How can I estimate kernel density, taking barriers (i.e. a priori zero-probability regions) into account?

At the moment, regions where the population cannot occur are clipped out for display. However, I imagine this is biasing the results, since observations near the zero-probability regions don't have as much impact as they might.

I have several thousand observations, so a tool or programmatic solution is a necessity. Bonus points for an implementation in ArcGIS or R.

  • 1
    One thing people do is when clipping out the regions of zero probability, they simply re-weight the densities back to unity. This is the simplest type of dasymetric mapping procedure. Preferably you would not want to smooth over such regions to begin with, but I'm not sure if any implementations of that exist. – Andy W Aug 20 '12 at 2:54
  • Related: gis.stackexchange.com/questions/30562/… – whuber Aug 20 '12 at 15:19

You can take boundaries properly into account when calculating kernels using the R package 'adehabitatHR' - see this link: http://cran.r-project.org/web/packages/adehabitatHR/vignettes/adehabitatHR.pdf

You have to use a relatively simple boundary line and have no locations that are erroneously on the wrong side of it.

This is the only proper implementation of this that I've found so far, but would love to know of any others as the oversimplification of coastline is a bit of a problem for my data.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.