I'd like to blend edges of overlapping DEMs such that the join is a smooth transition (for use with InSAR). Apparently the best interpolation kernel for this, according to the literature, is thin plate spline. Note that I don't want a blended image of the data (such that gdaldem might produce), I want to blend the actual data values themselves.
I had hoped gdalwarp's -cblend option might do it, but my testing shows that it seems to be a masking tool.
I realise that I could write a wrapper around scipy libraries to do this, but maybe someone has already done it?