A New Well-Balanced Reconstruction Technique for the Numerical Simulation of Shallow Water Flows with Wet/Dry Fronts and Complex Topography

: This study develops a new well-balanced scheme for the one-dimensional shallow water system over irregular bed topographies with wet/dry fronts, in a Godunov-type ﬁnite volume framework. A new reconstruction technique that includes ﬂooded cells and partially ﬂooded cells and preserves the non-negative values of water depth is proposed. For the wet cell, a modiﬁed revised surface gradient method is presented assuming that the bed topography is irregular in the cell. For the case that the cell is partially ﬂooded, this paper proposes a special reconstruction of ﬂow variables that assumes that the bottom function is linear in the cell. The Harten–Lax–van Leer approximate Riemann solver is applied to evaluate the ﬂux at cell faces. The numerical results show good agreement with analytical solutions to a set of test cases and experimental results.


Introduction
Non-linear shallow water equations, which can be derived by integrating the Euler equations in depth, have wide applications to surface flows, such as water flows of rivers, flood plains, practical dam failures, coastal regions and tides. Dam-breaks and floods may have enormous economic and human costs. Numerical solutions to the shallow water equations have been important in simulating floods and dam-break flows [1][2][3][4].
The present paper focuses on one-dimensional shallow water equation [5]: (1) where h(x, t) is the water depth, u(x, t) is the depth-averaged velocity, g is the gravitational constant, and B(x) is bottom topography. Subscripts denote partial derivatives. The friction system is ignored. The shallow water system can be written in vector form as ∂U ∂t Water 2018, 10 where U is the vector of conservable variables, F is the flux vector function, and S is the source vector, expressed as: A large number of numerical schemes based on Equation (1) have been developed [6][7][8][9]. A challenge in developing a successful numerical scheme is to preserve the steady states at discrete levels. A scheme is called well-balanced if it is capable of balancing the flux and source terms exactly, which gives the motionless water surface in a lake [10,11]: where z(x, t) is the water level. Many well-balanced schemes have been presented [12][13][14]. Another difficult issue is the presence of dry areas in the computational domain. In numerical oscillations, the water depth h may become negative near wet/dry fronts. The overall calculations then break down. Many numerical techniques that preserve the positive value of the water depth near wet/dry fronts have been put forward [15,16]. The surface gradient method (SGM) was proposed by Zhou et al. [17]. Compared with conventional data reconstruction methods based on the water depth, the SGM uses the water level as the basis of reconstruction. The advantage of the SGM used with source terms is that the SGM is generally suited to both steady and unsteady shallow water problems while the homogeneous terms are no more complicated than those of conventional methods. However, the SGM cannot be applied to shallow water flows over a vertical step and wet/dry fronts.
A modified surface gradient method (MSGM) was proposed using a fictitious cell at the vertical step by Zhou et al. [18]. Dae-Hong et al. [19] presented a revised surface gradient method (RSGM) assuming that the bottom topography in a grid is irregular rather than linear as in the SGM. However, the MSGM and RSGM cannot solve the wet/dry front with partially flooded cells. Kurganov and Levy [15] described a reconstruction adjustment for the partially flooded cell when the water depth becomes negative at integration points. Their correction ensures that the value of the water depth remains non-negative. However, the method is not well-balanced for very small water depth at flooded cells. Bollermann et al. [6] improved Kurganov's method and proposed a well-balanced reconstruction method for the shallow water equations with wet/dry fronts.
In this study, we focus on the Godunov-type scheme and a reconstruction technique based on the values defined at the cell center. The Harten-Lax-van Leer (HLL) approximate Riemann solver is used to evaluate the flux terms at the cell interface. We present a robust, well-balanced and non-negative scheme that assumes the bottom function B(x) is continuous and linear only in partially flooded cells and the topography is irregular in flooded cells. The scheme proposes a modified RSGM and develops a new technique of reconstruction for partially flooded cells near the wet/dry fronts.
The remainder of the paper is organized as follows. The SGM and RSGM are reviewed briefly in Section 2. Section 3 describes the reconstruction method at the wet cell and the partially flooded cell. A Godunov-type method is introduced in Section 4. Several theoretical benchmark tests and laboratory dam-break problems are used to verify the performance of the proposed scheme in Section 5. Finally, conclusions are briefly presented in Section 6.

Review of the SGM and RSGM
In this section, the SGM and RSGM that are suitable for flooded cells are briefly reviewed. A uniform grid with finite volume I i : [x i−1/2 , x i+1/2 ] of length ∆x is used. The cell-averaged value U is expressed by U(t) i = 1 ∆x I i U(x, t)dx, where U = (h, hu) T . Apparently, in the flooded cell, U i (x, t) = h i , h i u i T is the same as U i (x, t) = (h i , h i u i ) T .

SGM
In high-order accurate Godunov-type schemes, the values at the cell interface are computed by reconstruction based on data at the cell center. The monotone upstream-centered scheme for conservation laws (MUSCL) approach, which has second-order accuracy in space resulting from piecewise linear reconstruction from values defined at cell center, is widely applied in the framework of Godunov-type schemes [3,4].
For the continuity Equation (1), the water level z at the cell center is chosen to be reconstructed in the SGM. The water level z within a cell can be expressed as z(x, t) = z i + (x − x i )δz i (5) and the water depth h within a cell is where δz i is the gradient of z, calculated as in which G is a slope limiter that avoids the generation of spurious oscillations in the process of reconstruction at the cell interface. Researchers have proposed several slope limiters that have been demonstrated to have good performance in numerical experiments [20], such as the minmod limiter, van Leer limiter, and superbee limiter. Here, the minmod limiter, which is expressed as is selected. The values of z at the interfaces in a cell are and the values of water depth h at interfaces in a cell are The bed topography is defined at the cell interface and is assumed to have a piecewise profile within a cell in the SGM. The bed topography at the cell center is then expressed exactly as For the momentum equation in Equation (1), the conserved value hu is selected to be reconstructed and the solution can be formulated in a similar manner to the formulation for z. For the stationary case u = 0 and z = constant, which is the "lake at rest" steady state, the momentum flux in Equations (1) is discretized using a Godunov-type scheme as Under the condition z = constant, the value of δz i becomes zero and Equation (10) becomes According to Equation (11), Equation (13) can be rewritten as Equation (12) At the same time, the source term, which is the bed slope term of the momentum Equation (1), is discretized as , because the water level z is constant throughout the computational domain.
From Equations (15) and (16), it is concluded that ∂F/∂x = S, which means the SGM satisfies the exact C-property and can be called a well-balanced scheme.

RSGM
As presented in Equations (11) and (14), the SGM assumes that the arithmetic mean of the water depth at the two interfaces of a cell is the same as the water depth at the cell center of a cell. However, in the case of the RSGM, Dae-Hong Kim et al. pointed out that the assumption is valid only when the bottom slope of the cell is constant [19]. For irregular bathymetry, the SGM may produce unwanted computational results.
In the RSGM, Equation (14) is rewritten as because Equation (11) is not satisfied in general cases. Furthermore, Equation (15) becomes To balance the flux term and source term in the stationary case, F in Equation (18) should be the same as the source term when it is discretized using a central discretization scheme in a cell: From Equation (19), a new arithmetic formula is obtained: Equation (20) is similar to Equation (14) but has different implications; i.e., the water depth h in the source term should be discretized as (h i+ 1 2 + h i− 1 2 )/2, rather than h i . Finally, the source term is discretized as The reconstruction for the conserved value of hu is the same as the solution in the SGM. Further details have been given by Zhou et al. [17] and Dae-Hong Kim et al. [19].

New Reconstruction Technique
We assume that all the computed values h i are positive at a certain time step.

Reconstruction for the Flooded Cell
The flooded cell should satisfy For the flooded cell, the reconstruction method of the RSGM, which assumes that the bottom topography is irregular, is applied. In Section 2, we introduced the SGM and RSGM briefly. After the reconstruction, the values of h − i+1/2 or h + i−1/2 may be negative (Figure 1b), even though the average water depth h i is positive (Figure 1a). The RSGM is therefore modified as follows.
we reset the water level at the two sides of the cell (Figure 1c): The reconstruction for the conserved value of hu is the same as the solution in the SGM.
Further details have been given by Zhou et al. [17] and Dae-Hong Kim et al. [19].

New Reconstruction Technique
We assume that all the computed values i h are positive at a certain time step.

Reconstruction for the Flooded Cell
The flooded cell should satisfy For the flooded cell, the reconstruction method of the RSGM, which assumes that the bottom topography is irregular, is applied. In Section 2, we introduced the SGM and RSGM briefly. After the reconstruction, the values of  (2) If we reset the water level at the two sides of the cell ( Figure 1c): (3) If

Reconstruction at the Partially Flooded Cell
We assume the bottom topography function is linear in the partially flooded cell. The partially flooded cell, which has a dry area in the cell (Figure 2), may produce negative values of the water depth at cell interfaces in the reconstruction stage introduced in Section 2. In Figure 2, x w is the separation point between wet and dry areas. Here, we note that U = U in the partially flooded cell. The water level function z(x) is defined as Water 2018, 10, x FOR PEER REVIEW 6 of 20  1/ 2 i z − + also needs to be corrected considering the conservation of the total amount of water.

Reconstruction at the Partially Flooded Cell
We assume the bottom topography function is linear in the partially flooded cell. The partially flooded cell, which has a dry area in the cell (Figure 2), may produce negative values of the water depth at cell interfaces in the reconstruction stage introduced in Section 2. In Figure 2, w x is the separation point between wet and dry areas. Here, we note that U U ≠ in the partially flooded cell.
The water level function ( ) where ( ) ( ) The formula for the free surface is The total amount of water given by ∆x · h i is equal to the volume enclosed between the water level line and the bathymetry. Then, if cell i is fully flooded, it satisfies the condition where |(B x ) i | is the bed slope in cell i. The separation point x w can be derived from mass conservation: The formula for the free surface is For the partially flooded cell, we modify the water depth values h + i−1/2 and h − i+1/2 at cell interfaces obtained in the reconstruction step. First, we define that the water depth value z − i−1/2 reconstructed in the next flooded cell equals z + i−1/2 [6]. Immediately, it follows that h − i−1/2 = h + i−1/2 . Two cases are considered.
From Equation (29), we obtain The water level , and the well-balanced reconstruction in partially flooded cell i is complete. We replace the water level and water depth values at the + 1 2 i interface by those at the w x interface: where w x is obtained using Equation (27). To keep the well-balanced property, the value +   (Figure 4a), the reconstruction step is neglected in cell i. We replace the water level and water depth values at the i + 1/2 interface by those at the x w interface: where x w is obtained using Equation (27). To keep the well-balanced property, the value z + i+1/2 is subtracted from ∆z ( Figure 4b): Here we make two statements about the new reconstruction technique: (1) If the cell is fully flooded satisfying Equation (22), the modified RSGM presented in Section 3.1 is applied; (2) when the cell does not satisfy the conditions of Equation (22) or Equation (25), the cell is partially flooded. Two practical cases are considered in Section 3.2.

Implementation Using a Godunov-Type Method
The reconstruction method proposed in this paper can be incorporated into any Godunov-type method that requires data reconstruction. Here, the MUSCL method having second-order accuracy in space is used. The updating conservative formula is F are fluxes at cell interfaces. The HLL approximate Riemann solver is determined by Here we make two statements about the new reconstruction technique: (1) If the cell is fully flooded satisfying Equation (22), the modified RSGM presented in Section 3.1 is applied; (2) when the cell does not satisfy the conditions of Equation (22) or Equation (25), the cell is partially flooded. Two practical cases are considered in Section 3.2.

Implementation Using a Godunov-Type Method
The reconstruction method proposed in this paper can be incorporated into any Godunov-type method that requires data reconstruction. Here, the MUSCL method having second-order accuracy in space is used. The updating conservative formula is where F HLL i+1/2 and F HLL i−1/2 are fluxes at cell interfaces. The HLL approximate Riemann solver is determined by and wave speeds S are given by S i denotes the source term. If the cell is flooded, according to Equation (21) applied in the RSGM, the source term is discretized as If the cell is partially flooded, in which the bottom topography function is assumed to be linear, according to Equation (16) applied in the SGM, the source term is discretized as To prevent numerical instability near dry areas, the average velocity is defined by The time step ∆t is determined by the Courant-Friedrichs-Lewy (CFL) condition: where C c f l is the Courant number. C c f l = 0.5 is used in this paper.

Steady Flow over One-Bump Topography
Steady flow with one-bump topography is tested in a 25-m-long channel. The bathymetry is defined as Four cases are considered in testing the proposed scheme.

Still-Water Test for Well-Balanced Property
The first test is of the property to preserve the "lake at rest" in Equation (4) with wet/dry interfaces and 100 uniform cells. The initial water level is given by z x = max(0.1, B(x)) so that the bump is partially submerged. Simulations run for 500 s and the result is shown in Figure 5. The still water level remains unchanged and meets the analytical solution perfectly.

Transcritical Flow without a Shock
The unit discharge q = 1.53 m 2 /s is imposed at the upstream boundary and zero gradient condition is given at the downstream boundary. The initial water level and velocity are 0.4 m and 0 m/s, respectively. There are 100 computational cells. Figure 6 compares the numerical results and analytical solution. The surface profile has good agreement with the analytical solution. Furthermore, the comparison between the computed discharge and the theoretical results shows that the scheme is conservative. The relative error is given in Figure 7.

Transcritical Flow without a Shock
The unit discharge q = 1.53 m 2 /s is imposed at the upstream boundary and zero gradient condition is given at the downstream boundary. The initial water level and velocity are 0.4 m and 0 m/s, respectively. There are 100 computational cells. Figure 6 compares the numerical results and analytical solution. The surface profile has good agreement with the analytical solution. Furthermore, the comparison between the computed discharge and the theoretical results shows that the scheme is conservative. The relative error is given in Figure 7.

Transcritical Flow without a Shock
The unit discharge q = 1.53 m 2 /s is imposed at the upstream boundary and zero gradient condition is given at the downstream boundary. The initial water level and velocity are 0.4 m and 0 m/s, respectively. There are 100 computational cells. Figure 6 compares the numerical results and analytical solution. The surface profile has good agreement with the analytical solution. Furthermore, the comparison between the computed discharge and the theoretical results shows that the scheme is conservative. The relative error is given in Figure 7.

Transcritical Flow with a Shock
The unit discharge q = 0.18 m 2 /s is imposed at the upstream boundary and h = 0.33 m is given at the downstream boundary. The number of computational cells is 200. Figure 8 compares the computational results with the analytical solution. The relative error is given in Figure 9.

Transcritical Flow with a Shock
The unit discharge q = 0.18 m 2 /s is imposed at the upstream boundary and h = 0.33 m is given at the downstream boundary. The number of computational cells is 200. Figure 8 compares the computational results with the analytical solution. The relative error is given in Figure 9.

Transcritical Flow with a Shock
The unit discharge q = 0.18 m 2 /s is imposed at the upstream boundary and h = 0.33 m is given at the downstream boundary. The number of computational cells is 200. Figure 8 compares the computational results with the analytical solution. The relative error is given in Figure 9.

Transcritical Flow with a Shock
The unit discharge q = 0.18 m 2 /s is imposed at the upstream boundary and h = 0.33 m is given at the downstream boundary. The number of computational cells is 200. Figure 8 compares the computational results with the analytical solution. The relative error is given in Figure 9.

Subcritical Flow
A unit discharge q = 4.42 m 2 /s is imposed at the upstream boundary and h = 2 m is specified at the downstream boundary. The number of computational cells is 100. The numerical results are depicted in Figure 10 and have excellent agreement with the analytical solution. The relative error is given in Figure 11.
Water 2018, 10, x FOR PEER REVIEW 12 of 20

Subcritical Flow
A unit discharge q = 4.42 m 2 /s is imposed at the upstream boundary and h = 2 m is specified at the downstream boundary. The number of computational cells is 100. The numerical results are depicted in Figure 10 and have excellent agreement with the analytical solution. The relative error is given in Figure 11.

Drain on a Bump Bottom
In this test case, the bottom function is Equation (41) The left boundary condition is a free condition on h and zero on hu. The right condition is a dry bed outlet [13,21]. A total of 250 uniform cells are used in the computational domain. The results at different times (t = 10 s, 20 s, 30 s, 100 s, 1000 s) are presented in Figure 12. Since a dry bed outlet is imposed on the right boundary condition, the water is freely flowing out of the domain on the right.

Subcritical Flow
A unit discharge q = 4.42 m 2 /s is imposed at the upstream boundary and h = 2 m is specified at the downstream boundary. The number of computational cells is 100. The numerical results are depicted in Figure 10 and have excellent agreement with the analytical solution. The relative error is given in Figure 11.

Drain on a Bump Bottom
In this test case, the bottom function is Equation (41) The left boundary condition is a free condition on h and zero on hu. The right condition is a dry bed outlet [13,21]. A total of 250 uniform cells are used in the computational domain. The results at different times (t = 10 s, 20 s, 30 s, 100 s, 1000 s) are presented in Figure 12. Since a dry bed outlet is imposed on the right boundary condition, the water is freely flowing out of the domain on the right.

Drain on a Bump Bottom
In this test case, the bottom function is Equation (41)  The left boundary condition is a free condition on h and zero on hu. The right condition is a dry bed outlet [13,21]. A total of 250 uniform cells are used in the computational domain. The results at different times (t = 10 s, 20 s, 30 s, 100 s, 1000 s) are presented in Figure 12. Since a dry bed outlet is imposed on the right boundary condition, the water is freely flowing out of the domain on the right.

Dam-Break Problem over a Plane
In this section, we consider a dam-break problem over a plane [13]. The computational domain is [−15, 15] and the bottom is given by where the angle α will be defined as π/60 and −π/60 in this study. The initial condition is given by Concerning the boundary condition, the discharge q = 0 m 2 /s is imposed at x = −15 m, and a free boundary condition is considered at x = 15 m. In this numerical experiment, ∆x = 0.1 m.
The exact position of the wet/dry front is given by the following expression: where α is the angle in Equation (43) and h0 = 1 m in this numerical experiment. Depending on different values of α , various wet/dry fronts will appear. In this experiment, we consider two cases: α = π/60 and α = −π/60. Numerical results corresponding to the case α = π/60 (t = 0 s, 2.0 s, 2.5 s) are shown in Figure 13. The water level at times t = 0 s, 2.0 s and 2.5 s are shown in Figure 13a, and the comparison of the RSGM scheme and the present scheme at t = 2.0 s and 2.5 s near the wet/dry front is presented in Figure 13b. Observe in Figure 13b that the water level computed with the present scheme is tangent to the plane at the wet/dry front as was expected [22], and a small non-tangent front near the front appears with the RSGM scheme. The time evolution of wet/dry front location for α = π/60 and α = −π/60 are presented in Figure 14a,b, respectively. The analytical solutions are also shown in the figures for comparison. The conclusion can be drawn that the present scheme can predict the wet/dry front position well.

Dam-Break Problem over a Plane
In this section, we consider a dam-break problem over a plane [13]. The computational domain is [−15, 15] and the bottom is given by where the angle α will be defined as π/60 and −π/60 in this study. The initial condition is given by Concerning the boundary condition, the discharge q = 0 m 2 /s is imposed at x = −15 m, and a free boundary condition is considered at x = 15 m. In this numerical experiment, ∆x = 0.1 m.
The exact position of the wet/dry front is given by the following expression: where α is the angle in Equation (43) and h 0 = 1 m in this numerical experiment. Depending on different values of α, various wet/dry fronts will appear. In this experiment, we consider two cases: α = π/60 and α = −π/60. Numerical results corresponding to the case α = π/60 (t = 0 s, 2.0 s, 2.5 s) are shown in Figure 13. The water level at times t = 0 s, 2.0 s and 2.5 s are shown in Figure 13a, and the comparison of the RSGM scheme and the present scheme at t = 2.0 s and 2.5 s near the wet/dry front is presented in Figure 13b. Observe in Figure 13b that the water level computed with the present scheme is tangent to the plane at the wet/dry front as was expected [22], and a small non-tangent front near the front appears with the RSGM scheme. The time evolution of wet/dry front location for α = π/60 and α = −π/60 are presented in Figure 14a

Small Perturbation Test
This numerical experiment was proposed by LeVeque (1998) and also tested by Xing, Y. et al. [23,24] to demonstrate the performance of our scheme on a rapidly varying flow over a smoothed bed. A stationary state is perturbed by a pulse. Theoretically, the pulse splits into two waves propagating in opposite directions, in which the left wave travels over a horizontal topography, and the right-going wave propagates over a bump.
The computational domain is [0, 2]. The bottom topography with one bump is given by Here, ε is the height of perturbation constant. We take two cases: ε = 0.2 m (big pulse, see Figure 15), ε = 0.001 m (small pulse). The solution is taken at time t = 0.2 s with outflow boundary conditions and 200 cells. The results are shown in Figures 16 and 17, for ε = 0.2 (big pulse) and ε = 0.001 (small pulse), respectively. We compare the solutions with that obtained with 3000 cells. The

Small Perturbation Test
This numerical experiment was proposed by LeVeque (1998) and also tested by Xing, Y. et al. [23,24] to demonstrate the performance of our scheme on a rapidly varying flow over a smoothed bed. A stationary state is perturbed by a pulse. Theoretically, the pulse splits into two waves propagating in opposite directions, in which the left wave travels over a horizontal topography, and the right-going wave propagates over a bump.
The computational domain is [0, 2]. The bottom topography with one bump is given by Here, ε is the height of perturbation constant. We take two cases: ε = 0.2 m (big pulse, see Figure 15), ε = 0.001 m (small pulse). The solution is taken at time t = 0.2 s with outflow boundary conditions and 200 cells. The results are shown in Figures 16 and 17, for ε = 0.2 (big pulse) and ε = 0.001 (small pulse), respectively. We compare the solutions with that obtained with 3000 cells. The

Small Perturbation Test
This numerical experiment was proposed by LeVeque (1998) and also tested by Xing, Y. et al. [23,24] to demonstrate the performance of our scheme on a rapidly varying flow over a smoothed bed. A stationary state is perturbed by a pulse. Theoretically, the pulse splits into two waves propagating in opposite directions, in which the left wave travels over a horizontal topography, and the right-going wave propagates over a bump.
The computational domain is [0, 2]. The bottom topography with one bump is given by while the initial conditions are Here, ε is the height of perturbation constant. We take two cases: ε = 0.2 m (big pulse, see Figure 15), ε = 0.001 m (small pulse). The solution is taken at time t = 0.2 s with outflow boundary conditions and 200 cells. The results are shown in Figures 16 and 17, for ε = 0.2 (big pulse) and ε = 0.001 (small pulse), respectively. We compare the solutions with that obtained with 3000 cells. The figures show that the numerical solutions have good performance on simulating the wave propagated by a big or small pulse over a smoothed bed.

Evolution of Shorelines over a Parabolic Topography
Analytical solutions of one-dimensional shallow water equations were derived by Sampson et al. [25] for a perturbed flow with a parabolic topography. This provides a perfect test case for our scheme in dealing with the wet/dry front and bed slope. The parabolic bottom is defined by with constant h 0 and a. The computational domain is [−5000 m, 5000 m] with 200 uniform cells.
The analytical water level without the friction source term is given by where ω = 2gh 0 /a, and b is a given constant. The location of the wet/dry front is calculated by The relevant coefficients are a = 3000, b = 5 and h 0 = 10 for the test case. The initial water level is defined by Equation (49) at time t = 0.0 and zero discharge. The boundary condition has no effect on the numerical solutions in this case as the flow cannot reach the boundaries. Figure 18 shows the numerical water level at different time (t = 1000 s, t = 2000 s, t = 3000 s, t = 4000 s, t = 5000 s, t = 6000 s,) with the comparison of analytical solutions, where a nice agreement is observed.

Evolution of Shorelines over a Parabolic Topography
Analytical solutions of one-dimensional shallow water equations were derived by Sampson. et al. [25] for a perturbed flow with a parabolic topography. This provides a perfect test case for our scheme in dealing with the wet/dry front and bed slope. The parabolic bottom is defined by with constant h0 and a. The computational domain is [−5000 m, 5000 m] with 200 uniform cells. The analytical water level without the friction source term is given by The relevant coefficients are a = 3000, b = 5 and h0 = 10 for the test case. The initial water level is defined by Equation (49) at time t = 0.0 and zero discharge. The boundary condition has no effect on the numerical solutions in this case as the flow cannot reach the boundaries. Figure 18 shows the numerical water level at different time (t = 1000 s, t = 2000 s, t = 3000 s, t = 4000 s, t = 5000 s, t = 6000 s,) with the comparison of analytical solutions, where a nice agreement is observed.

Experiments of Dam-Break over a Triangular Obstacle
A laboratory experiment of dam-break flow over a triangular obstacle was carried out by the EU Concerted Action on Dambreak Modelling (CADAM) project. In this test, the numerical results applied in the presented scheme are compared with the experimental measurements. Figure 19 shows the experimental set-up, in which the dam site is located 15. The initial water depth is set to 0.75 m upstream of the dam with zero velocity and the rest of the domain is dry bed. A constant Manning coefficient n = 0.0125 is used throughout the domain. The upstream and downstream boundary conditions are assumed to be a solid wall and a free outlet, respectively. In the laboratory experiment, the dam suddenly fails and the water is released. The complex flow is associated with wave interactions, shocks, wetting and drying, and irregular bathymetry. Experimental and numerical time histories of the water depth at distances 17.5 m, 28.5 m and 35.5 m from the left boundary are plotted in Figure 20. It is concluded that the numerical results are satisfactory and the scheme can simulate complex flows with shock, irregular bathymetry and wetting/drying front.

Experiments of Dam-Break over a Triangular Obstacle
A laboratory experiment of dam-break flow over a triangular obstacle was carried out by the EU Concerted Action on Dambreak Modelling (CADAM) project. In this test, the numerical results applied in the presented scheme are compared with the experimental measurements. Figure 19 shows the experimental set-up, in which the dam site is located 15.

Experiments of Dam-Break over a Triangular Obstacle
A laboratory experiment of dam-break flow over a triangular obstacle was carried out by the EU Concerted Action on Dambreak Modelling (CADAM) project. In this test, the numerical results applied in the presented scheme are compared with the experimental measurements. Figure 19 shows the experimental set-up, in which the dam site is located 15.  The initial water depth is set to 0.75 m upstream of the dam with zero velocity and the rest of the domain is dry bed. A constant Manning coefficient n = 0.0125 is used throughout the domain. The upstream and downstream boundary conditions are assumed to be a solid wall and a free outlet, respectively. In the laboratory experiment, the dam suddenly fails and the water is released. The complex flow is associated with wave interactions, shocks, wetting and drying, and irregular bathymetry. Experimental and numerical time histories of the water depth at distances 17.5 m, 28.5 m and 35.5 m from the left boundary are plotted in Figure 20. It is concluded that the numerical results are satisfactory and the scheme can simulate complex flows with shock, irregular bathymetry and wetting/drying front. The initial water depth is set to 0.75 m upstream of the dam with zero velocity and the rest of the domain is dry bed. A constant Manning coefficient n = 0.0125 is used throughout the domain. The upstream and downstream boundary conditions are assumed to be a solid wall and a free outlet, respectively. In the laboratory experiment, the dam suddenly fails and the water is released. The complex flow is associated with wave interactions, shocks, wetting and drying, and irregular bathymetry. Experimental and numerical time histories of the water depth at distances 17.5 m, 28.5 m and 35.5 m from the left boundary are plotted in Figure 20. It is concluded that the numerical results are satisfactory and the scheme can simulate complex flows with shock, irregular bathymetry and wetting/drying front.

Conclusions
In this paper, we present a new reconstruction technique including flooded cells and partially flooded cells that preserves the non-negative values of water depth at the cell interface. Two cases are considered: (1) If the cell is flooded, our proposed modified RSGM is applied assuming that the bed topography is irregular in the cell; and (2) if the cell is partially flooded, a special reconstruction of the flow variables is developed where the bottom function is assumed to be linear in the cell. The scheme is well balanced and preserves positivity for wet and dry cells.

Conclusions
In this paper, we present a new reconstruction technique including flooded cells and partially flooded cells that preserves the non-negative values of water depth at the cell interface. Two cases are considered: (1) If the cell is flooded, our proposed modified RSGM is applied assuming that the bed topography is irregular in the cell; and (2) if the cell is partially flooded, a special reconstruction of the flow variables is developed where the bottom function is assumed to be linear in the cell. The scheme is well balanced and preserves positivity for wet and dry cells.

Conflicts of Interest:
The authors declare no conflict of interest.