Picking option {.EDITGROUP.GENNODES}GENERATE gives a choice of methods for creating interior nodes for the model grid. Pick the first method, CLUSTERS, and examine the resulting right hand panel. A0, A1 and A2 are the coefficients in a quadratic relating cluster area to local water depth. With the default values shown, where only A1 is non-zero, cluster area will be linearly proportional to water depth, that is, the Courant number will be approximately uniform throughout the domain. Pick 'Display Clusters' in the right hand panel, then ACCEPT, in order to see how the program then groups squares from the Cartesian grid into clusters whose areas are roughly proportional to water depth (see Figure 3.2). The program then computes the centres of area of all clusters formed, and displays these points, which are the potential locations for interior nodes for the model grid. In this case, using the values chosen, the message at the top of the screen should indicate that about 475 interior nodes have been created. Confirm that these nodes are satisfactory, return to the top menu by repeatedly picking QUIT, and then pick SAVE and FINAL to output a file containing the set of boundary and interior nodes now available for triangulation into a model grid. The resulting file should be identical to file mod.nod in trigrid/demodata and is plotted in Figure 3.3..

It is evident that the process of forming Cartesian grids and creating interior nodes by forming clusters can be carried out quite quickly after a little practice. This means that when tackling an entirely new problem, it is quite practicable to try many different choices of grid resolution and cluster parameters, until a satisfactory set of nodes is obtained. As will be discussed later, the model domain can be subdivided into polygons and interior nodes can be created for each polygon in turn, using individual Cartesian grids and different choices of mesh parameters, with the result that very flexible control over generation of interior nodes can be exercised. In fact, by using different 'depth' grids at different stages, the spacing of interior nodes and hence element size can be geared to different scalar fields in different parts of the domain.