You are on page 1of 26

Theoretical and Applied Climatology (2022) 148:1085–1110

https://doi.org/10.1007/s00704-022-03990-0

ORIGINAL PAPER

Wind energy potential assessment using Weibull distribution


with various numerical estimation methods: a case study in Mersing
and Port Dickson, Malaysia
Muhammad Aslam Mohd Safari1 · Nurulkamal Masseran2 · Muhammad Hilmi Abdul Majid2

Received: 11 October 2021 / Accepted: 18 February 2022 / Published online: 28 February 2022
© The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2022

Abstract
In this study, the Weibull distribution with various numerical estimation methods is utilized for the assessment of wind energy
potential in Mersing and Port Dickson, Malaysia, by considering different monsoon seasons, i.e., northeast and southwest
monsoons. This assessment is conducted based on hourly wind speed data obtained from the Malaysian Meteorological
Department and the Department of Environment, Malaysia. A total of 28 numerical estimation methods are presented,
and their performances are first investigated through a Monte Carlo simulation. The results of the Monte Carlo simulation
indicate that no particular method outperforms all other methods. In different settings, i.e., different sample sizes and shape
parameters, certain methods are more efficient than others in estimating the Weibull parameters. Based on the simulation
results, 24 out of 28 numerical estimation methods are utilized for wind energy potential assessment in Mersing and Port
Dickson. The performance efficiency of all the considered methods is evaluated using an integrated approach comprising four
goodness of fit criteria, the modified Kolmogorov − Smirnov, modified Anderson − Darling, modified Cramér–von Mises,
and power density error. Using this integrated approach allow us to obtain a single-value goodness of fit measure known as
the global score to determine the best method. Based on the best fitted Weibull model, the wind characteristics in both sites
are investigated. The wind characteristics during northeast monsoon is found to be favorable for wind energy generation in
Mersing with the northeast (0 − 40°) as the prominent wind direction. However, small-scale wind energy is the best option
for wind energy development in Mersing. On the other hand, the Port Dickson site demonstrates poor wind characteristics,
indicating that this area is not favorable for wind energy development.

Keywords  Weibull distribution · Numerical estimation methods · Wind characteristic · Wind energy · Comparative
analysis · Monsoon season

1 Introduction energy is being rapidly and extensively harnessed through-


out the world (Global Wind Energy Council 2020; Interna-
The utilization of renewable energy (RE) sources is a sus- tional Energy Agency 2020). Wind energy (or wind power)
tainable development initiative that aims to reduce depend- is the process of using wind turbines to convert wind into
ency on fossil fuel, address the crisis of energy shortages, a useful form of energy, such as electricity (Panwar et al.
and mitigate the effects of climate change (Panwar et al. 2011). As a major RE resource, wind energy has the poten-
2011). As one of the most well-known RE sources, wind tial to supply more than 40 times the annual global consump-
tion of electricity (Lu et al. 2009; Jung and Schindler 2017).
* Muhammad Aslam Mohd Safari Globally, in 2019, about 60.4 GW of wind energy capacity
aslam.safari@upm.edu.my was installed, and the total global capacity for wind energy is
1
currently greater than 651 GW, an increase of 10% from that
Department of Mathematics and Statistics, Faculty of 2018 (Global Wind Energy Council 2020). The utiliza-
of Science, Universiti Putra Malaysia, 43400 UPM Serdang,
Selangor, Malaysia tion and development of wind energy as an RE source offers
2 many advantages, including reduce operational cost (cost
Department of Mathematical Sciences, Faculty of Science
and Technology, Universiti Kebangsaan Malaysia, effectiveness), facile transportation, mitigate environmen-
43600 Bangi, Selangor, Malaysia tal pollution, and creating opportunities for employment,

13
Vol.:(0123456789)
1086 M. A. M. Safari et al.

economic activity, and research agenda (Masseran 2015a; maximum likelihood method, method of moments,
Abbasi and Abbasi 2016; Masseran and Razali 2016). graphical method, L-moment method, empirical method
To produce wind energy efficiently and economically, of Lysen, empirical method of Justus, energy pattern fac-
two crucial factors must be considered, the first being the tor method, and equivalent energy method (Allouhi et al.
location of the site of the wind system installation and the 2017; Guarienti et al. 2020; Shaban et al. 2020). In the
second identifying the wind speed characteristics at the previous research works, comparative study of several
site based on their statistical distributions (Akgül et al. NEMs for wind energy assessment is conducted in many
2016). The model used for the wind speed data must be countries such as Turkey (Akdağ and Güler 2015; Usta
appropriate, adequate, and accurate, as this model is used et al. 2018), Brazil (Rocha et al. 2012), Spain (Akdağ
to describe the variation and uncertainty of the wind speed and Güler 2018), India (Chaurasiya et al. 2018; Rahman
data for evaluating the available wind energy potential and Chattopadhyay 2020), Pakistan (Shoaib et al. 2019;
(Wais 2017a; Masseran 2018). Even a small error in mod- Sumair et  al. 2020), Egypt (Saleh et  al. 2012), Korea
eling the wind speed data will generate a large error in the (Kang et  al. 2018), and Cameroon (Signe et  al. 2019;
assessment of the potential energy output (Ucar and Balo Djiela et al. 2020; Kapen et al. 2020). For determining
2009; Jung et al. 2018). In the modeling of wind speed the best NEM, several goodness of fit criteria are used
data, a parametric distribution is often applied. To inves- such as root mean square error (RMSE), coefficient of
tigate the wind energy potential, researchers have used determination (R 2 ), Anderson–Darling statistic (AD),
various single parametric distributions to represent wind Kolmogorov–Smirnov statistic (KS), and chi-square sta-
regimes, including the Weibull, gamma, lognormal, Gum- tistic (χ 2). The purpose of the comparative study is to
bel, inverse Weibull, inverse Gaussian, generalized Lindley, determine the most efficient and suitable NEM in estimat-
etc. (Masseran 2015b; Akgül et al. 2016; Deep et al. 2020; ing Weibull parameters for the assessment of wind energy
Şahin and Türkeş 2020). Apart from the single parametric potential. Although many NEMs have been applied for
distribution, many authors have also used the mixture distri- determining Weibull parameters in the assessment of
bution (Hu et al. 2016; Shin et al. 2016; Mazzeo et al. 2018) wind energy potential, many other methods are available
and the non-parametric approach (Xu et al. 2015; Han et al. and can be considered for the same purpose. As such, the
2019) to estimate and evaluate the potential wind power in a comparative study of NEMs remains open to exploration,
particular region. Comprehensive reviews of assessments of investigation, and further discussion. In addition, we note
wind speed distribution can be found in (Wang et al. 2016; that the application of different GoF measures results in
Jung and Schindler 2017, 2019). different NEMs being identified as best for estimating
In the wind energy sector, the Weibull distribution is an Weibull parameters.
often used single parametric distribution (Wais 2017a, b). In this study, we conduct wind energy potential assess-
For instance, the Weibull model has been utilized for esti- ment in Mersing and Port Dickson, Malaysia using Weibull
mating the wind power corresponding to the wind turbine distribution by considering different monsoon seasons. In
capacity factor (Chang and Tu 2007; Shu et al. 2015; Arslan addition, we also present and consider of a total of 28 NEMs,
et al. 2020), assessing the performance of wind energy sys- many of which have been applied in assessments of wind
tems (Celik 2003, 2006), and mapping of wind resources and energy potential and then carry out a comparative study of
properties (Faghani et al. 2018; Jung and Schindler 2018; Li their performances. The contributions of this work are as
et al. 2018; Schallenberg-Rodríguez and Montesdeoca 2018; follows:
Silva et al. 2021). Some of the advantages that have made the
Weibull distribution popular in the wind industry include its 1. We analyze the wind speed data for wind energy assess-
flexibility, the fact that it has only two parameters, the ease ment in Mersing and Port Dickson, Malaysia, based on
with which its parameters are estimated, and its closed form different monsoon seasons.
expression (Carta et al. 2009). As the wind power output is 2. We conduct comprehensive comparative study of vari-
calculated based on the estimated model parameters, another ous NEMs for estimating Weibull parameters via Monte
important step in investigating wind energy potential when Carlo simulation.
using a parametric model such as the Weibull distribution is 3. We carry out comprehensive comparative study of vari-
the selection of an efficient method for estimating its param- ous NEMs based on wind speed data in Mersing and
eters. If the estimated Weibull parameters are biased, this Port Dickson and propose an integrated approach known
can cause an inaccurate estimate of the wind power output. as global score to determine the most efficient NEM.
The numerical estimation method (NEM) is the most
commonly used method for estimating Weibull param- Based on the results of this work, the overview on which
eters in wind energy potential assessment. Some examples monsoon seasons that contributes to the optimal wind power
of the NEMs are maximum likelihood method, modified output in the area of Mersing and Port Dickson, Malaysia, is

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1087

acquired. Also, the findings from this study provide insight support renewable energy technologies such as wind
and guidelines for practitioners and other researchers regard- energy (Safari et al. 2020). The geographical coordinates
ing the use of NEMs for estimating Weibull parameters in of Mersing site are 2° 27′ north and 103° 50′ east, while
assessments of wind energy potential. for Port Dickson site, the geographical coordinates are 2°
26′ north and 101° 51′ east. Figure 1 shows the location
of wind measurement site at Mersing and Port Dickson.
2 Site description, data, and monsoon In this study, the data set used consists of hourly wind
season in Malaysia speed and wind direction data obtained from the Malay-
sian Meteorological Department (Mersing site) and the
Generally, Malaysia as a whole experiences wet and Department of Environment Malaysia (Port Dickson site)
humid weather conditions, with average daily tempera- which measured at 10 m height. The data of Mersing is
ture is between 21 and 35 °C, and its annual rainfall is available from January 1, 2010, to July 18, 2013, and for
80% a year which ranging from 2000 to 2500 mm. The Port Dickson, the data is available from January 1, 2010,
daily wind speed in Malaysia is higher during the day, i.e., to December 31, 2013.
from sunrise to sunset, than that periods before sunrise The commercial wind turbine usually has different
and after sunset. Coastal areas, especially in east Penin- hub heights which are ranging from 25 to 160 m. Since
sular Malaysia, is considered as windiest area and could the observed wind speed data was measured at 10  m
benefit from wind energy. In this study, Mersing and Port height, it is crucial to make adjustment to the data so
Dickson are chosen as a case study area due to its poten- that the assessment of wind energy potential is carried
tial and strategic location for wind energy generation in out at appropriate hub height of wind turbine. In the lit-
Peninsular Malaysia. In addition, in particular to Mers- erature, the wind profile power law is often used to make
ing, it is reported that the rural residents are generally adjustment on the observed wind speed data at different

Fig. 1  Map of Peninsular
Malaysia where the red dots
denote the location of wind
measurement site at Mersing
and Port Dickson. Source:
Google (2021)

13
1088 M. A. M. Safari et al.

heights (Zhou et al. 2020). The wind profile power law is There are two monsoon seasons in Malaysia, i.e., the
expressed as follows: northeast monsoon and the southwest monsoon. The
( )𝛼 northeast monsoon season usually occurs in November
l and ends in March. During this period, the northern hemi-
v2 = v1 2 (1)
l1 sphere experiences winter where the temperature value
is low creating high air pressure zones. This causes air
Here, v1 is the measured wind speed data at height l1,
movement towards the Asian continent. At the same time,
v2 denotes the adjusted wind speed data at height l2, and α
the southern hemisphere experiences summer and its tem-
is the power law exponent. For neutral stability conditions,
perature values sometimes exceed 30° C. High temperature
value of α is assumed approximately 0.143 or 1/7 and most
conditions will create the low air pressure zones around
of previous studies apply this value for extrapolating the
the Australian continent and the surrounding oceans. As
wind speed data to a certain adequate height. Note how-
a result, there will be air movements from the Asian con-
ever that the value of α varies according to the atmosphere
tinent to the Australian continent. When this group of air
stability and geographic features (Crippa et al. 2021; Jung
moves out from the Asian continent, it will be refracted
and Schindler 2021). In fact, α is influenced by several
southeast by the Coriolis force and in turn produce the
parameters such as elevation, time of day, nature of terrain,
northeast monsoon wind system (Ku Kassim et al. 2007).
season, wind speed, temperature, and various thermal and
During this monsoon season, Malaysia experiences a wet
mechanical mixing parameters (Manwell et al. 2010). Thus,
season, and the wind speed ranges from 10 to 30 knots.
assuming a value of 0.143 may lead to less accurate wind
The southwest monsoon winds blow when the northern
extrapolation since using this value ignores both the tem-
hemisphere experiences summer from May to September.
poral and spatial variability of power law exponent. In this
During this season, the low air pressure zones are formed
study, the wind profile power law is utilized to extrapolate
in the northern hemisphere due to the high temperature
the measured wind speeds to adequate height at 80 m. To
conditions. At the same time, there will be high air pres-
reduce the error while extrapolating the wind speed, spa-
sure zones in the Australian continent and the surrounding
tially explicit but temporally constant mean value of α is
oceans as a result of its low temperature conditions. The
used. Jung and Schindler (2021) suggested the suitable value
air pressure differences between these two regions will
of α for Malaysia is about 0.211.
result in the formation of a pressure slope between them.
The wind speed and wind direction in Malaysia are
The cold air clusters from the Australian continent and
highly influenced by the occurrence of monsoon seasons.
surrounding oceans will move towards areas with low air

Fig. 2  The formation of a the northeast monsoon and b the southwest monsoon winds in Malaysia

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1089

pressure on the Asian continent. Once the wind crosses 1 ∑ 3 1 3


n

the equator, this wind group will be deflected northeast P= 𝜌 v = 𝜌v , (4)


2n i=1 i 2
by the Coriolis force resulting in the southwest monsoon
wind system (Ku Kassim et al. 2007). The formation of where vi is the observed wind speed (m/s), n is the sam-
both northeast and southwest monsoons is demonstrated ple size of the wind speed data, and ρ denotes the air
in Fig. 2. density (kg/m 3). The standard air density is taken to be
ρ = 1.225 kg/m3 (Arslan et al. 2014). Based on the Weibull
distribution, the wind power density, most probable wind
3 Weibull distribution function and wind speed, and maximum wind energy carrying by the wind
power density speed, denoted by P w (W/m 2 ), V mp (m/s), V me (m/s),
respectively, can be computed as follows (Islam et  al.
Assume that a random wind speed variable V follows a 2011):
Weibull distribution. The respective probability den-

sity function (PDF) and cumulative distribution func- ( )
2 ∫
1 1 3
tion (CDF) of its Weibull distribution are given by the Pw = 𝜌 v3 f (v;k, c)dv = 𝜌c3 Γ 1 + , (5)
2 k
following: 0

( ) [ ( )]
k v k−1 v k ( )
f (v;k, c) = exp − , v > 0, k > 0, c > 0 (2) 1 1∕k
c c c Vmp = c 1 − , (6)
k
and
( )
[ ( )] 2 1∕k
v k Vme = c 1 + , (7)
F(v;k, c) = 1 − exp − , v > 0, k > 0, c > 0, (3) k
c
where Γ(•) represents a gamma function, which can be
where v is the observed wind speed data (m/s), k denotes the expressed by:
shape parameter, and c represents the scale parameter (m/s). ∞
The dimensionless parameter k characterizes the shape of the
∫ (8)
wind speed distribution, whereby a high k value indicates a Γ(x) = tx−1 exp(−t)dt.
narrow distribution with winds concentrated around a cer- 0

tain value, and a low k value indicates a widely dispersed


wind distribution. This parameter value is high when there
is low variation in wind speed in a particular location. Note 4 Available numerical estimation methods
that when k = 2, the distribution is known as a Rayleigh dis- for estimating Weibull parameters
tribution. In most applications of the Weibull distribution
for modeling wind speed data, the k value is typically ranges In this section, we present a total of 28 NEMs that are used
from 1 to 3, i.e., k ∈ (1, 3] (Akdağ and Güler 2018). The for estimating the parameters of a Weibull distribution.
parameter c determines the quality of the wind, with a high
value for windy sites and a low value for sites with little 4.1 Graphical method (GM)
wind. Parameter c also depends on the average wind speed,
such that when the average wind speed is high, the value of By taking twice the logarithm of the Weibull CDF in Eq. (3),
c is also high. the equation for the GM can be expressed as the linear equa-
The wind power density is an important indicator of the tion as follows (Chang 2011):
potential of wind resources and for describing the amount
of wind energy at various wind speed values at a particu- 1n[−1n[1 − F(v)]] = k1n(v) − k1n(c). (9)
lar site or location (Mohammadi et al. 2016). Wind power
In this linear equation, k is the slope and –k ln(c) is the
density is also used to evaluate the performance of wind tur-
y-intercept of the line. Here, the values of F(v) are esti-
bines and to recommend the optimum type of wind turbine
mated using an empirical CDF and the parameter estimates
at a particular site (Chaurasiya et al. 2018). Based on the
obtained using the GM, ̂ kGM and ̂ cGM , can be determined via
observed wind speed data, the wind power density denoted
least squares regression.
by P (W/m2) can be determined using the following equation
(Pishgar-Komleh et al. 2015):

13
1090 M. A. M. Safari et al.

4.2 Method of moments (MoM) speed data, respectively, which can be computed as


follows:
The parameter estimates of the MoM are obtained by
1∑ ( )
n
equating the mean and standard deviation of the wind vl = 1n vi , (17)
speed data with the mean and standard deviation of the n i=1
Weibull distribution, which can be written as follows
(Zhang et al. 2020):
1 ∑[ ( ) ]2
n

( ) 𝜎l2 = 1n vi − vl (18)
1 n − 1 i=1
v = cΓ 1 + , (10)
k

[ ( ) ( )] 4.4 Method of multi‑objective moments (MUOM)


2 1 1∕2
𝜎 =c Γ 1+ − Γ2 1 + , (11)
k k
The MUOM is developed as a minimization of a multi-
where objective function based on the first three moments, which
correspond to the mean, variance, and skewness. It is defined
1∑
n
using the weighted sum method (scalarization) in the follow-
v= v, (12)
n i=1 i ing equation (Usta et al. 2018):
[ ( ) ]2 [ ( ) ]2 [ ( ) ]2
(19)
1 2 3
[ ]1∕2 𝜆1 cΓ 1 + − v + 𝜆2 c 2 Γ 1 + − v2 + 𝜆3 c3 Γ 1 + − v3
1 ∑(
k k k
n
)2
𝜎= vi − v (13)
n − 1 1=i where vr is r-th sample moment, which is given as follows:

1∑ r
n
Here, v denotes the mean wind speed and σ is the vr = v. (20)
standard deviation of the wind speed. Dividing the n i=1 i
square of Eq. (10) by the square of Eq. (11), we have
In addition, λ 1 , λ 2 , and λ 3 are weights in which
the following:
λ 1  + λ 2  + λ 3  = 1. Usta et  al. (2018) suggested that the
( )
value of each weight be 1/3, so each objective function
2 Γ2 1 + 1k
v
= ( ) ( ) has equal importance within the estimation procedure. In
(14)
𝜎2 Γ 1 + 2k − Γ2 1 + 1k this method, we use an optimization method to obtain the
parameter estimates ̂ kMUOM and ̂ cMUOM by minimizing the
Using Eq. (14), the parameter estimate ̂kMoM is deter- multi-objective function given in Eq. (19). In this study,
mined using a numerical method; in this study, we apply we applied this method using the nlminb function available
the bisection method. The parameter estimate ̂
cMoM is then in R software.
obtained using Eq. (10).
4.5 L‑moment method (LMM)
4.3 Method of logarithmic moment (MoLM)
The LMM estimates are determined by equating the
The log-moment estimates of the shape and scale param- L-moment of a sample to its population counterpart. The
eters of a Weibull distribution are given by the following first and second L-moments of the Weibull distribution are
(Teimouri et al. 2013): as follows:
√ ( )
1
𝜋2 𝜇1l = cΓ 1 + , (21)
̂
kMoLM = , (15) k
6𝜎 2l
( )( )
1 1
( ) 𝜇2l = cΓ 1 + 1 − 1∕k . (22)
k 2
vl − Ψ(1)
̂
cMoLM = exp , (16) The first two sample L-moments are as follows:
̂
kMoLM
1∑
n
where Ψ(1) = –0.5772156 and vl and 𝜎l2 are the sam- ml1 + = v = v, (23)
ple variance and the mean of the log-transformed wind n i=1 (i)

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1091

2 ∑ n
4.7 Empirical method of Justus (EMJ)
ml2 = (i = 1)v(i) − v, (24)
n(n − 1) i=1
In the empirical method proposed by Justus et al. (1978),
where v (i) is the ordered sample of the wind speed, parameters k and c are estimated using the following
v(1) ≤ v(2) ≤ ⋯ ≤ v(n) , and v denotes the mean wind speed. equations:
By equating 𝜇1l and 𝜇2l with ml1 and ml2 , respectively, and ( )−1.086
𝜎
performing some mathematical manipulation, an L-moments ̂
kEMJ = , (29)
estimator is obtained as follows: v

̂ 1n(2) v
kLMM = − ( ), cEMJ = (
̂ ),
ml (25) (30)
1n 1 − m2l Γ 1 + ̂1
1 kEMJ

where v and σ denote the mean and standard deviation of the


ml1 wind speed, respectively.
cLMM = − (
̂ ) (26)
1
1+ ̂
kLMM
4.8 Empirical method of Lysen (EML)
4.6 Alternative moment method (AMM)
In the empirical method introduced by Lysen (1983), the
parameter estimate ̂
kEML is computed using the same for-
The AMM introduced by Akdağ and Güler (2018) has wide-
mula used by the EMJ, Eq. (29). Parameter c is estimated
ranging parameter estimation capability. In this method, the
as follows:
parameter k is estimated as follows:
( )4 ( )3 ( )2 ( ) ( )−1∕̂kEML
a4 𝜎
+ a3 𝜎
+ a2 𝜎
+ a1 𝜎
+ a0 0.433 (31)
v v v v ̂
cEML = v 0.568 + ,
̂
kAMM ≈ (27) ̂
kEML
( )4 ( )3 ( )2 ( )
𝜎 𝜎 𝜎 𝜎
b4 + b3 + b2 + b1 + b0
v v v v where v is the mean wind speed.
where v denotes the mean wind speed, σ is the standard
deviation of the wind speed, and ai and bi for i = 0, 1, 2, 3,
4 are coefficients, the values of which are given in Table 1.
4.9 Maximum likelihood method (MLM)
The parameter c can then be determined explicitly using the
The MLM is well known to be an efficient method par-
following equation:
ticularly for a large sample size, in which parameters k
v and c are estimated using the following equations:
cAMM = (
̂ ),
Γ 1+ ̂1 (28)
kAMM ⎡∑ n
k
� � ⎤
⎢ vi 1n vi �n
� �⎥
1
=⎢ 1n vi ⎥
i=1
̂
kMLM − (32)
⎢ ∑ n n ⎥
⎢ vki i=1

⎣ i=1 ⎦

( )1∕̂kMLM
1 ∑ ̂kMLM
n
Table 1  Values of ai and bi coefficients when using AMM
̂
cMLM = v . (33)
a0 2.94843 b0 3.20694 × ­10−7 n i=1 i

a1 1.50722 b1 2.29887 Note that Eq. (32) must be solved numerically to obtain


a2 2.56734 b2 2.48525
kMLM . In this study, we applied the bisection method to
̂
a3 0.903164 b3 2.35103
solve Eq. (33).
a4 0.208995 b4 1

Source: Akdağ and Güler (2018)

13
1092 M. A. M. Safari et al.

4.10 Modified maximum likelihood method 4.12 Rank correlation method (RCM)


(MMLM)
Given the iterative characteristics of MLM, Teimouri and
The MMLM can be applied to estimate Weibull param- Nadarajah (2012) developed a simple estimator of the shape
eters when wind speed data are available in a frequency parameter of the Weibull distribution, which they called the
distribution format. In the MMLM, parameters k and c RCM. In the RCM, parameter k is estimated as follows:
are determined using the following respective equations
−1n(2)
(Seguro and Lambert 2000): ̂
kRCM = � � ��� �� ,
𝜎 n+1 (38)
1n 1 − 𝛾 √
⎡∑ k � � � � ∑
nb n nb � � � � ⎤−1 n−1
v 3
⎢ vi
1n vi g vi 1n vi g vi ⎥


̂
kMMLM = ⎢ i=1

i=1 ⎥ , (34) where v denotes the mean wind speed, σ is the standard devi-
⎢ ∑ nb � � P(v 0) ⎥
⎢ vki g vi ⎥ ation of the wind speed, and γ is the correlation between the
⎣ i=1 ⎦ wind speed data vi and their ranks ri. The parameter estimate
cRCM is determined explicitly using the same formula as that
̂
( )1∕̂kMMLM used in the MLM, Eq. (33), which can be given as follow:
∑ ̂k
nb
( )
P(v ≥ 0) i=1 i
1 (35)
̂
cMMLM = v MMLM g vi , ( )1∕̂kRCM
1 ∑ ̂kRCM
n
̂
cRCM = v (39)
n i=1 i
where vi is the wind speed central value in bin i, nb is the
number of bins, g(vi) denotes the frequency in the wind
speed variation within bin i, and P(v ≥ 0) is the probability 4.13 Maximum product of spacings method (MPSM)
that the wind speed equals or exceeds zero. Equation (34)
is solved numerically using the bisection method to obtain The MPSM requires maximizing the geometric mean of the
kMMLM , after which Eq. (35) is solved explicitly. In this
̂
spacings in the data, which are the differences between the
study, we applied the Freedman–Diaconis rule to compute values of the CDF at neighboring data points (Cheng and
the width of the bins (Freedman and Diaconis 1981). The Amin 1983). Let v(i) be an ordered sample of wind speed data,
optimal number of bins was then determined based on the v(1) ≤ v(2) ≤ ⋯ ≤ v(n) . The uniform spacing of a random sam-
computed width. ple from the Weibull distribution is defined as follows:
( ) ( )
4.11 Alternative maximum likelihood method Di = F v(i) ;k, c − F v(i−1) ;k, c , (40)
(AMLM) ∑
where F(x(0); k, c) = 0, F(x(n + 1); k, c) = 1, and ni=1 Di = 1 .
The simple calculation procedure known as AMLM was devel- In the MPSM, parameter estimates ̂ kMPSM and ̂cMPSM are
oped based on the iterative characteristics of MLM. In the obtained by maximizing the geometric mean of the spacings
AMLM, parameter k of the Weibull distribution is computed with respect to k and c:
as follows (Akdağ and Güler 2015): [ n+1 ]1∕(n+1)

⎡ ⎤
1∕2 G= Di , (41)
⎢ n(n − 1) ⎥ i=1
̂ 𝜋 ⎢ ⎥
kAMLM =√ �n � �n �2 ⎥ . (36)

6⎢ ∑ � � ∑ � � or, equivalently, by maximizing the following function:
n 1n2 vi − 1n vi ⎥
⎣ i=1 ⎦
1 ∑ ( )
i=1 n+1
H= 1n Di . (42)
Then, the parameter estimate ̂cAMLM is obtained explicitly n + 1 i+1
using the same formula as that used in MLM, Eq. (33), which
can be expressed as: In this study, we applied the nlminb function available in
R software to solve Eq. (42). Note that the MPSM is sensi-
( )1∕̂kAMLM tive to closely spaced observations, and particularly to ties
1 ∑ ̂kAMLM
n
̂
cAMLM = v . (37) (Cheng and Stephens 1989). When the ties are due to multi-
n i=1 i ple observations, repeated spacings are replaced by the cor-
responding likelihood.

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1093

4.14 Percentile method (PM) Table 2  Values of ei and fi coefficients when using the NEPFM
e0  − 0.220374 f0  − 1.27285
The quantile function of a Weibull distribution can be writ-
ten as follows: e1 3.27527 f1 3.69115
e2  − 5.78961 f2  − 2.60973
[ ]1∕k
vp = c −1n(1 − p) , 0 < p < 1. (43) e3 2.15143 f3  − 0.800468
e4 0.590396 f4 0.992007
By setting p = 1 − exp(− 1) ≅ 0.6321, we obtain the
Source: Akdağ and Güler (2015)
parameter estimate ̂
cPM as follows:

̂
cPM = v0.6321 , (44)
Table 3  Values of ci and di c0 0.225761 d0  − 0.35144
where v0.6321 is the 63.21th percentile of the wind speed data. coefficients when using the
NEPFM c1 0.134704 d1 0.711818
Then, by substituting ̂ cPM into Eq. (43), the percentile-based
estimator for k can be expressed as: Source: Akdağ and Güler
[ ] (2015)
1n −1n(1 − p)

kPM = ( ) ( ) , 0 < vp < v0.6321 (45)
1n vp − 1n v0.6321
̂ 3.69
kEFFM = 1 + 2 , (49)
As suggested by Hassanein (1971) and Seki and Yokoy- Epf
ama (1996), the optimum value of p is 0.31.

v
cEPFM = (
̂ )
4.15 Median and quartiles method (MQM) 1 (50)
Γ 1+ ̂
kEPFM
If the 25th percentile, median, and 75th percentile are v0.25 ,
The PDM is a variant of the EPFM, where the PDM
vm , and v0.75 , respectively, then parameter estimate ̂ kMQM is
estimate of parameter k is obtained using the following
determined as follows (Aukitino et al. 2017):
equation:
[ ]
1n 1n(0.25)∕1n(0.75) 1.572534 ( )
̂
kMQM = ( ) = ( ) (46) Γ 1 + 3
1n v0.75 ∕v0.25 1n v0.75 ∕v0.25 v3 k
Epf = ( )3 = ( ). (51)
v Γ3 1 + 1k
The parameter c can then be computed using the follow-
ing equation:
kPDM , Eq. (51) must be
To obtain the parameter estimate ̂
̂
cMQM =
vm solved numerically, and here we apply the bisection method
̂ (47) to do so. The PDM estimate ̂ cPDM of parameter c is then
1n(2)1∕kMQM
determined using Eq. (50).

4.16 Energy pattern factor and power density 4.17 Novel energy pattern factor method (NEPFM)
methods (EPFM and PDM)
The basic concept of the NEPFM introduced by Akdağ and
The EPFM proposed by Akdağ and Dinler (2009) has a Güler (2015) is to first match the first and third moments
simple formula and is easy to compute. The energy pattern of the dataset to the first and third moments of the distri-
factor Epf is defined as follows: bution. This method is based on considering the varia-
tion in the shape parameter between 1 and 15. Using this
v3 method, parameters k and c are estimated using the follow-
Epf = ( )3 , (48)
v ing equations:

where v is the mean wind speed and v3 is the sample mean ̂


k e4 E4 +e3 E3 +e2 E2 +e1 Epf +e0 ,
pf pf pf (52)
of the cubic wind speed, which can be calculated using NEPFM≈
f4 E4 +f3 E3 +f2 E2 +f1 Epf +f0
pf pf pf

Eq. (20). Parameters k and c can then be determined using


the following expressions:

13
1094 M. A. M. Safari et al.

( )
PWMBPM estimate for parameter c can then be obtained
v ̂2
kNEPFM + d 1̂
kEFFM + d0
̂
cNEPFM ≈ (53) as follows:
̂2
kNEPFM + c1̂
kEFFM + c0
⎛ ⎞
⎜ v3 ⎟
where Epf is an energy pattern factor that is calculated using ̂
cPWMBPM =⎜ � � ⎟, (58)
Eq. (48). The values of coefficients ei and fi for i = 0, 1, 2, 3, ⎜Γ 1 + ̂ 3

4 are given in Table 2, and those of coefficients ci and di for ⎝ kPWMBPM ⎠

i = 0, 1 are presented in Table 3. where v3 is the sample mean of the cubic wind speed.

4.18 Equivalent energy method (EEM)


4.20 Wind atlas analysis and application program
In the EEM, Weibull parameters k and c are obtained from the method (WAsPM)
equivalence of the energy densities obtained by the theoretical
curve and the observations (Rocha et al. 2012). In this method, The WAsPM is developed based on the following require-
the parameter estimate ̂kEEM is obtained by minimizing the ments (Aukitino et al. 2017):
following function:

⎡ ⎡ ⎛ ⎞
k
⎤ ⎡ ⎛� � ⎞
k
⎤⎤
2 1. Equal mean power densities based on the fitted Weibull

model and observed distribution.
nb
⎢ � �
1∕3
⎢ ⎜ vi − 1(Γ(1 + 3∕k))1∕3 ⎟ ⎥ ⎢ ⎜ vi (Γ(1 + 3∕k)) ⎟ ⎥⎥
⎢P vi − exp⎢−⎜ � �1∕3 ⎟ ⎥ + exp⎢−⎜ � �1∕3 ⎟ ⎥⎥
i=1 ⎢

⎢ ⎜
⎣ ⎝ v3 ⎟



⎢ ⎜
⎣ ⎝ v3 ⎟

⎥⎥
⎦⎦ (54) 2. The proportion of values above the mean observed wind

nb
� �2 speed for the fitted Weibull model must be the same to
= 𝜀i ,
i=1 those observed distribution.
where vi is the highest wind speed in bin i, nb is the number
Based on the first requirement above, the following solu-
of bins, P(vi) denotes the probability of the wind speed in the
tion for parameter c is obtained:
ith bin, and v3 is the sample mean of the cubic wind speed.
Note that to solve Eq. (54), we applied the nlminb function ⎛ ⎞
1∕3

available in R software. The parameter estimate ̂ cEEM is then ⎜ v3 ⎟


determined explicitly using the following expression: c=⎜ � �⎟ (59)
⎜Γ 1 + 3

1∕3 ⎝ k ⎠
⎛ ⎞
⎜ v3 ⎟ The empirical CDF, F(v), gives the proportion of values
̂
cEEM =⎜ � �⎟ , (55)
⎜Γ 1 + ̂ 3
⎟ that are less than v, thus 1–F(v) is the proportion of values
⎝ kEEM ⎠
that exceed v. Given the second requirement above, suppose
that X denotes the proportion of observed wind speeds that
exceed the mean observed wind speed, as follows:
4.19 Probability weighted moments based
on power density method (PWMBPM) 1 − F(v) = X. (60)

The PWMBPM is an innovative method for estimating the By equating X with the complementary
( ) CDF of the
parameters of a Weibull distribution with simple formulation Weibull distribution at v , 1 − F v;k, c  , and by applying
and less computation. The PWMBPM estimate for parame- some algebraic manipulations, the following equation is
ter k is obtained using the following expression (Usta 2016): obtained:
( )k
1n(2) v
̂
kPWMBPM = ( ) . = −1n(X). (61)
(56) c
1n C
By substituting Eq. (59) into Eq. (61), the WAsPM esti-
Here, C is directly determined by: mate ̂kWAsPM for parameter k is defined as the solution to the
v following equation:
C= ,

n
(57) k
2
v(i) (n − i) ⎛ ⎞
n(n−1)
i=1 ⎜ v ⎟
⎜� � ��1∕3 ⎟ = −1n(X). (62)
where v is the mean wind speed and v(i) is the ordered ⎜ v3 ∕Γ 1 + 3 ⎟
sample of the wind speed data, v(1) ≤ v(2) ≤ ⋯ ≤ v(n) . The
⎝ k ⎠

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1095

n [
To obtain the parameter estimate ̂
kWAsPM , X is first calcu- 1 ∑ ( ) 2i − 1 ]2
C(k, c) = + F v(i) ;k, c − (68)
lated using Eq. (60). Then, Eq. (62) is solved numerically 12n i=1 2n
using the bisection method. Having obtained the parameter
estimate ̂
kWAsPM , the WAsPM estimate ̂cWAsPM for parameter and
c can be determined using Eq. (59). ∑ 2i − 1 [ [ ( )] [ ( )]]
n
A(k, c) = −n − 1n F v(i) ;k, c + 1n 1 − F v(i) ;k, c .
i=1
n
(69)
4.21 Ordinary least squares and weighted least
squares methods (OLSM and WLSM) To solve Eqs. (67), (68), and (69), we applied the nlminb
function available in R software.
Suppose that v(1) ≤ v(2) ≤ ⋯ ≤ v(n) is the ordered sample of
wind speed from the Weibull distribution. It is known that: 4.23 Method of Mabchour (MoMab)
[ ( )]
E F v(i) =
i
(63) The MoMab was proposed by Mabchour (1999) for assess-
n+1 ing the wind energy potential in Morocco. The MoMab esti-
and mate for parameters k and c are obtained using the following
expressions:
[ ( )] i(n − i + 1)
Var F v(i) = . (64) [ ( )]0.51
(n + 1)2 (n + 2)
̂
kMoMab = 1 + 0.483 v − 2 , (70)

The OLS estimates ̂ kOLSM and ̂


cOLSM for parameters k and
v
c, respectively, are obtained by minimizing the following cMoMab = (
̂ ),
1 (71)
function: Γ 1+ ̂
kMoMab
n [ ]2
∑ ( ) i
L(k, c) = F v(i) ;k, c − (65)
i=1
n+1 4.24 Energy variance method (EVM)
Based on the results obtained by Eqs. (63) and (64), the The EVM considers the kinetic energy of wind, which is pro-
WLS estimates ̂ kWLSM and ̂cWLSM are determined by mini- portional to the square of the wind speed (Rahman and Chat-
mizing the following function: topadhyay 2020). In this simple method, parameters k and c
are estimated by the following equations:
∑ (n + 1)2 (n + 2) [ ( ) ]2
n
i
W(k, c) = F v(i) ;k, c − . (66)
i(n − i + 1) n+1 ⎛∑
1∕2
2⎞
n
i=1
⎜ vi ⎟
To solve the expressions in Eqs. (65) and (66), we applied ̂
kEVM =⎜ 2 ⎟
i=1
(72)
the nlminb function available in R software. ⎜ n𝜎 ⎟
⎜ ⎟
⎝ ⎠
4.22 Kolmogorov–Smirnov, Cramér‑von Mises,
and Anderson–Darling methods (KSM, CVMM, v
and ADM) cEVM = (
̂ ),
Γ 1+ ̂1 (73)
kEVM
The KSM, CVMM, and ADM are the class of the maximum
goodness of fit (MGF) estimation method. In general, the class where v and σ are the mean and standard deviation of the
of MGF methods estimates Weibull parameters by numeri- wind speed, respectively.
cally minimizing any of the empirical distribution function
statistics. Let v(i) be the ordered sample of wind speed data,
v(1) ≤ v(2) ≤ ⋯ ≤ v(n) . In the KSM, CVMM, and ADM, the
parameter estimates ̂ kKSM , ̂
kCVMM , ̂ kADM , ̂
cKSM , ̂
cCVMM , and 5 Integrated goodness of fit (GoF) measure
cADM are determined by minimizing, with respect to k and c,
̂
the respective functions D(k, c), C(k, c), and A(k, c), as follows: To examine the efficiency and accuracy of the NEMs, three
different GoF measures are considered—the modified KS
max ( ( ) i−1 i ( ))
1≤i≤n
D(k, c) = F v(i) ;k, c − , F v(i) ;k, c , (MKS), modified AD (MAD), and modified Cramér–von
n n Mises (MCVM) statistics proposed by Zhang (2002), which
(67)

13
1096 M. A. M. Safari et al.

are reported to be more powerful than the traditional GoF � � Zij

2𝜋 ∫
1 −t2

statistics. The MKS, MAD, and MCVM can be computed Φ Zij = √ e 2 dt. (79)
−∞
as follows:
( Step 4: Compute the GS by calculating the average score
i − 1∕2
1≤i≤n
max 1
MKS = (i − )log[ ] of the GoF statistics in step 3.
2 nF(v(i) ;k, c)
) (74) Φ(Zi1 )+Φ(Zi2 )+Φ(Zi3 )+Φ(Zi4 )
1 n − i + 1∕2 GS =  , (80).
+ (n − i + )log[ ] , 4
2 n(1 − F(v(i) ;k, c)) The NEM that produces the smallest GS is considered to
be the best method for estimating the Weibull parameters.
[ ( ) ( ( )) ]

n
logF v(i) ;k, c 1 − F v(i) ;k, c
MAD = − + log ,
i=1
n − i + 1∕2 i − 1∕2 6 Simulation study
(75)
A Monte Carlo simulation is conducted to investigate the
and
performances of the 28 NEMs in estimating the shape and
[ ( ( )−1 )]2 scale parameters of a Weibull distribution. The R statisti-

n
F v(i) ;k, c −1
MCVM = log . (76) cal software version 4.1.1 is used to perform this simula-
i=1
(n − 1∕2)∕(i − 3∕4) − 1 tion study. The details of the simulation design and results
are given in Supplementary Material. Overall, based on the
The power density error (PDE) is also utilized to evaluate simulation results, we can conclude no particular method
the performance of each NEM in estimating the wind power clearly outperforms the other methods in estimating Weibull
density. The PDE is obtained by parameters. For every particular case, i.e., for a certain
| P − Pw | sample size and certain shape parameter, different methods
PDE = || | × 100.
| (77) demonstrate the best performance. Without considering the
| P |
worst methods (WAsPM, MQM, PM and MoLM), all the
The global score (GS) based on the MKS, MAD, MCVM, methods perform somewhat similarly, with only a slight dif-
and PDE is then calculated to determine the best NEM. The ference in terms of Def. In fact, the difference in terms of
GS is computed as follows: Def between the ­1st and ­10th ranked methods is less than
0.75% for n = 300, less than 0.35% for n = 500, and less than
Step 1: Calculate the MKS, MAD, MCVM, and PDE val- 0.25% for n = 1000. Regarding the application of real wind
ues for each method. speed data, the methods that provide the best performance
Step 2: Standardize each of the GoF statistics by trans- will likely differ from those that obtained the best simulation
forming them into standard normal random variables. results, because the true values of the Weibull parameters are
unknown and the GoF measure must be applied to determine
kij − kj the best methods for estimating the Weibull parameters. We
Zij = , (78)
Sj advise the use of several GoF criteria to obtain a conclusive
result.
where zij is the standardized score, kij is the ith value of the This simulation study also obtained different results in
j GoF statistics, and kj and sj denote the mean and standard terms of PDE . We found that the best (or worst) methods
deviation of the j GoF statistics, respectively. for estimating Weibull parameters are not necessarily the
best (or worst) methods for estimating the wind power den-
Step 3: Transform the standardize score using the stand- sity. For instance, the WAsPM obtained the worst estima-
ard normal CDF to obtain a criteria value between 0 and tions of the Weibull parameters but demonstrated excellent
1.

Table 4  Descriptive statistics Data Sample size, n Min Max Mean Standard Skewness Kurtosis
of wind speed data for Mersing deviation
station
Overall 30,787 0.47 16.75 4.24 2.06 0.10 4.12
Northeast monsoon 12,937 0.47 16.75 4.96 2.28 0.58 3.16
Southwest monsoon 12,796 0.47 13.65 3.76 1.76 1.39 5.96

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1097

Fig. 3  Monthly mean wind


speed of Mersing site

Table 5  Descriptive statistics Data Sample size, n Min Max Mean Standard Skewness Kurtosis
of wind speed data for Port deviation
Dickson station
Overall 32,457 0.30 10.47 2.56 1.35 0.83 3.92
Northeast monsoon 12,986 0.30 8.63 2.63 1.37 0.77 3.21
Southwest monsoon 14,435 0.30 9.91 2.47 1.32 0.88 4.66

performance in estimating wind power density. The simula- 7.1 Descriptive statistics


tion results for PDE further support the claim that differ-
ent GoF measures can lead to different recommendations Table 4 provides descriptive statistics of the wind speed
for estimation methods. For the PDE, the best method is data for Mersing station. From Table 4, the minimum and
determined by that which minimizes the relative error of maximum wind speed are 0.47 m/s and 16.75 m/s, respec-
the empirical and theoretical power densities. We found the tively. Thus, the range of wind speed data is 16.28 m/s. The
EEM, WAsPM, and PWMBPM to provide high precision in mean wind speed for overall data is found to be 4.24 m/s,
estimating wind power density, with relatively small respec- and it can be observed that the mean wind speed during
tive PDE values close to zero. In this study, in order to obtain northeast monsoon, i.e., 4.96 m/s is higher than mean wind
a conclusive result, an integrated approach which consists speed during southwest monsoon, i.e., 3.76 m/s. In fact,
of four GoF criteria is proposed and adopted for determin- the monthly mean wind speed values in northeast monsoon
ing the best NEM in modeling the wind speed data using (except November) are mostly higher than the monthly wind
Weibull model (see Section 5). speed in southwest monsoon as shown in Fig. 3. In addition,
the wind speed standard deviation values are low, indicat-
ing that the data are close to the mean. Since the values of
7 Results and discussion skewness are positive and the kurtosis are greater than 3, it
is ascertained that the empirical wind speed distributions
As of the first step, we transform the wind speed data meas- are skewed to the right and its tails are heavier than normal
ured at 10 m height to wind speed data with 80 m height distribution.
using Eq. (1). Then, the transformed wind speed data are The descriptive statistics for Port Dickson wind speed
used throughout the analysis. data is presented in Table 5. It can be seen that the respective

13
1098 M. A. M. Safari et al.

Fig. 4  Monthly mean wind


speed of Port Dickson site

Table 6  Parameter estimates and GoF measures of Weibull distribution for overall wind speed data of Mersing
Method ̂
k ̂
c MKS MAD MCVM PDE (%) MKS(0,1) MAD(0,1) MCVM(0,1) PDE(0,1) GS Rank

GM 2.4171 4.8597 3.3588 411.8956 5043.3849 12.2531 0.5841 0.7010 0.9733 0.9092 0.7919 21
MoM 2.1751 4.8978 3.3477 259.2886 1910.0196 2.3391 0.3446 0.2705 0.2459 0.3262 0.2968 7
MUOM 2.0507 4.8461 3.3727 470.0247 2363.2394 0.0004 0.8363 0.8317 0.3788 0.1917 0.5596 17
LMM 2.2404 4.8972 3.3436 254.4569 2151.7267 4.8211 0.2656 0.2587 0.3137 0.4987 0.3342 11
AMM 2.1750 4.8978 3.3477 259.3189 1909.9863 2.3381 0.3446 0.2705 0.2459 0.3262 0.2968 8
EMJ 2.1954 4.8977 3.3459 240.5099 1951.3611 3.1375 0.3088 0.2262 0.2569 0.3796 0.2929 6
EML 2.1954 4.8997 3.3458 242.9138 1943.2623 3.0174 0.3068 0.2316 0.2547 0.3714 0.2911 5
MLM 2.1857 4.9077 3.3461 249.2746 1892.2438 2.1660 0.3127 0.2463 0.2412 0.3151 0.2788 4
MMLM 2.1920 4.9154 3.3451 260.7678 1879.0510 1.9516 0.2934 0.2741 0.2378 0.3015 0.2767 2
AMLM 2.4349 5.0240 3.3493 543.6817 3784.1345 3.5445 0.3776 0.9346 0.8104 0.4078 0.6326 19
RCM 2.2394 4.9329 3.3415 299.9456 1986.6606 2.6925 0.2290 0.3790 0.2665 0.3495 0.3060 10
MPSM 2.1848 4.9078 3.3461 249.1649 1890.2844 2.1247 0.3127 0.2461 0.2407 0.3124 0.2780 3
EPFM 2.1303 4.8977 3.3533 315.8846 1920.7414 0.4923 0.4637 0.4250 0.2487 0.2168 0.3386 12
PDM 2.1189 4.8975 3.3551 331.4242 1945.0298 6.98e − 05 0.5033 0.4709 0.2552 0.1916 0.3552 13
NEPFM 2.1189 4.8975 3.3552 331.4794 1945.1854 0.0019 0.5055 0.4710 0.2552 0.1917 0.3559 14
EEM 2.1683 4.9320 3.3466 272.8856 1810.8401 3.15e − 14 0.3225 0.3050 0.2205 0.1916 0.2599 1
PWMBPM 2.2404 4.9786 3.3398 363.8853 1869.2220 3.15e − 14 0.2016 0.5671 0.2353 0.1916 0.2989 9
WAsPM 1.9196 4.7329 3.4285 828.1834 3935.5363 3.15e − 14 1.0000 0.9999 0.8428 0.1916 0.7586 20
OLSM 2.3087 4.7484 3.3647 374.2366 4520.5609 15.3104 0.7047 0.5973 0.9326 0.9704 0.8012 22
WLSM 2.3008 4.8077 3.3542 286.3681 3538.7377 11.8658 0.4835 0.3411 0.7499 0.8973 0.6180 18
KSM 2.3445 4.7261 3.3725 474.1926 5675.1926 17.4780 0.8336 0.8394 0.9931 0.9886 0.9137 24
CVMM 2.3089 4.7484 3.3647 374.5148 4523.5342 15.3162 0.7047 0.5981 0.9329 0.9705 0.8015 23
ADM 2.2379 4.8293 3.3502 224.6542 2603.7415 8.6409 0.3966 0.1922 0.4572 0.7534 0.4498 16
EVM 2.2924 4.8963 3.3437 270.4319 2582.5189 6.6345 0.2674 0.2987 0.4502 0.6269 0.4108 15

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1099

Table 7  Parameter estimates and GoF measures of Weibull distribution for northeast monsoon wind speed data of Mersing
Method ̂
k ̂
c MKS MAD MCVM PDE MKS(0,1) MAD(0,1) MCVM(0,1) PDE(0,1) GS Rank

GM 2.4417 5.5760 3.3127 80.9920 564.9170 5.9355 0.4991 0.5367 0.9677 0.9932 0.7492 22
MoM 2.3136 5.6021 3.3105 66.6938 324.1458 0.7691 0.3506 0.3220 0.2919 0.3741 0.3347 7
MUOM 2.2641 5.5845 3.3146 89.5041 367.2204 8.77E − 05 0.6283 0.6635 0.4525 0.2308 0.4938 16
LMM 2.3266 5.6017 3.3100 61.9719 324.6109 1.2198 0.3191 0.2595 0.2935 0.4690 0.3352 9
AMM 2.3136 5.6021 3.3105 66.6833 324.1422 0.7700 0.3506 0.3219 0.2919 0.3743 0.3347 8
EMJ 2.3314 5.6015 3.3098 60.2699 325.8693 1.3850 0.3068 0.2385 0.2978 0.5046 0.3369 10
EML 2.3314 5.6033 3.3097 60.0417 324.6593 1.2945 0.3007 0.2358 0.2936 0.4851 0.3288 6
MLM 2.3206 5.6090 3.3100 63.1411 319.7299 0.6362 0.3191 0.2743 0.2770 0.3472 0.3044 5
MMLM 2.3233 5.6104 3.3099 62.0017 319.0190 0.6496 0.3129 0.2598 0.2746 0.3499 0.2993 3
AMLM 2.4520 5.6685 3.3100 120.9607 470.8121 1.4578 0.3191 0.9496 0.8190 0.5202 0.6520 21
RCM 2.3259 5.6114 3.3098 60.9270 318.6460 0.6817 0.3068 0.2465 0.2734 0.3564 0.2958 2
MPSM 2.3188 5.6093 3.3101 63.7673 319.6672 0.5613 0.3253 0.2825 0.2768 0.3324 0.3042 4
EPFM 2.2961 5.6026 3.3116 73.3289 330.0905 0.1531 0.4235 0.4188 0.3125 0.2567 0.3529 11
PDM 2.2919 5.6027 3.3119 75.0023 332.6617 0.0006 0.4440 0.4443 0.3216 0.2309 0.3602 12
NEPFM 2.2917 5.6027 3.3119 75.0705 332.7810 0.0033 0.4440 0.4454 0.3221 0.2313 0.3607 13
EEM 2.2515 5.5760 3.3162 96.6066 388.4722 2.25E − 14 0.7273 0.7572 0.5368 0.2308 0.5630 18
PWMBPM 2.3266 5.6247 3.3095 60.9154 312.9860 5.62E − 14 0.2888 0.2464 0.2550 0.2308 0.2552 1
WAsPM 2.1389 5.4937 3.3377 177.7472 733.1203 4.50E − 14 1.0000 0.9999 0.9998 0.2308 0.8076 24
OLSM 2.2660 5.5644 3.3151 92.0041 378.6222 1.1457 0.6606 0.6981 0.4977 0.4531 0.5774 19
WLSM 2.3225 5.5813 3.3108 66.2960 340.5845 2.1597 0.3701 0.3165 0.3504 0.6663 0.4258 14
KSM 2.3408 5.4903 3.3164 72.8100 520.4761 7.4283 0.7387 0.4110 0.9201 0.9995 0.7673 23
CVMM 2.2662 5.5643 3.3151 91.9072 378.4377 1.1591 0.6606 0.6968 0.4970 0.4560 0.5776 20
ADM 2.3056 5.5786 3.3117 73.1456 342.7056 1.7519 0.4303 0.4161 0.3582 0.5830 0.4469 15
EVM 2.3987 5.5990 3.3096 74.9688 406.4503 3.5688 0.2947 0.4438 0.6069 0.8830 0.5571 17

minimum and maximum wind speed are 0.30  m/s and wind speed data of Mersing site, respectively. For different
10.47 m/s for Port Dickson site. The mean for overall wind NEMs, we can see in Table 6 that the estimated parameter ̂ k
speed in Port Dickson is lower than Mersing site which is values range from 1.9196 to 2.4349 and those of estimated
2.56 m/s. In addition, as demonstrated in Fig. 4, there is not parameter ̂ c range from 4.7261 to 5.0240 m/s for overall
much different in terms monthly mean wind speed in both wind speed data. If we compare the estimated Weibull
monsoon seasons in Port Dickson. Similar to Mersing site, parameters between northeast monsoon and southwest mon-
the wind speed standard deviation values of Port Dickson soon as shown in Tables 7 and 8, the values of estimated
site are low, and the wind speed distribution is skewed to parameter ̂ k range from 2.1389 to 2.4520 and from 1.7325
the right and has heavier right tail as compared to normal to 2.8776 for northeast monsoon and southwest monsoon,
distribution. Based these characteristics of wind speed data respectively. Apart from that the values of estimated param-
of both sites, the Weibull distribution is considered as a suit- c range from 5.4937 to 5.6685 m/s for northeast mon-
eter ̂
able parametric model for describing the wind speed data. soon and from 3.9224 to 4.4503 m/s for southwest monsoon.
Given the high value of estimated parameter ̂ c for northeast
7.2 Assessment of wind energy potential using monsoon as compared to southwest monsoon, thus, we know
Weibull distribution that the Mersing area is windier during northeast monsoon
than during southwest monsoon.
The simulation results presented in Section 6 are used as a The estimated Weibull parameters and GoF results for
guideline for the selection of appropriate NEMs for estimat- the overall wind speed data, northeast monsoon wind speed
ing the Weibull parameters. For the analysis of wind speed data, and southwest monsoon wind speed data of Port Dick-
data, we considered only 24 NEMs, and excluded the four son site are shown in Tables 9, 10, 11, respectively. The
worst-performing NEMs, i.e., the PM, MQM, MoMab, and respective range of estimated shape parameter ̂ k for the
MoLM. Tables 6–8 show the estimated Weibull parameters overall, northeast monsoon and southwest monsoon are
and the GoF results for the overall wind speed data, north- 1.8175 − 2.1423, 1.7314 − 2.1914, and 1.9240 − 2.2147.
east monsoon wind speed data, and southwest monsoon Meanwhile, the estimated parameter ̂ c values are less than

13
1100 M. A. M. Safari et al.

Table 8  Parameter estimates and GoF measures of Weibull distribution for southwest monsoon wind speed data of Mersing
Method ̂
k ̂
c MKS MAD MCVM PDE MKS(0,1) MAD(0,1) MCVM(0,1) PDE(0,1) GS Rank

GM 2.6748 4.2000 3.4783 747.2987 12,968.0190 18.0395 0.5113 0.6305 0.6271 0.7829 0.6380 20
MoM 2.2620 4.2438 3.4229 292.4632 3439.5897 5.2329 0.2674 0.2291 0.2983 0.3563 0.2878 9
MUOM 1.9806 4.1443 3.5010 597.5213 3346.6119 0.0016 0.6157 0.4917 0.2954 0.2009 0.4009 16
LMM 2.4698 4.2379 3.4294 419.8127 6171.2281 11.5809 0.2931 0.3297 0.3880 0.5800 0.3977 15
AMM 2.2619 4.2438 3.4229 292.4651 3439.3771 5.2319 0.2674 0.2292 0.2983 0.3562 0.2878 8
EMJ 2.2808 4.2435 3.4218 291.2236 3581.6739 5.8857 0.2632 0.2283 0.3027 0.3783 0.2931 11
EML 2.2808 4.2449 3.4216 292.8383 3572.4225 5.7882 0.2625 0.2294 0.3025 0.3750 0.2923 10
MLM 2.2413 4.2494 3.4240 300.0497 3276.1236 4.1343 0.2717 0.2346 0.2933 0.3202 0.2799 5
MMLM 2.2445 4.2514 3.4235 301.8527 3284.6234 4.1133 0.2698 0.2359 0.2935 0.3195 0.2797 4
AMLM 2.7173 4.4503 3.4701 581.8759 9168.5233 3.3373 0.4730 0.4770 0.4936 0.2950 0.4346 17
RCM 2.4679 4.3442 3.4201 410.4687 5001.8209 4.7088 0.2568 0.3217 0.3485 0.3389 0.3165 13
MPSM 2.2397 4.2497 3.4241 300.5598 3265.0980 4.0566 0.2721 0.2350 0.2929 0.3177 0.2794 3
EPFM 2.1389 4.2445 3.4396 364.8096 2922.7945 0.4810 0.3354 0.2841 0.2825 0.2132 0.2788 2
PDM 2.1276 4.2444 3.4420 375.2561 2907.9219 8.02e − 05 0.3457 0.2925 0.2820 0.2008 0.2803 6
NEPFM 2.1276 4.2444 3.4420 375.3032 2908.0136 0.0015 0.3457 0.2926 0.2820 0.2009 0.2803 7
EEM 2.3641 4.3702 3.4129 444.9100 3610.3017 1.21e − 14 0.2303 0.3514 0.3036 0.2008 0.2716 1
PWMBPM 2.4698 4.4154 3.4199 511.9531 4544.8255 3.64e − 14 0.2560 0.4118 0.3335 0.2008 0.3005 12
WAsPM 1.7325 3.9224 3.7051 1164.2156 5557.1466 1.21e − 14 0.9964 0.9063 0.3671 0.2008 0.6177 19
OLSM 2.8772 3.9812 3.6458 1601.7482 39,105.3054 32.7591 0.9767 0.9907 0.9962 0.9823 0.9865 23
WLSM 2.8257 4.0205 3.5983 1379.8849 30,869.6969 30.1434 0.9242 0.9662 0.9731 0.9692 0.9582 22
KSM 2.6621 4.0305 3.5283 1005.5267 18,576.9284 27.3751 0.7304 0.8272 0.7958 0.9475 0.8252 21
CVMM 2.8776 3.9811 3.6460 1602.8515 39,148.8904 32.7664 0.9769 0.9907 0.9962 0.9824 0.9865 24
ADM 2.5063 4.1127 3.4605 604.0720 9301.5053 19.9660 0.4285 0.4979 0.4983 0.8302 0.5637 18
EVM 2.3590 4.2416 3.4209 293.0666 4380.7232 8.4232 0.2598 0.2296 0.3282 0.4674 0.3212 14

3.0300 m/s for overall, northeast monsoon and southwest • For sout hwest monsoon wind speed dat a,
monsoon wind speed data. In fact, the values of estimated EEM ≥ EPFM ≥ MPSM ≥ MMLM ≥ MLM.
parameter ̂ c for southwest monsoon are less than values of   Port Dickson:
c for northeast monsoon, suggesting that the Port Dickson
̂ • For overall wind speed data, PWMBPM ≥ EEM ≥ AML
is slightly windier during northeast monsoon than during M ≥ RCM ≥ EML.
southwest monsoon. Although the location of Port Dickson • For northeast monsoon wind speed data, PWMBPM ≥ M
is on the west coast of Malaysia (in Fig. 2), this area does not MLM ≥ MPSM ≥ MLM ≥ RCM.
experience high magnitude of wind speed during southwest • For southwest monsoon wind speed data, PWMBPM ≥ 
monsoon because the wind speed is blocked by the Island AMLM ≥ AMM ≥ MoM ≥ EML.
of Sumatra, Indonesia (Masseran and Razali 2016). Apart
from that, the southwest monsoon is characterized with low As shown in Tables 6–11, there are only slight difference
precipitation, less cloud, high outgoing long-wave radiation in the GS values of the top five methods for both sites. The
and often featured by dry epochs (Chenoli et al. 2018). graphical representations based on PDF and CDF plots of
On the basis of the GS values, which were computed from the Weibull distribution by the top five NEMs for the overall,
the MKS, MAD, MCVM, and PDE, we can conclude that northeast monsoon, and southwest monsoon wind speed data
the five best methods to estimate Weibull parameters for of Mersing and Port Dickson are displayed in Figs. 5, 6, 7,
Mersing and Port Dickson sites are as follows (rank from 8, 9 and 10. From these figures, it is clear that the Weibull
the best): distribution can adequately describe the wind speed data at
Mersing: both sites. The results show that the most efficient methods
for estimating the Weibull parameters are either PWMBPM
• For overall wind speed data, EEM ≥ MMLM ≥ MPSM ≥  or EEM. Moreover, as can be observed in Tables 6–11, the
MLM ≥ EML. EEM and PWMBPM, obtain very small PDE values close to
• For northeast monsoon wind speed data, PWMBPM ≥ R zero, which means that its estimated wind power density by
CM ≥ MLMM ≥ MPSM ≥ MLM. the Weibull distribution is almost identical to that calculated

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1101

Table 9  Parameter estimates and GoF measures of Weibull distribution for overall wind speed data of Port Dickson
Method ̂
k ̂
c MKS MAD MCVM PDE (%) MKS(0,1) MAD(0,1) MCVM(0,1) PDE(0,1) GS Rank

GM 2.0539 2.8842 3.3691 360.4586 2084.6059 5.4839 0.2656 0.2454 0.5028 0.6848 0.4246 13
MoM 1.9785 2.8875 3.3779 429.3162 1709.7095 1.4023 0.5032 0.6170 0.2465 0.2961 0.4157 11
MUOM 1.9196 2.8688 3.3927 499.1952 1816.3927 0.0016 0.8588 0.9031 0.3126 0.1882 0.5657 19
LMM 2.0151 2.8885 3.3723 393.7501 1811.9682 3.1944 0.3462 0.4161 0.3097 0.4644 0.3841 6
AMM 1.9785 2.8875 3.3779 429.3328 1709.6898 1.4015 0.5032 0.6171 0.2465 0.2960 0.4157 12
EMJ 2.0016 2.8882 3.3742 406.5512 1760.4455 2.5452 0.3980 0.4886 0.2770 0.4009 0.3911 7
EML 2.0016 2.8898 3.3740 406.0385 1754.5681 2.3782 0.3924 0.4857 0.2734 0.3849 0.3841 5
MLM 1.9884 2.8905 3.3760 418.5392 1717.6257 1.6240 0.4487 0.5570 0.2512 0.3154 0.3931 8
MMLM 1.9832 2.8881 3.3770 424.4236 1714.7825 1.5869 0.4774 0.5900 0.2495 0.3121 0.4073 10
AMLM 2.0792 2.9194 3.3646 330.0281 2135.6668 3.1558 0.1709 0.1303 0.5403 0.4606 0.3255 3
RCM 2.0149 2.8989 3.3715 390.7865 1773.8353 2.1311 0.3252 0.3996 0.2853 0.3616 0.3429 4
MPSM 1.9878 2.8906 3.3761 419.1320 1716.1312 1.5829 0.4516 0.5603 0.2503 0.3118 0.3935 9
EPFM 1.9623 2.8870 3.3810 446.0761 1701.4064 0.5696 0.5916 0.7047 0.2417 0.2287 0.4417 14
PDM 1.9515 2.8866 3.3833 457.6198 1708.1938 1.60e − 05 0.6544 0.7591 0.2456 0.1880 0.4618 15
NEPFM 1.9516 2.8866 3.3833 457.4807 1708.1165 0.0112 0.6544 0.7585 0.2456 0.1888 0.4618 16
EEM 1.9938 2.9091 3.3737 407.4746 1690.7068 1.21e − 14 0.3841 0.4939 0.2356 0.1880 0.3254 2
PWMBPM 2.0151 2.9199 3.3701 384.3511 1735.9153 1.21e − 14 0.2897 0.3644 0.2620 0.1880 0.2761 1
WAsPM 1.8175 2.8062 3.4369 658.5640 2721.9763 1.77e − 14 1.0000 0.9998 0.8818 0.1880 0.7674 24
OLSM 2.0985 2.8395 3.3747 336.3806 3088.3641 11.6763 0.4119 0.1505 0.9686 0.9785 0.6274 21
WLSM 2.0748 2.8648 3.3706 348.4828 2450.1565 8.2959 0.3022 0.1947 0.7524 0.8813 0.5327 18
KSM 2.1020 2.8380 3.3750 333.9794 3167.5687 11.9552 0.4204 0.1426 0.9776 0.9818 0.6306 23
CVMM 2.0986 2.8395 3.3747 336.3059 3090.2986 11.6825 0.4119 0.1503 0.9689 0.9786 0.6274 22
ADM 2.0309 2.8687 3.3729 385.2286 2014.4709 5.9366 0.3623 0.3691 0.4512 0.7237 0.4766 17
EVM 2.1423 2.8901 3.3674 345.5157 3157.7575 8.6518 0.2270 0.1832 0.9766 0.8980 0.5712 20

using measured data. This result is consistent with the simu- that the Port Dickson area is not suitable for wind energy
lation results, in which these two NEMs efficiently estimated development.
the wind power density.
Based on the best fitted Weibull model, the wind charac- 7.3 Wind direction analysis
teristics of both sites are studied. Tables 12 and 13 present
the wind characteristics of Mersing and Port Dickson sites, In wind energy potential assessment for a particular area,
respectively, involving the wind power density (Pw), most wind direction data also plays an important role. Knowing
probable wind speed (Vmp), and maximum wind energy car- the dominant directions of the wind enables the wind tur-
rying by the wind speed (Vme). It can be seen from Table 12 bine to be positioned in such a way so that the total amount
that the wind power density, most probable wind speed, of captured energy can be maximized, and the performance
and maximum wind energy carrying by the wind speed at of wind turbine can be optimized. Figures 11, 12, and 13
Mersing were highest during northeast monsoon with their shows the wind rose diagram and wind direction density
respective values 126.37 W/m 2, 4.42 m/s, and 7.34 m/s. of Mersing site for overall data, northeast monsoon data
On the other hand, the Mersing clearly shows poor wind and southwest data, respectively. Based on Fig. 11, it can
characteristics during southwest monsoon. Nevertheless, be seen that the Mersing wind regime has two dominant
with the wind characteristics presented in Table 12, the wind direction, i.e., northeast (0 − 40°) and southwest
Mersing area may not be suitable for large-scale electric- (230 − 270°). During northeast monsoon (Fig. 12), the wind
ity generation even during northeast monsoon. Small-scale direction is mostly between 0 − 50° and 250 − 360° with
wind turbines are probably the best option for wind energy wind speed more than 4.5 m/s is observed. While during
development in Mersing area. Based on Table 13, it can be southwest monsoon season (Fig. 13), the direction of wind
observed that the Port Dickson site has low values of wind is dominantly between 100 and 260° with large proportion
power density, most probable wind speed, and maximum of wind speed less than 4.5 m/s. Therefore, the prominent
wind energy carrying by the wind speed suggesting that wind direction of Mersing is northeast (0 − 40°) with pref-
this site has poor wind characteristics. Thus, it is clear erable wind speed for wind energy development. The wind

13
1102 M. A. M. Safari et al.

Table 10  Parameter estimates and GoF measures of Weibull distribution for northeast monsoon wind speed data of Port Dickson
Method ̂
k ̂
c MKS MAD MCVM PDE MKS(0,1) MAD(0,1) MCVM(0,1) PDE(0,1) GS Rank

GM 2.1717 2.9473 3.3503 116.3508 939.2568 10.3215 0.3200 0.2294 0.7250 0.9871 0.5654 20
MoM 2.0011 2.9666 3.3593 143.6553 642.5272 1.0061 0.4399 0.4731 0.3617 0.2796 0.3886 9
MUOM 1.9572 2.9529 3.3702 161.9512 737.8647 0.0011 0.5918 0.6494 0.4808 0.1874 0.4774 17
LMM 2.0229 2.9671 3.3555 135.9536 619.5166 2.0617 0.3879 0.3984 0.3344 0.3954 0.3790 8
AMM 2.0011 2.9666 3.3593 143.6615 642.5509 1.0053 0.4399 0.4731 0.3617 0.2795 0.3886 10
EMJ 2.0239 2.9671 3.3553 135.5953 618.7584 2.1112 0.3852 0.3950 0.3336 0.4011 0.3787 7
EML 2.0239 2.9688 3.3552 135.4177 616.9532 1.9487 0.3838 0.3933 0.3315 0.3823 0.3727 6
MLM 2.0262 2.9746 3.3544 134.0207 609.7400 1.4826 0.3731 0.3801 0.3231 0.3299 0.3516 4
MMLM 2.0282 2.9757 3.3540 133.2060 607.3802 1.4758 0.3678 0.3725 0.3204 0.3292 0.3475 2
AMLM 2.1914 3.0272 3.3433 133.9958 800.7390 3.5990 0.2376 0.3799 0.5610 0.5788 0.4393 14
RCM 2.0226 2.9735 3.3550 135.3454 613.3999 1.4183 0.3811 0.3926 0.3273 0.3229 0.3560 5
MPSM 2.0251 2.9748 3.3545 134.3662 610.4010 1.4156 0.3745 0.3834 0.3239 0.3226 0.3511 3
EPFM 1.9925 2.9663 3.3610 146.8204 655.3868 0.5764 0.4636 0.5042 0.3773 0.2377 0.3957 11
PDM 1.9811 2.9660 3.3635 151.0919 675.5291 2.62e − 05 0.4986 0.5461 0.4021 0.1873 0.4085 12
NEPFM 1.9812 2.9660 3.3635 151.0541 675.3405 0.0093 0.4986 0.5457 0.4018 0.1881 0.4086 13
EEM 1.9461 2.9467 3.3737 167.2479 772.3835 1.66e − 14 0.6388 0.6965 0.5250 0.1873 0.5119 19
PWMBPM 2.0229 2.9878 3.3542 133.7335 605.0511 1.54e − 14 0.3705 0.3774 0.3177 0.1873 0.3132 1
WAsPM 1.7314 2.8060 3.4933 325.3261 2158.2095 3.31e − 14 1.0000 1.0000 1.0000 0.1873 0.7968 24
OLSM 2.0231 2.9057 3.3633 142.7670 770.5798 8.0307 0.4958 0.4643 0.5227 0.9378 0.6052 22
WLSM 2.0437 2.9398 3.3554 131.8771 665.5501 5.7314 0.3865 0.3602 0.3897 0.8002 0.4842 18
KSM 2.0394 2.9034 3.3617 137.3492 788.0128 8.9912 0.4734 0.4117 0.5449 0.9661 0.5990 21
CVMM 2.0233 2.9056 3.3633 142.6900 770.8673 8.0475 0.4958 0.4636 0.5230 0.9384 0.6052 23
ADM 2.0059 2.9444 3.3604 144.4641 667.2005 3.4467 0.4552 0.4810 0.3918 0.5607 0.4722 16
EVM 2.1595 2.9686 3.3468 95.5515 805.2196 7.8979 0.2773 0.1049 0.5667 0.9327 0.4704 15

rose diagram and wind direction density of Port Dickson estimating Weibull parameters were assessed via a Monte
site for overall data, northeast monsoon data and southwest Carlo simulation. The simulation results revealed that no
data are respectively shown in Figs. 14–16. As shown in particular method clearly outperforms the others in estimat-
Figs. 14, 15 and 16, the Port Dickson wind regime only ing Weibull parameters. For different cases, i.e., with dif-
has one dominant wind direction even at different monsoon ferent sample sizes and different shape parameter values,
seasons which is centered around east (50–120°). However, different methods provide the best performance. In addi-
we can see that large proportion of wind speed at this site tion, when the sample size is large, there is not much differ-
are less than 4.5 m/s. ence in terms of the Def criteria of the top 10 NEMs, which
suggests that their performances are very similar. We also
found that the best method for estimating Weibull param-
8 Conclusion eters is not necessarily the best for estimating wind power
density and vice versa. For example, the WAsPM performed
In the assessment of wind energy potential using a Weibull worst in estimating Weibull parameters but demonstrated
distribution, NEMs are often used to estimate its param- excellent performance in estimating wind power density. In
eters. The use of an efficient NEM for estimating the Weibull terms of power density estimation, the EEM, WAsPM, and
parameters is essential to ensure accurate estimation of PWMBPM were highly accurate in estimating wind power
the wind power output. In this study, the Weibull model density, with their respective PDE values so small as to be
was utilized for the assessment of wind energy potential close to zero.
in Mersing and Port Dickson, Malaysia. We presented a The assessment of wind energy potential in Mersing and
total of 28 NEMs in this paper, including the GM, MoM, Port Dickson was conducted by considering two monsoon
MoLM, MUOM, LMM, AMM, EMJ, EML, MLM, MMLM, seasons, i.e., northeast monsoon and southwest monsoon.
AMLM, RCM, MPSM, PM, MQM, EPFM, PDM, NEPFM, We considered 24 out of the 28 NEMs and the remaining
EEM, PWMBPM, WAsPM, OLSM, WLSM, KSM, CVMM, four worst-performing NEMs, i.e., the PM, MQM, MoMab,
ADM, MoMab, and EVM and then their performances in and MoLM were excluded this assessments. In order to

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1103

Table 11  Parameter estimates and GoF measures of Weibull distribution for southwest monsoon wind speed data of Port Dickson
Method ̂
k ̂
c MKS MAD MCVM PDE MKS(0,1) MAD(0,1) MCVM(0,1) PDE(0,1) GS Rank

GM 2.0150 2.8538 1347.1617 251.0635 3.3962 1.2989 0.2657 0.2077 0.7106 0.3036 0.3719 14
MoM 2.0121 2.8384 1423.6989 269.0265 3.3933 1.6663 0.3022 0.2765 0.5205 0.3337 0.3582 4
MUOM 1.9240 2.8119 1331.7202 256.1946 3.4081 0.0021 0.2586 0.2263 0.9956 0.2084 0.4222 18
LMM 2.0700 2.8395 1651.7120 320.5609 3.3878 4.2460 0.4222 0.5165 0.1830 0.5648 0.4216 17
AMM 2.0121 2.8384 1423.6494 269.0113 3.3933 1.6654 0.3022 0.2765 0.5205 0.3337 0.3582 3
EMJ 2.0348 2.8389 1508.7931 291.5245 3.3904 2.8389 0.3454 0.3760 0.3255 0.4364 0.3708 13
EML 2.0348 2.8404 1505.4790 292.0955 3.3902 2.6683 0.3436 0.3786 0.3131 0.4210 0.3641 5
MLM 2.0093 2.8376 1409.6887 263.0472 3.3943 1.5685 0.2954 0.2524 0.5890 0.3256 0.3656 9
MMLM 1.9977 2.8308 1379.3735 252.7747 3.3960 0.9265 0.2808 0.2138 0.6986 0.2744 0.3669 10
AMLM 2.0458 2.8488 1473.4461 286.4233 3.3908 2.0441 0.3272 0.3524 0.3509 0.3659 0.3491 2
RCM 2.0697 2.8561 1616.7265 325.5057 3.3867 2.7276 0.4030 0.5408 0.1368 0.4263 0.3767 15
MPSM 2.0078 2.8378 1405.6126 261.8911 3.3945 1.4783 0.2934 0.2479 0.6025 0.3182 0.3655 8
EPFM 1.9824 2.8376 1367.6279 249.1042 3.3966 0.5589 0.2752 0.2009 0.7339 0.2469 0.3642 6
PDM 1.9711 2.8373 1347.2835 239.5118 3.3985 1.04E − 05 0.2657 0.1694 0.8301 0.2082 0.3684 11
NEPFM 1.9713 2.8372 1347.6065 239.6498 3.3985 0.0127 0.2659 0.1698 0.8301 0.2091 0.3687 12
EEM 2.0905 2.8948 1855.9906 394.9666 3.3830 1.33E − 14 0.5367 0.8309 0.0411 0.2082 0.4042 16
PWMBPM 2.0700 2.8855 1587.8264 335.4615 3.3855 1.33E − 14 0.3873 0.5890 0.0962 0.2082 0.3202 1
WAsPM 1.9549 2.8287 1350.3641 242.2465 3.3980 1.93E − 14 0.2672 0.1780 0.8072 0.2082 0.3652 7
OLSM 2.2144 2.8031 3748.2474 499.1684 3.3989 14.0053 0.9971 0.9875 0.8471 0.9919 0.9559 23
WLSM 2.1530 2.8158 2707.9441 433.7021 3.3898 10.4688 0.9022 0.9243 0.2889 0.9444 0.7650 22
KSM 2.1909 2.8044 1861.5424 345.8078 3.3872 6.4510 0.5398 0.6377 0.1567 0.7485 0.5207 19
CVMM 2.2147 2.8030 3752.9541 499.4309 3.3990 14.0179 0.9972 0.9876 0.8512 0.9920 0.9570 24
ADM 2.1118 2.8199 1920.1480 351.1876 3.3874 7.0188 0.5724 0.6622 0.1652 0.7882 0.5470 20
EVM 2.1679 2.8401 2654.2479 443.3375 3.3881 9.1986 0.8884 0.9398 0.1971 0.9034 0.7322 21

Fig. 5  Graphical representations
of fitted Weibull model based
on the top five NEMs for overall
wind speed data of Mersing site

evaluate the performance efficiency of all the considered It is found that the overall best NEM for both sites and for
NEMs, we used an integrated approach comprising four different monsoon seasons were determined to be either
GoF criteria, the MKS, MAD, MCVM, and PDE. Using the EEM or PWMBPM. The PDF and CDF plots showed
this integrated approach, we obtained a single-value GoF that the Weibull model realizes excellent fitting to the wind
measure known as the GS to determine the best method. speed data of Mersing and Port Dickson sites, which means

13
1104 M. A. M. Safari et al.

Fig. 6  Graphical representations
of fitted Weibull model based
on the top five NEMs for wind
speed data of Mersing site dur-
ing northeast monsoon

Fig. 7  Graphical representations
of fitted Weibull model based
on the top five NEMs for wind
speed data of Mersing site dur-
ing southwest monsoon

Fig. 8  Graphical representa-
tions of fitted Weibull model
based on the top five NEMs for
overall wind speed data of Port
Dickson site

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1105

Fig. 9  Graphical representations
of fitted Weibull model based
on the top five NEMs for wind
speed data of Port Dickson site
during northeast monsoon

Fig. 10  Graphical representa-
tions of fitted Weibull model
based on the top five NEMs for
wind speed data of Port Dickson
site during southwest monsoon

Table 12  Wind characteristics Data NEM ̂ ̂


c Pw (W/m2) Vmp (m/s) Vme (m/s)
at Mersing based on Weibull k
model
Overall EEM 2.1683 4.9320 90.30 3.71 6.67
Northeast monsoon PWMBPM 2.3266 5.6247 126.37 4.42 7.34
Southwest monsoon EEM 2.3641 4.3702 58.56 3.46 5.66

Table 13  Wind characteristics Data NEM ̂ ̂


c Pw (W/m2) Vmp (m/s) Vme (m/s)
at Port Dickson based on k
Weibull model
Overall PWMBPM 2.0151 2.9199 20.11 2.08 4.11
Northeast monsoon PWMBPM 2.0229 2.9878 21.46 2.13 4.20
Southwest monsoon PWMBPM 2.0700 2.8855 18.89 2.10 4.00

that the Weibull distribution is adequate for modeling and is windier than during southwest monsoon with their
describing the wind speed data. respective wind power density, most probable wind
Based on the best fitted Weibull model, the wind speed, and maximum wind energy carrying by the wind
characteristics at both sites were studied. The results speed were 126.37  W/m 2, 4.42  m/s, and 7.34  m/s. By
indicated that during northeast monsoon, Mersing area considering the wind characteristics of Mersing wind

13
1106 M. A. M. Safari et al.

Fig. 11  a Wind rose diagram and b wind direction density for overall data of Mersing site

Fig. 12  a Wind rose diagram and b wind direction density for northeast monsoon data of Mersing site

regime, small-scale wind turbines are the best option for Mersing for optimizing the performance of wind turbine
wind energy development in this area. Besides, northeast and for maximizing the total amount of captured energy.
(0 − 40°) was found to be the prominent wind direction in Meanwhile, the Port Dickson wind regime shows poor

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1107

Fig. 13  a Wind rose diagram and b wind direction density for southwest monsoon data of Mersing site

Fig. 14  a Wind rose diagram and b wind direction density for overall data of Port Dickson site

13
1108 M. A. M. Safari et al.

Fig. 15  a Wind rose diagram and b wind direction density for northeast monsoon data of Port Dickson site

Fig. 16  a Wind rose diagram and b wind direction density for southwest monsoon data of Port Dickson site

wind characteristics in terms of wind power density, most Supplementary Information  The online version contains supplemen-
probable wind speed, and maximum wind energy carrying tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00704-0​ 22-0​ 3990-0.
by the wind speed. Therefore, this area is not suitable for
wind energy development.

13
Wind energy potential assessment using Weibull distribution with various numerical estimation… 1109

Acknowledgements  The authors are indebted to the Malaysian Mete- Celik AN (2003) A simplified model for estimating the monthly perfor-
orology Department for providing the hourly mean wind speed and mance of autonomous wind energy systems with battery storage.
wind direction data. Additionally, the authors would like to thank the Renew Energy 28:561–572
editor and anonymous reviewer for their constructive comments, which Chang T-J, Tu Y-L (2007) Evaluation of monthly capacity factor of
helped us to improve the manuscript. WECS using chronological and probabilistic wind speed data: a
case study of Taiwan. Renew Energy 32:1999–2010
Author Contribution  Muhammad Aslam Mohd Safari: conceptualiza- Chang TP (2011) Performance comparison of six numerical methods
tion, methodology, software, formal analysis, writing—original draft, in estimating Weibull parameters for wind energy application.
project administration. Nurulkamal Masseran: supervision, writing— Appl Energy 88:272–282
review and editing, validation, funding acquisition. Muhammad Hilmi Chaurasiya PK, Ahmed S, Warudkar V (2018) Study of different
Abdul Majid: writing—review and editing, formal analysis, software, parameters estimation methods of Weibull distribution to deter-
validation. mine wind power density using ground based Doppler SODAR
instrument. Alexandria Eng J 57:2299–2311
Cheng RCH, Amin NAK (1983) Estimating parameters in continu-
Funding  This work was supported by the Universiti Kebangsaan
ous univariate distributions with a shifted origin. J R Stat Soc
Malaysia [grant number DIP-2018–038].
Ser B 45:394–403
Cheng RCH, Stephens MA (1989) A goodness-of-fit test using Moran’s
Data Availability  The wind speed and win direction data can be statistic with estimated parameters. Biometrika 76:385–392
requested from the Malaysian Meteorological Department and the Chenoli SN, Jayakrishnan PR, Samah AA et al (2018) Southwest
Department of Environment, Malaysia. monsoon onset dates over Malaysia and associated climatological
characteristics. J Atmos Solar-Terrestrial Phys 179:81–93
Code Availability  The R codes for all methods are available upon Crippa P, Alifa M, Bolster D et al (2021) A temporal model for vertical
request. extrapolation of wind speed and wind energy assessment. Appl
Energy 301:117378
Deep S, Sarkar A, Ghawat M, Rajak MK (2020) Estimation of the wind
Declarations  energy potential for coastal locations in India using the Weibull
model. Renew Energy 161:319–339
Conflict of Interests  The authors declare no competing interests. Djiela RHT, Kapen PT, Tchuen G (2020) Wind energy of Cameroon by
determining Weibull parameters: potential of a environmentally
friendly energy. Int J Environ Sci Technol 1–20
Faghani GHR, Ashrafi ZN, Sedaghat A (2018) Extrapolating wind data
References at high altitudes with high precision methods for accurate evalu-
ation of wind power density, case study: Center of Iran. Energy
Abbasi SA, Abbasi T (2016) Impact of wind-energy generation on cli- Convers Manag 157:317–338
mate: a rising spectre. Renew Sustain Energy Rev 59:1591–1598 Freedman D, Diaconis P (1981) On the histogram as a density estima-
Akdağ SA, Dinler A (2009) A new method to estimate Weibull tor: L 2 theory. Zeitschrift Für Wahrscheinlichkeitstheorie Und
parameters for wind energy applications. Energy Convers Manag Verwandte Gebiete 57:453–476
50:1761–1766 Global Wind Energy Council (2020) Global Wind Report 2019. Brus-
Akdağ SA, Güler Ö (2015) A novel energy pattern factor method for sels, Belgium
wind speed distribution parameter estimation. Energy Convers Google (2021) No Title. https://​maps.​googl​eapis.​com/​maps/​api/​geoco​
Manag 106:1124–1133 de/​json?​addre​ss=​Mersi​ng,+​Johor​&​key=​xxx
Akdağ SA, Güler Ö (2018) Alternative moment method for wind Guarienti JA, Almeida AK, Neto AM et al (2020) Performance analy-
energy potential and turbine energy output estimation. Renew sis of numerical methods for determining Weibull distribution
Energy 120:69–77 parameters applied to wind speed in Mato Grosso do Sul, Brazil.
Akgül FG, Şenoğlu B, Arslan T (2016) An alternative distribution Sustain Energy Technol Assess 42:100854
to Weibull for modeling the wind speed data: Inverse Weibull Han Q, Ma S, Wang T, Chu F (2019) Kernel density estimation model
distribution. Energy Convers Manag 114:234–240 for wind speed probability distribution with applicability to
Allouhi A, Zamzoum O, Islam MR et al (2017) Evaluation of wind wind energy assessment in China. Renew Sustain Energy Rev
energy potential in Morocco’s coastal regions. Renew Sustain 115:109387
Energy Rev 72:311–324 Hassanein KM (1971) Percentile estimators for the parameters of the
Arslan H, Baltaci H, Akkoyunlu BO et al (2020) Wind speed variability Weibull distribution. Biometrika 58:673–676
and wind power potential over Turkey: case studies for Çanakkale Hu Q, Wang Y, Xie Z et al (2016) On estimating uncertainty of wind
and İstanbul. Renew Energy 145:1020–1032 energy with mixture of distributions. Energy 112:935–962
Arslan T, Bulut YM, Yavuz AA (2014) Comparative study of International Energy Agency (2020) Renewables 2020. Paris, France
numerical methods for determining Weibull parameters for Islam MR, Saidur R, Rahim NA (2011) Assessment of wind energy
wind energy potential. Renew Sustain Energy Rev 40:820–825 potentiality at Kudat and Labuan, Malaysia using Weibull distri-
Aukitino T, Khan MGM, Ahmed MR (2017) Wind energy resource bution function. Energy 36:985–992
assessment for Kiribati with a comparison of different meth- Jung C, Schindler D (2017) Global comparison of the goodness-of-fit
ods of determining Weibull parameters. Energy Convers Manag of wind speed distributions. Energy Convers Manag 133:216–234
151:641–660 Jung C, Schindler D (2019) Wind speed distribution selection–A
Carta JA, Ramirez P, Velazquez S (2009) A review of wind speed review of recent development and progress. Renew Sustain
probability distributions used in wind energy analysis: case Energy Rev 114:109290
studies in the Canary Islands. Renew Sustain Energy Rev Jung C, Schindler D (2018) 3D statistical mapping of Germany’s wind
13:933–955 resource using WSWS. Energy Convers Manag 159:96–108
Celik AN (2006) A simplified model for estimating yearly wind frac- Jung C, Schindler D (2021) The role of the power law exponent in wind
tion in hybrid-wind energy systems. Renew Energy 31:105–118 energy assessment: a global analysis. Int J Energy Res 45:8484–8496

13
1110 M. A. M. Safari et al.

Jung C, Schindler D, Laible J (2018) National and global wind resource Saleh H, Aly AAE-A, Abdel-Hady S (2012) Assessment of different
assessment under six wind turbine installation scenarios. Energy methods used to estimate Weibull distribution parameters for wind
Convers Manag 156:403–415 speed in Zafarana wind farm, Suez Gulf, Egypt. Energy 44:710–719
Justus CG, Hargraves WR, Mikhail A, Graber D (1978) Methods for Schallenberg-Rodríguez J, Montesdeoca NG (2018) Spatial planning to
estimating wind speed frequency distributions. J Appl Meteorol estimate the offshore wind energy potential in coastal regions and
17:350–353 islands. Practical Case: the Canary Islands. Energy 143:91–103
Kang D, Ko K, Huh J (2018) Comparative study of different methods Seguro JV, Lambert TW (2000) Modern estimation of the parameters
for estimating weibull parameters: A case study on Jeju Island. of the Weibull wind speed distribution for wind energy analysis.
South Korea Energies 11:356 J Wind Eng Ind Aerodyn 85:75–84
Kapen PT, Marinette JG, David Y (2020) Analysis and efficient Seki T, Yokoyama S (1996) Robust parameter-estimation using the
comparison of ten numerical methods in estimating Weibull bootstrap method for the 2-parameter Weibull distribution. IEEE
parameters for wind energy potential: application to the city of Trans Reliab 45:34–41
Bafoussam, Cameroon. Renew Energy Shaban AH, Resen AK, Bassil N (2020) Weibull parameters evaluation
Ku Kassim KY, Ahmad A, Mahyam MI (2007) Keadaan laut perairan by different methods for windmills farms. Energy Rep 6:188–199
Semenanjung Malaysia untuk panduan nelayan. DPPSPM/DOF Shin J-Y, Ouarda TBMJ, Lee T (2016) Heterogeneous mixture distribu-
Li J, Wang X, Yu XB (2018) Use of spatio-temporal calibrated wind tions for modeling wind speed, application to the UAE. Renew
shear model to improve accuracy of wind resource assessment. Energy 91:40–52
Appl Energy 213:469–485 Shoaib M, Siddiqui I, Rehman S et al (2019) Assessment of wind
Lu X, McElroy MB, Kiviluoma J (2009) Global potential for wind- energy potential using wind energy conversion system. J Clean
generated electricity. Proc Natl Acad Sci 106:10933 LP – 10938. Prod 216:346–360
Lysen EH (1983) Introduction to Wind Energy. SWD Publication Shu ZR, Li QS, Chan PW (2015) Statistical analysis of wind character-
SWD 82–1, The Netherlands istics and wind energy potential in Hong Kong. Energy Convers
Mabchour H (1999) Etude modélisation et expérimentation des com- Manag 101:644–657
posants d’un système hybride couplant les énergies solaire et Signe EBK, Kanmogne A, Emmanuel GD, Meva’a L (2019) Comparison of
éolienne: Performances et méthodologie de dimensionnement seven numerical methods for determining Weibull parameters of wind
Manwell JF, McGowan JG, Rogers AL (2010) Wind energy for sustainable energy in Douala, Cameroon. Int J Energy Sect Manag
explained: theory, design and application. John Wiley & Sons Silva AR, Stosic T, Stosic B (2021) Wind speed persistence at the
Masseran N (2015a) Markov chain model for the stochastic behav- Fernando de Noronha archipelago, Brazil. Theor Appl Climatol
iors of wind-direction data. Energy Convers Manag 92:266–274 144:723–730
Masseran N (2018) Integrated approach for the determination of an Sumair M, Aized T, Gardezi SAR et al (2020) A novel method devel-
accurate wind-speed distribution model. Energy Convers Manag oped to estimate Weibull parameters. Energy Rep 6:1715–1733
173:56–64 Teimouri M, Hoseini SM, Nadarajah S (2013) Comparison of esti-
Masseran N (2015b) Evaluating wind power density models and their mation methods for the Weibull distribution. Statistics (ber)
statistical properties. Energy 84:533–541 47:93–109
Masseran N, Razali AM (2016) Modeling the wind direction behav- Teimouri M, Nadarajah S (2012) A simple estimator for the Weibull
iors during the monsoon seasons in Peninsular Malaysia. Renew shape parameter. Int J Struct Stab Dyn 12:395–402
Sustain Energy Rev 56:1419–1430 Ucar A, Balo F (2009) Evaluation of wind energy potential and electricity
Mazzeo D, Oliveti G, Labonia E (2018) Estimation of wind speed generation at six locations in Turkey. Appl Energy 86:1864–1872
probability density function using a mixture of two truncated Usta I (2016) An innovative estimation method regarding Weibull
normal distributions. Renew Energy 115:1260–1280 parameters for wind energy applications. Energy 106:301–314
Mohammadi K, Alavi O, Mostafaeipour A et al (2016) Assessing Usta I, Arik I, Yenilmez I, Kantar YM (2018) A new estimation
different parameters estimation methods of Weibull distribu- approach based on moments for estimating Weibull parameters
tion to compute wind power density. Energy Convers Manag in wind power applications. Energy Convers Manag 164:570–578
108:322–335 Wais P (2017a) A review of Weibull functions in wind sector. Renew
Panwar NL, Kaushik SC, Kothari S (2011) Role of renewable energy Sustain Energy Rev 70:1099–1107
sources in environmental protection: A review. Renew Sustain Wais P (2017b) Two and three-parameter Weibull distribution in avail-
Energy Rev 15:1513–1524 able wind power analysis. Renew Energy 103:15–29
Pishgar-Komleh SH, Keyhani A, Sefeedpari P (2015) Wind speed Wang J, Hu J, Ma K (2016) Wind speed probability distribution esti-
and power density analysis based on Weibull and Rayleigh dis- mation and wind energy assessment. Renew Sustain Energy Rev
tributions (a case study: Firouzkooh county of Iran). Renew 60:881–899
Sustain Energy Rev 42:313–322 Xu X, Yan Z, Xu S (2015) Estimating wind speed probability distribu-
Rahman SM, Chattopadhyay H (2020) A new approach to estimate tion by diffusion-based kernel density method. Electr Power Syst
the Weibull parameters for wind energy assessment: Case stud- Res 121:28–37
ies with four cities from the Northeast and East India. Int Trans Zhang J (2002) Powerful goodness-of-fit tests based on the likelihood
Electr Energy Syst 30:e12574 ratio. J R Stat Soc Ser B (Statistical Methodol) 64:281–294
Rocha PAC, de Sousa RC, de Andrade CF, da Silva MEV (2012) Zhang J, Zhang M, Li Y, et al (2020) Analysis of wind characteris-
Comparison of seven numerical methods for determining tics and wind energy potential in complex mountainous region in
Weibull parameters for wind energy generation in the northeast southwest China. J Clean Prod 274:123036
region of Brazil. Appl Energy 89:395–400 Zhou X, Qin J, Li HD et al (2020) A statistical method to construct
Safari MAM, Masseran N, Jedi A et al (2020) Rural public accept- wind speed at turbine height for study of wind power in China.
ance of wind and solar energy: a case study from Mersing, Theor Appl Climatol 141:419–432
Malaysia. Energies 13:3855
Şahin S, Türkeş M (2020) Assessing wind energy potential of Turkey Publisher's note Springer Nature remains neutral with regard to
via vectoral map of prevailing wind and mean wind of Turkey. jurisdictional claims in published maps and institutional affiliations.
Theor Appl Climatol 141:1351–1366

13

You might also like