search,grids,SA significance test
Trevor, a few notes in addition to Roger's answer:
Trevor Doerksen wrote:
I have two specific questions: 1) ... Also, I don't know how to mask a specific area (like the nice grid in the meuse data set) of interest within that grid.
Depends on where you start. Suppose you have a SpatialPolygons or SpatialPolygonsDataFrame object, you can generate a regular grid using spsample, or start with makegrid, convert to SpatialPixels, and then use overlay the result to obtain NA indices for grid points outside any polygon. If you don't have a polygon, you may select grid points within a certain distance to nearest observations by interpolating them with a search radius; again the NA pattern will reveal grid points too far away.
2) Is there a test to compute the significance of spatial autocorrelation in a variogram? The biological signal I'm modelling in my variograms is very weak, with a huge nugget. How large can the nugget be before it negates the spatially specific weights for kriging, thus rendering the prediction close to an inverse distance weighting scheme?
IMHO, insignificant spatial correlation in spatial data will usually indicate a too small sample. A large relative nugget may indicate that statistics under the independence assumption are reasonable, insignificant correlation still does not imply zero correlation. Large nugget effects lead to predictions equal to the grand (or local) mean or trend estimate, not the inverse distance weighted interpolated value. -- Edzer