Jennifer,
here are some notes that may be helpful for finding your answers:
1. The main factor here is the amount of spatial correlation. Try
to simulate large fields without correlation ("pure nugget"), and
then try to simulate large unconditional fields without nugget and
with long-range correlations (e.g. range the size of a field
dimension). You'll find that the nugget fields have fractions of 1's
very close to the pdf value (simple kriging mean), whereas the long
range ones vary much more. More data do of course constrain the
unconditional fields more, but the effect will still be there. Simply
put: white noise averages out fast over large fields, correlated noise
less fast.
2. Conditioning data should be honoured in the resulting simulations.
Make sure that the locations correspond _exactly_ to the grid cell
centers. One way to ensure this is to feed gstat with data that are
on a grid map (I'm not sure whether Idrisi's interface allows this).
Best regards,
--
Edzer
