Vous êtes sur la page 1sur 11

Science of the Total Environment 557–558 (2016) 91–101

Contents lists available at ScienceDirect

Science of the Total Environment

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

Abrupt state change of river water quality (turbidity): Effect of extreme


rainfalls and typhoons
Chih-Sheng Lee ⁎, Yi-Chao Lee, Hui-Min Chiang
Department of Environmental Engineering, Kun Shan University, Tainan City 71003, Taiwan

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Evaluation of impact of typhoons on


river quality.
• First attempt using statistical methods
for abrupt state changes of turbidity.
• Daily turbidity correlated with daily flow
rate for all the eleven events studied.
• Typhoon Morakot and subsequent rain-
fall events leads to high river turbidity.

High turbidity water: (left) in a reservoir; (right) raw water in a water treatment plant.

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

Article history: River turbidity is of dynamic nature, and its stable state is significantly changed during the period of heavy rainfall
Received 25 September 2015 events. The frequent occurrence of typhoons in Taiwan has caused serious problems in drinking water treatment
Received in revised form 16 February 2016 due to extremely high turbidity. The aim of the present study is to evaluate impact of typhoons on river turbidity.
Accepted 24 February 2016
The statistical methods used included analyses of paired annual mean and standard deviation, frequency distri-
Available online xxxx
bution, and moving standard deviation, skewness, and autocorrelation; all clearly indicating significant state
Editor: D. Barcelo changes of river turbidity. Typhoon Morakot of 2009 (recorded high rainfall over 2000 mm in three days, respon-
sible for significant disaster in southern Taiwan) is assumed as a major initiated event leading to critical state
Keywords: change. In addition, increasing rate of turbidity in rainfall events is highly and positively correlated with rainfall
Rainfall intensity intensity both for pre- and post-Morakot periods. Daily turbidity is also well correlated with daily flow rate for all
Flow rate the eleven events evaluated. That implies potential prediction of river turbidity by river flow rate during rainfall
Turbidity and typhoon events. Based on analysis of stable state changes, more effective regulations for better basin man-
Frequency distribution agement including soil-water conservation in watershed are necessary. Furthermore, municipal and industrial
Moving skewness/autocorrelation
water treatment plants need to prepare and ensure the adequate operation of water treatment with high raw
water turbidity (e.g., N 2000 NTU). Finally, methodology used in the present of this study can be applied to
other environmental problems with abrupt state changes.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction

Challenges of river basin management are shifting, from pollution


⁎ Corresponding author. sources and control (e.g., total maximum daily loads and assimilative
E-mail address: cslee@mail.ksu.edu.tw (C.-S. Lee). capacity analysis (Lee and Chang, 2005), non-point sources

http://dx.doi.org/10.1016/j.scitotenv.2016.02.213
0048-9697/© 2016 Elsevier B.V. All rights reserved.
92 C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101

management (Zushi et al., 2008)) to dealing with extreme rainfalls and uses. Location of the Gaoping weir close by meteorological station is
their effects, such as runoff (Frans et al., 2013) and flood (Hirabayashi shown in Fig. 1.
et al., 2013; Vaghefi et al., 2014). In particular, global warming may Three relevant time-series datasets were collected: rainfall, river
lead to extreme events of rainfall with changing local rainfall duration flow rate, and river turbidity. All were in the period of Jan 1, 2003 to
and intensity (IPCC, 2007; Wang et al., 2013). In Taiwan, the rainfall Dec 31, 2013, a span of 11-yr daily data with each over 4000 daily re-
trend of shorter duration and higher intensity is clearly evident by the cords. Daily rainfall data (Fig. 2a) were collected from Shipu meteoro-
analysis of daily rainfall record from 1950s to 2000s (Jhong et al., logical station of Central Weather Bureau, near the Gaoping river weir.
2009). Moreover, the probability of extreme rainfall events and crisis In addition, for heavy rainfall events, hourly rainfall data were collected
of water scarcity is increasing in southern Taiwan with uneven distribu- to determine rainfall intensity (total rainfall in an effective rainfall event
tion of rainfall (Jhong et al., 2009). Since 2009, problems of high turbid- divided by the rainfall duration). For example, an effective rainfall starts
ity (N2000 NTU) have caused significant concern to the public and 20:10 and ends next day 6:00, then the duration in dominator of rainfall
regulatory agencies in the Gaoping river. In Taiwan, rainfalls almost con- intensity is only 8 h. In the present study, effective rainfall refers to rain-
centrate in two seasons, plum season from April to May and typhoon fall more than 0.5 mm/h. Daily flow rate and turbidity data monitored at
season from June to September. The high rainfall intensity with shorter the Gaoping river weir (Fig. 2b and 2c) were obtained from Southern
duration (e.g., 15 mm during 1-h period) in rainy seasons certainly ex- Water Resources Bureau. River flow rate is continuously monitored
ceeds river capacity causing flood while significant amounts of water while river turbidity monitored daily at 7–8 am with a laboratory turbi-
are flowing into ocean. dimeter (HACH 2100N used in 2003–2011 and 2100AN in 2012 with
Therefore, in the past decades, the main problem facing river quality maximum detected limits: 4000 and 10,000 NTU, respectively). It is
management associated with heavy rainfalls mostly causing by ty- noted that the peak turbidity and peak daily flow rate does not occur
phoons is its high turbidity in river, due to landslide and erosion. In at the same time. Basically, the cyclic data variations of higher rainfall/
fact, erosion rate in the Gaoping river basin has been reported as flow rate/turbidity are due to typhoon events and other extremey
10–30 mm/yr (Dadson et al., 2003). Extremely high turbidity (more high ranfall events in rainy searsons. Since typhoon Morakot of 2009
than 6000 NTU) may lead to water treatment plant shutdown. Particu- was the major typhoon causing serious damages (Kao et al., 2012;
larly, in summer of 2009, typhoon Morakot (extremely high rainfall Huang and Montgomery, 2013), therefore in this study, two periods
over 2000 mm in three days (Aug 8–10)) caused serious disaster with are divided as pre-Morakot (2003–2008) and post-Morakot
complete shutdown of many water treatment plants. In general, turbid- (2009–2013) periods including Morakot and subsequent typhoons.
ity is caused by colloids; however such high turbidity during high rain-
fall events in Taiwan is due to silt, clay and suspended solids by 2.2. Statistical analysis
interaction of rainfall, upstream erosion and in-stream re-rolling sedi-
ment. In short, it is worthwhile investigating the impact of Morakot For better illustration of typhoon impact on river quality, statistical
and subsequent typhoons on river water quality (turbidity) in southern approaches (e.g., paired annual mean and standard deviation (SD), fre-
Taiwan. quency distribution, moving SD, skewness and autocorrelation) were
In particular, one needs to see how turbidity changes at the onset of, used in this study. Annual mean and SD of river turbidity are a paired in-
during and after the typhoon event. Identification of changes in turbid- dicator which can easily be analyzed and understood regarding the
ity pattern is of importance since it can reveal how previous stable states change of state at a yearly scale. Frequency distribution is an effective
have been changed during an extreme rainfall event, and assist water approach that can find out the extreme data from its long-tail pattern
treatment plants to plan strategies against sudden shutdown. This is for long-term data. Moving SD/skewness/autocorrelation are useful
analogous to perturbation caused by external elements and once exter- techniques that can identify the accumulated effects of rainfall events.
nal element is removed, it is slowly returned to the original state in eco-
system (Hirota et al., 2011; Veraart et al., 2012), lake eutrophication 2.2.1. Skewness and autocorrelation
(Hargeby et al., 2007; van Nes et al., 2007) and pollutant in rive Skewness is a statistic indicator to measure shape of data distribu-
(Tsuzuki, 2015). For example, Hargeby et al. (2007) reported two tion. For example, normal distribution exhibits a symmetric bell shape,
lakes in southern Sweden shifting repeatedly between clear water indicating that skewness is exactly equal to zero. However, when skew-
state and turbid water state in 1970–2004. They reported accumulation ness N 0, meaning distribution of skewing right, it is defined as positive
of nutrients leading to change of stable states. skewness. The larger positive skewness is, the more shape of skewing
In view of frequent occurrence of typhoons in Taiwan and their im- right becomes; and skewness b0 vice versa. Change of skewness will re-
pact on river water quality, this study was undertaken to address the veal the transition of stable states while comparing with its present
following questions: (1) What is the effect of Morakot and subsequent state (Chatfield, 2004; Dakos et al., 2012). Correlation coefficient is an-
typhoons on river turbidity? (2) Is the rate of increasing turbidity corre- other statistic indicator to identify the relationship of datasets x and y.
lated with rainfall intensity? (3) How long will it take to reach the orig- Similarly, autocorrelation is a statistic coefficient to identify the correla-
inal pre-typhoon state? And (4) what is the impact of subsequent tion of dataset x itself with different time interval. The range of autocor-
typhoons after Morakot on river turbidity. The results will be useful relation is in [−1, +1]. For example, autocorrelation approaching 1.0
for more effective regulations for soil-water conservation in watershed means data sparse (i.e., distributed wide range of values indicating
and adequate operation of water treatment plants with high source high variation) with positive correlation; autocorrelation approaching
water turbidity. To the best of knowledge, this is the first attempt to il- −1.0 indicating high variation with negative correlation; and autocor-
lustrate the cumulative effect of typhoons on river water quality relation approaching zero concentrating in a narrow range (Chatfield,
(turbidity). 2004; Ren and Watts, 2015).
For all n observations of data, xi, i = 1, 2, …, n. The SD is the root of
2. Materials and methods variance, given by

2.1. Gaoping river weir and watershed information


rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 n
The length of the Gaoping river is 171 km, the second longest one in SD ¼ ∑ ðx  xÞ2 ð1Þ
n i¼1 i
Taiwan, covering watershed area of 3257 km2. The upstream of Gaoping
weir was used for obtaining turbidity data. The weir is used for with-
drawing river water as source water for domestic and industrial water then, the skewness is defined as the ratio of the third momentum (m3)
C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101 93

Fig. 1. Map of the Gaoping river weir and watershed of the Gaoping river.

to the second momentum (m2) as: and the second 500-d moving SD (SD of days 2–501) as

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 n 1 501
∑ ðxi  xÞ3 SD ¼ ∑ ðx  xÞ2 ð5Þ
m3 n i¼1 500 i¼2 i
Skewness ¼ 3=2
¼h i3=2 ð2Þ
m2 n 2
n ∑i¼1 ðxi  xÞ
1
and so forth. All 500-d moving SD can be plotted with time for identify-
ing the trend of data. Calculation of moving skewness and autocorrela-
tion is similar as in Eqs. 4 and 5 for determining moving SD.
the lag-1 autocorrelation is defined as the correlation of the first n (i =
1, 2, ……, n) observed data and the next n (i = 2, 3, ……, n + 1) ob-
2.2.2. Frequency distribution
served data:
Observation of data can be feasibly assorted as m groups with ki data
in the i-th group (i = 1, 2, …, m). The frequency distribution for all ki can
n
∑i¼1 ðxi  x1 Þðxiþ1  x2 Þ be represented by a probability distribution as follows (Hirota et al.,
Autocorrelation ¼ h i1=2 h i1=2 ð3Þ 2011; Lee and Su, 2014):
n n
∑i¼1 ðxi  x1 Þ2 ∑i¼1 ðxiþ1  x2 Þ2
ki
Pðki Þ ¼ m  100%: ð6Þ
∑i¼1 ki
where x1 and x2 are means of the first n data and the next n data,
respectively.
Furthermore, for a long term time-series data, moving SD/skewness/ P(ki) is the probability of the group i; higher probability means high
autocorrelation is of use for evaluating the trend of data. In this study, frequency. The relationship of P(ki) to group i can help distinguish the
the trend of data is used to identify the cumulated effects of typhoons type of data distribution. The distribution can identify not only higher
and heavy rainfall events. Moving SD/skewness/autocorrelation is sim- peak of the group data, but also the profile of the whole group of data.
ilar to the concept of statistically moving average (Chatfield, 2004). For Theoretically, a short-tail distribution indicates that only few data be-
example, the first 500-d moving SD (meaning SD of days 1–500) of daily long to higher group i, meaning low frequency/probability of the ex-
long-term time series data is given by tremely large data. On the other hand, long-tail pattern shows
probability of extremely high data; the longer the tail is, the more ex-
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi treme of the dataset is. For example, in the case of high rainfall intensity
1 500 causing extremely high turbidity in river, the frequency of extremely
SD ¼ ∑ ðx  xÞ2 ð4Þ
500 i¼1 i high turbidity leading to long-tail distribution in the plot of frequency
94 C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101

Fig. 2. (a) Daily precipitation in period of 2003–2013, (b) daily flowrate, and (c) daily turbidity (arrows indicate typhoon Morakot of 2009).

vs. turbidity. The change of stable states can then be identified from starting point, SP), then leads to sharp increase to the highest value
shifting of shot tail to long-tail. (critical point, CP), and then rapidly decrease (Fig. 3a, as ending point,
EP) or took a long time (Fig. 3b) back to the original stable state. The en-
2.2.3. Increasing and recovery rates of turbidity tire transient state from SP to CP and back to EP can be characterized as
There are two typical types of change on river turbidity caused by increasing rate (IR) and recovery rate (RR), respectively. In this study, IR
rainfall events as shown in Fig. 3. Type I (Fig. 3a) clearly shows a sharp and RR are calculated from the following equations for significant rain-
increase of turbidity from the original state, reaching the peak value falls event:
and then returning to the original turbidity level just after several
ln ðCP=SPÞ
days. On the other hand, Type II (Fig. 3b) first also shows a sharp in- IR ¼ ð7Þ
t
crease of turbidity from the original state, but turbidity then kept de-
creasing and fluctuating: the turbidity does not return to the original − ln ðEP=CPÞ
state after 15 d of termination of rainfall event. Instead it may establish RR ¼ : ð8Þ
t
another pseudo state with turbidity far above the original pre-typhoon
level or even could not reach any stable state. It may take long time to It is noted that the duration between SP and EP for each event may
reach the original turbidity state or even cannot attain the pre- be different. It depends on a particular rainfall event based on the defi-
typhoon turbidity. nition of IR and RR. In addition, linear regression was used for several
Several points identified in Fig. 3 are straightforward and used for points instead of data from merely two points. To the best of our knowl-
calculating increasing and recovery rates of turbidity. High rainfall in- edge, this is the 1st attempt to use the term of IR in correlation with
tensity gradually increases turbidity concentration (identifying as other parameters to be discussed below.
C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101 95

Fig. 3. Cases of state change on river turbidity: (a) Type I: back to original stable state (e.g., event I in Table 1); and (b) Type II: change to new stable state (e.g., event VI in Table 1).

3. Results and discussion monitoring data (EPA of Taiwan, 2015) for the Lining Bridge monitoring
station (immediately upstream of the Gaoping weir) show average
3.1. Changes of paired annual mean and SD in river turbidity suspended solids (highly correlated with turbidity) had been greatly in-
creased from 8 mg/L in 2003 to 920 mg/L in 2013. In short, the long term
Total annual precipitation of rainfall is shown in Fig. 4a with the an- impact of Morakot along with subsequent typhoons is significant to ac-
nual mean and SD of river turbidity in Fig. 4b. For better illustration, data count for increased turbidity during the post-Morakot period.
for pre- and post-Morakot periods are separated. It shows that annual
rainfall of pre-Morakot period (2003–2009) varies between 1500 and 3.2. Shift of frequency distribution for river turbidity
3500 mm/yr with mean and SD of 2410 ± 740 mm/yr and that of
post-Morakot period (2009–2013) is approximately the same (mean For analysis of frequency distribution, five types of turbidity range
2400 mm/yr) but with SD of ±470 mm/yr. Clearly SD has been signifi- were used (Fig. 5) as: (1) adequate: below 100 NTU; (2) fair:
cantly reduced from pre- to post-Morakot period, while mean precipita- 100–1000 NTU; (3) high: 1000–6000 NTU; (4) extremely high:
tion remained the same, meaning lower deviation of rainfall for post- 6000–10,000 NTU; and (5) unacceptable: above 10,000 NTU. Adequate
Morakot. On the other hand, turbidity data illustrate both high mean type is suitable as raw water source in water treatment plants, and fair
and SD at post-Morakot period, as compared to both low mean and type requires the flexible operation in water treatment plants. In
SD at pre-Morakot period (Fig. 4b). For example, annual mean of turbid- Taiwan, the turbidity of 6000 NTU is one of the criteria for plant shut-
ity is 680 NTU during pre-Morakot period had been significantly in- down, thus high type of 1000–6000 NTU provides a significant warning
creased to 2100 NTU for post-Morakot period with SD from 1680 NTU to treatment plants. The historical highest turbidity values for pre- and
of pre-Morakot period sharply increasing to 4650 NTU of post- post-Morakot periods were 30,000 and 57,000 NTU, respectively. In
Morakot period. Clearly, the extremely high turbidity values Fig. 5, frequency of adequate turbidity of pre-Morakot is over 50% and
(e.g., N10.000 NTU) caused by typhoons are responsible for such high post-Morakot only 9% which is significantly different between these
average of turbidity of 2100 NTU. Moreover, regular water quality two periods. On the other hand, 5.4% of extremely high turbidity and
96 C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101

Fig. 4. (a) Annual mean of rainfall and (b) annual mean of turbidity at the periods of 2003–2008 (solid block) and 2009–2013 (empty block) with standard deviation (line) (arrows
indicate typhoon Morakot of 2009).

4.5% of unacceptable turbidity at post-Morakot period are much higher In brief, distribution of frequency clearly shows long-tail effect for
in comparison with those of pre-Morakot period (1.3% and 0.7%, respec- post-Morakot period (Fig. 5), also as evidence of existence of extremely
tively). The shifted frequency distribution clearly indicates that in the high turbidity. Again, frequency distribution shows states of turbidity
period of post-Morakot, higher turbidity leads to lower security for change from short-tail to long-tail through pre- to post-Morakot period.
water treatment plant and may consequently induce risk of water treat- In the climate system, for example, long-tail frequency distribution of
ment plant shutdown. events maybe a sign for extreme climate (Lenton, 2011).

3.3. Trends of moving SD, skewness and autocorrelation: Cumulated effects


of rainfall events

Moving averages of SD, skewness and autocorrelation have been


widely applied to explain critical transition, such as in ecological system
(Wang et al., 2012) and eutrophication of lake (Dakos et al., 2013). This
study has tried various moving days (300, 500, 1000, 1200, 1500, and
2000 d) for determining these statistical parameters. The results for all
the moving days data consistently show the cumulated effect for post-
Morakot period (Fig. 6 and Fig. SI-2). However, the results indicate
that the shorter the moving days is the more variations the pattern
will exhibit (see curves in Fig. 6). The variation of curves can demon-
strate more clearly the state of changes. Also it will be meaningful for
the moving days over a year long cycle (365 d). Therefore, 500-d data
were selected to explain the cumulated effect for post-Morakot period
(Fig. 6).
Fig. 5. Trubidity frequency distribution (solid block: pre-Morakot, empty block: post- For 500-d moving SD (Fig. 6a), it is clear that SD for different periods
Morakot). are different; from lower SD for pre-Morakot period (1200–3000, mean
C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101 97

Fig. 6. Turbidity for 500-day moving (a) standard deviation, (b) skewness, and (c) autocorrelation (arrows indicate typhoon Morakot of 2009; solid line is mean of pre-Morakot and dash
line mean of post-Morakot; square marks indicate typhoon/rainfall events for pre-Morakot and circle marks for post-Morakot).

1760) increasing to much higher for post-Morakot period (3000–6000, Morakot significantly raises SD from 1800 to 5000, and then keeps
mean 4580). This is consistent with previous discussion about paired 5000 more than six months till next significant rainfall event (event
mean and SD. In addition, after typhoon Morakot, SD remains higher VII) (Fig. 6a). Skewness is from 6 increasing to 14, keeping about 8 till
than that of pre-Morakot period implying cumulated effect of Morakot event VII (Fig. 6b), with autocorrelation from 0.4 up to 0.8, remained
and subsequent typhoons. Both skewness and autocorrelation also at 0.8 (Fig. 6c). Comparing with other typhoons and rainfall events, in-
have similar changes between pre- and post-Morakot periods (Figs. 6b creases in SD, skewness and autocorrelation for typhoon Morakot result
and 6c), but not as significant as SD. For example, 500-d skewness from its high rainfall intensity 16.5 mm/h (instantaneous rainfall inten-
(Fig. 6b) is from 4 to 12 (mean 6.1) for pre-Morakot period down to sity 86 mm/h, defined as maximum hourly rainfall in a rainfall event)
3–14 (mean 5.2) for post-Morakot period. Particularly, right skewed and high peak turbidity 57,000 NTU as well as long days (105 d) to
distribution results from extremely high turbidity in the period of reach pseudo state (Table 1). The exact role of typhoon Morakot of
post-Morakot. This is also consistent with previous discussion of fre- 2009 causing the subsequent high river turbidity remains to be seen.
quency distribution. Higher 500-d autocorrelation for post-Morakot pe-
riod (0.55–0.85, mean 0.72; Fig. 6c) also indicates more wide-range 3.4. Characteristics of event-based extreme rainfalls
distribution and higher variation of river turbidity than those for pre-
Morakot period (0.40–0.85, mean 0.68). In brief, the results clearly pres- As mentioned before, in Taiwan, typhoon season is from June to Sep-
ent cumulated effect of post-Morakot period (including Morakot and tember, and Plum season (pre-typhoon season) is from April to May. Al-
subsequent typhoons) on river turbidity. It is noted that the sharp in- most all extreme rainfalls appear in plum and typhoon seasons. During
creases of Fig. 6a–6c at the intersection of pre- and post-Morakot pe- 2003 to 2013, eleven significant rainfall events were analyzed (Table 1
riods are caused by the extreme high turbidity during typhoon and in Supplementary information Fig. SI-1) with eight events brought
Morakot, in Aug 8–10, 2009 (51,000, 52,000 and 57,000 NTU for the by typhoons. The three non-typhoon induced events were possibily
three days, respectively). These phenomena occurred in the results of caused by strongly advective clouds in the plum season (CWB of
all other moving days (Fig. SI-2). Taiwan, 2015) and all appeared in the period of post-Morakot period.
Further analysis of moving SD/skewness/autocorrelation can explain Note that there were two typhoons named Morakot in 2003 and 2009
which typhoon/rainfall exerts a major effect. The reason is that sharp in- (in this study, the typhoon Morakot is specified as Morakot of 2009).
crease/decrease of river turbidity may result in significant changes of In particular, events II/III, VII/VIII, and X/XI are three dual events, refer-
500-d SD/skewness/autocorrelation. Ten typhoon/rainfall events ring two independent events appearing one after another.
(events II–XI, Table 1) are marked in Fig. 6 for better illustration. Note As shown in Table 1, rainfall intensity varied from 6.3–15.7 mm/h
that event I is excluded because of not enough data for calculating (instantaneous rainfall intensity 20–70 mm/h) of pre-Morakot period
500-d SD/skewness/autocorrelation. It clearly shows that typhoon slightly extending to 3.5–19 mm/h (instantaneous rainfall intensity
98 C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101

Table 1
Typhoon-induced extreme rainfall and related high turbidity.

Rainfall Peak Turbidity Turbidity Day to pseudo state (d)


Flow rate (at peak Type of state of
Typhoon event intensity turbidity increasing recovery New Pseudo-original
turbidity) (m3/s) change
(mm/h) (NTU) rate (1/d) rate (1/d) state state

I. Morakota
13.3 11,000 730 4.6 2.3 – 7 I
(Aug 1–10, 2003)
II. Mindulle
6.3 24,700 15,300 2.6 0.5 4 90 II
(Jun 28–Jul 11, 2004)
III. Kompasu
4.8 4000 55 0.4 0.5 – 2 I
(Jul 15–22, 2004)
IV. Haitang
12.2 30,390 11,850 5.8 0.5 6 90 II
(Jul 12–24, 2005)
V. Kalmaeg
15.7 19,800 4330 4.6 0.5 8 25 II
(Jul 14–21, 2008)
VI. Morakot
16.5 57,000 9055 2.6 0.3 15 105 II
(Aug 4–14, 2009)
VII. No typhoon
7.8 30,000 1740 3.0 0.6 3 45 II
(May 21–26, 2010)
VIII. No typhoon
3.5 28,000 2300 0.5 0.5 – 6 I
(May 26–Jun 3, 2010)
IX. Fanapi
19.0 16,000 11,160 2.9 0.1 3 25 II
(Sep. 18–22, 2010)
X. No typhoon
7.8 50,000 12,400 1.8 0.4 7 25 II
(Jun 7–16,2012)
XI. Talim
6.4 27,000 5770 0.6 0.5 – 2 I
(Jun 17–26, 2012)
a
The same name as event VI since the name of typhoon is listed repeatedly by the committee of World Meteorological Organization.

Fig. 7. Relation of increasing rate and recovery rate for turbidity. (a) Rainfall intensity vs increasing rate of turbidity; and (b) flow rate (at the day of peak turbidity) vs recovery rate of
turbidity (solid circle and solid line: pre-Morakot, empty circle and dash line: post-Morakot).
C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101 99

Fig. 8. (a) Peak turbidity vs flow rate (solid circle and solid line: pre-Morakot, empty circle and dash line: post-Morakot); and (b) turbidity vs flow rate (empty circle: peak turbidity on the
rainfall events).

15.5–100 mm/h) of post-Morakot period. For example, event I recorded geomorophology may more or less lead to the contradiction. Also,
total 40 mm with duration of 3 h (3–6 pm of Aug 2, 2003) and instanta- flow rates (at the days of peak turbidity, Table 1) of pre-Morakot period
neous rainfall intensity 39.5 mm/h (recorded at 3 pm of Aug 2, 2003). ranged from 55 to 15,300 m3/s and changed to 1740–12,400 m3/s for
Rainfall duration is an important factor for rainfall intensity. For com- post-Morakot period. Flow rate profile of all 11 rainfall events is
parison, the rainfall intensity is 55 mm/h for 10-yr sewer design based shown in Fig. SI-1. The effects of flow rate on river turbity and RR will
on 2-h duration in Taipei, the capital of Taiwan. However, the highest be discussed later.
peak turbidity of five events at pre-Morakot period was only once CP varied from 4000 to 30,390 NTU of pre-Morakot period and sig-
over 30,000 NTU (event IV), but for post-Morakot period once over nificantly increased to 16,000–57,000 NTU in the post-Morakot period.
30,000 NTU (event VII), and even twice over 50,000 NTU (events VI EP also exhibites the same trend as CP, from 105 to 2850 NTU of pre-
and X). High turbidity events were all induced by typhoons in pre- Morakot period increasing to 5600–12,400 NTU of post-Morakot period.
Morakot period; however, in post-Morakot period, three high turbidity The SP, CP and EP of all 11 events are shown in Fig. SI-1. In particular,
events happened in plum seasons, not in typhoon seasons (events VII, event III (dual event followed by event II) exerts a low peak turbidity
VIII and X). In particular, even small rainfalls (but with significantly (4000 NTU) and low flow rate (55 m3/s), induced by typhoon Kompasu
high instantaneous rainfall intensity) also can cause high turbidity which is not an extreme rainfall.
(e.g., event VIII, 3.5 mm/h for rainfall intensity and 15.5 mm/h for in- IR and RR, calculated by Eqs. 6 and 7 are shown in Table 1 and Fig. SI-
stantaneous rainfall intensity). That implies that rainfall intensity 1; higher IR (larger than 3.0, 1/d) appears in pre-Morakot period, re-
plays an important role for high turbidity events. For example, rainfall sponsible for 60% of five events (0.4–5.8, 1/d). However, all IR are less
intensity of events VII and X (both 7.8 mm/h with higher turbidity) than or equal to 3.0, 1/d in post-Morakot period (six events, 0.5–3.0,
was higher than that of events II, III, and XI (6.3, 4.8 and 6.4 mm/h 1/d). It seems interesting that higher CPs appears in post-Morakot pe-
with lower turbidity). However, many other factors also affect observed riod but higher IR in pre-Morakot period.
high turbidity in river. For example, event IX (typhoon Fanapi) had Finally, “day to pseudo-original state” is affected by flow rate and IR
higher rainfall inmtensity but with lower peak turbidity. In fact, in the (Table 1). Both higher flow rate (more than 1000 m3/s) and IR (more
subsequent analysis (Sec. 5), it has been found that peak turbidity is than 1.0 1/d) resulted in prolonged period to reach “pseudo-original
weakly correlated with rainfall intensity. In addition, change of state” (over 25 d). In particular, typhoon Morakot of 2009 took 105 d
100 C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101

to reach the original pre-typhoon state. On the other hand, for those databases of turbidity, rainfall and flow rate. The state change of turbid-
with lower flow rate (b1000 m3/s) or lower IR (b1.0, 1/d), days to ity can be easily observed by using paired annual mean and SD, fre-
“pseudo-original state” are shorter (b 7 d). quency distribution and moving SD/skeweness/autocorrelation.
Increasing rate of turbidity in rainfall events is correlated with rainfall
3.5. Effect of rainfall intensity and flow rate intensity both for pre- and post-Morakot period. Daily turbidity is also
well correlated with daily flow rate for all the eleven events. Cumulated
This study further evaluates the effects of rainfall intensity and flow effect of post-Morakot (including Morakot and subsequent typhoons)
rate on river turbidity. Theorectically, rainfall intensity may directly in- clearly leads to high river turbidity in the Gaoping river.
duce erosion resulting in high turbidity runoff flowing into rivers. In this Prediction of river turbidity becomes possible as good correlation ex-
study, rainfall intensity has good correlation with IR (Fig. 7a), with ists between IR and rainfall intensity, as well as between RR and flow
higher correlation r = 0.86 for pre-Morakot period, and r = 0.74 for rate. Understanding types and characteristics of critical state change,
post-Morakot. The reason that correlation corfficient is not as good as as well as effects of rainfall intensity and flow rate, is essential for better
one would expect may be due to errors in calculating rainfall intensity, simulation and prediction of river turbidity. That will be beneficial for
the location of meteorological station, and few sampling points in deter- river basin management. Hopefully, the results of the present study
mining IR. It is thought that the high rainfall intensity may cause higher may shed some light as to change and trend of deteriorating environ-
river turbidity in rainfall events. However, peak turbidity is weakly cor- mental quality and assist regulatory agency in policy-making for con-
related with rainfall intensity (data not shown). trolling river turbidity problems.
For discussing flow rate, data of event III with very low flow rate
(55 m3/s) were discarded. Peak turbidity has high correlation with Acknowledgment
flow rate (Fig. 8a) with higher correlation r = 0.85 for pre-Morakot pe-
riod, but no correlation for post-Morakot period. The reason may be due Southern Water Resources Bureau of Taiwan providing river turbid-
to the fact that time for occurrence of peak turbidity and flow rate may ity and flow rate data is acknowledged.
not be the same. Flow rate (at the day of peak turbidity) also has nega-
tive correlation with RR (Fig. 7b), with r = 0.72 for pre-Morakot, and Appendix A. Supplementary data
r = 0.80 for post-Morakot. Similarly as for the correlation of rainfall in-
tensity and IR, the correlation corfficient is not as good as one would ex- Supplementary data to this article can be found online at http://dx.
pect and it may be due to errors in few smapling points in determining doi.org/10.1016/j.scitotenv.2016.02.213.
RR. Implication of the relationship indicates high flow rate leads to
lower RR with gradual decrease of river turbidity. This may raise the
References
risk of shutdown of water treatment plants if high turbidity
(N6000 NTU) (Fig. SI-3) still persists. Chatfield, C., 2004. The Analysis of Time Series: An Introduction. sixth ed. New York, NY,
Not surprising, daily turbidity has high correlation with daily flow Chapman & Hall/CRC.
Dadson, S.J., Hovius, N., Chen, H., Dade, W.B., Hsieh, M.-L., Willett, S.D., Hu, J.-C., Horng, M.-
rate, r = 0.70 for the total eleven typhoon and rainfall events J., Chen, M.-C., Stark, C.P., Lague, D., Lin, J.-C., 2003. Links between erosin, runoff var-
(Fig. 8b). For practical application, the relationship between flow rate iability and seismicity in the Taiwan orogen. Nature 426, 648–651.
and turbidity may assist the prediction of daily turbidity from daily Dakos, V., Carpenter, S.R., Brock, W.A., Ellison, A.M., Guttal, V., Ives, A.R., Kefi, S., Livina, V.,
Seekell, D.A., van Nes, E.H., Scheffer, M., 2012. Methods for detecting early warnings
flow rate. Turbidity forecasting models using statistical regression and of critical transitions in time series illustrated using simulated ecological data. PLoS
artificial neural networks have been used (Mukundan et al., 2013; One 7 (7), e41010.
Iglesias et al., 2014; Yang et al., 2014). For example, Mukundan et al. Dakos, V., van Nes, E.H., Scheffer, M., 2013. Flickering as an early warning signal. Theor.
Ecol. 6 (3), 309–317.
(2013) established regression correlation of event mean storm turbidity Frans, C., Istanbulluoglu, E., Mishra, V., Munoz-Arriola, F., Lettenmaier, D.P., 2013. Are cli-
and stream discharge in New York City. matic or land cover changes the dominant cause of runoff trends in the Upper Missis-
In Taiwan, typhoons bring significant amounts of rainfall resulting in sippi River Basin? Geophys. Res. Lett. 40 (6), 1104–1110.
Hargeby, A., Blindow, I., Andersson, G., 2007. Long-term patterns of shifts between clear
landslides and changes of geomorphology, and consequently contribute
and turbid states in Lake Krankesjon and Lake Takern. Ecosystems 10 (1), 29–36.
to increased turbidity in rivers. In the present study, the changes of geo- Hirabayashi, Y., Mahendran, R., Koirala, S., Konoshima, L., Yamazaki, D., Watanabe, S., Kim,
morphology were not monitored. However, the previous studies may K., Kanae, S., 2013. Global flood risk under climate change. Nat. Clim. Chang. 3,
816–821.
shed some light regarding the geomorphology changes due to typhoon.
Hirota, M., Holmgren, M., van Nes, E.H., Scheffer, M., 2011. Global resilience of tropical for-
For example, Tsai et al. (2010) utilized remote sensing and spatial anal- est and savanna to critical transitions. Science 334 (6053), 232–235.
ysis for assessment of landslides. They calculated the quantity of land- Huang, M.Y.-F., Montgomery, D.R., 2013. Altered regional sediment transport regime after
slide after typhoon Morakot in the region of southern Taiwan to be a large typhoon, southern Taiwan. Geology 41 (12), 1223–1226.
Iglesias, C., Martinez Torres, J., Garcia Nieto, P.J., Alonso Fernandez, J.R., Diaz Muniz, C.,
22,590 ha of area. Tsou et al. (2011) reported 25 × 106 m3 of sediment Pineiro, J.I., 2014. Turbidity prediction in a river basin by using artificial neural net-
caused from the landslide event of Shaolin village during typhoon works: a case study in northern Spain. Water Resour. Manag. 28 (2), 319–331.
Morakot. Lee and Fei (2011) calculated rainfall induced landslide quan- IPCC, 2007. Climate Change 2007: The Physical Science Basis. Contribution of Working
Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate
tity in the Gaoping river basin with 29.4, 57.6, and 192.8 million tons for Change. Cambridge University Press, Cambridge, UK.
typhoons Mindulle (in 2004), Haitang (in 2005), and Morakot (in Jhong, Y.-D., Kuo, C.-C., Chen, C.-S., 2009. The temporal variation of regional rainfall char-
2009), respectively. Also, Tseng showed that decreases of both slope acteristics in Taiwan. J. Taiwan Agric. Eng. 55, 1–18 (in Chinese).
Kao, C.M., Yang, B.M., Lee, M.S., Liu, S.F., 2012. Environmental assessment and restoration
of terrain and surface roughness on disturbance area were about 70% of typhoon Morakot disaster: a case study in Kaohsiung, Chinese Taipei. In K. Otto-
after event of typhoon Morakot. The most influenced factor for geomor- Zimmermann ed., Resilient Cities 2: Local Sustainability Volume 2, 75–87. Springer
phology change of mountain regions in Taiwan is the maximum daily Verlag: Dordrecht, Germany.
Lee, C.-S., Chang, S.-P., 2005. Interactive fuzzy optimization for an economic and environ-
rainfall, flowed by land use and the lithology (Shih, 2014). Thus, soil
mental balance in a river system. Water Res. 39 (1), 221–231.
conservation and sustainable land-use basin management, such as bal- Lee, C.-T., Fei, L.-U., 2011. Disaster analysis of landslides and sandstorms in Kao-Ping River
ance between economic development and water-quality protection, are basin. J. Prof. Eng. 58 (3), 30–45 (in Chinese).
Lee, C.-S., Su, P.-J., 2014. Tracking sinks of atmospheric methane using small world net-
essential to reduce typhoon induced soil erosion and landslides.
works. Chemosphere 117, 766–773.
Lenton, T.M., 2011. Early warning of climate tipping points. Nat. Clim. Chang. 1, 201–209.
4. Conclusions Mukundan, R., Pierson, D.C., Schneiderman, E.M., O'Donnell, D.M., Pradhanang, S.M., Zion,
M.S., Matonse, A.H., 2013. Factors affecting storm event turbidity in a New York City
water supply stream. Catena 107, 80–88.
In this study, critical state change of deteriorating water quality (as Ren, H., Watts, D., 2015. Early warning signals for critical transitions in power systems.
turbidity) was demonstrated by analyzing long term time-series Electr. Power Syst. 124, 173–180.
C.-S. Lee et al. / Science of the Total Environment 557–558 (2016) 91–101 101

Shih, H.-Y., 2014. Analyze Geomorphology Changes in Mountain Regions of Taiwan with Veraart, A.J., Faassen, E.J., Dakos, V., van Nes, E.H., Lürling, M., Scheffer, M., 2012. Recovery
Remote Sensing Image. National Cheng Kung University, Master thesis of Department rates reflected distance to a tipping point in a living system. Nature 481, 357–359.
of Resources Engineering (in Chinese). Wang, R., Dearing, J.A., Langdon, P.G., Zhang, E., Yang, X., Dakos, V., Scheffer, M., 2012.
Tsai, F., Hwang, J.–H., Chen, L.-C., Lin, T.-H., 2010. Post-disaster assessment of landslides in Flickering gives early warning signals of a critical transition to a eutrophic lake
southern Taiwan after 2009 Typhoon Morakot using remote sensing and spatial anal- state. Nature 492, 419–422.
ysis. Nat. Hazards Earth Syst. Sci. 11(3), 759–764. Wang, D., Hagen, S.C., Alizad, K., 2013. Climate change impact and uncertainty analysis of
Tsou, C.-Y., Feng, Z.-Y., Chigira, M., 2011. Catastrophic landslide induced by Typhoon extreme rainfall events in the Apalachicola River basin. Fla. J. Hydrol. 480, 125–135.
Morakot, Shaolin, Taiwan. Geomorphology 127 (3–4), 166–178. Yang, T.-M., Fan, S.-K., Fan, C., Hsu, N.-S., 2014. Establishment of turbidity forecasting
Tsuzuki, Y., 2015. Relationships between pollutant discharge and water quality in the riv- model and early-warning system for source water turbidity management using
ers from “better” to “worse” water quality. Ecol. Indic. 52, 256–269. back-propagation artificial neural network algorithm and probability analysis. Envi-
Vaghefi, S.A., Mousavi, S.J., Abbaspour, K.C., Srinivasan, R., Yang, H., 2014. Analyses of the ron. Monit. Assess. 186 (8), 4925–4935.
impact of climate change on water resources components, drought and wheat yield Zushi, Y., Takeda, T., Masunaga, S., 2008. Existence of nonpoint source of perfluorinated
in semiarid regions: Karkheh River Basin in Iran. Hydrol. Process. 28 (4), 2018–2032. compounds and their loads in the Tsurumi River basin, Japan. Chemosphere 71 (8),
van Nes, E.H., Rip, W.J., Scheffer, M., 2007. A theory for cyclic shifts between alternative 1566–1573.
states in shallow lakes. Ecosystems 10 (1), 17–28.