World Applied Sciences Journal 28 (7): 902-908, 2013

ISSN 1818-4952
IDOSI Publications, 2013
DOI: 10.5829/idosi.wasj.2013.28.07.946

Using Mann-Kendall and Time Series Techniques for Statistical

Analysis of Long-Term Precipitation in Gorgan Weather Station
1

M. Soltani, 1I. Rousta and 2Sh. S. Modir Taheri

Department of Climatology, Faculty of Geography, University of Tehran, Tehran, Iran

2
Department of Geography, Faculty of Human Sciences,
Islamic Azad University, Science and Research Branch, Tehran, Iran

Abstract: This study aims to analyze the long-term precipitation series using nonparametric methods (i.e. the
MannKendall and Time Series techniques) spanning from 1956 to 2005 in Gorgan station. The monthly total
of precipitation data collected from the Bureau of Meteorology and then the time series plots produced using
AnClim software.The outputs indicated that the direction of wintertime precipitation trend in polynomial
regression is, in general, downward in Gorgan station as well as the confidence limit bands are close to the
polynomial fit line in the middle periods.The normal distribution indicated a right skewness in precipitation time
series and the curve is completely unsymmetric and slightly bell-shaped. The spectrum of precipitation time
series (red-noise) in Blackman-Tukey method showed a continuous decrease of spectral amplitude with slightly
decreasing in precipitation frequency. As well as, the power spectra for analysis of cycles illustrated thatthere
are significant cycles of 5-years in the series. The results of u and u' time series diagrams in Mann-Kendall
method also indicated that there is not a meaningful trend or a sudden change (jump) in the time series.
A natural negative/descending trend is seen without jump, though.
Key words: Winter Precipitation Series
Frequency Mann-Kendall

Regression Normal Distribution

Gorgan Station

INTRODUCTION

non-parametric Mann-Kendall statistical test [15] Mann;

[16] Kendall to detect trend in recorded hydrologic time
series such as water quality and precipitation time series.
Trend analysis has proved to be a useful tool for effective
water resources planning, design and management since
trend detection of hydrological variables such as
precipitation provides useful information on the
possibility of change tendency of the variables in the
future [17] Yue and Wang. [18] Lettenmaier et al. looked
for evidences of long-term trends in precipitation, mean
temperature, temperature range and streamflow over the
continental USA by adopting the MannKendall test. [19]
Yue and Hashino studied long-term trends in Japanese
annual and monthly precipitation and generally found
significant negative trends. [20] Zhang et al. analyzed
precipitation totals and ratios of snowfall to total
precipitation in Canada during the 20th century and
pointed out a prevailing wetter pattern in Canadian
climate. [13] C g zoglu et al. investigated trend existences
in maximum, mean and low flows in Turkish rivers using

Time series analysis comprises methods for analyzing

time series data in order to extract meaningful statistics
and other characteristics of the data [1]. Bloomfield.
Time series data have a natural temporal ordering.
This makes time series analysis distinct from other
common data analysis problems, in which there is no
natural ordering of the observations [2]. Shumway. Time
series analysis is also distinct from spatial data analysis
where the observations typically relate to geographical
locations [3]. Boashash. Most studies concerning longterm climatologic time series trends have focused on
precipitation; [4]. Hirsch et al., [5]. Van Belle and Hughes,
[6]. Yu et al., [7] Kalayci and Kahya and [8] Kahya et al.
for water quality variables; [9] Zhang et al., [10] Burn and
Hag Elnur, [11] Kahya and Kalayci, [12] Kalayci and
Kahya and [13] C g zoglu et al. for streamflow; [14]
Turkes for precipitation. In addition, in recent years,
researchers and practitioners have frequently applied the
Corresponding Author:

Power Spectra Analysis

Mohsen Soltani, Department of Climatology, Faculty of Geography, University of Tehran, Tehran, Iran
Tel: +98 9119772934.

902

World Appl. Sci. J., 28 (7): 902-908, 2013

Thus, the difference and ratio of the winter total of

precipitation time series together with polynomial
regression, normal distribution, cumulative deviation,
autocorrelation, frequency and power spectra (PS)
dynamical analysis are considered. In addition, the
Mann-Kendall non-parametric test is analyzed for the
precipitation series in Gorgan weather station.
RESULTS
Time Series Technique: Time series analysis aims to
understand the temporal behavior of one of several
variables y(t). In fact, a time series is a sequence of
observations which are ordered in time (or space).
If observations are made on some phenomenon
throughout time, it is most sensible to display the data in
the order in which they arose, particularly since
successive observations will probably be dependent [23]
Box et al. The winter total of precipitation series together
with confidence limits 95% for the polynomial regression
trend line is shown in Fig. 2a.The time series in Gorgan
weather station indicates many fluctuations with time, so
that the long-term amount of the precipitation has
slightly decreased at the end years of the examined
period. The maxima and minima total of precipitation in the
winter series observations seen in 1958 and 1962,
respectively (Fig. 2a). In statistics, polynomial regression,
also known as polynomial least squares fitting, is a form
of nonlinear regression in which the relationship
between the independent variable x and the dependent
variable y is modelled as an nth order polynomial [24].
Cook and Weisberg. The bar plot (Fig. 2a) shows the
predictor (X axis-time) plotted against the response

Fig. 1: Geographic location of Gorgan weather station

used in this research and the underlying
topography for Golestan province - north of Iran
daily mean values for nearly 100 stations. They detected
trends in the majority of rivers located in western and
southern Turkey, as well as in some parts of central and
eastern Turkey.
Data and Methodology: The data used in this study
consists of the winter total of precipitation in Gorgan
weather station (Fig. 1) spanning from 1956 to 2005.
The station has continuous observation over 50 years,
which a period of 50-year is considered. Monthly total of
precipitation complied by the Bureau of Meteorology
(IRIMO) has been used to obtain seasonal total of
precipitation in winter period. The study analysis is
carried out using two procedures viz time series analysis
by AnClim software for time series analysis [21].
t pnek, as well as Mann-Kendall non-parametric test
using an Excel-based program developed by [22] Hemmati.

Fig. 2: (a) winter total of precipitation time series together with 95% confidence limits (blue dashed lines) for the
polynomial regression trend line (red bold line) and (b) the residual plot for same data as a. The winter total of
precipitation ranged from 1956 to 2005
903

World Appl. Sci. J., 28 (7): 902-908, 2013

(Y axis-precipitation). The polynomial fit (thick red line) is

shown surrounded by confidence limit bands (blue
dashed lines) showing the probable range of the
polynomial fit in the underlying precipitation data. The
trend indicates a slightly descending tendency toward the
end of the period. As can be seen, when the fit is good,
the confidence limit bands are close to the polynomial
fit line (i.e. middle periods - ranging from 1970 to 1990).
In contrast, once the fit is inappropriate, the confidence
limit bands are getting far away from the polynomial fit
line (i.e. beginning and end of the period).The residual
plot of the series in Gorgan station also shown in Fig. 2b.
A residual plot, by definition, is a graph that shows the
residuals on the vertical axis, which here it is precipitation
and the independent variable on the horizontal axis
namely time. Its aim is to show the relationship between
the independent variable and the response variable [25]
Westgard et al. The residual plot (Fig. 2b) shows that
there are some odd data or curved trends in the plot, the
average of the residuals is zero and the bars are not
equally represented about the x-axis. In fact, this residual
represents a slightly high difference between the
observed response variable viz precipitation series and
the value predicted by the regression line. On the other
hand, the smaller the variability of the residual values
around the regression line relative to the overall
variability, the better is the prediction. Therefore, the
wintertime total of precipitation (Y) and the time periods
(X) are not perfectly related, then there are both positive
and negative residual variances in the plot and the ratio of
variance is not around 0.0 (Fig. 2b).
The frequency of the precipitation time series (blue
bar-chart) with normal distribution overlay (red curve-line)
is shown in Fig. 3a. The normal distribution is the most
important and most widely used distribution in statistics.
The graph of the normal distribution depends on two
factors - the mean (skewness) and the standard deviation
(kurtosis). The mean of the distribution determines the
location of the center of the graph and the standard
deviation determines the height of the graph. The normal
distribution is produced by the normal density function
[26] Gentle as follow:
Y = [1/ * sqrt(2 )] * e-(x- )2/2

Table 1: Some statistical characteristics for winter total of precipitation

time series in Gorgan weather station
Length of the Series

50 years

Arithmetic Mean

173.8735 mm

Standard Deviation

66.6584`

Variance

4443.3487

Coefficient of Variance

38.34%

Coefficient of Skewness

2.1551

Coefficient of Kurtosis

8.4883

Maximal Precipitation

451.0 mm (1959)

Minimal Precipitation

75.0 mm (1963)

1st Quartile (25%)

118.6

Median

172.9

3rd Quartile (75%)

197.9

Extreme Value in Precipitation

1959 (481.0 mm)

In one hand, it can be seen that (Fig. 3a) some

distributions of the precipitation data have many more
observations on one side of the graph than the other.
Distributions with fewer observations on the right
(toward higher values) are said to be skewed right.
Thus, the winter total of precipitation in Gorgan station
with a normal distribution has a right skewness and the
curve is completely unsymmetric. On the other hand,
kurtosis is any measure of the peakedness of the
probability distribution of a real-valued random variance.
It is common to use an adjusted version of Pearson's
kurtosis, the excess kurtosis, to provide a comparison of
the shape of a given distribution to that of the normal
distribution.
Distributions with negative or positive excess
kurtosis are called platykurtic distributions or leptokurtic
distributions, respectively [27] Dodge. Therefore, since
the standard deviation value i.e.66.6584 as well as the
kurtosis value i.e. 8.4883 (Table 1) are kind of small, the
curve is slightly bell-shaped, as shown in Fig. 3a.
Note that a "gate" consisting of 1SD accounts for 68%
of the distribution or 68% of the area under the curve,
2SD accounts for 95% and 3SD accounts for >99%.
At 2SD, 95% of the distribution is inside the "gates,"
2.5% of the distribution is in the lower or left tail
and the same amount (2.5%) is present in the upper tail.
Some scientists in the field of statistics call this polygon
an error curve to illustrate that small errors from the mean
occur more frequently than large ones [27]. Gentle, as it
has taken place in Fig. 3a.
Fig. 3b indicates the cumulative deviation plot for
the series. The standard deviation is a numerical value
used to indicate how widely individuals in a group vary.
If individual observations vary greatly from the group
mean, the standard deviation is big; and vice versa.

where, X is a normal random variable, is the mean, is

the standard deviation, is approximately 3.14159 and e
is approximately 2.71828.
904

World Appl. Sci. J., 28 (7): 902-908, 2013

Fig. 3: (a) time series frequency (blue bar-chart) of the winter total of precipitation with a normal distribution overlay (red
curve-line); (b) plot of cumulative deviation; (c) Blackman-Tukey method for variance spectrum analysis in
frequency time series along with white noise (pink dashed line-WN) and red noise with 95% confidence limits
(red dashed line-RN);and (d) plot of autocorrelation coefficient with confidence limits 95% (red dashed lines). The
wintertotal of precipitation ranged from 1956 to 2005
The sum of the deviations across the entire set of all
observations from the overall sample mean is always zero
and the average deviation is zero [28] Xue and
Titterington as follows:
s=

developed by [21] Blackman and Tukey (1958) and is

based on the Wiener-Khinchin theorem, which states that
if the Fourier transform of a series x(t) is X(f) and if the
autocorrelation function of the series is R, then the
Fourier transform of R yields PX(f)=|X(f)|2 or the power
spectrum of x(t). In the Blackman-Tukey approach PX(f) is
estimated by:

n t ( x i2 )t-( x i )2t
n t (n t - 1)

where, nt( xi)t2 is the total of the sums of all the squared
individual values and ( xi)t2 is the square of the total of
the sums of all the individual values and nt is the total
number of measurements in the time period of interest.
Thus, the magnitude of the total precipitation indicates
that the size of the difference in the years ranging from
1978 to 1995 is big. While it represents a small deviation
differences at the beginning of the examined period.
Frequency, by definition, measures the number of
times something occurs in a specific amount of time. It is
also referred to as temporal frequency, which emphasizes
the contrast to spatial frequency [29] Zwillinger and
Kokoska (2010).Frequencyof the total of precipitation time
series witha Blackman-Tukey method in variance
spectrum analysis overlaytogether with white noise (pink
dashed line-WN) and red noise with 95% confidence
limits (red dashed line-RN) shown in Fig. 3c. This method

Px ( f ) =|

M 1

wkrk e 2 ifk |
k =0

where rk is the autocorrelation estimate at lag k, M is the

maximum lag considered and window length and wk is the
windowing function. Spectral analysis is an important tool
in climate research because it allows the variance of a time
series to be separated into contributions associated with
different time scales. It thus helps to understand better
the physical processes, which generate the variability
recorded in a time series. As can be seen in Fig. 3c, the
spectrum of precipitation time series (red-noise) shows a
continuous decrease of spectral amplitude with slightly
decreasing in precipitation frequency. Red noise, in fact,
in a random time series is associated with each years
value being correlated with the value of the year before.
If it is uncorrelated, then it is called white noise.
905

World Appl. Sci. J., 28 (7): 902-908, 2013

Fig. 4: (a) The power spectra(PS) analysis of winter total of precipitations frequency; and (b) same as a but for
dynamicalPS of frequency in 3D plot. The wintertotal of precipitation ranged from 1956 to 2005
In statistics, the autocorrelation of a random process
describes the correlation between values of the process
at different times, as a function of the two times or of the
time difference. It is a mathematical tool for finding
repeating patterns, such as the presence of a periodic
signal obscured by noise, or identifying the missing
fundamental frequency in a signal implied by its harmonic
frequencies. The definition of the autocorrelation between
times s and t is [30] Toros as follow:
R (s , t ) =

E[X t

t ](X s
t s

s)

38.34%, 2.1551 and 8.4883 in the winter total of

precipitation in Gorgan station. Some other statistical
information can be seen in Table 1.
Mann-Kendall Test: Sequential values u(t) and u'(t) from
the progressive analysis of the MannKendall test were
determined in order to see change of trend with time [31]
Sneyers (1990). u(t) is the same as the z values that are
found from the first to last data point. This test considers
the relative values of all terms in the time series (x1, x2, . .
., xn). The following steps are applied in sequence:

The magnitudes of xj annual mean time series, (j = 1,

. . ., n) are compared with xk, (k = 1, . . ., j-1). At each
comparison, the number of cases xj>xk is counted and
denoted by n j.
The test statistic tis then given by equation

where "E" is the expected value operator. The plot of

autocorrelation coefficient with confidence limits 95%
(red dashed lines) is shown in Fig. 3d. As can be seen
from the plot, the autocorrelation computed resulting in
number ranging from +1 to -1 for wintertime total of
precipitation time series in Gorgan weather station.
A considerable value of autocorrelation illustrates +0,
which represents a positive correlation, while some little
other values of -0 represents negative and anti-correlation
in the autocorrelation plot (Fig. 3d).The power spectra
(PS) method used for analysis of cycles in the winter total
of precipitation series in Fig. 4.Upon this method
(dynamical MESA), we can study how the cycles change
in time. Therefore, based on the plot there are significant
cycles of 5 years in the series and then a trend (cycle with
infinity length - it is on the very left side). It can also be
seen that the most frequency of precipitation occurred
between 0.1 and 0.2 (Fig. 4a).The same PS of frequency in
3D is shown in Fig. 4b, as well.
At the end, some important statistical characteristics
for precipitation series derived from the AnClim software
used in this research shown in table 1. According to the
table, for instance, the maximal and minimal of
precipitation series occurred in values of 451mm in
the year 1959 and 75mm in the year 1963, respectively.
The coefficients of variance, skew and kurtosis are in turn

t=

ni
i =1

The mean and variance of the test statistic are

n( n 1)
4
n(n 1)(2n + 5)
Var (t ) =
72
E (t ) =

The sequential values of the statistic u(t) are then

calculated as
u (t ) =

t E (t )
Var (t )

Similarly, the values of u'(t) are computed backward,

starting from the end of the series. The sequential version
of the MannKendall could be considered as an effective
way of locating the beginning year(s) of a trend [32] Hipel
and McLeod.
906

World Appl. Sci. J., 28 (7): 902-908, 2013

Fig. 5: A 5-year moving average with a polynomial trendoverlay (left-side) and u and u'time series diagrams (right-side)
for winter total of precipitation series in Gorgan weather station
The u and u' components were successively tested
for the winter total of precipitation series. Using graphical
presentation of the standardized test statistic, the
development of trend in the series was easily traced.
A 5-year moving average time series with a polynomial
trend overlay (left-side) and u and u' time series diagrams
(right-side) are shown in Fig. 5. The moving average time
series illustrates a deep descending trend over a 50-year
period. As of 1998 forward, the trend rose up till the end
of the period, though. The u and u' time series diagrams
also indicate thatthere is not a meaningful trend and/or a
sudden change (jump) in the time series, as the u & u'
lines are almost moving parallel to each other except for a
few contacts, which it has not resulted in achange in
direction
of
the
trend
(jump).Anatural
negative/descending trend is seen without jump, though.

autocorrelation showed that most of the data did reveal a

significant +1 correlation coefficient, which represents a
positive correlation. Furthermore, the magnitude of the
cumulative deviation indicated that the size of the
difference in the precipitation data was slightly big
in particular in the years ranging from 1978 to 1995.
The spectrum of precipitation time series (red-noise) in
Blackman-Tukey method showed a continuous decrease
of spectral amplitude with slightly decreasing in
precipitation frequency. As well as, the power spectra for
analysis of cycles showed there were significant cycles
of 5-years in the series.The results of the moving
average time series illustrated a deep descending trend
over a 50-year period. As of 1998 forward, the trend rose
up till the end of the period, though. The u and u' time
series diagrams also indicated that there is not a
meaningful trend or a sudden change (jump) in the time
series. Although a natural negative/descending trend was
seen without jump over the examined period.

CONCLUSIONS
Long-term analysis of winter total of precipitation
records at Gorgan weather stationin north of Iran has
been carried out through two procedures viz time series
technique and Mann-Kendall non-parametric test using
monthly total series. The trend in polynomial regression
indicated a slightly descending tendency toward the end
of the period. The confidence limit bands were close to
the polynomial fit line in the middle terms - ranging from
1970 to 1990, when the fit was good. The direction of
precipitation trend was, in general, downward in Gorgan
station. The residual plot showed some odd data or
curved trends in the plot, the average of the residuals was
zero and the bars were not equally represented about the
x-axis. In addition, the winter total of precipitation in
Gorgan station with a normal distribution had a right
skewness and the curve was completely unsymmetric and
slightly bell-shaped due to having small values in both
standard deviation and kurtosis.Analysis of the

ACKNOWLEDGMENTS
The authors wish to express their sincere thanks to
Petr t pnek- authorof AnClim software - for helpful
support during some series plots analysis used in this
research. The monthly total of precipitation data provided
by Iranian Meteorological Organization (IRIMO).
REFERENCES
1.
2.
3.

907

Bloomfield, P., 1976. Fourier analysis of time series:

An introduction. New York: Wiley
Shumway, R.H., 1988. Applied statistical time series
analysis, Englewood Cliffs, NJ: Prentice Hall
Boashash, B., 2003. Time-Frequency Signal Analysis
and Processing: A Comprehensive Reference,
Elsevier Science, Oxford, 2003 ISBN 0-08-044335-4

World Appl. Sci. J., 28 (7): 902-908, 2013

4.

5.

6.

7.

8.

9.

10.
11.
12.
13.

14.
15.
16.
17.

18.

Hirsch, R.M., J.R. Slack and R.A. Smith, 1982.

Techniques of trend analysing for monthly water
quality data. Water Resources Research, 18: 107-121.
Van Belle G, Hughes JP. 1984. Nonparametric tests for
trend in water quality. Water Resources Research 20:
127-136.
Yu, Y.S., S. Zou and D. Whittemore, 1993.
Nonparametric trend analysis of water quality data of
rivers in Kansas. Journal of Hydrology, 150: 61-80.
Kalayci, S. and E. Kahya, 1998. Detection of water
quality trends in the rivers of the Susurluk basin.
Turkish Journal of Engineering and Environmental
Science, 22: 503-514.
Kahya, E., S. Kalayci and Y. Burdurlu, 1998. Analysis
of trends in the water quality parameters of Gediz
basin, In Proceedings of 1st International Workshop
on Environmental Quality and Environmental
Engineering in the Middle East Region, Konya:
Turkey.
Zhang, X., K.D. Harvey, W.D. Hogg and T.R. Yuzyk,
2001. Trends in Canada streamflow. Water Resources
Research, 37: 987-998.
Burn, D.H. and M.A. Hag Elnur, 2002. Detection of
hydrological trends and variability. Journal of
Hydrology, 255: 107-122.
Kahya, E. and S. Kalayci, 2004. Trend analysis
of streamflow in Turkey. Journal of Hydrology,
289: 128-144.
Kalayci, S. and E. Kahya, 2004. Streamflow trends in
Sakarya basin. ARI, the Bulletin of the Istanbul
Technical University, 53: 79-84.
C g zoglu, H.K., M. Bayaz t and B. O noz, 2003.
Trends on the maximum, mean and low flows of
Turkish rivers. Journal of Hydrometeorology,
6: 280-295.
Turkes, M., 1996. Spatial and temporal analysis of
annual rainfall variations in Turkey. International
Journal of Climatology, 16: 1057-1076.
Mann, H.B., 1945. Nonparametric tests against trend,
Econometrica, 13: 245-259.
Kendall, M.G., 1975. Rank Correlation Methods,
Charles Griffin, London.
You, S. and CH. Wang, 2004. The Mann-Kendall test
modified by effective sample size to detect trend in
serially correlated hydrological series, Water
Resources Management, 18: 201-218.
Lettenmaier, D.P., E.F. Wood and J.R. Wallis, 1994.
Hydro-climatological trends in the continental United
States, (1948-88). Journal of Climate, 7: 586-607.

19. Yue, S. and M. Hashino, 2003. Long term trends of

annual and monthly
precipitation in Japan.
Journal of the AmericanWater Resources
Association, 39: 587-596.
20. Zhang, X., L.A. Vincent, W.D. Hogg and A. Niitsoo,
2000. Temperature and precipitation trends in
Canada during the 20th century. Atmospheric Ocean,
38: 395-429.
21. t pnek, P., 2008: AnClim - software for time series
analysis: Dept. of Geography, Fac. of Natural
Sciences, MU, Brno. 1.47 MB.
http://
www.climahom.eu/ AnClim.html
22. Hemmati, R., 2003. An Excel-based program to
determine the jump points in time series based on
Mann-Kendal method, available at http://
www.pishbin.blogsky.com/
23. Box, G.E.P., G.M. Jenkins and G.C. Reinsel, 1994. Time
Series Analysis: Forecasting and Control. 3rd ed.
Upper Saddle River, NJ: Prentice-Hall, 1994
24. Cook R. Dennis and Weisberg Sanford, 1982.
Residual and Influence in Regression (2013) New
York: Chapman and Hall. ISBN041224280X. Retrieved
23 February 2013
25. Westgard, J.O., P.L. Barry and E.F. Quam, 1998. Basic
QC practices: Training in statistical quality control for
healthcare laboratories. Madison, WI: Westgard
Quality Corporation.
26. Gentle, J.E., 2009. Computational Statistics, Springer,
ISBN978-0-387-98145-1., Retrieved 2010-08-06
27. Dodge, Y., 2003. The Oxford Dictionary of Statistical
Terms, OUP., ISBN 0-19-920613-9
28. Xue, J.H. and D.M. Titterington, 2011. The p-folded
cumulative distribution function and the mean
absolute deviation from the p-quantile, Statistics &
Probability Letters 81 (8): 1179-1182. Doi:10.1016/
j.spl.2011.03.014
29. Zwillinger, D. and S. Kokoska, 2010. CRC Standard
Probability and Statistics Tables and Formulae., CRC
Press. pp: 49-ISBN978-1-58488-059-2
30. Toros, H., 1993. Trend analysis of climatologic series
in Turkey. MSc thesis, Institute of Science
Technology, Istanbul Technical University
31. Sneyers, R., 1990. On the statistical analysis of series
of observations. World Meteorological Organization,
Technical Note 143, Geneva, Switzerland
32. Hipel, K.W. and A.I. McLeod, 1994. Time Series
Modelling of Water Resources and Environmental
Systems, Amsterdam: Elsevier. ISBN:0-444-89270-2

908