Selecting points on the edge of the study area to represent the surface was a complicated process. First the vertices of the boundary line were written out, and assigned elevations from elevation grid. Elevations which were (slightly) lower than 300 meters were changed to 300 meters.
Next the boundary line was converted to grid, and the cells with elevations over 300 meters were selected. Those cells that were local maxima within 1500 horizontal meters were selected and converted to points. This became the second component of the elevation model.
Next, all points where the boundary intersected areas higher than 300 meters were selected and added to the data model. Thus the edge of the study area is represented by lines at an elevation of 300 meters, except where the present elevation excedes 300 meters.
The interior points of the elevation model comprised all points above 300m which were local maxima over a specified radius. Points of equal elevation (we worked in integer meters) were both both included if they were at least 1500 meters apart, else the first tied peak was used.
All these data points became components of a (TIN) in which the [hypothetical] landscape is represented by triangles. Many analyses can be done very quickly on a TIN, and it can also be converted to a lattice of regularly speced (30-meters, in our case) lattice of points.