[R-sig-Geo] knearneigh() computing efficiency on large datasets

[Pierre Roudier]

Dear list,

I am processing gridded data like DEMs, imported in R from geotiff
files. Of course, such data is regularly gridded, but very often with
NA values. In the framwork of some tests, I have to generate a
neighbourhood of the DEM, so that to extract local extrema of the
layer. For this task, I wrote a function based on  the knearneig()
function from the spdep package, with k=4 or k=8, to generate and
analyse the neighbourhood of each point.

Unfortunately, I often have to process big datasets (let's say grids
with between  50,000 and 350,000 points), and that's working but
knearneigh() takes *hours* to process.

Does anybody would have any suggestion to improve the efficiency of this step ?

Thanks,

Pierre