Assessing different methods of generating a three-dimensional numerical model mesh for a complex stream bed topography

Abstract

Three-dimensional numerical models of flow over complex bed geometry are becoming widely used in river and coastal engineering. Boundary-fitted coordinate grids are typically used to deal with this problem in natural channels. Recently, a regular structured grid method based on numerical porosity has been developed for high-resolution gravel-bed models. A simpler alternative approach is to use a Cartesian mesh with an interpolated 3D solid object representing the river bed. The objective of this study is to assess the impact of these three methods of mesh generation on the simulated flow field above complex bed topography around stream deflectors in a laboratory setting. Results show marked differences between the three types of simulations when running mesh sensitivity analysis. Because the bed porosity approach uses the digital elevation model (DEM) information in each cell to represent bed topography, it requires a finer mesh resolution in order to reach an accurate solution. Although there was good qualitative agreement between the simulated flow fields and Acoustic Doppler Velocity measurements, the quantitative comparison showed relatively poor agreement for all mesh design types. However, the three mesh types were in good agreement when compared to each other for velocity and pressure variables (average correlation coefficient, r, of 0.95), with the 3D object bed and bed porosity showing the best agreement (r ¼ 0.97). Simulated turbulence variables (KE and EP), however, showed more scatter (r ¼ 0.85) and slopes markedly different from unity.

A three‐dimensional numerical model investigation of the impact of submerged macrophytes on flow dynamics in a large fluvial lake

“…To determine the refinement factor of the sub‐grid, a grid sensitivity analysis was conducted with three grid resolutions (of 76 × 55 m, 25 × 20 m, and 15 × 12 m) in the study zone (keeping a coarser fixed LSP grid size in all tests). Modelling results using the second grid refinement (factor of 3) revealed a percentage difference in maximum velocity from the finest resolution (refinement factor of 5) of <10%, which is considered a satisfactory threshold (Biron, Haltigin, Hardy, & Lapointe, ). The refinement factor of 3 was therefore used, resulting in average cell size in the study zone of approximately 25 m (Figure ).…”

“…To determine the refinement factor of the sub-grid, a grid sensitivity analysis was conducted with three grid resolutions (of 76 × 55 m, 25 × 20 m, and 15 × 12 m) in the study zone (keeping a coarser fixed LSP grid size in all tests). Modelling results using the second grid refinement (factor of 3) revealed a percentage difference in maximum velocity from the finest resolution (refinement factor of 5) of <10%, which is considered a satisfactory threshold(Biron, Haltigin, Hardy, & Lapointe, 2007). The refinement factor of 3 was therefore used, resulting in average cell size in the study zone of approximately 25 m(Figure 2).Delft3D allows utilisation of different roughness coefficientsincluding Manning's n. Although the simplest approach was to use a single roughness coefficient value for the entire domain, we preferred the use of a generalised roughness coefficient map, to better represent the spatial variations in flow resistance, while considering the scale of LSP.…”

Sediment Transport and Flow Dynamics Around a Restored Pool in a Fish Habitat Rehabilitation Project: Field and 3d Numerical Modelling Experiments

“…Furthermore, there is greater uncertainty in simulated turbulence quantities than mean velocities in 3D simulations using RANS turbulence models. For example, Biron et al (2007) observed considerable variations in turbulence quantities around deflectors solely due to the type of mesh used (boundary‐fitted coordinates, object bed or bed porosity methods). Unsteady simulations such as LES or DES (Detached Eddy Simulation) could be used to compare bed shear stress estimates using various methods such as Reynolds shear stress, turbulent kinetic energy and log‐law in flow fields near deflector‐like structures.…”

“…The advantages of this approach are that (1) a Cartesian grid can be used where the angle between the cells always remains at 90°, which solves the problem of highly skewed cells in the constriction zone near deflectors that would occur with boundary‐fitted coordinates, (2) different flow stages can easily be simulated with the same ‘object‐bed’ and (3) the mesh resolution can be modified easily without changing any other parameter. This method has been tested successfully around laboratory deflectors (Biron et al ., 2007).…”

Sensitivity of simulated flow fields and bathymetries in meandering channels to the choice of a morphodynamic model

“…To avoid spatial autocorrelation problems in statistical analyses (Fortin et al, 1989), 200 test points were randomly selected for each configuration to examine discrepancies among and between predicted (numerical simulations) and measured (flume experiments) values. Reduced major axis regression (RMA) is used instead of ordinary least square regression to account for potential errors in both the dependent and independent variables (Hardy et al, 2003;Biron et al, 2007) and to maintain the variance of observations in our predictions (Berterretche et al, 2005). Results of RMA analyses are presented in this paper for the M med but are available as supplementary material for the other two configurations (M low and M high ).…”