Purely manual construction of an irregular grid obeying any of the constraints described in the preceding section is difficult and time-consuming, even leaving aside the complicated book-keeping required to specify the geometry of the finished grid in a form suitable for subsequent use in numerical models. Simpson [1979] and Thacker [1980] reviewed many widely varied methods for handling all or part of the grid-preparation process by computer. Simpson classified the available methods for producing triangular grids into four types:

- curvilinear coordinate mapping of simply shaped regions with regular grids into regions with complicated boundaries.
- generating grids by subdividing initially coarse grids.
- boundary contraction methods in which grids are constructed inwards from the model boundaries.
- vertex triangulation methods in which vertices (nodes) are distributed through the model domain and then connected appropriately by a triangulation algorithm.

In principle, methods of type (4) should be inherently more flexible than the others; in fact, (1),(2) and (3) are in a sense subsets of (4). The method described in this report is a partially automated method of type (4). Much of the work published since Simpson's and Thacker's reviews has been concerned with the algorithmic properties of various methods (see for example Joe [1984]), with the aim of completely automatic generation of grids without errors or anomalies. The direction followed here is quite different. The human eye can assess many geometric properties of two-dimensional figures much better than any available computer program. With current graphic display units and interactive graphics software, it is practicable to include the modeller whenever necessary in the grid design process, while maintaining automatically an up-to-date description of the grid for subsequent modelling purposes. Partial automation, in this sense, of the most subjective stages of grid design is probably far more cost-effective than full automation at this time. The method used in TRIGRID for generating a set of suitably-spaced nodes for a model grid is outlined in Henry [1988] and Henry and Walters [1993]. The triangulation algorithm used was devised by Renka [1982] (alternatively, see Cline and Renka [1984]) and subsequently modified by Bova and Carey [1992] to handle non-convex domains, using a method devised by Vassberg [1989].