You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/286949879

MASCARET: A 1-D open-source software for flow hydrodynamic and water


quality in open channel networks

Article · January 2012

CITATIONS READS

33 1,669

4 authors, including:

Nicole Goutal Kamal el Kadi Abderrezzak


Électricité de France (EDF) Électricité de France (EDF)
70 PUBLICATIONS   937 CITATIONS    76 PUBLICATIONS   851 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Couplage des codes TELEMAC et Mascaret View project

Physical modelling of overtopping induced fluvial dike failure View project

All content following this page was uploaded by Kamal el Kadi Abderrezzak on 23 May 2018.

The user has requested enhancement of the downloaded file.


MASCARET : a 1-D open-source software for flow hydrodynamic and water
quality in open channel networks
N. Goutal, J.-M. Lacombe, F. Zaoui & K. El-Kadi-Abderrezzak
LNHE, EDF R&D, Chatou, France

MASCARET modeling framework is a set of numerical codes simulating one-dimensional (1-D) hydro-
environmental problems through a network of open channels. The governing equations underlying MAS-
CARET are the shallow water equations for unsteady flow propagation and the advection-dispersion equation
for water quality and contaminant transport. The hydraulic component accounts for floodplains and storage
areas, and flow at channel junction can be treated using a 2-D approach. Additionally, flood propagation over
dry beds (e.g. dam-break flows) and non-hydrostatic waves can be simulated. The user interface FUDAA-
MASCARET manages the input data, allocation of parameters, running of simulations and viewing outputs.
MASCARET can be easily compiled as a dynamic library, offering special interfaces to be used with three main
steps : Initialization, Run and Finalization of the calculation. With these features, MASCARET can be coupled
or integrated to other softwares without requiring significant efforts. Since July 2011, MASCARET is world-
wide distributed as an open-source code. In this paper, are presented various applications of this tool covering
both flow hydrodynamic and water quality. Model-data comparisons show the performance of the modeling
framework, and demonstrate the interest of this open-source software for the scientific community.

1 INTRODUCTION models are commercial or freeware.


The study of free surface water flow and water qual- MASCARET modeling framework is a set of
ity in open channels has many important applica- numerical codes simulating one-dimensional (1-D)
tions, such as flood prevention measures, becoming hydro-environmental problems through a network of
ever more ambitious, and accidental release of con- open channels. This modeling package has been
taminants in rivers. With the advances made in com- developed by Electricité de France-Recherche &
putational techniques over recent decades, numerical Développement (EDF-R&D) in collaboration with
models have become proactive tools for investigating Centre d’Etudes Techniques Maritimes et Fluviales
flow and water quality in networks of open channels. (CETMEF) over more than 25 years1 . MASCARET
The selection of one model for each specific problem is suitable for a wide range of engineering and envi-
is usually contingent on the knowledge about the sys- ronmental applications, from calculating simple back-
tem, the available measurements, and the specific ob- water profiles to modeling networks of open channels
jectives of the study. One-dimensional (1-D) models including floodplains and storage areas. Applications
require the least amount of field data, and the numer- include flood risk assessments and mapping, develop-
ical schemes used for solving the water and sediment ing management plans, flood alleviation scheme de-
governing equations are more stable and offer order signs, dam-break flows, and water pollution manage-
of magnitude gains in computational efficiency over ment. Since July 2011 the software package is world-
2-D and 3-D models (Sobey 2001). wide distributed as an open-source project for the ben-
Several one-dimensional (1-D) numerical models efit of students, engineers and researchers.
have been proposed in the literature (e.g., HEC- The remainder of the paper is organized as follows.
RAS (USACE 2002), MIKE1-D (DHI 2001), ISIS Section 2 details the main feature of the flow compo-
(HRW 2001) and every model has its own capabili- nent of MASCARET, including the various numerical
ties and limitations. The development of these mod- schemes employed for resolving the water governing
els has focused mainly on such aspects as numerical
schemes, flow-resistance relations and main channel- 1
http://innovation.edf.com/recherche-et-communaute-
floodplains flow exchange. Most of these numerical scientifique/logiciels/code-mascaret-41197.html

1
equations. In Section 3, the water quality and pollu- 2-D elements (Rissoan et al. 2002). In this last case,
tant transport module is described along with simu- MASCARET implements a simplified 1-D-2-D cou-
lated physical/chemical processes. Section 4 presents pling. The junction, see figure 1, is modelled by a 2-D
four applications of the model modelling framework, representation with 12 cells :
namely: estimation of uncertainty in flood predic-
tions for dyke design, data assimilation in the Durance • 6 triangles defining an hexagon, models the junc-
River (France), flow propagation due to cascade dam tion itself and can be defined as exchange cells;
break, transport of tritium in the Middle Loire River
(France). This is followed by conclusions drawn in • 6 quadrilaterals covering the 1-D domain for the
section 5. overlap coupling.

2 HYDRODYNAMIC MODELLING The junction hexagon has a geometry similar with


The well known shallow water equations can be the real one. It is built using the direction of reaches
solved with three different computational kernels. The and the width of valleys given by the user. This should
choice of the specific kernel depends on the case study lead to a good discharge repartition in the different
: steady or unsteady, with subcritical and/or supercrit- reaches.
ical flows.

2.1 SARAP Kernel


This kernel is for steady cases with subcritical, crit-
ical and supercritical conditions. The 1-D equations
are solved in one step with a finite difference scheme
where time derivatives are cancelled. A special treat-
ment is performed for the shock capturing with the so-
lution of a non-linear equation within the supercritical
region in order to determine the flow level at the point
of the supercritical-subcritical transition. The validity
of the shock position is checked with the principle of
conservation of momentum (Carlier 1998).
SARAP is extremely fast due to the need of only Figure 1: Overlap coupling for the junction
one iteration to solve the problem. It is recommended
in order to quickly find correct initial conditions for
the two others unsteady kernels, or when a large MASCARET kernel is suitable for every kind of
amount of calculations is necessary for a paramet- computational fluid dynamics. One will only take care
ric study (model calibration, optimisation, incertitude of time step restriction due to the Courant-Friedrichs-
study, etc.). Lewy condition if the explicit version of the time dis-
cretization is chosen.
2.2 REZO Kernel
REZO implements the classical finite difference 3 WATER QUALITY
Preissmann scheme for the unsteady subcritical so- 3.1 Governing equation and numerical scheme
lution of the shallow water equations (Preissmann The water quality module of MASCARET has been
1961). It is implicitly coupled with a code of water designed to compute time-dependent concentrations
storage areas in order to take into account floodplain of a series of constituents, which are primarily gov-
inundations. erned by the processes of advection, dispersion, and
This code is extremely robust and fast under sub- chemical/physical reactions.
critical conditions for operational use. The flow and advection-dispersion equations are
solved sequentially at each time step. First the flow
2.3 MASCARET Kernel routing module of MASCARET is called to provide
The last hydrodynamic computational kernel is based the time-dependent hydraulic parameters throughout
on a well-balanced finite volume Roe scheme (Roe the model domain, and then these variables are passed
1981) (Vazquez Cendon 1994) (Goutal and Maurel to the water quality module for the solute transport
2002). It has been designed to perform well with the simulation. The advection and dispersion terms are
simulation of dam break waves for EDF needs. It of- computed in two separate but consecutive steps (i.e.
fers special capabilities comparing to other codes like the well-known split-operator approach). The pure
the modelling of some non-hydrostatic waves (Bris- advection equation (i.e. the hyperbolic partial differ-
teau et al. 2011), or the modelling of junctions with ential equation) can be solved using two approaches:

2
• Method of characteristics, which can be applied • the computation code can be easily compiled as a
either to the conservative or non-conservative dynamic library for being used outside the GUI;
form of the advection term;
• the algorithm is divided into three main steps:
• Finite volume method based either on a first Initialisation, Run and Finalisation (IRF) of the
or second order upwind scheme. For the sec- calculation;
ond order scheme, a superbee limiter function
suppressing numerical oscillations has been in- • it has special interfaces written in C or Fortran
cluded. Because the numerical scheme is ex- language that define several entries in the code,
plicit, the time step is limited by the Courant- one entry per specific action;
Friedrichs-Levy (CFL) condition.
• it is also instantiable allowing multiple usage of
In the second step, an implicit finite difference the code in a single run.
scheme is applied to the pure dispersion equation.
This method has the advantage of being uncondi- With all these features, MASCARET is also ready
tionally stable. Regarding the boundary conditions, for an easiest coupling or integration with other soft-
the Dirichlet condition is used at the upstream end, wares while requiring no significant efforts. For the
whilst the Neumann condition (i.e. zero-gradient) is considered example, the SARAP kernel is used within
imposed at the downstream end. The longitudinal dif- the R2 software environment for statistical computing
fusion coefficient can be either calibrated using field (Bivand et al. 2008).
tracer studies or calculated using empirical formula- The case study is the Garonne river (France). We
tions. The water quality module incorporates disper- want to evaluate the impact of uncertainties on the wa-
sion coefficient formulas proposed by many authors, ter level corresponding to a 1000 years return period
see (El Kadi Abderrezzak et al. 2012). discharge. The uncertainties taken into account con-
cern :
3.2 Simulated water quality processes
The water quality module can be applied to simulate • the upstream discharge;
various processes in open channels, such as migra-
tion of nuclear power plants releases, water tempera- • the friction factors for the river main bed and the
ture evolution, oxygen demand, growth rate of a pop- floodplains;
ulation of phytoplankton, and radionuclides decay in-
• and on the river’s topography with a stochastic
cluding interaction with suspended material. The fate
process based on a kriging method (Krige 1951)
and transport of water quality constituents involving
a variety of chemical and physical processes (interac- Figure 2 shows some computed results on the water
tion between pollutants) is mathematically expressed levels along the river considering two and three kind
in the source terms of the advection-dispersion equa- of uncertainties.
tion. The following processes have been implemented These results, on the probability of water levels
: micro-pollutants,dissolved oxygen, phytoplankton along the river, are important in order to study the
biomass, water temperature and eutrophication. role of dykes for the protection of EDF power facil-
ities against extreme flood. Due to its low computa-
4 APPLICATIONS tion cost, the SARAP kernel is particularly convenient
Four recent applications are presented here after for such probabilistic studies that can require several
demonstrating some of the capabilities of MAS- thousands of computations.
CARET.
4.2 Data assimilation with REZO
4.1 Uncertainties with SARAP The assimilation of water level observations is done in
The first application concerns the uncertainty affect- order to improve the flood forecasting with the REZO
ing flood hazard assessments. Interest for this prob- kernel in real time conditions. The case study is a part
lem is increasing within the scientific community and of the Durance river (France) concerning hydroelec-
among decision makers (Bernardara et al. 2010). The tricity.
treatment of uncertainty both involves the proper def- Main uncertainties of the problem come from hy-
inition of a deterministic model and of the probabilis- drological conditions on the Durance upstream flow
tic uncertainties on its inputs (De Rocquigny et al. and lateral inflows of river #1 and #2, see figure 3.
2008). With such unknowns, it is not possible to model cor-
MASCARET is well adapted for such parametric rectly the water level at the dam as a function of time.
studies since it offers a modern interoperable inter-
2
face: http://www.r-project.org

3
431.6
Observation
Free run
Assim run
431.5

Water level at the dam (m)


431.4

431.3

431.2

431.1

ASSIMILATION FORECAST

431
0 5 10 15 20 25 30 35 40 45 50
Time (hours)

Figure 4: Water level forecasting with and without


data assimilation

Figure 2: 90% confidence interval analysis Table 1: Errors between measurements and computa-
tions on the water level at the dam
Error (cm) Free run Assim run
Mean 12.5 2.2
Max 21.6 7.8

Starting from an initial estimation of the control vec-


tor xb = (a0 , b0 , c0 )T , xa is given by the following
formula which is a Best Linear Unbiased Estimate
(BLUE):

xa = xb + (B −1 + H T R−1 H)−1 H T R−1 d (2)


where B is the background error covariance matrix,
Figure 3: Hydropower example on Durance (France)
R the observation covariance matrix, y0 the observa-
tion and d the innovation vector :
The data assimilation process will use all available d = y0 − H(xb ) (3)
measurements (water level at the dam and discharge at
the plant in a 24 hours period) to correct the upstream Figure 4 clearly shows the interest of the filter-
flow. This would lead to a significant improvement ing technique (assimilation) in order to find optimal
in the water level computation in both analysis and parameters that will modify the upstream discharge
forecast modes. (forecast). Compared with no corrections made on the
The stationary Kalman filter is chosen as the inflow, the data assimilation greatly improve the fore-
data assimilation technique to adjust the upstream cast of water level at the dam as indicated in table 1.
flow Q(t). The flow is corrected through a three- Due to the robustness and fastness of the Preiss-
parameters linear transformation (Ricci et al. 2011) mann scheme, the Rezo kernel is well suited for oper-
: ational problems. Once again, the code interoperabil-
ity makes its integration easy.
Q̂(t) = a × Q(t − b) + c (1)
4.3 Dam break simulation with MASCARET
The control vector xa = (a, b, c)T is calculated In order to illustrate the ability of the finite volume
through the comparison of simulations and observa- scheme to deal with dam-break shock wave, a case
tions. An observation operator H is defined to find an study concerning four dams in a real valley is pre-
equivalence relation between the model results and sented, see figure 5. The first dam upstream collapses
the measured values over the chosen time window. instantaneously at the beginning of the simulation.

4
The dam #2 collapses when the wave arrives while 4.4 Tritium transport with TRACER
the two others #3 and #4 downstream are supposed to The modelling system is applied for simulating the
resist all along the simulation. transport of tritium in the Middle Loire River (France)
(Fig. 1), 350km long, over the period 01/01/1999-
31/12/1999. Four main tributary streams feed the
450
river: Vienne, Indre, Cher and Maine. Four Nuclear
Main Dam
Junction #1
Water level
Bottom Power Plants (NPP) are located along the Middle
400
Loire River and one NPP is located along the Vienne
350 Dam #2
Junction #2
River. These NPP generate low-activity radioactive
300
liquid waste, including tritium, which is released into
Dam #3 the river in a controlled way.
Level (m)

250
The Middle Loire River is modelled as one continu-
200 Dam #4
Junction #3
ous reach with the tributaries Vienne, Indre, Cher and
150
Maine as inflows.
100

50

0
0 20 40 60 80 100
Distance along the main valley (km)

Figure 5: Initial water level conditions

This case includes several reaches where the mod-


elling of 2-D junctions is necessary.
The 1-D mesh has a size of about 2650 nodes. The
time step of the explicit scheme is variable depending
on a given Courant number smaller than 1. With these
settings, the results shown in figure 6 can be obtained
in approximately 1 mn on a standard workstation.

450
Water level at 31974 s
Water level at 15678 s
Figure 7: Location map showing the study reach
400 Water level at 5629 s
Water level at 1573 s
Bottom
350

300
Three hundreds and sixty-eight cross-sections
spaced at intervals of about 1 km are used to describe
Level (m)

250
the river geometry. The downstream boundary condi-
200 tion at MontJean sur Loire is a water stage-flow rate
150
curve. For the solute transport module, the tritium dis-
charge recorded at Belleville NPP during the period
100
01/01/1999 to 12/31/1999 is specified as the upstream
50 boundary. The tritium releases from Dampierre, Saint
0
Laurent and Chinon NPP on the same period are
0 20 40 60 80 100
Distance along the main valley (km)
introduced as lateral source terms. The tritium dis-
Figure 6: Water levels at different simulation times charge due to release from Civaux NPP is estimated
by applying the MASCARET System to the Vienne
River between Civaux NPP and Vienne-Loire junc-
tion. Monitoring of tritium concentration was per-
The validity of the MASCARET results is not formed during 1999 at Anger city.
shown for this particular case. However, several stud- Calibration of the hydraulic module (i.e. Stricklers
ies have already demonstrated the skill of the code for coefficients) is carried out using measured water level
such problems. See for example (Goutal and Maurel measurements collected along the Middle Loire River
2002) or (Malleron et al. 2011) for a comparison with at low, medium and high flow discharges. The Strick-
analytical solutions, measurements and other codes. ler coefficient is set at 30m−1/3 .s−1 . The longitudinal

5
dispersion coefficient is computed using the formula REFERENCES
proposed by (Seo and Cheong 1998). The first order Bernardara, P. et al. (2010). Uncertainty analysis in
finite volume method is used. The numerical run is flood hazard assessment: hydrological and hy-
carried out using a time step 10s and the space step is draulic calibration. Canadian Journal of Civil
200m. The courant number is set at 0.5. Figure 8 com- Engineering 37(7), 968–979.
pares the numerical predictions of the tritium concen-
tration against measurements at Anger city for two Bivand, R. S. et al. (2008). Applied spatial data
months. Generally, the model-data fit is reasonably analysis with R. Springer.
well. The model reproduces the magnitude and gen- Bristeau, M.-O. et al. (2011, August). Numerical
eral temporal evolution of the tritium concentration simulations of a non-hydrostatic shallow water
at Angers. The mean relative error is approximately model. Computers & Fluids 47(1), 51–64.
43%.
Carlier, M. (1998). Hydraulique générale et ap-
pliquée. Eyrolles.
De Rocquigny, E. et al. (2008). Uncertainties in
35 industrial practice — A guide to quantitative
Measurements
Predictions uncertainty management. Wiley-Blackwell.
30
DHI (2001). MIKE 11 Reference Manual. Ap-
25 pendix A Scientific background. Danish Hy-
draulic Institute.
Concentration (Bq/l)

20
El Kadi Abderrezzak, K. et al. (2012). Applica-
15 bility of longitudinal dispersion coefficient for-
mulas in 1-d numerical modeling of solute
10
transport in open channels. In River Flow 2012,
5
San Jose, Costa Rica.
Goutal, N. and F. Maurel (2002). A finite volume
0
26/06 03/07 10/07 17/07 24/07 31/07 07/08 14/08 21/08 28/08 04/09 solver for 1d shallow water equations applied
Date
to an actual river. International Journal for Nu-
Figure 8: Comparison between calculated and mea- merical Methods in Fluids 38, 1–19.
sured tritium concentration at Angers city
HRW (2001). ISIS 2001-User Manual. HR
Wallingford.
Krige, D. (1951). A statistical approach to some
basic mine valuation problem on the witwater-
srand. Journal of the Chemical, Metallurgical
5 CONCLUSIONS
and Mining Society of South Africa 52(6), 119–
The open-source MASCARET software is a system 139.
dedicated to the solution of 1-D shallow-water equa-
Malleron, N. et al. (2011). On the use of a high-
tions with transport of passive pollutants. This system
performance framework for efficient model
benefits from 25 years of research and development at
coupling in hydroinformatics. Environmental
EDF-R&D in collaboration with the CETMEF (tech-
Modelling & Software 26(12), 1747–1758.
nical branch of the Public works french department).
As detailed previously, the more appropriate scheme Preissmann, A. (1961). Propagation des intumes-
has been integrated for each flow regime: Preissmann cences dans les canaux et rivières. In 1er
scheme for subcritical flow and finite volume scheme congrès de l’Association Française de Calcul,
for unsteady supercitical flow. Moreover, a strong val- Grenoble, France.
idation has shown the robustness and relevance for
Ricci, S. et al. (2011). Correction of upstream flow
dealing with complex industrial studies. The last de-
and hydraulic state with data assimilation in
velopments highlights the capability of the system to
the context of flood forecasting. Hydrology and
evolve towards new applications : uncertainties, data
Earth System Sciences 15(11), 3555–3575.
assimilation which are of main importance for real
time studies and flood hazard assessments. In a near Rissoan, C. et al. (2002). 1d hydraulic simulation
future, the MASCARET system will offer the capa- of a dam break wave on the rhone river. In 5th
bility to deal with flood plains by coupling between International Conference on Hydro-Science &
1-D and 2-D models. Engineering, Warsaw, Poland.

6
Roe, P. (1981, October). Approximate rie-
mann solvers, parameter vectors and dif-
ference schemes. Journal of Computational
Physics 43(2), 357–372.
Seo, I. W. and T. S. Cheong (1998). Predict-
ing longitudinal dispersion coefficient in nat-
ural stream. Journal of Hydraulic Engineer-
ing 124(1), 25–32.
Sobey, R. (2001). Evaluation of numerical models
of flood and tide propagation in channels. Jour-
nal of Hydraulic Engineering 127(10), 805–
824.
USACE (2002). HEC-RAS Hydraulic Reference
Manual. U.S. Army Corps of Engineers.
Vazquez Cendon, M. (1994). Estudio de esquemas
descentrados para su aplicaciòn a las leyes
de conservaciòn hiperbòlicas con tèrminos
fuente. Ph. D. thesis, Universidad Santiago de
Compostella.

7
View publication stats

You might also like