International Journal of Physical Sciences Vol. 8(2), pp.

83-88, 16 January, 2013

Available online at http://www.academicjournals.org/IJPS
DOI: 10.5897/IJPS12.685
ISSN 1992 - 1950 2013 Academic Journals

Full Length Research Paper

Twenty-five years of wind data in south-eastern

locations of Nigeria: Modeling and prediction of
wind potential
F. C. Odo1,2*, D. O. Ugbor2 and P. E. Ugwuoke1
1
National Centre for Energy Research and Development, University of Nigeria, Nsukka, Nigeria.
2
Department of Physics and Astronomy, University of Nigeria, Nsukka, Nigeria.
Accepted 14 January, 2013

Weibull distribution is often invoked to interpret and predict wind characteristics needed for effective
design of wind power systems for different locations. In this paper, daily average wind data for Enugu
(6.4N; 7.5E), Onitsha (6.8N; 6.1E) and Owerri (5.5N; 7.0E) over a 25-year period is modeled in terms
of the Weibull distribution in order to accurately predict wind potentials for the locations. The monthly
and annual wind speed probability density distributions at 10 m meteorological height were analyzed
and the Weibull shape and scale factors were empirically determined for the locations. The predicted
and measured wind speed probability density distributions of the locations are compared and the
accuracy of the model determined for each location using Pearson product moment correlation
coefficient (r) and root-mean-square error (). We find r and to be 0.64, 1.40, 0.67, 1.17 and 0.93, 1.55,
respectively, for Enugu, Onitsha and Owerri. The results suggest that the model can be used, with
acceptable accuracy, for predicting wind energy output needed for preliminary design assessment of
wind machines for the locations.

Key words: Renewable energy-general, wind, Weibull distribution.

INTRODUCTION

Energy plays a central role in economic development and
industrialization of any nation. Fossil fuels have been the
major resources that supply the world energy demand.
However, fossil fuel reserves are limited and usage of
fossil fuels to generate energy has negative
environmental effects. The world energy demand is
continuously increasing with increasing population such
that the present fossil fuel reserves cannot meet this
demand (Kamau et al., 2010). As a result, energy policies
of many nations are geared towards ensuring a supply of
reliable, economical and environmentally friendly energy
resources in a form that supports the targets for growth
and social development (Ucar and Balo, 2009). Wind
energy applications have been recently (Weisser and
*Corresponding author. E-mail: finboc@yahoo.com.
Garcia, 2006) described as an economic and
environmentally friendly solution to the urgent energy
problems of many countries.
Wind is an effect caused as a result of pressure
differences over regions and heights in the atmosphere
resulting in bulk motion of air masses. The force carried
by the moving air mass (wind) can be harnessed for
useful purposes such as grinding grain (in windmills) and
generating electricity (in wind turbine generators). It is
estimated that between 1.5 to 2.5% of the global solar
radiation received on the surface of the earth is converted
to wind (Vosburgh, 1983). Hence, wind energy, which
contributes very little pollution and few greenhouse gases
to the environment, is a valuable alternative to the non-
renewable and environmentally hazardous fossil fuels
(Taylor, 1983). Thus, the utilization of wind energy has
been increasing around the world at an accelerating
pace.
84 Int. J. Phys. Sci.

The extent to which wind can be exploited as a source idealization of the distribution of wind speed over time for most
of energy depends on the probability density of locations. The function gives the probability of wind speed being in
a range of 1 m/s about a particular speed (v), taking into account all
occurrence of different speeds at the site, which is variations for the period covered by the statistics. The Weibull
essentially, site-specific. However, the development of distribution is a statistical function given (Walker and Jenkins, 1997;
new wind projects continues to be hampered by the lack Gipe, 2004) by:
of reliable and accurate wind resource data in many parts
k 1 k
of the developing world. To optimize the design of a wind k v v
energy conversion device, data on speed range over f (v) exp , (1)
which the device must operate to maximize energy c c c
extraction is required, which requires the knowledge of
Where f(v) is the probability density defined as the frequency of
the frequency distribution of the wind speed. Among the
occurrence of wind speed (v), c is the scale parameter (in unit of
probability density functions that have been proposed for m/s), which is closely related to the wind speed for the location, and
wind speed frequency distributions of most locations, the k is the dimensionless shape parameter, which describes the width
Weibull distribution has been the most acceptable and of the distribution and measures the probability of extraction of wind
forms the basis for commercial wind energy applications energy at a given characteristic wind speed. The Weibull
and software (Seyit and Ali, 2009). Some of the wind distribution is therefore characterized for any location by the two
parameters c and k. The cumulative form of the Weibull distribution
energy software based on the Weibull distribution F(v) which gives the probability of the wind speed exceeding the
includes the Wind Atlas Analysis and Application value v is expressed (Justus et al., 1978; Walker and Jenkins,
Program (WAsP) and the recently developed Nigerian 1997) as:
Wind Energy Information System (WIS).
k
In previous papers (Enibe, 1987; Ugwuoke et al., 2008; v
Odo et al., 2010), the theoretical potentials of wind at F( v ) exp (2)
various heights above the ground, based on annual c
average values of wind speed, have been assessed for
many Nigerian locations. These analyses were carried On the other hand, the power derivable from the wind is a cubic
function of the wind speed such that, in the Weibull distribution, the
out using measured data over various periods ranging power density (PA) of the wind at any speed is given (Seyit and Ali,
from 1 to 10 years. In these analyses, little or no attention 2009) by:
was given to the frequency distribution patterns of wind
speed over the studied periods for the locations. In this 1 3
2 0
paper, the frequency distribution of daily averages of P( A) v f ( v ) dv , (3)
wind speed for three locations in south-eastern Nigeria,
namely; Enugu, Onitsha and Owerri, over a longer period Where is the density of air. However, the power derivable from
of up to 25 years are examined. The observed data for the wind scales with the height (h) above the ground
these three locations are modeled in terms of the Weibull
1
distribution, to enable an accurate prediction of the wind h according to the Hellmans exponential law given
potentials of the locations and the results are compared.
The results of this analysis are expected to be very useful (Walker and Jenkins, 1997; Gipe, 2004) by:
to designers of wind turbines, for various wind energy
applications, for the locations.
v h

v0 h0
(4)

MATERIALS AND METHODS

In this study, we use 25 years (1978 to 2003) daily averages of
wind speed data at 10 m meteorological height, for Enugu (6.4N;
7.5E), Onitsha (6.8N; 6.1E) and Owerri (5.5N; 7.0E) obtained
from the data bank of Nigerian Meteorological Agency (NIMET).
The data gives information on the daily average wind speed
distributions of the locations over the study period, from which the
monthly and yearly average data were calculated for the current
analysis.
Weibull probability density function
Where h0 is any reference height and v0 is the wind speed at h0,
while is the Hellmans constant which varies from one location to
another. Equation 4 suggests that the derivable power increases
with increasing height only if the change in density of air is
negligible. It has been shown (Walker and Jenkins, 1997) that
within the troposphere (h 10 km) the density of air varies very little
for any location.
Analysis of Equation 3 using Equation 1 shows that the power
density could be expressed as a gamma function defined in
general x-variable (Dass, 1998) as:

The Weibull probability density distribution is a two-parameter

x 0
x n1e x dx . (5)
function characterized by a dimensionless shape (k) parameter and
scale (c) parameter (in unit of speed). It is a mathematical Using Equations 5 and 1 in Equation 3, the average wind power
 Odo et al. 85

7
ENUGU
Equation 7 therefore suggests that the probability of capturing the
ONITSHA wind by a turbine at a mean wind speed is small if the shape factor
6 OWERRI
is high for that location, since k could be used as a measure of
dispersion (Pallabazzer, 2003) in a distribution. This shows that
5
knowledge of the exact value of k provides preliminary information
4 on the wind speed regime for which wind turbines should be
v (m/s)

designed for optimum performance in any given location. Similarly,

3 wind speed is a real valued random variable and most locations
show wide dispersion (Justus et al., 1978) so that the use of mean
2 values as the characteristic speed for designs may not be very
reliable (Jaramillo and Borja, 2004).
1 In this paper, we use analytic method in which F(v) is plotted
against v on double logarithm scales and apply a one-dimensional
0
regression on the plots to obtain values for k and c for the locations.
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month
Fig. 1: Time series distributions of wind speed for the locations
Figure 1. Time series distributions of wind speed for the locations. RESULTS AND DISCUSSION

We calculated the monthly and annual average wind

speed distributions at 10 m meteorological height for the
Table 1. Wind speed distributions parameters for Enugu, Onitsha three locations over the studied period. The time series
and Owerri. distributions of the monthly average values for the
locations are shown in Figure 1, while the annual average
Location Latitude Longitude vmean (m/s) k c (m/s) values are shown in Table 1. The distributions give
Enugu 6.4N 7.5E 5.5 0.6 2.0 6.4 annual mean values of 5.5 0.6 m/s, 3.6 0.4 m/s and
Onitsha 6.8N 6.1E 3.6 0.4 1.5 5.6 3.3 0.3 m/s, respectively, for Enugu, Onitsha and
Owerri 5.5N 7.0E 3.3 0.3 1.9 5.1 Owerri. It could easily be observed from Figure 1 that the
distributions of the monthly average wind speed for the
three locations are fairly similar, peaking in the month of
density (Pav) based on Weibull distribution can be expressed (Ucar
March and having minimum values in November.
and Balo, 2009) in the form: However, wind speed is highest in Enugu and lowest in
Owerri. Perhaps, the result is as expected since
3 difference in wind speed distributions may be related to
1 the difference in altitude between the locations. To model
P( av)
1
3 k the data in terms of the Weibull distribution, we took twice
(6)
3 logarithm of Equation 2 to obtain
2 1
1 k
ln ln Fv k ln v k ln c . (8)

Where is the characteristic wind speed of the location. However,

meteorologists have characterized the distributions of wind speeds The plots of ln (-ln Fv) as a function of ln v for the three
for many of the worlds wind regimes in terms of the speed locations, on the same scale, are shown in Figure 2.
distribution patterns. For example, in temperate climate (mid Linear regression of the plots gives
latitudes), a typical shape parameter k 2 offers a good
approximation (Gipe, 2004). For k = 2, Equation 1 or 2 reduces to
Rayleigh wind speed distribution. Thus, the Rayleigh distribution is ln ln Fv 2.0 ln v 3.7 ,
a special case of the Weibull distribution developed for estimation
of wind potential in temperate climate locations. Wind
characteristics are essentially location specific and performance of ln ln Fv 1.5 ln v 2.6
real wind conversion devices which are designed based on the
Rayleigh distribution may greatly differ if actual wind conditions at and
the location differ from those standard speed distributions.
A method has been suggested (Iheonu et al., 2002; Gipe, 2004) for
estimating the shape (k) factor of a set of wind speed data using the ln ln Fv 1.9 ln v 3.1
mean wind speed (v) and standard deviation () in a simple relation
of the form:
respectively, for Enugu, Onitsha and Owerri. By
1.086
comparing each of the equations with Equation 8, the
k (7) values of k and c were deduced for each location. The
v summary of the results is shown in Table 1.
86 Int. J. Phys. Sci.

0.5
Furthermore, using the results for k and c obtained
0
from the regression analysis for each location in Equation
0 0.5 1 1.5 2 2.5 1, the probability densities (%) of occurrence of different
-0.5 wind speeds were determined. The different speeds used
for this analysis were chosen from the range 0 v 10
ln(-ln Fv)

-1
m/s as covered by observational data in eight (8)
-1.5 comparable speed bins, for effective overlap. The results
of the prediction were compared with those calculated
-2 ENUGU
ONITSHA from observed data. The measured and predicted
OWERRI
-2.5 Linear (ENUGU)
Linear (ONITSHA)
probability density distributions for different wind speed
Linear (OWERRI)
bins for the three locations are shown in Figure 3, while
-3
ln v (m/s)
the summary of the comparison is displayed in Table 2.
Fig. 2: Plot of ln(-ln Fv ) against ln v for the locations
Similarly, the suitability of the model in predicting wind
Figure 2. Plot of ln(-ln Fv) against lnv for the locations. potential for each location was determined using two non-
parametric statistics, namely, the Pearson product
moment correlation coefficient (r) and root-mean-square
30
error (). The Pearson correlation coefficient is defined
OBS
PRED
(Aalen, 1978) as:
25

1
(a)
N
2
2

i
20
y x
Fv (%)

i
f(v) %

r 1 N
i 1
15
, (9)
10

yi y 2

5 i 1

0
0 2 4 6 8 10 12 14
Where N is the number of observations in each data set,
v (m/s) y and x, respectively are the measured and predicted
Fig. 3 (a): Observed and predicted probability density for Enugu

25
probability density, while y is the mean of the measured
OBS
PRED values. This statistic varies from 0 (for a null association)
20 to 1 (for a perfect association). A correlation is
statistically significant at a set level of significance if r
15 (b) 0.5, otherwise, it is not significant. The present analyses
f(v) %
Fv (%)

give correlation coefficients r ~ 0.6, 0.7 and 0.9,

10
respectively, for Enugu, Onitsha and Owerri at 5% level
of significance. Thus, all the correlations are statistically
5
significant at 5% level.
0
On the other hand, the root-mean-square error is a
0 2 4 6 8 10 12 14 statistic that determines the degree of departure of two
v (m/s) data sets from a supposed association and is defined
Fig. 3 (b): Observed and predicted probability density for Onitsha
(Joanes and Gills, 1998) as:
35
OBS
PRED
30 1
1 2
2
25
(c)
y i xi (10)
20 N
f(v) %
Fv (%)

15

10
The results give 1.40, 1.17 and 1.55, respectively, for
Enugu, Onitsha and Owerri. Results of all these analyses
5
are summarized in Table 2.
0 Modeling and prediction of wind characteristics are
0 2 4 6 8 10 12 14
major design inputs in the development of wind power
v (m/s)
Fig. 3 (c): Observed and predicted probability density for Owerri
systems for any location. However, the wind speed
Figure 3. Observed and predicted probability density for: (a) distribution for many of the worlds wind regimes have
Enugu, (b) Onitsha, (c) Owerri. been characterized and wind power systems are

Table 2. Comparison of measured and predicted probability density.
vpeak (m/s)
Location
Measured Predicted
Enugu 5.2 4.4
Onitsha 3.8 2.6
Owerri 3.5 3.5
optimized based on these standard distributions. In fact,
the Rayleigh distribution (k = 2) is often employed by
most wind power system developers (Vosburgh, 1983).
The result is that many wind power systems perform
poorly in many different locations because the actual
wind conditions at the locations differ largely from those
standard distributions (Ramachandra et al., 2005; Al-
Mohamad and Karmeh, 2003).
It could easily be observed from the time series
distributions of the monthly average wind speed that the
three locations studied in south-eastern Nigeria show
similarity in climatology of wind. Perhaps, this observation
could be attributed to their proximity in geographical
extent. The three locations lie within a latitudinal stretch
of about 1.3 and longitudinal stretch of 1.4. Similarly,
the distributions do not show any latitudinal or
longitudinal dependence either. Thus, wind speed in
south-eastern locations of Nigeria is essentially location
specific, which may be driven by environmental factors,
rather than geographical dependence.
We have also shown in the results that the Weibull
shape factor is 2.0, 1.5 and 1.9, respectively for Enugu,
Onitsha and Owerri, while the corresponding scale factor
is 6.4, 5.6 and 5.1 m/s. The values of the shape factor
presented in this paper suggest that while the data for
Enugu and Owerri are in close agreement with Rayleigh
distribution, the data for Onitsha departs significantly from
the standard Rayleigh distribution. The results further
suggest that the wind speed distributions in the studied
locations are widely dispersed, with those of Enugu and
Owerri being much wider than that of Onitsha. The
implication of these results is that any wind turbine which
is optimized based on Rayleigh distribution may be
suitable for Enugu and Owerri, but not for Onitsha. It
therefore becomes necessary that wind turbines for utility
generation in these locations be designed locally rather
than relying on importation of already designed systems.
A more comprehensive wind speed evaluation and
energy assessment is achieved by the use of real life
frequency distribution. The frequency distribution
obviously indicates the percentage of the time of
occurrence of the various wind bins/spectra and provides
information on when a particular rated turbine in the
location is expected to yield power (Amonye and Hassan,
2010). Thus, the most frequently occurring wind spectra
Odo et al. 87
F(v) peak (%)
r
Measured Predicted
22.0 14.0 0.64 1.40
21.0 13.5 0.67 1.17
31.0 17.0 0.93 1.55
defines the characteristic wind speed for the location,
which is a more reliable value for the design of wind
turbines for any location than the average value, which is
affected by the skewness of the distribution (Iheonu et al.,
2002; Ucar and Balo, 2009). A major outcome of our
result is the statistical accuracy of the Weibull distribution
model in predicting wind potentials of the locations as
determined by the correlation coefficient. In fact, for
Owerri, the model gives almost a perfect prediction of the
characteristic wind speed. However, for all the locations,
the correlation coefficients are statistically significant.
These results show that the predicted and measured data
are acceptably related by the Weibull distribution. Hence,
within the region of overlap, the Weibull distribution can
be used, with acceptable accuracy, to predict wind
potentials for these locations.
On the other hand, it is obvious in Figure 3 that there is
a wide departure of the theoretical predictions from the
real life wind speed frequency distribution patterns of the
studied locations. In fact, for each of the locations, within
the region of overlap, the theoretical model appears to
underestimate the probability density of every speed bin.
This is further supported statistically by the large values
of root-mean-square error calculated for all the locations.
Perhaps, this departure is attributable to the coarse
approximations arising from daily averages which fail to
account for the short time-scale variations of wind
characteristics of the locations. Wind speed is a real
valued random variable and observations over smaller
time-scales, such as hourly averages, may help to
improve the results.
REFERENCES
Aalen OO (1978). An Introduction to Categorical Data Analysis. Ann.
Stat. 6:701.
Al-Mohamad A, Karmeh H (2003). Wind Energy Potential in Syria.
Renew. Energy 28:1039.
Amonye M, Hassan B (2010). Basic Steps in Wind Turbine
Development, TradeNews, Nigeria. P. 33.
Dass HK (1998). Advanced Engineering mathematics, S. Chand &
company Ltd, New Delhi. P. 1158.
Enibe SO (1987). A Method of Assessing Wind Energy Potential in a
Nigerian Location. Nig. J. Solar Energy 6:14.
Gipe P (2004). Wind Power: Renewable Energy for home Farm and
Business, Chelsea Green, USA. P. 17.
Iheonu EE, Akingbade FO, Ocholi M (2002). Wind Resource Variation
over Selected Sites in West Africa. Nig. J. Renew. Energy 10:43.
88 Int. J. Phys. Sci.

Jaramillo OA, Borja MA (2004). Wind Speed Analysis in La Ventosa, Ucar A, Balo F (2009). Investigation of wind characteristics and
Mexico: A Bimodal probability distribution case. J. Renew. Energy generation Potentiality in Uludag-Bursa, Turkey. J. Appl. Energy
29:1613. 86:333.
Joanes DN, Gill CA (1998). Comparing measures of sample Skewness Ugwuoke PE, Bala EJ, Sambo AS, Argungu GM (2008). Essessment of
and Kurtosis. J. Royal Stat. Soc. 47:183. Wind Energy Potentials for Electricity Generation in the Nigerian
Justus CG, Hargraves WR, Mikhail A, Graber D (1978). Methods for Rural Setting. Nig. J. Solar Energy 19:98.
Estimating Wind Speed Frequency distribution. J. Appl. Meteorol. Vosburgh PN (1983). Commercial Application of Wind Power, Van
17:350. Nostrand Rainhold, New York. P. 12.
Kamau JN, Kinyua R, Gathua J (2010). Six years of wind data for Walker JF, Jenkins S (1997). Wind Energy Technology, John Wiley &
Marsabit: An Analysis of wind energy potentials. J. Renew. Energy Sons, Chinchester. P. 24.
35:1298. Weisser D, Garcia RS (2006). Instantaneous wind Energy Penetration
Odo FC, Igwebuike MN, Ezugwu CI (2010). Preliminary Assessment of in Isolated Electricity Grids. J. Renew. Energy 30:1299.
Enugu Location for Setting up a 1kW Wind turbine for Electricity
Generation. Nig. J. Solar Energy 21:182.
Pallabazzer R (2003). Parametric analysis of wind sitting efficiency. J.
Wind Eng. 91:1329.
Ramachandra TV, Rajeev KJ, Vansee K, Shruthi B (2005). Wind
Energy Potential Assessment Spatial Decision Support System.
Energy Educ. Sci. Technol. 14:61.
Seyit AA, Ali D (2009). A New method to estimate Weibull parameters
for Wind Energy applications. Energy Convers. Manag. 50:1761.
Taylor RH (1983). Alternative Energy Sources for the Centralised
Generation of Electricity, Adam Hilger, Bristol. P. 56.