flood frequency analysis at river confluences – univariate vs

14
316 DOI: 10.5675/ICWRER_2013 Excerpt of the full paper "ICWRER 2013 | Proceedings" Download: www.water-environment.org/proceedings.html ICWRER 2013 | Flood Frequency Analysis at River Confluences … Flood Frequency Analysis at River Confluences – Univariate vs. Multivariate Extreme Value Statistics Jens Bender 1,2 · Thomas Wahl 2,3 · Christoph Mudersbach 1 · Jürgen Jensen 1,2 1 University of Siegen, Research Centre for Water and Environment · 2 University of Siegen, Institute of Advanced Studies · 3 University of South Florida, College of Marine Science Abstract In this study we analyze the combined flood probability at a confluence using different statistic procedures. The study is exemplarily carried out at the confluence of the rivers Ilz and Wolfsteiner Ohe in Germany, where long time series of the hourly discharge are available at both rivers upstream of the confluence as well as downstream. On the one hand we perform a univariate statistical flood frequency analysis upstream and downstream of the confluence as it is applied commonly at major German rivers. The aim is to determine the statistical relevant inflow of the tributary for several given design discharges at the main stream. On the other hand we perform a bivariate statistical analysis using Archimedean copula functions at both streams upstream of the confluence. Comparing the results highlights the limited capability of the univariate approach to determine the statistical relevant inflows from the tributary. In particular for higher return period discharges at the main stream, the resulting inflows from the tributary differ from the results of the bivariate statistical analysis. 1. Introduction Design discharges at rivers are required for many engineering purposes, e.g. numerical inundation modeling, design of flood protection structures, etc. As an alternative to rainfall-runoff modeling, a broad variety of univariate statistical methods are available to determine the design discharges (e.g. HQ100) along rivers based on recorded discharge data (see e.g. Rao and Hamed, 2000). However, gauge data upstream of a confluence does not provide information on the statistical relevant inflow from the tributary. Especially for flow modeling along a river the inflow boundary conditions of tributaries are of great interest. Not at least because inflows affect the water levels downstream of the confluence and can cause hydraulic effects, like turbulences and tailbacks. In Germany only few approaches exist to estimate the statistical relevant discharge boundary conditions at a tributary for a given design flood at the main stream and vice versa. One of those approaches is the so called "Confluence Formula" (German: Mündungsformel). Today, the Confluence Formula is a widely used approach to estimate the discharge conditions at river confluences in Germany. The formula was developed by the Federal Department

Upload: donhi

Post on 04-Jan-2017

236 views

Category:

Documents


8 download

TRANSCRIPT

Page 1: Flood Frequency Analysis at River Confluences – Univariate vs

316

DOI: 10.5675/ICWRER_2013

Excerpt of the full paper "ICWRER 2013 | Proceedings"Download: www.water-environment.org/proceedings.html

ICWRER 2013 | Flood Frequency Analysis at River Confluences …

Flood Frequency Analysis at River Confluences – Univariate vs. Multivariate Extreme Value Statistics

Jens Bender1,2 · Thomas Wahl2,3 · Christoph Mudersbach1 · Jürgen Jensen1,2

1 University of Siegen, Research Centre for Water and Environment · 2 University of Siegen, Institute of Advanced Studies · 3 University of South Florida, College of Marine Science

AbstractIn this study we analyze the combined flood probability at a confluence using different

statistic procedures. The study is exemplarily carried out at the confluence of the rivers

Ilz and Wolfsteiner Ohe in Germany, where long time series of the hourly discharge

are available at both rivers upstream of the confluence as well as downstream. On

the one hand we perform a univariate statistical flood frequency analysis upstream

and downstream of the confluence as it is applied commonly at major German rivers.

The aim is to determine the statistical relevant inflow of the tributary for several given

design discharges at the main stream. On the other hand we perform a bivariate

statistical analysis using Archimedean copula functions at both streams upstream of

the confluence. Comparing the results highlights the limited capability of the univariate

approach to determine the statistical relevant inflows from the tributary. In particular

for higher return period discharges at the main stream, the resulting inflows from the

tributary differ from the results of the bivariate statistical analysis.

1. Introduction

Design discharges at rivers are required for many engineering purposes, e.g. numerical

inundation modeling, design of flood protection structures, etc. As an alternative to

rainfall-runoff modeling, a broad variety of univariate statistical methods are available to

determine the design discharges (e.g. HQ100) along rivers based on recorded discharge

data (see e.g. Rao and Hamed, 2000). However, gauge data upstream of a confluence does

not provide information on the statistical relevant inflow from the tributary. Especially

for flow modeling along a river the inflow boundary conditions of tributaries are of great

interest. Not at least because inflows affect the water levels downstream of the confluence

and can cause hydraulic effects, like turbulences and tailbacks. In Germany only few

approaches exist to estimate the statistical relevant discharge boundary conditions at

a tributary for a given design flood at the main stream and vice versa. One of those

approaches is the so called "Confluence Formula" (German: Mündungsformel). Today, the

Confluence Formula is a widely used approach to estimate the discharge conditions at

river confluences in Germany. The formula was developed by the Federal Department

Page 2: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 317

DOI: 10.5675/ICWRER_2013

of Environment of the state Baden-Wuerttemberg on basis of gauge records at a single

confluence of two rivers. Due to the lack of available methods, however, the Confluence

Formula is often applied to other ungauged rivers neglecting different topographical and

hydrological boundary conditions of the catchments.

Several studies have been carried out considering the bivariate nature of this concern.

Morris and Calise (1987) estimated the joint flood probability at a tributary, which is

influenced by the water level in the main stream. They used bivariate density functions to

describe the dependence between the water level in the main stream and the tributary.

Raynal and Salas (1987) were the first who used the bivariate General Extreme Value

(bGEV) distribution in the context of joint flood risk at the confluence of the rivers Bear

and Dry Creek in California, USA. They compared four approaches to estimate the total

discharge downstream of a confluence: (i) with the sum of the recorded discharges, (ii)

under the assumption of a perfect linear dependence (i.e. correlation of one) between

the main stream and tributary discharges, (iii) under assumption of independence

(correlation of zero) between the discharges, and (iv) with the use of the bGEV considering

a correlation of r = 0.86. They suggested that the correlation coefficient was a suitable

measure to describe the dependence and they proposed the application of the bivariate

Normal distribution function for estimating the joint probability of floods at confluences.

Over the last years copula functions have been used for several multivariate hydrological

analyses. They were applied for rainfall frequency analysis (e.g. De Michele and Salvadori,

2003, Grimaldi and Serinaldi, 2006, Zhang and Singh, 2007), flood frequency analysis

considering peak flow and flood volume (e.g. Favre et al., 2004, Zhang and Singh, 2006,

Karmakar and Simonovic, 2009), drought frequency analysis (e.g. Shiau, 2006, Kao and

Govindaraju, 2010, Song and Singh, 2010a, b), storm surge modelling (e.g. Wahl et al.,

2012), and for several other multivariate problems.

For flood risk analyses at river confluences copulas were already used by Wang et al.

(2009). They presented a copula-based algorithm to determine the joint probability at a

river confluence using copulas of the Archimedean family. Using Monte-Carlo simulations

they determined the joint probability of a concurrent occurrence of flood events in both

streams.

NCHRP (2010) first presented a general approach to estimate the joint flood risk at

ungauged rivers for the design of road drainage structures. The study was based on 83

homogeneous distributed gauge pairs throughout the USA. Basically three multivariate

Page 3: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 318

DOI: 10.5675/ICWRER_2013

methods were applied: bivariate distribution functions, copula functions and the total

probability method. For the generalization of the results from individual gauge pairs they

used the ratio of the confluent catchments.

Chen et al (2012) used 4-dimensional Copula functions to model the coincident risk of

concurrent occurring flood magnitudes and dates at confluences. Case studies were the

upper Yangtze River in China and the Colorado River in the United States.

At German major rivers, where a dense network of discharge gauges exists, this bivariate

issue is however still reduced to a univariate problem. This allows the application of

common extreme value models to the data sets from all gauges along the stream. The

difference between two design discharges derived by two neighbored gauges is then

considered as the statistical relevant discharge of all tributaries between the gauges.

If there is more than one tributary between two gauges the inflow of the individual

tributaries is weighted according to the catchment sizes (BfG, 2009).

In the present study we conduct two analyses of the joint flood occurrence at a gauged

river confluence in Germany. First, the simplified univariate approach is applied as it is

done at the German major rivers. In a second step we use Archimedean copula functions

to determine the joint flood risk. The main intention of this study is to compare the

results of both approaches to determine their capability of assessing the joint flood risk

at the confluence.

2. Data

2.1. General This study is exemplarily carried out at the confluence of the river Ilz and its tributary

Wolfsteiner Ohe in Bavaria, Germany. Hourly discharge data are available at river Ilz

some 2.2 km upstream (gauge Schrottenbaummühle) as well as 3.6 km downstream

(gauge Kalteneck) of the confluence. The river Wolfsteiner Ohe has a gauge (Fürsteneck)

approx. 2.0 km upstream the confluence with the river Ilz (see Figure 1). All time series

were provided by the Federal Environment Authority of Bavaria and cover the period

from 1972 to 2011, i.e. 40-yr long time series without gaps or discontinuities are available

for the analyses.

Page 4: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 319

DOI: 10.5675/ICWRER_2013

Figure 1 Schematic illustration of gauge locations at the confluence of the rivers Ilz and Wolfsteiner Ohe.

River Ilz drains at gauge Schrottenbaummühle a total area of 364 km² upstream of the

confluence, whereas river Wolfsteiner Ohe contributes at gauge Fürsteneck an almost

equal drainage area of 370 km². The mean discharge at gauge Schrottenbaummühle

amounts to MQs = 7.6 m³/s with a standard deviation of ss = 7.7 m³/s. At gauge Fürsteneck and

gauge Kalteneck the mean discharges and the standard deviations amount to MQF = 8.5 m³/s, sF = 8.0 m³/s, and MQK = 16.6 m³/s, sK = 16.3 m³/s, respectively. The highest discharges

at all three gauges were observed on 21st December 1993 with HHQS = 208.1 m³/s at

gauge Schrottenbaummühle and HHQF = 192.4 m³/s at gauge Fürsteneck which in turn

resulted in a maximum discharge at gauge Kalteneck of HHQK = 416.9 m³/s. Although the

highest measured flood peaks occurred at the same day, thorough investigations show

that 13 of the total 40 annual maximum flood peaks at rivers Ilz and Wolfsteiner Ohe did

not occur concurrently within a time frame of ±7 days.

2.2. Measurement Inaccuracies

A general problem in flood frequency analysis based on discharge data sets is related to

inaccuracies of the measurements. In particular during extreme events the extrapolated

water level-discharge function (discharge curve) often leads to large errors since discharge

curves are calibrated on measurements during low or medium runoffs (Maidment, 1992).

In the case at hand considerable discrepancies between the sum of the gauge data

upstream and the gauge data downstream of the confluence can be found, especially

during flood events. Although no information is available on incorrect operation of any

of the gauges, these discrepancies would affect the results of the statistical analysis. For

that reason and by having the exemplary nature of this study in mind, the discharge

data at gauge Kalteneck is replaced by a synthetic time series derived by the sum of the

discharges observed at the two gauges located upstream:

Q t Q t Q tKalteneck syn Schrottenbaummühle Fürsteneck( ) ( ) ( )= +

,��� (1)

Page 5: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 320

DOI: 10.5675/ICWRER_2013

3. Methods

3.1. Simplified univariate ApproachThe univariate approach only uses the discharge time series at the main stream, up- and

downstream of the confluence, which are in this case the gauges Schrottenbaummühle

and Kalteneck (synthetic). A common univariate flood frequency analysis is carried out

for both time series using extreme value distribution functions. The annual maximum

discharges (AMAX) of the hydrological years (in Germany from 1 November until 30

October) is considered as the relevant flood indicator. For catchments smaller than

50,000 km², as it is the case here, Svensson et al. (2005) suggested considering a minimum

time interval of five days as independence criterion in order to separate consecutive

flood events.

In a next step, univariate distribution functions are fitted to the AMAX series of the gauges

at the main river upstream and downstream of the confluence. Although the General

Extreme Value distribution (GEV) is often being treated as one of the main distribution

functions for extreme value analyses (e.g. Coles, 2001), other distribution functions are

additionally fitted to the data set. The most appropriate distribution function is identified

by the minimum root mean square error (RMSE) of the empirical distribution and the

parametric distribution function. The distribution parameters are estimated with the

maximum likelihood approach (e.g. Rao and Hamed, 2000) and the plotting positions are

derived by following the approach proposed by Gringorten (1963). Further Information

about flood frequency analyses can be found e.g. in Rao and Hamed (2000).

Next, all relevant quantiles of the parametric distribution functions are determined. In

this study we focus on the common quantiles in flood frequency analyses for the non-

exeedance probabilities of P = 0.5, 0.8, 0.9, 0.95, 0.98 and 0.99 with the corresponding

annual return periods of T = 2, 5, 10, 20, 50 and 100 years. Since higher return periods

then these are usually of minor interest for design purposes, they are not considered in

this study. The difference of the downstream quantiles and the upstream quantiles are

treated as the statistical relevant inflow of the tributary for a given flood event at the

main stream with an annual return period of T.

3.2. Bivariate Approach using Copula Functions

3.2.1. Theoretical Background

Copulas are flexible joint distributions for modeling the dependence structure of two or

even more random variables. First mentioned by Sklar (1959), the joint behavior of two (or

Page 6: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 321

DOI: 10.5675/ICWRER_2013

more) random variables X and Y with continuous marginal distributions u = FX(x) = P(X ≤ x) and v = FY(y) = P(Y ≤ y) can be described uniquely by an associated dependence function

or copula-function C. In the bivariate case, the relationship between all (u,v) Є [0,1]² can

be written as

F x y C F x F y C u vX Y X Y, , , ,( ) ( ) ( )[ ] ( )= = (2)

where FX,Y (x,y) is the joint cumulative distribution function (cdf) of the random variables

X and Y.

A copula function with a strictly monotonically decreasing generator function

φ: [0,1] → [0,∞] with φ(1) = 0 belongs to the Archimedean Copula family. The general

form of one-parametric Archimedean copulas is

C u v u vθ ϕ ϕ ϕ, -( ) ( ) ( )[ ]= +1 (3)

where θ denotes the copula parameter. In this study three Archimedean copulas, namely

the Clayton, Frank, and Gumbel copulas are considered. They are relatively easy to

construct, flexible and capable to cover the full range of tail dependencies. The Clayton

copula has lower tail dependence, while the Frank copula has no tail dependence and

the Gumbel copula has strong upper tail dependence (Schölzl and Friedrichs, 2008). The

copula parameters are estimated based on the inversion of Kendall’s τ. This is possible as

there exists an expression for τ as a function of θ for Archimedean copulas (see Table 1).

Page 7: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 322

DOI: 10.5675/ICWRER_2013

Table 1: Archimedean copula functions considered for the present study and their generator functions, ranges for the copula parameters θ and functional relationship to Kendall’s τ.

Copula function Cθ Generator φ(t)**

Range of θ Functional relationship of θ to τ

Clayton or Cook-Johnson

u v− −−

+ − θ θ θ1

1

1θt − − 0,∞[ ) 2θ

θ +

Frank

( ) ( )1 11ln 1

1

θu θv

θ

e e

θ e

− −

− −− +

1ln1

θt

θee

−− −

( ) { }0, \−∞ ∞ ( )[ ]1

41 1 D θ

θ− − *

Gumbel or Gumbel-Hougaard

( ) ( )1

exp ln lnθ θ θu v− − + −

( )ln θt− [ )1,∞ 11 θ −−

* 1. Debye Function ( )1

0

1

1

θ

t

tD θ dt

θ e=

−∫** t = u or t = v

Further important features of copulas and information about the theoretical background

can be found e.g. in Nelsen (1999), who provided a detailed introduction to the subject.

4. Results

4.1. The univariate approachIn the univariate case we fitted distribution functions to the AMAX series of gauges

Schrottenbaummühle and the synthetic gauge Kalteneck. In both cases the GEV

distribution provides the best fit. Figure 2 shows the result from fitting the GEV including

the upper and lower 95% confidence levels.

Page 8: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 323

DOI: 10.5675/ICWRER_2013

Figure 2 Fitted GEV distributions to AMAX series of gauge Schrottenbaummühle upstream of the confluence (A) and the synthetic gauge Kalteneck downstream of the confluence (B) including the upper and lower 95% confidence levels.

Table 2 shows the numeric quantiles of both fitted distributions. It can be seen that the

analyzed quantiles of the downstream located synthetic gauge Kalteneck varies between

151.1 m³/s and 390.7 m³/s for P = 0.5 (T = 2 a) and 0.99 (T = 100 a), respectively. At gauge

Schrottenbaummühle the values vary between 80.6 m³/s and 214.7 m³/s for the same

quantiles.

Table 2: Quantiles of the fitted distributions at gauge Schrottenbaummühle and the synthetic gauge Kalteneck and their differences

P 0.5 0.8 0.9 0.95 0.98 0.99 [ - ]

Return period 2 5 10 20 50 100 [a]

HQT - Kalteneck (syn.) 151.1 207.8 248.2 289.1 345.6 390.7 [m³/s]

HQT – Schrottenbaumm. 80.6 111.1 133.3 156.2 188.3 214.5 [m³/s]

Difference 70.5 96.7 114.9 132.9 157.3 176.2 [m³/s]

Page 9: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 324

DOI: 10.5675/ICWRER_2013

The difference of the same quantiles is than treated as the statistical relevant inflow of

river Wolfsteiner Ohe for the corresponding flood events at river Ilz. These values range

between Q = 70.5 m³/s (P = 0.5, T = 2 a) and Q = 176.2 m³/s (P = 0.99, T = 100 a).

4.2. The bivariate approach using Copulas

In contrast to the univariate approach, the time series of gauge Kalteneck does not

play a role in the bivariate analysis. In this case, only the discharge series of both

gauges upstream of the confluence are considered (gauges Schrottenbaummühle and

Fürsteneck). Before choosing a suitable copula type, the marginal distributions need

to be fitted. Here, in turn, the question arises which data should be modeled. In many

bivariate statistical investigations mostly the AMAX series of both variables are modeled,

e.g. the annual maximum peak flow series and the annual maximum flood volume series,

despite considering possible different background of their genesis and physical relations.

Here, we model the AMAX series of the main stream at gauge Schrottenbaummühle

together with the concurrent flows of the tributary at gauge Fürsteneck. The fact that

13 of the total 41 annual maxima values did not occur within a time frame of ±7 days

confirms that this procedure is reasonable. For the sake of minimizing the randomness

of the concurrent discharges, we consider the maximum discharge at gauge Fürsteneck

within a time frame of ±7 days with reference to the occurrence time of the AMAX

values at gauge Schrottenbaummühle. We already showed that the GEV fits the AMAX

series at gauge Schrottenbaummühle best (see Figure 2). For the gauge Fürsteneck, the

2-parametric Weibull distribution provides the best fit (see Figure 3).

Figure 3 Fitted 2-parametric Weibull distribution to concurrent discharge values at gauge Fürsteneck with reference to the occurrence of the AMAX values at gauge Schrottenbaummühle

Page 10: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 325

DOI: 10.5675/ICWRER_2013

In order to choose the most appropriate Archimedean copula function, we applied the

Bayesian method for copula selection as suggested by Huard et al. (2005). The Gumbel

copula appears to the best model to describe the dependence structure between the

AMAX values at gauge Schrottenbaummühle and the concurrent discharges at gauge

Fürsteneck. With a given value of Kendall’s τ = 0.567 the copula parameter θ can be

calculated using the functional relationship of τ and the Gumbel copula parameter as

outlined in Table 1 to θ = 2.3077.The results from fitting the Gumbel copula (with the

above mentioned marginal distributions) to the data set are illustrated in Figure 4.

Figure 4 Isolines of equal return periods derived by the bivariate analysis compared with the results of the univariate approach.

The black crosses in Figure 4 show the observed AMAX values at gauge Schrotten-

baummühle and the concurrent discharges at gauge Fürsteneck. The grey dots are 10,000

synthetic values derived from the fitted copula and the marginal distribution functions

giving an optical impression of the goodness of the fit. The black lines illustrate the

bivariate isolines of equal quantiles or return periods, respectively. Here, the bivariate

return period of a certain quantile is defined as the inverse value of the non-exeedance

probability. Other approaches defining the return period in multivariate cases can be

found in Gräler et al. (2013).

For comparison purposes the blue crosses illustrate the results of the univariate method

as described in section 5.1. It can be seen, that for small return periods (up to T = 20 a),

the probabilities of the concurrent flows (on the ordinate) of both approaches agree

well. However, the values for higher return periods, i.e. T = 50 and 100 a, deviate by

up to 20 m³/s (12.5 %). This might be caused by the use of the Weibull distribution in

Page 11: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 326

DOI: 10.5675/ICWRER_2013

the bivariate case and the GEV in the univariate case; both distributions have different

curvatures in the range of higher return periods (see Figures 2A and 3).

Although the statistically relevant inflows of the tributary derived with the univariate

approach generally correspond with the results using copulas, the return periods of the

univariate discharge combinations are higher as compared to the bivariate approach.

This is highlighted in Figure 4 by the fact that the results from the univariate analysis

lie generally in the upper right of the isolines from the bivariate analysis. The discharge

with a return period of T = 10 a in the univariate case would be classified, despite almost

equal marginal values, as an event with a return period of T " 15 a in the bivariate case

(considering the above mentioned definition of bivariate return period). Using the

univariate approach assumes independent variables which is not valid in this case where

the modeled variables have a rank correlation of τ = 0.567.

5. Conclusion

The main objective of this study was to compare the results from univariate and bivariate

statistical extreme value analysis at a river confluence. The application of both methods

to a case study in Germany with long discharge time series at both confluent rivers shows

that the simplified univariate approach is capable to determine the statistical relevant

discharges of the tributary given design discharges for the return periods of T = 2, 5, 10

and 20 years at the main stream (i.e. the results of the univariate and bivariate approach

are similar). However, significant differences are found for higher return periods, i.e.

T = 50 and 100 a. Whether this result is transferable to other confluences, in particular

to major rivers, will be further investigated. Moreover, classifying the results of the

univariate approach according to bivariate probabilities shows that the return periods

are generally overestimated. This is due to the fact that using univariate approaches

does not allow for modeling the dependence structure of the two variables.

ReferencesBfG: Flusshydrologische Software (engl. river hydrologic software) – User Manual,

version 2.1.3, Bundesanstalt für Gewässerkunde (engl. German Federal Institute of

Hydrology), 2009

Chen, L., Singh, V.P., Shenglian, G., Hao, Z., and Li, T.: Flood Coincidence Risk Analysis

Using Multivariate Copula Functions, J. Hydrol. Eng., 742 – 755, 2012.

Coles, S.: An Introduction to Statistical Modeling of Extreme Values, Springer, ISBN

1-85233-459-2, 2001.

Page 12: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 327

DOI: 10.5675/ICWRER_2013

De Michele, C., and Salvadori, G.: A generalized Pareto intensity duration model of storm

rainfall exploiting 2-copulas, J. Geophys. Res., 108(D2), 1–11, 2003.

Favre, A-C., Adlouni, S., Perreault, L., Thiémonge, N., and Bobée, B.: Multivariate

hydrological frequency analysis using copulas, Water Resour. Res., 40(1), W01101,

2004.

Gräler, B., van den Berg, M. J., Vandenberghe, S., Petroselli, A., Grimaldi, S., De Baets, B.,

and Verhoest, N. E., C.: Multivariate return periods in hydrology: a critical and practical

review focusing on synthetic design hydrograph estimation, Hydrol. Earth Syst. Sci.,

17, 1281–1296, 2013.

Grimaldi, S., and Serinaldi, F.: Design hyetographs analysis with 3-copula function, Hydrol.

Sci. J., 51(2), 223–238, 2006.

Gringorten, I. I.: A plotting rule for extreme probability paper. J. Geophys. Res., 68, No. 3,

813–814, 1963.

Huard, D., Évin, G., and Favre, A.-C.: Bayesian copula selection. Computational Statistics

& Data Analysis 51, 809-822, 2006.

Kao, S., and Govindaraju, R. S.: A copula-based joint deficit index for droughts, J. Hydrol.

(Amsterdam), 380(1–2), 121–134, 2010.

Karmakar, S., and Simonovic, S. P.: Bivariate flood frequency analysis. Part 2: A copula-

based approach with mixed marginal distributions, J. Flood Risk Manage., 2(1), 32–44,

2009.

Maidment, D. R.: Handbook of Hydrology, McGraw-Hill, New York, 1992.

Morris, C.D., and Calise, S.J.: Bivariate Analysis of Concurrent Flooding. Hydrologic

Frequency Modeling: Proceedings of the International Symposium on Flood Frequency

and Risk Analysis, 14. – 17. May 1986, Baton Rouge, Louisiana, ed. Singh V. P., Reidel:

Dordrecht, 615 - 632, Netherlands, 1987.

NCHRP: Estimating Joint Probabilities of Design Coincident Flows at Stream Confluences.

National Cooperative Highway Research Program (NCHRP) Report 15-36. Washington,

USA, 2010.

Nelsen, R.B.: An introduction to copulas. Lecture Notes in Statistics, 139, Springer, New

York, 1999.

Rao, A.R., and Hamed, K.H.: Flood frequency analysis, CRC Press, New York, 2000.

Raynal, J.A. and Salas, J.D.: A Probabilistic Model for Flooding Downstream of the Junction

of Two Rivers. Hydrologic Frequency Modeling: Proceedings of the International

Symposium on Flood Frequency and Risk Analysis, 14. – 17. May 1986, Baton Rouge,

Louisiana, ed. Singh V. P., Reidel: Dordrecht, 595 - 601, Netherlands, 1987.

Schölzel, C., and Friederichs, P.: Multivariate non-normally distributed random variables

in climate research – introduction to the copula approach, Nonlinear Proc. Geophys.,

15, 761–772, 2008.

Page 13: Flood Frequency Analysis at River Confluences – Univariate vs

ICWRER 2013 | Flood Frequency Analysis at River Confluences … 328

DOI: 10.5675/ICWRER_2013

Shiau, J. T.: Fitting drought duration and severity with twodimensional copulas, Water

Resour. Manage., 20(5), 795–815, 2006.

Sklar, A.: Fonction de répartition à n dimensions et leurs marges. Publications de Institut

de Statistique Université de Paris, 8, 229–231, 1959.

Song, S., and Singh, V. P.: Meta-elliptical copulas for drought frequency analysis of periodic

hydrologic data, Stoch. Environ. Res. Risk Assess., 24(3), 425–444, 2010a.

Song, S., and Singh, V. P.: Frequency analysis of droughts using the plackett copula and

parameter estimation by genetic algorithm, Stoch. Environ. Res. Risk Assess., 24(5),

783–805, 2010b.

Svensson, C., W., Kundzewicz, Z., and Maurer, T.: Trend detection in river flow series: 2.

Flood and low-flow index series, Hydrological Sciences Journal, 50:5, -824, 2005.

Wang, C., Chang, N-B., and Yeh, G-T.: Copula-based flood frequency (COFF) analysis at the

confluence of river systems, Hydrological Processes 23, S. 1471 – 1486, 2009.

Wahl, T., Mudersbach, C., and Jensen, J.: Assessing the hydrodynamic boundary conditions

for risk analyses in coastal areas: A multivariate statistical approach based on Copula

functions, Nat. Hazards Earth Syst. Sci., 12, 495-510, 2012.

Zhang, L., and Singh, V. P.: Bivariate flood frequency analysis using the copula method, J.

Hydrol. Eng., 11(2), 150–164, 2006.

Zhang, L., and Singh, V. P.: Gumbel Hougaard copula for trivariate rainfall frequency

analysis, J. Hydrol. Eng., 12(4), 409–419, 2007.

Page 14: Flood Frequency Analysis at River Confluences – Univariate vs

DOI: 10.5675/ICWRER_2013

ICWRER 2013 | Proceedings | Disclaimer

German National Committee for the

International Hydrological Programme (IHP) of UNESCO and the

Hydrology and Water Resources Programme (HWRP) of WMO

Koblenz 2013

©IHP/HWRP Secretariat

Federal Institute of Hydrology

Am Mainzer Tor 1

50068 Koblenz · Germany

Telefon: +49 (0)261 / 1306 - 54 35

Telefax: +49 (0)261 / 1306 - 54 22

www.ihp-germany.de

DOI: 10.5675/ICWRER_2013

Disclaimer

Any papers included in these proceedings

reflect the personal opinion of the authors.

The publisher does not accept any liability

for the correctness, accuracy or complete-

ness of the information or for the obser-

vance of the private rights of any third parties.

Any papers submitted by the authors do

not necessarily reflect the editors’ opinion;

their publication does not constitute any

evaluation by the editors.

German IHP/HWRP National Committee

HWRP – Hydrology and Water Resources Programme of WMO

BfG – Federal Institute of Hydrology, Koblenz

International Hydrological Programme

United NationsEducational, Scientific and

Cultural Organization