Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

A Modified Muskingum Routing Approach For Floodplain Flows: Theory and Practice

Download as pdf or txt
Download as pdf or txt
You are on page 1of 37

1 A Modified Muskingum Routing Approach for Floodplain Flows:

2 Theory and Practice


3

4 J.J. O’Sullivana, S. Ahilana, M. Bruena


5 (jj.osullivan@ucd.ie) (sangar.ahilan@ucd.ie) (michael.bruen@ucd.ie)

a
7 School of Civil, Structural and Environmental Engineering, Newstead Building, University

8 College Dublin, Belfield, Dublin 4, Ireland.

9 Correspondence to: J.J. O’Sullivan (jj.osullivan@ucd.ie)

10 Tel: +353 1 7163213

11 Fax.: +353 1 7163297

12

13 Abstract

14 Hydrological or hydraulic flood routing methods can be used to predict the floodplain

15 influences on a flood wave as it passes along a river reach. While hydraulic routing uses both

16 the equation of continuity and the equation of momentum to describe the dynamics of river

17 flows, the simpler data requirements of hydrological routing makes it useful for preliminary

18 estimates of the time and shape of a flood wave at successive points along a river. This paper

19 presents a modified linear Muskingum hydrological routing method where the floodplain

20 effects on flood peak attenuation and flood wave travel time are included in routing

21 parameters. Developing the routing parameters initially involved routing hydrographs of

22 different flood peak and duration through a 1-dimensional model of a generalised river reach

23 in which a range of geometrical and resistance properties were varied. Comparison of

24 upstream and simulated downstream hydrographs for each condition investigated, allowed the

25 attenuation and travel time (storage constant, K, in standard Muskingum routing) of the flood

-1-
1 wave to be estimated. Standard Muskingum routing was then used to develop downstream

2 hydrographs for each K value together with assumed storage weighting factors (x) ranging

3 from 0 to 0.5. Flood peak attenuations were again determined through comparison of the

4 upstream and routed downstream hydrographs and with these, linear relationships between x

5 and these attenuations were developed. Actual weighting factors, corresponding to storage

6 constants, were subsequently determined using these relationships for all attenuations

7 determined from the 1-dimensional model simulations. Using multi-variate regression

8 analysis, the computed values of K and x were correlated to catchment and hydrograph

9 properties and expressions for determining both K and x in terms of these properties were

10 developed. The modified Muskingum routing method based on these regressed expressions

11 for K and x was applied to a case study of the River Suir in Ireland where good agreement

12 between measured and routed hydrographs was observed.

13

14 Keywords: Overbank flow; Flood routing; Muskingum routing; Hydraulic and hydrological

15 methods; floodplains; Modelling and Saint-Venant equations.

16

17 1 Introduction

18 Hydraulic or hydrological flood routing techniques are commonly used by engineers

19 and hydrologists to predict the temporal and spatial variations of a flood wave through a river

20 reach (Choudhury et al., 2002). The methodologies that have been developed vary in their

21 complexity with more analytically rigorous methods having increased capacity to better

22 accommodate the dynamics and influences of floodplain behaviour on the propagation of a

23 flood wave in a natural channel. The Muskingum method of hydrological flood routing is one

24 approach. The popularity of the Muskingum method derives primarily from its minimal data

25 requirements. Knowledge of topographical catchment conditions are not required to

-2-
1 understand the propagation of a flood wave as these dynamics are represented in a calibration

2 carried out using observed data (Gallati and Maione, 1977). The Muskingum approach

3 represents a hydrological flood routing technique and is based on the equation of continuity

4 and a relationship that describes storage in the system. At the opposite end of the spectrum,

5 full-scale dynamic wave models based on the Saint-Venant equations provide a more

6 sophisticated means of hydraulic flood routing (Tung, 1985). Hydraulic routing uses the

7 equation of continuity and a momentum balance equation (the Saint Venant equations) and

8 involves their numerical solution using finite difference or characteristic methods.

9 Simplification of the momentum equation produces approximate solutions (e.g. monoclinal

10 wave, convective diffusion) that are easier to calculate and may be adequate in specific cases.

11 Advantages and disadvantages exist for using both hydrological and hydraulic flood

12 routing techniques. Although hydraulic routing techniques can more adequately describe the

13 dynamics of unsteady flows in canals and rivers, these methods are more demanding in their

14 information inputs and require data to accurately represent the geometrical and resistance

15 characteristics of the main channel and floodplain. Initial and boundary conditions are also

16 required. Conversely, the data inputs and computational procedures for hydrological routing

17 techniques are considerably simpler (Singh, 1988) and these methods are useful where

18 preliminary estimates of the time and shape of a flood wave at successive points along a river

19 are required, or where budgetary constraints may not facilitate full hydraulic routing.

20 A modified linear Muskingum hydrological routing method suitable for floodplain

21 flows is presented in this paper. The method is based on the standard Muskingum method.

22 However, rather than determining the routing parameters through analyses of observed

23 upstream and downstream hydrographs for given flows, the proposed method estimates its

24 routing parameters from empirical relationships describing floodplain effects on flood peak

25 attenuation and flood wave travel time. The method therefore has similarities with the

-3-
1 Muskingum-Cunge method (Cunge, 1969) in that flood routing parameters are determined

2 from geometrical and resistance properties of the channel, thus avoiding the need for a

3 calibration process. Development of the method involved a combination of 1-dimensional

4 hydraulic modelling and standard Muskingum flood routing to determine storage constants, K

5 and storage weighting factors, x, for a range of catchment and hydrograph properties. Multi-

6 variate regression analysis was used to correlate these computed values of K and x to these

7 properties and expressions for determining both K and x were developed. The modified

8 Muskingum routing method based on these regressed expressions for K and x was applied to a

9 case study of the River Suir in Ireland where good agreement between measured and routed

10 hydrographs was observed. The method offers a simple and inexpensive method of

11 estimating the time and shape of a overbank flood wave as it progresses along a river channel

12 of low to moderate sinuosity and in which backwater and inertia influences are likely to be

13 small.

14

15 2 Muskingum Flood Routing


16 The Muskingum method of flood routing (McCarthy, 1938), based on a simple storage

17 – discharge relationship in river systems, is extensively used in river engineering (Gill, 1979).

18 The method performs best in river systems where inertia effects and backwater influences are

19 small and where model parameters are appropriately chosen to represent the hydraulic

20 behaviour of the system (Chang et al., 1983). The linear Muskingum model uses continuity

21 and storage relationships expressed as:

dS t
22 Continuity:  It  Ot (1)
dt

23 Storage: S t  KxI t  1  x O t  (2)

-4-
1 where St, It and Ot are simultaneous amounts of storage, inflow and outflow respectively at a

2 given time t, K is a storage constant expressing the ratio between storage and discharge in a

3 river reach and x is a dimensionless weighing factor that varies between 0 and 0.5 for natural

4 rivers. This weighting factor describes the relative importance of inflow and outflow to

5 storage. The storage time constant, K, equates closely to the flow travel time through the

6 river reach (McCuen, 1998). If K and x are known, routing is performed using:

7 Ot  C1 It  C2 It t  C3 Ot t (3)

8 in which, C1, C2 and C3 are routing coefficients given by:

0.5t  Kx
9 C1  (4)
1  x K  0.5t
Kx  0.5t
10 C2  (5)
1  x K  0.5t
 0.5t  1  x K
11 C2  (6)
1  x K  0.5t
12 where parameters are as defined for Eq. 2 and in which t is the time step and It-t and Ot-t

13 are the inflow and outflow discharges at time t-t. Once C coefficients (which sum to unity)

14 are determined, Eq. 3 is used repeatedly to determine outflow discharges, Ot, at any time.

15 Values for K and x that describe the storage characteristics of a river reach are usually

16 derived from observed upstream and downstream hydrographs extracted from historical flow

17 records. These methods are well reviewed and are broadly represented in five classifications:

18 (a) graphical method; (b) least squares method; (c) method of moments and method of

19 cumulants; (d) direct optimisation; and (e) those based on the Saint-Venant equations (Singh

20 and McCann, 1980). More recently, Yoon and Padmanabhan (1993), identified a further

21 three methods for linear model parameter estimation. These algorithms included forward and

22 backward optimisation using t-statistics, an outliers filtering estimation method and a

23 quadratic programming algorithm.

-5-
1 Graphical methods are commonly applied. The standard trial-and-error graphical

2 approach used by McCarthy (1938) for the linear Muskingum model involves plotting

3 xI t  1  x Ot  , known as the weighted discharge term, against the accumulated storage for
4 different assumed values of x. Different values of x in Eq. 2 produce a family of curves that

5 vary from being heavily looped to being reasonably linear. The particular value that generates

6 the narrowest loop and can be best fitted with a straight line is considered the best estimate of

7 x. The inverse slope of this line gives the required value of K. Although the graphical

8 method is generally satisfactory (Chow, 1964; Linsley et al., 1975; Viessman et al., 1972;

9 Wilson, 1990), it is time consuming to apply. Furthermore, no objective-selection criteria

10 exists for choosing the appropriate value of x and the method therefore, requires a level of

11 subjective interpretation to determine a value that optimises the linear relationship (Gelegenis

12 and Serrano, 2000; Yoon and Padmanabhan, 1993; Chang et al., 1983). Muskingum routing

13 parameters have also been estimated using a least-squares scheme based on minimising the

14 sum of squares of the deviations between observed storage and computed storage for given

15 inflow and outflow hydrographs (Gill, 1978; Birkhead and James, 1997; Al-Humoud and

16 Esen, 2006). The underlying principles of the graphical and the least-square methods are the

17 same and both methods should produce similar parameter values. The method of moments

18 and the method of cumulants are similar and are based on relating the first and second

19 moments or cumulants of the instantaneous unit hydrograph (IUH) of the Muskingum reach

20 to the Muskingum routing parameters, K and x (Dooge, 1973). The method of direct

21 optimisation is based on minimising the difference between observed and computed

22 hydrographs to determine directly the routing coefficients of the Muskingum model without

23 explicitly estimating K and x (Gelegenis and Serrano, 2000).

24 More recent advances in computer technologies have allowed the traditional

25 Muskingum routing methods to be linked with hydrodynamic software packages for the

-6-
1 analysis of surface water drainage in natural catchments (see for example the HEC-HMS and

2 TOPMODEL hydrological models). These are based on the Saint-Venant equations that are

3 derived from the principles of conservation of mass and momentum and can be written in

4 their 1-dimensional form as:

y V y
5 y V 0 (7)
t x x

V  y
6 V  gA  gS f  S 0   0 (8)
t x c x

7 (I) (II) (III) (IV)

8 where y is the flow depth, V is the flow velocity, g is the acceleration due to gravity, So is the

9 river bed slope, Sf is the slope of the energy line, x is the longitudinal distance and t is the

10 time. These equations are simultaneous, quasi-linear, first-order partial differential equations

11 of the hyperbolic type and are not amenable to general analytical solutions. The first term (I)

12 in Eq. 8 is the local inertia term, the second (II) is the convective inertia term, the third (III) is

13 the pressure differential term and the fourth (IV) accounts for the friction and bed slopes.

14 Numerical methods for solving the Saint-Venant equations are broadly classified in two

15 categories: (a) approximate methods; and (b) complete numerical methods. Approximate

16 methods are based on the equation of continuity only or on a significantly curtailed equation

17 of momentum.

18 Kinematic and diffusion wave models can be constructed to solve Eq. 7 and Eq. 8 by

19 assuming that the significance of some terms is negligible compared to that of others (Moussa

20 and Bocquillon, 1996). Models that neglect inertial terms (I and II) are known as diffusion

21 wave models (Cunge, 1969; Bajracharya and Barry, 1997; Moussa and Bocquillon, 2009) and

22 models that neglect both inertial and pressure terms (I, II, III) are referred to as kinematic

23 models (Smith, 1979). The full Saint-Venant equations have also been solved for Muskingum

24 routing applications using numerical techniques in channels with cylindrical and irregular

-7-
1 cross-sectional geometries without the need for simplifying assumptions (Amein and Fang,

2 1970; Dooge et al., 1982; Wang et al., 2006).

3 Cunge (1969) included the effects of geometrical and resistance properties of a river

4 reach in the original Muskingum method to develop the Muskingum-Cunge (M-C) model.

5 Cunge showed that the Muskingum formula for solving flood routing problems is identical to

6 a finite-difference approximation of the linearised diffusion wave equation, this equation

7 being derived from the Saint-Venant equations by neglecting the inertial terms. In the M-C

8 flood routing procedure, the necessity of calibration that characterises the Muskingum method

9 is not required and the routing parameters, K and x, are obtained from hydraulic properties of

10 the reach using:

x
11 K (9)
c

12 and

1 Q 
13 x 1   (10)
2 BS 0 cx 

14 where parameters are as described above and where c is the flood wave celerity, x is the

15 longitudinal channel distance increment, Q is the discharge and B is the average bed width of

16 the channel.

17 The time step, t, used in the M-C routing procedure is appropriately chosen to fully

18 define the shape of the inflow hydrograph. A dependency exists between t and x in M-C

19 routing procedures and the choice of increments is therefore important to ensure that x is not

20 significantly smaller than the distance travelled by the flood wave in a single time step, t.

21 This x interval chosen in the M-C routing method is therefore based on t, c, Q, the channel

22 top width and the longitudinal channel slope, So. The flood wave celerity, c, is obtained from

23 the slope of the discharge - area curve for a given discharge, Q. Details of the M-C method

-8-
1 are discussed in Volume III of the Flood Studies Report (FSR) (NERC, 1975). Tang et al.,

2 (1999a) investigated the properties of the M-C method for flood routing, using several

3 hypothetical flood hydrographs in a prismatic channel with significant floodplains. Results

4 indicated that the M-C method suffers from a loss of outflow volume which depends on bed

5 slope and floodplain roughness. Furthermore, it was observed that an initial leading edge dip

6 and trailing edge oscillation occur in the rising and recession limbs of the hydrograph

7 respectively. These oscillations become more significant as the roughness of the floodplain

8 increases, but gradually disappear with decreasing bed slope.

9 The standard linear Muskingum routing method assumes that K and x remain constant

10 and these are determined by analysis of measured inflow and outflow hydrographs. The

11 method therefore, does not accommodate changes in these parameters that would reflect more

12 accurately the routing of storm sequences in the river reach beyond the calibration range

13 (Kundzewicz and Strupczewski, 1982). More recent developments to the method however,

14 do allow for parameter variability with changing characteristics of the inflow hydrograph (see

15 for example Perumal, 1992a; Guang-Te and Singh, 1992; Al-Humoud and Esen, 2006).

16 Perumal (1992a) developed a multi-linear Muskingum flood routing method based on a time

17 distribution scheme. The physically based Muskingum method is used as the linear sub-

18 model in this method and the parameters are varied at each time step for the routing of

19 prescribed flow zones in the inflow hydrograph. Guang-Te and Singh (1992) proposed three

20 versions of the linear Muskingum method with variable parameters. In these methods, the

21 reach travel time (which depends on the storage, channel characteristics and discharge) is

22 obtained from a simplification of the Saint-Venant equations. Al-Humoud and Esen (2006)

23 proposed two approximate methods for the estimation of linear Muskingum flood routing

24 parameters. The first method requires the computation of the slopes of the inflow and outflow

25 hydrographs at their point of intersection, and the computation of the maximum storage

-9-
1 within the reach. The second method requires the computation of the inflow and outflow

2 hydrographs at two specific points.

3 Although the current paper is based on a linear Muskingum method, the relationship

4 between xI t  1  x Ot  and St is usually nonlinear and Gill (1978) proposed the following

5 storage relationship for the Muskingum model:

S t  KxI t  1  x O t 
m
6 (11)

7 where m is an exponent that defines the nonlinear relationship between accumulated storage

8 and weighted flow. This exponent cannot be directly determined from inflow and outflow

9 hydrographs and therefore, alternative parameter estimation methods have been presented (see

10 for example Gill, 1978; Tung, 1985; Yoon and Padmanabhan, 1993; Mohan, 1997; Kim et al.,

11 2001; Geem, 2006; Chu, 2009). Gill (1978) proposed a routing scheme for the nonlinear

12 model based on the segmented curve method where coefficients are determined by a least

13 squares method. However, the technique is somewhat arbitrary in its process of selecting

14 three points on the segmented curve for solving simultaneously the continuity (Eq. 1) and

15 storage (Eq. 11) equations of the nonlinear method (Tung, 1985). Tung (1985) proposed

16 procedures using the Hook-Jeeve (HJ) pattern technique in combination with simple linear

17 regression (LR), the conjugate gradient (CG), and the Davidon-Fletcher Powell (DFP)

18 techniques and used the state variables technique for routing. The approaches were compared

19 with Gill’s methodology which showed that HJ+CG and HJ+DFP techniques produced better

20 estimations of the routing parameters. Yoon and Padmanabhan (1993) proposed a nonlinear

21 least-squares regression technique which directly fits the nonlinear model. The estimation

22 proceeds iteratively from an initial assumption of the parameters using the Marquardt

23 algorithm (Marquardt, 1963). In addition, this method has internal logic that analyses the data

24 and computes reasonably accurate values of the parameters to be estimated. This serves to

-10-
1 expedite the optimisation process. Mohan (1997) suggested a calibration technique for

2 determining K, x and m based on a genetic algorithm that avoids the need to make initial

3 assumptions. Kim et al. (2001) proposed a harmony search algorithm for estimation of the

4 same parameters and observed that the technique outperformed other genetic, heuristic and

5 mathematical algorithms, evolutionary programming and non linear programming. Geem

6 (2006) presented a Broyden-Fletcher-Goldfarb-Shanno (BFGS) technique for parameter

7 estimation in nonlinear Muskingum models. The BFGS algorithm is a branch of the quasi-

8 Newton method based on mathematical gradients that searches for an optimised solution of

9 the unconstrained nonlinear equations. Chu (2009) combined a Fuzzy Inference System

10 (FIS), implemented in an adaptive network framework with a nonlinear Muskingum model to

11 estimate the outflow hydrograph. However, the calibration procedure for finding the correct

12 values of the three parameters K, x and m to determine this outflow hydrograph is complicated

13 (Kim et al., 2001).

14

15 3 Methods

16 A multi-stage process (Fig. 1) that included 1-dimensional HEC-RAS modelling of a

17 generalised river reach, standard Muskingum routing and regression analysis was adopted for

18 developing expressions for storage constants (K) and weighting factors (x) for use in the

19 modified Muskingum method that is presented. The process begins with implementing a

20 generalised HEC-RAS model using input flow hydrographs and geometric properties to

21 determine travel times and relative attenuations which are then input into a regression model

22 to develop equations for estimating Muskingum model parameters. The following paragraphs

23 explain the process in detail.

24

25 Fig. 1

-11-
1

2 3.1 HEC-RAS Model of Generalised River Reach

3 The HEC-RAS model of the generalised river reach was executed in dynamic mode

4 for an extensive range of geometrical and resistance properties. HEC-RAS (Hydraulic

5 Engineering Centre – River Analysis System) is a 1-dimensional link and node river model

6 developed by the US Army Corps of Engineers that discretises and solves the dynamic Saint-

7 Venant equations using an implicit, finite difference method. Theoretical hydrographs of

8 varying peak and also, hydrographs of varying duration were key inputs to the model.

10 3.1.1 Hydrographs of Varying Peak Flow

11 Hydrographs for a range of peak flows were developed using a methodology (Fig. 2)

12 and associated software from Work-Package 3.1 of the Irish Flood Studies Update (FSU)

13 programme (O’Connor and Goswami, 2010) that utilises a historical record of flow data.

14

15 Fig. 2

16

17 Any gauged location where a record of good quality data was available would

18 therefore be suitable and one such site with a long and continuous flow record was chosen.

19 Hydrograph development involved the following:

20 (1) The annual exceedence series of the maximum flood events for the selected site was

21 identified from recorded data (shown for a single event in Fig. 2 (a)).

22 (2) Annual exceedence flood hydrographs were isolated from the flow record by discarding

23 the complex segments on each side of the peak, leaving the single-peak hydrograph

24 component (Fig. 2 (b)).

25 (3) The isolated flood hydrographs were then standardised to have a peak value of unity by

-12-
1 dividing all its flow ordinates by the peak flow.

2 (4) This unit peak is assumed to represent the 100 percentile flow and hydrograph widths

3 were determined at percentiles of 98, 95, 90 …. 10 and 5.

4 (5) Widths corresponding to these flow percentiles were averaged over the annual

5 exceedence series and the rising limb of the theoretical flood hydrograph was

6 approximated by fitting a modified form of the Gamma curve to these widths. The shape

7 of the full unit hydrograph was obtained by the addition of an exponential recession curve

8 drawn from the point of inflection of the modified Gamma curve (Fig. 2 (c)).

9 (6) The required hydrographs were generated by scaling up the derived unit hydrograph for

10 peak flows of different return periods (2, 5, 25, 50, 100, 500 and 1000 years were used).

11 A base flow for the particular return periods was added at each ordinate (Fig. 2 (d)).

12

13 Fig. 3

14

15 Annual maximum flow series in most Irish catchments follow a Generalised Extreme

16 Value (GEV) Type I distribution (NERC, 1975) and based on this distribution, flood quantiles

17 for the hydrographs in Fig. 2 (Panel D) vary from 91.41 m3/s for the 2-year event to 153.90

18 m3/s for the 1000-year flood.

19

20 3.1.2 Hydrographs of Varying Duration

21 The hydrographs developed using this methodology have the same base length and

22 therefore flood volume is determined solely by the flood peak. The relationship therefore,

23 between flood volume and flood peak is not fully defined. Floods result from natural

24 processes and to fully account for the random nature of their generation, duration needs to be

25 included if flood volume is to be accurately related to peak. Although attempts to develop

-13-
1 relationships between volume and peak of direct runoff for catchments elsewhere in the world

2 have been made (see for example Rogers, 1980, Mimikou, 1983, Singh and Aminian, 1986),

3 no validated relationship exists for Irish catchments and an attempt to develop such a

4 relationship in the FSU was inconclusive (O’Connor and Goswami, 2010).

5 In the absence of a validated method that reflects the increased flood volumes that

6 would usually be associated with floods of longer duration, a simple approach where flood

7 duration is included independently of flood peak was implemented. The approach involved

8 developing a triangular hydrograph of the same volume as the 1000-year hydrograph in Fig. 2

9 (Panel D) and linking this volume to the hydrograph characteristics by:

1
10 Volume =  T B  QP (12)
2

11 where TB is the hydrograph base width and QP is the flood peak (153.90 m3/s). The duration

12 of the 1000-year hydrograph corresponding to this volume is approximately 265 hours. This

13 duration (TB) was linked to the time to peak, TP, by the Flood Studies Report (NERC, 1975)

14 relationship:

15 TB  2.52TP (13)

16 By further scaling the 1000-year hydrograph, the approach facilitated the development of a

17 second set of hydrographs of different flood durations (Fig. 2 (e)).

18

19 3.1.3 Geometrical and Resistance Properties in HEC-RAS Model

20 The basic HEC-RAS model geometry (Fig. 3) included basic bankfull (Bbf) and

21 floodplain (bfp) widths of 25 m and a bankfull depth (h) of 2.5 m. This bankfull depth

22 produced a bankfull flow for the flood hydrograph of the median flow having a 2-year return

23 period and ensured that floodplains in the generalised model would be active for larger floods.

24 The main channel side slopes and the floodplain boundaries were inclined at 45o giving

-14-
1 trapezoidal geometries in both the inbank and overbank zones. The basic hydraulic resistance

2 of the main channel (nmc) and floodplains (nfp) was expressed in terms of Manning’s n and

3 assigned values of 0.03 and 0.25 respectively. The high base value of nfp was chosen to

4 ensure that measurable attenuations were observed in model simulations. The basic model

5 length (L) was 50 km and its longitudinal slope of the floodplain (Sfp) was set at 1 m/km. In

6 total, 65 variations of these basic properties were tested in eight sets of simulations, denoted

7 by A to H and in which one property was varied at a time (Table A1). Case A investigated

8 the effect of channel length (L), Case B investigated the effect of the longitudinal floodplain

9 slope (Sfp), Case C varied the floodplain hydraulic resistance (nfp), Case D the floodplain

10 width (bfp), Case E, the transverse floodplain slope () and Case F, the main channel

11 hydraulic resistance (nmc). The influence of flood peak (QP) and flood duration (TB) was

12 explored by routing the two sets of hydrographs (Fig. 2 Panel D and Fig. 2 Panel E) through

13 the generalised model in the Case G and Case H simulations respectively. For each set of

14 simulations, the effect of changes in a specific property on flood attenuation and travel time

15 was examined by comparing input and output hydrographs.

16

17 3.2 Standard Muskingum Routing Model

18 The travel time of the peak of the flood wave determined for each simulation from the

19 input and output hydrographs in the HEC-RAS model was assumed equal to the storage

20 constant (K) in the standard Muskingum routing method (Fig. 1 Panel A). Estimation of

21 corresponding weighting factors (x) was more involved. For each value of K, in combination

22 with assumed values of x (increasing incrementally from 0 to 0.5 where low values reflect

23 high attenuation and vice versa), standard Muskingum flood routing of the inflow

24 hydrographs was repeatedly performed using Eq. 3, together with Eqns. 4, 5 and 6 to generate

25 a series of outflow hydrographs. Peak outflows were determined and comparison of these

-15-
1 with the peaks of the inflow hydrographs (Fig. 3 (b)) allowed a series of relative attenuations

2 to be determined, from:

Q P1  Q P 2
3 % Relative attenuation =  100 (14)
Q P1

4 where QP1 and QP2 are the peaks of the inflow and outflow hydrographs (Fig. 1 Panel A).

5 Linear relationships between these relative attenuations from the Muskingum routing and

6 assumed weighting factors (x) were developed. These relationships were used to determine

7 actual weighting factors for the relative attenuations calculated by comparison of the inflow

8 and outflow hydrographs for each HEC-RAS simulation (Fig. 1 Panel B). This produced a

9 weighting factor for each of the 65 simulations for which corresponding storage constants

10 were directly determined, covering the full range of geometrical and resistance properties that

11 were assessed for the different inflow hydrographs.

12

13 3.3 Regression Analysis

14 Using multi-variate regression analysis, the computed values of the storage constants

15 and weighting factors were correlated to catchment and hydrograph properties and

16 expressions for determining both K and x in terms of these properties were developed.

17

18 4 Results
19 Estimated values of K and x from both the HEC-RAS hydraulic modelling and the

20 standard Muskingum routing are shown in Table A1 and variations of these routing

21 parameters with catchment and hydrograph properties are shown in Fig. 4 and Fig. 5

22 respectively. Variations in both flood peak (QP) and transverse floodplain slope () were

23 shown to have only a small influence on storage weighting factors and were excluded from

24 the analysis.

-16-
1

2 Fig. 4

4 Fig. 5

6 Results confirm that increasing floodplain length (Case A) and width (Case D), as

7 noted by Wolff and Burges (1994), increases the capacity of the overbank zone to attenuate

8 and delay the propagation of a flood wave along a channel. Longitudinal slope (Case B) is

9 also important. Steep catchments have the capacity to convey floods more rapidly than those

10 that are more mildly graded and the increased conveyance is reflected in reduced travel time

11 (Fig. 4 (b)) and attenuation (Fig. 5 (b)). These trends are consistent with the findings of Wolff

12 and Burges (1994); Tang et al., (1999b) and Ghavasieh et al. (2006) where large attenuations

13 with sharp reductions in the variability of the cumulative distribution were observed in low

14 gradient catchments. Tang et al., (1999b) reported that the Muskingum Cunge flood routing

15 method suffers a certain amount of volume loss that depends on the longitudinal slope of the

16 channel. Increasing the lateral slope of floodplains (Case E) results in a geometry in which

17 overbank flow is continually redirected back towards the main channel. Channels therefore,

18 with steep lateral slopes will convey an increased proportion of the flood volume in the main

19 channel. Furthermore, floodplain resistance in the generalised river model was higher than

20 that in the main channel. The reduced attenuations and higher wave speeds with reduced

21 travel times (Fig. 4 (e)) in geometries with increasing transverse slopes are exacerbated given

22 that a diminishing proportion of the total flow is being influenced by the higher floodplain

23 roughness values. Increases in main channel (Case F) and floodplain resistance values (Case

24 C) produce increased flood attenuation and travel times.

-17-
1 The full influence of floodplains on flood wave attenuation is complex and is also

2 influenced by flow peak magnitude (Case G) and the resulting overbank depth. At low return

3 periods (typically less that 2 years), flows will not significantly inundate the floodplain and

4 will not be affected by the additional attenuation associated with the floodplain. Fig. 4 (g)

5 indicates that the wave speed is relatively high for these in-bank events. At flows that

6 produce low overbank depths (return periods of less than 5 years), floodplain influences

7 increase attenuation and travel times. Case H simulations assessed flood duration on relative

8 attenuation and flood wave travel time. When combined, flood peak (QP) and duration (TB)

9 define the flood volume. Floods with low volume hydrographs but sharp peaks and thus short

10 durations experience significantly higher attenuation than those with higher volumes. Floods

11 that are characterised by high volumes on the rising limb of the hydrograph will tend to

12 occupy floodplain storage that is available and once occupied; this storage is no longer

13 available for the remainder of the flood. The attenuation provided by the floodplain in these

14 cases is thus limited. In contrast, hydrographs with low rising limb volumes, disperse most of

15 the flood volume to storage and contribute therefore, to comparatively high relative

16 downstream attenuations.

17

18 5 Development of Muskingum Model Parameters

19 The influences of the geometrical, resistance and hydrograph properties on K and x in

20 Fig. 4 and Fig. 5 were included in a multi-variate regression analysis to generate expressions

21 for these routing parameters. The floodplain width (bfp) and the bankfull width (Bbf) were

22 expressed as a single parameter defined by (bfp/Bbf) that is consistent with other overbank

23 flow research (for example Knight and Shiono, 1996). The expressions are:

24

-18-
0.60
b 
L n n  fp  TB0.07
0.24
fp
0.42
mc

1 K  0.794  Bbf  (15)


0.06
 Q 
Sfp  
0.53 0.09 P 

2
 fpb  B bf 

L0.03 S0fp.16 TB0.39


3 x  0.035 0.05 (16)
 b fp 
  n  mc
0.06
n 0fp.008
 Bbf 

4 Application of Eq. 15 requires that  > 0 and therefore horizontal floodplains are

5 represented by a near-zero value of . Similarly, the equation assesses floodplain effects and

6 therefore peak flows should exceed bankfull discharge capacities in a given river reach. It

7 should be noted that Eq. 15 and Eq. 16 are based solely on the influences of the assessed

8 parameters on relative attenuations and delays of flood peaks determined by the HEC-RAS

9 modelling of the variations in the generalised river reach. The values of parameter exponents

10 are therefore based on the simulated data only and as with regression models of this type,

11 parameters that may intuitively be considered to be important do not necessarily come to the

12 fore in the analysis. The negative influence of floodplain roughness (nfp) in Eq. 16 is a case in

13 point. It would be expected that significant floodplain roughness would result in reasonably

14 large storage and yield low storage weighting factors. However, in the HEC-RAS modelling,

15 increasing floodplain resistance increased the proportion of flow being conveyed in the main

16 channel for all flows investigated, with the result that simulated attenuations were low. The

17 performance of these indices is shown in Fig. 6 where the routing parameters, K and x, are

18 plotted on linear scales against those calculated using Eq. 15 and Eq. 16.

19

20 Fig. 6

-19-
1

2 Fig. 6 indicates that Eq. 15 and Eq. 16 reproduce reasonably well the simulated data

3 for most of the geometrical, resistance and hydrograph properties. However, some limitations

4 exist. Simulated weighting factors (x) are shown to vary most significantly with flood

5 duration (TB). The poor fit in this case may result from the assumption of independence

6 between the flood peak (QP) and the flood duration (TB) that was made when including

7 duration as a parameter in the regression model. In addition, the clustering of nfp values in

8 Fig. 6 indicates the low influence of this property on the routing parameters determined using

9 the equations.

10

11 6 Illustration of Modified Muskingum Flood Routing Method

12 The routing method presented was applied to the River Suir in Co. Tipperary, Ireland.

13 The River Suir is typical of most Irish rivers in terms of its scale and its low main channel and

14 floodplain sinuosity. A 16.8 km reach between the New Bridge (Station No. 16008) and

15 Caher Park (Station 16009) gauging stations was tested (Fig. 7). Both stations, in addition to

16 a third station at Killardry (Station 16007), where the flow of the main tributary (River

17 Aherlowe) that joins the Suir between New Bridge and Caher Park is measured, are

18 characterised by good quality, digitised 15-minute flow records from 1954 to 2007. Other

19 less significant tributaries also join the river between these stations but are not gauged. The

20 catchment areas to the New Bridge and Caher Park gauging stations are 1120 km2 and 1602

21 km2 respectively and the area to Killardry on the Aherlowe River is 273 km2. The flood

22 history of the river in this area also indicates that significant floodplain inundation is frequent.

23

24 Fig. 7

25

-20-
1 Illustrating the method involved application of the modified Muskingum method using

2 Eq. 15 and Eq. 16 to route a selection of measured hydrographs at New Bridge through the

3 River Suir reach and comparing these at Caher Park to those obtained from both measured

4 data and from a HEC-RAS model of the river system. The HEC-RAS model was developed

5 from 35 recently obtained cross-sections between New Bridge and Caher Park that defined

6 the main channel geometry and floodplains to a width of approximately 30 m from the

7 channel banks of the Suir and Aherlowe rivers. This data was augmented by LIDAR data to

8 further define the floodplain topography to widths of approximately 500 m on each side of the

9 main channel. Longitudinal distances between measured cross-sections in the reach were

10 approximately 400 m and this resolution in the model was increased through cross-section

11 interpolation. The lower reach of the River Aherlowe was included in this model. At present,

12 85% and 70% of the Suir and Aherlowe catchments respectively are covered by grassland

13 pasture and this land use is dominant in the floodplains. A Manning’s n value of 0.05 for

14 grassland pasture with areas of brush described the hydraulic resistance of the floodplain and

15 a coefficient of 0.04 that is typical for a reasonably straight, clean channel at full stage with

16 some obstructions and marginal vegetation defined the main channel resistance (Cowan,

17 1956; Chow, 1959; Hollinrake and Millington, 1994).

18 The majority of natural hydrographs are complex and are characterised by kinks and

19 multiple peaks that reflect both the temporal variability of the storm and the spatial

20 heterogeneity of the catchment. Although it is theoretically possible to resolve a complex

21 hydrograph into a series of simple hydrographs, the routing of simple hydrographs from

22 isolated storm events extracted from long flow records was used in this study. Isolating these

23 events was a laborious task and was assisted by FSU hydrograph processing software that

24 facilitated the identification of simple hydrographs at the three gauging stations for three

-21-
1 specific storm events. These events related to periods in December 1954/ January 1955,

2 August 1986/ September 1986 and October 2004/ November 2004 (Fig. 8).

4 Fig. 8

6 Measured outflow hydrographs at Caher Park can potentially be influenced by a

7 number of tributary inflows (Fig. 7) for which no flow data is available. Their impacts on the

8 peak and timing of the flood hydrographs in the main River Suir, therefore is uncertain. To

9 ensure that the measured outflow hydrographs at Caher Park were not unduly influenced by

10 these tributaries and that measured data offers a basis for testing the modified method,

11 measured hydrographs at New Bridge and Killardry (Fig. 8) were routed through the HEC-

12 RAS model of the river system (that excludes the tributary network for which no data is

13 available) and compared to observed hydrographs at Caher Park. The good agreement

14 between the hydrographs routed in the HEC-RAS model and those measured at Caher Park

15 for the three events investigated (Fig. 9), indicates that the contribution of the tributary flows,

16 other than that from the Aherlowe River, is not significant.

17

18 Fig. 9

19

20 Parameters in Eq. 15 and Eq. 16 apply only to a single reach and cannot easily be

21 extrapolated to a river system that includes a tributary network. For validation of the

22 modified method therefore, Caher Park hydrographs for the three storm events that exclude

23 the contribution from the Aherlowe River were required. These were approximated by

24 routing the observed hydrographs at Killardry through a HEC-RAS model of the river from

25 this location to Caher Park and subtracting these hydrographs from those measured at Caher

-22-
1 Park. The resulting hydrographs are referred to as ‘adjusted’ hydrographs. Although this

2 process is somewhat artificial and backwater effects from interactions between the shared

3 floodplain of the main channel and tributary are not included in the analysis, the effects are

4 likely to be local and in the context of a 16.8 km reach, the approach was considered

5 acceptable. Comparison of the adjusted hydrographs with those determined by routing the

6 measured hydrographs at New Bridge using the modified Muskingum procedure allowed the

7 performance of the approach to be illustrated.

8 The testing of the Muskingum method using Eq. 15 and Eq. 16 was based on assigning

9 appropriate values to parameters that describe the geometry and hydraulic resistance of the

10 channel and floodplains together with the characteristics of the flood hydrographs.

11 Geometrical properties of the main channel and floodplain were determined from survey data

12 and where necessary, averaged over the river reach. The floodplain slope (Sfp) and length (L)

13 are FSU catchment descriptors for which numerical values are readily available. Main

14 channel and floodplain Manning’s resistances were estimated to be 0.04 and 0.05 respectively

15 and flood peaks and durations were determined from measured inflow hydrographs at New

16 Bridge. This data is summarised in Table 1 for the 1954/ ’55, 1986 and 2004 flood events.

17 For the initial testing of the method, the floodplain widths (bfp) in Table 1 were computed by

18 averaging (for 400 m intervals over the 16.8 km reach) the floodplain widths predicted from

19 the HEC-RAS model when routing the measured hydrographs at New Bridge to Caher Park.

20

21 Table 1

22

23 Caher Park hydrographs using the modified Muskingum method (referred to as

24 ‘Muskingum’) with these averaged floodplain widths are shown in Fig. 10 with the adjusted

-23-
1 Caher Park hydrographs. The Caher Park hydrographs from the HEC-RAS model (referred to

2 as ‘HEC-RAS’) are also shown for comparative purposes.

4 Fig. 10
5

6 Although strong correlations between the Muskingum and HEC-RAS routed

7 hydrographs are evident in Fig. 10, the usefulness of the proposed method as presented is

8 limited in that the floodplain widths are derived from hydraulic routing. To improve the

9 predictive capacity of the method, floodplain attenuation indicators (FAIs) that were

10 developed for the Irish Flood Studies Update (FSU) were utilised. FAIs are flood polygons

11 that define the active river floodplain in Irish catchments for the 10-year (Q10), the 100-year

12 (Q100) and the 1000-year (Q1000) floods from normal depth modelling at FSU nodes

13 (approximate intervals of 500 m) on the main river network. The approach is based on the

14 assumption that the median flood, Qmed with a return period of 2 years is equivalent to the

15 bankfull flow in all rivers. Given that bankfull recurrence intervals in many rivers are in the

16 order of 1 – 3 years (see for example Richards, 1982; Petit and Paquet, 1997; Castro and

17 Jackson, 2001), this simplifying assumption is reasonable. The median flood is determined

18 using an FSU relationship (Sweeney and Murphy, 2010) for ungauged catchments given by:

19

20 Qmed  1.237105 AREA0.937 BFI0.922 SAAR1.306 FARL2.217 DRAIND0.341 S10850.185


21 1 ARTDRAIN  0.408
(17)

22

23 where AREA (km2) is the catchment area of the river to the outlet point being considered,

24 S1085 (m/km) is the average slope of the river between 10% and 85% of its length from the

25 outlet, SAAR (mm) is the annual average rainfall on the catchment, FARL is a flood

26 attenuation factor for reservoirs and lakes, BFI is the baseflow index, DRAIND (km-1) is a

-24-
1 simple index that relates the length of the upstream hydrological network (km) to the area of

2 the gauged catchment (km2) and ARTDRAIN is an index of the arterial drainage extent

3 defined as the percentage area of benefiting lands with respect to the total catchment area.

4 Simple multiplication of Qmed by appropriate growth curve factors defined the

5 magnitudes of Q10, Q100 and Q1000. Floodplain flows were determined by subtracting Qmed

6 values from these flood quantiles and corresponding floodplain flow depths were determined

7 iteratively at all nodes using the Manning equation based on the geometry at that node and a

8 resistance coefficient that was consistent with the land use at the node. Incorporating these

9 depths into a Digital Terrain Model (DTM) facilitated the production of flood polygons for

10 Q10, Q100 and Q1000 for the Irish river network.

11 The return periods for the 1954/ ’55, 1986 and 2004 floods in the River Suir varied

12 between 5 and 10 years. The Q10 FAI was therefore the most relevant polygon from which to

13 estimate floodplain widths and an averaged value of 100 m was determined from the flood

14 extent at all nodes between New Bridge and Caher Park. Hydrographs developed from the

15 modified Muskingum approach using this floodplain width of 100 m are shown in Fig. 11

16 with those generated from hydraulic routing and those developed from observed data.

17

18 Fig. 11

19

20 6.1 Discussion of Results

21 Visual comparisons of the hydrographs in Fig. 10 and Fig. 11, although somewhat

22 subjective, provide a quick and simple means of assessing the performance of the modified

23 Muskingum routing method presented in this paper. Unsurprisingly, adjusted Caher Park

24 hydrographs in Fig. 10, correlate closely with those of the modified Muskingum method in

25 which floodplain widths were extracted from results of the HEC-RAS routing. However,

-25-
1 given that floodplain widths are based on outputs from a hydraulic model, this approach is of

2 limited use. More meaningful assessments of the method as a predictor of outflow

3 hydrographs are shown in Fig. 11. Here, good agreement is again observed between

4 hydrographs from the modified Muskingum method in which floodplain widths are

5 determined from the FAI catchment descriptor, and the HEC-RAS and adjusted hydrographs.

6 The goodness-of-fit between the Muskingum and adjusted hydrographs was qualitatively and

7 less subjectively tested using the range of goodness-of-fit criteria recommended by Jewitt and

8 Schulze (1999). Goodness-of-fit statistical tests measure the deviation of a simulated output

9 from an observed input data set and different tests are applied to assess different hydrograph

10 components. Root mean square errors (RMSE) determine the magnitude of error in the

11 computed hydrographs and were estimated using the relationship by Schulze et al. (1995)

12 given by:

 Qcomp( t )  Qobs ( t ) 
n 2

13 RMSE  i 1
for i = 1, 2, 3,……n (18)
n

14 where Qcomp(t) and Qobs(t) are the computed and observed discharges at a number of different

15 time steps n. Given that peak outflow is important in a single event hydrograph model,

16 percentage errors in computed and observed peak flow rates, peak timing and volume were

17 determined using the following equations (Green and Stephenson, 1985):

Qpcomp  Qpobs
18 Epeak   100 (19)
Qpobs

t pcomp  t pobs
19 Etime   100 (20)
t pobs

Vcomp  Vobs
20 Evolume   100 (21)
Vobs

-26-
1 where Epeak, Etime and Evolume are percentage errors in peak flow, timing and hydrograph

2 volumes respectively, Qp-comp and Qp-obs are computed and observed peak flows, tp-comp and

3 tp-obs are computed and observed times to peak flow and Vcomp and Vobs are computed and

4 observed hydrograph volumes.

5 Even though the RMSE, Epeak, Etime and Evolume statistics may assess model

6 performance effectively, differences in the shape of computed and observed hydrographs may

7 not be accounted for fully. To overcome this, Nash and Sutcliffe (1970) proposed a

8 dimensionless coefficient of model efficiency (E), given as:

 2 2

9 E  Fo 2 F (22)
Fo

in which F 2   Q obs( t )  Q comp( t )  and Fo2   Q obs ( t )  Q m  .


n 2 n 2
10
i 1 i 1

11 The coefficient of efficiency in Eq. 22 provides a well accepted measure of fit between

12 computed and observed hydrographs, its value increasing toward unity as the fit of the

13 simulated hydrograph progressively improves. Values of E that exceed 0.8 (Green and

14 Stephenson, 1986) are considered to reflect a good correlation between the adjusted and the

15 Muskingum and HEC-RAS hydrographs in this study. Results of these statistical tests where

16 the Muskingum and HEC-RAS hydrographs at Caher Park in Fig. 11 are compared to

17 adjusted hydrographs developed from observed data are shown in Table 2.

18

19 Table 2

20

21 Table 2 shows that the modified Muskingum method produces outflow hydrographs that

22 compare favourably with the adjusted hydrographs and are comparable to those developed

23 through HEC-RAS modelling. The differences that exist between both the Muskingum and

24 HEC-RAS hydrographs and the adjusted hydrographs are likely to result from the simplifying

-27-
1 assumptions inherent in both methods. In reality, main channel and floodplain momentum

2 exchanges in the River Suir may increase flood wave attenuation and decrease travel time.

3 However, the modified routing parameters were developed from a multi-variate analysis of

4 data from 1-D modelling of a generalised channel in which the energy losses from these

5 interactions are not included. Furthermore, the influences of geometrical, resistance and

6 hydrograph properties in this analysis were assessed independently in these analyses.

7 Similarly, main channel and floodplain interactions are unaccounted for in the HEC-RAS

8 hydrographs developed from 1-D modelling of the River Suir study reach. The impacts of

9 these interactions however, appear low and the relationships in Eq. 15 and Eq. 16 are

10 considered to provide reasonable estimates for K and x in the presented method. The

11 satisfactory performance of the modified method further implies that the storage equation in

12 Muskingum routing methods is a substitute for the momentum equation in hydraulic routing

13 approaches in typical Irish rivers and therefore it is reasonable to relate the routing parameters

14 to channel and flow characteristics (Perumal, 1992b).

15 It should be noted however that the limitations of the approach are the same as in most

16 hydrological or Muskingum flood routing methods and therefore, the results from this method

17 should be confirmed if applied to river reaches where backwater and inertia effects are

18 significant, where floodplain sinuosity is excessively high or where significant lateral

19 momentum exchanges between main channel and floodplain zones are influential. The

20 method however, does provide a simple and inexpensive method for obtaining preliminary

21 estimates of the time and shape of a flood hydrograph as it travels overbank along a river

22 reach.

23

-28-
1 7 Conclusions

2 A modified linear Muskingum routing method suitable for floodplain flows is

3 presented in this paper. Muskingum flood routing methods are based on storage – discharge

4 relationships in river systems and can satisfactorily produce outflow hydrographs in river

5 systems where inertia effects and backwater influences are small. However, values of the

6 routing parameters, x and K, that describe the storage characteristics of a river reach are

7 usually derived analytically from observed upstream and downstream hydrographs extracted

8 from historical flow records. The proposed method uses explicit relationships for K and x

9 described in terms of standard geometrical and resistance properties of channels with

10 floodplains together with properties of standard inflow hydrographs. The relationships were

11 based on regression analysis of computational data generated through 1-dimensional

12 modelling of a generalised river reach. The expressions are:


0.60
b 
L n n  fp  TB0.07
0.24
fp
0.42
mc L0.03 S0fp.16 TB0.39
13 K  0.794  Bbf  x  0.035
0.06 0.05
 Q   b fp 
Sfp  
0.53 0.09 P 

  n 
mc
0.06
n 0fp.008
 2 b fp  B bf   Bbf 

14 Application of the method requires that  > 0 and therefore horizontal floodplains are

15 represented by a near-zero value of . Furthermore, the modified method assesses floodplain

16 effects and therefore flows must be sufficient to produce out of bank conditions. Application

17 of the method in a reach of the River Suir, Co. Tipperary, Ireland was shown to produce

18 outflow hydrographs that compared favourably to those developed from measured flow

19 records.

20

21

22

-29-
1 Acknowledgements

2 The authors wish to acknowledge the financial support provided by the Irish Office of

3 Public Works (OPW) and the Centre for Water Resources Research (CWRR) in the School of

4 Architecture, Landscape and Civil Engineering in UCD for undertaking this research. The

5 OPW funding was provided through the Irish Flood Studies Update (FSU) programme.

8 References

9 Al-Humoud, J.M., Esen, I. I., 2006. Approximate methods for the estimation of Muskingum

10 Flood Routing Parameters. Water Resources Management., 20: 979-990.

11 Amein, M., Fang, C.S., 1970. Implicit flood routing in natural channels. J. Hydraulic

12 Engineering., ASCE, February, Paper no. 7773, 96(12):2481-2500.

13 Bajracharya, K., Barry, D.A., 1997. Accuracy criteria for linearised diffusion wave flood

14 routing. J. Hydrol., 195:200-217.

15 Birkhead, A.L., James, C.S., 1997. Synthesis of rating curves from local stage and remote

16 discharge monitoring using nonlinear Muskingum routing. J. Hydrol., 205: 52-65.

17 Castro, J.M., Jackson P.L., 2001. Bankfull discharge recurrence intervals and regional

18 hydraulic geometry relationships patterns in the Pacific Northwest, USA. Journal of the

19 American Water Resources Association, 37(5), pp. 1249-1262.

20 Chang, C.N., Singer, E.D.M., Koussis, A.D., 1983. On the mathematics of storage routing. J.

21 Hydrol., 61:357-370.

22 Choudhury, P., Shrivastava, R.K., Narulkar, S.M., 2002. Flood routing in river networks

23 using equivalent Muskingum inflow. J. Hydrol. Eng., 7(6):413-419.

24 Chow V.T., 1959. Open-channel Hydraulics. McGraw-Hill Book Company, New York,

25 ISBN 070107769.

-30-
1 Chow, V.T., 1964. Handbook of Applied Hydrology. McGraw-Hill, New York.

2 Chu, H.J., 2009. The Muskingum Flood routing model using a Neuro-Fuzzy approach.

3 KSCE Journal of Civil Engineering., 13(5):371-376.

4 Cowan, W.L., 1956. Estimating hydraulic roughness coefficients. Agricultural Engineering,

5 vol. 37, no. 7, pp. 473-475, July.

6 Cunge, J.A., 1969. On the subject of a flood propagation computation method (Muskingum

7 method). J. Hydraul. Res., 7 (2):205-230.

8 Dooge, J.C.I., 1973. Linear theory of hydrologic systems. U.S. Dep. Agric., Agric. Res.

9 Serv., Tech. Bull. No. 1468.

10 Dooge, J.C.I., Strupczewski, W.G., Napiorkowski, J.J., 1982. Hydrodynamic derivation of

11 storage parameters of the Muskingum model. J. Hydrol., 54:371-387.

12 Gallati, M., Maione, U., 1977. Perspective on mathematical models of flood routing. Proc. of

13 the workshop held at the IBM scientific Centre, Pisa, Italy, John Wiley and Sons,

14 London, pp. 169-179.

15 Geem, Z.W., 2006. Parameter estimation for the nonlinear Muskingum model using the

16 BFGS technique. Journal of Irrigation and Drainage Engineering., ASCE, 132(5):474-

17 478.

18 Gelegenis, J.J., Serrano, S.E., 2000. Analysis of Muskingum equation based on flood routing

19 schemes. J. Hydraulic Engineering., January, Technical note no. 18898, 5(1):102-105.

20 Ghavasieh, A.R., Poulard, C., Paquier, A., 2006. Effect of roughened strips on flood

21 propagation: Assessment on representative virtual cases and validation. J. Hydrol.,

22 318:121-137.

23 Gill, M.A., 1978. Flood routing by the Muskingum method. J. Hydrol., 36:353-363.

24 Gill, M.A., 1979. Critical examination of the Muskingum method. Nordic Hydrology., 10:

25 261-270.

-31-
1 Green, I.R.A., Stephenson, D., 1985. Comparison of Urban Drainage Models for Use in

2 South Africa. WRC Report No 115/6/86. Water Research Commission, Pretoria, RSA.

3 Green, I.R.A. and Stephenson, D., 1986. Criteria for comparison of single event models.

4 Hydrological Sciences Journal., 31(3):395-411.

5 Guang-Te, W., Singh, V.P., 1992. Muskingum method with variable parameter for flood

6 routing in channels. J. Hydrol., 134:57-76.

7 Hollingrake, P.G., Millington, R.J., 1994. Discharge and Roughness Assessment at UK

8 Gauging Stations, HR Wallingford, Report SR 286, March.

9 Jewitt, G.P.W., Schulze, R.E., 1999. Verification of the ACRU model for forest hydrology

10 applications. Water SA., 25(4):483-490.

11 Kim, J.H., Geem, Z.W., Kim, E.S., 2001. Parameter estimation of the nonlinear Muskingum

12 model using harmony search. JAWRA., 37(5):1131-1138.

13 Knight, D.W., Shiono, K., 1996. River Channel and Floodplain Hydraulics. In: Floodplain

14 Processes. Anderson et al (ed.), Wiley, 139-181.

15 Kundzewicz, Z.W., Strupczewski, W.G., 1982. Approximate translation in the Muskingum

16 model. Hydrological Sciences Journal., 27(1):19-26.

17 Linsley, R.K., Kohler, M.A., Paulhus, J.L.H., 1975. Hydrology for Engineers. McGraw-Hill,

18 New York, N.Y., 2nd ed., 340 pp.

19 Marquardt, D.W., 1963. An algorithm for least-squares estimation of nonlinear parameters.

20 J. Soc. of Industrial and Appl. Mathematics., 11:431-441.

21 McCarthy, G.T., 1938. The unit hydrograph and flood routing. Presented at Conf. North

22 Atlantic Div., U.S. Army Corps Eng., New London, CT. US Engineering Office,

23 Providence RI.

24 McCuen, R.H., 1998. Hydrologic Analysis and Design. Second Edition, Prentice Hall, NJ,

25 Chapter 10, section 10.5.1 Estimation of the Muskingum Routing Coefficients.

-32-
1 Mimikou, M., 1983. A study of Drainage Basin Linearity and Non Linearity. Journal of

2 Hydrol., 64: 113-134.

3 Mohan, S., 1997. Parameter estimation of nonlinear Muskingum models using genetic

4 algorithm. J. Hydraulic Engineering., ASCE, February, Paper no. 9812, 123(2):137-

5 142.

6 Moussa, R., Bocquillon, C., 1996. Criteria for the choice of flood-routing methods in natural

7 channels. J. Hydrol., 186:1-30.

8 Moussa, R., Bocquillon, C., 2009. On the use of the diffusive wave for modelling extreme

9 flood events with overbank flow in the floodplain. J. Hydrol., 374:116-135.

10 Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual model. J. Hydrol.,

11 10(2): 82-92.

12 NERC (Natural Environment Research Council), 1975. Flood Studies Report. Vol. I

13 Hydrological Studies. Vol II Meteorological Studies. Vol III Flood Routing Studies.

14 Vol IV Hydrological Data. Vol V Maps, London.

15 O’Connor, K.M., Goswami, M., 2010. Hydrograph Width Analysis – Final Report for WP

16 3.1 of Irish Flood Studies Update programme. Unpublished Report.

17 Perumal, M., 1992a. Multi linear Muskingum flood routing method. J. Hydrol., 133:259-

18 272.

19 Perumal, M., 1992b. The cause of negative initial outflow with the Muskingum method.

20 Hydrological Sciences., 37(4):391-401.

21 Petit, F., Pauquet, A., 1997. Bankfull discharge recurrence interval in gravel-bed rivers.

22 Earth surface processes and landforms., 22, pp. 685-693.

23 Richards, K. 1982. Rivers, Form and Process in Alluvial Channels. Methuen, 358 pp.

24 Rogers, W.F., 1980. A practical Model for Linear and Nonlinear Runoff. Journal of Hydrol.,

25 46: 51-78.

-33-
1 Schulze, R.E., Lynch, S.D., Smithers, J.C., Pike, A., Schmidt, E.J., 1995. Statistical output

2 from ACRU. In: Schulze RE (ed.) Hydrology and Agrohydrology: A text to accompany

3 the ACRU 3.00 Agrohydrological Modelling System. (Ch. 21). WRC Report No TT

4 69/95 Water Research Commission, Pretoria, RSA. AT21-1 - AT21-35.

5 Singh, V.P., McCann, R.C., 1980. Some notes on Muskingum method of flood routing. J.

6 Hydrol., 48:343-361.

7 Singh, V.P., 1988. Hydrologic systems. Prentice Hall, Eaglewood, Cliffs, NJ, USA.

8 Singh, V.P., Aminian H., 1986. An Empirical Relation Between Volume and Peak of Direct

9 Runoff. Water Resources Bulletin 22: 725-730.

10 Smith, A.A., 1979. A generalized approach to kinematic flood routing. J. Hydrol., 45:71-89.

11 Sweeney, J., Murphy, C., 2010. Estimation of the Index Flood At Ungauged Sites – Final

12 Report of Work-Package 2.3 of the Irish Flood Studies Update (FSU) programme.

13 Unpublished Report.

14 Tang, X., Knight, D.W., Samuels, P.G., 1999a. Variable parameter Muskingum-Cunge

15 method for flood routing in a compound channel. J. Hydraul. Res., IAHR, 37(5): 591-

16 614.

17 Tang, X., Knight, D.W., Samuels, P.G., 1999b. Volume Conservation in Variable Parameter

18 Muskingum-Cunge Method. J. Hydraul. Engrg., ASCE, June, 610-620.

19 Tang, X., Knight, D.W., Samuels, P.G., 2001. Wave speed – discharge relationship from

20 cross section survey. Proc. Instn. of Civil Engineers, Water & Maritime Engineering,

21 London, June, 148(2): 81-96.

22 Tung, Y.K., 1985. River flood routing by nonlinear Muskingum method. J. Hydraulic

23 Engineering., ASCE, December, Paper no. 20217, 111(12):1447-1460.

24 Viessman, W., Harbaugh, T.E., Knapp, J.W., 1972. Introduction to hydrology. New York: In

25 text educational publishers, 415p.

-34-
1 Wang, G. T., Yao, C., Okoren, C., Chen, S., 2006. 4-Point FDF of Muskingum method based

2 on the complete St Venant equation. J. Hydrol., 324:339-349.

3 Wilson, E.M., 1990. Engineering Hydrology. MacMillan, Hong Kong.

4 Wolff, C.G., Burges, S.J., 1994. An analysis of the influence of river channel properties on

5 flood frequency. J. Hydrol., 153:317-337.

6 Yoon, J., Padmanabhan, G., 1993. Parameter estimation of linear and nonlinear Muskingum

7 models. J. Water Res. Planning Mgmt. Div., ASCE, September/October, Technical

8 note no. 774, 119(5):600-610.

10 Appendix

11 Table A1

12

13

14

-35-
1 List of Figures

2 Fig. 1. Stages in the development of the modified flooding parameters.

3 Fig. 2. Development of theoretical flood hydrographs for routing in generalised river model.

4 Fig. 3. Notation describing cross-sectional geometry in model of generalised river reach.

5 Fig. 4. Influence of hydrograph and floodplain properties on Muskingum storage constant

6 (K).

7 Fig. 5. Influence of hydrograph and floodplain properties on Muskingum storage weighting

8 factor (x).

9 Fig. 6. Comparison of simulated storage weighting parameter (K) and storage time constant

10 (x) to that calculated using index plotted on linear scales.

11 Fig. 7. Study region of the River Suir catchment.

12 Fig. 8. Hydrographs at Killardry, New Bridge and Caher Park for 1954/ ’55 (a), 1986 (b) and

13 2004 (c) storm events.

14 Fig. 9. Measured and routed outflow hydrographs at Caher Park for assessment of flow

15 contribution from ungauged tributaries.

16 Fig. 10. Routed hydrographs of New Bridge (16008) at Caher Park (16009) using measured

17 data (‘adjusted’), HEC-RAS modelling (HEC-RAS) and the modified Muskingum

18 method (Muskingum) using floodplain widths extracted from the HEC-RAS model

19 simulations.

20 Fig. 11. Routed hydrographs of New Bridge (16008) at Caher Park (16009) using measured

21 data (adjusted), HEC-RAS modelling (HEC-RAS) and the modified Muskingum

22 method (Muskingum) using floodplain widths extracted from catchment floodplain

23 attenuation indicators (FAIs).

24

25

-36-
1 List of Tables

2 Table 1. Catchment descriptors for the 1954/ ’55, 1986 and 2004 flood events used in the

3 modified Muskingum routing method.

4 Table 2. Goodness of fit test results for Modified Muskingum (Muskingum) and HEC-RAS

5 hydrographs by comparison with Caher Park adjusted hydrographs.

7 Table A1. Summary of simulations and Muskingum routing parameters from HEC-RAS

8 modelling of generalised river reach.

10

-37-

You might also like