Second order discontinuous Galerkin scheme for compound natural channels with movable bed. Applications for the computation of rating curves

https://doi.org/10.1016/j.advwatres.2015.06.007Get rights and content

Highlights

  • Second order scheme (or higher) with stiff source term treatment.

  • Validation with field measurements and test cases shows excellent agreement.

  • Application to rating curves computation shows that in some cases, it can be necessary to use schemes with high order accuracy and stiff source terms treatment, unless a drastic grid refinement is perfomed.

Abstract

A new higher order 1D numerical scheme for the propagation of flood waves in compound channels with a movable bed is presented. The model equations are solved by means of an ADER Discontinuous Galerkin explicit scheme which can, in principle, reach any order of space–time accuracy. The higher order nature of the scheme allows the numerical coupling between flux and source terms appearing in the governing equations and, importantly, to handle moderately stiff and stiff source terms. Stiff source terms arise in the case of abrupt changes of river geometry such as in the case of hydraulic structures like bridges and weirs. Hydraulic interpretation of these conditions with 1D numerical modelling requires particular attention; for instance, a 1st order scheme might either lead to inaccurate solutions or impossibility to simulate these complex conditions. Validation is carried out with several test cases with the aim to check the scheme capability to deal with abrupt geometric changes and to capture the direction and celerity of propagation of bed and water surface disturbances. Validation is done also in a real case by using stage-discharge field measurements in the Ombrone river (Tuscany). The proposed scheme is further employed for the computation of flow rating curves in cross-sections just upstream of an abrupt narrowing, considering both fixed and movable bed conditions and different ratios of contraction for cross-section width. This problem is of particular relevance as, in common engineering practice, rating curves are derived from stage-measuring gauges installed on bridges with flow conditions that are likely to be influenced by local width narrowing. Results show that a higher order scheme is needed in order to deal with stiff source terms and reproduce realistic flow rating curves, unless a strong refinement of the computational grid is performed. This capability appears to be crucial for the computation of rating curves on coarse grids as it allows the modeling of abrupt contractions and jumps in bed bottom elevations, which often occur near cross-sections where stage measuring gauges are installed.

Introduction

In the present work, a higher order 1D Discontinuous Galerkin numerical scheme for the propagation of flood hydrograph over a compound channel with movable bed is presented. The scheme is designed in order to model the complex geometry variations occurring in natural channels, which are represented by source terms in the governing equations.

The model equations are represented by the coupled system of balance laws (SBL) formed by the 1D Shallow Water equations and the Exner sediments continuity equation, written for a compound natural channel. A bedload transport equation is used as closure condition for sediments dynamics.

Source terms appearing in the shallow water equations for natural channels are related to channel width, slope and friction. The presence of abrupt geometry variations, which, for instance, is common in cross-sections near stage measuring gauges in rivers, may lead to stiff source terms. The characteristic speed associated to such terms is much larger than the one associated to the flux term. If an explicit numerical scheme does not feature some appropriate treatment for stiff source terms, the numerical solution can be inaccurate and asymptotically inconsistent (see Dumbser et al. [3]) with wrong advection speed estimates or even instabilities on coarse grids.

In order to solve an SBL, a widely used method is the so-called source term splitting (see Toro [2]). It consists in splitting the SBL onto two sub-problems to be solved in sequence: a homogeneous problem, in which the source term in the original SBL is not taken into account, and an ODE containing only the source term and the time derivatives of state from the original SBL. The numerical method to solve the ODE can feature some appropriate treatment for stiff source terms if needed. This method is however not optimal when a scheme which is also capable of effectively reproducing stationary solutions is sought, as splitting methods may lead to oscillations near steady states.

The main reason that motivated the present work is the one of devising a scheme capable of treating moderately stiff and stiff source terms, while preserving other desired properties such as the well-balanced property, which allows steady states to be reproduced with a certain accuracy.

The scheme has been developed within the framework of the ADER-Discontinuous Galerkin (DG) methods, as proposed by Dumbser et al. [1]. As it will be clear from Section 3, it is necessary to employ a higher order scheme (a scheme with order of accuracy higher than the 1st), in order to preserve the coupling between the flux and the source term. Also, the nature of the model equations requires the use of a path-conservative formalism in order to treat liquid and solid dynamics in a coupled way.

In the literature, 1D shallow water numerical models have been recently proposed both using 1st order (e.g. Catella et al. [4], Audusse et al. [5]) and higher order (e.g. Caleffi et al. [6], Siviglia et al. [7] and Canestrelli et al. [8]) schemes.

It would still be possible to use a shallow water scheme not featuring a special treatment for source terms in order to simulate abrupt geometry changes in natural rivers, by refining the computational grid. However, the use of an ADER-DG strategy in such cases proves to be effective with no need for grid refinement and the resulting scheme is very stable and can be easily extended to even higher orders of accuracy.

The use of an ADER-DG strategy in the case of shallow water equations for natural channels has never been investigated in the scientific literature, to the knowledge of the authors.

The proposed numerical scheme can in principle reach any order of space–time accuracy. For the applications in the present work, the 2nd order accuracy has been found sufficient and represents a good compromise between accuracy and calculation time. This is partly due to information on river geometry being often low in accuracy.

The model is validated against several benchmarks: (i) water at rest in a non-prismatic channel, (ii) dam break problem with a moving strong shock, (iii) steady flow in a Venturi-type flume, (iv) subcritical flow in an irregular channel, and (v) propagation of a sediment hump near critical conditions. Moreover, the model is further validated against field measurements of water level-flow discharge during a flood in the Ombrone Pistoiese river in Tuscany (Italy).

The implications of modelling moderately stiff or stiff source terms in the case of flow rating curves are discussed. Importantly, conditions of non-uniformity are found in cross-sections just upstream of bridges where, typically, water-level gauges are installed for flow monitoring by making use of stage-discharge rating curves. Abrupt geometric changes, due to a rapid cross-section narrowing, can deeply affect the flow, leading for instance to backwater effects and transition to a supercritical state. As a result, rating curves may considerably deviate from the classical power law function assuming non-trivial shapes. Moreover, flow unsteadiness can produce a hysteretic behavior (see Schmidt and Yen [9] and Francalanci et al. [10]). In these conditions, reliable rating curves need to be developed by coupling filed measurements with 1D hydraulic numerical modelling.

The model is applied to the case of abrupt geometry changes, where flow can be more effectively reproduced by using a higher order scheme featuring some kind of treatment for stiff source terms; a 1st order scheme would lead to erroneous results, unless a drastic grid refinement was performed. Flow rating curves are derived in a schematic channel subject to a flood wave forcing in the cases of movable and fixed bed with a local width constriction (such as in the case of bridge piers). Various numerical tests are carried out considering different degrees of channel narrowing for given boundary conditions (i.e. input hydrograph, upstream bed level and downstream water level). Results are shown considering the present higher order scheme and its 1st order version. It appears that in these geometrical conditions, rating curves are better reproduced only when a higher order numerical scheme is considered. Importantly, a 1st order scheme can lead to wrong/inaccurate results although it does not become unstable.

The numerical scheme is introduced in Section 3, and its validation and application to the computation of rating curves are shown in Section 4. A discussion of the results and the conclusions are shown in Section 5.

Section snippets

Model equations

The model features the shallow water equations for a natural channel coupled with the Exner equation describing sediments mass conservation. A capacitive approach is used in the present work: the solid flow rate in the channel is assumed to be coincident with the solid flow rate as predicted by bedload transport formulas.

The three governing equations can be written as an SBL having the following form: Ut+dFdx=BUx+Sωxwhere (x, t) are the space–time coordinates, U(x, t) is the state vector,

Numerical scheme

In order to solve a hyperbolic SBL in the form of Eq. (1), the main idea is the one of writing it as an homogenous nonconservative system. The following trivial equation is therefore added to the system: ωt=0and an extended state vector is defined: U˜=(Uω)If URD, then U˜RD+1.

The following homogeneous hyperbolic system of equations is considered: U˜t+A˜(U˜)U˜x=0If Eq. (19) represents a nonconservative system (NCS), then the A˜(U˜)·U˜/x term cannot be written as the gradient of some

Validation of the numerical model

Validation of the present model is carried-out by performing numerical simulations of (i) water at rest in a non-prismatic channel, (ii) dam break problem with a moving strong shock, (iii) steady flow in a Venturi-type flume, (iv) subcritical flow in an irregular channel, (v) propagation of a sediment hump near critical conditions, and a real flood wave in the Ombrone Pistoiese river in Tuscany (Italy).

Discussion and conclusions

In the present work a 1D model based on a higher order numerical scheme for simulating flood wave propagation on a natural channel with a movable bed has been presented. An important feature of the scheme is represented by its ability at treating the moderately stiff or stiff source terms that may appear in the governing equations due to the occurrence of abrupt changes in the channel geometry.

After the validation of the scheme, its application to the computation of flow rating curves in a

Acknowledgments

The authors are grateful to the Tuscany regional authority ”Centro Funzionale di Monitoraggio Meteo Idrologico - Idraulico, Servizio Idrologico Regionale” for having contributed to the present research by providing the Ombrone river data used in the present work. Part of such data is not of public domain and was provided to the University of Florence under specific and broader agreements. The Editor Dr. Andrea Rinaldo, the Guest Editor Dr. Annunziato Siviglia and two anonymous referees are

References (31)

  • Garcia-NavarroP. et al.

    On numerical treatment of the source terms in the shallow water equations

    Comput Fluids

    (2000)
  • ToroE.F.

    Riemann solvers and numerical methods for fluid dynamics

    (2009)
  • CatellaM. et al.

    Conservative scheme for numerical modelling of flow in natural geoemtry

    J Hydraul Eng

    (2008)
  • AudusseE. et al.

    A fast and stable well-balanced scheme with hydrostatic reconstruction for shallow water flows

    SIAM J Sci Comput

    (2004)
  • SivigliaA. et al.

    Quasi-conservative formulation of the one-dimensional Saint-Venantexner model

    J Hydraul Eng

    (2008)
  • Cited by (7)

    • (Multi)wavelets increase both accuracy and efficiency of standard Godunov-type hydrodynamic models: Robust 2D approaches

      2020, Advances in Water Resources
      Citation Excerpt :

      The second-order discontinuous Galerkin (DG2) discretisation extends the finite volume principle by representing flow variables and terrain data as piecewise-planar fields (Gourgue et al., 2009; Trahan and Dawson, 2012; Dawson et al., 2013; Kesserwani and Wang, 2014; Vater et al., 2019). Previous studies identified practical benefits to adopting local piecewise-planar approximations in 2D hydrodynamic models to improve the capture of free-surface flow spillage over steep topography (Buttinger-Kreuzhuber et al., 2019; Sanders and Schubert, 2019), and to improve velocity predictions on coarse meshes leading to better rating curve predictions (Kesserwani, 2013; Kesserwani and Wang, 2014; Minatti et al., 2016). While DG2 solvers can improve predictions on coarse meshes, they suffer from excessive runtime cost ( > 3 times the cost of FV1) (Kesserwani, 2013).

    • (Multi)wavelets increase both accuracy and efficiency of standard Godunov-type hydrodynamic models

      2019, Advances in Water Resources
      Citation Excerpt :

      The Discontinuous Galerkin (DG) method extends the foundation of the FV1 approach by shaping local piecewise-polynomial solutions from a discrete (element-wise) formulation of the conservative model equation(s). DG methods significantly reduce numerical diffusion even on very coarse meshes (e.g. at a grid resolution exceeding 10 m2) and have excellent conservation properties (Minatti et al., 2016, Kesserwani, 2013, Kesserwani and Wang, 2014, Bokhove, 2005). Compared to a FV counterpart, the DG method has a much larger cost per mesh element in terms of data storage and computing time, and such cost is proportional to the desired order-of-accuracy.

    • Discontinuous Galerkin formulation for 2D hydrodynamic modelling: Trade-offs between theoretical complexity and practical convenience

      2018, Computer Methods in Applied Mechanics and Engineering
      Citation Excerpt :

      Compared to other popular alternatives, such as the continuous Galerkin and the finite volume methods, the DG method entails significantly larger computational costs in terms of degrees of freedom (or coefficients) per mesh element. Perhaps, a fairer comparison is to fix a prescribed level of accuracy and/or of conservativeness vs. runtime cost and mesh-size required, where DG methods outrival [37–43]. With DG methods, the number of coefficients needed are proportional to the desired order-of-accuracy, the spatial dimensionality involved and the number of equations forming the physical model in question, all of which define the complexity, both theoretical and computational, involved in a DG formulation.

    • Numerical modelling of river morphodynamics: Latest developments and remaining challenges

      2016, Advances in Water Resources
      Citation Excerpt :

      Almost all the papers contained in this SI simulate reach-scale morphodynamic processes using homogeneous sediment. The most widely modelling approach used is two-dimensional horizontal (2DH) [3,4,7,8,22] and in two cases one-dimensional [10,20]. Three papers [4,9,20] perform simulations considering sediment mixtures.

    View all citing articles on Scopus
    View full text