1. The grid
To allow flow to be computed numerically, space and time need to be discretised. For time are time-steps defined, and for space is a grid defined. In the sections below we elaborate on the grid specified for computations in the 1D and in the 2D surface water and groundwater domains.
3Di simulations can consist of a surface water, a groundwater and a 1D network component. The surface water and groundwater components are in principle a two-layer model, which can be coupled to the 1D network. Therefore, the grid of the surface water and groundwater domain are similar, while the 1D network is fundamentally different.
1.1. Computational grid for 2D domain
In 3Di we make use of a so-called structured, staggered grid. This implies for the surface water and the groundwater domain that computational cells are perfect squares, where the pressure/ water levels and volumes are defined in the cell centres and velocities and discharges are defined at the cell edges. This also yields for the vertical coupling. In the Figures below, an example is shown of a grid in a 2D domain and the two-layer coupling.
The hydrodynamic computations are based on the conservation of volume and momentum. In the next sections (Conservation of mass, 1D Flow, 2D Surface Flow, and 2D Groundwater flow), the methods concerned for the computations are discussed. However, in order to solve the equations, the domains in which they are valid, need to be defined. In the Figure above, the volume and momentum domains are shown.
1.1.1. Grid Refinement in 2D
The computational cost of a simulation is strongly related to the number of computational cells. One always needs to find a balance between grid resolution and computational time. There are often regions where the flow is more complex or where one requires results with a finer resolution. To optimize the computational cost and grid resolution, users can refine the grid locally. 3Di uses a method called quad-tree refinement. This means, that in space, refinements are placed by dividing neighboring cells by a factor 4 (Figures below and above). This is a simple refinement method that forces smooth grid variations, which enhances an accurate solution of the equation.
In case groundwater is taken into account, the grid refinement is present in both the surface layer and the groundwater layer.
1.1.2. Subgrid method
Flow is strongly affected by the bathymetry. Therefore, to simulate flow accurately, information of the bathymetry is crucial. To illustrate the importance of a well described bathymetry, consider the three flows in a flume shown shown in the Figure below. While the bathymetry differs only slightly, the flow behaves completely different.
Nowadays, detailed bathymetry information becomes more and more available. However, it is difficult to take detailed grid information into account, without a strong increase in computational cost. In search of an optimal balance between computational cost and accuracy, the so-called subgrid method (Casulli 2009, Casulli&Stelling 2011, Stelling 2012, Volp & Stelling 2013) is developed and implemented in 3Di.
The basic idea behind the subgrid method, is that water levels vary much more gradually than the bathymetry. In hydrodynamic computations, water levels are assumed to be uniform within a computational cell. Traditionally, this is also assumed for the bathymetry within such a cell. The subgrid-based approach allows the bathymetry to vary within one computational cell, while the water level remains uniform. In 3Di two grids are to be defined; a high resolution subgrid and a coarse(r) computational grid. All input data, such as the bathymetry, roughness and infiltration rates can be defined on the high resolution grid, while the computations are performed on the coarse computational grid. Volumes and cross-sectional areas are based using the high resolution bathymetry information. The variation of the bathymetry within a computational cell means that a cell can be dry, wet or partly wet. This approach has two implications:
The volume has a non-linear relation with the water level, because when the water level rises, the wet surface area increases is well. Such a system can be solved using a newton iteration. To compute the volumes at the next time step.
As we are allowed to have a non-linear relation between water level and volume, 3Di can deal automatically with flooding and drying. No artificial thresholds are necessary.
Users define for the grid generation a cell size (of the finest grid resolution) and the number of refinement layers. A computational cell consists always of an even number of subgrid cells. In addition, the user needs to define where and if refinements should be defined. One can define polygons or lines to indicate these areas and the refinement level. For a detailed example, see Flood model.
1.1.4. Some facts and figures:
The use of high resolution information goes hand in hand with large amounts of data. To compress this data, it is stored during the computations in tables. More information about this can be found in Tabulated data storage.
There are more variables defined at the high resolution grid; such as roughness, infiltration capacity and hydraulic connectivity. These will be introduced later in the documentation.
The numerical grid is generated based on some the following settings: - Grid space - Number of refinement levels - Grid refinements
The grid space is the measure of the dimension of a computational cell. It is important that a width and the height of a grid cell contains an even number of subgrid cells. In case the dimensions of a subgrid cell are 0.5 x 0.5 m2, the grid space can be 5.0 x 5.0 . In case the dimensions of a subgrid cell are 1.0 x 1.0 m2, the grid space can not be 5.0 x 5.0 m2. The grid space is defined in the v2_global_settings table and is the dimension of the smallest grid size. The kmax setting is the number of refinement levels. Locations where the refinements need to be defined can be added by a line using the v2_grid_refinement table, or by an area using v2_grid_refinement_area. In case two refinement levels are defined at the same location, 3Di will refine to the highest level indicated. 3Di will always aim at a minimum number of grid cells, it will coarsen the grid as fast as possible, but it will only be possible to do that in steps of four.
1.2. Computational grid for 1D domain
For studying the flow of narrow features in the landscape or sewer systems, it is advantageous to use one dimensional models. This allows for an extensive description of the system, without actually computing cross-flow phenomena. These are in those cases limited and the use of a 1D representation will reduce the computational cost. In 3Di 1D networks can be defined, representing open channels, manholes, weirs, orifices, culverts and closed pipes. There are several options to couple the 1D and the 2D domain (see Section 1D Flow). All options for the coupling allow for a fully integrated computation, this means that the full 1D and 2D systems are solved as one.
To compute the flow in a 1D network a grid is to be defined as well. Consistent with the grid defined for the 2D domain, a staggered grid is used again. Pressure/ water level points are allocated with a velocity point. An example can be seen in the Figure below.
1D networks can consist of open channels, closed pipes and various structures. More about the various options can be found in the Sections 1D flow over structures and Channels. The resolution of the 1D domain can be defined per 1D element.