How to reassign small polygons to its nearest neighbor in R (similar to gdal_sieve.py function)?

Question

I created a spatial polygon file from a classified gridded dataset using the grid2polygons function in R. I would now like to take that polygon file and merge any polygons containing less than 6 pixels with its nearest largest neighbor. In the image link below, you can see an example of small polygons circled in blue. I would like those merged with the larger surround polygon. Alternatively, if there is a way to create a minimum threshold of 6 clumped pixels prior to the conversion to polygon stage, any suggestions for that method would be great as well.

There is a function in Gdal: gdal_sieve.py that can do exactly this (http://www.gdal.org/gdal_sieve.html) if you would like to get an idea of exactly what I would like to do. Does anyone know of a way to do this in R?

Answer 1

You can call gdal_sieve using system2() in R if you have your system environment variables set up correctly, e.g.:

system2('gdal_sieve', args = c('-st', 6, infile.tif, outfile.tif))

then vectorise the output with

system2('gdal_polygonize', args = c(outfile.tif, '-f', 'GPKG', 'outfile.gpkg'))

You can do all this using GRASS via R too. The process is a bit more involved, but you have more control over some aspects, and it handles no-data areas properly. I've blogged it here but to summarise:

• read your raster into GRASS
• use r.to.vect to vectorise the raster
• use v.clean with tool 'rmarea' and an appropriate threshold to remove extraneous tiny polygons
• if you want something with smoother boundaries, convert the results of the above process back into a raster, then convert it to vector again using the flag 's'.

Be aware that the topology in the resulting outputs can be problematic and that no easy fixes appear to exist (gosh, I'd be happy to be wrong about this). So the outputs can be fine for interactive use, but subsequent geometric operations like intersect and dissolve can fail unless you make some manual edits.