That has to be one of those classic point patterns that really shouldn't be interpolated beyond the bounds of their individual clusters.
The blue area is apparently some formation? of some kind? and you had to use the only data available to you? Would that be a fair guess?
What is wrong with dealing with analysing the three clusters separately, perhaps using a small buffer around each cluster's convex hull. What do each of the clusters show? Are they hugely similar? different? Are there any distinct directional trends?
What is the proposition that the intervening space between the clusters shares properties or differences with each other?
What is the need to interpolate between the clusters?
More importantly, what is the need to extrapolate beyond the boundaries.
Just some food for thought and discussion, before you get into a 'which is the best interpolator' discussion