with respect to those employed in previous studies. First, it
10
Applied Energy 264 (2020) 114731
allows for year-round assessments of wind resources, as compared to
other classifications which are hampered by their limited applicability
to specific seasons. With this approach wind power output can be ex-
amined on a wide range of time scales, from the analysis of specific
episodes of anomalous power production to assessments of the seasonal
to interannual variability of wind resources. Second, our WRs are tai-
lored to the regions of study, namely the UK and the Iberian Peninsula,
which maximizes the response of wind power production to the at-
mospheric circulation.
The optimized approach presented here is able to account for up to
80% of the actual CFs over large parts of Spain, Germany, Benelux, the
British Isles and some regions of France on monthly scales (Fig. 5). Our
results improve those from other studies that have examined the impact
of large-scale circulation patterns on wind energy in the same regions.
Jerez et al. [18] used the main modes of atmospheric circulation
variability over the North Atlantic-European sector (NAO, EA and
SCAND patterns) to predict October to March monthly time series of
wind power averaged over Iberia from a regional climate model hind-
cast simulation. Their result (R2 = 0.41) seems to be consistent with
Correia et al. [16], who found similar winter correlations for Portugal
(R2 = 0.36–0.45). Zubiate et al. [21] extended the area of study to
western Europe, reproducing monthly average wind speeds during the
winter months in the British Isles (R2 = 0.57–0.72) and Portugal
(R2 = 0.46). Overall, our WR-based model performs similarly as the
previous analyses for Scotland and outperforms them for Iberia. We can
also compare our results with those from Cortesi et al. [32], who used a
set of four North Atlantic-European WRs for each month to reproduce
reanalysis near-surface wind speed variability across Europe. While the
skill of both models is qualitatively similar in winter, ours yields better
results in the transition seasons and summer (see e.g. their Fig. 6). This
is relevant for all sub-regions considered here, where wind production
in some of the transition seasons is comparable to that in winter. More
specifically, the average CFs in spring exceed those of winter for SW
and SE Iberia, whereas wind production in autumn and spring is
comparable to that in winter (above 80% of the average winter values)
for Scotland and the remaining Iberian regions, respectively.
These findings indicate that the adaptation of the spatial domain
used for the definition of WRs to the area of study, as done in the
present analysis, helps capture year-round wind power variability. This
improvement seems to compensate for the reduction in explained var-
iance that may occur when using yearly as compared to monthly or
seasonal WR classifications. Future work would improve the statistical
modelling framework used here by considering the within-type changes
of the atmospheric circulation, i.e. the deviations of the daily WRs from
their respective centroids. On the other hand, as mentioned above, our
annual WR classification allows applying a consistent group of WRs to
reproduce day-to-day wind speed variability during extreme events
regardless of the time of the year. As an illustration, we have applied
the catalogue of WRs to two recent periods of low wind in the United
Kingdom and high wind in Iberia, finding that the unusual occurrence
of unfavorable/favorable WRs largely explains the wind power energy
deficit/surplus in Scotland/Iberia.
These results emphasize that previous knowledge of WRs brings
improvements in wind power short-term forecasts, as compared to es-
timates based on the current weather (persistence) or on the “normal”
weather (climatology) in a given region. Understanding how short-term
wind speed variability at typical wind farm locations depends on the
occurrence of large-scale atmospheric circulation patterns has some
potential implications for monthly-to-seasonal predictability of wind
resources. This is of particular relevance for wind energy stakeholders
and managers, as the estimates of annual wind production and the as-
sociated budgets often rely on historical values. Note that numerical
models are considerably more skillful to reproduce large-scale circula-
tion patterns than local weather phenomena. In particular, for rela-
tively long time horizons, uncertainties derived from the influence of
local processes make the seasonal prediction of wind resources an
J.M. Garrido-Perez, et al.
almost unapproachable problem. Our methodology is particularly well
suited to seasonal forecasts of wind power resources because it is based
on the large-scale spatial configuration rather than on local aspects of
the atmospheric circulation. Moreover, the WR classification allows for
the treatment of model uncertainty, as it relies on the similarity of the
synoptic pattern to a predefined set of recurrent weather regimes within
an allowable limit. Similar “qualitative” approaches have been suc-
cessfully applied to account for structural model uncertainties in future
climate change projections [46]. Consequently, as the performance of
weather and seasonal climate forecasts of the atmospheric circulation
improves, the relationships established here will result in enhanced
forecast capability of wind generation. In this sense, recent studies have
shown substantial improvements in the seasonal prediction of the large-
scale atmospheric circulation over the Euro-Atlantic sector using both
model-based and statistical approaches (e.g. [47] and references
therein). Seasonal climate forecasting techniques have already been
applied with some success in other fields such as air pollution [48],
forest fires [49] or agriculture [50]. Hence, WRs like these presented
here can be employed in the future to forecast the intermittency of wind
and other renewable sources (e.g. photovoltaic).
7. Conclusions
In this work, we have explored the variability of wind power pro-
duction in the United Kingdom (UK) and the Iberian Peninsula from
daily to annual time scales. To derive wind production, we have used
wind speed data from a meteorological reanalysis dataset fed through a
typical power curve as well as high-resolution data available at more
than 100 wind farms in those regions. Wind capacity factors (CFs) from
both datasets display similar spatiotemporal patterns, indicating that
wind production is more effective in the UK than in Iberia as well as in
winter as compared to summer.
Wind power output has been related to weather patterns, as derived
from a customized set of weather regimes (WRs), in order to explore
their skill to drive wind power variability. To do this, we have modelled
sub-regional monthly time series of CFs by means of a multi-linear re-
gression model (MLRM) on the monthly frequencies of occurrence of
the WRs. The model yields skillful results for most seasons and sub-
regions, being able to explain up to around two thirds of the month-to-
month variability in regional power output. Overall, WRs are more
skillful in winter and the transition seasons than in summer, when
correlations decay due to the weakening of the large-scale circulation
and the reduced wind speeds in Europe. The reported response of wind
power resources to the atmospheric circulation during the transition
seasons is relevant as previous studies have mostly addressed the winter
season.
It is also remarkable that the skill of reconstructed CFs is compar-
able for areas with rather different regional climates such as Scotland
and most sub-regions in Iberia. While the MLRM performs only slightly
better in Scotland than in most Iberian sub-regions, we have identified
some regional differences in the sensitivity of wind production to the
atmospheric circulation. In particular, Southeastern Iberia is the area
where the model has most problems to reproduce the month-to-month
variability of wind energy generation, because of the prevalence of
local to mesoscale circulations over the large scale there. These results
advise against the use of national aggregate CFs when assessing the
atmospheric circulation patterns responsible for wind power variability
in regions where the wind speed distribution is highly sensitive to the
geographical location. Our findings also encourage the development of
regional forecast models for wind production, e.g. by using short-term
forecasts of WRs as predictors, which are less affected by model un-
certainty than local wind speed forecasts.
Compared to previous assessments, our approach has resulted in
improved capability to reproduce the evolution of wind energy re-
sources over the western European façade, including periods of anom-
alous production which are challenging for capacity planning in
11
Applied Energy 264 (2020) 114731
electricity generation. Our findings also show some predictive cap-
ability of WRs to reproduce the evolution of wind generation on longer
(monthly and seasonal) time scales. Although assessing the applic-
ability of our results to the long-term forecast of wind energy resources
is out of the scope of this study, this can be investigated in the future by
using seasonal forecast systems.
CRediT authorship contribution statement
Jose M. Garrido-Perez: Methodology, Software, Formal analysis,
Writing - original draft. Carlos Ordóñez: Conceptualization,
Methodology, Funding acquisition, Writing - review & editing. David
Barriopedro: Conceptualization, Methodology, Writing - review &
editing. Ricardo García-Herrera: Conceptualization, Writing - review
& editing. Daniel Paredes: Methodology, Writing - review & editing.
Acknowledgments
This study was partly funded by Iberdrola Renovables Energía
S.A.U. JMGP was supported by a predoctoral research grant awarded by
the Spanish Ministerio de Educación, Cultura y Deporte (FPU16/01972)
and CO by the Ramón y Cajal Programme of the Spanish Ministerio de
Economía y Competitividad [grant number RYC-2014-15036]. We also
acknowledge support from the Spanish National Projects STEADY
(CGL2017-83198-R) and JeDiS (RTI2018-096402-B-I00). We thank
ECMWF for providing the ERA-Interim reanalysis and Iberdrola for
making available the high-resolution CF data at their wind farms. The
authors thank two anonymous reviewers and the editors for their useful
comments.
Declaration Competing Interest
This work was partly funded by Iberdrola.
Appendix A. Supplementary material
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.apenergy.2020.114731.
References
[1] European Comission, 2018. EU Energy in figures. Statistical Pocketbook. ISSN:
2363-247X.
[2] Yang J, Chen B. Emergy-based sustainability evaluation of wind power generation
systems. Appl Energy 2016;177:239–46. https://doi.org/10.1016/j.apenergy.2016.
05.126.
[3] Zhang X, Ma C, Song X, Zhou Y, Chen W. The impacts of wind technology ad-
vancement on future global energy. Appl Energy 2016;184:1033–7. https://doi.org/
10.1016/j.apenergy.2016.04.029.
[4] Albadi MH, El-Saadany EF. Overview of wind power intermittency impacts on
power systems. Electr Power Syst Res 2010;80:627–32. https://doi.org/10.1016/j.
epsr.2009.10.035.
[5] Kang J, Wei Y, Liu L, Han R, Yu B, Wang J. Energy systems for climate change
mitigation: a systematic review. Appl Energy 2020;263:114602https://doi.org/10.
1016/j.apenergy.2020.114602.
[6] Katzenstein W, Apt J. The cost of wind power variability. Energy Policy
2012;51:233–43. https://doi.org/10.1016/j.enpol.2012.07.032.
[7] Ren G, Liu J, Wan J, Guo Y, Yu D. Overview of wind power intermittency: Impacts,
measurements, and mitigation solutions. Appl Energy 2017;204:47–65. https://doi.
org/10.1016/j.apenergy.2017.06.098.
[8] González-Aparicio I, Zucker A. Impact of wind power uncertainty forecasting on the
market integration of wind energy in Spain. Appl Energy 2015;159:334–49. https://
doi.org/10.1016/j.apenergy.2015.08.104.
[9] González-Aparicio I, Monforti F, Volker P, Zucker A, Careri F, Huld T, et al.
Simulating European wind power generation applying statistical downscaling to
reanalysis data. Appl Energy 2017;199:155–68. https://doi.org/10.1016/j.
apenergy.2017.04.066.
[10] Sanz Rodrigo J, Chavez Arroyo RA, Moriarty P, Churchfield M, Kosović B, Réthoré
PE, et al. Mesoscale to microscale wind farm flow modeling and evaluation. WIREs:
Energy Environ 2017;6(2):e214https://doi.org/10.1002/wene.214.
[11] Zhang F, Sun YQ, Magnusson L, Buizza R, Lin SJ, Chen JH, et al. What is the pre-
dictability limit of midlatitude weather? J Atmos Sci 2019;76(4): 1077–91. 10.
1175/JAS-D-18-0269.1.
J.M. Garrido-Perez, et al.
[12] Agüera-Pérez A, Palomares-Salas JC, de la Rosa JJG, Florencias-Oliveros O.
Weather forecasts for microgrid energy management: review, discussion and re-
commendations. Appl Energy 2018;228:265–78. https://doi.org/10.1016/j.
apenergy.2018.06.087.
[13] Foley AM, Leahy PG, Marvuglia A, McKeogh EJ. Current methods and advances in
forecasting of wind power generation. Renew Energy 2012;37:1–8. https://doi.org/
10.1016/j.renene.2011.05.033.
[14] Wang X, Guo P, Huang X. A review of wind power forecasting models. Energy Proc
2011;12:770–8. https://doi.org/10.1016/j.egypro.2011.10.103.
[15] Brayshaw DJ, Troccoli A, Fordham R, Methven J. The impact of large scale atmo-
spheric circulation patterns on wind power generation and its potential predict-
ability: a case study over the UK. Renew Energy 2011;36:2087–96. https://doi.org/
10.1016/j.renene.2011.01.025.
[16] Correia JM, Bastos A, Brito MC, Trigo RM. The influence of the main large-scale
circulation patterns on wind power production in Portugal. Renew Energy
2017;102:214–23. https://doi.org/10.1016/j.renene.2016.10.002.
[17] García-Bustamante E, González-Rouco JF, Navarro J, Xoplaki E, Luterbacher J,
Jiménez PA, et al. Relationship between wind power production and North Atlantic
atmospheric circulation over the northeastern Iberian Peninsula. Clim Dyn
2013;40:935–49. https://doi.org/10.1007/s00382-012-1451-8.
[18] Jerez S, Trigo RM. Time-scale and extent at which large-scale circulation modes
determine the wind and solar potential in the Iberian Peninsula. Environ Res Lett
2013;8. https://doi.org/10.1088/1748-9326/8/4/044035.
[19] Kirchner-Bossi N, García-Herrera R, Prieto L, Trigo RM. A long-term perspective of
wind power output variability. Int J Climatol 2015;35:2635–46. https://doi.org/10.
1002/joc.4161.
[20] van der Wiel K, Bloomfield HC, Lee RW, Stoop LP, Blackport R, Screen JA, et al. The
influence of weather regimes on European renewable energy production and de-
mand. Environ Res Lett 2019;14:094010https://doi.org/10.1088/1748-9326/
ab38d3/meta.
[21] Zubiate L, McDermott F, Sweeney C, O’Malley M. Spatial variability in winter
NAO–wind speed relationships in western Europe linked to concomitant states of
the East Atlantic and Scandinavian patterns. Q J R Meteorol Soc 2017;143:552–62.
https://doi.org/10.1002/qj.294.
[22] Trigo RM, Valente MA, Trigo IF, Miranda PMA, Ramos AM, Paredes D, et al. The
impact of North Atlantic wind and cyclone trends on European precipitation and
significant wave height in the Atlantic. Ann NY Acad Sci 2008;1146:212–34.
https://doi.org/10.1196/annals.1446.014.
[23] Michelangeli P-A, Vautard R, Legras B. Weather regimes: recurrence and quasi
stationarity. J Atmos Sci 1995;52:1237–56. https://doi.org/10.1175/1520-
0469(1995) 052<1237:WRRAQS>2.0.CO;2.
[24] Vautard R. Multiple weather regimes over the North Atlantic: analysis of precursors
and successors. Mon Weather Rev 1990;118:2056–81. https://doi.org/10.1175/
1520-0493(1990) 118<2056:MWROTN>2.0.CO;2.
[25] Vrac M, Yiou P. Weather regimes designed for local precipitation modeling: ap-
plication to the Mediterranean basin. J Geophys Res Atmos 2010;115:1–23. https://
doi.org/10.1029/2009JD012871.
[26] Ayarzagüena B, Barriopedro D, Garrido-Perez JM, Abalos M, de la Cámara A,
García-Herrera R, et al. Stratospheric connection to the abrupt End of the 2016/
2017 Iberian Drought. Geophys Res Lett 2018;45:12639–46. https://doi.org/10.
1029/2018GL079802.
[27] Cassou C. Intraseasonal interaction between the Madden-Julian Oscillation and the
North Atlantic Oscillation. Nature 2008;455:523–7. https://doi.org/10.1038/
nature07286.
[28] Cattiaux J, Vautard R, Cassou C, Yiou P, Masson-Delmotte V, Codron F. Winter 2010
in Europe: a cold extreme in a warming climate. Geophys Res Lett 2010;37:1–6.
https://doi.org/10.1029/2010GL044613.
[29] Quesada B, Vautard R, Yiou P, Hirschi M, Seneviratne SI. Asymmetric European
summer heat predictability from wet and dry southern winters and springs. Nat
Clim Chang 2012;2:736–41. https://doi.org/10.1038/nclimate1536.
[30] Yiou P, Goubanova K, Li ZX, Nogaj M. Nonlinear Processes in Geophysics Weather
12
Applied Energy 264 (2020) 114731
regime dependence of extreme value statistics for summer temperature and pre-
cipitation. Nonlinear Process Geophys 2008;15:365–78. https://doi.org/10.5194/
npg-15-365-2008.
[31] Grams CM, Beerli R, Pfenninger S, Staffell I, Wernli H. Balancing Europe’s wind-
power output through spatial deployment informed by weather regimes. Nat Clim
Chang 2017;7:557–62. https://doi.org/10.1038/NCLIMATE3338.
[32] Cortesi N, Torralba V, González-Reviriego N, Soret A, Doblas-Reyes FJ.
Characterization of European wind speed variability using weather regimes. Clim
Dyn; 2019. 10.1007/s00382-019-04839-5.
[33] Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, et al. The ERA-
Interim reanalysis: Configuration and performance of the data assimilation system.
Q J R Meteorol Soc 2011;137:553–97. https://doi.org/10.1002/qj.828.
[34] Matsueda M, Palmer TN. Estimates of flow-dependent predictability of wintertime
Euro-Atlantic weather regimes in medium-range forecasts. Q J R Meteorol Soc
2018;144:1012–27. https://doi.org/10.1002/qj.3265.
[35] Santos JA, Corte-Real J, Leite SM. Weather regimes and their connection to the
winter rainfall in Portugal. Int J Climatol 2005;25:33–50. https://doi.org/10.1002/
joc.1101.
[36] Yiou P, Cattiaux J, Ribes A, Vautard R, Vrac M. Recent trends in the recurrence of
north atlantic atmospheric circulation patterns. Complexity 2018;2018:1–8.
https://doi.org/10.1155/2018/3140915.
[37] Jézéquel A, Yiou P, Radanovics S. Role of circulation in European heatwaves using
flow analogues. Clim Dyn 2018;50:1145–59. https://doi.org/10.1007/s00382-017-
3667-0.
[38] Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, et al. The NCEP
NCAR 40-Year Reanalysis Project. Bull Am Meteorol Soc 1996;77:437–72. https://
doi.org/10.1175/1520-0477(1996) 077<0437:TNYRP>2.0.CO;2.
[39] Staffell I, Pfenninger S. Using bias-corrected reanalysis to simulate current and
future wind power output. Energy 2016;114:1224–39. https://doi.org/10.1016/j.
energy.2016.08.068.
[40] Molod A, Takacs L, Suarez M, Bacmeister J. Development of the GEOS-5 atmo-
spheric general circulation model: evolution from MERRA to MERRA2. Geosci.
Model Dev. 2015;8:1339–56. https://doi.org/10.5194/gmd-8-1339-2015.
[41] Rienecker MM, Suarez MJ, Gelaro R, Todling R, Bacmeister J, Liu E, et al. MERRA:
NASA’s modern-era retrospective analysis for research and applications. J Clim
2011;24:3624–48. https://doi.org/10.1175/JCLI-D-11-00015.1.
[42] Wilks DS. Statistical methods in the atmospheric sciences, 3rd ed., Academic Press,
Oxford, UK; 2011.
[43] Beck C, Jacobeit J, Jones PD. Frequency and within-type variations of large-scale
circulation types and their effects on low-frequency climate variability in Central
Europe since 1780. Int J Climatol 2007;27:473–91. https://doi.org/10.1002/joc.
1410.
[44] Met Office, 2018. Summer 2018 Summary. < https://www.metoffice.gov.uk/
climate/uk/summaries/2018/summer > (accesed July 2019).
[45] García-Herrera R, Garrido-Perez JM, Barriopedro D, Ordóñez C, Vicente-Serrano
SM, Nieto R, et al. The European 2016/17 Drought. J Clim 2019;32:3169–87.
https://doi.org/10.1175/jcli-d-18-0331.1.
[46] Zappa G, Shepherd TG. Storylines of atmospheric circulation change for European
regional climate impact assessment. J Clim 2017;30(16):6561–77. https://doi.org/
10.1175/JCLI-D-16-0807.1.
[47] Wang L, Ting M, Kushner PJ. A robust empirical seasonal prediction of winter NAO
and surface climate. Sci Rep 2017;7(1): 1–9. 10.1038/s41598-017-00353-y.
[48] Shen L, Mickley LJ. Seasonal prediction of US summertime ozone using statistical
analysis of large scale climate patterns. Proc Natl Acad Sci 2017;114(10):2491–6.
https://doi.org/10.1073/pnas.1610708114.
[49] Shabbar A, Skinner W, Flannigan MD. Prediction of seasonal forest fire severity in
Canada from large-scale climate patterns. J Appl Meteorol Climatol
2011;50(4):785–99. https://doi.org/10.1175/2010JAMC2547.1.
[50] Klemm T, McPherson RA. The development of seasonal climate forecasting for
agricultural producers. Agric For Meteorol 2017;232:384–99. https://doi.org/10.
1016/j.agrformet.2016.09.005.