Hindawi

Advances in Meteorology
Volume 2020, Article ID 3638423, 11 pages
https://doi.org/10.1155/2020/3638423

Research Article
Comparison of Weibull Estimation Methods for Diverse Winds

Ferhat Bingöl
Izmir Institute of Technology, Department of Energy Systems Engineering, 35430 Urla, Izmir, Turkey

Correspondence should be addressed to Ferhat Bingöl; ferhatbingol@iyte.edu.tr

Received 22 October 2019; Revised 13 March 2020; Accepted 13 June 2020; Published 6 July 2020

Academic Editor: Pedro Salvador

Copyright © 2020 Ferhat Bingöl. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

Wind farm siting relies on in situ measurements and statistical analysis of the wind distribution. The current statistical methods
include distribution functions. The one that is known to provide the best fit to the nature of the wind is the Weibull distribution
function. It is relatively straightforward to parameterize wind resources with the Weibull function if the distribution fits what the
function represents but the estimation process gets complicated if the distribution of the wind is diverse in terms of speed and
direction. In this study, data from a 101 m meteorological mast were used to test several estimation methods. The available data
display seasonal variations, with low wind speeds in different seasons and effects of a moderately complex surrounding. The results
show that the maximum likelihood method is much more successful than industry standard WAsP method when the diverse
winds with high percentile of low wind speed occur.

1. Introduction ideally preferable. The final aim is to use the collected data
for creating sector wise and generalized wind statistics for
One of the first and most crucial steps for wind farm in- the measurement location. The generalized wind statistics
vestment is to investigate the characteristics of the local wind can be used to create regional wind climates, also known as
resources. The aim is to select the optimum locations for the wind atlases. Since the atlas represents the larger domain,
turbines, maximizing income and minimizing cost. The one can statistically transfer the in situ measurements to
analysis most commonly relies on in situ measurements desired wind turbine locations. After several iterations of the
collected through conventional meteorological masts (met. process, the best production numbers are calculated, which
masts) [1, 2]. Recently, remote sensing devices with vertical will maximize the income of the wind farm
wind profiling capabilities have also been accepted by the For many years, wind data have been represented by the
market, but these devices have limitations, mostly related to Rayleigh distribution function [7, 8]. Although they are
the complexity of the terrain [3–6]. Therefore, the meth- similar and come from the same statistical methodology, for
odology of spatial modeling based on point measurements is almost over three decades, the Weibull distribution function
practically the only acceptable methodology for the wind site has been the choice of experts, mainly because it is con-
assessment step. A conventional met. mast is usually erected sidered to be more representative of the natural wind
on a location that is considered representative of the area of characteristics [9]. One of the first studies that used the
interest. Sometimes more than one met. mast may be Weibull function as the main statistical methodology was the
necessary if the wind characteristics are particularly variable European Wind Atlas [10], which leads to the industry
in an area. Cup anemometers, vanes, sonics, and atmo- standard Wind Atlas Methodology. The Weibull distribution
spheric sensors are used as measurement devices. It is im- function (equation (1)) can be presented as a cumulative
portant that data collection has a high recovery rate in order (CDF) (equation (2)) or probability (PDF) (equation (3))
to accurately capture the wind resources, and a rate of 90% is density distribution function where U is the wind speed, A is
2 Advances in Meteorology

the scale of the wind speed (m/s), and k is the unitless shape ignored; meanwhile, the final recovery rate for the channels
parameter [11]. dropped down to 99%. The data can be considered to be of
1 high quality in terms of availability. The air density was
U � AΓ􏼒1 + 􏼓, (1) measured to be 1.19 kg/m3 on average and only changes
k
±0.01 kg/m3 throughout the day in monthly and yearly
k average but can sometimes reach 1.01 kg/m3 in the occa-
F(U) � 1 − e− (U/A) , (2)
sional 10-minute sample.
The location of the met. mast was chosen for research
k U k− 1 − (U/A)k purposes (Figure 1), such that each wind sector has different
P(U) � 􏼒 􏼓 e . (3)
A A combinations of terrain and roughness classifications. The
main wind direction is the first sector, S1, where the
The common way of using the distribution function is to
northerly wind occurs more than 45% of the time. S1 is
fit measured wind speed frequency data to the probability
located 5 km away from coastline and has a flat terrain at
density function to get the relationship between the wind
50 m a.s.l., covered with grass. S2 is only 1 km away from the
speed and the wind frequency with two-parameter Weibull
coastline and is under the influence of the sudden roughness
distribution. Thereafter, one can define the observational
change. S3, S4, and S5 are occupied by a small village with
climate of the location by only recording two characteristics,
narrow streets, where the tallest houses are approximately
A and k.
6.5 m high. The village is built parallel to the coastline and
Although the method looks straightforward, there are
the sea is nearly 1.25 km from the met. mast at its closest
known problems. The most important one is diverse wind
location. S6 and S7 are occupied by the IZTECH campus,
conditions, occurring with high frequency during the year,
with the buildings closest to the mast being 4 m high at
which add extra uncertainty to computational fitting pa-
maximum, while further away (> 12 m distance), there are
rameters; this is the case in all used methods [12]. In this
taller university buildings but more sparsely located. A hill
study, a dataset with high recovery rate from a semicomplex
parallel to the coastline with an average height of 350 m a.s.l.
terrain at 101 m height was used to compare different es-
covers the grounds of S8–S12. The vegetation of these sectors
timation methods of the Weibull parameters. The aim was to
is characterized by low bushes (max. 30 cm) mixed with
perform a comparison between the most widely used
grassland. The only unique sector within this range is sector
methods for a representative height for common wind
S9, where 5 units of 3 MW wind turbines are located more
turbines.
than 5 km away from the mast. Due to the long distance, it is
In the next section, studies on distribution function for
assumed by the author that the met. mast is not under the
wind energy use and specifically Weibull distribution are
influence of these turbines (Table 2).
discussed, followed by the experimental setup and presen-
Vertical wind profiles were calculated with all the data
tation of available wind data used with five different esti-
and the yearly statistics show an almost logarithmic vertical
mation methods. Finally, the results are compared and
wind speed profile except for the minor speed up at 30 m
discussed.
height (Figure 2(a)). The wind directional turn shows minor
terrain effects above 50 m where the averaged wind direction
2. Experimental Setup and General becomes nearly stable (Figure 2(b)). All five wind speed
Wind Characteristics channels were also analyzed for diurnal statistics, which
demonstrate that the measurement location is characterized
Izmir Institute of Technology (IZTECH) is located on the by an almost laminar flow at night and quite unstable
west coast of Turkey, around 40 km to the west of the city of conditions during daytime (Figure 3). Based on the yearly
Izmir at the village of Urla. The peninsula surrounding the averaged wind speed values, the omnidirectional wind shear
area has more than 20% of the total installed wind power was calculated with the power law function (equation (4)) as
capacity of Turkey. In August 2017, a mast with 101 m height 0.18, being within the safety limits of 0.0 and 0.2 as the IEC
was erected in the university campus with several instru- 61400-1 standard suggests. The dominant northerly wind
ments mounted (Table 1) at N38.3332° E26.6326° geographic direction’s wind shear is even lower, at 0.09. The wind
coordinates. The reasoning of the new met. mast is to make sectors from the urban areas, S3 to S5, display wind shear
academic studies about wind speed distributions, atmo- and turbulence values above the acceptable ranges; it must
spheric stability, and complex terrain short-term predic- be noted, however, that few data points were available for
tions. Therefore, the mast is equipped with also two 3D sonic these sectors, hindering the wind shear calculations, as
anemometers and measurement devices in different heights. discussed later in the manuscript.
Data collection started on the 1st of August 2017 and the
α
campaign is ongoing. The best recovery rate, with over 99.9% U1 z
for a full year’s data, is available between the 21st of De- � 􏼠 1􏼡 . (4)
U2 z2
cember 2017 and the 21st of December 2018, totaling 52191
of 10-minute samples. The available data were filtered with Monthly statistics of the top-mounted anemometer,
two rules: (i) the standard deviation of the channels is bigger WS101, show high wind speeds during the fall and winter
than zero and (ii) wind speeds below a calm threshold level periods, while the averaged wind speeds get lower in the
for cup and vane of 0.3 m/s and 0.4 m/s, respectively, are summer and spring as expected due to the local weather
Advances in Meteorology 3

Table 1: Available instrumentation on the IZTECH met. mast with Two-parameter Weibull function is used with each dataset
recovery rate for the period between 21 December 2017 and 21 split with sector wise wind direction; each holds 30° sections.
December 2018. The wind speed and the wind direction are The relationship between the mean wind speed U and the
measured with sonic 3D anemometer at 10 m and 52 m, while other Weibull parameters, similar to equation (1), is given in
heights are measured with conventional anemometers.
equation (5); A is the scale and k is the shape parameter as
Parameter Unit Heights (m) before. Mean value of the wind speed can be calculated.
Wind speed (m/s) 101, 76, 52, 30, 10 Power density, Pd, can also be computed from the third
Wind direction ° 98, 74, 52, 28, 10 moment of the equation (see (6)). Gamma (Γ) is the Euler
Temperature (°C) 90, 35, 3 Gamma Function (equation (7)).
Relative humidity (%) 90, 35, 3
Pressure (Pa) 90, 2 1
U � A · Γ􏼒1 + 􏼓, (5)
k
1 3
Pd � ρA3 Γ􏼒1 + 􏼓, (6)
2 k
Group I ∞
S1 Γ(n) � 􏽚 e− t tn− 1 dt. (7)
S12 S2
0

S11 S3
Several different methods have been studied extensively
Group IV
in the past by the wind energy community. A review of the
S10 S4
literature shows a great variety of formulations in estimating
Group II the Weibull distribution parameters (Table 5). The literature
S9 S5 includes Empirical Method (EM), Power Density Method
(PDM), Graphical Method (GM), Maximum Likelihood
S8 S6 Method (MLM), Modified (Weighted) Maximum Likeli-
S7
hood Method (MMLM), Moment Methods (MM), and Least
Group III Square Method (LSM). It is also observed that the Wind
Atlas Analysis and Application Program (WAsP) (official
website: http://www.WAsP.dk) method has a limited use in
Figure 1: The center being the met. mast location: the figure shows
such studies. Most of the studies using the WAsP method
up to 3 km radius with 500 m circle steps size. Wind sectors are split
into 12 wind direction sectors, 30° each, starting from first sector through the WAsP software are not comparing it to other
being centered at 0°. White numbers are the sector numb. methods [34, 35], but see [36]. In the current study, a se-
lection of methods from the literature was used for the
estimation of Weibull parameters.. Elimination of methods
conditions (Table 3). Wind data were split into 12 equal wind is done through similarities. Between similar methods of EM
direction sectors, based on the vane measurements at 98 m and PDM and between GM and LSM, the EM and the LSM
coupled with the top-mounted cup anemometer at 101 m. are chosen, respectively. Moment methodologies have
Directional wind speed statistics when the top-mounted similarities with WAsP; therefore they are ignored. Five
anemometer is coupled with the closest wind vane at 98 m methods are shortlisted for the study: EM, MLM, MMLM,
show highest energy density at sector 1 and sector 7, which LSM, and WAsP.
are centered at 0° and 180°, respectively (Table 4).

3. Distribution Analysis 3.1. Empirical Method (EM). One of the most well-known
and simple methods for estimating the Weibull param-
In the last decade, several other distribution functions have eters is the Empirical Method (EM), derived from the
been put forward/have been proposed, among which the power density function [25, 27, 29, 30, 32]. The energy
most commonly used in the literature seem to be the fol- factor can be calculated as the ratio between the mean of
lowing: Rayleigh, Weibull, Lognormal, Gamma, Pearson, the sum of cubes of all wind speeds and the cube of the
Kappa, Erlang, and Gumble or bimodals of these [13–17]. mean wind speed (equation (8)). After Ef is calculated,
Among the referenced studies, the general conclusion is that one can easily use a derived numerical solution of k and A
the two- or three-parameter Weibull distribution is the best (equation (9)). The simplicity of the equation makes it
distribution function to describe the wind characteristics very useful for initial calculations, but it has also been
[11, 18]. This is also reflected in industry applications, where observed that the fit is not as good as that in other
any major wind energy or wind field modeling tool employs methods when there is low wind speed and high tur-
the two-parameter Weibull function (e.g., WAsP and bulence, which makes the Weibull PDF not a smooth
WindPRO). curve. Nevertheless, it was chosen for this study as a
In the current study, selected dataset is also used for reference formulation in order to explore the possible
analysis using sector wise Weibull distribution function. differences with more advanced methods.
4 Advances in Meteorology

Table 2: Sector-wise wind frequency and topographical characteristics. The sectors are grouped based on similarities of roughness and
obstacles.
Sector Center Fre. (%) Group Terrain Roughness Obstacles
1 0 45.65 None
I Grassland
2 30 11.21 Small cottages
3 60 1.95
4 90 1.41 II Flat; 50 m a.s.l. Urban area Village with max 6.5 m tall houses
5 120 1.3
6 150 3.7
III University with tall buildings
7 180 12.59
8 210 7.59 None
9 240 2.75 Distant wind farm
10 270 2.09 IV Hilly; up to 350 m a.s.l. with gentle slope Grassland
11 300 4.66 None
12 330 5.07

100 100

80 80
Heighta.g.l.(m)

Heighta.g.l.(m)

60 60

40 40

20 20

0 0
2 3 4 5 6 7 8 335 340 345 350 355 360
Wind speed (m/s) Wind direction (º)
(a) (b)

Figure 2: 1 full-year averaging of all data. Wind speed (a) and wind direction (b).

Γ(1 +(3/k)) U3 k within the desired limits with equation (11), the scale pa-
Ef � � , (8) rameter A can be computed with equation (12).
Γ3 (1 +(1/k)) U3
􏽐ni�1 Uki ln Ui 􏼁 􏽐ni�1 ln Ui 􏼁
3.69 k� − , (11)
k�1+ 2 , 􏽐ni�1 Uki n
(9)
Ef
(1/k)
n
U ⎝􏽘 Uk ⎞
A �⎛ ⎠ . (12)
i
A� . (10) i�1
Γ(1 +(1/k))

3.3. Modified Maximum Likelihood Method (MMLM).

3.2. Maximum Likelihood Method (MLM). Another well- There is an alternative version of MLM, which has a
known and widely used method is the Maximum Likelihood modification on the wind frequencies [20, 21, 26, 27, 30, 32].
Method (MLM) [20–29]. In MLM, the k parameter is found The method is mostly preferred when there is a large amount
with the iteration of equation (10) for n number of samples with of missing data. All wind frequency values are weighted
the initial shape parameter with the value of k � 2. After finding based on the available data; therefore the method is called
Advances in Meteorology 5

5
Wind speed (m/s)
4

0
0 5 10 15 20 25
Hour of the day
WS101 WS30
WS76 WS10
WS52
Figure 3: Diurnal averaged wind speed for all the measurement heights. The thick and dashed lines are longest and shortest day of the year
for the measurement location during the campaign, respectively.

Table 3: Monthly wind speed statistics (all in m/s) of the top-mounted cup anemometer, WS101.
Month Recovery (%) Mean Median Min Max Std. dev.
Jan 98.25 6.557 5.123 0.305 23.611 5.391
Feb 99.68 5.511 5.141 0.306 16.776 3.354
Mar 99.66 6.523 6.365 0.305 19.775 3.511
Apr 98.4 4.396 3.499 0.3 19.561 3.654
May 99.26 5.472 4.528 0.3 15.704 3.958
Jun 99.49 5.127 5.227 0.306 13.92 2.97
Jul 99.48 6.476 6.682 0.315 15.445 3.143
Aug 99.64 7.736 8.32 0.304 15.472 3.498
Sep 99.56 7.016 6.452 0.301 20.19 4.217
Oct 98.75 5.617 5.291 0.308 16.658 3.433
Nov 99.91 7.221 7.078 0.348 18.321 3.667
Dec 99.53 6.447 5.538 0.301 23.628 4.432
Total 99.3 6.185 5.825 0.3 23.628 3.933

Table 4: Directional wind speed statistics of the top-mounted cup anemometer, WS101, based on the closest wind vane, WD98.
Direction Rec. Fre. Mean Median Min Max Std. dev.
Sector Range (%) (%) (m/s) (m/s) (m/s) (m/s) (m/s)
1 345°–15° 99.75 45.65 7.564 7.511 0.403 23.628 3.696
2 15°–45° 98.9 11.21 5.921 4.629 0.403 23.611 4.871
3 45°–75° 95.83 1.95 1.614 1.355 0.405 13.227 1.127
4 75°–105° 95.19 1.41 1.772 1.369 0.418 7.588 1.15
5 105°–135° 93.11 1.3 1.442 1.138 0.403 8.046 1.022
6 135°–165° 97.12 3.7 4.074 2.542 0.401 19.775 3.912
7 165°–195° 98.94 12.59 6.057 5.992 0.404 19.2 3.684
8 195°–225° 98.28 7.59 5.163 5.161 0.402 16.151 2.836
9 225°–255° 96.03 2.75 4.693 4.817 0.405 13.211 2.541
10 255°–285° 95.26 2.09 4.756 4.784 0.412 13.536 2.601
11 285°–315° 98.17 4.66 5.52 5.316 0.401 14.8 2.804
12 315°–345° 97.52 5.07 4.648 4.04 0.403 16.122 3.122
Total 98.71 100 6.22 5.857 0.401 23.628 3.919
6 Advances in Meteorology

Table 5: Selected studies related to the estimation of Weibull distribution in the last 20 years and the estimation of Weibull parameter
methods used. WAsP method is not fully explained and presented in literature, except the technical documents of the WAsP software.
Study EM PDM GM MLM MMLM MM LSM
Garcia [19] x x x
Seguro [20] x x x
Donk [21] x x x
Ramirez [22] x x x
Carta [23] x x
Cellura [24] x
Akdag [25] x x x x
Saleh [26] x x x x
Khahro [27] x x x x x
Arslan [28] x x
Mohammadi [29] x x x x x
Katinas [30] x x x x
Ali [31] x
Kang [32] x x x x
Polnumtiang [33] x

Modified Maximum Likelihood Method (MMLM) (some- considered to be more realistic by the method’s developers
times it is also called Weighted Maximum Likelihood [38]. Nevertheless, that does not mean it is more effective for
Method in literature). Equation (11) is rewritten with the diverse frequencies distributions; therefore it has been se-
addition of the frequency by knowing that all the samples are lected as one of the methods to be studied. The reference
above 0 m/s (equation (13)). The same iteration from MLM documentation about the method is not detailed but the
is applied to calculate the modified scale parameter Am as in WAsP method is one of the most used methodologies
equation (14). through the WAsP software family.
The first step in the method is to define a parameter that
k
􏽐ni�1 Ui m ln Ui 􏼁f Ui 􏼁 􏽐ni�1 ln Ui 􏼁f Ui 􏼁 gives the probability of the wind speeds above the mean
km � k
− , (13) value, which can be calculated through the cumulative
􏽐ni�1 Ui m f Ui 􏼁 n
density function (CDF) with the Weibull parameters
(1/km ) (equation (2)). If the mean wind speed is applied to the
n
⎝1 􏽘 Ukm f U 􏼁⎞
⎠ function, the cumulative density will be the total sum of the
Am � ⎛ i . (14)
probabilities of the wind speeds below the mean values;
n i�1 i
therefore 1-F (U) becomes the proportion of the values
above the mean value, Up (equation (16)).

3.4. Least Square Method (LSM). A less preferred method Up � e− (U/A) . (16)
but one that is known to be more accurate for diverse
frequencies distributions is the Least Square Method (LSM) When both sides of the equation are taken as logarithmic
[21, 23, 30, 37]. The Weibull function is transformed into a normal (equation (17)), one can write the parameter A as a
linear function format as in “y � Gain · x + Offset.” In order function of k or vice versa and calculate both in two steps. The
to make this transition, log normal of both sides of equation shape parameter A can be written as a function of k through
(2) is taken and this leads to the desired format as in the power density function (equation (6)), which is equal to
equation (15). The left side of the equation can be computed the mean of cube sum of wind speed samples (equation (18)).
from the wind speed variable and a fit algorithm can be used If parameter A is singled out, it can be written as a function of
to calculate k and A. k(equation (19)). If equations (17) and (19) are merged as in
equation (20), one can use iterative numerical steps to solve k
1 and place the result in equation (19) to calculate A.
ln􏼢ln􏼠 􏼡􏼣 � k ln(U) − k ln(A). (15) k
1 − F(U) U
ln􏼐Up 􏼑 � − 􏼠 􏼡 , (17)
A

3 1 n
A3 Γ􏼒1 + 􏼓 � 􏽘 U3i , (18)
3.5. WAsP Weibull Method. This method uses a different k n i�1
approach for estimating the Weibull parameters; it assumes
that the wind speed above the mean value is most likely to (1/3)
(1/n) 􏽐ni�1 U3i
create the maximum power. Therefore, Weibull fitting by A�􏼢 􏼣 , (19)
using only the data above the mean wind speed value is Γ(1 + (3/k))
Advances in Meteorology 7

0.08 EM MLM MMLM LSM WAsP

0.06
Frequency

0.04

0.02

0.00
0 5 10 15 20
Wind speed (m/s)
Figure 4: Omnidirectional results. Number of samples is 52559.

Table 6: Comparison results from four different groups.

Sector Power density measured (W/m2) Method k A (m/s) RMSE [10− 3] R2 (%) Power density fitted (W/m2)
EM 1.64 6.9 3.16 84.80 327
MLM 1.53 6.8 2.93 86.20 350
All 341 MMLM 1.68 7.3 3.08 83.80 381
LSM 1.42 6.9 2.79 86.40 394
WAsP 1.79 7.2 3.49 81.70 330
EM 1.91 8.1 2.23 91.00 444
MLM 1.82 8.1 2.19 90.80 460
1-2 460 MMLM 1.95 8.6 2.07 91.50 505
LSM 1.68 8.2 2.21 89.30 513
WAsP 2.02 8.3 2.3 90.80 448
EM 1.53 6.1 5.03 66.30 252
MLM 1.39 6.1 4.51 71.60 269
6-7 275 MMLM 1.55 6.6 5.04 60.50 296
LSM 1.34 6 4.36 72.80 279
WAsP 1.71 6.5 5.82 54.40 260
EM 1.82 5.5 5.62 71.80 146
MLM 1.63 5.5 5.33 71.50 158
8–12 155 MMLM 1.85 6 5.21 71.20 179
LSM 1.42 5.6 5.19 67.00 188
WAsP 2.06 5.8 5.95 70.60 149

k wind speed bins of 0.5 m/s and all available data for the
U
ln􏼐Up 􏼑 � − ⎣
⎡
(1/3)
⎦⎤ . (20) selected sectors were processed for the calculation.
􏼂(1/n) 􏽐ni�1 U3i /Γ(1 + (3/k))􏼃
4. Results
Omnidirectional data and grouped sectors were used in
In order to understand the accuracy of the Weibull esti-
the analysis (see Table 2). Sectors were grouped based on the
mation method, most of the studies address the question as a
common roughness and obstacle types of the sectors. Group
statistical error and calculate the root mean square error
I includes sectors 1 and 2, with the northerly winds and the
(RMSE) based on the measured data.
highest number of samples. The urban area located in sectors 􏽶������������
􏽴
3, 4, and 5 constitutes Group II, which has the lowest
1 n 2
number of samples and a total energy density calculated to RMSE � 􏽘 y − xi 􏼁 , (21)
be close to zero. Therefore, the results for Group II might be n i�1 i
misleading; nevertheless they are still presented for the sake
of completeness. The university zone, sectors 7 and 8, where yi is the measured value and xi is the Weibull-pa-
constitutes Group III and the sectors covered with hills from rameters-based calculated value. However, this method can
8 to 12 form Group IV. The collected data were analyzed by be misleading because it places the same importance on
8 Advances in Meteorology

0.08 EM MLM MMLM LSM WAsP

0.06
Frequency

0.04

0.02

0.00
0 5 10 15 20
Wind speed (m/s)
Figure 5: Group I: sectors 1 and 2; wind direction between 345° and 45°. Number of samples is 29629.

0.08 EM MLM MMLM LSM WAsP

0.06
Frequency

0.04

0.02

0.00
0 5 10 15 20
Wind speed (m/s)
Figure 6: Group III: sectors 6 and 7; wind direction between 135° and 195°. Number of samples is 8580.

every range of wind speed and the total sum of errors is not for the methods EM, MLM, and WAsP, which is already
weighted by the possible energy density. Therefore, in the within the limits of statistical uncertainty [39], while fre-
current study, the accuracy is also evaluated through the quency levels show that almost 25% of the whole data is
power density function because this is the real effective below 3 m/s, which is the cut-in wind speed for most tur-
difference between estimation methods when it comes to bines. When the sector wise results are observed
calculating the wind energy production. It is common to (Figures 5–7 and Table 6), it is also seen that low wind speeds
calculate the power density based on two datasets as in below 3-4 m/s are common and can affect the calculations
equation (22), where ρ is the air density and fm is the fre- for the all groups. Among the compared methods, MLM
quency for the given wind speed range. The values can be appears to be the best performing overall, with a maximum
used to drive an error percentage value, ε (equation (23)), error of 3% in power density calculations at the sectors
compared to the measured value. without urban areas. The WAsP gives the second lowest
percentage error, arriving at similar results to those of EM,
n
1 with the difference being that the WAsP method has a better
Pm � 􏽘 ρU3i fm Ui 􏼁, (22)
i�1
2 fit for wind speeds close to the mean level, which is in-
tentional by the design of the methodology, as was described
Pm − P in subsection “WAsP Weibull Method.” MMLM and LSM
ε �􏼔 􏼕 · 100. (23) produce high error percentages nearly in every sector and
P
even in the omnidirectional fit. RMSE and R2 values show an
The omnidirectional results (Figure 4 and Table 6) show inverse relationship as is expected. When the RMSE in-
a minor error in power density prediction, as low as ± 3-4% creases, the R2 parameter decreases, and vice versa, but there
Advances in Meteorology 9

0.08 EM MLM MMLM LSM WAsP

0.06
Frequency

0.04

0.02

0.00
0 5 10 15 20
Wind speed (m/s)
Figure 7: Group IV: sectors 8–12; wind direction between 195° and 345°. Number of samples is 11799.

is no clear relationship between these statistical measures Several wind data analysis tools in the world use WAsP
and the calculated power density estimation errors. Weibull fitting method as the core estimation method for
Weibull parameters through the WAsP software. Never-
theless, it is observed that, for the diverse winds, Maximum
5. Conclusion Likelihood Method is much more stable and has a better
Yearly 10-minute statistics from a 101 m met. Mast and top- correlation with measured data. One should create obser-
mounted cup anemometer data have been used for com- vational wind statistical data by means of MLM and apply
parison of Weibull parameter estimation. The measurement the Weibull parameters to the model even if the used model
period is from 27 December 2017 to 27 December 2018 for a is WAsP software.
total of 1 full year with over 99% recovery rate. The data has In this study, wind speed data in the form of 10-minute
almost 25% low wind speeds, which causes difficulties on the statistics from a 101 m met. mast with top-mounted cup
estimation, which is the core reason of the study. Vertical anemometers are used to compare different methods for the
wind characteristics from the met. mast show that the estimation of Weibull parameters. The measurement period
measurement location has wind shear within limits of the was from 27 December 2017 to 27 December 2018, that is,
standards and vertical directional turn is negligible between one full year of data with over 99% recovery rate. A high
cup anemometer and the vane used in the study. Basic proportion of the dataset (over 20%) consisted of low wind
statistics of monthly and 12 equal wind sectors are calcu- speeds, which are known to cause difficulties in the pa-
lated. Sectors 2 to 6 are excluded from the study due to the rameter estimation, and this is the main driver that moti-
low recovery rate and urban areas closer than 500 m. The vated the study. The vertical wind characteristics from the
results leading to the conclusion are made with the other met. mast show that the measurement location has wind
sectors. shear within the limits of the industry standards. Basic
The omnidirectional and sector wise calculations for five statistics of monthly and 12 equal wind sectors were cal-
different estimation methods are tested and the results are culated. The sectors are divided into four groups (Figure 1).
presented with observed statistics. The distribution fits are The results leading to the conclusion were produced with
analyzed through root mean square errors (RMSE) and R2 data from these groups. Nevertheless, Group II does not
parameters in addition to the power density error function have enough amount of data to compare and present the
(ε). The results show that the Maximum Likelihood method methods; therefore, it is excluded from the results.
(MLM) has the best performance. A clear link has not been The omnidirectional and sector wise calculations for
found between ε and RMSE and/or R2 contrary to the several five different estimation methods were tested and the re-
previously cited studies, where conclusions are made based sults are presented with the observed statistics. The dis-
on these statistical parameters. Other estimation methods tribution fits were analyzed through root mean square
show high uncertainty and power density calculation error. errors (RMSE) and R2 parameters in addition to the power
These two results lead to the fact that power density error density error function (ε). The results show that the
estimation is the most effective way of checking the dis- Maximum Likelihood method (MLM) had the best per-
tribution fit quality even though similar studies. Based on formance for the dataset. A clear link has not been found
Group III results, it can be said that the low number of between ε and RMSE and/or R2 contrary to several previous
samples with low wind speeds causes high uncertainty and studies, where conclusions on performance were made
deviation in any method. based on these statistical parameters. Selected list of these
10 Advances in Meteorology

referenced studies is in Table 5. The other estimation [8] S. Mathew, K. P. Pandey, and A. Kumar.V, “Analysis of wind
methods tested showed high uncertainty and higher power regimes for energy estimation,” Renewable Energy, vol. 25,
density calculation errors. These two results lead to the no. 3, pp. 381–399, 2002.
conclusion that power density error estimation is the most [9] R. E. Luna and H. W. Church, “Estimation of long-term
effective way of checking the distribution fit quality of an concentrations using a “universal” wind speed distribution,”
estimation method even though similar studies only focus Journal of Applied Meteorology, vol. 13, no. 8, pp. 910–916,
1974.
on the statistical terms of RMSE and/or R2. Based on the
[10] I. B. Troen and E. L. Petersen, European Wind Atlas, Risø
results of Group III, it is evident that low number of
National Laboratory, Roskilde, Denmark, 1989.
samples combined with low wind speeds causes high un- [11] D. A. Stewart and O. M. Essenwanger, “Frequency distri-
certainty and deviation in any method. bution of wind speed near the surface,” Journal of Applied
Several wind data analysis tools in the world use the Meteorology, vol. 17, no. 11, pp. 1633–1642, 1978.
WAsP Weibull fitting method as the core estimation method [12] D. M. Deaves and I. G. Lines, “On the fitting of low mean
for Weibull parameters through the WAsP software. Nev- windspeed data to the weibull distribution,” Journal of Wind
ertheless, it is demonstrated that, in the case of diverse Engineering and Industrial Aerodynamics, vol. 66, no. 3,
winds, the Maximum Likelihood Method is much more pp. 169–178, 1997.
stable and has a better correlation with measured data. For [13] J. Zhou, E. Erdem, G. Li, and J. Shi, “Comprehensive eval-
these types of datasets, it is advisable to create observational uation of wind speed distribution models: a case study for
wind statistical data by means of MLM and apply the North Dakota sites,” Energy Conversion and Management,
Weibull parameters to the model. vol. 51, no. 7, pp. 1449–1458, JUL 2010.
[14] E. C. Morgan, M. Lackner, R. M. Vogel, and L. G. Baise,
“Probability distributions for offshore wind speeds,” Energy
Data Availability Conversion and Management, vol. 52, no. 1, pp. 15–26, JAN
2011.
The data used to support the findings of this study are
[15] T.P. Chang, “Estimation of wind energy potential using
available from the corresponding author upon request. different probability density functions,” Applied Energy,
vol. 88, no. 5, pp. 1848–1856, 2011.
[16] T. B. M. J. Ouarda, C. Charron, J. Y. Shin et al., “Hosary.
Conflicts of Interest Probability distributions of wind speed in the UAE,” Energy
The author declares that there are no conflicts of interest Conversion and Management, vol. 93, pp. 414–434, 2015.
[17] P. Wais, “A review of Weibull functions in wind sector,”
regarding the publication of this article.
Renewable and Sustainable Energy Reviews, vol. 70,
pp. 1099–1107, 2017.
Acknowledgments [18] K. Conradsen, L. B. Nielsen, and L. P. Prahm, “Review of
weibull statistics for estimation of wind speed distributions,”
This project has been supported by the Scientific and
Journal of Climate and Applied Meteorology, vol. 23, no. 8,
Technological Research Council of Turkey (TÜBİTAK) pp. 1173–1183, 1984.
(Grant no. 215M384). [19] A. Garcia, J. L. Torres, E. Prieto, and A. De Francisco, “Fitting
wind speed distributions: a case study,” Solar Energy, vol. 62,
References no. 2, pp. 139–144, 1998.
[20] J. V. Seguro and T. W. Lambert, “Modern estimation of the
[1] IEC, Wind Turbines-Part 1: Design Requirements (IEC
parameters of the Weibull wind speed distribution for wind
standart 61400-1), IEC, Geneva, Switzerland, 3rd edition,
energy analysis,” Journal of Wind Engineering and Industrial
2005.
Aerodynamics, vol. 85, no. 1, pp. 75–84, 2000.
[2] MEASNET, Evaluation of Site-specific Site Conditions,
[21] S. J. van Donk, L. E. Wagner, E. L. Skidmore, and J. Tatarko,
MEASNET, Madrid, Spain, 2016.
“Comparison of the weibull model with measured wind speed
[3] F. Bingöl, J. Mann, and D. Foussekis, “Conically scanning
lidar error in complex terrain,” Meteorologische Zeitschrift, distributions for stochastic wind generation,” Transactions of
vol. 18, no. 2, pp. 189–195, 2009. the ASAE, vol. 48, Article ID 503510, 2 pages, 2005.
[4] S. Bradley, Y. Perrott, P. Behrens, and A. Oldroyd, “Cor- [22] P. Ramirez and J. A. Carta, “The use of wind probability
rections for wind-speed errors from sodar and lidar in distributions derived from the maximum entropy principle in
complex terrain,” Boundary-Layer Meteorology, vol. 143, no. 1, the analysis of wind energy. A case study,” Energy Conversion
pp. 37–48, 2012. and Management, vol. 47, no. 15-16, pp. 2564–2577, 2006.
[5] N. Vasiljevia, M. José, L. M. Palma et al., “Methodology for [23] J. A. Carta and P. Ramı́rez, “Analysis of two-component
atmospheric multi-doppler lidar experiments,” Atmospheric mixture Weibull statistics for estimation of wind speed dis-
Measurement Techniques, vol. 10, no. 9, pp. 3463–3483, 2017. tributions,” Renewable Energy, vol. 32, no. 3, pp. 518–531,
[6] J. Mann, R. Menke, N. Vasiljević, J. Berg, and N. Troldborg, MAR 2007.
“Challenges in using scanning lidars to estimate wind re- [24] M. Cellura, G. Cirrincione, A. Marvuglia, and A. Miraoui,
sources in complex terrain,” The Science of Making Torque “Wind speed spatial estimation for energy planning in Sicily: a
from Wind 2018 - Politecnico di Milano (POLIMI), vol. 1037, neural kriging application,” Renewable Energy, vol. 33, no. 6,
Article ID 072017, 2018. pp. 1251–1266, 2008.
[7] M. Jamil, S. Parsa, and M. Majidi, “Wind power statistics and [25] S. A. Akdag and D. Ali, “A new method to estimate weibull
an evaluation of wind energy density,” Renewable Energy, parameters for wind energy applications,” Energy Conversion
vol. 6, no. 5-6, pp. 623–628, 1995. and Management, vol. 50, no. 7, pp. 1761–1766, JUL 2009.
Advances in Meteorology 11

[26] H. Saleh, A. A. E. Azm Aly, and S. A.- Hady, “Assessment of

different methods used to estimate Weibull distribution pa-
rameters for wind speed in zafarana wind farm, Suez Gulf,
Egypt,” Energy, vol. 44, no. 1, pp. 710–719, 2012.
[27] S. F. Khahro, K. Tabbassum, A. Mahmood Soomro, L. Dong,
and X. Liao, “Evaluation of wind power production pro-
spective and Weibull parameter estimation methods for
Babaurband, Sindh Pakistan,” Energy Conversion and Man-
agement, vol. 78, pp. 956–967, 2014.
[28] T. Arslan, Y. M. Bulut, and A. Altın Yavuz, “Comparative
study of numerical methods for determining Weibull pa-
rameters for wind energy potential,” Renewable and Sus-
tainable Energy Reviews, vol. 40, pp. 820–825, 2014.
[29] K. Mohammadi, O. Alavi, M. Ali, N. Goudarzi, and
M. Jalilvand, “Assessing different parameters estimation
methods of Weibull distribution to compute wind power
density,” Energy Conversion and Management, vol. 108,
pp. 322–335, 2016.
[30] V. Katinas, M. Marčiukaitis, G. Gecevičius, and
A. Markevičius, “Statistical analysis of wind characteristics
based on weibull methods for estimation of power generation
in Lithuania,” Renewable Energy, vol. 113, pp. 190–201, 2017.
[31] S. Ali, S.-M. Lee, and C.-M. Jang, “Statistical analysis of wind
characteristics using Weibull and Rayleigh distributions in
Deokjeok-do Island-incheon, South Korea,” Renewable En-
ergy, vol. 123, pp. 652–663, 2018.
[32] D. Kang, K. Ko, and J. Huh, “Comparative study of different
methods for estimating weibull parameters: a case study on
Jeju island, South Korea,” Energies, vol. 11, no. 2, 2018.
[33] S. Polnumtiang and K. Tangchaichit, “Wind speed and power
characteristics of Kalasin province, Thailand,” IOP Conference
Series: Earth and Environmental Science, vol. 150, Article ID
012020, 2018.
[34] B. Prahm, T. Vihma, E. Atlaskin et al., “Production of the
finnish wind atlas,” Wind Energy, vol. 16, no. 1, pp. 19–35,
2013.
[35] N. Nawri, G. N. Petersen, H. Bjornsson et al., “The wind
energy potential of Iceland,” Renewable Energy, vol. 69,
pp. 290–299, 2014.
[36] B. Jourdier and D. Philippe, “Errors in wind resource and
energy yield assessments based on the weibull distribution,”
Annales Geophysicae, vol. 35, no. 3, pp. 691–700, 2017.
[37] P. Ramirez and J. A. Carta, “Influence of the data sampling
interval in the estimation of the parameters of the Weibull
wind speed probability density distribution: a case study,”
Energy Conversion and Management, vol. 46, no. 15-16,
pp. 2419–2438, 2005.
[38] WAsP 12 Documentation, Last Edited 2018-06-01, Depart-
ment of Wind Energy, Technical University of Denmark
(DTU), Roskilde, Denmark, 2018.
[39] S.-D. Kwon, “Uncertainty analysis of wind energy potential
assessment,” Applied Energy, vol. 87, no. 3, pp. 856–865, 2010.