The grid interval is a critical choice because for all methods and most point sets it will determine how well the information in the point set is represented. Always err on the small side when determining this parameter. It will run longer, but you will preserve as much information as possible that way. Only in the case of small datasets with unclustered distributions would you use a smaller interval than that predicted from the procedure below:
- Determine a grid spacing so that you have about 20000 nodes in the mesh. This is what is referred to as the default later.
- Determine the smallest spacing between points. Exclude the few that may be very close, have similar z values or are identical points. These are handled internally. Using 1/2 of this value, compare it to the default, then choose the larger.
- Determine a spacing that is about 1/4 the breadth of the smallest anomaly you expect or wish to see in the final surface. IF this is larger than the value chosen above, select it. Round down the value you have to something rational or meaningful in the coordinate system you are using.