RESEARCH ARTICLE | MAY 12 2023

Flood frequency analysis of annual maximum stream flows

for Kuantan River Basin
Abdullah Mukmin Ahmad; Noor Suraya Romali  ; Sumiliana Sulong

AIP Conference Proceedings 2688, 040002 (2023)

https://doi.org/10.1063/5.0111746

CrossMark

 
View Export
Online Citation

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

Articles You May Be Interested In

Bat algorithm and neural network for monthly streamflow prediction

AIP Conference Proceedings (November 2018)

Simulation of upper Kuantan River basin streamflow using SWAT model

AIP Conference Proceedings (September 2015)

Sedimentation rate and 210Pb sediment dating at Kuantan and Kemaman harbours, Malaysia
AIP Conference Proceedings (February 2019)
 Flood Frequency Analysis of Annual Maximum Stream
Flows for Kuantan River Basin
Abdullah Mukmin Ahmada) Noor Suraya Romalib) and Sumiliana Sulongc)

Faculty of Civil Engineering Technology, Universiti Malaysia Pahang, Lebuhraya Tun Razak,
26300 Gambang, Pahang, Malaysia

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

a)
abullahmukmin@gmail.com
b)
Corresponding author: suraya@ump.edu.my
c)
sumiliana@yahoo.com

Abstract. Precise stream flow forecasting is essential in water management. Effective usage of flow estimates gives
considerable assistance to water resources planning. In this study, five common distribution models, namely Generalized
Extreme Value (GEV), Generalized Pareto (GP), Log-Pearson 3, Weibull (3P), and Log-normal were employed to
identify the most appropriate probability distribution and to forecast the streamflows for the Kuantan River Basin. Yearly
peak flow data from the Bukit Kenau station from 1977 to 2013 was used in the study. The best-fitted distribution model
was evaluated using the Kolmogorov-Smirnov (K-S) goodness-of-fit test (GOF). The results show that Generalized
Extreme Value (GEV) is the best-fitted flood distribution model, with a P-value of 0.997, followed by the Generalized
Pareto, Log Pearson (3), Weibull, and Log-normal. The estimated peak flow values for 5, 10, 25, 50, 100, 200, 500 and
1000-year ARIs were 1569 m3/s, 1984 m3/s, 2560 m3/s, 3030 m3/s, 3535 m3/s, 4080 m3/s, 4868 m3/s, and 5521 m3/s
respectively. The results of the peak flow for different ARIs might benefit future flood models and risk assessments
conducted in this area of study.

INTRODUCTION
In recent years, there has been considerable interest in characterizing flood behavior in data-scarce areas due to
the increasing number of high-profile flood events [1]. This has generated interest in the development of end-user
models for emergency response, re-insurance, aid, and development. However, the lack of globally available
discharge estimates is preventing these technologies from being used. One of the most fundamental issues in
catchment hydrology is estimating high flows in unmeasured catchments.
The analysis of flood frequency can be defined as an approach without dimensions which utilizes probability
distributions derived from the historic records of peak discharge taken at many riverside gauge stations. This data is
used to connect the extent of extreme events to how frequently they occur [2]. Deriving the flood frequency curve
using probability distribution functions is the most widely used statistical method in hydrology. It is a promising and
straightforward way to address this problem [3]. Several studies have found that flood frequency and magnitude are
increasing in various places around the world [4]. Flood frequency study was undertaken by Rahman et al. [5] with
the aim of selecting the distribution models best suited to fit the data for the yearly maximum flood levels in
Australia. Their study involved the leading 15 probability distributions, as recommended by experts. These were as
follows: normal, two-parameter lognormal (LN2), three-parameter lognormal (LN3), two-parameter gamma (G2),
Pearson 3, Gumble EV1, GEV, LP3, exponential, Weibull, generalized logistic (GL), logistic, five-parameter
Wakeby (WAK5), four-parameter Wakeby (WAK4), and GPA. The researchers used four goodness-of-fit tests: the
Akaike information criterion, the Anderson–Darling test, the Bayesian information criterion, and the Kolmogorov–
Smirnov test. The outcome was the identification of the three leading distribution models which are the log-Pearson
3, Generalized Extreme Value (GEV), and Generalized Pareto distributions. Guru and Jha [6] used the data from the
flood series of Annual Maximum (AM) and Peak over Threshold (POT) when conducting an analysis of flooding in

World Sustainable Construction Conference Series 2021

AIP Conf. Proc. 2688, 040002-1–040002-6; https://doi.org/10.1063/5.0111746
Published by AIP Publishing. 978-0-7354-4483-6/$30.00

040002-1
the Mahanadi River system in Eastern Central India, specifically in the Tel basin. With each goodness-of-fit test,
more successful results were derived from the AM flood data series utilizing the Generalized Pareto (GP)
distribution. Meanwhile, with each goodness-of-fit test, the most successful outcomes were obtained using the
LogNormal (3P) distribution, following by using GP distributions for the POT flood data series. A recent study was
conducted by Samantaray and Sahoo [7] in Mahanadi river basin. Monthly flow data from four stations on the
Mahanadi River was involved namely Rampur, Sundargarh, Jondhra, and Basantpur. The most successful outcomes
were given by GEV at the gauge stations of Rampur, Sundergarh, and Jondhra; followed by LP III. Meanwhile, at
Basantpur, LP III provided the best fit, after which came GEV distribution model.
In the Malaysian context, a quantitative goodness-of-fit test was conducted by Mohd Daud et al. [8] to ascertain
which probability distribution was best suited to delineating Peninsular Malaysia’s yearly maximum rainfall series.
The study investigated the most frequently used probability distributions (eight in all): two-parameter Gumbel and
gamma, generalized Pareto (GPA), Pearson type 3 (PE3) and log-Pearson type 3 (LP3), three-parameter generalized
extreme value (GEV), generalized normal (GNO), and the five-parameter Wakeby. The study found that GEV was
the distribution best suited to delineating Peninsular Malaysia’s yearly peak rainfall series. In a similar study
conducted in the Kuantan River Basin, Lar Win and Win [9] stated that the same distribution model i.e. GEV was

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

the most appropriate model to use when describing annual peak daily precipitation level at four riverside stations.
Meanwhile, Salarpour et al. [10] noted that the distribution that best fitted the Johor river basin peak flows was
Generalized Pareto. A methodology based on an Archimedean Copula (t-Copulo) was used to model the joint
dependence structures of the attributes, quantity, and duration of the flooding. Similar results were achieved during
research by Romali and Yusop [11] in the Johor region in the south of Malaysia. The researchers determined that the
distribution model most appropriate for describing the Segamat River’s annual peak flows was Generalized Pareto.
Recently, a flood frequency study by Mohamad Hamzah et al. [12] was conducted in the Pelabuhan Klang area in
Malaysia. The study aimed to determine the magnitude of high tide at selected return periods at Pelabuhan Klang.
Several distributions models, namely Weibull, Gumbel, Generalized Extreme Value, Generalized Pareto, and
Lognormal (3 parameters), were tested. The Weibull distribution was found to best fit the annual maximum data of
high tides at Pelabuhan Klang.
Floods are common types of disaster in Malaysia, so forecasting the peak discharge at various intervals is
necessary for constructing hydraulic structures and flood-control measures. The availability of the recorded peak
discharge data is critical in flood frequency research, yet it is often limited or unavailable. As a result, estimating the
peak flood for various return periods is challenging [13]. Historical reports indicate how the Kuantan River Basin
(KRB) suffered numerous serious flood events in 2001, 2013, and 2016. These were caused by the city’s rapid
expansion and construction projects. In geographical terms, the events occurred in the east coast region of
Peninsular Malaysia, where extensive precipitation is experienced during the northeast monsoons [14].
Conducting flood frequency analysis in Kuantan River Basin is essential, while deciding the most appropriate
flood frequency analysis approach is critical to enhancing the reliability of water infrastructures. However, studies
on flood frequency analysis in the KRB remain scarce and the only available study was undertaken in 2014 by Lar
Win and Win [9]. Hence, this study attempts to assist the prevention and mitigation of floods in this flood-prone area
by determining the most appropriate flood frequency analysis approach to use in the KRB. The commonly used
flood distribution models were tested using the peak yearly flows of the Kuantan River Basin to select the best
distribution models and produce the most accurate predictions, based on the Kolmogorov-Smirnov (K-S) goodness-
of-fit (GOF) test (GOF).

METHODOLOGY
Illustrated in Figure 1 is the method used to analyze the flood frequency in the Kuantan River Basin. The initial
step was to prepare the data, which involved the collection of the annual rainfall statistics from the Department of
Drainage and Irrigation Malaysia (DID). The next step was the selection of the maximum annual flows from the
given data and to discard any missing or duplicated peak flow data. After that, using the Easyfit Software, the
selected data was input for probability distribution fitting using the selected distribution models. The Easyfit
software analyzed the peak flow data and ranked the best distribution models with a goodness-of-fit test (GOF),
based on Kolmogorov-Smirnov (K-S). The first-ranked distribution model was chosen to calculate the flow peaks
for different ARIs in the Kuantan River Basin [11].

040002-2
 Data Extracting maximum Input in Easyfit
preparation annual flow software

Fitting the data to selected Selected best Calculate ARI

distribution (Gen fitted based using selected
Pareto/GEV/Weibull on P-value distribution

FIGURE 1. Methodology used to analyze the flood frequency to determine the Average Recurrence Interval (ARI) in KRB

Study Area and Data

The Kuantan River Basin (KRB) is a key watershed of the city of Kuantan, in the southeast of Peninsular

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

Malaysia. The basin begins at Sungai Lembing and passes through the city, before ultimately draining into the South
China Sea. The Kuantan area is frequently impacted by flooding since extensive precipitation occurs annually from
November to December. This means that rivers spill over and floods affect low-lying areas, in which social and
economic human activity occurs [15]. The gauging station on the Kuantan river is located on the upper part of the
basin, near to the confluence of Kuantan River and Ulu Sg Kuantan. The Kenau flow station is the only flow gauge
in the Kuantan River Basin. A total of 33 annual peak flow data points from 1977 to 2013 were obtained from the
Department of Irrigation and Drainage of Malaysia (DID); these were used for the model distribution analysis. The
location of the flow station is shown in Figure 2.

FIGURE 2. Location of the study area. Adopted from Othman et al. [16]

Probability Distribution and Goodness of Fit (GOF)

To analyze the annual maximum flow, five models of probability distribution were selected: Generalized
Extreme Value (GEV), Generalized Pareto (GP), Log-Pearson 3, Weibull (3P), and Log-normal. The basis of this
choice was the historical flood frequency studies conducted in Malaysia with the commonly used distribution
models [8-12]. The Kolmogorov-Smirnov (KS) test for goodness-of-fit (GOF) ascertains if a particular distribution
fits when using a specifically collected dataset. The fit qualities for the dataset in the study were ranked using a
calculation of the statistical parameters [7].

040002-3
 Quantile Estimation of Flood Frequency Models
After the distribution model is fitted in EasyFit Software, the parameters of the distribution can be obtained from
the software to estimate the peak flow of different return period, T using quantile estimates, XT. The return period
corresponds to the exceedance probability(F), as shown in equation (1):
ଵ
F=1- ் (1)

where F is the probability of having a flood magnitude XT or smaller. General equation (2) is used for calculating XT
as proposed by Chow [17]:

xT u1'  K T P2 (2)

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

where KT is the frequency factor which is a function of the return period and of the parameters of the distributions.
The equation used to determine the XT for the GEV flood distribution models in this study is as shown in equation
(3) [18]. Where σ is the continuous scale parameters, k is the shape parameters, and T is year.
.
σ 1
Xt = ൣ1- ቄ-log ቀ1- ቁቅ ^k] (3)
k T

RESULT AND DISUCSSION

Figure 3 shows the annual floods that occurred from 1977 to 2013 at the Bukit Kenau flow station. On many
occasions in 1985, 1993, 1994, 1998, 2001, 2004, 2006, and 2009, serious flooding was experienced, where
maximum flows of over 1000 m3/s were recorded. The highest flow was recorded in 2001, with a value of 2287
m3/s. This massive flood was due to the continuous heavy rainfall during the northeast monsoon that affected
Kuantan communities and land severely, with hundreds of victims losing their property and assets [15].

2500

2000
Q (METERCUBE/S)

1500

1000

500

0
1977 1981 1985 1993 1997 2001 2005 2009 2013
YEARS

FIGURE 3. The annual peak flow for Bukit Kenau gauging station from 1997 till 2013

The selected distribution models’ parameters were calculated and generated using the Easyfit software, as
presented in Table 1. The parameters (α, k) stand for the shape parameters; (σ, β) and (μ, γ) represent the parameters
of continuous scale and continuous location, respectively. The peak flow estimations can be calculated using these
parameters, according to the quantile estimation given in Equation (3).

040002-4
 TABLE 1. Fitting results for probability distribution of annual flood

# Distribution
1 Gen. Extreme Value
2 Gen. Pareto
3 Log-Pearson 3
4 Weibull (3P)
5 Lognormal (3P)
Parameters
k=0.11389 V=405.35 P=485.52
k=-0.21223 V=846.81 P=72.031
D=18.827 E=-0.19714 J=10.052
D=1.118 E=718.27 J=78.89
D=0.7058 P=6.5117 J=77.659

TABLE 2. Goodness-of-fit test ranking by Kolmogorov-Smirnov test.

Kolmogorov
# Distribution Smirnov

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

P-value Rank
1 Gen. Extreme Value 0.99733 1
2 Gen. Pareto 0.99618 2
3 Log-Pearson 3 0.98813 3
4 Weibull (3P) 0.98194 4

5 Lognormal (3P) 0.84384 5

TABLE 3. Peak flow for 5, 10, 25, 50, 100, 200, 500 and 1000-year ARI using GEV distribution model

Return period, T Flood discharge, Q

(years) (m3/s)
5 1569
10 1984
25 2560
50 3030
100 3535
200 4080
500 4868
1000 5521

Table 2 ranks the probability distributions based on the goodness-of-fit test ranking using the Kolmogorov-
Smirnov test. According to Romali and Yusop [11], if the p-value from the GOF test using K-S represents a value
near to one, this indicates a better fit. EasyFit software ranked Gen. Extreme Value (GEV) first. Gen. Pareto was
second, Log-Pearson 3 was third, followed by Weibull (3P) and Lognormal (3P) in fourth and fifth place,
respectively, based on the goodness-of-fit test ranking using the Kolmogorov-Smirnov. These results explain why
Gen. Extreme Value (GEV) was the chosen distribution model used to estimate the peak flows in the Kuantan River
Basin. This is in agreement with the outcomes reported by Lar Win and Win [9], showing that to describe the annual
Kuantan River Basin maximum precipitation, GEV was the most suitable distribution. In addition, Mohd Daud et al.
[8] obtained similar results in their flood frequency study in Peninsular Malaysia, where GEV was found to be the
most suitable distribution model. Table 3 shows the GEV distribution models used to estimate the peak flows for 5,
10, 25, 50, 100, 200, 500 and 1000-year ARIs. They are 1569 m3/s, 1984 m3/s, 2560 m3/s, 3030 m3/s, 3535 m3/s,
4080 m3/s, 4868 m3/s, and 5521 m3/s respectively.

040002-5
 CONCLUSION
This study analyzed the flood frequency that occurred using 33 annual maximum streamflow data points from
the gauge station at Bukit Kenau. The research aim was to choose the distribution models best suited to the Kuantan
River Basin. Five probability distribution models were tested to identify the best probability model that could be
used to predict flow estimations for different ARIs. As a result, GEV showed the highest p-value (0.977), which
indicated it had the best fit compared to the other probability distribution models tested. The peak flow estimations
for the Kuantan River Basin for 5, 10, 25, 50, 100, 200, 500 and 1000-year ARI were 1569 m3/s, 1984 m3/s, 2560
m3/s, 3030 m3/s, 3535 m3/s, 4080 m3/s, 4868 m3/s, and 5521 m3/s. The outcome of this study can be applied in future
hydraulics and hydrological studies of the Kuantan River Basin.

ACKNOWLEDGMENTS
The authors would like to thank Ministry of Education (MOE) for providing financial support under
Fundamental Research Grant Scheme (FRGS) number FRGS/1/2019/TK01/UMP/02/2 (University reference

Downloaded from http://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0111746/17471329/040002_1_5.0111746.pdf

RDU1901155). Besides, technical support and encouragement from Faculty of Civil Engineering Technology and
Department of Drainage and Irrigation (DID) Malaysia are appreciated.

REFERENCES
1. A. Smith, C. Sampson, and P. Bates, Water Resources Research, 51(1), 539–553, (2014).
2. M. S. Bhat, A. Alam, B. Ahmad, B. S. Kotlia, H. Farooq, A. K. Taloor, and S. Ahmad, Quaternary
International, 507, 288–294 (2019).
3. G. Tegegne, A. M. Melesse, D. H Asfaw, and A. W Worqlul, Hydrology, 7(3), 1-21, (2020).
4. K. Vormoor, D. Lawrence, L. Schlichting, D. Wilson, and W. K. Wong, Journal of Hydrology, 538, 33–48
(2016).
5. A. S. Rahman, A. Rahman, M. A. Zaman, K. Haddad, A. Ahsan and M. Imteaz, Nat Hazards, 69, 1803-1813
(2013).
6. N. Guru and R. Jha, Aquatic Procedia, 4, 427-434 (2015).
7. S. Samantaray and A. Sahoo, H2Open Journal, 3(1), 189–207, (2020).
8. Z. Mohd Daud, A.H. Mohd Kassim, M. N. Mohd Desa and V. T. V. Nguyen, FRIEND 2002-Regional
Hydrology: Bridging the Gap between Research and Practice, (Proceedings of the Fourth International
FRIEND Conference held at Cape Town, South Africa), 274, 61–68, (2002).
9. N. Lar Win, and Win, International Journal of Science and Research (IJSR), 3(8), 977-983 (2014).
10. M. Slarpour, Z. Yusop, F. Yusof, S. Shahid and M. Jajarmizadeh, Journal of Applied Sciences, 13(7), 1021-
1028 (2013).
11. N. S Romali, and Z. Yusop, MATEC Web of Conferences, 103, 04003, 1-9, (2017).
12. F. Mohamad Hamzah, H. Tajudin, S. M. Syed Abdullah, E. Toriman, and H. Juahir, International Journal of
Engineering Trends and Technology (IJETT), Editor’s Issues, 92-96 (2020).
13. T. K. Drissia, V. Jothiprakash, and A. B Anitha,Water Resources Management, 33(3), 1013–1037, (2019).
14. S. M. Zaidi, A. Akbari, and W. M. F Ishak, International Journal of Civil Engineering and Geo-Environment, 5
(2014).
15. M. Y. Safiah Yusmah, L. J. Bracken, Z. Sahdan, H. Norhaslina, M. D. Melasutra, A. Ghaffarianhoseini, S.
Sumilian and A. S. Shereen Farisha, Natural Hazards, 101, 551-571 (2020).
16. N. Othman, N.S Romali, S.R Samat, and A.M Ahmad, IOP Conference Series: Materials Science and
Engineering, 1092(1), 012028 (2021).
17. V. T. Chow, Handbook of Applied Hydrology, McGraw-Hill (1964).
18. H. Kim, S. Kim, H Shin, and J.H Heo, Journal of Hydrology, 547, 557–574, (2017).

040002-6