Advances in Water Resources 82 (2015) 1–15

Contents lists available at ScienceDirect

Advances in Water Resources

journal homepage: www.elsevier.com/locate/advwatres

Assessment of trends in point rainfall using Continuous Wavelet

Transforms
Md. Mamunur Rashid a,⇑, Simon Beecham a, Rezaul Kabir Chowdhury b
a
Centre for Water Management and Reuse, School of Natural and Built Environments, University of South Australia, Mawson Lakes, SA 5095, Australia
b
Department of Civil and Environmental Engineering, United Arab Emirates University, Al Ain 15551, United Arab Emirates

a r t i c l e i n f o a b s t r a c t

Article history: The existence of trends in hydro-climatic variables such as rainfall is an indication of potential climate
Received 20 October 2014 variability and climate change and the identification of such trends in rainfall is essential for the planning
Received in revised form 18 March 2015 and design of sustainable water resources. This study focuses on identifying existing trends in annual,
Accepted 13 April 2015
seasonal and monthly rainfall at thirteen stations in the Onkaparinga catchment in South Australia during
Available online 18 April 2015
the period 1960–2010. A relatively new trend detection approach, which combines a Continuous Wavelet
Transform (CWT) with the Mann Kendall (MK) test, was applied in this study. The original rainfall time
Keywords:
series was decomposed to different periodic components using a CWT and then the MK test was applied
Continuous Wavelet Transform
Mann–Kendall trend test
to detect the trends. One station showed a statistically significant (at the 5% level) negative trend for
Periodicity annual rainfall. Winter rainfall exhibited significant positive trends at four stations. In the case of
Trend analysis monthly rainfall, significant positive trends were observed in June (at seven stations), November (at
Rainfall one station) and December (at one station). The study showed that the periodic components might have
significant trends even when there are no significant trends in the original data. The periodic component
that dominates the trend in the original data varies from season to season. A sequential Mann–Kendall
analysis was found useful for identifying the trend turning points. Most of the trends, whether positive
or negative, started during the mid-1970s to mid-1980s. The technique developed in this study may also
be applied for trend detection of other hydro-climatic variables in other catchments, particularly where
temporal and spatial variabilities are high.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction continent [15,18,33,53]. Moreover, a strong spatial variation in

Hydro-meteorological variables are useful for detecting changes
in the climate over time [8]. Understanding significant trends in
rainfall is vital for sustainable water resources management [59].
The existence of trends in hydro-climatic variables such as rainfall,
temperature, humidity, evapotranspiration and stream flow is an
indication of climatic variability and climate change [6]. Changes
in the trends of observed rainfall are caused by variations in the cli-
matic environment, rainfall intensity and seasons [7,18,53,56].
Investigation of rainfall trends is therefore essential for identifica-
tion of the potential impacts of climate change on a region’s water
resources and this should be monitored regularly [2].
Australian rainfall has strong spatial and temporal variability
[15,49]. Previous investigations of rainfall trends reveal a variety
of results across Australia. Both positive (increasing) and negative
(decreasing) trends have been observed in different parts of the
⇑ Corresponding author. Tel.: +61 416504600.
E-mail address: mdmamunur.rashid@mymail.unisa.edu.au (M.M. Rashid).
rainfall trends has been observed even at the regional scale
[14,16]. Taschetto and England [53] examined the trends in
Australian rainfall from 1970 to 2006. They observed a pattern in
annual rainfall trends as being positive across the west and nega-
tive over the east part of Australia and this east–west pattern is
exhibited by the trends in summer and autumn rainfall. South
Australian annual total rainfall shows evidence of a decreasing
trend during the last half of the twentieth century [38]. Murphy
and Timbal [33] described the impact of recent low rainfall from
1997 to 2006 on the overall trends of rainfall across South-eastern
Australia since 1950. They also observed that most of the rainfall
decline occurred during autumn (March–May). Increasing and
decreasing trends in South Australian rainfall can be associated
with changes in heavy rainfall and decreases in the density of
low pressure systems in the region [53]. Because of reduced advec-
tion of moist warm air into South Australia from the northwest and
moist cold air along the southern coastline, their interaction
become limited which results in rainfall deficiencies in South
Australia [18].

http://dx.doi.org/10.1016/j.advwatres.2015.04.006
0309-1708/Ó 2015 Elsevier Ltd. All rights reserved.
2 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15

Observed time series data are often used for historical trend approximation carries most of the trend components and in order
analysis. The Mann–Kendall (MK) trend test [26,29] is the most to better represent the MK trend analysis, the respective approxi-
widely used non-parametric test for detecting monotonic trends mation components need to be added [35,36]. Whereas in case of
in rainfall, temperature, stream flow and water quality CWT, development of the periodic component is straightforward
[3,12,13,24,31,37,46,58]. Non-parametric methods are less affected and the MK trend analysis can be applied directly to each periodic
by the existence of outliers in the data series [28]. Previous studies component. In addition, we investigate how annual, seasonal and
have used the MK test for detecting trends in Australian rainfall monthly rainfall trends change over time which is useful to iden-
[4,15,16]. The MK test is robust as its performance is not affected tify the non-stationarity in the trends.
even if the time series has a skewed distribution [39]. Moreover,
this test is resilient to non-stationarity, non-linearity and missing
2. Study area and data
values [1,40,59]. A notable limitation of the MK test is that it can-
not handle data that have serial correlation, which is often found in
This study focuses on identification of rainfall trends across the
rainfall data [40,60,61]. Hamed and Rao [21] therefore proposed a
Onkaparinga catchment in SA. The catchment is approximately
modified MK test which can handle autocorrelated data. Sneyers
25 km southeast of the city of Adelaide in the Mount Lofty
[51] developed the sequential MK test which is a further advance-
Ranges (MLR) and has an area of 553 km2. There is a significant
ment on the MK test as it is able to examine the changes of trend in
gradient ranging from the low lying coastal plain to an elevation
rainfall data with time [35,40,41,58].
of 700 m in the upper catchment. It has been selected as the
A relatively recent development in signal processing is the well-
Application Test Bed for the Goyder Institute climate change pro-
known wavelet transform. It is a powerful tool which is capable of
ject (https://www.goyderinstitute.org). The water resources of
providing simultaneous time and frequency representations of a
the Onkaparinga catchment are crucial in SA because they form a
signal. Wavelet transforms can decompose the original time series
significant source of water supply for metropolitan Adelaide as
into several periodicities. Lower periodicities contain high fre-
well as providing water to farm dams and for the natural environ-
quency components (rapidly changing events) and higher periodic-
ment to maintain the bio-diversity. Heneker and Cresswell [23]
ities represent the low frequency components (slowly changing
studied the climate change impacts in the MLR and found that
events) of the original signal. The wavelet transform has been suc-
there was a 30% potential reduction in the annual runoff in the
cessfully used in the analysis of various hydrological and meteoro-
Onkaparinga catchment. Moreover, Teoh [54] identified that farm
logical variables to determine the non-stationary trends and
dams divert about 8–10% (4.5 GL) of the surface water and this
periodicities [4,27,34,41,52,57,62]. The results obtained from a
was forecast to increase to around 7–10 GL under the current man-
wavelet analysis can be used to detect the trends in different peri-
agement policies, as at 2002. This would have a profound impact
odic components as well as their contribution to the trends in the
on water security for Adelaide. Therefore the results of this study
original time series. Partal and Küçük [40] used both the wavelet
will be useful for water managers and policy makers involved in
transform and the MK test to detect non-linear trends in different
sustainable water resource management and climate change adap-
periodic components of annual total rainfall in Turkey. They
tation for the Onkaparinga catchment.
observed that a 16 year periodic component was the dominating
Around 60% of Adelaide’s municipal water is supplied from the
component for producing the trend in the original annual rainfall
Onkaparinga catchment. This catchment is also important for its
time series. Zume and Tarhule [63] observed inter-annual to deca-
valuable contribution for meeting local irrigation demand. The
dal variability in annual precipitation when they applied wavelet
Onkaparinga catchment is hydrologically very well instrumented,
transforms and the MK test to analyse the variability and trends
partly because of its importance as a water supply catchment
in rainfall and stream flow in north western Oklahoma over the
and partly because it includes the Willunga Basin Super Science
period 1983–2003. Nalley et al. [35] applied a wavelet transform
Site, which is funded by the Australian Commonwealth
and the MK test together to analyse the trends in annual, seasonal
Government’s Super Science program for the development of sci-
and monthly precipitation and stream flow in southern Ontario
entific infrastructure. The median annual rainfall over the area is
and Quebec over the period 1954–2008. They observed that the
approximately 770 mm but this varies with a strong gradient from
periodic components that affect the trend of the original series
approximately 400 mm near the coast to 1170 mm in upstream
vary from one station to another. While a number of research stud-
areas [54]. The catchment has a strong seasonal rainfall variation
ies have been performed to identify trends in the original rainfall
with less rainfall in summer (December–February) and higher
data at different temporal resolutions in Australia [4,14–
rainfall during winter (June–August) [5]. Thirteen rainfall stations
16,33,48,50,53], no studies have investigated the trends in various
were selected within and around the catchment as shown in
periodic components and their influence on the trend of the origi-
Fig. 1. These rainfall stations were previously used by Teoh [54]
nal observed rainfall time series.
who found that the temporal homogeneity of rainfall data at these
In this study we have used a Continuous Wavelet Transform
stations was satisfactory. Seven out of these selected stations
(CWT) and the MK test to investigate the existing trends in both
match with the stations selected by Heneker and Cresswell [23]
the original series and in various periodic components of the
who used a network of 20 stations for the assessment of potential
annual, seasonal and monthly rainfall in the Onkaparinga catch-
climate change on the water resources across the MLR. Details of
ment in South Australia (SA). CWT can be used to decompose a ser-
the stations used in this study are given in Table 1. Rainfall data
ies into any scale or frequency while discrete wavelet transforms
from 1960 to 2010 at these stations were collected from the
(DWTs) are limited to a discrete number of scales, which are
Bureau of Meteorology (BOM), Australia. For detail statistical prop-
mostly multiples of the power of two of the average sampling
erties of rainfall data used in this study, please refer to Rashid et al.
intervals [47]. Moreover, the CWT is capable of detecting, extract-
[45].
ing and reconstructing non-linear long term trends making it suit-
able for hydrology and climate research [1]. Furthermore the
development of the periodic components using CWT proposed in 3. Methodology
this study is useful over DWT when there is a necessity to con-
struct a periodic component for a scale other than dyadic. In In this study, annual, seasonal and monthly rainfall have been
DWT, the signal is decomposed into two components, namely analysed at thirteen rainfall stations in the Onkaparinga catchment
detail and approximation or residual. The literature shows that this in SA to identify the existence of trends. First, the original rainfall
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15 3

Fig. 1. Onkaparinga catchment and spatial location of rain gauges (G1 to G13).

data series were decomposed to different periodic components
using a CWT. An Inverse Continuous Wavelet Transform (ICWT)
was used to reconstruct various periodic rainfall series from their
corresponding wavelet coefficients. MATLAB wavelet functions
were used to calculate the CWT and ICWT. The rainfall series were
tested for autocorrelation in order to check for the existence of any
serial correlation in the data series. Trend detection methods were
selected based on the result of the autocorrelation test. Due to the
existence of significant autocorrelation in some of the original and
decomposed series, the modified MK test proposed by Hamed and
Rao [21] was used for trend detection. The sequential MK test was
applied to both the original and periodic time series. The most
influencing periodic component for the trend in the original series
was identified by comparing the plot of the progressive sequential
MK Z-values, u(t), of the original data and the periodic components.
Moreover, their level association was evaluated using Spearman’s
rank correlation coefficient (q). The sequential MK graph consists
of curves of progressive and retrograde series of the MK Z-values
that were used to identify the trend starting point for the original
time series of annual, seasonal and monthly rainfall at all thirteen
stations.
3.1. Decomposition and reconstruction of series using CWT
Considering x(t) as a continuous time series with t ranging
from 1 to þ 1; then the wavelet function (w) that depends
on the non-dimensional time parameter (g) can be defined
as [41]:
4 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15

Table 1
Details of rainfall stations used in the study.

BOM station ID Station code Latitude Longitude (decimal Elevation (m) Annual median rainfall
(decimal degree) degree) (1960–2010)
023726 G1 34.9 138.87 459 887
023750 G2 34.96 138.74 487 1034
023707 G3 35.01 138.76 445 987
023720 G4 35.03 138.81 341 779
023709 G5 35.06 138.66 376 897
023713 G6 35.1 138.79 370 746
023710 G7 35.11 138.62 267 756
023730 G8 35.18 138.76 356 821
023753 G9 35.27 138.56 104 615
023704 G10 35.01 138.65 305 838
023721 G11 35.06 138.56 170 657
023722 G12 34.93 139.01 365 577
023733 G13 35.06 138.85 363 716

 
1 ts 3.2. Periodicities in rainfall series
wðgÞ ¼ wðs; sÞ ¼ pffiffi w ð1Þ
s s
The CWT was applied to rainfall time series to obtain informa-
where t represents time, s is the time step on which the wavelet is tion about the periodic structure of the observed rainfall series. The
iterated over the time series and s is the scale ranging from 0 to 1. CWT power spectrum and global wavelet spectrum of typical
To be a wavelet, the function w(g) should have zero mean and be annual and seasonal rainfall series are presented in Fig. 2.
localised in the time–frequency space [55]. The wavelet coefficient Although the strongest variability in the global wavelet spectrum
of CWT for a time series x(t) with an equal time interval dt is defined varies with stations and temporal resolutions (annual and sea-
as the convolution of x(t) with the scaled and translated version of sonal), in general the stronger variability occurs between 2 and
w(g) [41]: 16 year scales. In some cases a few higher scales such as 32 and
Z þ1   64 year scales also showed strong variability but these scales were
1 ts not considered in this study because production of periodic com-
Wðs; sÞ ¼ pffiffi xðtÞw dt ð2Þ
s 1 s ponents at these scales might not be reliable [35,41]. In wavelet
transforms, the selection of scale to construct the periodic compo-
where the symbol (⁄) indicates the complex conjugate. A smooth nent of the original series is related to the length of the original
variation of scale (s) and time step (s) will reveal a two dimensional series. In this study we have considered rainfall data from 1960
matrix of wavelet power, |W(s, s)|2. This will show the amplitude of to 2010 as the data prior to 1960 are of lesser quality. It is not prac-
a specific frequency and how it varies with time [55]. tical to construct a periodic component using a relatively higher
The CWT can also be defined as the inverse Fourier transform of scale like 32 years or 64 years from a series of 51 data points. It
a product of Fourier transforms, which is called the Continuous has been observed in previous studies that the strongest variability
Wavelet Transform via the Inverse Discrete Fourier Transform in South Australian rainfall generally occurs in annual and decadal
(CWTFT) [55]. periodicities [4,44,52]. Previous studies also have identified that
Z South Australian rainfall shows strong coherency with different
^f ðxÞpffiffisw
þ1
1 ^  ðsxÞejxs dx
Wðs; sÞ ¼ ð3Þ climate indices such as SOI, DMI and Niño3.4 at or around the 2,
2p 1 4, 8 and 16 year scales [22,44]. The intension of this study was to
select some scales which can represent inter-annual and decadal
where ^f is the Fourier transform of the signal, w^  is the Fourier
periodicities of the rainfall and assess the trends in the original
transform of the wavelet and ejxs is the analysing function used and different periodic components. In this study 2, 4, 8 and 16 year
in the Fourier transform. scales were selected to develop the periodic components from the
The waveforms can be reconstructed by the inverse CWT original rainfall series.
(ICWTFT) utilising the delta functions [19]. The ICWTFT is classi-
cally presented in the double integral form and can be simplified 3.3. The Mann Kendall test method
to a single integral formula with some restriction upon the nature
of the wavelet. In the Mann–Kendall test, for a time series x1, . . . xn the null
 Z 1  hypothesis H0 states that the samples are independent and identi-
1 ds
xðtÞ ¼ 2Re hxðtÞ; WðtÞi ð4Þ cally distributed while the alternative hypothesis H1 for a two-
W W;d 0 s3=2 sided test states that the distribution of xi (i = 1, 2, . . ., n  1) and
xj (j = i + 1, . . ., n) are non-identical for all i, j 6 n; with i – j. The test
where Re[. . .] is the real part and h , i denotes the inner product. For
statistic S is defined as [25]:
this analysis two key criteria of the mother wavelet should be met
(1) the mother wavelet must have a real valued Fourier transform X
n1 X
n

and (2) The Fourier transform of the wavelet has support only on S¼ signðxj  xi Þ ð5Þ
i¼1 j¼iþ1
the set of non-negative frequencies. The Morlet wavelet [32] satis-
fies these conditions and was used for this study. The Morlet wave- 8
let has been used before for analysis of rainfall series [4,15,20,42]. A < þ1 if ðxj  xi Þ > 0
>
detailed description of this method is available in MathWorks [30]. Signðxj  xi Þ ¼ 0 if ðxj  xi Þ ¼ 0 ð6Þ
>
:
This method can be applied to decompose and reconstruct the ser- 1 if ðxj  xi Þ < 0
ies for any scale.
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15 5

Fig. 2. Continuous wavelet power spectrum and global wavelet spectrum of typical rainfall series for (i) annual (ii) summer (iii) autumn (iv) winter and (v) spring. The thick
curved black line represents the cone of influence (COI). The strength of power (%) in the contour image is labelled by a colour bar. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)

 
where xi and xj are the ranked values of the data and n is the length n
S0t ¼ fnðn  1Þð2n þ 5Þ=18g  ð9Þ
of the record. For independently and identically distributed data ns
with zero mean, the variance (St) of the statistic (S) is given
where n=ns is the correction factor for autocorrelation in the data-
as [37]:
" #, set. The correction term is defined as:
X
q
 X
n1
St ¼ nðn  1Þð2n þ 5Þ  t p ðtp  1Þð2t p þ 5Þ 18 ð7Þ n 2
¼ 1 þ ðn  iÞðn  i  1Þðn  i  2Þqs ðiÞ
p¼1 ns
 nðn  1Þðn  2Þ i¼1
where tp is the number of ties for the pth value and q is the number ð10Þ
of tied values. The Mann–Kendall test statistic, Z, can be estimated
where qs(i) is the autocorrelation function of the ranks of the obser-
as:
vations, which is computed using the inverse of Eq. (9) [21].
8
S1
p ffiffiffi
>
> if St > 0
>
< St 3.4. Sequential Mann Kendall test
Z ¼ 0 if St ¼ 0 ð8Þ
>
>
>p
: ffiffiffi
Sþ1
if St < 0 The sequential Mann Kendall test [51] is used to observe the
St
change of trend with time by estimating the series of progressive,
A positive Z value indicates that there is a positive trend u(t), and retrograde, u0 (t), values which are then standardized with
whereas a negative value represents a negative trend in the time zero mean and unit standard deviation. The values of xj
series. In order to test the significance of the monotonic trends at (j = 1, 2, . . ., n) are compared with xi (i = 2, 3, . . ., n  1) and the num-
level p, the absolute value of Z is compared with Z1p/2 and if ber of cases xj > xi is counted for each comparison and denoted by
Z > Z1p/2, then the trend is significant at a level p. Here Z1p/2 is nj. Then the test statistic t, as well as its mean and variance are cal-
obtained from the standard normal cumulative distribution. In this culated by Eqs. (11)–(13), respectively.
study we have used a significance level of p = 0.05.
X
j
The existence of autocorrelation in an observed time series can tj ¼ nj ð11Þ
lead to misinterpretation of the trend test results obtained from 1
the Mann–Kendall test, as described above. A positive serial corre-
lation in the observations would increase the chance of a resulting nðn  1Þ
EðtÞ ¼ ð12Þ
significant trend, although the time series may not have any trend 4
[17,21,41]. Hamed and Rao [21] proposed a modified Mann–
Kendall test which is suitable for both autocorrelated and indepen- jðj  1Þð2j þ 5Þ
Varðt j Þ ¼ ð13Þ
dent data. They proposed an empirical formula to estimate the cor- 72
rected variance as:
6 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15
When the mean and variance are estimated, the progressive,
u(t) and retrograde, u0 (t) sequential values are calculated by the
following equation starting from the first and last data of the time
series, respectively.
t j  EðtÞ
uðtÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Varðtj Þ
In a trend test, a null hypothesis is accepted at a significance
level if juðtÞj 6 uðtÞ1a=2 , where uðtÞ1a=2 is the critical value of the
standard normal distribution with a probability exceeding a/2. In
this study a is set as 0.05. A positive value of u(t) or u0 (t) denote
an upward trend while negative values indicate a downward trend
in the time series. When the values of u(t) and u0 (t) are plotted
against time, the intersection point of the progressive and retro-
grade curves represents the approximate turning point of trend
within the original time series. When either the progressive or ret-
rograde curve exceed certain limits before or after the intersection
point, this trend turning point is considered significant at the cor-
responding level i.e. 1.96 for the 5% significance level.
4. Results and discussion
4.1. Autocorrelation
It is important to check the autocorrelation properties of the
data before making decision on the selection of trend detection
test. Fig. 3 shows the correlogram of annual and seasonal total
rainfall (all correlograms are not shown here). For annual and sea-
sonal rainfall, even though only a few stations show significant (at
the 5% level) lag 1 autocorrelation, all stations show non-signifi-
cant autocorrelation at various lags. Rashid et al. [43] also identi-
fied both significant and non-significant lag 1 autocorrelations in
these rainfall stations for different temporal resolution (daily,
monthly, seasonal and annual) rainfall.
4.2. Annual rainfall analysis
The CWT approach was used to decompose the total annual
rainfall series into four identical resolutions including 2 year peri-
odicity (C1), 4 year periodicity (C2), 8 year periodicity (C3) and
16 year periodicity (C4). For example, Fig. 4 shows the normalised
annual rainfall for the original and decomposed series for station
G4.
Statistically significant decreasing trends in the original annual
rainfall were observed at station G4 with a MK Z-value of 2.27, as
Fig. 3. Autocorrelation function for different lags, (a) annual rainfall at station G1, (b) summer (DJF) rainfall at station G4, (c) autumn (MAM) rainfall at station G9, (d) winter
(JJA) rainfall at station G12, and (e) spring (SON) rainfall at station G3. Horizontal solid lines represent the confidence limits at a = 5%.
shown in Table 2. Taschetto and England [53] also observed
decreasing trends in annual total rainfall in South Australia during
recent decades. Over the decade (1997–2006) the mean rainfall
across south-eastern Australia has been approximately 14% lower
than the climatological mean from 1961 to 1990 [33]. The MK Z-
ð14Þ values of decomposed annual rainfall series (C3 and C4) become
statistically significant for some of the stations although these
were not significant for the original series at those stations. This
implies that the higher periodic components are dominating the
trends in annual rainfall in the Onkaparinga catchment. Annual
rainfall has a long term slowly generating trend (Fig. 4). Low peri-
odicity time series show very abrupt changes with time and do not
contain any component of statistically significant trend (Table 2). It
can be seen that the periodic components of the original rainfall
data show a combination of positive and negative MK Z-values at
all rainfall stations. This indicates that the trend in the original
rainfall data originates from the combined effects of trends in dif-
ferent periodic components. Also, it is not necessarily the case that
the trend sign in all the periodic components will follow the sign of
the trend in the original data.
Table 3 shows the relationship between the original and differ-
ent periodic components of annual rainfall at all thirteen stations
in terms of Spearman’s rank correlation coefficients of the progres-
sive MK Z-values. Table 3 indicates that generally higher periodic
components (C4) show higher correlation with the original series.
However, for stations G1 and G8, the highest correlation coeffi-
cients (significant at the 5% level) were observed for the 2 year
periodic component (C1). A trend in the 8 year (C3) periodic com-
ponent was most influential in the original annual rainfall at sta-
tions G3, G5 and G11. In contrast, the trends in the annual
rainfall at stations G2, G4, G6, G7, G9, G10 and G12 were domi-
nated by the 16 year (C4) periodicity. Fig. 5 shows the changes of
the MK Z-values with time for the original data and for the periodic
components C1–C4 for rainfall station G4. It should be noted that
due to space limitations, the results for every station are not pre-
sented graphically. It is seen that the higher periodic component
(C4) is more harmonious with the original data than the other
lower periodic components which was also reflected in the corre-
lation analysis.
Fig. 6 shows the plot of the progressive, u(t) and retrograde, u0 (t)
series of MK Z-values against time for annual rainfall. There is a
potential trend turning point during 1976–1985 for rainfall sta-
tions G1, G3, G4, G7, G9, G12 and G13. Other rainfall stations did
not show any potential trend changing point. At stations G3 and
G12, a positive trend started in 1981 and 1977, respectively. On
the other hand at stations G1, G4, G7, G9 and G12 a negative trend
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15 7

Fig. 4. Normalized annual rainfall for the original and four decomposed periodic components of C1 (2 year), C2 (4 year), C3 (8 year) and C4 (16 year) series at rainfall station
G4.

Table 2
Mann–Kendall test statistics (Z-values) for the original and decomposed periodic series of annual rainfall.

Data G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

Original 0.78 0.03 1.30 2.27* 0.41 0.23 0.82 0.23 0.89 0.18 0.03 0.90 0.78
C1 0.13 0.00 0.38 0.26 0.37 0.08 0.18 0.16 0.06 0.18 0.05 0.26 0.94
C2 0.39 0.32 0.59 0.13 0.48 0.19 0.02 0.02 0.40 1.57 0.45 0.45 1.08
C3 1.34 0.16 0.28 0.94* 1.60 0.75 0.81 0.68 2.01* 0.67 0.63 0.73 2.27*
C4 1.01 0.02 0.99 0.88 1.16 2.11* 0.68 1.17 0.94 0.66 1.14 1.73 1.13
*
Represents significance at a = 5%.

Table 3
Comparison of progressive sequential MK Z-values of the original and decomposed series (C1 to C4) of annual rainfall at different stations.

G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

C1 0.41* 0.34* 0.06 0.26 0.32* 0.29* 0 0.50* 0.27 0.2 0.15 0.32* 0.03
C2 0.17 0.2 0.1 0.03 0.21 0.13 0.08 0.23 0.31* 0.36* 0.33* 0.31* 0.26
C3 0.29* 0.02 0.46* 0.09 0.51* 0.22 0.04 0.31* 0.06 0.32* 0.59* 0.34* 0.05
C4 0.4* 0.55* 0.31* 0.41* 0.24 0.35* 0.54* 0.47* 0.5* 0.41* 0.09 0.45* 0.04
*
Represents significance at a = 5%.

started in 1986, 1976, 1976, 1981 and 1980 respectively. In rainfall
station G4, although the progressive series, u(t) and retrograde ser-
ies, u0 (t) did not intersect they did show a continuous decreasing
trend starting from 1976 which became significant at the 95% con-
fidence level after 1997. At station G6, a positive trend in annual
rainfall started in 1971 and then reduced to approximately a zero
trend after 1990. In general, annual rainfall in the Onkaparinga
catchment showed a decreasing trend starting from 1975.
4.3. Seasonal rainfall analysis
In the case of seasonal rainfall, the original rainfall series were
decomposed into four different decomposition level including
2 year (C1), 4 year (C2), 8 year (C3) and 16 year (C4) were consid-
ered for the MK trend analysis. The seasons were classified as
summer (December–February), autumn (March–May), winter
(June–August) and spring (September–November). Table 4 shows
8 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15
Fig. 5. Sequential progressive Mann–Kendall Z statistics for different periodic series of annual rainfall at station G4. The solid and dashed lines represent the original and
decomposed series, respectively.
the MK Z-values of winter rainfall. Statistically significant (at the
5% level) positive trends were observed in the original series of
winter rainfall at stations G3, G6, G8 and G12. No significant trend
was observed in any station for the original series of summer,
autumn and spring rainfall. However a significant negative trend
was observed for 4 year periodicity for autumn rainfall at station
G5. In the case of winter rainfall, a significant positive trend was
observed for the 4 year periodic component at station G3. For rain-
fall stations G6, G8 and G12 although there was a significant trend
in the original winter rainfall series, no significant trends were
observed in the decomposed series (C1–C4). It seems that winter
season rainfall has a very slowly generating trend which might
be dominated by higher periodicities than the ones considered in
this study. Observed positive trends in winter rainfall have also
been reported in earlier studies. Taschetto and England [53]
observed positive trends in winter rainfall over certain regions of
southern Australia. Earlier studies have shown that South
Australia’s winter rainfall is significantly correlated with the
Southern Oscillation Index (SOI) [14]. The Indian Ocean Dipole
(IOD) has also been identified as a dominant driver of winter rain-
fall in South Australia [11,18].
The statistics for the relationship between the progressive MK
Z-values of the original and different periodic components for
winter rainfall are listed in Table 5 (only the winter season is
shown here as an example). In winter, at station G8 the highest
statistically significant correlation with the original series was
observed for the 2 year periodic component. Stations G7 and
G13 showed the highest correlation for the 4 year periodic com-
ponent; stations G1, G4 and G9 showed the highest correlation
for the 8 year periodic component and station G6 showed the
highest correlation for the 16 year periodic component. Stations
G2, G3, G5, G10, G11 and G12 did not show any significant cor-
relation between the four periodic components and the original
series. Although the decomposed series of winter rainfall did
not show any significant trends over the period 1960–2010, sig-
nificant correlation was observed between the progressive MK
Z-values of the original and decomposed series. In summer, the
16 year periodic component had a significant influence on the
trend of the original rainfall series. In autumn, the highest
statistically significant correlation was observed for the 2 year
(at station G1), 4 year (at station G3), 8 year (at stations G6, G7,
G8, G11, G12 and G13) and 16 year (at stations G2, G4, G5, G9
and G10) periodic components. For spring rainfall at nine stations
the progressive MK Z-values for the 16 year periodic component
showed the highest significant correlation with the original ser-
ies. The results reveal that the temporal variability of trend in
the original series is significantly influenced by the different peri-
odic components. It is also evident that the periodic component
that dominates the trend in the original data varies from season
to season. The seasonal variation might be due to the variation in
the dominant climate drivers for seasonal rainfall.
Figs. 7 and 8 represent the progressive, u(t) and retrograde,
u0 (t) MK Z-values of the sequential Mann–Kendall test against
time for summer (DJF) and winter (JJA) season, respectively
(autumn and spring are not shown here). Fig. 6 shows that an
increasing mode of trend was observed for summer rainfall from
1967 to 1975 in all rainfall stations which start to decrease after
1975 and has been continuing during the present time. Two dis-
tinct step change point have identified in all rainfall station one
of which in 1970 (changed to increasing trend) and other one
in 1984 (changed to decreasing trend). However, stations G10
and G12 showed positive trends in the recent time period from
1994 to 2010. In the case of autumn rainfall, negative trends
were observed to dominate for most of the time over the period
1960–2010 for all stations (not shown here). Murphy and Timbal
[33] mentioned that about 60% of the rainfall decline in south-
eastern Australia was due to the higher decline in autumn rain-
fall. Taschetto and England [53] also observed significant decline
in autumn rainfall across south-eastern Australia over the second
half of the last century. A positive trend was observed for autumn
rainfall over a short period which started during 1980–1985 and
changed to a decreasing trend during 1990–1995. It implies that
most of the reduction in autumn rainfall occurred after 1995
which is also consistent with the findings of Murphy and
Timbal [33]. Starting of a positive trend in winter rainfall was
identified during 1985–1990 at most of the stations (Fig. 8).
Significant positive trends in winter rainfall were observed at sta-
tions G3, G5, G8, G10, G11 and G12 within the period 1990–
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15 9

Fig. 6. MK Z-values (y-axis) against time for annual rainfall at stations G1 to G13. The solid line represents the progressive series, u(t), and the dashed line represents the
retrograde series, u0 (t). Horizontal dashed lines are the confidence limits (a = 5%).

Table 4
MK test statistics (Z-values) for the original and decomposed periodic series of winter (JJA) rainfall.

Data G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

Original 0.91 1.16 2.52* 0.00 1.26 1.82* 0.59 1.8* 0.08 1.36 1.45 2.08* 1.45
C1 0.16 0.23 0.32 0.27 0.79 0.11 0.26 0.14 0.52 0.34 0.30 0.23 0.22
C2 0.58 0.45 3.13* 0.52 0.68 0.50 0.99 0.32 0.55 0.57 0.63 0.60 0.34
C3 0.57 0.89 0.80 0.78 1.11 0.86 0.97 0.77 1.78 1.23 0.91 0.77 0.53
C4 0.73 0.95 1.61 1.11 0.43 0.09 0.31 1.48 0.60 0.34 0.29 1.66 1.20
*
Represents significance at a = 5%.

1992. At station G6, although u(t) and u0 (t) did not cross each
other, u(t) showed a continuous increasing trend starting from
1970 which become significant in 2005 (Fig. 8). For spring rainfall
(SON), no significant trend was observed in any station over the
period 1960–2010. However, eight out of thirteen stations
showed insignificant negative trend. A change point of the trend
was observed in 1980s. Starting of a positive trend was observed
in 1980s in stations G3, G7, G11 and G12. On the other hand sta-
tions G1, G4, G9 and G13 showed decreasing trend started in
early 1980s.
10 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15

Table 5
Comparison of progressive sequential MK Z-values of the original and decomposed series (C1 to C4) of winter (JJA) rainfall.

G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

* * *
C1 0.31 0.10 0.10 0.39 0.01 0.27 0.01 0.38 0.07 0.06 0.22 0.05 0.01
C2 0.00 0.11 0.10 0.21 0.06 0.30* 0.37* 0.06 0.18 0.05 0.11 0.25 0.37*
C3 0.47* 0.26 0.07 0.48* 0.15 0.17 0.32* 0.17 0.34* 0.22 0.24 0.09 0.24
C4 0.19 0.23 0.02 0.27 0.02 0.36* 0.30* 0.09 0.04 0.12 0.08 0.15 0.02
*
Represents significance at a = 5%.

Fig. 7. MK Z-values (y-axis) for the sequential Mann–Kendall test against time (year, x-axis) for summer (DJF) rainfall at stations G1 to G13. The solid line represents the
progressive series, u(t), and the dashed line represents the retrograde series, u0 (t). Horizontal dashed lines are the confidence limits (a = 5%).

4.4. Monthly rainfall analysis
The original rainfall series of different months were decom-
posed into four different levels including 2 year (C1), 4 year (C2),
8 year (C3) and 16 year (C4) and MK trend analysis was applied
to the original and each decomposed series. We have analysed
the trends of the rainfall for each month of the year separately
and identified how different monthly rainfall influences the trend
of the corresponding seasonal rainfall over the period 1960–2010
comparing the progressive series of MK Z-values. It was found wor-
thy to use monthly analysis to identify the dominating month for a
seasonal trend. Tables 6 and 7 show the MK test statistics (Z-val-
ues) for the original and decomposed periodic series of January
and June rainfall, respectively. Fig. 9 shows the plot of the progres-
sive, u(t), and retrograde, u0 (t), series of MK Z-values against time
for different months of the year at rainfall station G3. It should
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15 11

Fig. 8. MK Z-values (y-axis) for the sequential Mann–Kendall test against time (year, x-axis) for winter (JJA) rainfall at stations G1 to G13. The solid line represents the
progressive series, u(t), and the dashed line represents the retrograde series, u0 (t). Horizontal dashed lines are the confidence limits (a = 5%).

Table 6
MK test statistics (Z-values) for the original and decomposed periodic series of January rainfall.

Data G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

Original 0.57 0.03 0.03 1.23 0.54 0.70 0.42 0.66 0.17 0.57 0.27 0.28 0.56
C1 0.05 0.02 0.17 0.15 0.23 0.05 0.15 0.35 0.11 0.21 0.15 0.00 0.27
C2 0.84 0.34 0.18 1.69 0.03 0.45 0.44 0.45 0.10 0.07 0.40 0.27 0.62
C3 0.84 2.17* 1.04 1.31 0.83 2.02* 0.92 2.52* 0.90 0.91 0.68 0.83 1.20
C4 1.19 0.66 0.96 2.17* 0.81 2.21* 0.73 1.49 0.19 0.02 0.20 0.37 1.64
*
Represents significance at a = 5%.

be noted that the results for other stations are not shown here due test of December rainfall showed that in general there was an
to space limitations. increasing trend at all stations starting around 1965 which became
In December, a significant (at the 5% level) positive trend was significant during the period 1970–1972. After this time, the trend
observed in the original series at station G12. The sequential MK starts to decrease. From 1985 to the present time, the MK Z-values
12 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15

Table 7
MK test statistics (Z-values) for the original and decomposed periodic series of June rainfall.

Data G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13

* * * * * * *
Original 1.70 1.72 2.39 0.84 2.67 2.38 2.02 1.51 0.82 2.76 3.14 3.30 1.50
C1 0.02 0.00 0.10 0.10 0.19 0.06 0.22 0.16 0.38 0.15 0.28 0.24 0.00
C2 0.67 0.54 0.49 0.45 0.62 0.37 0.68 0.23 0.62 0.54 0.39 0.58 0.26
C3 1.20 1.15 1.47 1.30 1.09 1.45 1.22 1.24 1.41 0.96 1.23 0.70 1.85
C4 0.85 1.19 1.44 1.19 3.58* 0.51 1.31 0.13 1.97* 0.66 0.46 1.67 0.77
*
Represents significance at a = 5%.

Fig. 9. MK Z-values (y-axis) for the sequential Mann–Kendall test against time (year, x-axis) for different months of the year at station G3. The solid line represents the
progressive, u(t), and the dashed line represents the retrograde, u0 (t) series. Horizontal dashed lines are the confidence limits (a = 5%).

were observed to vary around zero for most of the stations. But at
rainfall station G12 a significant positive change was observed
around 1987 which continues to the present time. In January, no
significant trend was observed in the original series, as shown in
Table 6. However significant negative trends were observed at a
few stations for 8 year and 16 year periodicities. According to the
sequential MK test results, in general for most of the rainfall sta-
tions an increasing mode of trend was observed to start around
1970 and this became significant (at the 5% level) over the period
1970–1975. The trend decreased over the period 1975–1985 and
the MK Z values were observed to be around zero for the rest of
the time after 1985. Rainfall stations G10 and G12 showed positive
trends over the period 1995–2010. A step change point was
observed around 1970 (changed to increasing trend) for all rainfall
stations. For February rainfall, no significant trend was observed in
the original series over the period 1960–2010. However, a step
change was observed (changed to decreasing trend) from 1980 to
1985. In general, at most of the rainfall stations the trend of the
February rainfall starts to decrease around 1980 and this was
found to be significant (at the 5% level) around 1985. After this
time, the MK Z values starts to increase and finally turned to
around zero between 1995 and 2010. This study reveals that the
January and February rainfall have influences on the negative trend
of the summer rainfall whereas the positive trend is generally
influenced by the December rainfall over the period 1960–2010.
Two distinct step changes observed around 1970 and 1985 in sum-
mer season were dominated by the January and February rainfall,
respectively. The recent increasing trend in summer season at rain-
fall stations G10 and G12 is influenced by the December rainfall.
In the case of March, April and May rainfall, no significant
trends were observed for the original rainfall series at any station.
However, significant trends were observed for higher periodic
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15
components. A step change (to a positive trend) was observed in
March rainfall at all rainfall stations between 1980 and 1985.
Over the period 1995–2010, the MK Z-values were observed to
be around zero (no trend) for all stations. A sequential MK test of
April rainfall shows that there is a continuous negative trend exist-
ing over the period 1990–2010 at all stations. The progressive, u(t),
series of MK Z-values of May rainfall showed that there was a sig-
nificant negative trend over the period 1995–2000. The negative
trend observed in the autumn rainfall was influenced by April
and May rainfall. On the other hand, the positive trend observed
in the autumn rainfall over the period 1980–1995 was dominated
by the observed trends in the March rainfall.
In June, there is strong evidence of the existence of an increas-
ing trend. Table 7 shows that seven of the thirteen stations had sig-
nificant positive trends over the period 1960–2010. Chowdhury
and Beecham [15] observed a positive trend in Adelaide’s June
rainfall. The observed trend in the June rainfall is stimulated by
the effect of climate drivers. They found that the positive trend
of June rainfall in SA is associated with the SOI. The sequential
MK test showed that a positive trend started at all rainfall stations
during the period 1985–1990 and this became significant around
the year 2000. The result is also consistent with the finding of
Chowdhury and Beecham [15]. They observed a positive trend
starting year in Adelaide’s June rainfall in 1987. No stations
showed significant trends for the original series for July and
August rainfall. However, significant trends were observed in the
8 and 16 year periodic series. According to the sequential MK
results, there were no significant changes in the trends of July
and August rainfall over the period 1960–2010. The trends
observed in the winter rainfall in the Onkaparinga catchment were
mostly dominated by changes in the June rainfall. The recent pos-
itive trend observed in the winter rainfall starting during the per-
iod 1985–1990 was also influenced mainly by changes in June
rainfall.
In September and October, no station showed significant trends
for the original rainfall series for the period 1960–2010. The
sequential MK trend result showed significant positive trends over
the period 1975–1980 at stations G6 and G11. For October rainfall,
no significant temporal changes were observed in the trends over
the period 1960–2010. In the case of November rainfall, a signifi-
cant positive trend was observed at station G12 over the period
1960–2010. After the 1980s, a decreasing trend was observed at
rainfall stations G1, G2, G4, G9 and G13.
5. Conclusion
Decomposition of the original rainfall time series to various
periodicities using a CWT technique and subsequent application
of the MK test on those series was found to be useful for detection
of trends in annual, seasonal and monthly rainfall at thirteen rain-
fall stations in the Onkaparinga catchment in South Australia. A
statistically significant positive trend was identified in the original
series of annual rainfall at station G4 (Z = 2.27). For winter (JJA)
season rainfall, significant positive trends were observed at sta-
tions G3, G6, G8 and G12. The trends in rainfall for the other three
seasons were not statistically significant. In the case of monthly
rainfall, significant positive trends were observed for June (at sta-
tions G3, G5, G6, G7, G10, G11 and G12), November (at station
G12) and December (at station G12) rainfall.
This investigation has demonstrated how various periodic com-
ponents affect trends in original data. Application of the MK test to
the original series of rainfall data and to its various periodic com-
ponents indicates that it is not obvious that the sign of the trend in
the periodic components should be the same as the sign of the
trend in the original data. The sign of the trend in the original data
13
is the combined effect of the positive and negative signs of the
trends in the different periodic components. Periodic components
might have significant trends even though the original data does
not have any significant trend. The relationship between the pro-
gressive MK Z-values of the original rainfall and its different peri-
odic components in terms of the Spearman’s rank correlation
coefficient (q) is useful for identifying the periodic component that
dominates the trend in the original series. The dominant periodic
components vary from season to season and from one station to
another. These variations are due to the location of stations, sea-
sonal variations in rainfall, long-term changes in climate and vari-
ations in the influencing climate drivers for seasonal rainfall.
A sequential MK analysis is useful for identifying the potential
starting point of a trend and to assess the temporal variability of
trends over a period of time. This is achieved by plotting the MK
Z-values of progressive, u(t), and retrograde, u0 (t), series against
time. A strong temporal variability of trends was observed for most
of the rainfall series over the period 1960–2010. The potential
trend starting points were different for different temporal resolu-
tions (annual, seasonal and monthly) and for different stations.
However, whether positive or negative, most of the trends started
during the mid-1970s to mid-1980s.
Several studies have described the potential causes of rainfall
variability and trends in SA. It has been observed that the telecon-
nection between rainfall and different climate drivers significantly
varies from one season to another. The El Niño Southern Oscillation
(ENSO) has been identified as one of the dominant drivers of win-
ter rainfall in SA [11,14,18], whereas spring rainfall is strongly cor-
related with Dipole Model Index (DMI) [9,10]. Recent studies have
also identified that these teleconnections are non-stationary and
that they can vary at different temporal scales [22,44]. These stud-
ies also observed that for a specific period of time different periodic
components of the rainfall series are influenced by different cli-
mate drivers. This might cause differences in the magnitude and
sign of the trends for different periodic components and their com-
bined effect can modulate the trends in the original rainfall series.
Future studies could further investigate how the dominant period-
icities and the large scale climate driver cycles are linked.
The outcomes of this study provide details of the trends in
annual, seasonal and monthly rainfall at thirteen rainfall stations
in the Onkaparinga catchment in South Australia. In addition, this
study provides some baseline information about the periodic com-
ponents that affect the trends in the original data. These time–fre-
quency properties of trends can be used to investigate how natural
variability can affect rainfall trends. Moreover, the technique
developed in this study may also be applied in trend detection of
rainfall series in other catchments, particularly where temporal
and spatial variabilities are high.
Acknowledgements
This study was funded by the South Australian Government’s
Goyder Institute for Water Research through Grant C.1.1. The
Goyder Institute also provided additional scholarship funding for
the first author. The researchers are also grateful to the
Australian Bureau of Meteorology for providing meteorological
data.
References
[1] Adamowski K, Prokoph A, Adamowski J. Development of a new method of
wavelet aided trend detection and estimation. Hydrol Process
2009;23:2686–96. http://dx.doi.org/10.1002/hyp.7260.
[2] Ampitiyawatta A, Guo S. Precipitation trends in the Kalu Ganga basin in Sri
Lanka. J Agric Sci 2009;4:10–8. http://dx.doi.org/10.4038/jas.v4i1.1641.
[3] Antonopoulos VZ, Papamichail DM, Mitsiou KA. Statistical and trend analysis
of water quality and quantity data for the Strymon River in Greece. Hydrol
14 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15
Earth Syst Sci Discuss 2001;5:679–92. http://dx.doi.org/10.5194/hess-5-679-
2001.
[4] Beecham S, Chowdhury R. Temporal characteristics and variability of point
rainfall: a statistical and wavelet analysis. Int J Climatol 2010;30:458–73.
http://dx.doi.org/10.1002/joc.1901.
[5] Beecham S, Rashid M, Chowdhury RK. Statistical downscaling of multi-site
daily rainfall in a South Australian catchment using a generalized linear model.
Int J Climatol R Meteorol Soc 2014. http://dx.doi.org/10.1002/joc.3933.
[6] Birsan M-V, Molnar P, Burlando P, Pfaundler M. Streamflow trends in
Switzerland. J Hydrol 2005;314:312–29. http://dx.doi.org/10.1016/
j.jhydrol.2005.06.008.
[7] Brunetti M, Colacino M, Maugeri M, Nanni T. Trends in the daily intensity of
precipitation in Italy from 1951 to 1996. Int J Climatol 2001;21:299–316.
http://dx.doi.org/10.1002/joc.613.
[8] Burn DH, Hag Elnur MA. Detection of hydrologic trends and variability. J
Hydrol 2002;255:107–22. http://dx.doi.org/10.1016/S0022-1694(01)00514-5.
[9] Cai W, Cowan T, Sullivan A. Recent unprecedented skewness towards positive
Indian Ocean Dipole occurrences and its impact on Australian rainfall. Geophys
Res Lett 2009;36:L11705. http://dx.doi.org/10.1029/2009GL037604.
[10] Cai W, van Rensch P, Cowan T. Influence of Global-Scale Variability on the
Subtropical Ridge over Southeast Australia. J Clim 2011;24:6035–53. http://
dx.doi.org/10.1175/2011JCLI4149.1.
[11] Cai W, van Rensch P, Cowan T, Hendon HH. Teleconnection pathways of ENSO
and the IOD and the mechanisms for impacts on Australian rainfall. J Clim
2011;24:3910–23. http://dx.doi.org/10.1175/2011JCLI4129.1.
[12] Caloiero T, Coscarelli R, Ferrari E, Mancini M. Trend detection of annual and
seasonal rainfall in Calabria (Southern Italy). Int J Climatol 2011;31:44–56.
http://dx.doi.org/10.1002/joc.2055.
[13] Cannarozzo M, Noto L, Viola F. Spatial distribution of rainfall trends in Sicily
(1921–2000). Phys Chem Earth Parts A/B/C 2006;31:1201–11. http://
dx.doi.org/10.1016/j.pce.2006.03.022.
[14] Chambers LE. South Australian rainfall variability and trends. BMRC Research
Report No. 92, Bureau of Meteorology, Australia; 2003.
[15] Chowdhury RK, Beecham S. Australian rainfall trends and their relation to the
southern oscillation index. Hydrol Process 2010;24:504–14. http://dx.doi.org/
10.1002/hyp.7504.
[16] Chowdhury RK, Beecham S. South Australia rainfall – trends and climate
drivers, water and climate: policy implementation challenges. In: Practical
responses to climate change national conference 2012, Canberra, Australia,
1–3 May; 2012.
[17] Cox D, Stuart A. Some quick sign tests for trend in location and dispersion.
Biometrika 1955:80–95. http://www.jstor.org/stable/2333424.
[18] Evans AD, Bennett JM, Ewenz CM. South Australian rainfall variability and
climate extremes. Clim Dyn 2009;33:477–93. http://dx.doi.org/10.1007/
s00382-008-0461-z.
[19] Farge M. Wavelet transforms and their applications to turbulence. Ann Rev
Fluid Mech 1992;24:395–458. http://dx.doi.org/10.1146/
annurev.fl.24.010192.002143.
[20] Gedalof Ze, Smith DJ. Interdecadal climate variability and regime-scale shifts
in Pacific North America. Geophys Res Lett 2001;28:1515–8. http://dx.doi.org/
10.1029/2000GL011779.
[21] Hamed KH, Rao RA. A modified Mann–Kendall trend test for autocorrelated
data. J Hydrol 1998;204:182–96. http://dx.doi.org/10.1016/S0022-
1694(97)00125-X.
[22] He X, Guan H. Multiresolution analysis of precipitation teleconnections with
large-scale climate signals: A case study in South Australia. Water Resour Res
2013;49:6995–7008. http://dx.doi.org/10.1002/wrcr.20560.
[23] Heneker TM, Cresswell D. Potential impact on water resource availability in
the mount lofty ranges due to climate change. Technical report DFW 2010/03,
Department of Water, for the Goverment of South Australia; 2010.
[24] Hirsch RM, Slack JR, Smith RA. Techniques of trend analysis for monthly water
quality data. Water Resour Res 1982;18:107–21. http://dx.doi.org/10.1029/
WR018i001p00107.
[25] Hirsch RM, Slack JR. A nonparametric trend test for seasonal data with serial
dependence. Water Resour Res 1984;20:727–32. http://dx.doi.org/10.1029/
WR020i006p00727.
[26] Kendall MG. Rank correlation methods. London, UK: Griffin; 1975.
[27] Kirkup H, Pitman A, Hogan J, Brierley G. An initial analysis of river discharge
and rainfall in coastal New South Wales, Australia using wavelet transforms.
Aust Geog Stud 2001;39:313–34. http://dx.doi.org/10.1111/1467-8470.00149.
[28] Lanzante JR. Resistant, robust and non-parametric techniques for the analysis
of climate data: Theory and examples, including applications to historical
radiosonde station data. Int J Climatol 1996;16:1197–226. http://dx.doi.org/
10.1002/(SICI)1097-0088(199611)16:11<1197::AID-JOC89>3.0.CO;2-L.
[29] Mann HB. Nonparametric tests against trend. Econ J Econ Soc 1945;13:245–59.
http://www.jstor.org/stable/1907187.
[30] MathWorks. Continuous wavlet transform; 2012. <http://www.mathworks.
com.au/help/wavelet/gs/continuous-wavelet-transform.html#bsu1pdf>.
[31] Mohsin T, Gough WA. Trend analysis of long-term temperature time series in
the Greater Toronto Area (GTA). Theor Appl Climatol 2010;101:311–27. http://
dx.doi.org/10.1007/s00704-009-0214-x.
[32] Morlet J, Arens G, Fourgeau E, Giard D. Wave propagation and sampling
theory; part I, complex signal and scattering in multilayered media.
Geophysics 1982;47:203–21. http://dx.doi.org/10.1190/1.1441328.
[33] Murphy BF, Timbal B. A review of recent climate variability and climate change
in southeastern Australia. Int J Climatol 2007;28:859–79. http://dx.doi.org/
10.1002/joc.1627.
[34] Nakken M. Wavelet analysis of rainfall–runoff variability isolating climatic
from anthropogenic patterns. Environ Model Softw 1999;14:283–95. http://
dx.doi.org/10.1016/S1364-8152(98)00080-2.
[35] Nalley D, Adamowski J, Khalil B. Using discrete wavelet transforms to analyze
trends in streamflow and precipitation in Quebec and Ontario (1954–2008). J
Hydrol 2012;475:204–28. http://dx.doi.org/10.1016/j.jhydrol.2012.09.049.
[36] Nalley D, Adamowski J, Khalil B, Ozga-Zielinski B. Trend detection in surface air
temperature in Ontario and Quebec, Canada during 1967–2006 using the
discrete wavelet transform. Atmos Res 2013;132:375–98. http://dx.doi.org/
10.1016/j.atmosres.2013.06.011.
[37] Nasri M, Modarres R. Dry spell trend analysis of Isfahan Province, Iran. Int J
Climatol 2009;29:1430–8. http://dx.doi.org/10.1002/joc.1805.
[38] Nicholls N, Lavery B. Australian rainfall trends during the twentieth century.
Int J Climatol 2006;12:153–63. http://dx.doi.org/10.1002/joc.3370120204.
[39] Onoz B, Bayazit M. The power of statistical tests for trend detection. Turk J Eng
Environ Sci 2003;27:247–51.
[40] Partal T, Küçük M. Long-term trend analysis using discrete wavelet
components of annual precipitations measurements in Marmara region
(Turkey). Phys Chem Earth Parts A/B/C 2006;31:1189–200. http://dx.doi.org/
10.1016/j.pce.2006.04.043.
[41] Partal T. Wavelet transform-based analysis of periodicities and trends of
Sakarya basin (Turkey) streamflow data. River Res Appl 2010;26:695–711.
http://dx.doi.org/10.1002/rra.1264.
[42] Prokoph A, Barthelmes F. Detection of nonstationarities in geological time
series: wavelet transform of chaotic and cyclic sequences. Comput Geosci
1996;22:1097–108. http://dx.doi.org/10.1016/S0098-3004(96)00054-4.
[43] Rashid M, Beecham S, Chowdhury R. Assessment of statistical characteristics
of point rainfall in the Onkaparinga catchment in South Australia. Hydrol Earth
Syst Sci Discuss 2013;10:5975–6017. http://dx.doi.org/10.5194/hessd-10-
5975-2013.
[44] Rashid M, Beecham S, Chowdhury RK. Influence of climate drivers on
variability and trends in seasonal rainfall in the Onkaparinga catchment in
South Australia: A wavelet approach. In: 13th International conference on
urban drainage (ICUD) 2014, 7–12 September, Kuching, Sarawak, Malaysia;
2014.
[45] Rashid MM, Beecham S, Chowdhury RK. Statistical characteristics of rainfall in
the Onkaparinga catchment in South Australia. J Water Clim Change 2014.
http://dx.doi.org/10.2166/wcc.2014.031.
[46] Río Sd, Herrero L, Fraile R, Penas A. Spatial distribution of recent rainfall trends
in Spain (1961–2006). Int J Climatol 2011;31:656–67. http://dx.doi.org/
10.1002/joc.2111.
[47] Rioul O, Vetterli M. Wavelets and signal processing. Signal Process Mag IEEE
1991;8:14–38.
[48] Shi G, Cai W, Cowan T, Ribbe J, Rotstayn L, Dix M. Variability and trend of North
West Australia rainfall: observations and coupled climate modeling. J Clim
2008;21:2938–59. http://dx.doi.org/10.1175/2007JCLI1908.1.
[49] Smith I, McIntosh P, Ansell T, Reason C, McInnes K. Southwest Western
Australian winter rainfall and its association with Indian Ocean climate
variability. Int J Clim 2000;20:1913–30. http://dx.doi.org/10.1002/1097-
0088(200012)20:15<1913::AID-JOC594>3.0.CO;2-J.
[50] Smith I. An assessment of recent trends in Australian rainfall. Aust Meteorol
Mag 2004;53:163–73.
[51] Sneyers R. On the statistical analysis of series of observations. World
Meteorological Organization, Technical note 143, Geneva, Switzerland; 1991.
[52] Speer MS, Leslie LM, Fierro AO. Australian east coast rainfall decline related to
large scale climate drivers. Clim Dyn 2011;36:1419–29. http://dx.doi.org/
10.1007/s00382-009-0726-1.
[53] Taschetto AS, England MH. An analysis of late twentieth century trends in
Australian rainfall. Int J Climatol 2009;29:791–807. http://dx.doi.org/10.1002/
joc.1736.
[54] Teoh KS. Estimating the impact of current farm dams development on the
surface water resources of the Onkaparinga River Catchment. DWLBC Report
2002/22, Department of Water, Land and Biodiversity Conservation, for the
Government of South Australia; 2003.
[55] Torrence C, Compo GP. A practical guide to wavelet analysis. Bull Am Meteorol
Soc 1998;79:61–78. http://dx.doi.org/10.1175/1520 0477(1998)079<
0061:APGTWA>2.0.CO;2.
[56] Ventura F, Rossi Pisa P, Ardizzoni E. Temperature and precipitation trends in
Bologna (Italy) from 1952 to 1999. Atmos Res 2002;61:203–14. http://
dx.doi.org/10.1016/S0169-8095(01)00135-1.
[57] Yan Z, Tsimplis MN, Woolf D. Analysis of the relationship between the North
Atlantic oscillation and sea-level changes in northwest Europe. Int J Climatol
2004;24:743–58. http://dx.doi.org/10.1002/joc.1035.
[58] Yang Y, Tian F. Abrupt change of runoff and its major driving factors in Haihe
River Catchment, China. J Hydrol 2009;374:373–83. http://dx.doi.org/10.1016/
j.jhydrol.2009.06.040.
[59] Yenigün K, Gümüs V, Bulut H. Trends in streamflow of the Euphrates basin,
Turkey. Water Manag 2008;161:189–98. http://dx.doi.org/10.1680/
wama.2008.161.4.189.
[60] Yue S, Pilon P, Phinney B, Cavadias G. The influence of autocorrelation on the
ability to detect trend in hydrological series. Hydrol Process 2002;16:1807–29.
http://dx.doi.org/10.1002/hyp.1095.
 M.M. Rashid et al. / Advances in Water Resources 82 (2015) 1–15 15

[61] Yue S, Pilon PJ, Phinney B. Canadian streamflow trend detection: impacts of [63] Zume JT, Tarhule A. Precipitation and Streamflow Variability in Northwestern
serial and cross-correlation. Hydrol Sci J 2003;48:51–63. http://dx.doi.org/ Oklahoma, 1894–2003. Phys Geogr 2006;27:189–205. http://dx.doi.org/
10.1623/hysj.48.1.51.43478. 10.2747/0272-3646.27.3.18.
[62] Zhou W, Wang X, Zhou T, Li C, Chan J. Interdecadal variability of the
relationship between the East Asian winter monsoon and ENSO. Meteorol
Atmos Phys 2007;98:283–93. http://dx.doi.org/10.1007/s00703-007-0263-6.