Professional Documents
Culture Documents
Binata Roy 1; A. K. M. Saiful Islam, Ph.D. 2; G. M. Tarekul Islam, Ph.D. 3; Md. Jamal Uddin Khan 4;
Biswa Bhattacharya, Ph.D. 5; Md. Hazrat Ali, Ph.D. 6; Abu Saleh Khan 7; Md. Saiful Hossain 8;
Gopal Chandra Sarker 9; and Nahruma Mehzabeen Pieu 10
Abstract: A danger level at a river location is the level above which a flood may cause loss and damage to nearby lives, crops, and
homesteads. Currently, in Bangladesh, the danger level is designed and updated considering the riverine flood during the monsoon season
(June–September). However, during the premonsoon season (March–May), flash floods, almost every year, damage crops in some parts of the
northeast Haor (saucer- or bowl-shaped large tectonic depression) region. Damage to Boro rice, the only crop that is grown in this region, is
catastrophic to the local population and threatens the food security of the country. At present, there is no warning or danger level to inform the
community to save the Boro crops from flash floods in that region. Hence, this study attempts to establish a flash flood danger level at 40
locations in the northeast region of Bangladesh based on flood frequency analysis and elevations of the river floodplains. The homogeneity
and adequacy of the data series was tested by performing the run test, lag 1 auto correlation, Mann-Kendall test, and Mann-Whitney test. After
passing the data quality tests, flood frequency analysis was conducted using six widely used distributions: lognormal, lognormal Type III,
Pearson Type III, log-Pearson Type III, and Gumbel and generalized extreme value (GEV) distributions. Best regional distribution was
selected from the probability plot correlations coefficient (PPCC), normalized root mean square error (NRMSE), and mean absolute error
(MAE) results. Among the candidate distributions, GEV distribution was found to provide the best results. Then, the danger level was fixed
on the basis of the flooding inundation scenarios of different return periods at which the stage is at or near the floodplain level. The results
showed that for most of the stations, the 2.33-year flood level coincides with the existing floodplain level. For the rest of the stations, flash
flood danger levels were fixed at the 5-year return period’s flood level. The recommended flash flood danger levels were found much lower
than the existing monsoon flood danger levels. For example, at Sunamganj station, the existing flood danger level is 8.25 mPWD (Public
Works Datum) whereas the recommended flash flood danger level is only 6.41 mPWD. At some stations, the difference is too prominent, even
more than 5 m, which can cause severe destruction for the Boro rice cultivation during a flash flood. It is expected that flash flood danger
levels estimated in this study may assist in flash flood forecasting in the northeast Haor region and provide an effective warning at the time of
necessity, thus minimizing the loss and damage by the flash flood to some extent. DOI: 10.1061/(ASCE)HE.1943-5584.0001760. © 2019
American Society of Civil Engineers.
Author keywords: Bangladesh; Danger level; Flash flood; Flood frequency analysis; Haor region.
1 5
Lecturer, Institute of Water and Flood Management, Bangladesh Associate Professor, Hydroinformatics Chair Group of the Integrated
Univ. of Engineering and Technology, Dhaka 1000, Bangladesh. Email: Water Systems and Governance, UNESCO-IHE Institute for Water Education,
binata_roy@iwfm.buet.ac.bd P.O. Box 3015, 2601 DA Delft, Netherlands. Email: b.bhattacharya@
2
Professor, Institute of Water and Flood Management, Bangladesh unesco-ihe.org
6
Univ. of Engineering and Technology, Dhaka 1000, Bangladesh (corre- Professor, Dept. of Civil Engineering, Chittagong Univ. of Engineering
sponding author). Email: akmsaifulislam@iwfm.buet.ac.bd; saiful3@ and Technology, Chittagong 4349, Bangladesh. Email: pdrmhali@gmail.com
7
gmail.com Deputy Executive Director, Operation, Institute of Water Modelling,
3
Professor, Institute of Water and Flood Management, Bangladesh House #496, Rd. #32, New DOHS, Mohakhali, Dhaka 1206, Bangladesh.
Univ. of Engineering and Technology, Dhaka 1000, Bangladesh. Email: Email: ask@iwmbd.org
8
tarek@iwfm.buet.ac.bd Superintending Engineer, Processing and Flood Forecasting Circle,
4
Research Assistant, Institute of Water and Flood Management, Bangladesh Water Development Board, 72 Green Rd., Dhaka 1215,
Bangladesh Univ. of Engineering and Technology, Dhaka 1000, Bangladesh. Bangladesh. Email: se.pffc@bwdb.gov.bd
9
Email: jamal919@gmail.com Project Director, Local Government Engineering Dept., Haor
Note. This manuscript was submitted on April 16, 2018; approved on Infrastructure and Livelihood Improvement Project, Agargoan, Dhaka
October 1, 2018; published online on January 31, 2019. Discussion period 1207, Bangladesh. Email: gopalpdhilip@gmail.com
10
open until June 30, 2019; separate discussions must be submitted for in- Research Assistant, Institute of Water and Flood Management,
dividual papers. This paper is part of the Journal of Hydrologic Engineer- Bangladesh Univ. of Engineering and Technology, Dhaka 1000, Bangladesh.
ing, © ASCE, ISSN 1084-0699. Email: nahruma@gmail.com
into the Haors within 3–6 h (Nowreen et al. 2013; NWS 2018). In the importance of providing a danger level for flash floods, separate
from monsoon flood, for saving the Boro crop–based agriculture of
most of the cases, farmers do not get enough time to harvest their
the northeast region has not been done yet, let alone its implica-
standing Boro crop which is the only crop of this region. Boro is
tions. Hence, this study is designed to recommend flash flood
cultivated over almost 80% of the Haor areas while only 10% of the
danger levels for the northeast Haor region with the following ob-
area is used to cultivate another rice variety known as Aman rice jectives: (1) to select a representative regional probability distribu-
(Huda 2004). A study by Khan et al. (2012) showed that about 70% tion function for the gauge stations of the northeast region, and
of the crop in this area is damaged by flash floods, which impacts (2) to fix up a new flash flood danger level by flood frequency
the national economy. The flash flood of 2010 damaged crops over analysis. In this study, the flood frequency analysis has been
152,000 ha (1,520 km2 ) causing loss of 775,000 t of crops worth conducted with the maximum annual water level which is the first
BDT 13.18 billion (Daily Star 2010) while the flash flood of 2016 step to determine the recurrence interval of an event. Once the
damaged about 50,000 ha (500 km2 ) of paddy in more than 100 recurrence interval is known, the flash flood danger level is recom-
villages of Sunamganj and Sylhet (Deshwara 2016). mended by evaluating the flooding scenarios of different return
Conventional structural flood control measures like submersible periods (2.33–100 year) at which the water levels reach or are
embankments levees, rubber dams etc. have already been built, and nearer to the floodplain at the full stage.
more are being built by the Bangladesh Water Development Board
(BWDB) and Local Government Engineering Department (LGED)
to protect the Boro crops from the flash floods (Hossain and Nishat Study Area
1996). A submersible embankment is a low-height embankment
specially designed for the protection of Boro crops from the flash Flash floods usually occur in the northeastern and southeastern
flood in the premonsoon season, which remains submerged during (Chittagong) regions of Bangladesh. However, the northeast is
monsoon and cooperates in the lateral migration of fish between highly susceptible to flash floods due to its surrounding hilly areas
river and floodplain (Mondal et al. 2004). About 47 submersible and saucer- or bowl-shaped large tectonic depressions. Besides,
embankments covering 5,000–6,000 ha (50–60 km2 ) each have Jhuming, or shifting cultivation (popularly known as Sweden cul-
been constructed between 1975 and 1990s by BWDB to save the tivation or cultivation of slash and burn), is a more prevalent and
Boro crop (Rahman and Salehin 2013). However, due to the de- suitable form of cultivation in the southeastern hill tracks rather
fective construction of the submersible embankments, erosion and than Boro cultivation (Baten et al. 2010). So premonsoon flash
breaching of the embankments during flood flow, and manmade floods are not a threating factor for agriculture in the southeastern
breaches for navigation/boat crossing, the submersible embank- regions of Bangladesh. The core northeast Haor region extends over
ments are not serving the design purpose to their full capability seven districts—Sunamganj, Habiganj, Netrokona, Kishoreganj,
(Alam 2016). Besides, the submersible embankments also have neg- Sylhet, Moulavibazar, and Brahmanbaria—covering an area of
ative impacts on the Haor environment and ecology (Khan 2010) as about 1.99 million hectares and accommodating about 19.37 million
well as cause drainage congestion after a flash flood. Considering people (CEGIS 2012). About 373 Haors are located in these dis-
the disadvantages of structural measures in this area, the nonstruc- tricts comprising about 85,900 ha (859 km2 ) of the area, which is
tural measures of flood risks management are widely preferred for around 43% of the total Haor region of Bangladesh (CEGIS 2012).
more sustainable and long-term solutions to water-related problems The Hakaloki Haor, Sumir Haor, Dakhar Haor, Tanguyar Haor,
with minimum costs for operation, maintenance, rehabilitation, and Gungiajuri Haor, Mukhar Haor, Kaowadighir Haor etc. are the ma-
replacement (Faisal et al. 1999; Kundzewicz and Menzel 2003; jor Haors within this area (Alam et al. 2010). In this study, 40 water
Kryžanowski et al. 2014). Flood forecasting and warning is the most level stations located within this area were selected. These sta-
widely used tool for nonstructural mitigation of flood damage tions spread over the major rivers of this region—Bhogai-Kangsa,
(Montz and Gruntfest 2002; UNECE 2000). In Bangladesh, the Dhalai, Dhanu-Baulai-Ghorautra, Jadukata, Khowai, Kushiyara,
flood forecasts are made by the Flood Forecasting and Warning Monu, Someswari, Surma-Meghna, Sutang, Piyan, Sari-Gowain,
Center (FFWC) of BWDB which is the nationally mandated organi- and Mogra. The northeast Haor area is well known for its unique
zation for all kinds of flood forecasting. These flood forecasts are hydroecological characteristics. It is a mosaic of wetland habitats
made with respect to danger levels defined by the FFWC. Unfortu- including rivers, streams, and irrigation canals, large areas of sea-
nately, these danger levels have been defined focusing only on ex- sonally flooded cultivated plains, and hundreds of Haors and Beels
treme events during the monsoon floods considering the period from (permanent water bodies) (Hussain and Salam 2007; Sumon and
June to September (FFWC 2018). No danger level for the flash Islam 2013). The climate of this area is subtropical monsoonal with
flood is fixed by the FFWC for forecasting flash floods yet. The an average annual rainfall of approximately 4,000 mm (NERP FAP -
danger level for flash floods is usually much lower than the danger 6 1995). Geographically, this area is surrounded by the hilly regions
level of monsoon floods. Due to the lack of proper flash flood fore- of India—Meghalaya on the north, Tripura and Mizoram on the
casting using danger level of monsoon floods, the farmers of the south, and Assam and Manipur on the east. Hydrologically, it lies
Haor region cannot take proper initiatives to save the Boro crop in in the Meghna basin of the Bengal delta. Fig. 1 shows the selected
their fields. Therefore, fixation of a reliable danger level for flash water level stations of the northeast region. The Haor system of the
flood–prone areas of the northeast region is essential which will northeast region and the selected stations as well as the respective
Fig. 1. Water level stations of the Bangladesh Water Development Board in the northeast region of Bangladesh used for this study. (Inset map data
from ESRI, HERE, DeLorme, MapmyIndia, © OpenStretMap contributors, and the GIS user community.)
Upazilas, districts, and rivers are given in Fig. S1 and Table S1, of the rivers; (4) there was not much variation in the rising and fall-
respectively. ing limb of the discharge and water level rating curves of the river
system of the northeast region; and (5) the planform of the rivers of
this region is comparatively stable. Considering all the aforemen-
Data and Methods tioned, the flood frequency analysis was done using water level data
rather than conventional discharge data.
Data
Methodology
Based on the data availability, the 3-h water level data from 1996 to
2014 for the 40 stations mentioned in Fig. 1 were collected from Flood frequency analysis is an important phase for fixing up the
BWDB. The cross sections of rivers near these stations were col- danger level of an area (IFCDR 1995). Selecting the best frequency
lected from both BWDB and the Institute of Water Modeling distribution available to fit the hydrologic data series for the north-
(IWM). Considering the short time of the concentration of flash east region is the first step. Three goodness-of-fit statistics were
floods (NWS 2018), we used 3-h water level data for the period considered, namely, normalized root mean square error (NRMSE),
1996–2014. All the water level or cross section data collected in probability plot correlations coefficient (PPCC), and mean absolute
this study were in terms of Public Works Datum (PWD) which is error (MAE). The flash flood danger levels for the 40 stations of the
approximately 0.46 m (1.5 ft) below the mean sea level (FFWC northeast region were fixed considering the flooding scenarios of
2018). different return periods (from 2.33 to 100 years) at which the water
Conventionally, flood frequency analysis is conducted using the level reaches or is nearer to the floodplain at full stage. The step-by-
discharge data (Stedinger et al. 1993; McMahon and Srikanthan step methodology of the frequency analysis followed by this study
1981; Rossi et al. 1984; Ahmad et al. 1988). But in this study, flood is shown in Fig. S2.
frequency analysis was conducted using water level data primarily
for the following reasons: (1) the inadequacy of the discharge sta- Data Analysis
tions compared with the water level stations in the northeast Haor
region—there are only 33 discharge gauge stations whereas the Annual Maximum Series Preparation
number of water level stations is 61; (2) 3-h water level data were There are two different models for analyzing extreme hydrologic
collected in these water level stations whereas the discharge data events: partial duration series (PDS) and annual maximum series
were collected twice or thrice in a month and required the prepa- (AMS). In this study, the flood frequency analysis was conducted
ration of a rating curve to convert them into continuous discharge with annual maximum water level time series. For the return peri-
data; (3) the flash flood warning was provided using the water level ods less than 10 years, annual maximum series was lower than
the flash floods, which matures and is ready for harvesting by the being used by many hydrologists for a very long time (Mann 1945;
middle of May (BWDB 2006). Hence, the duration of this analysis Kendall 1975; Hamed 2008; Mondal et al. 2012; Ahmad et al.
was set by considering the end of the harvesting period of Boro 2015). Hence, in this study, the nonparametric Mann-Kendall test
crops in that region. was used to statistically detect the presence of monotonic upward
or downward trend of the variable over time. If x1 ; x2 : : : xn is the
Homogeneity Assessment of the Data Series time series of length n, then the Mann-Kendall test statistic S is
A primary condition for frequency analysis is that the time series given by Eqs. (3) and (4)
data should be homogeneous over time, implying that the time
series must be random, independent, and stationary. In this study, X
n−1 X
n
the homogeneity of the water level data series of all gauge stations S¼ sgnðxj − xi Þ ð3Þ
i−1 j−iþ1
was checked by quantitative statistical tests for randomness, inde-
pendence, and stationarity at 5% significance level.
Randomness Test. For a probabilistic frequency analysis, it is 1 for x > 0
required that samples be pure random. A sample for which the sam- Where; sgnðxÞ ¼ 0 for x ¼ 0 ð4Þ
pling procedure is entirely by chance is called an unbiased, or ran- −1 for x < 0
dom, sample. Randomness in a hydrological time series means that
the data arise from natural causes. Randomness happens when the If the null hypothesis H0 (there is no trend in the time series) is
rivers or streams are regulated by upstream dams, regulators, or true, then S is normally distributed with
any flood control structures. Here, the northeast river system is not EðSÞ ¼ 0 ð5Þ
regulated by any large dam or flood control structure. However,
statistical test is necessary to justify this fact. Randomness can be
nðn − 1Þð2n þ SÞ
tested by both parametric and nonparametric tests. Generally, in VðSÞ ¼ ð6Þ
nonparametric tests, no underlying distribution is assumed. Another 18
feasible alternative is the turning point test which is a nonparametric where EðSÞ is the mean and VðSÞ is the variance of S. Then the
method for determining if a time series is random. This holds for Mann-Kendall Z is given by Eq. (7)
the run test as well. For the current study, the nonparametric two-
tailed run test was carried out at 5% significance level to test the S−1
pffiffiffiffiffiffiffiffiffiffi for S > 0
randomness of the data set (Bradley 1968; McGhee 1985; Adeloye VðSÞ
and Montaseri 2002). Z¼ 0 for S ¼ 0 ð7Þ
Independence Test. The independence test refers to the mutual Sþ1
independence of the random variables. If the data is independent, pffiffiffiffiffiffiffiffiffiffi for S < 0
it means there is no persistency. Otherwise, the series is persistent. VðSÞ
The persistence is normally quantified in terms of a serial correla-
tion coefficient. Therefore, a two-tailed lag 1 auto correlation test is A positive value of S indicates an increasing trend and vice versa
performed to check the independency of the data at 5% significance while test statistic Z gives the significance level of rejecting the null
level (Ngongondo et al. 2011). The lag 1 auto correlation is com- hypothesis.
puted using Eq. (1) while the limits of the auto correlation of an Once the trend is detected, the magnitude of trend is determined
independent series at 5% significance level are computed using using standard least-squares slope test, Theil–Sen approach (TSA),
the Eq. (2) or trimmed and Winsorized slope test. However, the standard meth-
ods for estimating and testing slopes are problematic when data are
Pn−k not normally distributed and residual variances are heterogeneous.
ðx − x̄Þðxiþk − x̄Þ
rk ¼ i¼1Pni 2
ð1Þ Different studies show that Theil–Sen slopes are more powerful
i¼1 ðxi − x̄Þ slope estimators than trimmed and Winsorized slopes (Moses and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Klockars 2008). Hence, in this study, TSA slope was used at 5%
−1 1.96 N − k − 1 significance level (Thiel 1950; Sen 1968). The TSA slope β is
rk ¼ ð2Þ
N−k−1 given by Eq. (8)
where N = sample size; and k is the lag. Haan (1986) discussed xj − xi
β ¼ median ð8Þ
another test of significance for the serial correlation coefficient de- j−i
veloped by Anderson (1942) for a circular, normal, stationary time
series. However, the assumption of a circular test series is unreal- Homogeneity of the mean test checks the consistency of the
istic (Haan 1986). Hence, this alternative was not attempted in this mean of the data series over time. The two-sample t-test is one
study. of the most used statistical procedures to test it. However, this test
Assessment of Stationary. Stationarity tests check the consistency assumes that the variable in question is normally distributed in
of the statistical parameters (mean, variance, standard deviation the two groups. When this assumption is in doubt, the nonparamet-
etc.) over time (Challis and Kitney 1990). A hydrologic time series ric Mann-Whitney test is suggested as an alternative to the t-test.
likelihood of occurrence of flood events at a particular location. expressed in a general form as described by Cunnane (1978) using
However, it is important to select a representative probability dis- the following equation [Eq. (10)]:
tribution function for fitting the extreme event at a particular loca-
tion. There is no universally accepted frequency distribution model ði − aÞ
Fi ¼ ð10Þ
for frequency analysis of an extreme event. Many distribution ðn þ 1 − 2aÞ
functions are being used depending on the characteristics of the
particular event (rainfall, runoff, flood flow, low flow, water level, where Fi is the plotting probability; and i is the rank in ordered
pollutant loadings, groundwater, etc.) and site. Widely used prob- observations in which i ¼ 1 presents the smallest observation in
the sample. In this study, the Blom formula with a ¼ 0.375 was
ability distribution functions in hydrology are normal distribution,
used for the LN, LN3, P3, and LP3 distributions (Vogel and Kroll
lognormal (LN) distribution, exponential distribution, gamma dis-
1989); the Gringorten formula with a ¼ 0.44 was used for the
tribution, Pearson Type III (P3), log-Pearson Type III (LP3), ex-
Gumbel distribution (Vogel 1986); and the Cunnane formula with
treme value distribution I (Gumbel), extreme value distribution II
a ¼ 0.4 was used for the GEV distribution (Chowdhury et al.
(Frechet), extreme value distribution III (Weibull), generalized Pa-
1991).
reto distribution, and so forth (Stedinger et al. 1993). In this study, Normalized Root Mean Square Error. NRMSE is a frequently
six distributions—lognormal, lognormal Type III (LN3), Pearson used goodness-of-fit method used by the Natural Environment Re-
Type III, log-Pearson Type III, extreme value distribution I/Gumbel search Council (NERC 1975), Karim and Chowdhury (1995), and
distribution, and generalized extreme value (GEV) distribution— Hosking and Wallis (1993) along with other methods for ranking
were used. These distributions have been widely used in hydrology the candidate distributions. While root mean square error (RMSE)
for a very long time (Benson 1968; Chang and Moore 1983; represents the dimensional standard deviation of the differences
Hosking et al. 1985; FEMA 2004; Mukherjee 2013; Izinyon and between computed and observed value, NRMSE is a dimension-
Ajumuka 2013). These six distributions are highly also used for the less statistic to show this difference with respect to a certain value.
frequency analysis of discharge or water level data in Bangladesh Different types of data are normalized differently in different liter-
(Karim and Chowdhury 1995; IFCDR 1995). The Gumbel distri- ature. Although there is no consistent means of normalization in the
bution is used by the BWDB, which is a mandated organization literature, common choices are the mean or the range (the differ-
for designing and implementing of all flood control projects and ence between the maximum and minimum value) of the observed
hydraulic structures in Bangladesh. The log-Pearson Type III dis- data (CIRP 2016). In this study, RMSE was normalized with re-
tribution has been used in the National Water Plan while the gen- spect to the corresponding observed value. A value of the NRMSE
eralized extreme value distribution has been used in the Flood near 0 suggests good fitness of the distribution. The NRMSE of a
Action Plan of Bangladesh. The probability distribution functions distribution fitted to the observed water level data at a station is
of the six distributions are presented in Table S2. computed using the following equation [Eq. (11)]:
X
of 2.33-year return period is good enough to fix the danger level. Danger Level Determination
But the elevation of the floodplain is also an important criterion, as Generally, the danger level of a location is set in a manner to min-
mentioned by Adhikari et al. (2014). They fixed the danger level of imize flood damage and risk to life safety. This requires estimation
the monsoon flood for the two major rivers of Nepal by comparing of the elevations for protected areas, then computing the return
the elevations of the floodplain with the levels from frequency period in which those elevations would be exceeded. Therefore, in
analysis. A similar approach was adopted in this study. Flood fre- this study, GEV distribution was first selected for conducting fre-
quency, along with the elevations of the river floodplains were con- quency analysis in order to set flash flood danger levels. Then, the
sidered and, finally, the danger levels were fixed considering the elevations of the bank lines of each of the 40 forecasting stations
return periods at which the water level reached the floodplain or were compared with the levels from the frequency analysis. Finally,
were nearer to the floodplain at full stage. the danger levels were fixed considering the return periods at which
the water level was reached or was nearer to floodplain at the full
stage.
Results and Discussions For example, the flooding scenarios of 2.33- to 100-years return
period was examined for Khaliajuri station (SW 72) on the Dhanu-
Data Homogeneity Analysis Baulai-Ghorautra river, as shown in Fig. 2(a). From the analysis, it
was found that at station SW 72 on the 2.33-year return period the
In this study, the homogeneity and adequacy of the water level data surface water level would be just at the floodplain stage. Therefore,
series of all gauge stations were checked by quantitative statistical for this station, water level at the 2.33-year return period was
tests for randomness, independence, and stationarity at 5% signifi- selected as the danger level.
cance level. For satisfying these criteria, nonparametric run test, Fig. 2(b) shows the water level scenarios of 2.33- to 100-year at
lag 1 auto correlation test, Mann-Kendall test, and Mann-Whitney Sutang Railway Bridge (SW 280) station on Sutang river. From the
test were performed. The results of these four nonparametric homo- figure, it is observed that at this station, the water surface would be
geneity test results are given in Table 1. just at the floodplain stage for a flood of 5-year return period.
The homogeneity test results showed that 37 stations out of the Hence, for this station, the danger level was defined at 5-year return
selected 40 stations show randomness in run test results (z value is period.
greater than 1.96). The independence test showed that four stations A similar assessment was done for the rest of the selected sta-
failed in independency test, whose value was greater than the prob- tions, and a flash flood danger level was assigned to each of these
able limits of correlation at 5% significance level. stations assessing different flooding scenarios of 2.33- to 100-year
In case of stationarity test, both Mann-Kendall and Mann- return period. Table 3 shows the proposed flash flood danger level
Whitney tests were computed. Nonparametric Mann-Kendall test at 40 stations along with the existing monsoon flood danger level at
showed that only 1 of the 40 stations showed significant trend while 20 stations updated in FFWC website.
the rest of the stations showed almost no upward or downward From Table 3, it is clearly observed that at most of the stations
trend. Thus, it can be concluded that the selected stations were free on 2.33-year flood level, the water surface is at the floodplain level
of any significant trend. On the other hand, the Mann-Whitney or nearer to floodplain level. Hence, for these stations, the proposed
test result showed that six stations were not stationary. Altogether, flash flood danger levels were marked at the 2.33-year flood level.
10 out of 40 stations failed in at least one test, while SW 131.5 failed For the rest of the stations, flash flood danger levels were fixed at
in all the tests of homogeneity. These stations could be thrown out 5-year return period. Fig. 3 shows the locations of these 40 stations
for further analysis. But considering their importance, these stations and surrounding river systems along with the danger level recom-
were not excluded from the analysis. So, the regional flood fre- mended for each of the stations.
quency analysis was conducted on all these 40 stations. The difference between the proposed flash flood danger level
and the existing danger level is shown in Fig. 4, which shows that
the flash flood danger level recommended in this study is much
Regional Flash Flood Frequency Analysis
lower than the existing monsoon flood danger level. At some sta-
As stated previously, a total of six widely used frequency distribu- tions the differences are more prominent, ranging from 0.15 to
tion functions—lognormal, lognormal Type III, Pearson Type III, 5.21 mPWD with an average difference of nearly 1.50 mPWD.
log-Pearson Type III, extreme value distribution I/Gumbel distribu- Though this difference is not much significant for monsoon flood,
tion, and generalized extreme value distribution—were fitted. The it can cause severe destruction to the Boro rice cultivation during a
best-fit regional distribution was selected applying three goodness- flash flood. Hence, the flash flood danger level should be fixed sep-
of-fit criteria: PPCC, NRMSE, and MAE. A ranking scale was used arately from the existing monsoon danger level considering the
for comparing the result of the goodness-of-fit outcomes. Consid- flash flood period only and should be updated in the FFWC website
ering the goodness-of-fit results at a site, each distribution was as soon as possible. Furthermore, it is also recommended that for
given a rank between 1 and 6. Rank 1 is for the best fitting distri- the purpose of safety, a warning level can be designed (0.3–0.5 m)
bution and Rank 6 is for the worst fitting one. Table 2 shows the below the proposed danger level on the basis of topography, actual
rank of distributions of the 40 selected gauge stations based on the field condition, and importance of the respective sites for Boro rice
goodness-of-fit analysis. cultivation. Concurrently, FFWC officials should use these warning
Table 1. Homogeneity (randomness, independence, and stationarity) test results of data series
Ramdomness test Independence test Stationarity test
© ASCE
Run test Lag 1 auto correlation Mann kendall Mann whitney
Mann-Kendall trend Sen’s slope
Auto Lower Upper
Station ID Station Name n Z Value 5% Significance Lag correlation r limit r1 limit r2 5% Significance Z Value 5% Significance Q Z value 5% Significance
SW 3 Brahmanbaria Rly Bridge 19 −1.0975 S 1 −0.0095 −0.5525 0.4275 S −0.9452 S −0.0200 1.38804 S
SW 34 Nakuagaon 19 −0.4291 S 1 −0.0772 −0.5525 0.4275 S −0.8407 S −0.0327 0.73485 S
SW 36 Jaria-Jhanjail 19 −0.2243 S 1 0.1511 −0.5525 0.4275 S 0.6297 S 0.0283 0.44907 S
SW 36.1 Mohanganj 19 0.7227 S 1 0.1359 −0.5525 0.4275 S −1.2610 S −0.0545 1.55134 S
SW 67 Kamalganj 19 −1.1713 S 1 0.4214 −0.5525 0.4275 S −0.7697 S −0.0342 2.00042 I
SW 72 Khaliajuri 18 −0.2243 S 1 0.2493 −0.5525 0.4275 S −0.2273 S −0.0113 1.34722 S
SW 73 Itna 19 1.0994 S 1 −0.0705 −0.5525 0.4275 S 0.2799 S 0.0117 0.53072 S
SW 74 Dilalpur 19 −0.2243 S 1 −0.0465 −0.5525 0.4275 S −0.3151 S −0.0053 0.9798 S
SW 131.5 Laurergarh Saktiarkhola 19 −2.1184 I 1 0.4546 −0.5525 0.4275 I −2.7656 I −0.1122 3.26599 I
SW 135A Juri_Cont_silghat 13 −0.0946 S 1 −0.0094 −0.7198 0.5198 S 1.2812 S 0.0984 −1.21429 S
SW 138 Sofiabad 18 −1.3105 S 1 0.6012 −0.5727 0.4394 I −0.4545 S −0.0610 1.94265 S
SW 157 Ballah 18 −1.8234 S 1 0.4402 −0.5727 0.4394 I −0.2273 S −0.0180 1.23623 S
SW 158 Shaistaganj 19 −0.6978 S 1 0.1654 −0.5525 0.4275 S −1.6093 S −0.0967 1.91877 S
SW 158.1 Chunarghat 19 −2.0674 I 1 0.2301 −0.5525 0.4275 S −0.6997 S −0.0562 1.67382 S
SW 159 Habiganj 19 −2.9765 I 1 0.4157 −0.5525 0.4275 S −0.8065 S −0.0367 2.00042 I
SW 172.5 Amalshid 19 0.2492 S 1 −0.1902 −0.5525 0.4275 S −0.2799 S −0.0386 0.85732 S
SW 173 Sheola 19 0.2492 S 1 −0.1911 −0.5525 0.4275 S 0.0000 S −0.0041 0.53072 S
SW 174 Fenchuganj 19 0.2492 S 1 −0.1374 −0.5525 0.4275 S 0.6297 S 0.0600 0.04082 S
SW 175.5 Sherpur 19 0.3573 S 1 −0.1214 −0.5525 0.4275 S 1.0152 S 0.0600 −0.04082 S
SW 192 Motiganj 11 −0.9331 S 1 0.1938 −0.8180 0.5680 S 0.0000 S −0.0060 0.63901 S
SW 201 Monu Rly. Bridge 19 0.3430 S 1 0.4275 S S 2.49031 I
05019004-7
−0.0612 −0.5525 −1.3994 −0.0727
SW 202 Moulvi Bazar 19 −0.6978 S 1 0.2544 −0.5525 0.4275 S −0.8397 S −0.0582 2.08207 I
SW 233A Jaflong_Spill 12 0.4472 S 1 −0.0092 −0.7644 0.5422 S −1.5772 S −0.0998 1.52122 S
SW 251 Sarighat 19 −0.6978 S 1 0.0821 −0.5525 0.4275 S 0.2099 S 0.0333 −0.04082 S
SW 252.1 Salutikar 17 −0.9740 S 1 0.0138 −0.5953 0.4524 S 0.9895 S 0.0642 0.04811 S
J. Hydrol. Eng.
Table 2. Rank of distributions for the 40 gauge stations based on goodness of fit analysis
Number of sites receiving ranks
Goodness of fit Distribution 1 2 3 4 5 6 Total of ranks
PPCC Lognormal 0 2 3 14 18 3 177
Lognormal Type III 2 7 3 9 14 5 161
Pearson Type III 8 18 9 4 1 0 92
Log-Pearson Type III 2 9 20 7 1 1 119
Gumbel distribution 2 0 2 2 5 29 215
Downloaded from ascelibrary.org by Bangladesh University of Engineering and Technology on 02/10/19. Copyright ASCE. For personal use only; all rights reserved.
Fig. 2. Water level at (a) Khaliajuri (SW 72); and (b) Sutang Railway Bridge (SW 280) station at different return periods.
Table 3. Recommended flash flood danger level for the northeast Haor regions
Proposed FFWC
© ASCE
flash flood Return Monsoon flood
Station danger level period danger level Nearby Haor
number River name Station name Station ID District (mPWD) (year) (mPWD) (submersible embankment crest level, mPWD)
1 Anderson khal Brahmanbaria Railway SW 3 Brahmanbaria 2.97 2.33 5.5 Dattakhola Haor
Bridge
2 Bhogai-Kangsa Nakuagaon SW 34 Sherpur 20.98 2.33 22.4 —
3 Bhogai-Kangsa Jaria-Jhanjail SW 36 Netrokona 8.08 5 9.75 —
4 Bhogai-Kangsa Mohanganj SW 36.1 Netrokona 5.81 2.33 — Medar Beel-2 Haor, Jaldhara/Keuti Haor, Dingi pota Haor,
Kankhola bil Haor (Sunair Haor system)
5 Dhalai Kamalganj SW 67a Moulvibazar 19.58 2.33 19.82 Kauadighi Haor (9.1–12.4).
6 Dhanu-Baulai-Ghorautra Khaliajuri SW 72 Netrokona 5.31 2.33 — Chakua Haor, Jagannathpur Haor, Kirtonkholar Haor,
Pangachia Haor
7 Dhanu-Baulai-Ghorautra Itna SW 73 Kishoreganj 4.46 2.33 — Uttarer Band Haor, Dakhshiner Band Haor
8 Dhanu-Baulai-Ghorautra Dilalpur SW 74 Kishoreganj 3.29 2.33 — Dulalpur Haor, Humairpur (4.83)
9 Jadukata Laurergarh Saktiarkhola SW Sunamganj 7.8 5 8.53 Angurali Haor (7.10), Lubar Haor, Matian Haor (7.3),
131.5a, b, c, d Balda Gulaghat Haor
10 Juri Juri_Cont_silghat SW 135A Moulavibazar 10.45 2.33 — Hakaluki Haor
11 Korangi Sofiabad SW 138c Habiganj 9.53 2.33 — Gungiajuri Haor
12 Khowai Ballah SW 157c Habiganj 21.93 2.33 21.64 —
13 Khowai Chunarghat SW 158 Habiganj 14.92 2.33 — Gungiajuri Haor
14 Khowai Shaistaganj SW 158.1b Habiganj 12.39 2.33 — Gungiajuri Haor
15 Khowai Habiganj SW 159a, b Habiganj 9.35 2.33 9.5 Ikram-Sangar Haor, Gazaria Beel Haor, Gungiajuri Haor
16 Kushiyara Amalshid SW 172.5 Sylhet 13.34 2.33 15.85 Balait Haor
17 Kushiyara Sheola SW 173 Sylhet 12.86 5 13.5 Nateshwar Haor, Adinabad Haor, Muria Haor
18 Kushiyara Fenchuganj SW 174 Sylhet 9.71 5 — Hakaluki Haor, Dhamrir Haor,
05019004-9
19 Kushiyara Sherpur SW 175.5 Sylhet 8.44 5 9 Banaiya Haor, Muktarpur Haor
20 Lungla Motiganj SW 192 Moulvibazar 8.7 2.33 — Hail Haor (9.76–10.36)
21 Monu Monu Railway Bridge SW 201a Moulvibazar 17.15 2.33 18 Kauadighi Haor (9.1–12.4)
22 Monu Moulvibazar SW 202a Moulvibazar 10.57 2.33 11.75 Kauadighi Haor (9.1–12.4)
J. Hydrol. Eng.
and danger levels during the flash flood forecasting and warning.
This will help the farmers to estimate the upcoming risk of flash
ered for saving crops of that area. Unfortunately, the crest levels of
the submersible embankments in the northeast region are not prop-
erly documented. Moreover, breaching of the embankments after
the flash flood is a common phenomenon. The fixation of the danger
level considering these incorrect elevations may provide ineffective
or misleading outcomes. In this context, we have not considered the
Monsoon flood
—
—
—
—
—
2.33
proposed flash flood danger level is 7.8 mPWD whereas the sub-
5
5
(mPWD)
Sunamganj
Netrokona
Netrokona
Sylhet
Hence, the proposed flash flood danger level will also serve the pur-
pose of the embanked stations.
Furthermore, the present study was completely done on observed
data and we used 3-h water level data. However, an extremely
Station ID
SW 337
SW 310
SW 326
SW 296
Lubachara
Conclusions
River name
Lubachara
Noyagang
Flash flood is the major natural disaster in the northeast Haor region
Mogra
Mogra
Titas
ture of the region and thereby jeopardizes the lives and livelihood of
Mann-Whitney test.
Mann-Kendall test.
crops from the flash floods, they are not fully capable of serving the
36
37
38
39
40
d
a
Fig. 3. Proposed flash flood danger level (mPWD) for the selected water level station based on the return period of the 2.33- or 5-year flood.
Fig. 4. Existing monsoon flood danger level and proposed flash flood danger level.
warnings are considered as one of the important means for risk re- recommended on the basis of flood frequency analysis and eleva-
duction. But, prior to that, establishing a reliable flash flood danger tion of the nearby river floodplain. The result shows that the rec-
level for the northeast region is essential. Hence, in this study, a ommended flash flood danger level is much lower than the existing
reliable danger level at 40 water level stations in the region has been monsoon flood danger level. At some stations, the difference is
flood danger level. Furthermore, it is also recommended that for the impact on the Haor area: A case study of Karchar Haor project.” M.Sc.
purpose of safety, a warning level can be designed (0.3–0.5 m) be- thesis, Institute of Water and Flood Management, Bangladesh Univ. of
low the proposed danger level on the basis of topography, actual Engineering and Technology.
field condition, and importance of the respective sites for Boro rice Alam, M. S., M. A. Quayum, and M. A. Islam. 2010. “Crop production in
cultivation. However, one of the limitations of this study is that the the Haor areas of Bangladesh: Insights from farm level survey.”
danger level is solely determined on the basis of flood frequency Agriculturists 8 (2): 88–97. https://doi.org/10.3329/agric.v8i2.7582.
analysis and the elevations of the river floodplains. For fixing up the Anderson, R. L. 1942. “Distribution of serial correlation coefficient.” Ann.
danger level, the crest levels of the submersible embankments of the Math. Stat. 13 (1): 1–13. https://doi.org/10.1214/aoms/1177731638.
nearby Haors should also be considered. However, the crest levels of Armstrong, W. H., M. J. Collins, and N. P. Snyder. 2012. “Increased
most of the submersible embankments are not correctly measured frequency of low-magnitude floods in New England.” J. Am. Water Re-
sour. Assoc. 48 (2): 306–320. https://doi.org/10.1111/j.1752-1688.2011
and change with time due to subsidence or improper maintenance.
.00613.x.
In the current study, some of the submersible embankments’ eleva-
Baten, M. A., N. A. Khan, R. Ahammad, and K. Misbahuzzaman. 2010.
tions were collected from BWDB, and, at some stations, the eleva- Village common forests in Chittagong hill tracts, Bangladesh: Balance
tions of the proposed danger levels were cross checked with those of between conservation and exploitation. Dhaka, Bangladesh: Unnayan
the submersible embankments. However, these are not adequate to Onneshan-The Innovators.
provide a complete idea of that region. Hence, it is felt that surveys Benson, M. A. 1968. “Uniform flood-frequency estimating methods for
on a regular basis are required to collect the crest levels of the sub- federal agencies.” Water Resour. Res. 4 (5): 891–908. https://doi.org/10
mersible embankments. The crest levels of submersible embank- .1029/WR004i005p00891.
ments at a given location also need continuous verification. Besides, Black, P. E. 1996. Watershed hydrology. 2nd ed. New York: Taylor and
contemporary real-life case studies of modeling techniques could Francis.
enrich the quality of this study by capturing a better flood peak. Bradley, J. V. 1968. Distribution-free statistical tests. No. 04; QA278.
It is expected that incorporation of recent land topography and de- 8, B7. Englewood Cliffs, NJ: Prentice-Hall.
tailed and updated elevations of the submersible embankments BWDB (Bangladesh Water Development Board). 2006. Mathematical
along with incorporation of modeling techniques may give more modelling with hydrological studies and terrestrial survey under the
Haor rehabilitation scheme. Dhaka, Bangladesh: BWDB.
accurate danger level and thereby help the flood forecasting author-
CARE (Cooperative for Assistance and Relief). 2010. “Project completion
ity in precise flash flood forecasting and early warning and protect
flood risk reduction activities in Sunamganj (FRRAS) phase-II.” Ac-
the Boro crop of the northeast region from the flash flood. cessed January 19, 2019. http://www.carebangladesh.org/publication
/Publication_4718927.pdf.
CEGIS (Center for Environmental and Geographic Information Serv-
Acknowledgments ices). 2012. Master plan of Haor Area. Volume 1, summary report.
Dhaka, Bangladesh: Government of People’s Republic of Bangladesh,
The work has been supported by a research project on ‘Developing Ministry of Water Resources, Bangladesh Haor and Wetland Develop-
Flash Flood Early Warning System, Capacity Building and Knowl- ment Board.
edge Management for Haor Region of Bangladesh’ currently being Challis, R. E., and R. I. Kitney. 1990. “Biomedical signal processing (in
carried out in Institute of Water and Flood Management (IWFM) of four parts).” Med. Biol. Eng. Comput. 28 (6): 509–524. https://doi.org
Bangladesh University of Engineering and Technology funded by a /10.1007/BF02442601.
Haor Area Infrastructure and Livelihood Improvement (HILIP) Chang, S. K., and S. M. Moore. 1983. “Flood frequency analysis for small
Project of Local Govt. Engineering Department (LGED) sponsored watersheds in southern Illinois.” Water Resour. Res. 19 (2): 277–282.
by the International Fund for Agricultural Development (IFAD). https://doi.org/10.1111/j.1752-1688.1983.tb05326.x.
Chau, K. W. 2007. “A split-step particle swarm optimization algorithm in
The authors would like to acknowledge the use of hydrological data
river stage forecasting.” J. Hydrol. 346 (3–4): 131–135. https://doi.org
from Bangladesh Water Development Board (BWDB). /10.1016/j.jhydrol.2007.09.004.
Chen, X. Y., K. W. Chau, and A. O. Busari. 2015. “A comparative study of
population-based optimization algorithms for downstream river flow
Supplemental Data forecasting by a hybrid neural network model.” Eng. Appl. Artif. Intell.
46: 258–268. https://doi.org/10.1016/j.engappai.2015.09.010.
Figs. S1 and S2 and Tables S1–S4 are available online in the ASCE Chowdhury, J. U., J. R. Stedinger, and L. H. Lu. 1991. “Goodness-of-fit
Library (www.ascelibrary.org). tests for regional generalized extreme value flood distribution.” Water
Resour. Res. 27 (7): 1765–1776. https://doi.org/10.1029/91WR00077.
CIRP (Coastal Inlets Research Program). 2016. Coastal Inlets Research
References Program (CIRP) Wiki—Statistics. Washington, DC: US Army Corps
of Engineers.
Adeloye, A. J., and M. Montaseri. 2002. “Preliminary stream flow data Cunnane, C. 1978. “Unbiased plotting positions—A review.” J. Hydrol.
analyses prior to water resources planning study.” Hydrol. Sci. J. 37 (3–4): 205–222. https://doi.org/10.1016/0022-1694(78)90017-3.
47 (5): 679–692. https://doi.org/10.1080/02626660209492973. Cunnane, C. 1989. Statistical distributions for flood frequency analysis.
Adhikari, T., B. Rakhal, L. D. Maharjan, and P. S. Kushi. 2014. Determi- WMO No. 718; Operational Hydrology Rep. (OHR) No. 33. Geneva:
nation of flood warning and danger of Mohana and Macheli river. World Meteorological Organization.
doi.org/10.1016/S1462-0758(00)00004-2. tests for time series analysis: Application to hydrological time series.”
FEMA. 2004. Federal Emergency Management Agency of the United Hydrol. Sci. J. 53 (2): 353–366. https://doi.org/10.1623/hysj.53.2.353.
States. Washington, DC: FEMA. Mann, H. B. 1945. “Non-parametric tests against trend.” Econmetrica
FFWC (Flood Forecasting and Warning Centre). 2018. “Bangladesh water 13 (3): 245–259. https://doi.org/10.2307/1907187.
development board.” Accessed January 19, 2019. http://www.ffwc Mann, H. B., and D. R. Whitney. 1947. “On a test of whether one of two
.gov.bd/. random variables is stochastically larger than the other.” Ann. Math.
Filliben, J. J. 1975. “The probability plot correlation test for normality.” Stat. 18 (1): 50–60. https://doi.org/10.1214/aoms/1177730491.
Technometrics 17 (1): 111–117. https://doi.org/10.1080/00401706 McGhee, J. W. 1985. Introductory statistics. New York: West Publishing.
.1975.10489279. McMahon, T. A., and R. Srikanthan. 1981. “Log Pearson III distribution—
Fotovatikhah, F., M. Herrera, S. Shamshirband, K. W. Chau, S. Is it applicable to flood frequency analysis of Australian streams?”
Faizollahzadeh Ardabili, and M. J. Piran. 2018. “Survey of computa- J. Hydrol. 52 (1–2): 139–147. https://doi.org/10.1016/0022-1694(81)
tional intelligence as basis to big flood management: Challenges, re- 90100-1.
search directions and future work.” Eng. Appl. Comp. Fluid Mech. Mondal, A., S. Kundu, and A. Mukhopadhyay. 2012. “Rainfall trend
12 (1): 411–437. https://doi.org/10.1080/19942060.2018.1448896. analysis by Mann-Kendall test: A case study of north-eastern part
Gumbel, E. J. 1945. “Flood estimated by probability methods.” Eng. of Cuttack district, Orissa.” Int. J. Geol. Earth Environ. Sci. 2 (1):
News-Rec. 134 (24): 97–101. 70–78.
Haan, C. T. 1986. Statistical methods in hydrology. Ames, IA: Iowa State Mondal. M. K., A. F. M. Saleh, M. N. Hossan, and S. I. Bhuiyan. 2004.
University Press. “Strategies for increasing the productivity of rice areas affected by flash
Hamed, K. H. 2008. “Trend detection in hydrologic data: The Mann- flood.” In Rice research and development in the flood-prone ecosystem,
Kendall trend test under the scaling hypothesis.” J. Hydrol. 349 (3–4): edited by S. I. Bhuiyan. Los Baños, Philippines: International Rice
350–363. https://doi.org/10.1016/j.jhydrol.2007.11.009.
Research Institute.
Hosking, J. R. M., and J. R. Wallis. 1993. “Some statistics useful in regional
Montz, B. E., and E. Gruntfest. 2002. “Flash flood mitigation: Recommen-
frequency analysis.” Water Resour. Res. 29 (2): 271–281. https://doi
dations for research and applications.” Global Environ. Change Part B:
.org/10.1029/92WR01980.
Environ. Hazard. 4 (1): 15–22. https://doi.org/10.1016/S1464-2867(02)
Hosking, J. R. M., J. R. Wallis, and E. F. Wood. 1985. “An appraisal
00011-6.
of the regional flood frequency procedure in the UK Flood Studies
Moses, T., and A. Klockars. 2008. “An evaluation of standard, alternative,
Report.” Hydrol. Sci. J. 30 (1): 85–109. https://doi.org/10.1080
and robust slope test strategies.” J. Mod. Appl. Stat. Methods 7 (1):
/02626668509490973.
77–93. https://doi.org/10.22237/jmasm/1209614760.
Hossain, G. M., and A. Nishat. 1996. “Environmental considerations for
Mukherjee, M. K. 2013. “Flood frequency analysis of River Subernarekha,
water resources development in Haor areas of northeastern Bangladesh.”
In Proc., North American Water and Environment Congress and India, using Gumbel’s extreme value distribution.” Int. J. Comput. Eng.
Destructive Water, 1063–1068. New York: ASCE. Res. 3 (7): 12–19.
Huda, M. K. 2004. “Experience with modern and hybrid rice varieties in NERC (Natural Environment Research Council). 1975. Flood studies
Haor ecosystem: Emerging technologies for sustainable rice produc- report. Vol. 1. London: Natural Environment Research Council.
tion.” In Proc., Twentieth National Workshop on Rice Research and NERP FAP - 6 (Northeast Regional Water Management Plan, Bangladesh
Extension in Bangladesh, 19–21. Gazipur, Bangladesh: Bangladesh Flood Action Plan 6). 1995. Final Rep. Dhaka, Bangladesh: NERP
Rice Research Institute. FAP - 6.
Hussain, M., and A. Salam. 2007. Basic service delivery advocacy: Review Ngongondo, C. S., C. Y. Xu, L. M. Tallaksen, B. Alemaw, and T. Chirwa.
report. Dhaka, Bangladesh: DEW. 2011. “Regional frequency analysis of rainfall extremes in Southern
IFCDR (Institute of Flood Control and Drainage Research). 1995. “Flood Malawi using the index rainfall and L-moments approaches.” Stochastic
frequency analysis.” Vol. 1 of Component of the study on revision Environ. Res. Risk Assess. 25 (7): 939–955. https://doi.org/10.1007
of flood danger levels in Bangladesh, edited by J. U. Chowdhury, /s00477-011-0480-x.
M. Y. Rana, and M. Salehin. Dhaka, Bangladesh: BWDB. Nowreen, S., S. B. Murshed, A. K. M. S. Islam, and B. Bhaskaran. 2013.
Izinyon, O. C., and H. N. Ajumuka. 2013. “Probability distribution models “Change of future climate extremes for the Haor Basin area of
for flood prediction in Upper Benue river basin.” Civ. Environ. Res. Bangladesh.” In Vol. 2 of Proc., 4th Int. Conf. on Water and
3 (2): 63–74. Flood Management (ICWFM 2013), 545–556. Dhaka, Bangladesh:
Karim, M. A., and J. U. Chowdhury. 1995. “A comparison of four distri- Bangladesh Univ. of Engineering and Technology.
butions used in flood frequency analysis in Bangladesh.” Hydrol. Sci. J. Nowreen, S., S. B. Murshed, A. S. Islam, B. Bhaskaran, and M. A. Hasan.
40 (1): 55–66. https://doi.org/10.1080/02626669509491390. 2015. “Changes of rainfall extremes around the Haor Basin areas of
Kendall, M. G. 1975. Rank correlation methods. 4th ed. London: Charles Bangladesh using multi-member ensemble RCM.” Theor. Appl. Clima-
Griffin. tol. 119 (1–2): 363–377. https://doi.org/10.1007/s00704-014-1101-7.
Khan, A. A. 2010. “Impact of a submersible embankment project on Haor NWS (National Weather Service). 2018. National weather service. Silver
based livelihood.” M.Sc. thesis, Institute of Water and Flood Manage- Spring, MD: NOAA.
ment, Bangladesh Univ. of Engineering and Technology. Rahman, R., and M. Salehin. 2013. “Flood risks and reduction ap-
Khan, M. N. H., M. Y. Mia, and M. R. Hossain. 2012. “Impacts of flood proaches in Bangladesh.” In Disaster risk reduction approaches in
on crop production in Haor areas of two Upazillas in Kishoregonj.” Bangladesh, edited by R. Shaw, F. Mallick, and A. Islam. Tokyo:
J. Environ. Sci. Nat. Resour. 5 (1): 193–198. https://doi.org/10.3329 Springer.
/jesnr.v5i1.11581. Rossi, F., M. Fiorentino, and P. Versace. 1984. “Two-component extreme
Kryžanowski, A., M. Brilly, S. Rusjan, and S. Schnabl. 2014. “Structural value distribution for flood frequency analysis.” Water Resour. Res.
flood-protection measures referring to several European case studies.” 20 (7): 847–856. https://doi.org/10.1029/WR020i007p00847.