Hubert wrote:
Basically: as small as possible, as large as necessary. Try it with your existing log data.
I think Hubert has summed it up nicely. Ideally, you would use as low an interpolation value as possible, but we appreciate that a lot of users like to "fill in the holes" of maps by using larger values. It is very important to remember that all interpolation is only guess work, and the larger the value, the more likely the guess is to be wrong. In the case of large interpolation values, which can be up to 200m, there is a lot of scope for error.