In river engineering problems, the flow cannot always be approximated by two-dimensional models. The complexity of flow, especially on the deformed beds and irregular banks, plays an important role in the flow structures, sediment transport and the morphological changes of the rivers. Therefore, simulation on the three-dimensional structures of the flow is essential in prediction of realistic bed shear stresses and can be a suitable estimation for sediment transport and bed morphodynamics.

Several depth integrated two-dimensional models are implemented in iRIC. They demonstrated a great success in the simulation of large scale engineering problems. However, they are not able to capture the three-dimensional structures of the flow even in the presence complex geometries. They estimate the flow roughly ignoring the vertical flow structures and approximate the flow under the assumption of hydrostatic pressure. The phenomena such as secondary flow, turbulence eddies, horseshoe vortices, etc. cannot be observed under shallow-water estimation.

NaysCube was a solution for these problems, in which the flow can be solved as three-dimensional, using Reynolds Averaged Navier-Stokes (RANS) model. NaysCube, in its turn, is able to solve a wide range of three-dimensional problems. It is based on curvilinear coordinate system with k-epsilon model as a turbulence closure. NaysCube was successful to simulate the range of three-dimensional problems. However, NaysCube is based on RANS which remove all flow fluctuations, and lead to a smooth flow field. This is not true in reality hence NaysCube fails in simulation of highly complex geometries, in which the turbulence gets significantly affected by the geometry.

Here, we introduce NaysEddy as a three-dimensional solver based on large-eddy simulations (LES). It solves the flow in more details using Cartesian grids with ghost-cell immersed boundary methods. This solver is flexible, accurate and it is able to solve the complex problems with a great success. It is tested under extremely complex conditions, moving boundaries, complex bed topography, and flows with relatively high Froude numbers, etc., and the solver shows its capability in the simulation of those problems. This solver applies the full Navier-Stokes equations without approximation.

However, the computational load of this solver is one order lager than that of Nays2D. The solver is effective for reproducing local phenomena of rivers for a short time span, rather than reproducing a large section of the river for a long time span.