Arab J Geosci

DOI 10.1007/s12517-012-0652-0

ORIGINAL PAPER

Stochastic analyses of maximum daily rainfall series recorded

at two stations across the Mediterranean Sea
Tefaruk Haktanir & Saleh Bajabaa & Milad Masoud

Received: 30 May 2012 / Accepted: 31 July 2012

# Saudi Society for Geosciences 2012

Abstract Independence, stationarity, homogeneity, trend, Introduction

and periodicity tests are applied on 48-year-long com-
plete and 79-year-long incomplete maximum daily rain- Many studies in the last two decades indicate a change in
fall series recorded at Alexandria, Egypt, and on 61- climate in the twentieth century, which is expected to pro-
year-long maximum daily rainfall series recorded at tract into the twenty-first century, caused mainly by release
Antalya, Turkey, which are located at the southeastern of excessive amounts of greenhouse gasses into the atmo-
and northeastern shores of the Mediterranean Sea. The sphere since 1970s (e.g., Anderson et al. 2010; IPCC 2007,
results indicate no significant trend and no periodicity in 2008, 2011; NZCCO 2008; Sen et al. 2012a, b). Yet, a few
mean, and both series are independent and homoge- studies may be found arguing that various mechanisms of
neous. Linear regression trend test applied to the 10 % the planet will counter the anthropogenic effects somehow
highest part of the Alexandria series indicated a signif- (e.g., Ball 1992; Nandargi and Dhar 2011). According to
icant increasing trend. Next, frequency analysis is ap- many publications, an increase in extreme precipitation is
plied on each of these series by the probability expected worldwide even at regions where annual average
distributions of Gumbel, general extreme-values, three- rainfall has a decreasing trend (e.g., Burn et al. 2011; Collins
parameter log-normal, Pearson-3, log-Pearson-3, log- 2009; Douglas and Fairbank 2011; Fujibe et al. 2005;
logistic, generalized Pareto, and Wakeby. The distribu- Groisman et al. 2005; Guo 2006; IPCC 2007, 2008, 2011;
tions, except for the generalized Pareto and Wakeby, Kundzewicz et al. 2005; NZCCO 2008).
pass the χ2 and Kolmogorov–Smirnov goodness-of-fit The Intergovernmental Panel on Climate Change, known
tests at 90 % probability. By visual inspection of the by the acronym IPCC, is a unit comprised of experts estab-
plots of histograms together with the probability density lished by the United Nations to organize evaluations of
functions, and by the results of the χ2, Kolmogorov– climate change information. The climatic simulation models
Smirnov, and probability plot correlation coefficient such as ECHAM4 of Max Planck Institut für Meteorologie
tests, the general extreme-value distribution whose are executed on high-capacity computers with various emis-
parameters are computed by the method of probability- sions scenarios, and their effects on future meteorological
weighted moments is deemed to be suitable for these events are predicted. A comment in chapter 2 of a technical
two maximum daily rainfall series. paper by IPCC (IPCC 2008) is: “Theoretical and climate
model studies suggest that, in a climate that is warming due
Keywords Maximum daily rainfall . Tests for trend . to increasing greenhouse gases, a greater increase is
Independence . Stationarity . Homogeneity . Frequency expected in extreme precipitation, as compared to the mean.
analysis … It is very likely that heavy precipitation events will
become more frequent.”
In New Zealand, there are laws about the effects of the
climate change on public utilities like storm water drainage
T. Haktanir (*) : S. Bajabaa : M. Masoud
King Abdulaziz University, Water Research Centre,
systems and about the measures to be taken by regional
Jeddah, Saudi Arabia councils and local authorities against those effects
e-mail: [email protected] (NZCCO 2008). So, for New Zealand, the topic of climate
 Arab J Geosci

change is not an issue to be disputed but rather a fact Turkey, in order to (1) look for a possible trend due to the
recognized by law. The effect of climate change on extreme effect of climate change, (2) check for independence, statio-
rainfalls is quantified as increases in magnitudes of various narity, homogeneity, and (3) apply frequency analysis to
return period rainfalls expressed in percentages 50 and predict large return-period maximum daily rainfalls.
100 years later (e.g., Table 5.2 in NZCCO 2008).
In a recent study by Burn et al. (2011), the Mann–Kendall
trend test is applied to annual maximum rainfalls of dura- Materials and methods
tions from 1 to 24 h in the State of British Columbia in
Canada, and generally increasing trends in these series are Maximum daily rainfall series recorded at Alexandria
detected. Fujibe et al. (2005) investigated the data of ex- and Antalya
treme rainfalls of three different durations: 1, 4, and 24 h
recorded at 8, 46, and 61 stations in Japan, respectively, The second largest city in Egypt, Alexandria, known as “The
having continuous record lengths of at least 80 years within Pearl of the Mediterranean,” has an atmosphere that is more
the period of 1898–2003 for determination of trends. In Mediterranean than Middle Eastern; its ambience and cultural
conclusion, an increasing trend in extreme precipitation all heritage distance it from the rest of the country, although it is
over Japan is detected. actually only 225 km from Cairo. Alexandria has an arid
One notices just a few studies, yet, containing results climate (climate classification BWh by Köppen and Geiger
which express stationarity in floods and extreme rainfalls 1936), but the prevailing north winds, blowing across the
(e.g., Kundzewicz et al. 2005; Nandargi and Dhar 2011; Mediterranean, gives the city a different climate from the desert
Villarini et al. 2011). An interesting example is given by hinterland. The city’s climate shows Mediterranean character-
Douglas and Fairbank (2011), who, having applied the istics, namely mild, variably rainy winters and hot summers
Mann–Kendall and the linear regression trend tests on an- that, at times, can be very humid; January and February are the
nual maximum daily rainfall series recorded at 48 locations coolest months, with daily maximum temperatures typically
in northeastern America with an average record length of ranging from 12 °C to 18 °C and minimum temperatures that
80 years, say: “The trend analysis over the time period could reach 5 °C. Alexandria experiences violent storms, rain,
1954–2005 indicated that annual maximum daily rainfall and sometimes hail during the cooler months. July and August
was amazingly stationary; …” They detected trend for the are the hottest and driest months of the year, with an average
period 1954–2008 only when they included the last 3 years daily maximum temperature of 30 °C.
of the recorded data between 2005 and 2008 (Douglas and The climate of Antalya (Turkey) is typically Mediterranean,
Fairbank 2011). Similarly, the last sentence of the paper by and the rainy season is from November to March. Air temper-
Nandargi and Dhar (2011) is: “Hence, it is somewhat baf- atures are in the interval of 30 °C–35 °C during the period June
fling as to whether climate change has any impact on ex- through September. December, January, and February are the
treme rainfall events in the entire Himalayan region, coolest months, with daytime temperatures averaging
especially, in the recent years of the period 2001–2007.” about 15 °C and night-time temperatures always well
The annual maximum daily rainfall is the maximum of so above freezing (http://www.dmi.gov.tr).
many daily rainfalls occurring within a hydrologic year. So, Alexandria is located on the southeastern shore of the
in a recorded series of maximum daily rainfalls, there are as Mediterranean Sea at coordinates of 31.2°N and 29.92°E.
many elements as the length of record in years. On the Antalya is a city located at the northeastern shore of the
geographical region enclosing the Mediterranean Sea, it is Mediterranean Sea at coordinates of 36.9°N and 30.68°E.
highly unlikely for the atmosphere to produce incessant The birds-eye distance between them is 633 km, and their
rainstorms longer than 24 h. Therefore, the rainfall records longitudinal positions are not too far apart as both are close to
in those countries flanking the Mediterranean Sea are taken the 30°E meridian. Antalya is situated at the northern tip of the
for a maximum period of 1 day. The maximum daily rainfall Antalya Gulf. The Taurus Mountain Range extends all along
is the major cause of intense floods in moderate-size and from the southwestern end of the Anatolian Peninsula to its
large watersheds, and therefore they deserve special atten- southeastern end, paralleling the shoreline from a distance of
tion. For example, for determination of flood inundation about 20 to 30 km. There are no similar mountain ranges
maps in wadis in arid regions as suggested by Sen et al. behind Alexandria, and she is on a fairly plain terrain. The
(2012a), the critical flood peak must be calculated which is map of this part of the Mediterranean Region is given in
dependent on plausible estimation of maximum daily rain- Fig. 1. Antalya is subjected to the orographic effects of the
fall. The objective of this study is to carry out a stochastic sea-ward slopes of 3,000-m-high Taurus Mountains surround-
analysis on series of annual maximum daily rainfalls ing her from all sides except for the sea direction.
recorded at two important cities on the shores of the The rainfall gauging at Alexandria started as early as the
Mediterranean Sea, Alexandria in Egypt and Antalya in 1900s. There are two 8-year discontinuities in the data, and
Arab J Geosci

1950–2010. Table 1 gives the maximum daily rainfalls data

recorded at these two cities.

Tests for trend

The plot of a series of a hydrologic variable versus time over

the total length of the observation period is useful for visual
investigation. A possible trend, a sudden change, and even a
periodicity may reveal themselves by a careful visual obser-
vation of the plotted values. So, the figure of the magnitudes
of a hydrological event versus time usually provides a
qualitative evaluation. About the merit of plots in statistical
analysis in general, Helsel and Hirsch (2002, ch. 2) say:
“Graphs provide visual summaries of data which more
quickly and completely describe essential information than
do tables of numbers.”
The Mann–Kendall, linear regression, and Spearman’s
rho tests are commonly applied for detecting trend in hy-
drologic series (e.g., Burn et al. 2011; Collins 2009;
Douglas and Fairbank 2011; Ehsanzade et al. 2010; Fujibe
Fig. 1 Map showing the eastern part of the Mediterranean Sea and et al. 2005; Groisman et al. 2005; Helsel and Hirsch 2002,
locations of Alexandria (Egypt) and Antalya (Turkey) ch. 12; Hipel and McLeod 1994, ch. 23; Onoz and Bayazit
2003; Rao et al. 2011; Yue et al. 2002). In this study, the first
the periods of the available records are: 1900–1947, 1956– two of these are applied, being influenced by Yue et al.
1979, and 1988–1994, each one being continuous in itself. (2002) who showed that the Spearman’s rho test gives
All tests are applied on the first 48-year-long segment, but results almost the same as the Mann–Kendall test.
the linear regression test is applied to the series from the
beginning till the end, taking into account the chronological Mann–Kendall trend test
positions of the missing years. And, the frequency analysis
also is applied to the 79-element total series of Alexandria The Mann–Kendall (M-K) statistic of a recorded series, which
data. The Antalya data are complete over the period of is ordered with respect to years, as originally proposed by

Table 1 Values of the maximum

daily rainfalls (in mm) recorded Maximum daily rainfalls (in mm) recorded at Alexandria, Egypt
at Alexandria (Egypt) and Ant- Beginning year: 1900; number of records, 79
alya (turkey) 28.0 19.0 28.0 22.0 21.0 47.0 28.0 17.0 32.0 30.0 33.0 27.0
13.0 40.0 55.0 10.0 47.0 24.0 18.0 46.0 13.0 18.0 32.0 19.0
27.0 20.0 19.3 53.0 18.0 25.2 35.5 7.7 40.0 32.9 34.5 23.0
18.0 18.0 62.0 18.0 34.6 18.7 12.1 17.6 19.5 23.0 21.8 26.0
– – – – – – – – 14.6 28.7 20.0 27.8
33.0 24.0 22.3 13.0 28.0 11.8 65.0 15.7 17.0 44.0 53.0 29.7
35.0 28.5 54.3 30.0 15.0 28.0 21.0 10.0 – – – –
– – – – 29.0 38.0 14.0 27.0 97.0 29.0 24.0

Maximum daily rainfalls (in mm) recorded at Antalya, Turkey

Beginning year: 1950; number of records, 61
99.6 212.9 232.4 139.5 233.2 66.0 163.9 69.0 134.5 145.9 186.6 129.0
122.6 146.0 71.5 151.9 210.1 245.7 155.0 334.5 185.6 94.3 163.2 120.9
283.5 160.7 126.0 170.2 109.7 199.2 152.4 178.3 89.1 105.8 236.1 185.1
120.9 63.7 108.1 131.0 125.7 277.9 155.9 125.3 167.9 428.6 194.0 205.8
184.0 124.7 216.5 253.2 261.9 238.2 220.3 153.3 248.6 117.2 52.4 199.0
128.8
 Arab J Geosci

Kendall (1975), is computed by (e.g., WMO 2009, ch.5): exist. The regression line fitted to the variable suspected
of having a trend over years is:
n
1 X
X n  
SM
k ¼ sign xi
xj ð1Þ x ¼ A þ B  ðtimeÞ ð4Þ
i
1 j¼iþ1
Where x is the hydrological variable in concern, maxi-
where sign(xi −xj) assumes a value of +1 if xi >xj and −1 mum daily rainfall in this study, time is time in years,
if xi <xj, and n is the total number of elements of the which takes on integer values from 1 to n for a com-
series. The variance of SM−K is given as: plete series, B is the estimate of the population slope
VarðSM
k Þ ¼ ½n  ðn
1Þ  ð2n þ 5Þ=18 ð2Þ term β calculated out of the available recorded series,
and A is the estimate of the population intercept α. The
The expected value of SM−K is zero for an identically essence of this test is to check the hypothesis: if β,
distributed random variable without trend (Kendall which is the Expected value of B (β0E[B]), is equal to
1975). Therefore, the standardized variable given below zero? The intercept being zero or not does not affect the
obeys a standard normal distribution for sufficiently existence of trend. Assuming no trend, which means
large series. assuming β 00, the reduced variate in the two-sided
. confidence interval becomes (e.g., Ross 2004, ch. 9):
Z ¼ ðSM
K
0Þ ½VarðSM
k Þ0:5 ð3Þ
( " #)0:5 ,( )0:5
X
n
2
X
n
2
If the magnitude of the Z statistic computed by Eq. 3 t¼ ð n
2Þ  ði
aaiÞ B ½xi
ðA þ B  iÞ
i¼1 i¼1
remains within the interval: −Zcr<Z<Zcr, where Zcr is the
ð5Þ
critical value in the standard normal table for a tail
probability equaling half of the probability level of Where n is the total number of elements in the recorded
rejection initially chosen by the analyst, then the hy- series (series length in years), i is the ith year of record,
pothesis that Expectation[SM−K]00 is true, which means being 1 for the first year regardless of its calendar
there is no trend. The reason for halving the tail prob- value, which is the independent variable in the trend
ability is because a two-sided confidence interval is test by regression, and aai is the arithmetic average of
evaluated for the Z statistic. Another way of deciding integer numbers from 1 to n. If the t value computed by
on the existence of trend by the Mann–Kendall test is to Eq. 5 turns out to be smaller than the critical t value
compute the probability of exceedence (Pex) of the Z taken from the Student’s t distribution for half of a
statistic given by Eq. 3 and check if this Pex is greater critical level of probability of exceedence (Pexcr) and
or smaller than half of the critical level of probability degree of freedom of n–2, then the hypothesis that β00
(Pexcr). If Pex is smaller than Pexcr/2, this will indi- is correct, which means there is no trend. Conversely, if
cate the existence of a trend. For an optimistic view, Pex of the absolute value of the magnitude of t com-
meaning, an initial tendency for rejecting trend, the puted by Eq. 5 given by the Student’s t distribution for
critical probability is taken as a small value like 1 %. a degree of freedom of n–2 is greater than Pexcr/2, this
Conversely, for an analysis tending to accept the exis- means there is no trend. Both approaches lead to the
tence of trend, a fairly large probability, like 10 %, is same conclusion.
assumed. It is noticed that, usually, a critical probability An unpublished study done by the authors in which
level of 5 % is adopted by relevant studies. all these five tests were applied on 175 series of daily
maximum rainfalls recorded at rain-gauging stations all
Linear regression trend test over Turkey having record lengths in the interval: 39≤
n≤73, revealed that both the M-K and linear regression
This test is based on the idea that, if there is either an (LR) tests agreed on 145 of these 175 series. The
increasing or a decreasing significant trend in the mag- Mann–Kendall test and the other three tests which are
nitude of the investigated hydrologic variable, then its summarized below necessarily must be applied on chro-
mean should be either increasing or decreasing with nologically complete series of recorded hydrologic data.
respect to years, and the slope term of the simple linear The LR test however can be applied to any series
regression fitted to the chronological values of the having discontinuities in records, provided that the years
recorded series will be statistically significant. In other of the missing elements are known. This is because the
words, the expected value of the slope coefficient independent variable in the LR model is the years,
should be different from zero, which means the two- which are integer numbers from 1 up to the last year
sided confidence interval of the slope coefficient must minus the first year of record, skipping the integer
not comprise zero, in order for a significant trend to numbers of the missing years. So, this peculiarity of
Arab J Geosci

,
the LR method is a definite advantage of it over the M- X
n
1
2
Xn
K test for trend detection. Performing the regression Q¼ ðxiþ1
xi Þ ðxi
aaxÞ2 ð6Þ
computations with the actual numerical values of the i¼1 i¼1

years would reveal the same results, but usage of inte-

gers from 1 to (n+number of missing years) instead of Where xi is the ith element of the sample series and aax is
4-digit year numbers may be better for numerical sta- the arithmetic average of the series. Under the assumption
bility in intermediate computations. that the xi’s are independent, the expected value of the Q
statistic is equal to 2, and Q obeys a special distribution,
Tests for independence, stationarity, and homogeneity whose critical values are given by Bierkens (2006, ch. 4) for
commonly used tail probabilities. Alternatively, for large
In relevant literature, one can find many statistical tests sample series, Q is approximately normally distributed with
to check for a time series to be independent and iden- mean 2 and variance as given below.
tically distributed, stationary, and homogeneous (e.g.,  
Helsel and Hirsch 2002; Kite 1977; Rao and Hamed σ2Q ¼ 4  ðn
2Þ n2
1 ð7Þ
2000; WMO 2009). Practically, stationarity means inde- Therefore, for fairly large series, comparing the magni-
pendence from time. A hydrologic variable must be tude of the Q statistic computed by Eq. 6 with a critical
independent, stationary, and homogeneous in order to value from its table yields the same result as comparing the
establish a relationship between its magnitude and its Z statistic below with the critical value from the standard
probability of exceedence (or non-exceedence) by ap- normal table versus a tail probability equaling Pexcr/2.
plying a probability distribution to it, which is common- .n . 0:5 o
ly known as frequency analysis. Otherwise, the serial Z ¼ ð Q
2Þ 2  ½ ð n
2Þ n2
1 ð8Þ
effects need to be separated, the frequency analysis be
carried out on the randomized series, and the temporal
components be quantitatively superposed. Presence of Wald–Wolfowitz test for stationarity
trend causes deviation from this basic assumption.
Significant serial correlation is another effect distorting In order to test if the elements of a recorded series are realiza-
the temporal independence. All the measured magni- tions of independent and stationary occurrences, as originally
tudes of a random variable, which are elements of a proposed by Wald and Wolfowitz in 1943, the following
sample series, must occur as governed by the same statistic, known as Wald–Wolfowitz’s R statistic is suggested
population probability distribution in order for them to (e.g., Rao and Hamed 2000, ch. 1; WMO 2009, ch. 5).
be homogeneous. For example, all of the measured
flood peaks must ideally be either extreme-rainfall- X
n
1
induced floods or snow-melt runoffs, and not combina- R¼ xi  xiþ1 þ x1  xn ð9Þ
tions or even superpositions. Possible effect of 11-year i¼1
sunspot cycles on hydrologic variables may be another
R is normally distributed with mean and variance given below.
cause jeopardizing true randomness. But, such cyclic
 
effects are not detected on extreme quantities like an- μR ¼ S12
S2 =ðn
1Þ ð10Þ
nual instantaneous flood peaks and maximum daily rain-
falls, usually. If for some reason there is a periodic
 
variation in the mean of a hydrologic variable, this σ2R ¼ S22
S4 =ðn
1Þ
μ2R
deterministic component must be taken into account
 
along with the random rule of the governing probability þ S14
4 S12 S2 þ 4 S1 S3 þ S22
2 S4 =ðn
1Þ=ðn
2Þ
distribution. ð11Þ
In this study, next to the two trend tests, commonly used
tests for independence, stationarity, and homogeneity are also Where the terms Sj’s are defined as Sj 0n·MOj, in which MOj
applied, and brief information about them is given below. is the jth moment with respect to the origin of the sample
series, and μR and σR2 are the mean and variance of R.
Von-Neumann’s test for independence Hence, the standardized variate is computed by
Z ¼ ð R
μ R Þ =σ R ð12Þ
In order to test if the elements of a recorded series are
realizations of independent occurrences, von Neumann in If the magnitude of the Z statistic is smaller than the
1941 suggested the following statistic, known as von- critical value (Zcr) given by the standard normal distribution
Neumann’s Q statistic (e.g., Bierkens 2006): for an exceedence probability equaling Pexcr/2 (Z<Zcr),
 Arab J Geosci

then the null hypothesis is true and the series is assumed to probability of the absolute value of Z (Pex) is greater than
be stationary. Or, if the exceedence probability of the abso- Pexcr/2 (Pex>Pexcr/2), then the series is homogeneous.
lute value of Z (Pex) is greater than half of the test exceed- Both ways give the same result.
ence probability (Pex>Pexcr/2), then the series is stationary. The Mann–Whitney test is also applied to two sample
Both ways give the same result. series to check whether they can be assumed to come from a
common population probability distribution. Assuming the
Mann–Whitney test for homogeneity lengths of the two series are p and q, the one having the
shorter length becomes the first series, and the combined
In order to test if the elements of a recorded series are series having p+q(0n) elements is then ranked in ascending
realizations obeying the same population probability order. In this case, R equals the summation of the rank
distribution, as originally proposed by Mann and numbers of the first series in the combined series. The rest
Whitney in 1947, the following statistic, known as of the computations and analyses are the same as above. The
Mann–Whitney’s U statistic is suggested (e.g., Rao and result in this case is the decision that the two sample series
Hamed 2000, ch. 1; Helsel and Hirsch 2002, ch. 12; can be assumed to have a common probability distribution if
WMO 2009, ch. 5). First, the sample series is split into the null hypothesis is accepted.
two pieces, such that p≤q, p and q denoting the lengths
of the first and the second parts, and next the original Spectral density plot for sun-spot effect periodicity
series is ranked in ascending order. Obviously, p+q0n.
Secondly, the summation of the rank numbers of the Assuming an annual hydrologic variable, like annual flows
first half of the series is computed, which is denoted by at a section of a natural stream, instantaneous maximum
R. Thirdly, V and W statistics are computed as follows. flow rate in a hydrologic year, the greatest one of so many
tr-minute-duration rainfalls occurring within a hydrologic
V ¼ R
½pðp þ 1Þ=2 ð13Þ
year period at a geographical location, etc., exhibits a deter-
ministic periodic behavior in mean being affected by a few
harmonics, the Fourier series depicting the periodic cycling
W ¼pq
V ð14Þ
of the mean can be written as (e.g., Kottegoda 1980, ch. 2;
V gives the number of times an element in the first Yevjevich 1972, ch. 3)
segment follows an element of the second segment as both PX
=2
1 PX
=2
1
are in their original order. The magnitude of the Mann– μt ¼ μ þ ai sinð2p i t =PÞ þ bi cosð2p i t =PÞ ð18Þ
Whitney’s U statistic is the smaller one of V and W. For a i¼1 i¼1
series having more than 20 elements, U is approximately
Where i’s are harmonics, t is time in years, αi and βi are
normally distributed with mean and variance given below.
Fourier coefficients, and P must be even.
μ U ¼ p q=2 ð15Þ The quantity known as the spectral density is defined as
(e.g., Kottegoda 1980, ch. 2; Yevjevich 1972, ch. 3):
" #
, ( ) X
M
   k 
X  Sk ¼ 1 þ 2 rj cosðp k j=M Þ =M ; k ¼ 1; 2; . . . ; M ð19Þ
σ2U ¼ p q ½ n ðn
1 Þ n3
n =12
Tj3
Tj =12 j¼1
j¼1

ð16Þ Where M is a truncation point corresponding to the

limiting frequency, and rj is the j-lagged sample serial
Here, Tj is the number of elements tied at the jth rank correlation coefficient, which is computed by the classic
whose magnitudes are equal to each other, and k is the equation below.
number of tied groups in the total ranked series. If there
are m equal elements at any rank (m ties), then the arithmetic  . n   0:5 o
rj ¼ Covariance xi ; xi
j Varianceðxi ÞVariance xi
j ð20Þ
average of m integer rankings is ascribed to each one as its
rank number. Hence, the standard variate is computed by
Next, Sk is smoothed by the below equation, which is
Z ¼ ðU
μU Þ=σU ð17Þ known as the Hanned spectral density, after J. von Hann,
If the magnitude of the Z statistic computed by Eq. 17 is who suggested it originally.
smaller than the critical value (Zcr) given by the standard Sk ¼ Sk
1 =4 þ Sk =2 þ Skþ1 =4 ð21Þ
normal distribution for an exceedence probability equaling
Pexcr/2 (Z<Zcr), then the null hypothesis is true and the The plot of S*k versus frequency ðf ¼ 1=ð2M =k Þ; k ¼ 1;
series is assumed to be homogeneous. Or, if the exceedence 2; . . . ; M :Þ reveals sharp jumps for frequency values at which
Arab J Geosci

the series investigated has significant periodicity. For serially Pexcr01 %. This is true with the Alexandria series also,
independent series, however, the S*k −f plot exhibits a shape of which passes the Wald–Wolfowitz test at 1 %, while failing
irregularly zigzagging ups and downs in a fairly narrow band it at 5 %. According to Table 3, there is no significant trend
(various examples are given in Kottegoda 1980, ch. 2 and in mean of the Alexandria maximum daily rainfalls even for
Yevjevich 1972, ch. 3). If the peaking values of an S*k −f plot Pexcr010 %.
protrudes above the upper bound of the p% confidence interval Figures 2 and 3 show the plots of the maximum daily
(C.I.ub), then this means at that frequency and at the rainfalls versus years recorded at Alexandria and Antalya.
corresponding period there is significant periodicity of the in- Visual inspection of Figs. 2 and 3 when they had the plots of
vestigated random variable. Usually, p090 %, or 95 %, or 99 %, the measured elements of the recorded series versus years
depending on the choice of the analyst. It has been shown that only might suggest slightly increasing trend. The quantita-
the quantity ν·S*k is a random variable obeying a Chi2 (χ2) tive results of both the M-K and the LR tests, however,
distribution with a degree of freedom of ν0(2.67)·n/M (e.g., indicated no significant trend in both, even for Pexcr0
Kottegoda 1980, Ch. 2). Hence, by the classic confidence 10 %. In short, both the 79-element maximum daily rainfalls
interval expression, the magnitude of the upper bound of the series of Alexandria and 61-element series of Antalya ex-
p% confidence interval is estimated by the below equation (e.g., hibit no significant trend in mean, but the lines of these
Kottegoda 1980, ch. 2; Yevjevich 1972, ch. 3). insignificant trends in means are still shown in Figs. 2 and 3.
Being influenced by (1) the comments in various relevant
C:I:ub ¼ c cr2 =v ð22Þ
publications referred to in the “Introduction” section about
Where, χcr is the quantile of the random variable obey-
2 an increasing trend in extreme precipitation being expected
ing the χ2 distribution with a degree of freedom of ν having worldwide and by (2) the appearances of the plots of the
an exceedence probability of (1−p)/2 % for a p% confi- measured points in Figs. 2 and 3, another linear regression
dence interval. test is applied to the extremes of the recorded elements.
With this purpose, those elements of a recorded series
which are greater than or equal to the quantile having
approximately a 10-year return period (T) are chosen.
Results and discussion Assuming the recorded series is fairly long, the classical
California plotting position formula, which is T ≈ n/j,
Results of the trend, independence, homogeneity, where j is the rank number of the jth element in the
and stationarity tests and discussion series ordered in descending sequence and n is the total
number of elements of the series (sample length), is
All the tests considered in this study necessitate chronolog- used to separate those elements of the series having
ically complete recorded series, the LR trend test being an exceedence probabilities equal to or smaller than 10 %
exception to this rule. Therefore, the LR test for the (Pex ≤0.10). In other words, the element whose rank
Alexandria series is applied in two different manners in this number is closest to (0.1)⋅n is determined, and those
study. First, the longest complete part, which is the first 48- elements greater than or equal to it are separated as
year segment, is used together with the Antalya series for all extreme events of the recorded series. Table 4 gives
the five tests of trend, independence, homogeneity, and the results of the linear regression applied on six greater
stationarity, whose results are given in Table 2 with a critical elements of the Antalya series and the results of that
probability level of 5 % (Pexcr05 %). And next, the LR test applied on nine greater elements of the Alexandria
was applied also on the 79-year-long series of Alexandria series, according to which the regression to the nine
having two 8-year gaps between 1948 through 1955, and extreme elements of the 79-element Alexandria series
1980 through 1987 (inclusive), whose results are given in indicates a highly significant increasing trend. This re-
Table 3. As seen in Table 2, the Antalya series is trend-free, sult concurs with the general comments about an in-
independent, stationary, and homogeneous, and as can be creasing trend in extreme precipitation expected worldwide
seen by the numerical results, these are true even for Pexcr0 for Alexandria (Egypt). A similar result does not exist for
10 % also. While the 48-year-long complete part of the Antalya, however, and the linear regression applied to the
Alexandria series passes four of these five tests, it is not six greater elements of the 61-element Antalya series indicates
stationary according to the Wald–Wolfowitz test. Most rel- no significant trend. The city of Antalya being located at the
evant studies use 5 % probability for these tests. Choosing edge of Antalya Bay (Fig. 1) surrounded by high mountains
1 % probability would increase the chance of decision for from three sides may have reduced the impact of global
no-trend, independence, stationarity, and homogeneity, and climate change on Antalya maximum daily rainfall patterns
naturally, the number of series investigated in a geographi- and may have caused the rainfalls to be heavily affected by
cal region passing the tests increases with a choice of local, mainly orographic, conditions.
 Arab J Geosci

Table 2 Results of the Mann–Kendall and linear regression trend tests, von-Neumann independence, Wald–Wolfowitz stationarity, and Mann–
Whitney homogeneity tests applied on the maximum daily rainfalls (in millimeters) series recorded at Alexandria (Egypt) and at Antalya (Turkey)

Maximum daily rainfalls (mm) recorded at Alexandria, Egypt

The number of elements in this series048
Result of the Mann–Kendall trend test:
The M-K statistic and its standard deviation0101, 112.51
The standardized variate00.898 and its tail probability018.5 %
There is no trend because 18.5>2.5
Result of the linear regression trend test:
The coefficients of the linear regression: X0A+B*time are A029.24, B0−0.094 correlation coefficient0−0.109
The t statistic of B0−0.742 and its tail probability023.1 %
There is no trend because 23.1>2.5
Result of the von-Neumann independence test:
For a probability of 5 % the series is independent because von-Neumann’s Q statistic is>the critical value 2.615>1.5351
Result of the Wald–Wolfowitz stationarity test:
The W-W statistic, its mean and SD are00.32658E+05, 0.34650E+05, 0.97391E+03
The standardized variate0−2.045 and its tail probability02.0 %
The series is not stationary because 2.0<2.5
Result of the Mann–Whitney homogeneity test:
The M-W statistic, its mean, and SD are0259.0, 288.00, 48.44
The standardized variate 0 −0.599 and its tail probability 0 27.5 %
The series is homogeneous because 27.5>2.5

Maximum daily rainfalls (mm) recorded at Antalya, Turkey

The number of elements in this series061
Result of the Mann–Kendall trend test:
The M-K statistic and its standard deviation0−207, 160.70
The standardized variate01.288 and its tail probability 0 9.9 %
There is no trend because 9.9>2.5
Result of the linear regression trend test:
The coefficients of the linear regression: X0A+B*time are A0149.53 B00.630 correlation coefficient00.162
The t statistic of B01.258 and its tail probability010.7 %
There is no trend because 10.7>2.5
Result of the von-Neumann independence test:
For a probability of 5 % the series is independent because von-Neumann’s Q statistic is>the critical value, 1.803>1.5834
Result of the Wald–Wolfowitz stationarity test:
The W-W statistic, its mean, and SD are00.17714E+07, 0.17387E+07, 0.35536E+05
The standardized variate00.920 and its tail probability017.9 %
The series is stationary because 17.9>2.5
Result of the Mann–Whitney homogeneity test
The M-W statistic, its mean, and SD are0423.5, 465.00, 69.32
The standardized variate0−0.599 and its tail probability027.5 %

Table 3 Result of the linear regression trend test applied on all 79-element maximum daily rainfall series of Alexandria

The number of elements in this series079

Result of the linear regression test:
The coefficients of the linear regression X0A+B*time are A025.639, B00.0561; correlation coefficient00.106
The t statistic of B00.930 and its tail probability017.8 %
There is no trend because 17.8>2.5
Arab J Geosci

Next to the five tests applied individually on these two

series, the Mann–Whitney homogeneity test was also ap-
plied to the combined series of the first part of Alexandria
plus Antalya. Because the length of the first complete por-
tion of the Alexandria data set is shorter than the length of
the Antalya set, in combination, the 48-element Alexandria
series was taken as the first series to which the 61-element
Antalya series was attached as the ensuing second series.
Actually, the 48 elements of the Alexandria series were
divided by its arithmetic average, and the 61 elements of
the Antalya series were divided by its arithmetic aver-
age prior to combining the two series, which is the
regular procedure of this test actually. And by the
symbols used herein, with two series of lengths p048
and q 061, the Mann–Whitney homogeneity test was
applied to this combined series of length n0p+q0109
also. As given in Table 5, according to the Mann–
Whitney homogeneity test, the maximum daily rainfall
series recorded at Alexandria and Antalya, although
their means are appreciably different from each other, can
Fig. 2 Plots of (1) recorded maximum daily rainfall values at Alexan- be assumed to come from a common population probability
dria (Egypt) over the period: 1900–1994, (2) sample mean of 79- distribution.
element series, (3) trend line for the sample mean, which is statistically
insignificant, (4) trend line for the quantiles having exceedence prob- Spectral density plots for possible periodicity for sun-spot
abilities smaller than 10 %, which is statistically significant
effect

The Hanned spectral density (S*k) versus frequency (

f ¼ 1=ð2M =k Þ; k ¼ 1; 2; . . . ; M ) relationship is computed
separately for the 61-element Antalya maximum daily series
and for the 48-element complete part of Alexandria series as
summarized in the pertinent subsection above, with the
purpose of detecting any possible period close to 11 years.
Kottegoda (1980) recommends an integer value between n/
10 and n/5 for M, which are 6 and 12 for Antalya, and 5 and
10 for Alexandria, respectively. Here, M is taken as 11 for
both Alexandria and Antalya. Figures 4 and 5 show the
diagrams of these relationships. Other values for M, up to
22, are also tried and the resultant diagrams are observed,
which exhibit similar appearances. From Figs. 4 and 5, it
is clear that there is no periodicity for a period about
11 years, whose frequency is f≈0.091, for both Antalya
and Alexandria. For Alexandria, however, the spectral
density diagram makes a small perturbation above the upper
bound of 95 % confidence interval at a frequency of (1/2.7).
Theoretically, this would suggest a periodicity with a period
of 2.7 years. Because the extension of the spectral
density line above the upper bound is not too high
and because 2.7 is not a reasonable number like 3 or
4 years, it is believed that this slight perturbation cannot
Fig. 3 Plots of (1) recorded maximum daily rainfall values at Antalya be taken as revelation of a serious periodicity. In a
(Turkey) over the period: 1950–2010, (2) sample mean of 61-element
series, (3) trend line for the sample mean, which is statistically insig-
similar study, Iqbal et al. (2011) also found no period-
nificant, (4) trend line for the quantiles having exceedence probabilities icity related to the 11-year sun-spot cycles on rainfall
smaller than 10 %, which is statistically insignificant series recorded at various cities in Pakistan.
 Arab J Geosci

Table 4 Results of the linear regression trend tests applied on those parts having elements with exceedence probabilities smaller than 10 % of the
maximum daily rainfalls series of Alexandria (Egypt) and of Antalya (Turkey)

Maximum daily rainfalls (mm) recorded at Alexandria, Egypt

The number of elements in this series079
The linear regression test is repeated on the truncated series
The truncation value is the magnitude of the quantile of the series having a probability of exceedence of approximately 0.10, which is047.0 for this
series. Hence, the elements and their dates of the truncated series are 1905, 47.0, 1914, 55.0, 1916, 47.0, 1927, 53.0, 1938, 62.0, 1966, 65.0, 1970,
53.0, 1974, 54.3, 1992, 97.0
The number of elements in the truncated series09
Result of the linear regression test on the truncated series:
The coefficients of the linear regression X0A+B*time are A043.828, B00.3378, correlation coefficient00.691
The t statistic of the slope coefficient04.488
The tail probability of this t statistic00.0 %
There is increasing trend because the slope coefficient is positive and 0.0<2.5

Maximum daily rainfalls (mm) recorded at Antalya, Turkey

The number of elements in this series061
The linear regression test is repeated on the truncated series
The truncation value is the magnitude of the quantile of the series having a probability of exceedence of approximately 0.10, which is0253.2 for
this series. Hence, the elements and their dates of the truncated series are 1969,334.5, 1974, 283.5, 1991, 277.9, 1995, 428.6, 2001, 253.2, 2002,
261.9
The number of elements in the truncated series06
Result of the linear regression test on the truncated series:
The coefficients of the linear regression X0A+B*time are A0337.397, B0−0.7764 correlation coefficient0−0.164
The t statistic of B0−1.148 and its tail probability02.8 %
There is no trend because 12.8>2.5

Plots of histograms of the recorded series together in hydrologic design, estimation of extreme hydrologic events
with the probability density functions of candidate having return periods much greater than 100 years, like
probability distributions 10,000 years, and even longer, is often needed especially for
important structures. For example, in Chapter 5 of a relevant
Although a flood peak or a maximum daily rainfall having an report by the World Meteorological Organization (WMO
average return period (T) of 100 years sounds to be a hydro- 2009), it is stated that: “Design problems generally require
logic event of very extreme nature, actually it is not. For information on very rare hydrological events, namely those
example, the probability of occurrence of a 100-year-return- with return periods much longer than 100 years.” Because the
period flood at least once at a hydraulic structure having an probability of exceedence (Pex) of maximum daily rainfalls at
economical life of 100 years is 63 % and not 1 %. Therefore, a location, for example, equals the inverse of the chosen

Table 5 Result of the Mann–Whitney homogeneity test applied on the maximum daily rainfalls (in millimeters) recorded at Alexandria (Egypt)
and Antalya (Turkey)

Mann–Whitney test for checking if two sample series can be assumed to have a common population probability distribution
The first sample series is: Maximum daily rainfalls (mm) recorded at Alexandria, Egypt
Length, arithmetic average, variation coefficient, and skewness coefficient of the first sample series are p048, AA02.6925E+01, VC00.452,
SC01.023
The second sample series is: Maximum daily rainfalls (mm) recorded at Antalya, Turkey
Length, arithmetic average, variation coefficient, and skewness coefficient of the second sample series are q061, AA01.6906E+02, VC00.409,
SC01.039
Results of the Mann–Whitney homogeneity test:
The M-W statistic, its mean, and SD are01397.0, 1464.00, 163.81
The standardized variate0−0.409 and its tail probability034.1 %
The two series are homogeneous because 34.1>2.5
Arab J Geosci

of the candidate probability distributions gives a visual eval-

uation as to the suitability of the distributions to the measured
series within its recorded range, which is a commonly applied
practice of comparing candidate distributions, and many pack-
age computer programs provide these plots. Nevertheless,
because the right-tail behavior of a probability distribution is
more important on the magnitude of large return period
events, it is commonly argued that the overall matching of a
density function to the observed histogram in the recorded
range cannot reflect the goodness of a probability distribution
in estimating right-tail quantiles. Yet, both such a plot and its
quantitative version, the χ2 goodness-of-fit test, are still com-
monly used in helping to decide on a suitable probability
distribution (e.g., Rao and Hamed 2000; Seckin et al. 2010).
Briefly, the χ2 goodness-of-fit test quantitatively compares the
histogram of the recorded series with the histogram of the
series of the same length given by a candidate probability
distribution within the range of the recorded series. The histo-
grams of the observed series are drawn concurrently with
scale-adjusted density functions of the probability distribu-
tions considered. Because a distribution with different magni-
Fig. 4 Plot of Hanned spectral density for the 48-element complete
tudes for its parameters is actually another distribution, there
part of the maximum daily rainfall values recorded at Alexandria
(Egypt) over the period: 1900–1947 are around 30 different distributions in this study. Drawing all
of them on the same figure will create congestion and there-
fore, these plots are drawn with a few of them in one graph.
average return period (Pex01/T) and the right-tail part, even Figure 6 shows the histogram of the 79-element maximum
the far right-tail part, of its density function of the chosen daily rainfall series recorded at Alexandria, and Fig. 7 shows
probability distribution has the governing effect on the ex- the same information for the 61-element Antalya series. As
treme quantile. The plot of the histogram of the recorded series
together with the plots of the scale-adjusted density functions

Fig. 6 Plots of (1) the histogram of the 79-element series of the

Fig. 5 Plot of Hanned spectral density for the 61-element (complete) maximum daily rainfalls recorded at Alexandria (Egypt), and (2)
series of the maximum daily rainfall values recorded at Antalya scale-adjusted probability density functions of the probability distribu-
(Turkey) over the period: 1950–2010 tions deemed suitable in this study based on goodness-of-fit tests
 Arab J Geosci

point for a two-parameter distribution. This plot is a surface

and not a curve for a four- or a five-parameter distribution.
Because mostly three-parameter distributions are used for
hydrologic frequency analysis, however, the comparison of
the population curves with the scatter of points plotted by
sample L-coefficients is common. The three-parameter dis-
tributions of three-parameter log-normal (LN3), general ex-
treme values (GEV), log-logistic (LL), Pearson-3 (P3), and
log-Pearson-3 (LP3) are widely used, including this study,
and a myriad of publications are available about them (e.g.,
Cunnane 1989; Rao and Hamed 2000; WMO 2009).
Accordingly, the curves of the population L-KC versus L-
SC for distributions of LN3, GEV, P3, and LL are drawn
herein also and are shown in Fig. 8 along with two points
formed by plotting sample L-KC versus L-SC for both
Alexandria and Antalya maximum daily rainfall series. As
can be seen in this figure, the sample point of Antalya series
is close to the curve of the GEV distribution and to the point
of the Gumbel distribution. By the classic goodness-of-fit
Fig. 7 Plots of (1) the histogram of the 61-element series of the tests of χ2, Kolmogorov–Smirnov (K-S), and probability
maximum daily rainfalls recorded at Antalya (Turkey) and (2) scale- plot correlation coefficient, both the GEV and the Gumbel
adjusted probability density functions of the probability distributions
deemed suitable in this study based on goodness-of-fit tests distributions by all four parameter estimation methods per-
formed well. The magnitude of the sample skewness coef-
ficient (SC) of the Antalya series is 1.04, and this is very
seen by the results presented so far, although they are on the close to the population SC of the Gumbel distribution,
northern and southern shores of the Mediterranean Sea at which is 1.14. Interestingly, closeness of the sample and
close longitudinal distance to each other, the sample means population SCs, results of the classic goodness-of-fit tests,
of these two series are much different from each other. This is and the L-KC versus L-SC plots all are in agreement for the
probably because Alexandria is in an arid region on a fairly
plane terrain whereas Antalya is in a humid region with a lot of
orographic effect of 3,000-m-high Taurus Mountains sur-
rounding Antalya.

Plots of sample L-kurtosis coefficient (L-KC) versus

L-skewness coefficient (L-SC) together with theoretical
plots L-KC versus L-SC of some probability distributions

A recent approach for choosing a suitable probability distri-

bution for regionalized frequency analyses by the L-
Moments method as developed by Hosking and Wallis
(1997) depends on evaluation of closeness of the plotted
points of sample L-kurtosis coefficients (L-KC) versus sam-
ple L-skewness coefficients (L-SC) to the plots of the same
L-coefficients of the candidate probability distributions,
which seems to have gained popularity (e.g., Karim and
Chowdhury 1995; Kumar and Chatterjee 2005; Peel et al.
2001; Saf 2009; Seckin et al. 2011). The closeness of the
sample points to the theoretical curves can be observed
visually, and the quantitative evaluation of this is done by
a statistic symbolized by ZDIST, details of which are given in Fig. 8 Plots of (1) the sample L-coefficients of kurtosis versus skew-
ness of the maximum daily rainfalls series of Alexandria (Egypt) and
the book by Hosking and Wallis (1997). of Antalya (Turkey) and (2) the population L-coefficients of kurtosis
The plot of the population L-KC versus L-SC is a curve versus skewness of four three-parameter probability distributions con-
for a three-parameter probability distribution, and it is a sidered in this study
Arab J Geosci

Antalya series. Because the Gumbel distribution is a special In a relevant report by the World Meteorological
two-parameter version of the three-parameter GEV distribu- Organization (Cunnane 1989), the probability distributions:
tion, however, generally GEV should be preferred over Gumbel, LN2, LN3, GEV, LL (the same as generalized logis-
Gumbel. Agreement of the results of the classic goodness- tic), P3, LP3, generalized pareto (GP), and Wakeby are listed
of-fit tests with the closeness of the points of sample L-KC as distributions used for frequency analysis of both annual
versus L-SC to the curves of population L-KC versus L-SC maximum flows and annual maximum rainfalls. In another
does not always occur and is not a generalization. As seen in relevant report by WMO (2009), the Halphen distribution is
Fig. 8, for the Alexandria series also, the point of sample L- added to this list. The five-parameter Wakeby distribution is
KC versus L-SC is close to the population curve of the GEV praised and considered as a parent distribution by some sta-
distribution. And it is close to the population curve of the LL tistical arguments (e.g., Bobee and Rasmussen 1995; Hosking
distribution also. and Wallis 1997, p. 205). Because it does not have an analyt-
ical probability density function, both the methods of
Frequency analyses moments and maximum-likelihood are not available for the
Wakeby distribution. Expressed by its five parameters, and in
Except for the Wald–Wolfowitz stationarity test on some cases four parameters, the quantile function (inverse of
Alexandria series, both series passed all five tests. Hence, the cumulative distribution function) of the Wakeby distribu-
it can be concluded that according to the results of relevant tion is suitable for the probability-weighted moments (PWM)
tests, both of the fairly long maximum daily rainfall series method for computing the magnitudes of its parameters
recorded at Alexandria (Egypt) and Antalya (Turkey) are (Greenwood et al. 1979; Cunnane 1989). In a similar study,
independent, homogeneous, trend-free, and non-periodic. Chargui et al. (2012) applied the two-parameter log-normal
Hence, classical frequency analyses are applied to both the and exponential distributions to extreme rainfalls series
79-element maximum daily rainfall series of Alexandria and recorded in central Tunisia. In another similar study, Subyani
to the 61-element series of Antalya. Figures 9 and 10 show (2011) applied frequency analyses on annual maximum daily
the frequency curves of quantiles versus average return rainfall series recorded at eight stations in western Saudi Arabia
periods by probability distributions judged herein to be region using the Gumbel and log-Pearson-3 distributions and
more successful along with the elements of the series plotted found that the Gumbel distribution was a better fit. In this
by the Cunnane plotting position formula in a Gumbel graph study, frequency analysis on both the 79-element series of
paper. maximum daily rainfalls recorded at Alexandria and 61-

Fig. 9 Frequency curves of

those probability distributions
deemed suitable for the
maximum daily series of
Alexandria (Egypt) along with
the recorded elements of the
series plotted by the Cunnane
plotting position formula
 Arab J Geosci

Fig. 10 Frequency curves of

those probability distributions
deemed suitable for the
maximum daily series of
Antalya (Turkey) along with the
recorded elements of the series
plotted by the Cunnane plotting
position formula

element series recorded at Antalya is applied using almost all daily rainfall series recorded at Alexandria from 1900 through
probability distributions recommended by WMO, which are: 1994 with two 8-year discontinuities in between also indicates
Gumbel, LN3, GEV, LL, P3, LP3, GP, and Wakeby. The no trend in it. But the linear regression test applied to 9 of 79
parameters of these distributions are computed by the methods elements having exceedence probabilities smaller them 10 %
of moments, maximum-likelihood, and PWM, except indicates a highly significant increasing trend. Spectral density
Wakeby for which only the PWM method is applied by the analysis applied on these series does not indicate any period-
method presented in Appendix 4 of the report by WMO icity due to 11-year sun-spot cycles. Frequency analyses on
(Cunnane 1989). Information can be found about these prob- these series indicate that the general extreme values probabil-
ability distributions and the parameter estimation methods in ity distribution whose parameters are computed by the method
relevant literature (e.g., Bobee and Rasmussen 1995; Cunnane of probability-weighted moments (GEV-PWM) is a suitable
1989; Haktanir and Horlacher 1993; Haktanir et al. 2010; Rao model for extrapolating to magnitudes of high return periods.
and Hamed 2000; WMO 2009). The results of the χ2 and K-S The sample mean of the maximum daily rainfall
tests indicate that all distributions except GP and Wakeby pass series of Antalya is about six times higher than that of
these goodness-of-fit tests at 90 % probability for the Antalya Alexandria. These two locations are situated at the
series and LN3, LL, P3, LP3, and GEV distributions with all northeastern and southeastern shores of the Mediterranean
parameter estimation methods pass the χ2 and K-S tests at Sea, and both are longitudinally very close to each other being
90 % probability for the Alexandria series. In a similar study, very near to the 30° E meridian. Although they are at the
Iqbal and Ali (2012) found the GEV distribution as the best-fit crossing sides of the Mediterranean Sea, Alexandria is in an
distribution for annual extreme rainfall series recorded at arid region, while Antalya is in a humid region, being about
various cities of the Punjab Region. 633 km away from each other. The strong orographic effect of
the 3,000-m-high Taurus Mountains range surrounding
Antalya from three sides add considerably to this large differ-
Conclusions ence in means.
The results of the Mann–Kendall and the linear regression
According to commonly used statistical tests applied on trend tests indicate that the “impact of climate change” does
continuous 48-year-long maximum daily rainfall series not seem to have yet noticeably acted on the means of the
recorded at Alexandria (Egypt) and on continuous 61-year- maximum daily rainfall series of Antalya and Alexandria. But
long maximum daily rainfall series recorded at Antalya a statistically highly significant increasing trend in maximum
(Turkey), no trends are detected in both these series. The linear daily rainfalls having return periods 10 years and longer is
regression trend test applied on the 79-element maximum determined at Alexandria.
Arab J Geosci

References IPCC (2008) Climate change and water. Technical Paper of the
Intergovernmental Panel on Climate Change, [Bates, B.C.;
Kundzewicz, Z.W.; Wu, S., Palutikof, J.P. (Editors)]. IPCC
Anderson BT, Hayhoe K, Liang X-Z (2010) Anthropogenic-induced Secretariat, Geneva.
changes in twenty-first century summertime hydroclimatology of IPCC (2011) Managing the risks of extreme events and disasters to
the Northeastern U.S. Climate Change 99(3-4):403–423 advance climate change adaptation. A special report of Working
Ball T (1992) An iconoclast’s view of climatic change. Can Water Res Group I and Working Group II of the Intergovernmental Panel on
J 17(2):151–160 Climate Change [Field, C. B.; Barros, V.; Stocker, T.F.; Qin, D.;
Bierkens MFP (2006) Stochastic hydrology (GEO4-4420). Department Dokken, D.; Ebi, K.L.; Mastrandrea, M. D.; Mach, K. J.; Plattner,
of Physical Geography, Utrecht University, The Netherlands G.-K.; Allen, S. K.; Tignor, M. and P. M. Midgley (editors)].
Bobee B, Rasmussen PF (1995) Recent advances in flood frequency Cambridge University Press, Cambridge, United Kingdom and
analysis. Rev Geophys, Supp, US Nat Report to Int Union of New York, NY, USA
Geodesy Geophys 1991–1994:1111–1116 Iqbal MJ, Ali M (2012) A probabilistic approach for estimating return
Burn DH, Mansour R, Zhang K, Whitfield PH (2011) Trends and period of extreme annual rainfall in different cities of Punjab. Arabian
variability in extreme rainfall events in British Columbia. Can J Geosciences, in production. doi:10.1007/s12517-012-0548-z
Water Res J 36(1):67–82 Iqbal MJ, Quamar J, Yousufzai MAK (2011) Spectral analysis of local
Chargui S, Slimani M, Cudennec C (2012) Statistical distribution of climatic fluctuations. Arabian J Geosciences 4(1–2):291–298
rainy events characteristics and instantaneous hyetographs gener- Karim MA, Chowdhury JU (1995) A comparison of four distributions
ation (Merguellil watershed in central Tunisia). Arabian J Geo- used in flood frequency analysis in Bangladesh. Hydrological Sci
sciences, in production. doi:10.1007/s12517-011-0440-2 J 40(1):55–66
Collins MJ (2009) Evidence for changing flood risk in New England Kendall MG (1975) Rank Correlation Methods. Griffin, London
since the late 20th century. J Am Water Res Assoc 45(2):279–290 Kite GW (1977) Frequency and risk analysis in hydrology. Water
Cunnane, C.(1989) Statistical distributions for flood frequency analy- Resources Publications, Fort Collins, CO
sis. WMO Operational Hydrology Report No.33. World Meteo- Köppen W, Geiger R (1936) Das Geographische System der Klimate.
rological Organization, Geneva, Switzerland. In: Köppen W, Geiger R (eds) Handbuch der Klimatologie. Verlag
Douglas EM, Fairbank CA (2011) Is precipitation in Northern New Gebrüder Bornträger, Berlin
England becoming more extreme? Statistical analysis of extreme Kottegoda NT (1980) Stochastic water resources technology. The
rainfall in Massachusetts, New Hampshire, and Maine and MacMillan Press Ltd, Hong Kong
updated estimates of the 100-year storm. J Hydrologic Eng, Kumar R, Chatterjee C (2005) Regional flood frequency analysis using
ASCE 16(3):203–217 L-moments for North Brahmatputra Region of India. J Hydrologic
Ehsanzade E, El Adlouni S, Bobee B (2010) Frequency analysis Eng, ASCE 10(1):1–7
incorporating a decision support system for hydroclimatic varia- Kundzewicz ZW, Ulbrich U, Brücher T, Graczyk D, Krüger A,
bles. J Hydrologic Eng, ASCE 15(11):869–881 Leckebusch G, Menzel L, Pińskwar I, Radziejewski M,
Fujibe F, Yamazaki N, Katsuyama M, Kobayashi K (2005) The in- Szwed M (2005) Summer floods in Central Europe—climate
creasing trend of intense precipitation in Japan based on four- change track? Nat Hazards 36(1/2):165–189
hourly data for a hundred years. SOLA 1:2005–2012 Nandargi S, Dhar ON (2011) Extreme rainfall events over the Hima-
Greenwood JA, Landwehr JM, Matalas NC, Wallis JR (1979) Proba- layas between 1871 and 2007. Hydrological Sci J 56(6):930–945
bility weighted moments: definition and relation to parameters of NZCCO (2008) Climate change effects and impacts assessment: a
several distributions expressible in inverse form. Water Resources guidance manual for local governments in New Zealan–2nd edi-
Res 15(5):1049–1054 tion, Climate Change Office of the Ministry for the Environment
Groisman PY, Knight RW, Easterling DR, Karl TR, Hegerl GC, Razuvaev New Zealand.
VN (2005) Trends in intense precipitation in the climate record. J Onoz B, Bayazit M (2003) The power of statistical tests for trend
Climate 18:1326–1350 detection. Turkish J Eng Environ Sci, Tubitak 27:247–251
Guo Y (2006) Updating rainfall IDF relationships to maintain urban Peel MC, Wang QC, Vogel RM, McMahon TA (2001) The utility
drainage design standards. J Hydrologic Eng, ASCE 11(5):506– of L-moment ratio diagrams for selecting a regional probability
509 distribution. Hydrological Sci J 46(1):147–155
Haktanir T, Horlacher HB (1993) Evaluation of various distributions Rao AR, Hamed KH (2000) Flood frequency analysis. CRC Press,
for flood frequency analysis. Hydrological Sci J 38(1):15–32 Washington
Haktanir T, Cobaner M, Kisi O (2010) Frequency analyses of annual Rao AR, Azli M, Pae LJ (2011) Identification of trends in Malaysian monthly
extreme rainfalls from 5 min to 24 h. Hydrological Processes runoff under the scaling hypothesis. Hydrological Sci J 56(6):917–929
24:3574–3588 Ross SM (2004) Introduction to probability and statistics for engineers
Helsel, D. R., and Hirsch, R. M. (2002) Statistical methods in water and scientists, 3rd edn. Elsevier Academic Press, USA
resources, techniques of water-resources investigations of the Saf B (2009) Regional flood frequency analysis using L-moments for
United States, book 4, hydrologic analysis and interpretation, the Buyuk and Kucuk Menderes River Basins of Turkey. J Hy-
chapter A3, 〈http://water.usgs.gov/pubs/twri/ twri4a3/〉. drologic Eng, ASCE 14(8):783–794
Hipel KW, McLeod AI (1994) Time series modeling of water resources Seckin N, Yurtal R, Haktanır T, Doğan A (2010) Comparison of
and environmental systems. Elsevier, The Netherlands probability weighted moments and maximum likelihood methods
Hosking, J.R.M. and Wallis J.R. (1997). Regional frequency analysis used in flood frequency analysis for Ceyhan River Basin. Arabian
—an approach based on L-Moments. Cambridge, Cambridge J Sci Eng 35(1B):49–69
University Press. Seckin N, Haktanir T, Yurtal R (2011) Flood frequency analysis of
IPCC (2007) Climate Change 2007: physical science basis. Contribu- Turkey using L-moments method. Hydrological Processes
tion of Working Group I to the Fourth Assessment Report of the 25:3499–3505
Intergovernmental Panel on Climate Change, 2007. [Solomon, S.: Sen Z, Khiyami HA, Al-Harty SG, Al-Ammawi FA, Al-Balkhi AB,
Qin, D.; Manning, M.; Chen, Z.; Marquis, M.; Averyt, K.B.; Al-Zahrani MI, Al-Hawsawy HM (2012a) Flash flood inundation
Tignor, M.; Miller, H.D., (Editors)] Cambridge Univ. Press, map preparation for wadis in arid regions. Arabian J Geosciences,
New York. in production. doi:10.1007/s12517-012-0614-6, 1-10
 Arab J Geosci

Sen Z, Al AlSheikh A, Al-Turbak A, Al-Bassam A, Al-Dakheel A (2012b) WMO (2009) Guide to hydrological practices, volume II, management
Climate change impact and runoff harvesting in arid regions. Arabian of water resources and application of hydrological practices.
J Geosciences, in production. doi:10.1007/s12517-011-0354-z, 1-9 WMO-No. 168, sixth edition, World Meteorological Organiza-
Subyani AM (2011) Hydrologic behaviour and flood probability for tion, P.O. Box 2300, CH-1211, Geneva 2, Switzerland.
selected arid basins in Makkah Area, western Saudi Arabia. Yevjevich, V. (1972) Stochastic processes in hydrology. Water
Arabian J Geosciences 4(5–6):817–824 Resources Publns, Fort Collins. CO.
Villarini G, Smith JA, Baeck ML, Krajewski WF (2011) Examining Yue S, Pilon P, Caradias G (2002) Power of the Mann-Kendall and
flood frequency distributions in the Midwest U.S. J Am Water Spearman’s rho tests for detecting monotonic trends in hydrolog-
Res Assoc 47(3):447–463 ical series. J Hydrology 259:254–271