Vous êtes sur la page 1sur 12

+ MODEL

Available online at www.sciencedirect.com

Journal of Hydro-environment Research xx (2013) 1e12


www.elsevier.com/locate/jher

Research paper

Risk and uncertainty analysis for dam overtopping e Case study:


The Doroudzan Dam, Iran
Ehsan Goodarzi a,*, Lee Teang Shui b, Mina Ziaei a
a
Department of Civil and Environmental Engineering, Georgia Institute of Technology, 251, 10th St. NW, APT A608, Atlanta, GA 30318, USA
b
Department of Water Resources Engineering, Faculty of Engineering, University Putra Malaysia, 3rd Floor, 43400 Serdang, Selangor, Malaysia
Received 29 August 2010; revised 29 January 2013; accepted 4 February 2013

Abstract

There is a growing tendency to assess the safety levels of existing dams by using mathematical and statistical methods. In this study, the
application of risk and uncertainty analysis to dam overtopping is presented for Doroudzan Reservoir located at the south part of Iran. The main
objective of the overtopping analysis of dams is estimating the height of water in the reservoir under various inflows and comparing the
computed results with the dam crest elevation. Hence, the main steps of this study include univariate flood frequency analysis of annual
maximum inflows to estimate the peak flows in various return periods, generate inflow hydrographs based on the estimated peak flows, and route
the hydrographs through the reservoir to compute the maximum height of the water in reservoir. In this study, the spillway discharge coefficient,
quantile of peak flows, and initial water surface level are subject to uncertainty, and the Monte-Carlo simulation (MCS) and Latin hypercube
sampling (LHS) are applied to perform the uncertainty analysis. In addition to inflows, the effect of different wind speeds on the probability of
overtopping has been considered. The results demonstrated that both increasing water level and wind speed have significant impact on the risk of
overflowing.
Ó 2013 International Association for Hydro-environment Engineering and Research, Asia Pacific Division. Published by Elsevier B.V. All rights
reserved.

Keywords: Risk analysis; Uncertainty analysis; Overtopping; Flood; Wind; Reservoir

1. Introduction quantitative support for decision makers, but also helps to find
the most effective options for decision-making. For instance,
The risk concept has a long history and has been a main engineers could never have designed systems such as great
aspect of life since the beginnings of human experience. Ap- bridges, dams, sewer systems, and so on, without some form
plications of risk and safety analyses have been developed of risk assessment.
simultaneously by expanding various facets of technology in An efficient way to manage water resources is dam con-
all branches of science, such as engineering and environment. struction that creates reservoir to storage water and distributes
The main intentions of risk and safety analyses are to identify it at the right time into downstream districts. Reservoirs have
existing system threats and predict possible outcomes in the significant roles in water resource engineering in which their
future to provide clearer ideas for making the best possible proper design, construction and maintenance contribute
decisions. In other words, risk analysis not only provides considerably toward fulfilling water supply requirements and
minimizing the risk of water shortages. The proper design of a
dam’s spillway and the flood control capacity of a reservoir
can ensure the safety of a dam and prevent any undesirable
* Corresponding author. Tel.: þ1 4049446539.
E-mail addresses: ehsan.goodarzi@gatech.edu (E. Goodarzi), tslee@eng. problems such as overtopping. The design flood of reservoirs
upm.edu.my (L.T. Shui), mina.ziaei@gatech.edu (M. Ziaei). is usually computed based on univariate flood frequency

1570-6443/$ - see front matter Ó 2013 International Association for Hydro-environment Engineering and Research, Asia Pacific Division. Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jher.2013.02.001

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

2 E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12

analysis of peak discharges in which the hydrograph of design construction by focusing on the upstream water surface
flood is routed through the reservoir to determine the discharge elevation for flooding condition. Kuo et al. (2007) presented
capacity of the spillway gates. However, dams still suffer the procedure and application of risk and reliability analysis of
overtopping and this problem comprises about one third of all Feitsui Dam overtopping by considering five uncertainty
uncontrolled breach failures (ICOLD, 1973). Traditionally, the analysis methods (MFOSM, RPEM, HPEM, LHS, and MCS)
approach to dam design focuses deterministic analysis on and four initial water levels in five return periods. Li and Zhao
extreme events such as probable maximum flood (PMF). PMF (2010) introduced a time-dependent reliability method to
is calculated based on Probable Maximum Precipitation predict the risk of dam failures due to increasing wave effect
(PMP) which is the greatest rainfall rates for specified dura- and decreasing structural capacity. A stochastic process also
tions and is used to predict the largest flood volume at a dam was proposed by them to model the time variant and random
site. PMF assumes the risk of dam failure is zero and there is nature of severe waves. In another study, Goodarzi et al.
no danger for downstream properties and population. How- (2012) presented the application of risk and uncertainty
ever, PMF has two important disadvantages that should be analysis to dam overtopping based on univariate and bivariate
considered by dam engineers; variation of PMF over time with flood frequency analyses by applying the Gumbel logistic
regards to climate changes, and, lack of proportional balance distribution.
between costs, benefits and risk of dam failure (Stedinger et al. In this study, overtopping risk and uncertainty analyses
1996). By improving the mathematical and statistical models, based on univariate flood and wind speed frequency analyses
increasing the ability of computer programs, and accessibility are presented. The uncertainty factors are spillway discharge
to more data records for longer periods; it is time to move coefficient, quantile of peak flows, and initial water surface
from the deterministic approaches in engineering design to level. The Monte-Carlo simulation (MCS) and Latin hyper-
probabilistic methods that consider higher order uncertainty in cube sampling (LHS), as two effective sampling approaches
the variables and models. The probabilistic point of view helps are applied to perform the uncertainty analysis in this study. In
engineers to produce a distribution or range of performance addition, the effect of different wind speeds in the overtopping
predictions with related probabilities of occurrence instead of risk analysis, as a main gap in the available literature, has been
a single performance prediction. considered in this study.
In the past, some studies were carried out to consider the
risk and reliability analysis in dam safety. Wood (1977) 2. Flood model (reservoir routing)
evaluated the overtopping risk for an embankment dam by
applying the integral transformation approach. Cheng et al. The main objective of the overtopping analysis is esti-
(1982) evaluated the risk of overtopping by applying various mating the height of water in the reservoir under various in-
methods including; direct integration method, Monte Carlo flows and wind speeds and comparing the computed results
simulation, mean value first-order second-moment (MFOSM), with the dam crest elevation. The continuity equation as a
and advanced first-order second-moment (AFOSM), and known flood model that is frequently used in reservoir engi-
compared the results of different approaches with each other. neering applications, is:
The Committee on the Safety of Existing Dams (1983) offered
a risk index for overtopping and structural failures, and dis- ds
Qin  Qout ¼ ð1Þ
cussed the concept of risk-based design in hydrosystem en- dt
gineering. Singh and Snorrason (1982, 1984) studied the
where are reservoir inflow and outflow (m3/s), S is storage
historical overtopping dam failure of some earth filled dams
(m3), and t is time (s). The implementation form of reservoir
and found a strong correlation between breach width and
routing is:
height of dam. Bowles (2001) studied the tolerable risk
concept in hydrosystem engineering and presented some ex- Qint þ Qintþ1 Qoutt þ Qouttþ1 Stþ1  St
amples for tolerable risk criteria in dam safety. Yanmaz and  ¼ ð2Þ
2 2 Dt
Beser (2004) applied bivariate flood frequency analysis to
estimate the overtopping risk of a detention dam. Wang and where Qint and Qintþ1 are inflow into the reservoir (m3/s), Qoutt
Bowles (2005) studied different breach locations of an and Qouttþ1 are outflow from the reservoir (m3/s), St and Stþ1
earthen dam due to wave overtopping. Their results showed are reservoir storage (m3) at t and tþ1, respectively, and Dt is
that wind direction, as well as the wind speed, have an effect time interval (s).
on the location of the breach. Kwon and Moon (2006) intro- The maximum water height in the reservoir could be esti-
duced three major innovations to improve overtopping risk for mated by solving equation (2) step by step. Time interval Dt
existing dams based on stochastic concepts. Their innovations determines the length of each step in the reservoir routing and
are; using non-parametric probability density estimation output precision will be increased with decreasing Dt. In this
methods for selected variables, applying Latin hypercube study, a time interval of 30 min was selected to reduce
sampling to improve the efficiency of Monte Carlo simulation, uncertainty due to the highest water level possibility, which
and finally using Bootstrap re-sampling technique to deter- may occur between t and tþ1. The fourth order RungeeKutta
mine initial water surface level in the reservoir. Marengo is applied to solve reservoir routing throughout this
(2006) studied the probability of overtopping during dam investigation.
Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12 3

3. Wind model where A0 ¼ 0.6392(H/H0)2.0256, A1 ¼ 0.1804(H/H0)0.391, and


H is wave height (m) and it can be equaled the significant
Wind blowing on the free water surface in the reservoir weight height (Wang and Bowles, 2005). The significant wave
and generate waves. The height of generated wave is func- height Hs (m), which is the average of the highest one-third of
tion of the wind duration, the wind velocity, the depth of the waves of a given group or spectrum can be calculated by
water in the reservoir, and the fetch length or length of water the following equation (USBR, 1992):
surface over which the wind has blown. When wind blows
over a water surface, it applies a horizontal stress on the Hs ¼ 0:00237V 1:23 F0:5 ð6Þ
water surface and resulted in piling up of water above water
level of the reservoir. If the water elevation is very close to where, V is the wind speed over the water surface (km/h), and
the crest, the generated waves might wash over and increase F is the fetch length (km). Finally, the total wave height (Hw)
the probability of overflowing. In this case, wind set-up and which is an integration of the wind set-up and wave run-up in
wave run-up are applicable factors in evaluating the effect of the reservoir can be calculated as follow:
wind speed on the water surface elevation in reservoirs. Hw ¼ YR þ YS ð7Þ
Hence, there is a requirement to make a relationship between
the wind return period (Tw) and wind speed in the desired
return period ðVTw Þ to calculate wind set-up, wave run-up, 4. Risk model
and the total height of the water in the reservoir. When
wind hits the beach, a set-up is created and it causes the System failure occurs when the system fails to perform its
mean level of water goes up and stay higher than the normal objective functions and does not meet its requirements. In
water level. This phenomenon is called wind set-up. The other words, reliability or the probability of non-failure was
USBR (1992) has provided the following equation to defined as ability of system to fulfill their design functions and
compute wind set-up; it is commonly measured by studying the interaction of load
(L) and resistance (R) based on probabilistic analysis. When a
V2F system is reliable, the resistance of system exceeds the load,
Ys ¼ ð3Þ while if load exceeds resistance, the system cannot achieve the
62772D
defined purposes and it is unreliable. Identifying load and
where, Ys (m) is wind setup, F is the fetch length (Km), V is resistance is a fundamental issue in risk analysis and it
the wind speed over the water surface (Km/hr) and D (m) is noticeably depends on the type of hydraulic structure and
mean water depth along the fetch length. problem physics. Mathematically, the reliability of a system ps
On the other hand, if wave approaches a structure, such as is defined as (Tung et al. 2005):
when a wave hits the wall of a dam, part of the energy will be
destroyed in term of turbulence, and the rest of the energy will ps ¼ 1  pf ¼ PðL  RÞ ð8Þ
be used to run-up the upstream slope of the dam. Therefore,
wave run-up is defined as the vertical difference between the where, P([$]) is the probability of failure. Risk also can be
highest water level caused by the run-up on the dam and the represented as (Singh et al. 2007):
water level at the slope foot. According to the height of the a ¼ Risk ¼ PðZ < 0Þ ð9Þ
run-up, it can be determined whether overtopping occurs or
not. This parameter is a function of the measured character- where, Z is performance function. The main performance
istics of the wave, including significant wave height (Hs), the functions commonly used in hydrosystem engineering are;
wavelength (L), slope of the dam body (q), roughness, and the Z ¼ R  L, Z ¼ (R/L)  1, and Z ¼ ln (R/L). According to the
permeability of the dam body. Hughes (2004) presented an above equations, the system will be safe if Z > 0, while for
equation, which was based on the wave moment flux as Z < 0the system is not safe and failure occurs, and the
follows: condition Z ¼ 0 is called limit state function or failure
 1=2 surface.
YR Mf
¼ 3:84 tan q$ ð4Þ
H0 r$g$H02 5. Uncertainty analysis

where YR is the maximum run-up of regular waves (m), H0 is Nature has immense variability, and the information
water depth from the bed to the current water elevation (m), available to quantify this variability is usually limited. Hence,
Mf is depth integrated wave moment flux per unit width, r is planning, design, operation, and management of civil and
the density of water (Kg/m3), and q is the embankment slope. environmental engineering systems are greatly affected by the
Hughes (2004) also presented an empirical relationship for the unexpected action of uncertain natural events which lead to
estimation of the momentum flux parameter as follow: uncertain system performance. The various system compo-
   A1 nents are subject to different kinds of uncertainty in hydro-
Mf h
¼ A0 ð5Þ system engineering, and hence, it is very difficult to assess the
r$g$H02 max gT 2 system’s behavior with distinct certainty. One problem
Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

4 E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12

regarding the different uncertain variables in complex and number of iterations and simulations. Although the accuracy
non-linear models is deriving the PDF of uncertain variables of Monte Carlo simulation highly depends on the number of
and determining the appropriate statistical moments or prob- iterations and generated random numbers, performing simu-
ability distribution of model outputs. Furthermore, any anal- lations by applying this method is easier due to computer
ysis in the real world is based on historical recorded data, advancements. The computer programs are used to generate
while usually historical records are not long enough and the different random variables regarding the defined probability
data includes all sorts of errors. Hence, sampling as an distributions for input variables and then recalculate the model
applicable method to compound several random input values repeatedly to simulate all possible outcomes.
can be applied to get results with appropriate accuracy. However, increasing sample size in sampling based
Sampling can be defined as the procedure of selecting an methods can reduce sampling errors, while simulation process
individual from a specific statistical population to evaluate and computer time for generating random variates will be
characteristics of the entire population. In other words, it is increased. On this basis, there are some variance reduction
the extrapolation from sample to the population, and helps techniques to increase the precision of Monte Carlo simulation
engineers to improve quality of data and saving in time and outcomes without the need to increase sample size (Tung et al.
cost. The estimated results can be analyzed statistically to 2005). Some of the most important methods of variance
predict the behavior of the system and measure the risk of reduction are antithetic-variate technique, control variates,
overtopping more precisely. As, the accuracy of the sampling importance sampling technique, Latin hypercube sampling
methods strongly depends on the sample size, a large number (LHS), correlated sampling, and stratified sampling technique.
of samples (20,000 for MCS and 10,000 for LHS) were LHS is one of the main variance reduction techniques that can
considered in this study to increase the precision of the cal- increase the efficiency of the output statistics parameters. This
culations. As the Latin hypercube sampling can converge with method frequently used to decrease the number of necessary
smaller sampling, its sample size is considered half of the runs of Monte Carlo simulation to achieve a reasonably ac-
MCS technique. curate random distribution. In this method, the range of each
variable is divided into m non-overlapping intervals with the
5.1. Sampling techniques (MCS and LHS) equal probability 1/m. Then, a random variate is selected from
each range with regards to the desired probability distribution
The Monte Carlo simulation, as one of the most famous and (Singh et al. 2007). A simple and primary algorithm for
widely used numerical methods, is a numerical simulation that applying the LHS method is:
replicates stochastic variables according to a certain statistical
distribution. The basic part of this method is iteration and 1. Divide the range of input variables into the number of m;
generation of random variables from a specific range. In other 2. Generate M uniform random number from U(0,1/M );
words, it is a numerical simulation which replicates stochastic 3. Perform random permutation;
input random variables from a particular probability distribu- 4. Determine random variates (xi,j) by applying following
tion to model desired process. To generate continuous random equation:
numbers based on the Monte Carlo simulation, assume X as a
 
random variable and Fx(X ) as its cumulative distribution 1 
function (CDF), the inverse function for any value of u w u xi;j ¼ F1
j Pi;j  ri;j ð11Þ
m
(0,1) can be written as:

X ¼ F1
x ðuÞ ð10Þ Where ri,j and Pi,j are random number and random permuta-
tion, respectively (Kwon and Moon, 2006).
where F1 x ðuÞ is the inverse function and u has a uniform Based on the LHS method, each generated random
distribution on (0,1). variate is placed in a separate interval with the equal
It should be noted that the continuous probability distri- probability of 1/m. For example, Fig. 1 shows the range of
butions in hydrosystem engineering is strictly uptrend for all each variable for m ¼ 5 non-overlapping intervals with
random variables X and thus, there is a unique relationship equal probability of 1/5 ¼ 0.2. On the other hand, the
between Fx(x) and u as u ¼ Fx(X ). To generate m random generated random variates from the Monte Carlo (MC)
variables using the CDF-inverse method, the following steps technique are randomly distributed and there may be more
should be repeated m times: than one random variate, or no random variate placed in an
equal probability area. Fig. 2 illustrates the main differences
1. Draw a uniform random variate as uwu(0,1), (random between MCS and LHS techniques. As it can be seen from
number generator), this figure, with the LHS sampling strategy each row and
2. Find x such that x ¼ F1
x ðuÞ. each column is filled by a black circle which represents a
generated random variate. Whereas based on MCS method,
However, there are two major concerns with Monte Carlo some rows and columns do not contain any black circle and
simulation; 1) it needs large computations to generate random some rows and columns have been filled with more than one
values, and 2) accuracy of results strongly depends on the random variate.
Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12 5

Table 1
Physical characteristics of Doroudzan Reservoir.
Type Earth-fill
Height 57 (m)
Crest length 710 (m)
Crest width 10 (m)
Fill volume 4.8 (106 m3)
Volume 993 (106 m3)
Dead storage 133 (106 m3)
Spillway type Ojee spillway

This dam is a major source of water supplying for


112,000 ha agricultural land and domestic-industrials and
power plants requirements of Shiraz, the capital of Fars
province, and Marvdasht and Zarghan as two other main cities
of Fars province. In addition, any problems with the Dor-
oudzan Dam will undoubtedly immerse two valuable heritage
Fig. 1. Non-overlapping probability area for m ¼ 5 with equal probability of 1/5. sites, the Pasargadae and Persepolis monuments which date
back to 515 BC, located downstream of the dam.
All inflows, reservoir storage, evaporation, and releases
6. Case study from 1975 to 2008 have been collected by the Fars Ministry of
Energy Data Center land based/surface data collection. Team
6.1. Doroudzan Dam members collected all available meteorological data including
inflows, water elevation, rainfall, temperature, etc. for each
Doroudzan Dam is one of the most important dams in the station along the Kor River, and the recorded data were ported
south part of Iran. The preliminary studies and investigations in Microsoft Excel workbooks for data quality assurance and
of the dam were carried out in the years 1963e1966 and dam quality control.
construction was started and completed in 1970 and 1974,
respectively. The basin of this multipurpose earth-fill dam is 6.2. Outlier test
situated near northwest Shiraz on the Kor River and in the
Bakhtegan lake catchment area. The Kor river watershed is In the first step of this study, an outlier test is applied for
0 0
between longitude 51 43 and 52 54 East and latitude 30 080 34-year (1975e2008) annual maximum discharges to deter-
 0
and 32 00 . The elevation of the highest watershed point is mine the data that are departed from the trend line. Without
3749 m from the mean sea level and is located northwest of the outlier test, the data point will not follow the trend of the
the watershed. The total volume and dead storage of the assumed population regardless of the probability distribution.
reservoir are 993 and 133 (106 m3), respectively. Basic tech- Data which are departed from the trend line occur in either
nical information of Doroudzan Dam, the schematic view of upper and lower tails, are called high and low outlier,
dam, and its basin are shown in Table 1and Fig. 3, respectively. In this study, outlier analysis (high and low
respectively. outlier) was implemented using the Bulletin 17B approach
(McCuen, 2005). According to the outcomes of this test, there
is a low event datum and it should be omitted from the annual
maximum flood series. Therefore, the number of recorded data
was reduced to 33.

6.3. Flood frequency analysis

Different statistical distributions were fitted to the annual


maximum floods in order to estimate the peak flows in various
return periods. The distributions that are fitted to the maximum
annual discharges at the significance level of 0.05 are; Gumbel
Max, General Extreme Value (GEV), Gamma, Log-Gamma,
Log-Logistic, Log-normal, Normal, and Pearson 5 (3P). Af-
terward, a goodness-of-fit test was applied for choosing the
best distribution based on the Chi Square test (Table 2).
Although the result of test demonstrated that all considered
distributions could be selected for the recorded flood data, the
Fig. 2. The generated random variates based on MCS and LHS methods. GEV distribution fits better than others and it is selected to
Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

6 E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12

Fig. 3. The schematic view of Doroudzan Dam and its basin.

predict the peak discharges in different return periods. speeds and minimum duration to reach maximum wave height
Furthermore, the mean and standard deviation of the estimated are computed in 2, 10, 20, 50, and 100-years return periods
peak discharge for the desired return periods were computed and the results is presented in Table 5.
based on the frequency analysis and the results are presented
in Table 3. 6.5. Statistical characteristics of uncertain variables

6.4. Wind frequency analysis The following variables are subject to uncertainty in this
study:
There are two main directions for the wind speed in the
Doroudzan basin, the south-west and west directions. As the 1. Peak floods in different return periods, (I ): the main rea-
speed of west-wind is higher than south-west, and its direction is sons for considering peak floods as uncertain variables are
along with the fetch length (Fig. 4), the west wind data have error in data recording, lack of data, and lateral inflow into
been used in this study to evaluate the wind set-up and wave run- the reservoir. The values for the mean and standard devi-
up in the reservoir of dam. Different statistical distributions are ation of peak floods are presented in Table 3. It is
fitted to the 34 years (1975e2008) annual maximum wind important to note that the estimated peak discharges based
speeds in order to estimate the maximum speeds in several re- on GEV distribution have been used to generate inflow
turn periods. In this case, the following distributions are used at hydrographs, and then, the generated hydrographs are
the significance level of 0.05; Gumbel max, General extreme routed into the reservoir to compute the maximum water
value (GEV), Gamma, Log-Gamma, Gamma 3P, Weibull, height.
Weibull 3P, Log-normal 3P, Normal, Pearson 5 (3P). A 2. The initial water level, (H0); the average depth of water in
goodness-of-fit test is applied to find the best distribution by the reservoir was computed based on the observed and
using the Chi Square test and the results is presented in Table 4. recorded water elevation over 33 years (1975e2008). The
Based on the goodness-of-fit test, the General Extreme mean and standard deviation of water depth were 43.16
Value (GEV), Gumbel Max, Log-Pearson, and other used (m) and 1.63 (m), respectively. In addition, six other
distributions can be considered for wind frequency analysis. In depths (with 1.5 m increments) have been assumed as
this study, the GEV distribution is selected and the wind initial depths in order to consider the effect of changing
initial water depth on the probability of overtopping. The
considered depths are 43.16, 44.66, 46.16, 47.66, 49.16,
Table 2 50.66, and 52.16 m.
Goodness of fit test of maximum annual flood. 3. Spillway discharge coefficient (C ). Its mean and standard
Probability distribution Chi Square deviation have been assumed as 2.05 and 0.069,
Statistic value Table value Remark respectively.
Gumbel Max 0.318 9.487 Ok
GEV (General Extreme Value) 0.288 9.487 Ok
Gamma 0.322 11.07 Ok Table 3
Log-Gamma 2.132 9.487 Ok Mean and standard deviation of peak discharges in various return periods.
Log-Logistic 1.441 9.487 Ok
Log-normal 0.415 7.814 Ok T-year 2-Years 10-Years 20-Years 50-Years 100-Years
Normal 4.938 9.487 Ok mI 524.191 755.388 871.876 1048.4 1201.14
Pearson 5 (3P) 3.618 9.487 Ok sI 21.56 52.74 78.83 126.30 173.85

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12 7

Table 5
Value of wind speed and minimum duration to reach maximum wave height.
T-year CDF V (km/hr) tmin (hr)
2 0.500 53.05 0.24
10 0.900 66.66 0.22
20 0.950 71.61 0.21
50 0.980 77.83 0.20
100 0.990 82.35 0.20

the log form of performance function was selected. Hence, the


form of performance function (Z ) can be considered as follow:
 
R
Z ¼ ln ð12Þ
L
More information on various performance function forms
and their application to hydraulic engineering systems are
presented by Yen (1979).
The next important and fundamental step is identification of
the load and resistance in the dam overtopping risk analysis.
These two parameters are strongly dependent on the type of
Fig. 4. The wind rose in Doroudzan basin. hydraulic structure and physics of the desired problem. As,
overtopping happens when the flood outlet cannot release
water fast enough and water rises above the dam and spills
The specifications of input parameters such as mean (m),
over, the height of the water in the reservoir (Hmax) and the
standard deviation (s), and the probability distribution func-
height of dam (HR) can be considered as load and resistance of
tion (PDF) that have been fitted to the random uncertain data
system, respectively. Therefore, the overtopping probability
are presented in Table 6.
with respect to the performance function due to different in-
flows and wind speeds can be expressed as follows (Singh
et al. 2007):
7. Overtopping risk model  
HR
Zf ¼ ln ð13Þ
As stated above, the performance function of an engineer- Hmax
ing system can be described in several forms in which the
selection of each form depends on the distribution type of and
desire performance function. In this study, the system out-  
HR
comes have been compared with the log-normal and normal Zfw ¼ ln ð14Þ
Hmax þ Hw
distributions, and the goodness-of-fit test was applied to
choose the appropriate distribution based on the Anderson- where, Zf is flood performance function, Zfw is flood and wind
Darling test and the results is presented in Table 7. performance function, HR is dam crest height, Hw is the total
The results of the test revealed that log-normal distribution wave height, andHmax is the highest water level during a flood
fits the available data better than normal distribution; and thus, event which can be calculated from reservoir routing equation.
Finally, the overtopping probability will be computed as
Table 4 follow:
Goodness of fit test for maximum annual wind speed.  
Probability distribution Chi Square m
Risk ¼ 1  B z ¼ 1  BðbÞ ð15Þ
Statistic value Table value Remark sz
Gumbel Max 0.611 7.814 Ok
GEV (General Extreme Value) 0.254 7.814 Ok
Gamma 0.820 7.814 Ok
Log-Gamma 0.627 9.487 Ok Table 6
Gamma (3P) 0.376 9.487 Ok Statistical properties for uncertain parameters.
Weibull 1.176 7.814 Ok
Variable Type PDF m s
Weibull (3P) 1.266 7.814 Ok
Log-normal (3P) 0.464 9.487 Ok H0 Random Normal 43.16e52.16 1.6341
Normal 1.192 7.814 Ok I Random GEV Table 3 Table 3
Pearson 5 (3P) 0.560 9.487 Ok C Random Normal 2.05 0.069

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

8 E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12

Table 7 Table 9
Goodness-of-fit tests for the system outcomes based on Anderson-eDarling Overtopping risk using LHS method due to different inflows.
test. H0 (m) T
T-year Probability distribution Statistic value Table value 2-Year 10-Year 20-Year 50-Year 100-Year
2-year Normal 0.824 2.50 43.16 5.66E-11 4.56E-10 7.83E-10 3.08E-08 8.06E-07
2-year Log-normal 0.176 2.50 44.66 5.95E-10 3.35E-09 6.78E-09 3.94E-07 4.60E-06
10-year Normal 0.827 2.50
46.16 4.04E-09 1.54E-07 2.69E-07 5.63E-06 4.40E-05
10-year Log-normal 0.171 2.50 47.66 7.80E-07 5.32E-06 9.80E-06 3.61E-05 8.73E-05
20-year Normal 0.872 2.50 49.16 5.70E-06 3.79E-05 4.25E-05 9.63E-05 1.77E-04
20-year Log-normal 0.203 2.50 50.66 6.33E-05 1.76E-04 3.78E-04 4.09E-04 7.76E-04
50-year Normal 0.894 2.50
52.16 9.95E-04 3.04E-03 4.43E-03 4.90E-03 6.12E-03
50-year Log-normal 0.212 2.50
100-year Normal 0.870 2.50
100-year Log-normal 0.215 2.50

in which, b is the reliability index indicator and is defined as


the ratio of mean of the performance function (mz) to its
standard deviation (sz).

7.1. Overtopping risk due to floods

The probability of overtopping is calculated for various


floods with 2, 10, 20, 50, and 100-year return periods and
considering peak discharges, initial water level in the
Fig. 5. Overtopping risks in different return periods for H0 ¼ 44.66 (m).
reservoir, and spillway discharge coefficient as uncertain
variables. All uncertain variables are assumed to be inde-
pendent variables, while the Monte-Carlo simulation (with a 7.2. Overtopping risk due to floods and wind
sample size of 20,000) and Latin hypercube sampling (with
a sample size of 10,000) are applied for uncertainty anal- After obtaining wind speeds in different return periods, the
ysis. The probability of overtopping due to floods in wind set-up and wave run-up are calculated based on the
different return periods and initial water levels are presented equations provided by USBR. The wind set-up and wave run-
in Tables 8 and 9. up are functions of the initial water level in the reservoir, and
Based on these tables, by increasing the initial water level hence, they are also subject to uncertainty. It is important to
in each step, the probability of overtopping (in a constant re- note that, there was no strong correlation between wind speed
turn period) was raised for both uncertainty approaches and inflows (Corr ¼ 0.152), and, thus, the wind speeds and
adopted in this study. The trends of computed risks indicated flood values are separately generated. Fig. 8 shows a scatter
that the calculated probabilities with Latin hypercube sam- plot as useful tool to graphically determine association be-
pling are slightly higher than the outcomes of Monte Carlo tween two maximum annual floods and wind speeds.
technique. Figs. 5 and 6 show the results of overtopping risks Hence, the highest water level in the reservoir and total
for two initial water levels; 44.66 (m), and 49.16(m) based on height of wave are calculated individually, afterward, the total
both MCS and LHS methods. Furthermore, the trend of
changing overtopping risk versus the initial water elevations in
the reservoir for both MCS and LHS methods in 2-year and
50-year return periods is shown in Fig. 7.

Table 8
Overtopping risk using Monte Carlo method due to different inflows.
H0 (m) T
2-Year 10-Year 20-Year 50-Year 100-Year
43.16 1.13E-11 1.67E-10 4.36E-10 2.60E-08 5.30E-07
44.66 3.60E-10 2.91E-09 4.99E-09 1.88E-07 2.53E-06
46.16 3.47E-09 7.95E-08 1.69E-07 2.28E-06 2.99E-05
47.66 2.95E-07 2.88E-06 5.96E-06 2.46E-05 3.13E-05
49.16 2.31E-06 1.46E-05 2.45E-05 4.02E-05 9.52E-05
50.66 4.58E-05 1.16E-04 2.59E-04 3.30E-04 3.89E-04
52.16 8.78E-04 2.75E-03 3.21E-03 3.73E-03 5.23E-03
Fig. 6. Overtopping risks in different return periods for H0 ¼ 49.16 (m).

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12 9

Fig. 7. Overtopping risk vs. the initial water elevations in the reservoir for both MCS and LHS methods in 2-year and 50-year flood return periods.

water elevation which is the sum of these two factors is return periods for Tw ¼ 2, 50, and 100-year and H0 ¼ 49.16 m
assigned as final water elevation in the risk analysis. However, are presented in Fig. 9.
many combinations of inflows, wind speeds, and water ele-
vations have been considered to cover the most likely condi- 8. Conclusions and discussions
tions that will probably happen in the reservoir. The
overtopping risks due to different floods and wind speeds in The overall procedure of risk analysis in this study in-
five return periods and four initial water levels are evaluated cludes; frequency analysis of floods and wind speeds, reservoir
by MCS and LHS uncertainty approaches and the results are routing, and integration of wind set-up and run-up to calculate
presented in Tables 10 and 11. In addition to tables, the the highest water elevation in the reservoir. Afterward, the
overtopping risks based on the MCS method versus different probability of overtopping was assessed by applying MCS and

Fig. 8. Scatter plot to graphically determine association between maximum annual inflows and wind speeds.

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

10 E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12

Table 10
Risk of Overtopping due to flood and wind using MCS.
TW H0 (m) T
2-Years 10-Years 20-Years 50-Years 100-Years
2-Years 47.66 3.45E-07 3.03E-06 6.46E-06 2.56E-05 3.25E-05
49.16 3.63E-06 1.78E-05 3.81E-05 9.18E-05 2.09E-04
50.66 5.86E-05 1.51E-04 3.82E-04 5.60E-04 9.34E-04
52.16 1.19E-03 3.32E-03 4.70E-03 6.91E-03 8.74E-03
10-Years 47.66 5.81E-07 3.83E-06 1.48E-05 4.44E-05 7.67E-05
49.16 3.92E-06 1.92E-05 4.12E-05 9.94E-05 2.25E-04
50.66 6.31E-05 1.63E-04 4.12E-04 6.05E-04 1.00E-03
52.16 9.83E-04 3.58E-03 5.08E-03 7.49E-03 9.01E-03
20-Years 47.66 6.36E-07 4.19E-06 1.62E-05 4.87E-05 8.34E-05
49.16 4.22E-06 2.06E-05 4.43E-05 1.07E-04 2.42E-04
50.66 6.66E-05 1.72E-04 4.35E-04 6.39E-04 1.06E-03
52.16 1.33E-03 3.71E-03 5.27E-03 7.78E-03 9.61E-03
50-Years 47.66 7.99E-07 5.24E-06 2.04E-05 6.11E-05 1.03E-04
49.16 5.29E-06 2.57E-05 5.56E-05 1.35E-04 3.01E-04 Fig. 9. Flood-wind overtopping risks in different return periods for H0 ¼ 49.16 (m).
50.66 8.34E-05 2.15E-04 5.45E-04 8.02E-04 1.30E-03
52.16 1.67E-03 4.64E-03 6.60E-03 9.77E-03 1.62E-02
100-Years 47.66 1.07E-06 7.00E-06 2.73E-05 8.18E-05 1.36E-04 Table 12
49.16 7.08E-06 3.44E-05 7.44E-05 1.81E-04 4.01E-04 Percentage of increasing overtopping risk in different H0 and Tw ¼ 20-years.
50.66 1.11E-04 2.87E-04 7.29E-04 1.07E-03 1.71E-03 H0 T ¼ 50 T ¼ 100
52.16 2.23E-03 6.19E-03 8.83E-03 1.31E-02 2.17E-02
LHS MCS LHS MCS
47.66 e e e e
49.16 133.11% 119.71% 191.35% 190.17%
50.66 392.03% 497.20% 276.24% 338.02%
LHS methods as two of the most used sampling methods in
52.16 1117.97% 1117.53% 829.82% 806.60%
water resources engineering, and considering the quantile of
flood peak discharge, initial depth of water in the reservoir,
and spillway discharge coefficient as uncertain variables. Table 13
Based on the achieved results, by increasing the initial Percentage of increasing overtopping risk in different wind speed and
water level in each step, the probability of overtopping (in a H0 ¼ 47.66 m.
constant return period) was raised for both uncertainty ap- TW T ¼ 50 T ¼ 100
proaches adopted in this study. To show the effect of LHS MCS LHS MCS
increasing initial water level in the reservoir on the risk of 2 44.60% 4.07% 7.10% 3.83%
overtopping, the percentage of increasing risk in different 10 52.08% 80.49% 11.91% 145.05%
20 63.99% 97.97% 19.13% 166.45%
50 98.34% 148.37% 42.04% 229.07%
Table 11
Risk of Overtopping due to flood and wind using LHS.
TW T
water levels, constant wind return period (Tw ¼ 20), and two
H0 (m) 2-Years 10-Years 20-Years 50-Years 100-Years
flood return periods (T ¼ 50, and 100) using both MCS and
2-Years 47.66 7.95E-07 5.93E-06 1.07E-05 5.22E-05 9.35E-05 LHS method is presented in Table 12. For example in T ¼ 50,
49.16 6.85E-06 4.39E-05 5.04E-05 1.22E-04 2.70E-04
H0 ¼ 49.16 m, and using MCS method the value of risk is
50.66 7.02E-05 1.98E-04 3.97E-04 6.00E-04 1.01E-03
52.16 1.21E-03 3.34E-03 4.74E-03 7.00E-03 8.92E-03 increased 119.71% compared with risk of overtopping in
10-Years 47.66 8.78E-07 5.93E-06 1.90E-05 5.49E-05 9.77E-05 H0 ¼ 47.66 m. It is important to note that the LHS stratifies
49.16 6.14E-06 4.23E-05 5.34E-05 1.30E-04 2.86E-04 cumulative distribution function (CDF) into several sub-
50.66 6.47E-05 2.01E-04 4.27E-04 6.44E-04 1.08E-03 regions and forces the input variables to be better than for
52.16 1.00E-03 3.68E-03 5.27E-03 7.98E-03 1.00E-02
simple random sampling, and hence, the estimated results are
20-Years 47.66 9.34E-07 6.28E-06 2.04E-05 5.92E-05 1.04E-04
49.16 6.44E-06 4.07E-05 5.65E-05 1.38E-04 3.03E-04 different from MCS method. On the other hand, the results
50.66 6.81E-05 1.79E-04 4.50E-04 6.79E-04 1.14E-03 revealed that wind speed could have a great impact on reser-
52.16 1.35E-03 3.80E-03 5.46E-03 8.27E-03 1.06E-02 voirs situated in windy areas, and the probability of over-
50-Years 47.66 1.10E-06 7.34E-06 2.45E-05 7.16E-05 1.24E-04 topping has been increased by increasing wind speeds in
49.16 6.51E-06 4.18E-05 6.78E-05 1.65E-04 3.62E-04
different return periods. To illustrate the effect of wind speed
50.66 8.49E-05 2.23E-04 5.61E-04 8.42E-04 1.38E-03
52.16 1.69E-03 4.74E-03 6.80E-03 1.03E-02 1.72E-02 on the overtopping risk, the percentage of increasing risk due
100-Years 47.66 1.37E-06 9.10E-06 3.15E-05 9.23E-05 1.57E-04 to various wind speeds for both LHS and MCS methods in
49.16 8.30E-06 4.05E-05 8.66E-05 2.11E-04 4.62E-04 H0 ¼ 47.66 m, and two flood and four wind speed return
50.66 1.13E-04 2.95E-04 7.44E-04 1.11E-03 1.79E-03 periods is presented in Table 13. Based on the results in
52.16 2.25E-03 6.29E-03 9.02E-03 1.36E-02 2.27E-02
Tw ¼ 10, T ¼ 50 and using LHS method, the overtopping risk
Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12 11

due to wind speed increases 52.08% rather than when the wind T Flood return period
speed is not considered in risk analysis. Tw Wind return period
In summary, the inclusion the uncertainty of key variables ui Uniform random number
results in an expanded range of overtopping risks in different V Wind speed over the surface of water (km/hr)
return periods, and provide significant information for decision VTw Wind speed in desired return period (km/hr)
makers to identify the critical parameters needed to effectively xi,j Random variates
monitor, and detect the events that indicate a developing z Performance function
failure mode. Dt Time interval (s)
a0 Risk
Acknowledgments a Reliability
r Density of water (kg/m3)
The authors would like to acknowledge their appreciations b Reliability index indicator
to Dr. Yung-Chia Hsu, Ph.D. Candidate, Dept. of Civil Engi- m Mean of variable
neering, National Taiwan Univ., Taipei for helping in the s Standard deviation
analysis of results, Eng. Naser Shokri, Ministry of Energy Fars q Slope of the dam body
Regional Water Authority, and Majid Mirzaei, PhD candidate
of University Putra Malaysia, for helping out in data collection
through this study.
References
Nomenclature
Bowles, D.S., 2001. Evaluation and Use of Risk Estimates in Dam Safety
Decision Making. 20-Year Retrospective and Prospective of Risk-based
Symbols Decision-making. ASCE, Santa Barbara, California, 17e32.
Cheng, S.T., Yen, B.C., Tang, W.H., 1982. Overtopping Risk for an
cms Cubic meter per second
Existing Dam. Civil Engineering Studies, Hydraulic Engineering Series
C Spillway discharge coefficient No. 37.
D Mean water depth along the fetch length (m) Committee on the Safety of Existing Dams, 1983. Water Science and Tech-
F Fetch length (km) nology Board. Commission on Engineering and Technical Systems and
F1
x Inverse function National Research Council. Safety of Existing Dams’ Evaluation and
H0 Mean of water elevation from bottom (m) Improvement. National Academy Press, Washington, DC.
Goodarzi, E., Mirzaei, M., Ziaei, M., 2012. Evaluation of dam overtopping
H Wave height (m) risk based on univariate and bivariate flood frequency analyses. Canadian
H1 Height difference between the crest of spillway and Journal of Civil Engineering 39 (4), 374e387.
initial water level (m) Hughes, S.A., 2004. Estimation of wave run-up on smooth impermeable slopes
H2 Height difference between the crest of dam and initial using the wave momentum flux parameter. Coastal Engineering 51,
water level (m) 1085e1104.
International Commission on Large Dams, 1973. Lessons from Dam Incidents,
HS Significance wave height (m) reduced ed. ICOLD, Paris.
Hmax Height of water in the reservoir (m) Kuo, J.T., Yen, B.C., Hsu, Y.C., Lin, H.F., 2007. Risk analysis for dam
HR Height of dam (m) overtoppingeFeitsui reservoir as a case study. Journal of Hydraulic En-
ht Wind setup (m) gineering 133, 955e963.
Kwon, H., Moon, Y., 2006. Improvement of overtopping risk Evaluations
hr Wave run-up (m)
using probabilistic concepts for existing dams. Springer 20, 223e237.
hw Total weight height (m) Li, C.Q., Zhao, J.M., 2010. Time-dependent risk assessment of combined
h Depth of water from the bed to the current water overtopping and structural failure for reinforced concrete coastal struc-
elevation (m) tures. Journal of Waterway, Port, Coastal, and Ocean Engineering 136,
Qin Inflow (cms) 97.
k Number of uniform random numbers McCuen, R., 2005. Hydrologic Analysis and Design. Pearson Prentice Hall,
New Jersey.
km Kilometer Marengo, H., 2006. Case study: dam safety during construction, lessons of the
L Wavelength (m) overtopping diversion works at Aguamilpa Dam. Journal of Hydraulic
Lf Load Engineering 132, 1121e1127.
m meter Singh, K.P., Snorrason, A., 1982. Sensitivity of Outflow Peaks and Flood
MCM Million cubic meters Stages to the Selection of Dam Breach Parameters and Simulation Models.
Technical Report 289. State Water Survey Division at the University of
Mf Depth integrated wave moment flux per unit width Illinois, USA.
P[$] Probability of. Singh, K.P., Snorrason, A., 1984. Sensitivity of outflow peaks and flood stages
Pi,j Random permutation to the selection of dam breach parameters and simulation models. Journal
Qout Outflow (cms) of Hydrology 68, 295e310.
R Resistance Singh, V.P., Jain, S.K., Tyagi, A., 2007. Risk and Reliability Analysis.
American Society of Civil Engineers.
ri,j Random number Stedinger, J.R., Heath, D.C., Thompson, K., 1996. Risk Analysis for Dam
S Storage (MCM) Safety Evaluation: Hydrologic Risk. U.S. Army Corps of Engineers
t Time (s) Institute for Water Resources: Cornel University.

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001
+ MODEL

12 E. Goodarzi et al. / Journal of Hydro-environment Research xx (2013) 1e12

Tung, Y.K., Yen, B.C., Melching, C.S., 2005. Hydrosystems Engineering Reli- Wood, E.F., 1977. An analysis of flood levee reliability. Water Resources
ability Assessment and Risk Analysis. McGraweHill Professional, New York. Research 13, 665e671.
USBR, 1992. Freeboard Criteria and Guidelines for Computing Freeboard Yanmaz, A.M., Beser, M.R., 2004. On the reliability based-safety analysis of
Allowance for Storage Dams. US Dept of the Interior, Bureau of Recla- the Porsuk Dam. Turkish Journal of Engineering and Environmental Sci-
mation, Denver, CO. ence 29, 309e320.
Wang, Z., Bowles, D.S., 2005. Dam breach simulations with multiple breach Yen, B.C., 1979. Safety factor in hydrologic and hydraulic engineering design.
locations under wind and wave actions. Advances in Water Resources 29, In: McBean, E.A. (Ed.), Reliability in Water Resources Management.
1222e1237. Water Resources Publications, Highlands Ranch, pp. 389e407.

Please cite this article in press as: Goodarzi, E., et al., Risk and uncertainty analysis for dam overtopping e Case study: The Doroudzan Dam, Iran, Journal of
Hydro-environment Research (2013), http://dx.doi.org/10.1016/j.jher.2013.02.001

Vous aimerez peut-être aussi