GRID CONSIDERATIONS

PROBLEM DOMAIN
  • Use well-defined, permanent natural boundaries when possible.
  • If a boundary is not permanent (e.g. a ground-water divide) anticipate potential future variations, and either accommodate them from the start or be prepared to monitor appropriately and make adjustments later.
  • Most approaches to grid development require substantial time and effort to make substantial changes to the model grid. If you are using one of these approaches it is often better to construct the large, more elaborate grid from the start.
  • To illustrate some of the concepts on this page, we will use a simplistic basin. Click here to view it: example basin.

EXTERNAL INACTIVE GRIDS
  • Orient the grid so as to have as few nodes external to the model domain as possible unless other criteria in this list cause you to do otherwise. This will minimize input and output file size and make data management easier.
  • Click here to view an example of an alternative grid orientation.

FLOW DIRECTION
  • Orient the grid such that the primary flow direction in the model area is aligned with the rows and columns of the grid. Flow calculations are oriented along rows and columns, so diagonal flows are calculated in a stair-step manner, thus orienting the rows and columns in the direction of flow will reduce errors.
  • Click here to view an example of a grid orientation that is more closely orient with the flow directions.

ANISOTROPY
  • Orient the grid such that the rows and columns of the grid coincide with the major axes of the hydraulic conductivity ellipsoid. Flow calculations are oriented along rows and columns in MODFLOW and it is assumed you have aligned the grid with the anisotropy. As you can imagine, this can cause problems if some layers have different anisotropy orientations.

MINIMIZE # OF CELLS
  • Fewer cells will yield a model that is easier to manage and executes more quickly. However, there is a trade off between cell size and accuracy, with smaller cells providing a more accurate answer. So, while trying to keep the number of cells low, add detail as required to cope with the following issues.

BOUNDARIES BETWEEN FEATURES / HIGH CONTRAST IN MATERIAL PROPERTIES
  • Use a more detailed grid along boundaries or at contacts where conditions change abruptly. This will allow accurate spatial definition of the boundary or contact. It also allows for definition of a gradual transition in parameter values at a contact, which can reduce calculation errors or convergence trouble. If the cells are small, such a gradation is a fairly good approximation of the actual transition.
  • Click here to view the grid with more detail around the contacts.

STRESS AREAS (steep gradients)
  • The gradient between finite difference cells is represented by a straight line. We all know that a curve is better approximated by many small straight line segments rather than a few large segments. Similarly, when the gradient is steep, the solution approximates field conditions better if many small cells are used.
  • Click here to view the grid with more detail around the wells.

OBSERVATION POINTS/AREAS OF INTEREST
  • Although we can interpolate values of head, concentration, or flow rate between cells it is more accurate and more convenient to have cells positioned at locations where the modeler needs to know the simulated result.
  • Click here to view the grid with more detail around the observation point.

SYMMETRY
  • Be on the look out for symmetry that may allow you to cut your model size in half or more. This is a common occurrence when simulating engineered features. For example, in a homogeneous material that doesn't experience any irregular, or asymmetric external stresses, the parallel drain problem would require only a 2D cross sectional model from the centerline to one drain. If you were simply simulating radial flow to a well, only a small sliver that widens with distance from the well would be needed. Many codes offer a 2D radial option. In this case the cell width into the computer screen increases with distance from the well by the function 2 pi r (the circumference of a circle). Beware not to add a sand lens at some distance from the well into you radial cross section, unless the lens happens to be shaped like a doughnut!

RELATIVE SIZE OF ADJACENT GRIDS (1.5)
  • If adjacent grids have substantially different size, then truncation errors may occur in the matrix solution. To avoid problems maintain a maximum size difference of 1.5 for adjacent cells. For example adjacent cells of 5 and 7.5m are fine, but 5 and 10 m is too much.

ORTHOGONAL DIRECTIONS (100:1)
  • The aspect ratio is less critical. It is generally acceptable for the ratio of length to width, or width to length, to be 100:1.

FUTURE SOLUTE TRANSPORT
  • Transport modeling frequently requires much smaller cells than flow modeling. In order to provide accurate velocities for the transport model it is best to anticipate the cell sizes needed for the transport model and use them in the flow model from the beginning. For now, do not be concerned with this, as we will talk about the grid requirements of transport models later in this course.



Smaller cell sizes provide more accuracy, but one wonders how small is small enough?