Full text of DOI:10.1080/02508060108686933

This article was downloaded by: [Florida International University]
On: 31 December 2014, At: 22:06
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
Water International
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/rwin20
Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
Daniel Cluis a & Claude Laberge b
a University of Quebec , Sainte-Foy, Quebec, Canada
b STATEX , Sainte-Foy, Quebec, Canada
Published online: 22 Jan 2009.
To cite this article: Daniel Cluis & Claude Laberge (2001) Climate Change and Trend Detection in Selected Rivers within the
Asia-Pacific Region, Water International, 26:3, 411-424, DOI: 10.1080/02508060108686933
To link to this article: http://dx.doi.org/10.1080/02508060108686933
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions

International Water Resources Association
Water International, Volume 26, Number 3, Pages 411–424, September 2001
Climate Change
and Trend Detection
within the Asia-Pacific
Region
in Selected Rivers
Daniel Cluis, University
of Quebec, Sainte-Foy
Claude Laberge, ST
ATEX, Sainte-Foy,
, Quebec, Canada,
Quebec, Canada
and
Abstract: Global clim
ate change is currently
of large rivers, w
hich can be considered
an issue of great concern.
This phenomenon
a regional integrator
of the local
was
studied using the runoff
precipitation occurring
streamflow records stored
in their basins. The
long-term stationarity
in the databank of the
Global Runoff D
ata Center (GRDC)
and the possibility of
trends in
at the Federal
Institute of Hydrology
in Koblenz (Germany)
were studied. Runoff
series that are geographically
distributed throughout
records originating from
78 rivers
the whole Asia-P
acific
with long monthly runoff
region were selected for
analyzed: the mean
yearly, and the maxim
study. For each of the
selected rivers, three
um and minimum m
onthly discharges. These
time series were constr
ucted and
series were sub-
mitted to a two-tier analysis.
First, a segmentation
the segmented series
were submitted to a specialized
procedure developed by
Hubert was applied
to assess
trend detection softw
are.
their stationarity. Then
The results show that
about two-thirds of
the series have rem
more changing levels
(37/78) than the m
ained stationary and
that the monthly
ean (26/78) and maxim
um (18/78)
minimum runoff exhibited
runoff. Most of the detected
graphic expansion and
urbanization in A
changes occurred during
the 1960s and 1970s,
a period of rapid dem
o-
were developed,
a number of large dam
sia, when irrigation
and other water uses
especially in tropical areas.
and reservoirs were
completed. Since these
During the same period
and within the area studied,
s
anthropic interventions
evidence supporting global
could be at the origin
climate change.
acific rivers, segmentation,
non-parametric tests.
of the changes
in runoff, there is no
Keywords: Climate change,
regionally consistent
trend detection, Asia-P
availability of an international databank for river runoff
constitutes a precious tool that should be exploited for re-
Introduction
Global climate change is currently an issue of great gional analyses.
concern and is often blamed for the occurrence of very
The Global Runoff Data Center (GRDC) in the Ger-
low frequency, extreme meteorological events. In this re- man Federal Institute of Hydrology (BfG) at Koblenz
gard, the runoff from large rivers can be considered a re- (Germany) operates under the auspices of the World
gional integrator of the local precipitation occurring in their Meteorological Organization (WMO). One of its objec-
basins. Folland et al. (1990) and Mann et al. (1999) showed tives is to collect the discharge time series of the world’s
an increase in global temperature variations over the last rivers, to store them in a unified data bank with a consis-
two decades. IPCC (1996) also concluded that there has tent format, and to disseminate this acquired information
been an increase in global mean surface air temperature for scientific use. This exchange of data allows interest-
o
th
by between 0.3 and 0.6 C since the late 19 century. These ing regional syntheses to be made, exploiting information
warmer temperatures over the last two decades could im- otherwise disseminated at the country level. Such an avail-
ply some changes in the water balance of different parts ability of regional data leads to a better global knowledge
of the world. Leith and Wakefield (1998) and Zhang et al. of the river regimes (mean values and seasonal distribu-
(in press) have shown that warmer temperatures have had tion of discharges), as well as the availability of surface
an effect on the hydrology of streams in Canada. IPCC water resources which constitute an important part of the
(1996) also stated that warmer temperatures will lead to terrestrial hydrologic cycle.
more vigorous hydrological cycle, translating into prospects
On the scientific front, this data bank constitutes a
for more severe droughts and/or floods in some places major contribution to the water budgets of the world’s
and less severe droughts and/or floods in other places. oceans and to Global Circulation Models (GCM), which
Hence, more investigation must be performed to evaluate are an increasingly important tool for providing a better
the effects of warmer temperatures on hydrology and the understanding of the phenomena driving the Earth’s cli-
411

412
D. Cluis and C. Laberge
matic environment. In this period of postulated climatic
changes and of devastating “El Niño” effects, it provides
an unbiased reference against which hypotheses can be
statistically tested and assessed.
In a more practical way, water availability constitutes
a vital but scarce and dwindling resource for many coun-
tries, which limits possibilities for current and future food
self-sufficiency. For these mostly tropical and equatorial
countries, any change in the long-term availability of wa-
ter will be fundamentally important for their survival and
future well being, both economically and politically.
ing to the availability of long time series within the da-
tabase and their adherence to the selection criteria.
They are grouped into five regional geographical sub-
sets to allow possible regionalization of the obtained
results. These five subsets are: Oceania-Pacific (19),
Southeast Asia (9), Far East Asia (25), Indian Subcon-
tinent (11), and Central Asia (13). The location of the
gauging stations is shown in Figure 1 with each station
identified by its number from Table 1.
According to GRDC procedures, countries provide
The purpose of this paper is to investigate the long- their discharge data for storage in the database and are
time trends of selected Asian and Oceanian river discharges solely responsible for the quality of these data. Lacking
chosen in WMO Regions 2 and 5 by examining and test- information about the quality and homogeneity of the data,
ing whether the eventuality of structural changes (trends) the non-parametric trend detection techniques as described
in the discharge data is related to possible modifications and used later in this paper seem to be the most appropri-
either of regional climatic conditions or land and water ate techniques, even if some detection power is lost in
uses within the river basins. For the purposes of this study, exchange for robustness. On the other hand, very little is
two statistical tools are used: the segmentation procedure known about land and water uses in the water basin areas
(Hubert et al., 1989) and DETECT specialized software controlled by the stations; this is also true for historic
(Cluis, 1988; Cluis et al., 1989) for trend detection in time changes within the river basin, human interventions, deri-
series with characteristics such as non normal distribu- vations or impoundments that might have occurred over
tion, auto-correlation, seasons, outliers, and non-equidis- the duration of the discharge records. These uncertainties
tant sampling intervals.
need to be considered in the interpretation of the results
obtained in this study, so that conclusion may eventually
be derived from these results.
Data Description
The selected stations within the Asia-Pacific region
The data were directly selected from the GRDC data used for this study are presented in Table 1. The table
bank using the GRDC Catalogue Tool software (Version presents the following information for each river: its GRDC
2.1 for Windows 95-NT). This software allows a re- station number, a number for identification on Figure 1,
searcher to query for data according to specific succes- country location code, name of the river and of the related
sive selection criteria such as aggregation frequency (daily gauging station, longitude and latitude, watershed area, first
or monthly), WMO regions (six continental entities) or sub- and last full years of operation, percentage of missing data,
region numbers (regional entities or watersheds), river and total length of the record in years. The data extracted
name or GRDC station number, country code, range of from the database and used throughout the analysis are
operational years, and size of river basin.
the monthly discharges from which yearly values were
Once the query file for stations is completed, the GRDC compounded. The few monthly missing values were first
database system extracts the selected data and provides filled in using the long-term monthly mean values. This
them to the user as an ASCII file. In this case, stations procedure was generally applied with the exception of cases
with monthly records from WMO Regions 5 (Oceania- where such a synthetic value would become a yearly maxi-
Pacific) and 2 (Asia) were extracted from the GRDC mum or minimum; in such cases, an interpolated value
database. A working data set of 78 stations was obtained calculated between successive months was preferred to
by using the following criteria as selection guidelines:
generate an occasional missing monthly value. Three se-
ries of yearly values were then created for analysis:
• Length of operation: The selected stations present a
record of a minimum of 25 years of continuous opera- • A mean yearly series obtained from the 12 monthly
tion until recent years, with generally less than 5 per- values,
cent missing data (eight stations were selected with a • A yearly series of monthly maximum values, abbrevi-
proportion of missing values greater than 5 percent but
smaller than 10 percent).
• Regional representativity: The selected stations should
drain large areas, making them representative of their
climatic regions and less sensitive to local meteorologi-
cal events.
ated as maximum monthly series,
• A yearly series of monthly minimum values, abbrevi-
ated as minimum monthly series.
The first series should allow the detection of temporal
changes in the mean level of the series, and the two last
• Geographical distribution: The chosen stations are dis- series reflect the change in levels of extreme (high or low)
tributed within the whole Asia-Pacific region accord- events over time.
IWRA, Water International, Volume 26, Number 3, September 2001

Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
413
on arborescences that avoids testing the bulk of the pos-
sible combinations. The search for the optimal segmenta-
tion is completed by a constraint applied to the produced
segmentations; segments will only be accepted if the means
of contiguous segments are significantly different; this can
be tested using the contrast concepts introduced by Scheffé
(1959) and presented by Dagnelie (1970). The Scheffé
test limits the order of the segmentations. Once the opti-
mal segmentation is obtained, the residuals (differences
between data values and the local segmentation level) are
tested for independence (Wald-Wolfowitz, 1943).
This procedure makes no hypothesis about the distri-
butional or persistence structure of the data. The authors
tested the reliability of their procedure using Monte-Carlo
simulations on constructed stationary series and found that
the Scheffé test on the absence of contrast often falsely
rejected the stationarity hypothesis, i.e., over-segmenting
the stationary series. In fact, the significance level of the
procedure is not related in a simple way to that of the
Scheffé test itself. Due to the fact that the procedure has
a tendency to over-segment, the procedure has been and
can be successfully used as an exploratory analysis to
determine which series should be studied in a complete
trend detection analysis.
Figure 1. Location of rivers studied in Asia and Oceania. The number
representing each river is shown in Table 1. CA = Central Asia; FEA
= Far-East Asia; ISC = Indian Subcontinent; SEA = Southeast Asia;
and OP = Oceania-Pacific.
Trend Detection Using the DETECT Software
Most of the classical statistical tests and techniques
have been developed with restricting hypotheses of nor-
Statistical Approach
Hubert’s Segmentation Procedure
As a preliminary analysis, a segmentation procedure mality and independence. It is well known that real-life
was applied to yearly series (minimum, mean and maxi- data diverge from these theoretical considerations. Most
mum). The segmentation procedure was developed by natural resource data exhibit distributions that are not nor-
Hubert et al. (1989) and has found many applications, es- mal and generally positively skewed, and they often simul-
pecially for testing for homogeneity and stationarity in the taneously present all three types of persistence: short-term
means of West African precipitation and discharge records. persistence, annual seasonality and eventually some long-
In the present study, the segmentation procedure is used term trends. Furthermore, extreme values (such as outli-
to reduce the number of series to be studied by the DE- ers) have a determining impact on the results of classical
TECT software. Without this preliminary step, 234 series parametric procedures.
would have to be studied by the time-consuming interac-
tive DETECT software.
To deal with this type of “messy” data, which are more
the rule than the exception in nature, classical parametric
Essentially, the segmentation procedure determines, methods exhibit several weaknesses. Montgomery and
for a record of given length, the optimal segmentation of Loftis (1987) studied the effects of non-normality, unequal
this series into two, three, four, etc., segments of constant variances, temporal persistence, seasonal fluctuations, and
levels (stepwise change). The sense of “optimal” here is unevenly spaced data on the results obtained using the
that the Root Mean Square Error between the measured Student’s t-test, and they showed that this test should not
data and the model (the different levels of each segment) be used if the samples have different distributions, unequal
is minimal.
variances, or lengths. In addition, seasonal variations or
For a series of length n, the number of possible seg- temporal persistence invalidate the results. Helsel (1987)
mentations into m segments N(n,m) can be expressed as has also shown that parametric procedures are largely in-
the number of possible groupings of (m - 1) objects se- fluenced by messy data.
lected out of a population of (n - 1) objects:
Robust techniques (i.e., techniques that give accept-
able results, even if the basic theoretical hypotheses are
not fully respected) are largely considered as the best avail-
able option. Several robust techniques are available, for
example M-Estimators (Huber, 1981) and non-parametric
ꢂ3 −ꢃꢀꢁ
ꢀꢂ3ꢄ2ꢀ =
ꢂ2 −ꢃꢀꢁꢂ3 − 2ꢀꢁ
This number quickly becomes very large and Hubert tests. In the present study, only non-parametric robust tech-
et al. (1989) have developed an optimization algorithm based niques are discussed and used.
IWRA, Water International, Volume 26, Number 3, September 2001

414
D. Cluis and C. Laberge
Table 1. Characteristics of the Selected Rivers by Area
Percent Dura-
Missing tion
Data (Years)
# on
Fig. 1
Country
Code
Watershed
Area
Begin
Year
End
Year
River
Station
Latitude
Longitude
Oceania-Pacific Area
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Darling River
Fitzroy
Daly
Bourke Town
The Gap
Mount Nancar
Ingham
Mount Bundy
Miva
Glenaladale
Coonooer
Frying Pan Creek
Mittagang Cross.
Nymboida
Serpent. Falls
Ouen-Kout
Goulet
Gore Hbr
Houpoto
Taringamutu
Mandamus
Sth Diadem
AU
AU
AU
AU
AU
AU
AU
AU
AU
AU
AU
AU
NC
NC
NZ
NZ
NZ
NZ
NZ
3009S
2310S
1383S
1863S
1292S
2595S
3775S
3644S
4304S
3618S
2998S
3237S
2078S
2223S
4610S
3786S
3886S
4279S
4447S
14594E
15010E
13241E
14613E
13165E
15250E
14737E
14330E
14684E
14909E
15272E
11601E769
16499E247
16685E69
16895E
17765E
17524E
17255E
16973E57
386,000
135,860
47,000
8,805
5,700
4,830
3,900
2,670
2,097
1,891
1,660
1944
1965
1970
1916
1957
1910
1938
1890
1949
1927
1909
1911
1956
1958
1961
1958
1963
1957
1964
1993
1995
1995
1996
1995
1995
1987
1993
1994
1993
1993
1992
1984
1984
1993
1990
1994
1990
1994
2.9
2.5
3.4
1.4
0.6
0
2.4
1.3
0.9
1.1
2.6
0.9
1.4
0
49
30
25
80
38
85
49
103
45
66
84
81
28
26
32
32
31
33
30
Herbert River
Mary River (1)
Mary River (2)
Mitchell River
Avoca River
Huon River
Murrumbidgee
Nymboida
Serpentine
Tipindje
Riviere Des Lacs
Mataura
Motu
Ongarue
Hurunui
Ahuriri
3,465
1,393
1,075
1,070
0
1.7
0
4.2
0
Far East Asia Area
13970E8,588
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
Tone
Kurihashi
Ishikari-Ohasha
Ojiya
Hirakata
Senoshita
Hankou
JP
JP
JP
JP
JP
CI
CI
CI
CI
CI
CI
CI
3613N
4312N
3730N
3480N
3353N
3058N
4577N
4023N
3463N
2890N
3492N
2385N
2317N
6967N
6242N
6958N
7070N
5225N
5643N
4843N
5063N
2418N
2465N
2270N
2305N
1938
1986
6.1
6.3
48
32
23
23
23
1.2
4.4
63
53
43
5.2
33
27
1.8
27
1
0
1.9
53
0.9
0
33
25
25
25
Ishikari
Shinano
Yodo
Chikugo
Changjiang
Songhuajiang
Yongding
Jinghe
Wujiang
Huanghe
Beijiang
Dongjiang
Yana
Penzhina
Indigirka
Lena
14153E12,697
13880E9,719
13563E7,281
13080E2,315
1954
1965
1965
1965
1986
1988
1988
1988
1865
1898
1988
1986
1982
1947
1987
1987
1938
1984
1937
4.5
4.2
4.2
1986
1987
11428E
12658E
1,488,036
391,000
121
89
Haerbin
Guanting
Zhangjiashan
Gongtan
Huayuankou
Hengshi
11560E42,500
10860E43,200
10835E58,300
11365E
11327E34,013
11430E25,325
1925
1933
1939
6.6
7.6
9.1
730,036
1988
41
46
1954
1960
1
0
Boluo
CI
Dzanghky
Kamenskoe
Vorontsovo
Kusur
Sretensk
Kluchi
Khabarovsk
Komsomolsk
Lu-Shui
Dahan
Ailiao
Laonong
RS
RS
RS
RS
RS
RS
RS
RS
TW
TW
TW
TW
13533E
16603E71,600
216,000
1984
1957
3.6
14735E
12765E
11772E
305,000
2,430,000
175,000
1994
1994
1985
0.8
1985
1990
57
59
88
1935
1897
Shilka
Kamchatka
Amur (1)
Amur (2)
Li-Wu
Yufeng
Sandimen
Xinfadaqiao
16105E45,600
1931
1984
13505E
13712E
12150E435
12118E335
12063E408
12065E812
1,630,000
1,730,000
1960
1897
1933
1993
1989
1989
1989
88
57
0
0
0
0
1964
1964
1964
Southeast Asia Area
45
46
47
48
49
50
51
52
53
Pampanga
Konga
Kelatan
Mekong (1)
Nam Chi
Nam Mun
Nan
San Agustin
kangay
Guillemard Bridge
Mukdahan
yasothon
PH
PH
MS
TH
TH
TH
TH
TH
TH
1517N
1808N
577N
1653N
1578N
1522N
1777N
2027N
1740N
12078E6,487
12070E534
10215E11,900
10473E 391,000
10415E43,100
1946
1947
1950
1974
1976
1986
1925
1991
1956
1988
5.7
6.1
28
29
7.7
1991
0.6
1991
2.9
37
0.4
37
1.1
32
66
35
1954
1956
Ubon
10487E 104,000
10055E13,300
Sirkit Dam
Chiang Saen
Nakhon Phanom
Mekong (2)
Mekong (3)
10010E
10480E
189,000
373,000
1961
1962
1991
1991
1
3.3
30
29
IWRA, Water International, Volume 26, Number 3, September 2001

Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
415
Table 1. Characteristics of the Selected Rivers by Area (continued)
Percent Dura-
Missing tion
Data (Years)
# on
Fig. 1
Country
Code
Watershed
Area
Begin
Year
End
Year
River
Station
Latitude
Longitude
Indian Subcontinent Area
54
55
56
57
58
59
60
61
62
63
64
64
Mahaweli Ganga
Gin Ganga
Karnali River
Kali Gandaki (1)
Kali Gandaki (2)
Tamur River
Peradeniya
Agaliya
Chisapani
Setibeni
Kotagaon Shringe
Mulghat
Harlinge Bridge
Farakka
Barashetra
Polavaram
Vijayawada
Jamtara
SB
SB
NE2864N
NE
NE
NE2693N
BW
IN
727N
618N
8058E1,189
8020E681
8129E42,890
8360E6,630
8435E11,400
8733E5,640
8903E
8792E
-
8178E
1950
1928
1962
1964
1964
1965
1984
1989
2.8
1.7
0
0.3
4.2
34
61
31
29
21
21
2.1
0
0
1993
1993
1985
1986
1934
1949
1947
1902
1901
1974
2801N
2775N
0
Ganges R. (1)
Ganges R. (2)
2408N
2500N
NE-
846,300
935,000
-
299,320
251,355
1949
1989
1985
1978
1979
1979
0.3
55
36
31
77
78
Sapt
Kosi
Godavari
Krishna
Narmada
IN
IN
IN
1692N
1652N
2302N
7
6.3
25
8062E
7993E16,576
Central Asia Area
66
67
68
69
70
71
72
73
74
75
76
77
78
Amu-Darya
Zaravchan
Gunt
Vakhsh
Biya
Ob
Tom (1)
Tom (2)
Tura
Yenisei
Syr-Darya
Ural
Chatly
Dupuli
Khorog
Tutkaul
Biysk
Salekhard
Novokuznetsk
Tomsk
Tiumen
Igarka
UZ
TA
TA
TA
RS
RS
RS
RS
RS
4228N
3938N
3753N
3833N
5252N
6657N
5375N
5658N
5715N
6748N
4405N
5085N
4117N
5970E
450,000
1932
1940
1932
1895
2,949,998
1931
1994
1985
1967
1985
1930
1985
1990
1985
1936
1930
1915
1990
1973
1.3
0
1.6
0
1994
0
0
2.1
62
45
35
90
0
91
25
89
0
42
64
6777E10,200
7152E13,700
6930E31,200
8527E36,900
6653E
8710E29,800
8487E57,000
6553E58,500
8650E
6705E
5128E
1894
1965
1896
0
RS
2,440,000
219,000
190,000
1933
1995
1984
1984
0
59
54
69
Tyumen-Aryk
Kushum
Uch-Kurgan
KZ
KZ
KG
7
4.3
57
Naryn
7210E58,400
Non-parametric techniques are based on the ranks of (autocorrelation and seasons).
the data within the sample. As such, they are unaffected
by the shape of the distribution and are also robust to out- Non-parametric Tests Adapted to
liers, as extreme values are limited to the maximum ranks Autocorrelation and Seasonal Dependence
even if they are very far from the bulk of the data. Al- The adapted non-parametric techniques consider two
though classical non-parametric trend tests such as the particular types of trends: (1) a stepwise (or jump-in-the-
Mann-Whitney and the Spearman tests are very useful mean) trend where, at some point in time, a sudden change
for the detection of monotonic or stepwise trends, they do of levels occurs as the result of some intervention; mean
not address the problems of temporal persistence and of levels before and after this date are compared using an
seasonal fluctuations often found in hydrological data.
adaptation of the Mann-Whitney test, to test if they are
The main developments in non-parametric procedures significantly different; and (2) a progressive, monotonic
adapted to messy data were obtained during the last 25 evolution of the series level over time. In this case, adap-
years (Lettenmaier, 1976; Hirsch et al., 1982; Hirsch and tation of the Spearman’s or Kendall’s test can be applied,
Slack, 1984) to exploit the water quality data bases result- using time as the independent variable.
ing from monitoring programs established as a result of
The statistics of tests for trend include a level or loca-
environmental concerns; these data bases were the ar- tion parameter estimate (mean, median, slope, etc.) and a
chetype of messy data since they exhibited all character- scale estimate for the level or location estimate (variance,
istics associated to messy data. It was thus quite difficult standard deviation, etc.). The scale estimate for the level
to answer the very practical question as to whether the or location estimate is generally related to the number of
state of the environment was improving or deteriorating, observations and to the variance of the sample. The ob-
which was and still remains, a very pertinent question. In servations in a sample are autocorrelated when succes-
these developments, authors have adapted non-paramet- sive observations in time are correlated meaning that each
ric tests to allow for trend detection, without being influ- observation contains some part of the information already
enced by other types of short-term interdependence available in the previous and following observations. If the
IWRA, Water International, Volume 26, Number 3, September 2001

416
D. Cluis and C. Laberge
Table 2. Effective Number of Independent Observations
for Various Combinations of Autocorrelation
Coefficients r and Series Lengths n
number of autocorrelated samples rises for a fixed and
given period, then the sampled observations become more
and more autocorrelated and the variance of the mean
more and more underestimated. An equivalent number of
independent observations, n*, lower than the actual num-
ber of dependent observations, (n), can be defined to ob-
tain an adequate estimate of the variance of the sample
mean. Each (dependent) observation has an information
content I = n*/n.
Lettenmaier (1976), using results from Matalas and
Langbein (1962), presents the relationship between n and
n* for a general autocorrelation structure and for a simple
lag-1 autoregressive Markovian (AR(1)) process. The lat-
ter process is important, as most natural processes locally
follow a Markovian-type structure reflecting the progres-
sive loss of memory of the phenomenon.
1
r
n=10
n=25
n=50
n=75
n=100 n=200
1
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
0,95
8,4
7
21
17
14
11
8,8
6,8
5
3,4
2
1,5
41
34
27
21
17
13
9,3
6,1
3,2
2
62
50
41
33
25
19
14
8,9
4,5
2,6
82
67
54
43
34
25
18
12
5,8
3,2
164
134
108
86
67
50
36
23
11
5,7
5,8
4,7
3,9
3,1
2,4
1,8
1,4
1,2
the power of non-parametric procedures varied between
In Table 2, the effective number of independent ob- 85 percent and 96 percent of that of their parametric coun-
servations (n*) is presented for some combinations of lag- terparts. In fact, when tested with a whole range of asym-
1 autocorrelation coefficients r and series lengths n metrical distributions, their power generally exceeded that
assuming that the underlying process is an AR(1). For of traditional parametric techniques.
1
example, a series of length 100 and of correlation coeffi-
cient of 0.3 is equivalent, for application of trend detection ric tests in a practical way, interactive software has been
tests, to a series of only 54 independent observations. written (Cluis, 1988; Cluis et al., 1989) and accepted as a
To exploit the new, previously described non-paramet-
Lettenmaier (1976) studied, using Monte-Carlo simu- Canadian contribution to the HOMS program (module
lations, the power of Spearman’s Rho test against linear K55.2.01) of the World Meteorological Organization
monotonic trends, and the power of the Mann-Whitney (WMO). This software, written in Fortran 77, is composed
test against step trends for series presenting an AR(1) of stand-alone modules which are executed in succession,
persistence structure. This author found that the docu- using a series of intermediate data files to transfer interim
mented power curves obtained in the case of independent results downstream from the first modules. It performs
samples were relevant for the dependent sample case if the following operations:
an equivalent number of independent observations n* was
used, instead of the actual length n of the sample.
After this breakthrough, Hirsch et al. (1982) investi-
gated the case of the seasonal fluctuations present in the
• Reading of the input data in an appropriate format;
display of the time-series; interactive appraisal and
elimination of obvious outliers.
vast majority of hydrological series. Kendall’s test (Lehman • Analysis of the frequency of sampling; ANOVA on
and D’Abrera, 1975) is used for each recognized sea-
sonal sub-series, and the resulting statistics were added
together. This property was exploited to assess if a global
months; interactive grouping of months into seasons
and test the equality of the means of the selected sea-
sons (groupings of months).
trend was present. Unfortunately this test could not be • Choice of an equi-spaced working interval, seasonal
applied if both persistence and seasonality were simulta-
neously present in the series. Hirsch and Slack (1984) in-
vestigated this last problem. In addition, Van Belle and
Hughes (1984), presented a new method for determining
if a trend was caused by a particular season.
At this point, a complete set of non-parametric tests
for monotonic and stepwise trend detection was available
for independent/autocorrelated, seasonal/non-seasonal time
series. The decision tree for choosing the appropriate test
according to the structure of the series and to the type of
trend is presented in Table 3. This set of non-parametric
or non-seasonal, with several options for filling-in miss-
ing data; analysis of the persistence structure of the
working series using significance levels for the sampled
autocorrelation coefficient.
• Analysis with inertia graphics (Mass-curves and
CUSUM functions, Cluis; 1983; Doerffet et al., 1991)
in order to assess the nature of a possible trend
(stepwise or monotonic) and also its eventual time of
occurrence.
Given the above information, the software performs
tests is well adapted to the real structure of hydrological the trend test adapted to the data, tests the significance of
data, but as they have been developed only recently, their the results and calculates the parametric values pertaining
power has only been partially established (Berryman et to each segment. The correspondence between the trend
al., 1988) and they often rely on Monte-Carlo simulations model and the data is computed as an RMSE (Root Mean
to validate performances. Nevertheless, Bradley (1968) Square Error), which has to be minimized in order to re-
demonstrated that even under the worst case situations, tain the best-fitted alternative. In a single time-series, the
IWRA, Water International, Volume 26, Number 3, September 2001

Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
417
Table 3. Set of Non-parametric Tests for Monotonic and Stepwise
Trend Detection Available for Independent/Dependent,
Seasonal/Non-Seasonal Time Series
into segments. This makes sense, since low flow values
are most likely to reflect local anthropic interventions as
flow diversions for irrigation purposes take place in the
dry season.
Type of
Trend
Persistence
Seasonality
Appropriate Test
Trend Detection Using the DETECT Software
All series that had been segmented by Hubert’s pro-
cedure were submitted to the specialized non-parametric
tests included in the DETECT software (Table 3). As al-
ready discussed, this software takes into account the sea-
sonal and/or persistence structures of the series and
redirects the treated series towards the adequate adapted
test. In fact, these characteristics (reduced seasonal
sub-series lengths, effective number of independent ob-
servations n*) are at the root of the recognized over-seg-
mentation properties (falsely rejecting the stationarity
Monotonic Markovian
trend
No seasons
With seasons
Lettenmaier/Spearman
Hirsh and Slack
persistence
No
persistence
No seasons
With seasons
Spearman/Kendall
Kendall seasonal
Stepwise
trend
Markovian
persistence
No seasons
With seasons
Lettenmaier/
Mann-Whitney
Hirsch and Slack
Mann-Whitney
Kendall seasonal
No
persistence
No seasons
With seasons
software may have to be rerun several times if there was
more than one change in level during the length of the
record or if computing for either monotonic or stepwise
structures lead to non-clearly discriminating RMSE. All
the choices made by the user are written in a report file
for further analysis of the statistical results related to the
different options run for the same series.
Table 4. Summary Results, for Each Area, of the Segmentation and
Trend Analyses Applied to the ThreeTypes of Series Investigated
(mean yearly, monthly maximum, and minimum discharge series)
Segmentation
Trend Detection
Down-
Type of
Series
ward Upward No
Trend Trend Trend
Region
Yes
No
Oceania
Pacific
(19 rivers)
mean
maximum
minimum
4
3
7
15
16
12
3
2
2
1
1
1
15
16
16
Results
The results presented here constitute a fraction of the
actual work done with the GRDC databank. The reader
interested in more detailed results should refer to Cluis (1998).
Far East
Asia
(25 rivers)
mean
maximum
minimum
11
6
18
14
19
7
6
4
5
4
1
10
15
20
10
Hubert’s Segmentation Procedure
Southeast
Asia
(9 rivers)
mean
maximum
minimum
3
4
8
6
5
1
3
3
3
0
1
4
6
5
2
We used the segmentation software developed and
provided by Hubert et al. (1989) and ran it on the three
yearly series of interest, i.e., the mean, maximum, and
minimum monthly series. It was applied to the discharge
data of the selected rivers of the Asia-Pacific region as
described in Table 1. For these 234 runs, the significance
level of 0.01 for the Scheffé test was used and, in addition,
we limited the investigation to a maximum of three seg-
mentations per record.
The summary results are presented in Table 4. Figure
2 presents the results for each station, indicating whether
a segmentation is significant for one, two or three of the
series (minimum, mean and maximum). Table 4 and Fig-
Indian mean
Subcontinent maximum
4
4
5
8
8
7
3
1
2
1
2
2
8
9
8
(12 rivers)
Central Asia
(13 rivers)
Total
minimum
mean
maximum
minimum
5
4
10
8
9
3
3
3
2
2
0
6
8
10
5
mean
maximum
minimum
27
21
48
51
57
30
18
13
14
8
5
23
52
60
41
ure 2 show that about 32 percent of the stations (and 60 hypothesis) of the procedure developed by Hubert et al.
percent of the series) are not segmented at all during their (1989).
record periods, and only 22 percent of the stations are
The summary results of trend detection are presented
segmented for all three series (minimum, mean, and maxi- in Table 4 and Figure 3, and specific results for each re-
mum). The least segmented series is the yearly series of gion are presented in the following sections.
monthly maximums, which generally presents the relatively
larger standard deviations, followed by the series of yearly
means and then by the series of monthly minimums. One
Oceania-Pacific
The results pertaining to each station within the
can note that, from the three studied yearly series (yearly Oceania-Pacific area are presented in Table 5; one can
means, monthly maximum, and monthly minimum), the see that four minimum monthly discharge series that had
series of monthly minimums are the most highly truncated been previously segmented, when revisited by this actual
IWRA, Water International, Volume 26, Number 3, September 2001

418
D. Cluis and C. Laberge
1970s while all stations with upward trends showed changes
in the late 1970s, and 1980s. These results suggest that
the opposite directions in trends for the Oceania-Pacific
area could be time related rather than spatially related if a
regional pattern can be deduced for this area. Only two
stations showed significant trends in all three series (mini-
mum, mean, and maximum): the Avoca River with up-
ward step trends in 1988 and the Murrumbidgee River
with downward step trends in 1961.
Far-East Asia
The results for Far East Asia presented in Table 6
show that three minimum monthly discharge series and
one mean monthly series that had been previously seg-
mented exhibited no trend when revisited with the DE-
TECT software. Further investigation of Table 6 and Figure
3 illustrates that all northern stations exhibited upward
trends taking place between 1960 and 1985. In the center
of the area, a majority of the stations showed downward
trends with changes in the 1960s, 1970s, and early 1980s,
the main exception being the Songhuajiang River which
showed upward step trends in all three series but the
changes in mean appeared in the 1930s and 1940s. Finally,
three stations in the south of the area show upward step
trends in the minimum series with changes in the 1960s
Figure 2. Representation of the segmentation results for all studied
rivers. The size of the dots indicates the number of series segmented
among the monthly minimum, monthly maximum, and yearly mean
series.
step, exhibited no trend (in the mean) after having been and 1970s. These results suggest that the opposite direc-
submitted to the non-parametric tests. Further investiga- tions in trends for the Far East Asia area could be ex-
tion of Table 5 and Figure 3 shows that all stations with plained by regional patterns since adjacent rivers tend to
downward trends showed changes in the 1960s and early produce similar trend patterns. Only three stations showed
Table 5. Trend Detection Analysis Results for the Yearly Mean,
the Minimum and Maximum Monthly Discharges in the Oceania-Pacific Area
Minimum Monthly Discharges
Mean Yearly Discharges
Trend
Maximum Monthly Discharges
Trend Period
Trend
Type
Period
Level*
Period
Level*
River
Period
Period
Type
Period
Type Level*
Daly
1970
1974
1973
1995
-
Step
8.9
17.9
No segmentation identified
No segmentation identified
No segmentation identified
Herbert
1916
1996
No trend
No trend
5.4
1916
1961
1971
1960
1970
1996
-
Step
Monotic
113
76
171-39
Mitchell
Avoca
1938
1987
2.24
No segmentation identified
No segmentation identified
1890
1988
1987
1993
-
Step
0.19
5.7
1890
1896
1988
1895
1987
1993
-
316
6.4
378
1890
1896
1988
1895
1987
1993
-
1240
32.6
2010
Step
Step
Step
Step
Huon
1949
1994
No trend 18.3
113
No segmentation identified
No segmentation identified
Murrumbidgee 1927
1961
1960
1993
1927
1961
1960
1993
1190
457
1927
1961
1960
1993
3730
1480
Step
55
Step
Step
Serpentine
1911
1992
No trend
6.2
1911
1971
1970
1992
-
Step
6.2 1911
0.49 1971
1970
1992
-
Step
29.1
2.19
*For monotonic trends, period level represents the value estimated for the first and last year in the period.
IWRA, Water International, Volume 26, Number 3, September 2001

Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
419
Table 6. Trend Detection Analysis Results for the Yearly Mean,
the Minimum and Maximum Monthly Discharges in the Far East Asia Area
Minimum Monthly Discharges
Mean Yearly Discharges
Maximum Monthly Discharges
Trend
Type
Period
Level*
Trend
Type
Period
Level*
Trend
Type
Period
Level*
River
Period
Period
1938 1960
1961 1985 Step
Period
Tone
1938 1947 No trend
1948 1973
1974 1985 Step
94
112
79
288
222
1938 1962
1963 1985 Step
742
535
Shinano
Yodo
1965 1988 Monotonic 295-207
1965 1988 No trend
530
No segmentation identified
No segmentation identified
1965 1983
1984 1988 Step
127
86
1965 1976
1977 1988 Step
316
234
Chikugo
1965 1974
1975 1988 Step
36
45.3
No segmentation identified
No segmentation identified
No segmentation identified
Changjiang
1865 1904
1905 1985 Step
6,370
7,230
1865 1953
1954 1985 Step
23,700
22,300
Songhuajiang 1898 1946
1947 1987 Step
108
302
1898 1928
1929 1987 Step
895
1360
1898 1931
1932 1987 Step
2,500
3,580
Yongding
1925 1942 No trend
1943 1970
1971 1988 Step
9.5
15.4
9.3
1925 1949
1950 1964 Step
1965 1988 Step
39
1925 1962
126
58.6
50.3 1963 1988 Step
21.1
Jinghe
1933 1986 No trend
17.2
No segmentation identified
No segmentation identified
Huanghe
1947 1955
1956 1988 Step
576
385
1947 1966
1967 1988 Step
1,610
1,260
No segmentation identified
No segmentation identified
No segmentation identified
Beijang
1954 1969
1970 1969 Step
201
268
No segmentation identified
No segmentation identified
Dongjiang
1960 1973
1974 1987 Step
208
351
Yana
1938 1984 No trend
0.42 No segmentation identified
No segmentation identified
No segmentation identified
Indigirka
1937 1984
1985 1994 Step
7.1
10.2
No segmentation identified
No segmentation identified
Lena
1935 1978
1979 1994 Step
1.1
1.9
No segmentation identified
No segmentation identified
Shilka
1897 1985 No trend
3.63 1898 1982
398
627
1984 1095 Step
Kamchatka 1931 1984 Monotonic 323-438
1931 1959
1960 1984 Step
736
826
1931 1959
1960 1984 Step
1780
1990
Amur (1)
1897 1946
1947 1985 Step
486
765
1897 1955 No trend
8370
1897 1953 No trend
1954 1985 Monotonic 25300-
16900
20800
1956 1985 Monotonic 10600-6640
No segmentation identified
Amur (2)
Yufeng
1933 1978
1979 1990 Step
877
1660
No segmentation identified
No segmentation identified
No segmentation identified
1964 1967
1968 1989 Step
1964 1989 Monotonic
1110
1780
1430-1980
*For monotonic trends, period level reprsents the value estimated for the first and last year in the period.
IWRA, Water International, Volume 26, Number 3, September 2001

420
D. Cluis and C. Laberge
significant trends in all three series (minimum, mean, and
maximum): the Tone and Yongding Rivers with downward
step trends in the 1960s and 1970s, and the Songhuajiang
River with upward step trends in 1930s and 1940s.
Southeast Asia
The results for Southeast Asia are presented in Table
7. Only one minimum monthly discharge series that had
been previously segmented exhibited no trend when revis-
ited with the DETECT software. Further investigation of
Table 7 and Figure 3 shows that all trends take place be-
tween 1960 and 1982. In general, the stations showed
downward step trends, with almost all exceptions being in
minimum monthly discharge series: the Nam Chi, Nam
Mun, and Nan Rivers showing upward step trends in the
1960s but no trends in the mean and maximum series, and
the Mekong (2) River showing an upward step trend in
the 1970s but downward trends in the mean and maxi-
mum series. These results suggest that the trends in the
South East Asia area could be explained by regional pat-
terns since a vast majority of stations produce similar trend
patterns.
Figure 3. Representation of the trend detection results for all the
studied rivers. The size of the arrows indicates the number of series
segmented among the monthly minimum, monthly maximum, and
yearly mean series. the direction of the arrows indicates The direction
of the trend.
Indian Subcontinent
The results for the Indian Subcontinent Area are pre-
Table 7. Trend Detection Analysis Results for the Yearly Mean,
the Minimum and Maximum Monthly Discharges in the Southeast Area
Minimum Monthly Discharges
Mean Yearly Discharges
Maximum Monthly Discharges
Trend
Type
Period
Level*
Trend
Type
Period
Level*
Trend
Type
Period
Level*
River
Period
Period
Period
Bonga
1947
1967
1966
1976
Step
Step
1.9
0.9
1947
1960
1959 No trend
1976 Monotonic
26.3
42.8-4.2
1947 1958 No trend 98.6
1959 1976 Monotonic 163-358
Kelantan
1950
1962
1961
1986
317
231
No segmentation identified
No segmentation identified
Step
Step
Step
Step
Step
Step
Mekong (1) 1925
1951
1950
1991
1550
1410
1925
1967
1966
1991 Step
8350
7270
1925 1973
1974 1991 Step
24400
20500
Nam Chi
Nam Mun
Nan
1954
1967
1966
1991
5.83
37.8
No segmentation identified
No segmentation identified
No segmentation identified
No segmentation identified
No segmentation identified
No segmentation identified
1956
1967
1966
1991
18.4
62
1956
1968
1967
1988
16.5
27.8
Mekong (2) 1961
1972
1971
1991
764
839
1961
1972
1971
1991 Step
2950
2600
1961 1971
1972 1991 Step
8080
6400
Mekong (3) 1962
1991
No trend 1440
No segmentation identified
1962 1981
1982 1991 Step
21400
17800
*For monotonic trends, period level represents the value estimated for the first and last year of the period.
IWRA, Water International, Volume 26, Number 3, September 2001

Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
421
Table 8. Trend Detection Analyses Results for the Yearly Mean,
the Minimum and Maximum Monthly Discharges in the Indian Subcontinent
Minimum Monthly Discharges
Mean Yearly Discharges
Maximum Monthly Discharges
Period
Level*
Trend
Type
Period
Level*
Trend
Type
Period
Level*
River
Period
Trend Type
Period
Period
Gin Ganga
1928 1957
1958 1989
No trend 19.5
Monotonic 26.3-11.3
No segmentation identified
No segmentation identified
No segmentation identified
No segmentation identified
Kali Gandaki (1)1964 1993
Monotonic 31.2-55.7
1964 1976
1977 1993
284
247
Step
Step
Step
Step
Kali Gandaki (2)No segmentation identified
1964 1968
1969 1985
530
457
Sapt Kosi
Krishna
No segmentation identified
1947 1969
1961 1979
1540
1250
1947 1969
1970 1979 Step
4510
5440
1901 1979
No trend
21.7
1901 1960
1961 1979
1780
1250
1901 1960 Monotonic9590-3960
Ganges R. (1) 1934 1974
1975 1989
1950
1130
No segmentation identified
1934 1945
1946 1989 Step
35200
42500
Step
Godavari
Narmada
1902 1979
Monotonic
23.7-120
No segmentation identified
No segmentation identified
No segmentation identified
1949 1974 No trend
1730
*For monotonic trends, period level represents the value estimated for the first and last year of the period.
sented in Table 8. Only one minimum monthly discharge were much more prone to changing levels than the mean
and one maximum monthly discharge series that had been and maximum ones. This reflects the fact that even small
previously segmented exhibited no trend according to the impoundments constructed for various water uses such as
nonparametric tests. Further investigation of Table 8 and irrigation, municipal or industrial uses can significantly
Figure 3 shows that almost all trends taking place between change the levels of the low flows. Conversely, dams and
1960 and 1975 are downward trends. These results sug- reservoirs can be managed and operated in such a way to
gest that the trends in the Indian Subcontinent area could guarantee a residual minimal flow in the river at all times,
be explained by regional patterns since a vast majority of for navigation or ecological purposes.
stations produce similar trend patterns and these trend Also to be noted is the large magnitude of the histori-
patterns are very similar to those of the adjacent South- cal changes in levels demonstrated during the analysis by
east Asia region.
some Australian rivers. It is also apparent that on the In-
dian Subcontinent and in Southeast Asia, many rivers have
exhibited a steady downward trend starting at the end of
Central Asia
The results for the Central Asia area are very similar the 1960s until now, possibly reflecting increased water
to those observed in the Far East Asia area: The northern use for irrigation, industrialization, or municipal uses. Also,
th
stations (between the 55 and 75 parallels) exhibiting one can clearly appreciate the historical fate of the rivers
th
th
upward trends and southern stations (around the 45 par- Amu-Darya and Syr-Darya flowing into the Aral sea, but
allel) showing downward trends. These results, added to largely diverted in recent times for widespread irrigation
the results of the adjacent Far East Asia area, suggest the of cotton fields
presence of a regional trend pattern.
Table 10 synthesizes, by region, the number of occur-
rences of shifts in levels by decades, as compounded for
all the considered series (mean yearly, maximum monthly,
and minimum monthly discharges). It provides the count
Discussion
Tables 4, 5, 6, 7, 8, and 9 show that about 65 percent of series for which levels shifted during a given decade.
of the studied series exhibited no change in their mean, One can see that most of the changes occurred during the
minimum, or maximum levels during their record period. 1960s and 1970s, a period of rapid demographic expan-
One can also see that the runoff of rivers in Southeast sion, and, consequently, of the development of irrigation,
Asia have exhibited decreasing trends with time. For all especially in tropical regions.
the regions, it is also clear that minimum monthly runoffs
During the same period, a large number of dams and
IWRA, Water International, Volume 26, Number 3, September 2001

422
D. Cluis and C. Laberge
Table 9. Trend Detection Analysis Results for the Yearly Mean,
the Minimum and Maximum Monthly Discharges in Central Asia
Minimum Monthly Discharges
Mean Yearly Discharges
Maximum Monthly Discharges
Trend
Type
Period
Level*
Trend
Type
Period
Level*
Trend
Type
Period
Level*
Period
Period
1931
1958 1973
Period
1931
1961 1973 Step
Amu-Darya 1931 1957
1958 1973
Step
Step
516
180
1520
1150
3540
2730
Step
Step
Biya
1895 1985
Monotonic
46-62
1895 1910
1911 1984
425
489
No segmentation identified
Ob
1930 1994
1894 1985
Monotonic 2530-3970
No segmentation identified
56.7-80.8 No segmentation identified
No segmentation identified
No segmentation identified
1965 1990 No trend
Tom (1)
Tom (2)
Monotonic
1965 1979
1980 1990
112
149
No segmentation identified
4620
Step
Tura
1896 1985
No trend
24.1 No segmentation identified
No segmentation identified
No segmentation identified
Yenisei
1936 1969
1970 1995
3.92 1936 1972
6.6 1973 1995
17.7
18.6
Step
Step
Step
Step
Syr-Darya
Ural
1930 1964
1965 1984
334
133
1930 1960
1961 1984
673
384
1930 1960
1961 1984 Step
1310
761
1915 1953
1954 1981
38
No segmentation identified
No segmentation identified
Step
57.4
136
Naryn
1933 1990
No trend
1933 1970
1971 1990
393
316
1933 1973
1974 1990 Step
1010
703
Step
*For monotonic trends, period level represents the value estimated for the first and last year of the period.
Table 10. Counts per Decade of the Occurrence of Upward and Downward Trends in Each of the Five Areas
Decade
Region
Trend
‘00
‘10
‘20
‘30
‘40
‘50
‘60
‘70
‘80
‘90
Oceania
Pacific
Downs
Ups
2
4
3
1
3
Far East
Asia
Downs
Ups
4
1
6
2
3
6
1
2
1
1
1
2
1
Southeast
Asia
Downs
Ups
2
1
4
3
3
1
1
Indian
Subcontinent
Downs
Ups
1
1
2
2
2
1
1
1
Central
Asia
Downs
Ups
2
1
4
2
2
2
1
1
reservoirs were completed (Vörösmarty et al., 1997; (1976) and Krishna Rivers (1974, 1982, 1984) in India, the
ICOLD, 1984, 1988), modifying the historical regimes of Syrdaria (1957, 1965) and Ural Rivers (1958) in
rivers. This has been the case within the watersheds of Kazakhstan, the Yenisei (1967) and the Ob Rivers (1957)
some of the larger rivers studied here, such as the in Russia, the Narin River (1978) in Kirghiztan, and the
Murrumbidgee (1956) and Darling Rivers (1960) in Aus- Beijiang (1973) Dongjiang (1974) and Yellow or Huanghe
tralia, the Nan River (1972) in Thailand, the Godavari River (1960 and 1968) in China. Some of these impound-
IWRA, Water International, Volume 26, Number 3, September 2001

Climate Change and Trend Detection in Selected Rivers
within the Asia-Pacific Region
423
ments could be at the origin of the results presented here. tistics in environmental studies and in trend detection in
In a future study, it would be interesting to submit the time series. Dr. Laberge can be contacted at STATEX,
same data set to new statistical analyses in order to as- 3332 de la Paix, Sainte-Foy, Québec, Canada, G1X 3W6.
sess a possible regional change over time in the internal Email: claude.laberge@ sympatico.ca
variability of the runoff series as suggested by some stud-
ies on the “El Niño” phenomenon. Such a study could at-
tempt to relate monthly or seasonal runoff values to the
lagged values of the Southern Oscillation Index (SOI),
whose frequency occurrence of low and high extreme
values appears to be closely related to climate change.
Discussions open until March 1, 2002.
References
Berryman, D., B. Bobee, D. Cluis, and J. Haemmerli. 1988. “Non-
parametric Tests for Trend Detection in Water Time Series.”
Water Resources Bulletin 24, No. 3: 545–556.
Acknowledgements
Bradley, J. V. 1968: Distribution-free Statistical Tests. Englewood
Cliffs, New Jersey, USA: Prentice-Hall.
Director of the Global Runoff Data Center at Koblenz, Cluis, D. 1983: “Visual Techniques for the Detection of Water
The authors wish to express gratitude to Dr. W. Grabs,
Germany for his warm reception and to the GRDC staff Quality Trends: Double-mass Curves and Cusum Functions.”
for facilitating access to its database and helping adapt its Environmental Monitoring Assessment 3: 173–184.
format to the statistical softwares used in this study and Cluis, D. 1988: “Environmental Follow-up: A Mixed Parametric
for the scientific discussion with regards to the methods
used and the interpretation of the results. Also, heartfelt
thanks are extended to Drs. D. Kraemer and A. Askew Cluis, D., C. Langlois, R. Van Coillie, and C. Laberge. 1989. “De-
and Non-parametric Approach.” Environmental Software 3,
No. 3: 117–121.
of the Hydrology and Water Resources Department of
the World Meteorological Organization in Geneva, Swit-
zerland. This work was conducted during part of a sab-
batical leave of absence from the water research centre
velopment of a Software Package on Trend Detection in Tem-
poral Series.” Statistical Methods for the Assessment of Point
Source Pollution. Chapman and El Shaarawi, eds. CCIW:
329–341.
INRS-EAU, University of Quebec, Canada and was made Cluis, D. 1998. “Analysis of Long Runoff Series of Selected
possible by the financial support of the World Meteoro-
logical Organization (WMO).
Rivers of the Asia-Pacific Region in Relation with Climate
Change and El Ninõ Effects.” GRDC Report No.21, Koblenz,
Germany.
Doerffel, K., G. Herfurth, V. Liebich, and E. Wendlandt. 1991.
“The Shape of CUSUM - An Indicator for Tendencies in
Time Series.” Fresenius Journal Analytic Chemistry 341:
519–523.
About the Authors
Daniel Cluis earned an engi-
neering degree in hydraulics and a
doctorate in fluid mechanics from the Dagnelie P. 1970. Théorie et méthodes Statistiques, Vol 1 and 2.
University of Grenoble (France). Gembloux: Pages 378 and 451.
Since 1970, he has taken part in the Folland, C.K., T.R. Karl, and K. Ya. Vinnikov. 1990. “Observed
development of the water research
center of the University of Québec
(INRS-Eau). His fields of interest
Climate Variations and Changes.” Climate Change: the IPCC
(Intergovernmental Panel on Climate Change) Scientific As-
sessment. Cambridge: Cambridge University Press: 195–238.
include the mathematical modelling Helsel, D.R. 1987. “Advantages of Non-parametric Procedures
of contaminant transportation at the watershed scale and for Analysis of Water Quality Data.” Journal of Hydrologi-
the adaptation of statistical methods to practical environ- cal Sciences 32, No.2: 179–190.
mental problems. Dr. Cluis can be contacted at INRS- Hirsch, R.M., and J.R. Slack. 1984. “A Non-parametric Trend
Eau, University of Québec, P.O. Box 7500 Sainte-Foy,
Québec, Canada, G1V 4C7. Email daniel_cluis@inrs-eau.
uquebec.ca.
Test for Seasonal Data with Serial Dependence.” Water Re-
sources Research 20, No, 6: 727–732.
Hirsch, D.R., J.R. Slack, and R.A Smith. 1982. “Techniques for
Trend Analysis for Monthly Water Quality Data.” Water
Resources Research 18, No. 1: 107–121.
Claude Laberge earned a
bachelor’s degree in statistics and a
master’s degree in mathematics (sta- Huber, P.J. 1981. Robust Statistics. New York, New York, USA:
tistics option) from Université Laval Wiley.
and completed his Ph. D. in water Hubert, P., J.P. Carbonnel, and A. Chaouche. 1989. Segmenta-
science at INRS-Eau (branch of the
Université du Québec). He founded
STATEX , a statistical consultant
tion des séries hydrometeorologiques. Application à des
séries de précipitations et de débits de l’Afrique de l’Ouest.
Journal of Hydrol. 100: 349–367.
firm, in 1990. His domains of spe- ICOLD. 1984,1988. World Register of Dams. First edition and
cialization are the application of sta- updating. International Commission on Large Dams, Paris.
IWRA, Water International, Volume 26, Number 3, September 2001

424
D. Cluis and C. Laberge
IPCC. 1996. “Climate Change 1995.” The IPCC second scientific
assessment. J.T. Houghton, L.G. Meira Filho, B.A. Callander,
N. Harris, A. Kattenberg, K. Maskell, eds. Cambridge: Cam-
bridge University Press: 572 pages.
ences, Uncertainties, and Limitations.” Geophysical Re-
search Letters 26, No. 6: 759–762.
Montgomery, R.H. and J.C. Loftis. 1987. “Applicability of the t-
test for Detecting Trends in Water Quality Variables.” Water
Resources Bulletin 23, No. 4: 653–662.
Lehman, E.L. and H. D’Abrera. 1975. Nonparametrics: Statisti-
cal Methods Based on Ranks. San Francisco, California, USA:
Holden Day.
Scheffé, M. 1959. The Analysis of Variance. New York, New
York, USA: Wiley.
Leith, R.M.M. and P.H. Wakefield. 1998. “Evidence of Climate Van Belle, G. and J.P. Hughes. 1984. “Non-parametric Tests for
Change Effects on the Hydrology of Streams in South-cen-
tral British-Columbia.” Canadian Water Resources Journal
23, No. 3: 219–230.
Trend in Water Quality.” Water Resources Research 20, No.
1: 127–136.
Vörösmarty, C.J., K. Sharma, B. Fekete, A.H. Copeland, J. Holden,
J. Marble, and J.A. Lough. 1997. “The Storage and Aging of
Continental Runoff in Large Reservoir Systems of the World.”
Ambio 26: 210–219.
Lettenmaier, D.P. 1976. “Detection of Trends in Water Quality
Data from Records with Dependents Observations.” Water
Resources Research 12, No. 5: 1037–1046.
Matalas, N.C. and W. B. Langbein. 1962. “Information Content Wald, A. and J. Wolfowitz. 1943. “An Exact Test for Random-
of the Mean.” Journal of Geophysical Research 67, No. 9:
3441–3448.
ness in the Non-parametric Case Based on Serial Correla-
tion.” Annual of Mathematic Statistics 14: 378–388.
Mann, M.E., R.S. Bradley, and M.K. Hughes. 1999. “Northern Zhang, Z., K.D. Harvey, W.D. Hogg, and T.R. Yuzyk. In press.
Hemisphere Temperature During the Past Millennium: Infer- “Trends in Canadian Streamflow.” Water Resources Research.
IWRA, Water International, Volume 26, Number 3, September 2001