Hi Andrew,
Your question is a little bit unclear. It would helpful to have a minimal working example, but from what I understand, I don't think you need to interpolate anything, and griddata is not necessary, as your matrices are already gridded.
For a rainfall matrix R with corresponding probabilities P, you can get a logical mask of all the grid points meeting a condition like
mask = P>=0.05 &P<0.10;
which would have ones wherever P is between 0.05 and 0.10, and zeros elsewhere. Then the corresponding rainfall in those grid cells is
R(mask)