ArticlePDF Available

Stochastic analysis and simulation of hydrometeorological processes associated with wind and solar energy

Authors:

Abstract and Figures

The current model for energy production, based on the intense use of fossil fuels, is both unsustainable and environmentally harmful and consequently, a shift is needed in the direction of integrating the renewable energy sources into the energy balance. However, these energy sources are unpredictable and uncontrollable as they strongly depend on time varying and uncertain hydrometeorological variables such as wind speed, sunshine duration and solar radiation. To study the design and management of renewable energy systems we investigate both the properties of marginal distributions and the dependence properties of these natural processes, including possible long-term persistence by estimating and analyzing the Hurst coefficient. To this aim we use time series of wind speed and sunshine duration retrieved from European databases of daily records. We also study a stochastic simulation framework for both wind and solar systems using the software system Castalia, which performs multivariate and multi-time-scale stochastic simulation, in order to conduct simultaneous generation of synthetic time series of wind speed and sunshine duration, on yearly, monthly and daily scale.
Content may be subject to copyright.
1
Stochastic analysis and simulation of hydrometeorological processes
associated with wind and solar energy
Georgios Tsekouras and Demetris Koutsoyiannis
Department of Water Resources and Environmental Engineering, Faculty of Civil
Engineering, National Technical University of Athens, Heroon Polytechneiou 5, GR-157 80
Zographou, Greece(georgtsek@yahoo.gr)
Abstract
The current model for energy production, based on the intense use of fossil fuels, is both
unsustainable and environmentally harmful and consequently, a shift is needed in the
direction of integrating the renewable energy sources into the energy balance. However, these
energy sources are unpredictable and uncontrollable as they strongly depend on time varying
and uncertain hydrometeorological variables such as wind speed, sunshine duration and solar
radiation. To study the design and management of renewable energy systems we investigate
both the properties of marginal distributions and the dependence properties of these natural
processes, including possible long-term persistence by estimating and analyzing the Hurst
coefficient. To this aim we use time series of wind speed and sunshine duration retrieved from
European databases of daily records. We also study a stochastic simulation framework for
both wind and solar systems using the software system Castalia, which performs multivariate
and multi-time-scale stochastic simulation, in order to conduct simultaneous generation of
synthetic time series of wind speed and sunshine duration, on yearly, monthly and daily scale.
Keywords: wind speed, sunshine duration, long term-persistence, Hurst coefficient,
marginal distributions, multivariate stochastic simulation
2
1. Introduction
The drawbacks of conventional energy sources including their negative environmental
impacts emphasize the need to integrate renewable energy sources, like wind and solar, into
the energy balance. However these energy sources, unlike fossil fuels on which the past
model was based on, are unpredictable, at both short and long run, and uncontrollable as they
are associated with relevant hydrometeorological processes which are characterized by high
variability and uncertainty [1]. As a result, the estimation of renewable energy sources
requires analyzing hydrometeorological data, especially wind speed, sunshine duration and
solar radiation time series.
Such hydrometeorological time series are typically modelled as stationary stochastic
processes. In this case their probability distribution does not change in time. It is thus
necessary, in order to characterize uncertainty, to assume a theoretical distribution function
that fits properly the data. Normal distribution is suitable for most hydrometeorological
processes at an annual or higher time scale due to the Central Limit Theorem. However,
variables of lower time scales, such as monthly or daily, are characterized by skewness and
thus, non-normal distributions are more appropriate for their representation. In particular, the
sunshine duration is represented as a random variable bounded from both below and above,
and thus a theoretical distribution with these properties should be chosen for its
representation.
Various studies have been performed in order to examine theoretical distribution
functions fitting properly wind speed and relative sunshine duration data. The Weibull and
Gamma distributions have been found to be appropriate to represent wind speed data while
the Beta distribution has been assumed to be suitable for the relative sunshine duration. Carta
et al. [2] and Zhou et al. [3] proposed that both the Weibull and the Gamma distribution are
suitable for hourly wind speed representation in the Canary Islands and in North Dakota
3
respectively. Garcia et al. [4] and Yilmaz and Celik [5] suggested that the Weibull
distribution fits satisfactorily hourly wind speed data in Spain and Turkey. Darbandi et al. [6]
noted that the Gamma distribution fits well the maximum annual wind speed data in Iran.
Concerning the relative sunshine duration, Sulaiman et al. [7] and Bashahu and Nsabimana
[8] examined the fit of the Beta distribution to daily data in Malaysia and Africa and
concluded that it fitted satisfactorily the data in several cases. Chia and Hutchinson [9]
examined the possibility of fitting the Beta distribution to daily relative cloud duration time
series in Australia.
In addition to the marginal distributional properties, the fact that climate is not static
but exhibits fluctuations at all time scales should also be taken into consideration for the
effective exploitation of renewable energy sources. In a stochastic framework, these
fluctuations are quantified through the dependence of the processes in time [10]. While in
several studies these processes have been regarded as independent identically distributed
random variables, particularly at coarse time scales such as monthly or annual, other studies
have verified the presence of dependence, sometimes long-range dependence, which has been
also termed as long-term persistence (LTP), self-similar behaviour, or the Hurst-Kolmogorov
dynamics. This long-range dependence expresses the tendency of similar physical events to
cluster in time and its quantification is expressed through the Hurst coefficient, the
mathematics of which is discussed later. The Hurst-Kolmogorov dynamics has been identified
by several researchers in various environmental variables [11]. The British engineer Hurst
[12] was the first to notice this behaviour while studying Nile’s runoff and, earlier, the
Russian mathematician Kolmogorov [13] had proposed a mathematical model consistent with
this behaviour when studying turbulence. Later, this behaviour was also observed in other
variables such as river runoff [14,15,16,17,18,19,20], temperature [21,22,23,24,25,17,26,27],
4
climatic indices such as North Atlantic oscillation [28] and decadal Pacific oscillation [29]
and atmospheric pressure fields [30].
So far, the concept of LTP has been incorporated in the modelling of runoff time
series applied to reservoir sizing design studies, where it has been found that neglect of LTP
usually leads to undersizing of the reservoir storage [31-34]. However, the Hurst-Kolmogorov
behaviour has been found to be omnipresent in the most natural processes [11]. Consequently,
it will strongly affect the design variables of any engineering project related to the
exploitation of natural phenomena, like a large scale hybrid renewable energy system
comprising wind and solar energy [35], as it expresses a general natural variability,
manifested by the aforementioned fluctuations in multiple time scales [36,10]. Such large
scale renewable energy projects aim not only at maximizing the profit but also the reliability
in satisfying energy demand or energy target via storage of the excess energy in each time
step of the project’s operation, usually by means of pumped storage hydropower facilities.
Therefore, neglect of this clustering of natural events at the design phase may lead to
undersizing of the facilities for energy storage or overestimation of the reliability of produced
energy.
A prerequisite for the investigation of Hurst-Kolmogorov dynamics is the analysis of
time series having long time period records so that their fluctuations on multiple time scales
can be detected. For this reason a sufficient time length (which is generally regarded 100
years or more and no less than 70-80 years) is required. However, hydrometeorological data
records, relevant to wind and sun, started in the beginning of the previous century or later and
as a consequence time series of such a length are not often available.
Once the marginal distribution and the dependence structure of the wind speed and
sunshine duration processes are identified, another relevant issue is to construct a simulation
model that can simulate such processes for arbitrarily long times, respecting the identified
5
stochastic properties of each process, as well as the cross-correlation between the two
processes. This issue is not trivial as typical stochastic generators may not be able to generate
processes with non-normal distributions and with autocorrelations departing from a Markov
process.
This study aims to perform a comprehensive analysis of existing data with sufficient
length of observation, concerning wind speed and sunshine duration, retrieved from
measuring stations all over Europe, in order to investigate the properties of their marginal
distributions and mainly to detect the possible presence of LTP by estimating and analyzing
the Hurst coefficient values of both processes.
At the same time, it also aims to construct a general stochastic simulation scheme, able
to reproduce the statistical characteristics of the natural series including their long term
variations. By this, the unpredictable fluctuations characterizing the natural resources (wind
and solar radiation) can be taken into account effectively in the design and management of
renewable energy systems that comprise both wind and solar systems, and possibly
hydropower systems with pumped storage, which provide means for energy storage. The
framework is particularly useful to study the reliability in fulfilling the energy demand or the
energy production target. This is attempted by using the multivariate stochastic simulation
system, Castalia, for the simultaneous generation of long synthetic time series, on the concept
of the steady-state simulation, of both variables, being statistically equivalent to the historical
ones at all time scales including the large scale, annual and overannual, in which the Hurst-
Kolmogorov behaviour dominates.
2. Description of data
Wind speed and sunshine duration records used in the analysis were retrieved from measuring
stations whose data are available online. The criteria for station selection were:
6
a) A minimum acceptable record length of 70 years so as the Hurst coefficient to be
estimated with some accuracy.
b) The existence of metadata related to wind speed data and especially information about
the measurement height above ground.
c) A maximum number of three homogeneous periods, in each of which the
measurement height above ground is constant (stations with too frequent changes may
not be reliable).
However, records of this time length are rare worldwide. In particular, in some
continents (Asia, Africa) there is lack of data series of this length. Even in Europe and North
America most of the free access records refer to a period of 50 years or less. Furthermore,
there is total lack of long solar radiation records. However, this variable can be indirectly
estimated by sunshine duration [37].
After an extensive search, wind speed and sunshine duration records fulfilling these
criteria were retrieved exclusively from European databases, the KNMI Climate Explorer, the
European Climate Assessment & Data (ECA&D) and the Deutscher Wetterdienst databases.
A total number of 20 wind speed records and 21 sunshine duration records were found. In
addition, wind speed time series from Dublin Airport and Valentia stations were found from
the website of the Irish Meteorological Service in chart form and were digitized using the
Engauge software. The data retrieved consist of daily records except for the ones of the Irish
stations whose records are annual averages. In Tables 1 and 2 stations details are shown.
As known, wind speed at a certain site increases with height and thus, the analysis of
wind speed time series requires a single measurement height above ground during the time
period of operation. However, the collected daily time series do not refer to a single height
above ground due to changes of the observation height at certain time periods in measuring
stations. Based on the assumption of process stationarity, this problem was overcome by
7
modifying the data of all the time periods except for the last period of records for each station
so as to refer to a unified altitude (data homogenization). Specifically the following procedure
was applied:
a) The daily time series were grouped to time periods where the measurement height
above ground is constant.
b) The average value of each homogeneous period was estimated.
c) The daily values of each period were multiplied with the ratio of the last period
average value to each period average value.
Thus, all records refer to the last measurement height above ground. Neglect of this
modification would lead to incorrect analysis, especially Hurst coefficient overestimation,
because, as exemplified in Figure 1, the annual raw time series at De Bilt station present a
sudden shift owing to the change of measurement height above ground. It is noted though that
the procedure followed may result in negative bias for the Hurst coefficient, because the
assumption that time averages of different periods are equal, which lies behind step (c), is not
true in general and may result in too stable time series.
In addition to wind speed gauge data, the potential of using Reanalysis data for further
investigation of LTP in other sites was also examined. Data were obtained from the
NCEP/NCAR database in grid mode and their reliability was checked by comparing their
records with the station-based ones. This was achieved by selecting the same longitude and
latitude with this of measuring stations. As can be seen in Appendix A, Reanalysis data are
unreliable as their annual average values are higher than the station-based ones even though
they refer to the same or a lower altitude and consequently they are not used any further.
8
3. Analysis
3.1. Properties of marginal distributions
While wind speed is a non-negative variable not bounded from above, sunshine duration has a
domain bounded from both below and above and its upper bound varies in different sites.
Thus, the relative sunshine duration is used instead, which varies in the interval [0,1] as it is
the ratio of the sunshine duration, n, to the astronomical day length, N. The latter depends on
the latitude and time of the year.
In this study, it was verified that both the Weibull and the Gamma distributions are
suitable for wind speed data (as summarized in the Introduction), mainly using graphical
depictions like that in Figure 2. Wind speed on daily scale is characterized by positive
skewness as shown in Figure 2, representing the frequency histogram of the time series at the
Eelde station. The probability density functions of the Gamma and the Weibull distributions
are nearly indistinguishable from each other and fit satisfactorily the data. Thus, both
distribution functions can be regarded as appropriate for the representation of wind speed.
Considering the fact that the Gamma distribution is used in several software systems
(including the software Castalia used in this study as described below) for generation of
synthetic time series, it is useful to also associate the distribution of the relative sunshine
duration with the Gamma distribution by performing a nonlinear transformation so that the
transformed variable takes values in the interval [0,+∞). Specifically, denoting the relative
sunshine duration variable as X, the logarithmic transformation Y= ln(1 X) is defined,
whose values belong indeed to [0,+∞), the value X = 0 corresponds to Y = 0 and X
approaching to 1 corresponds to Y approaching to +∞. As shown in Figure 3, the Gamma
distribution fits satisfactorily the empirical distribution of the transformed variable Y for the
relative sunshine duration data of the Eelde station.
9
Thus, assuming that Y has Gamma distribution with shape parameter κ and scale
parameter λ, it is proved (see Appendix B) that the probability density function of the variable
X is:
 
 
(1)
The probability density functions of both the theoretical distribution (1) and the Beta
distribution are fitted to the frequency histogram of the historical time series of the relative
sunshine duration as shown in Figure 3. These two distributions are almost indistinguishable
from each other and they fit very well the historical data.
3.2. Long-term persistence
3.2.1. The presence of the LTP in wind speed and sunshine duration
The basic condition for the investigation of LTP is the elimination of seasonality so that the
Hurst coefficient estimation be reliable and not affected by periodic fluctuations. As a result,
it is preferable to analyze annual time series. Bakker and Bart [38] analyzed annual
geostrophic wind speed, deriving from sea level pressure, in Northwestern Europe, and
identified LTP using the maximum likelihood method for the simultaneous estimation of
standard deviation and the Hurst coefficient, according to Tyralis and Koutsoyiannis [39].
However, in most relevant studies, time series of lower time scales were used,
especially daily and hourly ones. Haslett and Raftery [40] proposed the existence of long term
persistence in daily wind speed time series, which were firstly deseasonalized and then
adjusted to an ARFIMA model. Bouette et al. [41], based on the former study, adjusted a
GARMA model using the periodogram and Whittle methods and analyzing both seasonality
and long term persistence. The results of the methods indicated that seasonality is the
dominant phenomenon in these time series.
10
Furthermore, several researchers tried to investigate LTP using the so-called
Detrended Fluctuation Analysis [42]. Feng et al. [43] and Kavasseri and Nagarajan [44] used
Multi-Fractal Detrended Fluctuation Analysis to analyze daily and hourly wind speed time
series in China and North Dakota respectively, having previously deseasonalized the data.
Feng et al. concluded that no specific behaviour can be determined while Kavasseri and
Nagarajan noted that wind speed time series are multi-fractal processes, as suggested by the
range of their multi-fractal spectrum. Kocak [45] noted, after analyzing hourly wind speed
time series in Turkey using Detrended Fluctuation Analysis, that the scaling region is divided
into two parts. In the first one, time series fluctuate randomly while in the second one they are
characterized by long term correlations.
Additionally, Rehman and Siddiqi [46] estimated the Hurst coefficient in daily wind
speed time series in Saudi Arabia using the wavelet method, but no exact result was extracted
as these processes presented either persistence or anti-persistence. Finally Benth and Saltyre
[47] claimed that daily wind speed data in Lithuania can be modelled using autoregression
models of third and fourth order which indicates short term memory.
As far as sunshine duration is concerned, Liu et al. [48] noted intense self-similar
behaviour in annual time series in China using the original Hurst’s range analysis which led to
remarkably high Hurst coefficient values. Tsekov [49] also used the Detrended Fluctuation
Analysis in order to examine the existence of LTP in daily sunshine duration time series in
Bulgaria and concluded that data do not indicate long term correlations. Finally, Harrouni and
Guessoum [50] investigated LTP in both daily and annual global solar radiation time series in
Panama and USA using the fractal dimension method and suggested that this process is
characterized by anti-persistence.
11
3.2.2. Stochastic representation of LTP and estimation of the Hurst coefficient
Let Xi denote a hydrometeorological process with i=1,2,…., denoting discrete time (years).
Furthermore, let its mean be μ :=E[Xi], its autocovariance γj :=cov[Xi,Xi+j], its autocorrelation
ρj :=corr[Χii+j] = γj/γ0 (j=0,±1, ±2,…) and its standard deviation σ := γ01/2. Let k be a positive
integer representing time scale of aggregation for the Xi process. The mean aggregated
stochastic process on that time scale is:
()
( 1) 1
1
: ik
k
ii
l i k
XX
k 
(2)
Hydrometeorological processes characterized by LTP can be represented by a
stochastic process known as simple scaling stochastic (SSS) process or a Hurst-Kolmogorov
(HK) process [17,51]. By definition of the HK process, the variance of the mean aggregated
stochastic process is a power law of k with exponent 2H 2, where H is the Hurst coefficient:
( ) 2 2 0
Var[ ] ,
kH
i
Xk
(3)
The value of H varies from 0 to 1. A value in the interval (0,0.5) indicates anti-
persistence, which denotes that an increase in the values of time series is followed by a
subsequent decrease, a behaviour that is not normally observed in nature. A value 0.5
corresponds to a purely random process known as white noise. A value in the interval (0.5,1)
indicates persistence and implies positive autocorrelation. This behaviour is visualized as
multiyear fluctuations and apparent trends, as can be seen in Figure 4, which depicts a
sunshine duration time series characterized by a high value of Hurst coefficient (H=0.9) and
for comparison, a time series of a white noise with the same mean and standard deviation.
From equation (3) it is observed that if standard deviation (square root of variance) is
plotted against time scale on a logarithmic graph, known as the climacogram, a straight line is
formed with slope H 1. Wind speed and sunshine duration data support this behaviour, as an
approximation, as can be seen in the climacograms of Figure 5. Based on this scaling property
12
Koutsoyiannis [17] proposed a method for the simultaneous estimation of standard deviation
σ and the Hurst coefficient H using the least square error of sample standard deviation
estimates (LSSD). The method was later examined by Tyralis and Koutsoyiannis [39] and
Sheng et al. [52] and was found to perform very satisfactorily. In general, the method can be
perceived as a fitting of a straight line in a climacogram like that of Figure 5 using linear
regression of the mean aggregated standard deviation σ(k) on time scale k. The Hurst
coefficient is estimated from the slope of the climacogram using sample standard deviation
estimates on all time scales. The only difference in the LSDD method is that it explicitly
considers the bias of the standard statistical estimator of standard deviation on all time scales.
3.2.3. Results of the LSSD method
The results derived from the LSSD method are summarized in Figure 6 in terms of frequency
histograms of the Hurst coefficient values of the historical time series. Concerning both
variables, markedly high Hurst coefficient values are estimated and it is remarkable that no
value lies in the interval (0,0.5) i.e., no time series is characterized by anti-persistence. The
majority of the values of both variables lie in the interval (0.6,0.9) and it is notable that nearly
15% of the values lie in the interval (0.9,1) which indicates very strong self-similar behaviour.
In general, the results of the method imply that both wind speed and sunshine duration time
series are characterized by Hurst-Kolmogorov behaviour.
3.2.4. Representative value of Hurst coefficient
In order to assume a representative value of the Hurst coefficient for the historical time series
of both variables, the same number of synthetic time series is generated, each one of the same
mean value, standard deviation and record length with the respective historical one, but to
define the autocorrelation function a unique value of the Hurst coefficient is used, the same
for all time series. For this generation the multiple time-scale fluctuation approach [16] is
13
used. This method implies that a weighted sum of three exponential functions of time lag can
approximate satisfactorily the autocorrelation function of a simple scaling process on the
basic time scale. The generated process is the sum of three independent AR(1) processes.
In the end of the generation process the sample value of the Hurst coefficient is
estimated for all synthetic time series using the LSSD method. The frequency histogram of
these H values is then compared to the histogram of the historical H values. By trying
consecutive initial theoretical values of H until a convergence between the histogram of
historical and synthetic timeseries is achieved, it is concluded that the value H = 0.84 can be
regarded as representative for both variables as can be seen in Figure 6.
4. Modelling
4.1. The software system Castalia
Castalia is a software system that was initially introduced for the simulation of
hydrometeorological processes, like rainfall, runoff and evaporation. It performs multivariate
stochastic simulation on annual, monthly and daily scale using the three-parameter Gamma
distribution function in order to generate synthetic time series. The program preserves the
essential marginal statistics, specifically mean value, standard deviation and skewness, as well
as the joint second order statistics, namely auto- and cross-correlation, on these three time
scales. To estimate these statistics, historical daily time series are used as input data and
subsequently, historical monthly and annual time series are created by aggregating daily data
inside the system. Furthermore, Castalia reproduces the LTP considering it as a special
instance of a parametrically defined generalized autocovariance function [53].
The generation of the synthetic time series is performed in three time-aggregation
levels. Initially, a theoretical autocovariance function is defined for each annual variable,
which can include LTP. The dependence structure is reproduced through the Symmetric
Moving Average (SMA) model, whose parameters are estimated using the annual marginal
14
and joint second order statistics and the autocovariance function [53]. The SMA model is
used for the generation of the synthetic annual time series. Following this, auxiliary monthly
time series are generated using the Periodic Autoregressive (PAR(1)) model, whose
parameters are estimated using the monthly marginal and joint second order statistics. A
disaggregation process (called a coupling transformation) modifies the auxiliary monthly time
series in order to become consistent with the annual ones [54]. The arising deviations in the
statistics that must be preserved, which may be caused by the disaggregation process, are
overcome through a repetitive Monte-Carlo process that accompanies the disaggregation
process. By this, a statistically independent sequence of monthly variables is found which
approximates the annual value. The synthetic daily time series are then generated in a similar
manner by disaggregation of monthly values into daily.
4.2. Preservation of probability of zero values
Sunshine duration can take on a zero value with nonzero probability and thus the preservation
of the historical probability of zero values is important. This is similar to rainfall, which is an
intermittent process and the Castalia program, which was originally developed to simulate
rainfall, can handle the intermittent behaviour on the daily time scale [55]. This can be
achieved by appropriate choice of the parameters that are used for this aim into the system
after the generation of daily auxiliary time series. These include:
a) The parameters λ1 and λ2 that are used to adjust the probabilities k1 and k2 of a Markov
chain model. These probabilities are adjusted by the parameters λ1 and λ2, respectively,
of the historical probability of zero values of each variable. Specifically, k1 expresses
the probability of a zero value occurring in the current time step if there is also a zero
value in the previous time step while k2 expresses the probability of a zero value
occurring in the current time step if there is a non-zero value in the previous time step.
15
b) A parameter k3 which expresses the probability of the values of all variables in the
current time step being zero if one of them is zero.
c) The parameters of a round-off rule and specifically a threshold l0 and a percentage π0,
where the rule implies that a percentage π0 of values below this threshold are
converted into zero values.
Note that all three procedures described above generate zero values and thus they all
contribute to the final frequency of zero values.
4.3. Results
4.3.1. First application Untransformed variables
In the first application of the program 1000-year synthetic time series are generated in 8
measuring stations having both wind speed and sunshine duration records (16 variables in
total). The length of the simulation period (1000 years) was determined as explained in
Appendix C. Although the domain of the sunshine duration is bounded from above and below
and thus, the Gamma distribution is not appropriate to represent this variable, data are initially
imported into the system without any transformation to test the program performance in a fast
application without pre-processing of data. For this reason possible deviations concerning the
skewness of the sunshine duration variable are expected.
The model parameters are calculated automatically by the program except for those
controlling the preservation of the probability of zero values, which are determined by a trial
and error procedure. For the latter, the following values were finally adopted π0 = 0.9 for
l0=0.4; λ1 = 0.3, λ2 = 0.1; k3= 0.
Application results are presented in Figures 7-10 for one of the stations (Eelde). It is
observed that the statistical characteristics, the Hurst coefficient as well as the cross-
correlation coefficient of both processes are preserved on annual and monthly scale. On daily
scale wind speed statistics and cross-correlation coefficient are also satisfactorily preserved.
16
However, the skewness of the sunshine duration variable is overestimated. This divergence is
caused, as expected, by the fact that the domain of the sunshine duration variable differs from
the domain of the Gamma distribution, used for the generation of the time series. As a
consequence, several synthetic sunshine duration values, being higher than the upper bound of
the variable domain, are generated due to the fact that the domain of Gamma distribution is
the [0,+∞). These values are obviously unrealistic and lead to skewness overestimation.
4.3.2. Second application transformed sunshine duration
In order to overcome the overestimation of the skewness which was encountered in the first
application, a second application is performed using the previous wind speed data and the
proposed logarithmic transformation of sunshine duration data in each of the 8 stations. After
the generation of the synthetic time series, the Y daily time series are manually subject to the
reverse transformation so as the synthetic daily time series of the sunshine duration to arise.
Then these time series are compared with the historical ones.
The model parameters controlling the preservation of the probability of zero values are
also determined by a trial and error procedure. For the latter, the following values were finally
adopted π0 = 0.9 for l0 =0; λ1 = 0.3, λ2 = 0.1; k3 = 0.
Both the Hurst coefficient and the marginal statistical characteristics of the wind speed
process are preserved satisfactorily on all time scales as in the first application. Application
results for the sunshine duration are presented in Figures 11-14 for the same station as before
(Eelde). The Hurst coefficient and the statistical characteristics on annual and monthly scale
are also generally preserved. The only deviations observed are related to monthly skewness,
which is underestimated, and monthly autocorrelation, which is overestimated. However,
these features are not of great importance as the design focuses on the daily scale. It is
remarkable that the cross-correlation of the variables is preserved even though the logarithmic
transformation is used in the simulation. On daily time scale, skewness is now satisfactorily
17
reproduced and the only existing deviation refers to autocorrelation which is slightly
overestimated. The cross-correlation of the variables, which is high as the time scale of the
design is daily, as well as the probability of zero values are also preserved. The deviations
mentioned above on monthly and daily time scales are caused due to the fact that the iterative
Monte Carlo process and the adjusting-disaggregation procedures are performed inside the
program, so they are applied to the logarithmic transformation of the variable instead of the
actual one.
5. Conclusions
Simulation of wind speed and sunshine duration gains interest because these processes are
associated with renewable energy production. Knowing the marginal distribution as well as
the stochastic structure of these processes, including LTP, is important as the solar and wind
energy strongly depend on them.
The results derived from the analysis using data from Europe indicate that the Gamma
and the Weibull distributions are suitable for representing wind speed. For the relative
sunshine duration a theoretical distribution function, which is a transformation of the Gamma
distribution, is derived. This is practically indistinguishable from the Beta distribution and
they both fit satisfactorily to relative sunshine duration data.
Both processes are found to be characterized by strong self-similar behaviour with
values of the Hurst coefficient, H, (estimated using the LSSD method) markedly higher than
the value H = 0.5, corresponding to white noise, for the majority of the time series. After
implementing a Monte Carlo approach, a unique value H = 0.84 is found to be representative
for both variables. However, it should be noted that, due to the small amount of samples,
further investigation is required. Also, as the observations have started in their majority, all
over the world, after 1940, the record lengths are not quite sufficient to support a safe
estimation of the Hurst coefficient.
18
The detailed empirical analysis of the stochastic properties allows stochastic
simulation of the two processes which is particularly useful for the rational design of a
renewable energy system comprising wind and solar energy. Specifically, synthetic time
series of wind speed and sunshine duration data are simultaneously generated on annual,
monthly and daily scale using the software system Castalia.
Castalia performs this stochastic simulation satisfactorily as the marginal and joint
second order statistics as well as the LTP of the historical data retrieved from measuring
stations with common observations are well reproduced. An overestimation of the skewness
of daily sunshine duration which arises when raw data are used is attributed to the bounded
domain of the variable and it can be overcome by using the proposed logarithmic
transformation. Some deviations which arise in the monthly skewness and autocorrelation
when the transformed variable is used are not of great importance due to the fact that the time
scale of the design is daily while the overestimation of the daily autocorrelation is slight.
Thus, considering that Castalia was initially developed for the simulation of rainfall, runoff
and evaporation, it is also confirmed that this program can conduct stochastic simulations of a
wide spectrum of hydrometeorological variables on three important time scales, annual,
monthly and daily.
Acknowledgements: We thank two anonymous reviewers for the constructive review
comments on an earlier version of this manuscript as well as Andreas Efstratiadis, Stefanos
Kozanis, Yannis Dialynas and Panagiotis Dimitriadis for their help during the preparation of
the study.
19
References
[1] Koutsoyiannis D, Makropoulos C, Langousis A, Baki S, Efstratiadis A, Christofides
A, et al. Climate, hydrology, energy, water: recognizing uncertainty and seeking
sustainability. Hydrol Earth Syst Sci 2009;13:247-57.
[2] Carta JA, Ramirez P, Velasquez S. A review of wind speed probability distributions
used in wind energy analysis: case studies in the Canary Islands. Renew Sust Energ
Rev 2009;13:933-55.
[3] Zhou J, Erdem E, Li G, Shi J. Comprehensive evaluation of wind speed distribution
models: a case study for North Dakota sites. Energy Convers Manag 2010;51:1449-
58.
[4] Garcia A, Torres JL, Prieto E, de Francisco A. Fitting wind speed distributions: a case
study. Sol Energy 1998;62:139-44.
[5] Yilmaz V, Çelik HE. A statistical approach to estimate the wind speed distribution:
the case of Gelibolu region. Dogus¸ Üniversitesi Dergisi 2008;9:122-32.
[6] Darbandi A, Aalami MT, Asadi H. Comparison of four distributions for frequency
analysis of wind speed. Env Nat Resour Res 2012;2.
[7] Sulaiman YM, Hlaing OWM, Wahab MA, Zakaria A. Application of beta distribution
model to Malaysian sunshine data. Renew Energy 1999;18:573-9.
[8] Bashahu M, Nsabimana JC. Statistical analysis of sunshine duration measurements in
Burundi using beta distributions. In: World Conference on Physics and Sustainable
Development Durban Poster No102244, South Africa 2005.
[9] Chia E, Hutchinson MF. The beta distribution as a probability model for daily cloud
duration. Agric For Meteorol 1991;56:195-208.
[10] Koutsoyiannis D. Hurst-Kolmogorov dynamics and uncertainty. J Am Water Resour
Assoc 2011;47:481-95.
20
[11] Markonis Y, Koutsoyiannis D. Climatic variability over time scales spanning nine
orders of magnitude: connecting Milankovitch cycles with Hurst-Kolmogorov
dynamics. Surv Geophys 2013;34:181-207.
[12] Hurst HE. Long term storage capacities of reservoirs. Trans Am Soc Civ Eng
1951;16:776-808.
[13] Kolmogorov AN. Wienersche Spiralen und einige andere interessante Kurven in
Hilbertschen Raum. Dokl Akad Nauk 1940;26:115-8.
[14] Eltahir EAB. El Niño and the natural variability in the flow of the Nile river. Water
Resour Res 1996;32:131-7.
[15] Jiang T, Zhang Q, Blender R, Fraedrich K. Yangtze Delta floods and droughts of the
last millennium: abrupt changes and long term memory. Theor Appl Climatol
2005;82:131-41.
[16] Koutsoyiannis D. The Hurst phenomenon and fractional Gaussian noise made easy.
Hydrol Sci J 2002;47:573-95.
[17] Koutsoyiannis D. Climate change, the Hurst phenomenon, and hydrological statistics.
Hydrol Sci J 2003;48:3-24.
[18] Radziejewski M, Kundzewicz ZW. Fractal analysis of flow of the river Warta. J
Hydrol 1997;200:280-94.
[19] Sakalauskiene G. The Hurst phenomenon in hydrology. Environ Res Eng Manag
2003;3:16-20.
[20] Wang G, Jiang T, Chen G. Structure and long-term memory of discharge series in
Yangtze River. Acta Geogr Sin 2006;61:47-56.
[21] Alvarez-Ramirez J, Alvarez J, Dagdug L, Rodriguez E, Carlos Echeverria J. Long
term memory dynamics of continental and oceanic monthly temperatures in the recent
125 years. J Phys A 2008;387:3629-40.
21
[22] Bloomfield P. Trends in global temperature. Clim Change 1992;21:1-16.
[23] Fraedrich K, Blender R. Scaling of atmosphere and ocean temperature correlations in
observations and climate models. Phys Rev Lett 2003;90:1-4.
[24] Fraedrich K, Blender R, Zhu X. Continuum climate variability: long-term memory,
extremes, and predictability. Int J Mod Phys B 2009;23:5403-16.
[25] Koscielny-Bunde E, Bunde A, Havlin S, Roman HE, Goldreich Y, Schellnhuber HJ.
Indication of a universal persistence law governing atmospheric variability. Phys Rev
Lett 1998;81:729-32.
[26] Varotsos C, Kirk-Davidoff D. Long-memory processes in ozone and temperature
variations at the region 60_ Se60_ N. Atmos Chem Phys 2006;6:4093-100.
[27] Yano J, Blender R, Zhang C, Fraedrich K. 1/f e noise and pulse-like events in the
tropical atmospheric surface variabilities. Q J Roy Meteorol Soc 2004;130:1697-721.
[28] Stephenson DB, Paven P, Bojariu R. Is the North Atlantic oscillation a random walk?
Int J Climatol 2000;20:1-18.
[29] Khaliq MN, Gachon P. Pacific decadal oscillation climate variability and temporal
pattern of winter flows in Northwestern North America. J Hydrometeorol
2010;11:917-33.
[30] Giannoulis S, Ioannou C, Karantinos E, Malatesta L, Theodoropoulos G, Tsekouras G,
et al. Long term properties of monthly atmospheric pressure fields, EGU General
AssemblyIn Geophysical research abstracts, vol. 14; 2012. Vienna, 4680.
[31] Koutsoyiannis D., Reliability concepts in reservoir design. Water Encyclopedia, Vol.
4, Surface and Agricultural Water, edited by J. H. Lehr and J. Keeley, 259265,
Wiley, New York, 2005.
[32] Kottegoda NT. Stochastic water resources technology. London: Macmillan Press;
1980. p. 384.
22
[33] Klemes V, Srikanthan R, McMahon TA. Long-memory flow models in reservoir
analysis: what is their practical value? Water Resour Res 1981;17:737-51.
[34] Lettenmaier DP, Burges SJ. Operational assessment of hydrologic models of long-
term persistence. Water Resour Res 1977;13:113-24.
[35] Tsekouras G, Ioannou C, Efstratiadis A, Koutsoyiannis D. Stochastic analysis and
simulation of hydrometeorological processes for optimizing hybrid renewable energy
systems, EGU General Assembly In Geophysical research abstracts, vol. 15; 2013.
Vienna.
[36] Koutsoyiannis D. Nonstationarity versus scaling in hydrology. J Hydrol
2006;324:239-54.
[37] Koutsoyiannis D, Xanthopoulos Th. Engineering hydrology (in Greek). 3rd ed.
Athens: National Technical University of Athens; 1999. p. 418.
[38] Bakker AMR, Van den Hurk J J M Bart. Estimation of persistence and trends in
geostrophic wind speed for the assessment of wind energy yields in Northwest Europe
4. Clim Dynam 2012;39:767-82.
[39] Tyralis H, Koutsoyiannis D. Simultaneous estimation of the parameters of the Hurst-
Kolmogorov stochastic process. Stoch Env Res Risk A 2011;25:21-33.
[40] Haslett J, Raftery AE. Space modeling with long memory dependence: assessing
Ireland’s wind power resource. Appl Statv 1989;38:1-50.
[41] Bouette JC, Chassagneux JF, Sibai D, Terron R, Charpentier A. Wind in Ireland: long
memory or seasonal effect? Stoch Env Res Risk A 2006;20:141-51.
[42] Peng C-K, Buldyrev SV, Havlin S, Simons M, Stanley HE, Goldberger AL. Mosaic
organization of DNA nucleotides. Phys Rev E 1994;49:1685-9.
[43] Feng T, Fu Z, Deng X, Mao J. A brief description to different multi-fractal bevaviors
of daily wind speed records over China. Phys Lett A 2009;373:4134-41.
23
[44] Kavasseri RG, Nagarajan R. A multifractal description of wind speed records. Chaos
Soliton Fract 2005;24:165-73.
[45] Kocak K. Examination of persistence properties of wind speed records using
detrended fluctuation analysis. Energy 2009;34:1980-5.
[46] Rehman S, Siddiqi AH. Wavelet based Hurst exponent and fractal dimensional
analysis of Saudi climatic dynamics. Chaos Soliton Fract 2009;40:1081-90.
[47] Benth JS, Saltyte L. Spatial-temporal model for wind speed in Lithuania. J Appl Stat
2011;38:1151-68.
[48] Liu Liu, Xu ZX, Huang JX. Statial-temporal variation and abrupt changes for major
climate variables in the Taihu Basin, China. Stoch Env Res Risk A 2011;26:777-91.
[49] Tsekov M. Detrended fluctuation analysis of weather records from local place in south
Bulgaria. C R Acad Bulgare Sci 2003;56:8-25.
[50] Harrouni S, Guessoum A. Using fractal dimension to quantify long-range persistence
in global solar radiation. Chaos Soliton Fract 2009;41:1520-30.
[51] Koutsoyiannis D. A random walk on water. Hydrol Earth Syst Sci 2010;14:585-601.
[52] Sheng H, Chen YQ, Qiu TS. Tracking performance and robustness analysis of Hurst
estimators for multifractional processes. IET Signal Process 2012;6:213-26.
[53] Koutsoyiannis D. A generalized mathematical framework for stochastic simulation
and forecast of hydrologic time series. Water Resour Res 2000;36:1519-33.
[54] Koutsoyiannis D. Coupling stochastic models of different time scales. Water Resour
Res 2001;37:379-92.
[55] Koutsoyiannis D, Onof C, Wheater HS. Multivariate rainfall disaggregation at a fine
timescale, Water Resour Res, 2002:39, doi:10.1029/2002WR001600
[56] Koutsoyiannis D. Stochastic hydrology (in Greek). 4th ed. Athens: National Technical
University of Athens; 1997. p. 312.
24
[57] Koutsoyiannis D. On the quest for chaotic attractors in hydrological processes.
Hydrolog Sci J 2006;51:1065-91.
25
Appendix A: Assessment of Reanalysis data
Figure A.1 compares station-based and reanalysis data for the location of the De Bild station
at an annual basis. It can be seen that the two series do not correlate well. In addition the
reanalysis data series has annual average values higher than the station-based ones even
though the former refer to lower altitude than the latter.
Figure A.1: Mean annual wind speed time series from station-based (20m above ground) and
Reanalysis data (10m above ground) at De Bilt station.
Appendix B: Proof of equation (1)
Let X, Y, where 0< X <1 be random variables so that:
(B.1)
(B.2)
The event {Χ x} is identical to the event {Υ 1(x)}. As a result, the distribution functions
of Υ and Χ are linked through the equation:
              
(B.3)
The variables are continuous, and the function g is differentiable with derivative equal to:
      
(B.4)
Τhe probability density function of the variable X is a function of the density of the variable Y
[56]:
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
1940 1960 1980 2000 2020
Average annual wind
speed (m/s)
Year
Reanalysis
Station based
26
 = 
 =

  ]
(B.5)
As the variable Y for the continuous part of the distribution (Y > 0) is represented by the
Gamma distribution function with shape parameter κ > 0 and scale parameter λ > 0, the
probability density function of the variable X is:
 

     
(B.6)
or
(B.7)
and finally
(B.8)
Appendix C: Determination of the length of simulation period
In the estimation, through stochastic simulation, of a probability p, from a sample with
relatively large size N of independent identically distributed random variables, it is known
that the length of the confidence interval of the estimate is [31,57]:
2 z(1+γ)/2 [p (1 p) / N]0.5 = 2 c p
(1)
where γ is the confidence coefficient, za is the a-quantile of the standard normal distribution,
and c is the acceptable relative error. Solving for N, the minimum sample size Nmin that is
required for estimating the probability p is:
Nmin= (z2(1 + γ) / 2 / c2) (1/p 1)
(2)
For a daily step of design, a confidence coefficient γ = 0.95, an acceptable relative error c =
10% and probability of failure p = 1‰, the required sample size (number of days of
simulation) needed to obtain an accurate estimate of p on a daily basis is Nmin ≈ 384000
(days), which is rounded off to a simulation period of 1000 years. It is noted though that, as
the hydrometeorological processes are not independent in time, the required period should be
greater than 1000 years.
27
Tables
Station
Country
Longitude(ο)
Latitude (ο)
Measurement height
above ground (m)
Altitude
(m)
Source
Period of
record
Vlissingen
Netherlands
3.60 E
51.45N
15
8
ΚΝΜΙ
1909-2011
Regenburg
Germany
12.10 E
49.00 N
15
377
ECA&D
1946-2011
Potsdam
Germany
13.10 E
52.40 N
37.7
81
ECA&D
1911-2011
Maastricht
Netherlands
5.78 E
50.92 N
10
114
ΚΝΜΙ
1906-2011
Karlsruhe
Germany
8.40 E
49.00 N
47.4
113
ECA&D
1945-2011
Hohenpeissenberg
Germany
11.01 E
47.48 N
15
980
ECA&D
1939-2011
Hof
Germany
11.90 E
50.30 N
12
624
ECA&D
1946-2011
Giessen
Wettenberg
Germany
8.38 E
50.36 N
10
203
ECA&D
1939-2011
Eelde
Netherlands
6.58 E
53.13 N
10
3.5
ΚΝΜΙ
1904-2011
Dresden
Germany
13.80 E
51.10 N
10
220
ΚΝΜΙ
1941-2011
De Kooy
Netherlands
4.78 E
52.92 N
10
0.5
ΚΝΜΙ
1908-2011
De Bilt
Netherlands
5.18 E
52.10 N
20
1.9
ΚΝΜΙ
1904-2011
Tarifa
Spain
5.35 W
36.00 N
10
32
ECA&D
1945-2012
Bjoernoeya
Norway
19.0 E
74.50 N
10
16
ECA&D
1920-2011
San Sebastian
Spain
2.54 W
43.18 N
10
251
ECA&D
1933-2011
Dublin Airport
Ireland
6.30 W
53.42 N
-
85
MET
EIREANN
1944-2010
Valentia
Ireland
10.24 W
51.94 N
-
9
MET
EIREANN
1940-2010
Fokstua
Norway
9.17 E
62.07 N
10
975
ECA&D
1923-2011
Karasjok
Norway
25.52 E
69.47 N
10
133
ECA&D
1939-2011
Sula
Norway
8.45 E
63.85 N
10
6
ECA&D
1938-2011
Table 1: Stations with wind speed measurements
28
Station
Country
Longitude(ο)
Latitude (ο)
Altitude (m)
Source
Period of record
Alicante
Spain
0.60 W
38.30 N
82
ECA&D
1939-2011
Vlissingen
Netherlands
3.60 E
51.45N
8
KNMI
1907-2011
Lindenberg
Germany
14.10 E
52.20 N
112
ECA&D
1907-2011
Potsdam
Germany
13.10 E
52.40 N
81
ECA&D
1901-2011
Maastricht
Netherlands
5.78 E
50.92 N
114
KNMI
1906-2011
Karlsruhe
Germany
8.40 E
49.00 N
113
ECA&D
1936-2011
Hohenpeissenberg
Germany
11.01 E
47.48 N
980
ECA&D
1937-2011
Zugspitze
Germany
10.98 E
47.42 N
2960
ECA&D
1901-2011
Zuerichfluntern
Switzerland
8.57 E
47.38 N
556
ECA&D
1901-2011
Eelde
Netherlands
6.58 E
53.13 N
3.5
KNMI
1906-2011
Geneve
Switzerland
6.10 E
46.20 N
238
ECA&D
1901-2011
De Kooy
Netherlands
4.78 E
52.92 N
0.5
KNMI
1909-2011
De Bilt
Netherlands
5.18 E
52.10 N
1.9
KNMI
1901-2011
Granada
Spain
3.59 W
37.17 N
680
ECA&D
1942-2010
Aachen
Germany
6.10 E
50.80 N
264
ECA&D
1935-2010
Basel
Switzerland
7.61 E
47.56 N
265
ECA&D
1901-2011
Lugano
Switzerland
8.97 E
46.00 N
276
ECA&D
1901-2011
Madrid
Spain
3.65 W
40.40 N
687
ECA&D
1920-2011
Saentis
Switzerland
9.40 E
47.20 N
2500
ECA&D
1901-2011
Valencia
Spain
0.38 W
39.48 N
35
ECA&D
1938-2011
Zagreb
Croatia
16.03 E
45.82 N
123
ECA&D
1900-2011
Table 2: Stations with sunshine duration measurements
29
Figures
Figure 1: Annual averages of wind speed time series at De Bilt station before and after
modification; three different periods were identified: 1904-1960 (37 m above ground), 1961-
1992 (10 m above ground), 1993-2011 (20 m above ground).
Figure 2: Frequency histograms and probability density functions of historical daily wind
speed time series at Eelde station in June.
2.5
3
3.5
4
4.5
5
5.5
6
1900 1920 1940 1960 1980 2000 2020
Average annual
wind speed (m/s)
Year
Raw time
series
Modified time
series
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.25
0.75
1.25
1.75
2.25
2.75
3.25
3.75
4.25
4.75
5.25
5.75
6.25
6.75
7.25
7.75
8.25
8.75
9.25
9.75
10.25
Probability density
Wind speed (m/s)
Wind speed
frequency
histograms
Gamma
distribution
Weibull
distribution
30
Figure 3: Frequency histograms and probability density functions of the logarithmic
transformation of historical daily sunshine duration time series (above) and the relative
sunshine duration time series (below) at Eelde station in June.
0
0.25
0.5
0.75
1
1.25
1.5
0.125
0.375
0.625
0.875
1.125
1.375
1.625
1.875
2.125
2.375
2.625
2.875
3.125
3.375
3.625
Probability density
Transformed relative sunshine duration
Transformed
sunshine duration
frequency
histograms
Gamma
distribution
0
0.25
0.5
0.75
1
1.25
1.5
1.75
2
0.05
0.15
0.25
0.35
0.45
0.55
0.65
0.75
0.85
0.95
Probability density
Relative sunshine duration
Relative sunshine
duration
frequency
histograms
Theoretical
distribution (1)
Beta distribution
31
Figure 4: Annual and 10-year averages of sunshine duration time series at Vlissingen station
(above) and a series of white noise with the same mean and standard deviation (below).
Figure 5: Climacograms of wind speed (Valentia, G.Wettenberg) and sunshine duration
(Vlissingen, De Bilt) time series.
2.5
3
3.5
4
4.5
5
5.5
6
1900 1920 1940 1960 1980 2000 2020
Sunshine duration (h)
Year
Average
annual time
series
10 year
average
2.5
3
3.5
4
4.5
5
5.5
6
1900 1920 1940 1960 1980 2000 2020
Year
Average
annual time
series
10 year
average
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1 00.2 0.4 0.6 0.8 1
log (σ(k) /σ)
log (k)
G.Wettenberg
(Η=0.515)
Valentia
(Η=0.915)
Vlissingen
(Η=0.895)
De Bilt (Η=0.667)
32
Figure 6: Comparison of Hurst coefficient frequency histograms of historical and synthetic,
with a theoretical H = 0.84, time series for wind speed (above) and sunshine duration (below).
Figure 7: Comparison of Hurst coefficient of historical and synthetic time series at the Eelde
station; first application.
0%
5%
10%
15%
20%
25%
30%
0.5-0.6 0.6-0.7 0.7-0.8 0.8-0.9 0.9-1
Percentage of stations
Hurst coefficient
Historical
time series
Synthetic
time series
[3AR(1)]
0
0.1
0.2
0.3
0.4
0.5
0.5 - 0.6 0.6 - 0.7 0.7 - 0.8 0.8 - 0.9 0.9 -1
Percentage of stations
Hurst coefficient
Historical
time series
Synthetic
time series
[3AR(1)]
0
0.2
0.4
0.6
0.8
1
Wind speed Sunshine duration
Hurst coefficient
Historical
Synthetic
33
Figure 8: Comparison of annual, monthly and daily standard deviations and skewness
coefficients of historical and synthetic wind speed and sunshine duration time series at the
Eelde station; first application.
0
0.5
1
1.5
2
2.5
3
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Standard deviations
(m/s)
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
0
1
2
3
4
5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Standard deviations
(h)
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Skewness coefficient
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
-0.5
0
0.5
1
1.5
2
2.5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Skewness coefficient
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
34
Figure 9: Comparison of annual, monthly and daily auto- and cross-correlation coefficient of
historical and synthetic wind speed time series at the Eelde station; first application.
Figure 10: Comparison of the probability of zero sunshine duration of the historical and
synthetic time series at the Eelde station; first application.
-0.05
0.05
0.15
0.25
0.35
0.45
0.55
0.65
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Autocorrelation
coefficient
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
0
0.1
0.2
0.3
0.4
0.5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Autocorrelation
coefficient
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
-0.40
-0.35
-0.30
-0.25
-0.20
-0.15
-0.10
-0.05
0.00
0.05
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Crosscorrelation
factor
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
0
0.1
0.2
0.3
0.4
0.5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Probability of zero
values
Historical (daily)
Synthetic (daily)
35
Figure 11: Comparison of Hurst coefficient of historical and synthetic time series at the Eelde
station; second application.
Figure 12: Comparison of annual, monthly and daily standard deviations and skewness
coefficients of historical and synthetic sunshine duration time series at the Eelde station;
second application.
0
0.2
0.4
0.6
0.8
1
Wind speed Sunshine duration
Hurst coefficient
Historical
Synthetic
0
1
2
3
4
5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Standard deviations
(h)
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
-0.5
-0.1
0.3
0.7
1.1
1.5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Skewness
coefficioent
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
36
Figure 13: Comparison of annual, monthly and daily auto- and cross-correlation coefficient of
historical and synthetic sunshine duration time series at the Eelde station; second application.
Figure 14: Comparison of the probability of zero sunshine duration of the historical and
synthetic time series at the Eelde station; second application.
0
0.1
0.2
0.3
0.4
0.5
0.6
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Autocorrelation
coefficient
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
-0.35
-0.25
-0.15
-0.05
0.05
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Crosscorrelation
coefficient
Historical (monthly)
Synthetic (monthly)
Historical (daily)
Synthetic (daily)
0
0.1
0.2
0.3
0.4
0.5
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Year
Probability of zero
values
Historical (daily)
Synthetic (daily)
... On CN Links: [207], Epidemics: [208,209], Message, memes: [210,211] Of CN Links: [212,213,214], Nodes: [215], Hubs: [216], Communities: [217], Person's social signature: [218], Patterns: [219,220], Processes Rumors: [221], Epidemics: [222], Emotions: [223] Economics and Market Analysis Inflation [224,225,226,227,228,226,229,227,230,231,228] Expected Returns [232,233,234,233,235,236] Exchange Rates and Crypto-currencies [237,238,239,240,241,242,243,244,245,246,246,247] Stock Market [15,248,249,250,251,252,253,254,255,256,257,258] Market Analysis [259,260,261,262,263,264,265,266,267,268,269,270,271] [272,273,274] Non-equilibrium Thermodynamics Non-equilibrium systems [23,59,60,61,275,276,277,278,279,280,281,282,283] Synthetic time series Fourier filtering [284,285,286,287,288] Hydrology [289,290,290,291,292,293,294] Wind [295] Optimization and planning Robust Optimization [24,296,297,298,299,300,301] Planning [302,303,304] Health and biomedical sciences Biomedicine [27,305,306,307,308] Sport [309] ...
... In fact, the simulation of hydrological systems is a constant source of different methods for synthetic time series construction with persistence properties [290]. In [295] real and simulated time series of wind speed and sunshine duration are used study the design and management of renewable energy systems. The paper analyzes both the properties of marginal distributions and the dependence properties of these natural processes, including possible long-term persistence by estimating and analyzing the Hurst coefficient of the series. ...
... The paper analyzes both the properties of marginal distributions and the dependence properties of these natural processes, including possible long-term persistence by estimating and analyzing the Hurst coefficient of the series. These two papers ( [289] and [295]) use the software Castalia [292,293] to generate the synthetic time series with persistence properties, which implements the framework defined in [294]. In [291] a new method is presented to generate stationary multi-site hydrological time series. ...
Preprint
Full-text available
Persistence is an important characteristic of many complex systems in nature, related to how long the system remains at a certain state before changing to a different one. The study of complex systems' persistence involves different definitions and uses different techniques, depending on whether short-term or long-term persistence is considered. In this paper we discuss the most important definitions, concepts, methods, literature and latest results on persistence in complex systems. Firstly, the most used definitions of persistence in short-term and long-term cases are presented. The most relevant methods to characterize persistence are then discussed in both cases. A complete literature review is also carried out. We also present and discuss some relevant results on persistence, and give empirical evidence of performance in different detailed case studies, for both short-term and long-term persistence. A perspective on the future of persistence concludes the work.
... The use of stochastic models for generating long synthetic data, to be input to deterministic models, is a common practice in water resources and other environmental sciences [12]. The literature reports numerous modelling attempts for representing wind, solar and hydrological drivers through statistical and, less often, stochastic approaches [13][14][15][16]. The latter offer a consistent basis for process description, since they also account for dependencies in space, i.e. among correlated processes [17], and time. ...
Article
Full-text available
As the share of renewable energy resources rapidly increases in the electricity mix, the recognition, representation, quantification, and eventually interpretation of their uncertainties become important. In this vein, we propose a generic stochastic simulation-optimization framework tailored to renewable energy systems (RES), able to address multiple facets of uncertainty, external and internal. These involve the system's drivers (hydrometeorological inputs) and states (by means of fuel-to-energy conversion model parameters and energy market price), both expressed in probabilistic terms through a novel coupling of the triptych statistics, stochastics and copulas. Since the most widespread sources (wind, solar, hydro) exhibit several common characteristics, we first introduce the formulation of the overall modelling context under uncertainty, and then offer uncertainty quantification tools to put in practice the plethora of simulated outcomes and resulting performance metrics (investment costs, energy production, revenues). The proposed framework is applied to two indicative case studies, namely the design of a small hydropower plant (particularly, the optimal mixing of its hydro-turbines), and the long-term assessment of a planned wind power plant. Both cases reveal that the ignorance or underestimation of uncertainty may hide a significant perception about the actual operation and performance of RES. In contrast, the stochastic simulation-optimization context allows for assessing their technoeconomic effectiveness against a wide range of uncertainties, and as such provides a critical tool for decision making, towards the deployment of sustainable and financially viable RES.
... Therefore, based on the scientific boost, the climacogram (and not the other two metrics) was found to be adequate for the identification and model building of a stochastic process. Since then, interest in the scale domain and the climacogram estimator has increased, and the climacogram has been implemented in education material [49], and has been used to identify the LTP behaviour in various scientific studies, such as 2D precipitation fields [50], multidimensional spatiotemporal domain [51], paleoclimatic temperature [52] and precipitation [53,54], Bayesian statistical models of rainfall and temperature [55], higher-order moments of skewness and kurtosis vs. scale in grid turbulence [26], annual precipitation [56], water demand [57], daily river flows [58], precipitation and temperature for a bivariate drought analysis [59], wind and solar energy [60], water-energy nexus [61], solar radiation [62], wave height and period [63], daily streamflow [64], and monthly temperature and precipitation ( [65,66]), annual streamflow ( [30,66]), ecosystem variability [67], 2D rock formations [68], urban streamflows [69], global temperature and wind of resolution spanning 10 orders of magnitude from ms to several decades [70], disaggregation schemes from daily to hourly rainfall and runoff [71], hourly wind and daily precipitation [26], fine scale precipitation [3,22,[72][73][74][75][76][77][78], fine scale wind [26,70], 2D landscapes [79,80], flood risk assessment [81], bridge scour [82], art [83][84][85], spatiotemporal evolution of clustering [86], alternative statistical moments (e.g., L-moments, K-moments; [3,87]), comparison of multi-source data [88], daily extreme global temperature [89], hourly offshore and coastal wind for energy assessment [90], and in many other applications [3,18]. ...
Article
Full-text available
To seek stochastic analogies in key processes related to the hydrological cycle, an extended collection of several billions of data values from hundred thousands of worldwide stations is used in this work. The examined processes are the near-surface hourly temperature, dew point, relative humidity, sea level pressure, and atmospheric wind speed, as well as the hourly/daily streamflow and precipitation. Through the use of robust stochastic metrics such as the K-moments and a secondorder climacogram (i.e., variance of the averaged process vs. scale), it is found that several stochastic similarities exist in both the marginal structure, in terms of the first four moments, and in the secondorder dependence structure. Stochastic similarities are also detected among the examined processes, forming a specific hierarchy among their marginal and dependence structures, similar to the one in the hydrological cycle. Finally, similarities are also traced to the isotropic and nearly Gaussian turbulence, as analyzed through extensive lab recordings of grid turbulence and of turbulent buoyant jet along the axis, which resembles the turbulent shear and buoyant regime that dominates and drives the hydrological-cycle processes in the boundary layer. The results are found to be consistent with other studies in literature such as solar radiation, ocean waves, and evaporation, and they can be also justified by the principle of maximum entropy. Therefore, they allow for the development of a universal stochastic view of the hydrological-cycle under the Hurst–Kolmogorov dynamics, with marginal structures extending from nearly Gaussian to Pareto-type tail behavior, and with dependence structures exhibiting roughness (fractal) behavior at small scales, long-term persistence at large scales, and a transient behavior at intermediate scales.
... Since the research about SASR and PSD on Mars does not take terrain influences into account, the research about SASR and PSD taking terrain influences into account is an unexplored field in the current study of solar radiation on Mars. A large number of papers have studied the SASR and PSD over rugged terrains for different sites on Earth [5,6,16,19,22,23,58,59,63,[66][67][68]. On the contrary, little attention has been paid to Mars. ...
Article
Full-text available
Solar radiation may be shielded by the terrain relief before reaching the Martian surface, especially over some rugged terrains. Yet, to date, no comprehensive studies on the spatial structure of shielded astronomical solar radiation (SASR) and the possible sunshine duration (PSD) on Mars have been conducted by previous researchers. Previous studies generally ignored the influences of the terrain on the SASR and PSD, which resulted in a corresponding unexplored field on SASR. The purpose of this paper is to study the Martian spatial-temporal structure of SASR and the PSD under terrain influences. In this paper, the theory of Earth's SASR, the previous Martian SASR model and the theory of planetary science were combined to propose the SASR model that can be applied to Mars. Then, with the spectrum method theory of geography, we defined two new concepts of spectrums to explore the spatial-temporal distribution of SASR and PSD in different Martian landforms. We found SASR and PSD on Mars were significantly influenced by terrain relief and latitude and showed sufficient regularity, which can be concluded as a gradual attenuation with terrain relief and a regularity of latitude anisotropy. The latitude anisotropy feature is a manifestation of the terrain shielding effect. With the latitude varying, SASR and PSD at different temporal scale generally showed different features with those of Earth, which may be attributed to the imbalanced seasons caused by Martian moving orbits and velocity. Compared to PSD, SASR showed more regular variation under terrain relief and was more influenced by the terrain relief which revealed that SASR is more sensitive to terrain relief than PSD. Additionally, the critical area is a quantitative index to reflect the stable spatial structure of SASR and PSD in different landforms and may be viewed as the minimum test region of sample areas. The corresponding result of the experiments herein indicated that either spectrum can effectively depict the spatial-temporal distribution of SASR and PSD on Mars under terrain relief and deepen the understanding of the variation of SASR and PSD influences by terrain. The critical area of either spectrum can be employed to explore and determine the stable spatial structure of SASR and PSD in different landforms. The proposed Martian SASR model and the new spectral method theory shed new light on revealing the spatial-temporal structure of SASR and PSD under terrain influences on Mars.
... To circumvent this deficiency in the presence of long-range dependence, long memory autoregressive fractionally integrated moving average (ARFIMA) stochastic models (see e.g. [7]) have been used to a lesser extent. Although the afore-mentioned two classes could successfully account for stochastic variation, it is well-known that the global horizontal irradiance (GHI) data series recorded at the earth's surface is a mixture of deterministic (extra-terrestrially) and stochastic components due to the unpredictability of the prevailing atmospheric conditions. ...
Article
Full-text available
The global horizontal irradiance data recorded at the earth's horizontal surface is a mixture of deterministic (extra-terrestrially) and stochastic components due to the ever-changing atmospheric conditions. The Box-Jenkins short memory stochastic models and their hybrid versions have been used successfully to forecast solar irradiance data. However, these models lack robustness and faulter for distant horizon forecasting such as more than 2 hours in the hourly case. Using a quantile regression model as both a robust benchmark and robustification model, we remedy this drawback by using long memory models and their hybrid versions in 24 hour ahead irradiance forecasting at three sites in South Africa (RVD, UPR and SUN). Results show that the inherent long memory is anti-persistent in all models but one that shows persistence. The forecasts from these long memory-based models are generally within the prediction interval of the benchmark model forecasts for two sites. Based on the RMSE, rRMSE, MAE, rMAE and the pinball loss function at the central quantile level and two extreme quantile levels, forecasts obtained through robustification via forecasts combinations with the quantile regression yield the best performance. The predictive ability of the best model before forecasts combination against the quantile regression model combined (robustified) model based Diebold-Mariano test rejected the null hypothesis of the same accuracy. As a follow up to distinguish which has a superior predictive accuracy, "Murphy diagrams" were used. For the RVD and UPR sites, the model based on forecast combination has the best predictive accuracy while the quantile regression based benchmark model has the best accuracy for the SUN data.
... Without considering atmospheric attenuation, the SD refers to the possible sunshine duration (PSD) while considering shielding effect and the PSD without consideration of the shielding effect (Z. Q. Li and Weng 1988) (note that, in this paper, PSD refers to the former definition). The first PSD definition is of great importance because it is necessary for the computation of many meteorological factors, e.g., solar energy (Tsekouras and Koutsoyiannis 2014), astronomical solar radiation (X F Qiu et al. 2004;Mohammadi et al. 2015) and global solar radiation (Mohammadi et al. 2015). Researchers have studied the spatial structure of the monthly PSD in Pakistan (Ambreen et al. 2015), and it has been found that the terrain shielding effect may affect the spatial structure of PSD in mountains (Zeng et al. 2003). ...
Article
Full-text available
Possible sunshine duration (PSD) determines the solar energy received by the surface of Earth. The objective of this research is to explore the spatial structure of PSD which is closely related to slope and PSD. We chose 40 sample areas representing diversified landforms in China and then computed the PSD by using the digital elevation models (DEMs) with a resolution of 30 m for the sample areas. We define the concept of the spectrum which reveals the relation between the PSD and slope over an analysis window on the DEM, and propose the method for deriving the spectrum. Using this method, we derive the corresponding spectra in a year or in a season in the study areas when the analysis window is larger than the corresponding value of area, the stable area. The value of the spectrum generally decreases with the slope; the stable area is positively related to the variation in the slope and that of the relief degree. The spectrum may be useful for computing PSD while considering the shielding effect quickly; the stable area may be used as the minimum value of the area in studies on the spatial structure of PSD; this method might be a quantitative one in geography for revealing the spatial structure of other meteorological factors.
Article
This article aims to analyze the daily fluctuations of the time series of wind speed in some municipalities in the State of Bahia, Brazil from January 2009 to December 2018 with the approach of sliding windows. The analysis will be performed, mainly, with the method known in the literature as Detrended Fluctuation Analysis (DFA) able to identify and measure autocorrelation in non-stationary time series on different time scales (Peng et al., 1994). To meet the proposed objective, we chose five cities of Bahia, by methodological option: Barreiras, Feira de Santana, Guanambi, Salvador and Vitória da Conquista. The results indicated a persistent and statistically significant behavior at the level of 5% for all the studied cities and period. The description with sliding windows (w=365) found a predominance of relative variation above 15%, kurtosis and positive asymmetry. Our findings can be used as one more proposal to model wind speed data and contribute to research related to climatological data.
Article
Persistence is an important characteristic of many complex systems in nature, related to how long the system remains at a certain state before changing to a different one. The study of complex systems’ persistence involves different definitions and uses different techniques, depending on whether short-term or long-term persistence is considered. In this paper we discuss the most important definitions, concepts, methods, literature and latest results on persistence in complex systems. Firstly, the most used definitions of persistence in short-term and long-term cases are presented. The most relevant methods to characterize persistence are then discussed in both cases. A complete literature review is also carried out. We also present and discuss some relevant results on persistence, and give empirical evidence of performance in different detailed case studies, for both short-term and long-term persistence. A perspective on the future of persistence concludes the work.
Article
Among the several technologies for solar energy recovery, parabolic solar collectors have emerged as one of the most promising due to their performance, which can be enhanced if nanofluids are employed as heat transfer fluids instead of the traditional alternatives. The inherent time-dependent behavior of solar radiation profiles forces the solar thermal plants to be operated aided with controllers able to reject these strong disturbances. While traditional controllers can be employed for this aim, more advanced techniques such as Model Predictive Control are suggested since this optimal-control based method can be tuned to minimize operating costs, among some other features. The main objective of this work is to implement an MPC controller to a nanofluid-based solar thermal power plant in order to evaluate its performance to reject disturbances on the solar radiation profile in an efficient manner. An off-line nonlinear programming optimization was deployed so we could compare the response of the on-line MPC implementation on a strict enough basis. Furthermore, the performance of MPC controllers is affected by how well does the modeling of the system is able to stick to reality, thus, it is important to test if the controller is robust enough to deal with uncertainty that might be introduced as modeling errors. Results indicate that MPC controllers are suitable for their implementation on these kinds of power plants since they steer the system to achieve desired conditions by smoothly manipulating the decision variable, even in the scenarios where a substantial cascade-effect modeling error was imposed in the parameters of the nanofluid.
Article
Full-text available
Wind energy is renewable and environment friendly. It is an alternative clear energy source compared to the fossil fuels that pollute the lower layer of atmosphere. The most important parameter of the wind energy is the wind speed. Statistical methods are useful for estimating wind speed because it is a random phenomena. For this reason, wind speed probabilities can be estimated by using probability distributions. An accurate determination of probability distribution for wind speed values is very important in evaluating wind speed energy potential of a region. In this study, first, we tried to determine appropriate theoretical pdf (probability density function) by comparing 10 pdf for the wind speed data measured for Gelibolu region. In determining proper pdf , an approach consisting of 3 goodness of fit tests and fitted graphics have been used.
Article
Full-text available
A methodology for spatial-temporal disaggregation of rainfall is proposed. The methodology involves the combination of several univariate and multivariate rainfall models operating at different timescales, in a disaggregation framework that can appropriately modify outputs of finer timescale models so as to become consistent with given coarser timescale series. Potential hydrologic applications include enhancement of historical data series and generation of simulated data series. Specifically, the methodology can be applied to derive spatially consistent hourly rainfall series in rain gages where only daily data are available. In addition, in a simulation framework the methodology provides a way to take simulations of multivariate daily rainfall (incorporating spatial and temporal nonstationarity) and generate multivariate fields at fine temporal resolution. The methodology is tested via a case study dealing with the disaggregation of daily historical data of five rain gages into hourly series. Comparisons show that the methodology results in good preservation of important properties of the hourly rainfall process such as marginal moments, temporal and spatial correlations, and proportions and lengths of dry intervals as well as a good reproduction of the actual hyetographs.
Conference Paper
Full-text available
The drawbacks of conventional energy sources including their negative environmental impacts emphasize the need to integrate renewable energy sources into energy balance. However, the renewable sources strongly depend on time varying and uncertain hydrometeorological processes, including wind speed, sunshine duration and solar radiation. To study the design and management of hybrid energy systems we investigate the stochastic properties of these natural processes, including possible long-term persistence. We use wind speed and sunshine duration time series retrieved from a European database of daily records and we estimate representative values of the Hurst coefficient for both variables. We conduct simultaneous generation of synthetic time series of wind speed and sunshine duration, on yearly, monthly and daily scale. To this we use the Castalia software system which performs multivariate stochastic simulation. Using these time series as input, we perform stochastic simulation of an autonomous hypothetical hybrid renewable energy system and optimize its performance using genetic algorithms. For the system design we optimize the sizing of the system in order to satisfy the energy demand with high reliability also minimizing the cost. While the simulation scale is the daily, a simple method allows utilizing the subdaily distribution of the produced wind power. Various scenarios are assumed in order to examine the influence of input parameters, such as the Hurst coefficient, and design parameters such as the photovoltaic panel angle.
Conference Paper
Full-text available
We assess the statistical properties of atmospheric pressure time series retrieved from a large database of monthly records. We analyze the short and long term properties of the time series including possible trends, persistence and antipersistence. We also analyze times series of climatic indices which are based on the atmospheric pressure fields, such as the North Atlantic oscillation index and the El Niño-Southern Oscillation index.
Article
Full-text available
According to the traditional notion of randomness and uncertainty, natural phenomena are separated into two mutually exclusive components, random (or stochastic) and deterministic. Within this dichotomous logic, the deterministic part supposedly represents cause-effect relationships and, thus, is physics and science (the "good"), whereas randomness has little relationship with science and no relationship with understanding (the "evil"). We argue that such views should be reconsidered by admitting that uncertainty is an intrinsic property of nature, that causality implies dependence of natural processes in time, thus suggesting predictability, but even the tiniest uncertainty (e.g., in initial conditions) may result in unpredictability after a certain time horizon. On these premises it is possible to shape a consistent stochastic representation of natural processes, in which predictability (suggested by deterministic laws) and unpredictability (randomness) coexist and are not separable or additive components. Deciding which of the two dominates is simply a matter of specifying the time horizon of the prediction. Long horizons of prediction are inevitably associated with high uncertainty, whose quantification relies on understanding the long-term stochastic properties of the processes.
Book
Sections: Introduction and climate; Analysis of hydrologic time series; Probability functions and their use; Linear stochastic models; Special properties and models; Statistical treatment of floods; Probability theory applied to reservoir storage; Stochastic programming methods in systems engineering; Applied decision theory. (92 refs.)
Article
Based on the long-term monthly discharge series of Yichang, Hankou and Datong stations, this paper has explored the discharge series structure of periodicity, trend and abrupt changes as well as its long-term momery with Singular Spectum Analysis, and Detrended Fluctuation Analysis in employment. The results revealed that: 1) there is a significant 15a periodicity in the discharge series of the Yangtze River, and the 15a periodicity has been disturbed since the 1970s; 2) two abrupt and negative changes of Yichang series respectively in 1926 and 1970 have been detected in Hankou series in 1954; two abrupt changes occurred in Datong series respectively in 1955 and 1988, the former negative and the latter positive; and 3) the performance of DFA revealed a long memory in the discharge series of the Yangtze River, and the larger the upslope contributing area is, the stronger the long memory, which exhibits a cumulative effect.
Article
Wind climate in Northwest Europe is subject to long-term persistence (LTP), also called the Hurst phenomenon. Ignorance of LTP causes underestimation of climatic variability. The quantification of multi-year variability is important for the assessment of the uncertainty of future multi-year wind yields. Estimating LTP requires long homogeneous time series. Such series of wind observations are rare, but annual mean geostrophic wind speed (U) can be used instead. This study demonstrates a method to estimate the 10-year aggregated mean U for the near and the far future and its uncertainty in Northwest Europe. Time series of U were derived from daily sea level pressure from the European Climate Assessment Dataset. Minor inhomogeneities cannot be ruled out, but were shown to hardly affect the estimated Hurst exponent (Ĥ). A maximum likelihood method was adjusted to remove the biases in (Ĥ). The geostrophic wind speed over the North Sea, the British Isles and along the Scandinavian coast are characterised by statistically significant H between 0.58 and 0.74, (H = 0.5 implies no LTP). The additional affect of the parameter uncertainty is estimated in a Bayesian way and is highly dependent on the record length. The assessment of structural changes in future wind fields requires general circulation models. An ensemble of seventeen simulations (ESSENCE) with one single climate model (ECHAM5/MPI-OM) was used to evaluate structural trends and LTP. The estimated Ĥ in the ESSENCE simulations are generally close to 0.5 and not significant. Significant trends in U are found over large parts of the investigated domain, but the trends are small compared to the multi-year variability. Large decreasing trends are found in the vicinity of Iceland and increasing trends near the Greenland coast. This is likely related to the sea ice retreat within the ESSENCE simulations and the associated change in surface temperature gradients.
Article
The increase in negative effects of fossil fuels on the environment has forced many countries to use renewable energy sources, especially wind energy. Wind speed is the most important parameter of the wind energy. Probability distributions are useful for estimating wind speed because it is a random phenomenon. This study analyzes wind speed frequencies using wind data from Tabriz synoptic station in Iran. Four different distributions are fitted to the maximum annual wind from station, and parameters of the distributions are estimated using the method of maximum likelihood and the method of moments. Calculations are performed with Mathematica, a computer algebra system developed by Wolfram Research. The advantage of using this software is that the symbolic, numerical, and graphical computations can be combined and that all quantities can be accurately calculated; in particular, there is no need to resort to any approximate methods for the calculation of quantiles. There is a ready-to-use command for calculating quantiles from distributions that are built in Mathematica, while for other distributions they can be easily and accurately calculated by inverting the cumulative distribution functions or by solving nonlinear equations where the inversion is not possible. The best distribution is selected based on the root mean square error (RMSE), the coefficient of determination (R 2), and the probability plot correlation coefficient (PPCC). The results indicate that the best performance can be obtained by the Gamma distribution.
Article
In this study, the authors focus on the tracking performance and the robustness of 12 sliding-windowed Hurst estimators for multifractional processes with linear trend local Hölder exponent, noisy multifractional processes and multifractional processes with infinite second-order statistics. Four types of multifractional processes are synthesised to test the tracking performance and robustness of these 12 sliding-windowed Hurst estimators. They are (i) noise-free multifractional process; (ii) multifractional process corrupted by 30-dB signal-to-noise ratio (SNR) white Gaussian noise; (iii) multifractional process corrupted by 30-dB SNR impulse noise; and (iv) multifractional stable process, which has no finite second-order statistics. Furthermore, the standard error of different sliding-windowed Hurst estimators are calculated in order to quantify the accuracy and robustness. This study provides a guideline and principle in the selection of Hurst estimators for noise-free multifractional process, noise-corrupted multifractional process and multifractional process with infinite second-order statistics. The results of this analysis show that the sliding-windowed Kettani and Gubner̈s method provides the best-tracking performance for multifractional processes with linear trend local Hölder exponent and good robustness to noise.