You are on page 1of 9

Simple, Robust, and Efficient Algorithm for

Gradually Varied Subcritical Flow Simulation


in General Channel Networks
Dejun Zhu1; Yongcan Chen2; Zhiyong Wang3; and Zhaowei Liu4
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Abstract: This paper details the development of a method for subcritical flow modeling in channel networks by using the implicit finite-
difference method. The method treats backwater effects at the junction points on the basis of junction-point water stage prediction and
correction (JPWSPC). It is applicable to flows in both looped and nonlooped channel networks and has no requirement on the flow directions.
The method is implemented in a numerical model, in which the Saint-Venant equations are discretized by using the four-point implicit
Preissmann scheme, and the resulting nonlinear system of equations is solved by using the Newton-Raphson method. With the help of
JPWSPC, each branch is computed independently. This guarantees the simplicity, efficiency, and robustness of the numerical model.
The model is applied to two hypothetic channel networks and a real-life river network in South China. All the networks contain both branched
and looped structures. The simulated results compare well with the results from literature or the measurements. DOI: 10.1061/(ASCE)HY
.1943-7900.0000356. © 2011 American Society of Civil Engineers.
CE Database subject headings: Critical flow; Open channels; Networks; Finite difference method; Algorithms; Flow simulation.
Author keywords: Subcritical flow; Open channel; Network; Finite-difference scheme; Junction-point water stage prediction and
correction (JPWSPC).

Introduction Fread (1973) used an iterative method to simulate transient flow


in a channel network that assumes tributary flows as lateral inflows.
Flows in open-channel networks are governed by the one- Because the method needs to distinguish the principal and tributary
dimensional Saint-Venant equations. These nonlinear hyperbolic branches, which is impossible for looped networks, it is primarily
partial differential equations cannot be solved analytically in com- applicable to nonlooped networks. More recently, algorithms with
plex conditions. Among different numerical methods, the implicit specific node-numbering schemes were proposed to reduce the
finite-difference method has been widely used (Choi 1993; Fread bandwidth of the resulting solution matrix (Choi 1993; Nguyen
1973; Islam et al. 2005; Ji 1998; Nguyen and Kawano 1995; Sen and Kawano 1995). However, the use of a specific node-numbering
and Garg 1998, 2002). It is quite challenging to model subcritical scheme for a general channel network is difficult, especially when
flows in a channel network consisting of a number of intercon- the network is looped. Another category of widely applied algo-
nected channels because of backwater effects at the channel junc- rithms separate out the end-node variables first, then employ the
tions. The entire network needs to be considered as a single unit, end-node variables to model each branch independently one by
and flow in all channels needs to be simulated simultaneously one (Islam et al. 2005; Ji 1998; Sen and Garg 1998, 2002). These
(Islam et al. 2005). Simultaneous solution of the entire channel net- algorithms generally comprise three phases: first, establish branch
work requires a large computer memory and enormous computa- equations; second, solve the resultant global matrix to get end-node
tional effort. To reduce this computational difficulty, a number of variables; third, substitute the end-node variables into the branch-
algorithms for flows in channel networks have been developed. wise matrix to get variables at each interior node. Therefore, these
algorithms are referred to as three-phase algorithms subsequently in
1
Post-doctoral Researcher, State Key Laboratory of Hydroscience this paper. Three-phase algorithms are applicable in both looped
and Engineering, Tsinghua Univ., Beijing 100084, China (corresponding and branched channel networks, but they could be unstable when
author). E-mail: zhudj02@mails.tsinghua.edu.cn the initial conditions are inaccurate or the time step and subreach
2
Professor, State Key Laboratory of Hydroscience and Engineering,
length are not carefully chosen (Lv et al. 2007). In addition, solving
Tsinghua Univ., Beijing 100084, China. E-mail: chenyc@mail.tsinghua
.edu.cn the global matrix, which has a dimension of 4M × 4M (M is the
3
Ph.D. Student, State Key Laboratory of Hydroscience and Engineer- number of branches), may be time-consuming when M is large.
ing, Tsinghua Univ., Beijing 100084, China. E-mail: wang-zy@mails This paper presents details of a novel method proposed by Zhu
.tsinghua.edu.cn et al. (2009). Like the three-phase algorithms, the method is also
4
Associate Professor, State Key Laboratory of Hydroscience and Engi- applicable to both looped and nonlooped channel networks, and has
neering, Tsinghua Univ., Beijing 100084, China. E-mail: liuzhw@tsinghua no limitations on the flow directions. An advantage of the method is
.edu.cn that there is no need to establish and solve the global branch equa-
Note. This manuscript was submitted on March 9, 2010; approved on
tions. Instead, the method treats backwater effects at the channel
October 28, 2010; published online on December 22, 2010. Discussion per-
iod open until December 1, 2011; separate discussions must be submitted junctions by means of junction-point water stage prediction and
for individual papers. This paper is part of the Journal of Hydraulic En- correction (JPWSPC). Such a treatment is simple, robust, efficient,
gineering, Vol. 137, No. 7, July 1, 2011. ©ASCE, ISSN 0733-9429/2011/ and easy to implement in programming. The outline of this paper is
7-766–774/$25.00. as follows: the governing equations and their discretization are

766 / JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011

J. Hydraul. Eng. 2011.137:766-774.


given in the following two sections; then an overview of the ∂Fmj ∂Fmj ∂Fmj
a2j;1 ¼ ; a2j;2 ¼ ; a2j;3 ¼ ;
JPWSPC method is presented; the details of the method are given ∂Qj ∂Aj ∂Qjþ1
in the next section; and in the succeeding section, the proposed
∂Fmj
model is applied to two hypothetical channel networks and a a2j;4 ¼
real-life river network in South China, with results comparing well ∂Ajþ1
with the results from literature and the observed data; the conclu-
sions are reached in the last section. where the functions Fc and Fm = discretized continuity and
momentum equations, respectively; RFc and RFm = residuals of
Fc and Fm, respectively; Δ = difference between two iterations.
Governing Equations

The well-known one-dimensional Saint-Venant equations describe Overview of the JPWSPC Method
the conservation of mass and momentum, which may be expressed
The correct water stage at the junctions should satisfy the relation
as follows:
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

described by Eq. (3). Suppose values of the inflow and outflow


∂A ∂Q discharges at a junction point are positive and negative, respec-
þ q¼0 ð1Þ tively. It can be concluded from the nature of subcritical flow that,
∂t ∂x if the predicted water stage at a junction point is higher than the
correct value, the sum of the calculated discharges (called the
net discharge hereinafter) is negative. Conversely, if the predicted
 
∂Q ∂ Q2 ∂Z value is lower, the net discharge is positive. To meet the condition
þ þ gA þ gASf ¼ 0 ð2Þ described by Eq. (3) (i.e., the net discharge equals zero), the junc-
∂t ∂x A ∂x
tion-point water stage can be corrected iteratively until the net dis-
charge is almost zero. It is on this principle that the proposed
where A = cross-section area; Q = discharge; q = rate of lateral
JPWSPC method is based, and details of the prediction and correc-
inflow; Z = water stage; g = gravity acceleration; and Sf = frictional
tion processes are illustrated in the next section.
slope. For channel flow, Sf may be estimated by Manning’s
For each branch, there are 2N variables, where N is the number
equation, Sf ¼ n2 QjQj=A2 R4=3 , where n = Manning’s roughness
of nodes per branch; Eqs. (5) and (6) constitute a set of 2ðN  1Þ
coefficient; and R = hydraulic radius. The hydraulic conditions
equations obtained from the N  1 interior reaches. Two more
at junctions can be expressed by mass and energy conservation
equations per branch are obtained corresponding to the two end
equations, as shown in Eqs. (3) and (4). More complex conditions
nodes after applying either internal or external boundary condi-
(Kesserwani et al. 2008) are not included in this paper.
tions. It can be seen from Eqs. (3) and (4) that variables of different
X X X branches are interlinked at the junctions. By imposing a predicted
Q¼ Qi  Qo ¼ 0 ð3Þ or corrected water stage at each junction point, which serves as the
internal condition, each branch can be computed independently.
There is no need of establishing and solving global branch equa-
tions, thus the memory requirement and computational time are
Zi ¼ Zo ð4Þ
greatly reduced.
The prediction and correction of junction-point water stages is
where subscripts i and o = inflow and outflow, respectively.
carried out synchronously with the Newton-Raphson iteration. The
flowchart of the proposed algorithm is illustrated in Fig. 1. For the
sake of comparison, the flowchart of the widely used three-phase
Discretization of the Governing Equations algorithms, which are also applicable to general networks, is illus-
In this study, Eqs. (1) and (2) are discretized using the four-point trated in Fig. 2.
implicit Preissmann scheme. This results in a system of nonlinear The differences between the algorithms are highlighted with
equations, and the system is then solved by using the iterative virtual boxes. It can be seen that the complicated process of
Newton-Raphson method. Application of the Newton-Raphson establishing and solving branch equations is replaced by the simple
method for the jth reach, which links two consecutive nodes (cross prediction and correction of junction-point water stages. In the pro-
sections) j and j þ 1, results in the following equations: posed algorithm, each individual branch is simulated separately
with external or internal boundary conditions, i.e., junction-point
water stage, at end nodes. Since the coefficient matrix of the linear
a2j1;1 ΔQj þ a2j1;2 ΔAj þ a2j1;3 ΔQjþ1 þ a2j1;4 ΔAjþ1 þ RFcj
system is banded, efficient solution algorithms can thus be em-
¼0 ð5Þ ployed. The required number of computations for each branch with
N nodes per iteration is 38N (Fread 1973), which equals the total
number of computations required in the first and third phases of the
three-phase algorithms (Islam et al. 2005; Ji 1998; Sen and Garg
a2j;1 ΔQj þ a2j;2 ΔAj þ a2j;3 ΔQjþ1 þ a2j;4 ΔAjþ1 þ RFmj ¼ 0 ð6Þ 1998, 2002). The number of computations required for the junc-
tion-point water stage prediction and correction, which has the
where j ¼ 1, 2, 3…, and same order as the number of iterations, is negligible compared
to that of solving the global matrix. Therefore, the proposed algo-
∂Fcj ∂Fcj ∂Fcj rithm significantly reduces the number of computations per itera-
a2j1;1 ¼ ; a2j1;2 ¼ ; a2j1;3 ¼ ;
∂Qj ∂Aj ∂Qjþ1 tion by avoiding the formation and solution of the global matrix.
As mentioned previously, solving the 4M × 4M sparse global
∂Fcj
a2j1;4 ¼ matrix leads to the high computational cost in the three-phase al-
∂Ajþ1 gorithms when the total number of branches, M, is large. Moreover,

JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011 / 767

J. Hydraul. Eng. 2011.137:766-774.


input data branch independently within one round of iteration, so the algo-
rithm is extremely easy to implement in programming.

yes running time


output results is greater Method Details
than endtime?
Predictor Step
no
A time step is divided into a predictor and several following cor-
define boundary
conditions
rector steps. During the predictor step, the junction-point water
stages at the end of the previous time step, or initial conditions
for the first time step, are used as predicted values. Using these
predict junction stages water stages together with the newly updated boundary conditions,
the water stage and discharge at each node, including the junction
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

model branches points, are calculated.

Corrector Step

no meet the yes


The predictor step is followed by a series of corrector steps. As
update variables and depicted in Fig. 3, Node A with an x-coordinate of x0 represents
precision
correct junction stages
criteria? a junction point, and UA and AD are two branches joining at Node
A, and the arrow represents the flow direction.
The positive characteristic curve λþ is defined by Eq. (7)
according to the Saint-Venant equations (Tan 1998) along the
Fig. 1. Flowchart of the proposed algorithm
upstream reach UA:
rffiffiffiffiffiffiffi
dx Q A
¼ þ g ð7Þ
input data dt A B
where B = water surface width. The characteristic relation among
the flow variables valid along the positive characteristic curve may
yes running time be described by the following equation:
output results is greater
than endtime? Z A rffiffiffiffiffi !
Q g
d þ da ¼ gðS0  Sf Þdt ð8Þ
A 0 ab
no
define boundary where b = water surface width corresponding to the cross-section
conditions
area of a; and S0 = bed slope. Variation of the right-hand side of
Eq. (8) is assumed negligible during the prediction-correction pro-
set up branch equations
(phase 1) cess; integrating Eq. (8) along the positive characteristic passing
through ðx0 ; t 0 þ ΔtÞ in the x  t space results in a relation
solve branch equations described by Eq. (9)
(phase 2) Z A rffiffiffiffiffi
Q g
þ da ¼ C ð9Þ
backward substitution
A 0 ab
where C = constant determined by the discharge and water stage on
the curve at the time of t 0 . Differentiating both sides of Eq. (9) with
no meet the yes respect to water depth h leads to the following relation:
update variables
precision
(phase 3)
criteria? t flow
λ - λ+

Fig. 2. Flowchart of three-phase algorithms


t0+∆t

the accuracy of solving the global equations greatly affects the reli-
t0
ability of the final results. When the initial conditions are inaccurate
or the time step and subreach length are not carefully chosen, three-
phase algorithms has been reported to be unstable (Lv et al. 2007).
Moreover, similar to the three-phase algorithms, the proposed o x0 x
algorithm is also amenable to parallel processing, as the simulation
of each branch can be carried out concurrently with the help of the U A D
external and internal boundary conditions, i.e., junction-point water
Fig. 3. Characteristic curves
stage. In addition, the present algorithm computes each individual

768 / JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011

J. Hydraul. Eng. 2011.137:766-774.


dQi conditions, to carry out a round of Newton-Raphson iteration
dhi ¼ pffiffiffiffiffiffiffiffiffiffiffi ð10Þ throughout the network.
Qi Bi =Ai  gAi Bi
3. Correct the water stage at each junction point by using
Similarly, for the downstream reach AD, where the junction Eq. (17).
point is at its upstream end, the Saint-Venant equations give the 4. Use the corrected water stages at the junctions as the revised
following relation: internal boundary conditions and carry out another round of
Newton-Raphson iteration throughout the network.
dQo 5. Repeat Steps 3 and 4 until flow variables at each computational
dho ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi ð11Þ
Qo Bo =Ao þ gAo Bo node meet the precision criteria, i.e., jΔQ=Qj and jΔA=Aj do
not exceed the prescribed tolerance.
Once the discharge of each branch end-node is calculated,
adding Pup thesePdischarges
P provides
P the net discharge at the junc-
tion as Q ¼ Qi  Qo . If Q ≠ 0, the water stage is then Results and Discussion
corrected, with the aim of bringingP
the net discharge to zero. Thus,
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

the net discharge is adjusted by Δ Q, which satisfies the follow- The performance of the proposed algorithm are compared with those
ing relation: of two three-phase algorithms in this section. The algorithm of Sen
X X X X and Garg (2002), the algorithm of Schaffranek (see Islam et al.
QþΔ Q¼ Qþ ΔQ ¼ 0 ð12Þ 2005), and the proposed algorithm are called Algorithm-1,
Algorithm-2, and Algorithm-3, respectively. The Gauss-elimination
Eq. (12) is transformed into Eq. (13) by using Eqs. (10) and (11): method with maximum pivot strategy is used to solve the global
X matrix in the three-phase algorithms. The model was run on a Pen-
X pffiffiffiffiffiffiffiffiffiffiffi
Qþ Qi Bi =Ai  gAi Bi Δhi tium 4 personal computer, and a tolerance of 0.001 was set for
jΔQ=Qj and jΔA=Aj in all the simulations. Because all the algo-
X 
pffiffiffiffiffiffiffiffiffiffiffiffiffi rithms are found to produce almost the same results, only results
 Qo Bo =Ao þ gAo Bo Δho ¼ 0 ð13Þ from the proposed JPWSPC-based algorithm are presented herein.
Three cases were tested in total, of which two are hypothetical
Because ΔZ ¼ Δh and Z i ¼ Z o , Eq. (14) is derived from Eq. (13): networks taken from Islam et al. (2005) and Sen and Garg (2002),
P k respectively. The boundary conditions were digitized from the
Q curves presented in the papers because no exact data were available.
ΔZ ¼ P pffiffiffiffiffiffiffiffiffiffiffi
k P pffiffiffiffiffiffiffiffiffiffiffiffiffi ð14Þ
½ gAi Bi  Qi Bi =Ai  þ ½ gAo Bo þ Qo Bo =Ao  A small amount of discrepancy is expected between the present
results and theirs because of the inaccuracies in estimating the
where ΔZ = water stage increment between two consecutive boundary conditions from the graphs. The results of Islam et al.
corrector steps; the superscript k = kth corrector step; and k ¼ 0 in- (2005) and Sen and Garg (2002) were also extracted from the
dicates the predictor step. For simplicity, the denominator of Eq. (14) curves presented in the original papers and compared with the
is approximated by Ac =Δt, then present results. In Case 3, a real-life problem, the observed data
P at the end nodes were employed as the boundary conditions,
Δt Qk
ΔZ k ¼ ; k ¼ 0; 1; 2… ð15Þ and the observed data at two internal cross sections were compared
Akc with the computed results.
where Δt = the time step; Ac = artificial storage area hereinafter and Case 1
defined by Eq. (16), in which α = an adjustable parameter accounting
for the approximations made in the derivation: Hypothetical Network-1 is shown in Fig. 4, which was also studied
 Xh by Islam et al. (2005). Relevant characteristics for the channels are
pffiffiffiffiffiffiffiffiffiffiffi i given in Table 1. Upstream and downstream boundary conditions
Ac ¼ α gAi Bi  Qi Bi =Ai are specified at nodes 1–7 and 14, respectively, as illustrated
Xhpffiffiffiffiffiffiffiffiffiffiffiffiffi i in Fig. 5.
þ gAo Bo þ Qo Bo =Ao Δt ð16Þ Comparisons of the computed hydrographs and the results
of Islam et al. (2005) at two nodes are illustrated in Fig. 6. Some
differences are observed, especially the early and peak values. The
The stability increases with Ac , while convergence speed de-
creases with it. The value of α is recommended to be 1.0–2.0 based
1
on the experience of the authors; for all the cases in this paper, α is (1)
fixed at 1.2. Finally, the junction-point water stage is corrected as 8
(2) (8)
follows:
11
2 (12)
Z kþ1 ¼ Zk þ ΔZ k ; k ¼ 0; 1; 2… ð17Þ (9) 13 (14) 14
3 (3) 9
(11) (13)
(4)
Solution Procedure 4
12
For clarity, the computation steps within one time step are summa- (5) (10)
rized as follows: 5
(6) 10
1. Use the values at the end of the previous time step, or initial 6
stages for the first time step, as the predicted junction-point
7 (7)
water stages.
2. Use the predicted junction-point water stages as the internal
Fig. 4. Channel Network-1 (Islam et al. 2005, ASCE)
boundary conditions, together with the external boundary

JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011 / 769

J. Hydraul. Eng. 2011.137:766-774.


Table 1. Channel Characteristics for Network-1 (Data from Islam et al. 2005)
Channel Manning’s Number
number Length (m) Bed width (m) Side slope Bed slope coefficient of reaches
1,2,8,9 1,500 10 1∶1 0.00027 0.022 15
3,4 3,000 10 1∶1 0.00047 0.025 30
5,6,7,10 2,000 10 1∶1 0.00030 0.022 20
11 1,200 10 Vertical 0.00033 0.022 12
12 3,600 20 Vertical 0.00025 0.022 36
13 2,000 20 Vertical 0.00025 0.022 20
14 2,500 30 Vertical 0.00016 0.022 25
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Discharge Hydrograph at Node 1 7 Stage Hydrograph at Node 14


16 3
15 2.9
14 2.8
Q/m3⋅s -1

13 2.7

Z/m
12 2.6
11 2.5
10 2.4
9 2.3
0 10 20 30 40 0 10 20 30 40
t/h t/h

Fig. 5. Boundary conditions for Network-1 (Islam et al. 2005, ASCE)

Islam et al.
Zhu et al.
Discharge Hydrograph at Node 11 Discharge Hydrograph at Node 12
15 15
14 14
13 13
Q/m3⋅s -1
Q/m3⋅s -1

12 12
11 11
10 10
9 9
8 8
0 10 20 30 40 0 10 20 30 40
t/h t/h

Stage Hydrograph at Node 11 Stage Hydrograph at Node 12


3.9 3.9
3.7 3.7
3.5 3.5
3.3 3.3
Z/m

Z/m

3.1 3.1
2.9 2.9
2.7 2.7
2.5 2.5
0 10 20 30 40 0 10 20 30 40
t/h t/h

Fig. 6. Computed hydrographs for Network-1

maximum difference is about 2.0%, which appears at the peak of Table 2. Comparison of the Algorithms in Network-1
the discharge hydrograph at Node 12. This may be attributed to the Global Time per
aforementioned reason, e.g., the differences between the boundary matrix Simulation iteration
conditions employed. Algorithm size TNI time (s) ANI (ms)
To compare the efficiency and storage requirement of different
Algorithm-1 3,136 3,297 4.949 2.061 1.501
algorithms, the global matrix size, number of iterations and simu-
Algorithm-2 3,136 3,302 4.857 2.064 1.471
lation times are presented in Table 2, in which Δt ¼ 90 s. By
Algorithm-3 — 3,311 3.023 2.069 0.913
avoiding establishing the global branch equations, the present

770 / JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011

J. Hydraul. Eng. 2011.137:766-774.


10.0m3/s
b-j: jth node of b th branch algorithm reduces the storage requirement. As was pointed out in
4-1 : External boundary points the method overview, the present algorithm also reduces simulation
1-1 : Junctions time per iteration by avoiding solving the global matrix.
4-6 14.8m3/s
4.8m /s3 2-6 Table 2 shows neither the total number of iterations (TNI) nor
5-1 the average number of iterations (ANI) per Δt experiences much
5-6 15.6m3/s increase, so the algorithm is about 39% faster than the two three-
2-1 1-11
10.0m3/s 7-1 5.2m3/s 3-6 phase algorithms.
0.8m 3/s
3 3-1
15.2m /s 9-6 Case 2
10.0m3/s
7-6 9-1 Hypothetical Network-2 is shown in Fig. 7, which was also studied
8-6 6-1 6-610-1 14.4m3/s by Sen and Garg (2002). The branches are numbered and discre-
8-1 tized, as shown in Fig. 7. The steady state discharges are also
10-11 indicated in the picture and relevant characteristics are given in
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

Table 3. The upstream boundary condition is specified at Node


Fig. 7. Channel Network-2 (Sen and Garg 2002, ASCE)
8-1, as illustrated in Fig. 8. The downstream boundary conditions

Table 3. Channel Characteristics for Network-2 (Data from Sen and Garg 2002)
Manning’s Number
Channel number Length (m) Bed width (m) Side slope Bed slope coefficient of reaches
1, 10 2,000 100 1∶2 0.0001 0.025 10
2, 4, 5, 7, 8 1,000 50 1∶2 0.0002 0.025 5
3, 6, 9 1,000 75 1∶2 0.0001 0.025 5

Sen and Garg


Zhu et al.
Discharge Hydrograph at Node 8-1 Discharge Hydrograph at Node 1-1
50 30
40 25
20
Q/m3⋅s -1

Q/m3⋅s -1

30
15
20
10
10 5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t/h t/h

Discharge Hydrograph at Node 7-6 Discharge Hydrograph at Node 6-1


6 50
4
40
2
Q/m3⋅s -1

Q/m3⋅s -1

0 30
-2 20
-4
-6 10
-8 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t/h t/h

Discharge Hydrograph at Node 3-6 Discharge Hydrograph at Node 10-11


40 40
35 35
30 30
Q/m3⋅s -1

Q/m3⋅s -1

25 25
20 20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
t/h t/h

Fig. 8. Computed hydrographs for Network-2

JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011 / 771

J. Hydraul. Eng. 2011.137:766-774.


Table 4. Comparison of the Algorithms in Network-2 are specified at the other external boundary nodes with a constant
Global Time per depth of 5 m and remain unchanged with time.
matrix Simulation iteration The computed discharge hydrographs at a few key nodes are
Algorithm size TNI time (s) ANI (ms) illustrated in Fig. 8. Flow reversal in Branch 7 can be observed.
The proposed JPWSPC method performs well for such a complex
Algorithm-1 1,600 756 0.456 1.575 0.604
network. While the time-varying boundary conditions are imposed
Algorithm-2 1,600 756 0.456 1.575 0.603 on all the eight external end nodes in Network-1, only one time-
Algorithm-3 — 761 0.332 1.585 0.436 varying boundary condition is imposed in Network-2; thus, the
errors in the boundary conditions have a relatively smaller effect on
the results of Network-2, and the discrepancy between the present
Table 5. Comparison of the Algorithms in the Refined Version of results and those of Sen and Garg (2002) are very small. The global
Network-2 matrix size, number of iterations and simulation time are listed in
Global Time per Table 4, in which Δt ¼ 180 s. It shows that the three algorithms
behave similarly as in Network-1; the present algorithm reduces
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

matrix Simulation iteration


Algorithm size TNI time (s) ANI (ms) the storage requirement and is 28% faster than the two three-phase
algorithms.
Algorithm-1 1,600 1,502 0.939 1.565 0.625 The storage requirement and simulation time saved by the
Algorithm-2 1,600 1,502 0.936 1.565 0.623 present algorithm depend on the size of the global matrix, and
Algorithm-3 — 1,501 0.686 1.564 0.457 in turn depend on the number of branches in a network. Hence,
the advantage of the proposed algorithm over the three-phase algo-
rithms depends on the ratio of the number of branches to the total
number of nodes in the network. In an extreme situation, if a net-
work consists of only a single branch, the storage requirement and
Sanya East River Junction simulation time needed are the same for all of the algorithms. The
greater is the ratio, the more advantageous is the proposed algo-
) rithm over three-phase algorithms.
(2
Jinjiling Bridge To further compare the algorithms, the model was applied to a
(1 refined version of Network-2. The channel characteristics were
)
(3) kept unchanged, but the subreach length were halved. This nearly
doubled the total number of nodes in the system. The results are
(4)

identical to those shown in Fig. 8. The global matrix size, number


Sanya East Bridge
(5)

Ocean Sanya West Bridge of iterations, and simulation time are also presented in Table 5, in
Legend
(6)
River
which Δt ¼ 90 s. Table 5 shows the proposed algorithm is 26%
Sanya River Mouth Shoreline faster than the two three-phase algorithms.
Gauging Station
Case 3
Fig. 9. Location map of the Sanya River system The Sanya River is the most important river in Sanya City, a famous
tourist city in South China. Flows in the studied segments, as illus-
trated in Fig. 9, are tidal. The river network contains branches and
Table 6. River Characteristics for the Sanya River System
loops. A network of 6 branches and 82 computational nodes, with
lengths between two nodes ranging between 200 and 300 m, is used
River Length Average Manning’s Number of to model the Sanya River system (see Table 6). A constant dis-
number (m) bed slope coefficient reaches charge of 4:972 m3 =s at the Sanya East River Junction and the ob-
1 2,800 0.00015 0.05 14 served stage hydrograph at Jinjiling Bridge are adopted as the
2 5,000 0.0005 0.05 25 upstream boundary conditions; the tidal level fluctuation at the river
3 250 0 0.05 1 mouth is used as the downstream boundary condition; the time step
4 3,000 0.00017 0.05 15 Δt ¼ 90 s. Fig. 10 depicts the stage hydrographs at Jinjiling Bridge
and Sanya River Mouth.
5 4,700 0.0001 0.05 18
Fig. 11 illustrates comparison between the observed and simu-
6 600 0 0.05 3
lated stage hydrographs, and generally good agreements can be

Jinjiling Bridge Sanya River Mouth


2 2
1.5 1.5
1 1
Z/m

Z/m

0.5 0.5
0 0
-0.5 -0.5
0 5 10 15 20 25 0 5 10 15 20 25
t/h t/h

Fig. 10. Upstream and downstream boundary conditions

772 / JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011

J. Hydraul. Eng. 2011.137:766-774.


Simulated
Observed
Sanya East Bridge Sanya West Bridge
2 2
1.5 1.5
1 1

Z/m
Z/m
0.5 0.5
0 0
-0.5 -0.5
0 5 10 15 20 25 0 5 10 15 20 25
t/h t/h

Fig. 11. Comparison of observed and simulated stage hydrographs


Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

observed. The computed tidal height is 2.009 m and 2.035 m at the Notation
Sanya East and West Bridges, respectively, whereas the observed
values are 2.01 m and 2.05 m., respectively. The maximum differ- The following symbols are used in this paper:
ences appear right after the peak levels, which are approximately A = cross-section area;
0.1 m and 0.13 m at the Sanya East and West Bridges, respectively. Ac = artificial storage area;
The agreement shows the JPWSPC method is also applicable to C = constant;
real-life complex river networks under tidal influences. Fc = function representing discretized continuity equation;
Fm = function representing discretized momentum equation;
g = gravity acceleration;
Conclusions h = water depth;
N = number of nodes per branch;
Details of the development of the JPWSPC method for subcritical n = Manning’s roughness coefficient;
channel network flows are given in this paper. The method incor- Q = discharge;
porates the backwater effects at the junctions of channel networks q = rate of lateral inflow;
R = hydraulic radius;
and iteratively solves for the water levels and discharges. This
RFc = function representing residual of discretized
method does not require any specific node-numbering strategy
continuity equation;
or the need to form and solve the global branch equation, making
RFm = function representing residual of discretized
it a generic and efficient method. In the algorithm based on the momentum equation;
JPWSPC method, each individual branch is computed independ- Sf = friction slope;
ently by means of the junction-point water stage prediction and S0 = bed slope;
correction, so the algorithm is extremely easy to implement in t = time;
programming. Similar to the three-phase algorithms, the present x = longitudinal coordinate;
algorithm is also generally applicable and amenable to parallel Z = water stage;
processing. α = adjustable parameter; and
The proposed algorithm has been compared with the widely Δt = time step.
used three-phase algorithms in three examples. The results obtained
Subscripts
with the present algorithm are found to be nearly identical to those
obtained with three-phase algorithms, and compare well with the i = index for incoming channels;
observed data. Moreover, the present algorithm has been shown to j = index for computational point; and
reduce the storage requirement and the simulation time by avoiding o = index for outgoing channels.
establishing and solving the global branch equations. The extent Superscripts
of which the storage requirement and simulation time are reduced
k = index for corrector step.
depends on the size of the global matrix, and in turn, depends on the
number of branches in a network. This study suggests that the pro-
posed JPWSPC method is promising for modeling flows in channel
networks.
References
Choi, G. W., and Molinas, A. (1993). “Simultaneous solution algorithm for
channel networks modeling.” Water Resour. Res., 29(2), 321–328.
Acknowledgments Fread, D. L. (1973). “Technique for implicit dynamic routing in rivers with
tributaries.” Water Resour. Res., 9(4), 918–926.
The authors acknowledge the financial support from the National Islam, A., Raghuwanshi, N. S., Singh, R., and Sen, D. J. (2005).
Natural Science Foundation of China (Project No. 50779026), “Comparison of gradually varied flow computation algorithms for
and the National Basic Research and Development Program of open-channel network.” J. Irrig. Drain Eng., 131(5), 457–465.
Ji, Z. (1998). “General hydrodynamic model for sewer/channel network
China (Project No. 2006CB403304). The valuable suggestions systems.” J. Hydraul. Eng., 124(3), 307–315.
from Professor Binliang Lin from Cardiff University, UK, and Kesserwani, G., Ghostine, R., Vazquez, J., Mose, R., Abdallah, M., and
Dr. Dongfang Liang from Cambridge University, UK, are also Ghenaim, A. (2008). “Simulation of subcritical flow at open-channel
gratefully acknowledged by the first author. junction.” Adv. Water Resour., 31(2), 287–297.

JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011 / 773

J. Hydraul. Eng. 2011.137:766-774.


Lv, M. Y., Jiang, W., and Zhan, J. M. (2007). “Study of treatment for Sen, D. J., and Garg, N. K. (2002). “Efficient algorithm for gradually varied
linking conditions at junction points of river networks.” Yellow River, flows in channel networks.” J. Irrig. Drain Eng., 128(6), 351–357.
29(3), 31–32 (in Chinese). Tan, W. Y. (1998). Computational shallow water dynamics—Application
Nguyen, Q. K., and Kawano, H. (1995). “Simultaneous solution for of the finite volume method, Tsinghua University Press, Beijing, 22–34
flood routing in channel networks.” J. Hydraul. Eng., 121(10), (in Chinese).
744–750. Zhu, D. J., Chen, Y. C., and Wang, Z. Y. (2009). “A novel method for gradu-
Sen, D. J., and Garg, N. K. (1998). “Efficient solution technique for ally varied subcritical flow simulation in general channel networks.”
dendritic channel networks using FEM.” J. Hydraul. Eng., 124(8), 33rd IAHR Congress: Water Engineering for a Sustainable Environ-
831–839. ment, Vancouver, Canada, 6327–6335.
Downloaded from ascelibrary.org by New York University on 05/16/15. Copyright ASCE. For personal use only; all rights reserved.

774 / JOURNAL OF HYDRAULIC ENGINEERING © ASCE / JULY 2011

J. Hydraul. Eng. 2011.137:766-774.

You might also like