AKDE bounary corrections

[Christen Fleming]

I just pushed an update to GitHub that allows for hard boundaries to be enforced in akde() at the kernel level, which is statistically better than correcting the entire UD afterwards. This will be described in an upcoming manuscript.

The below plot is of a fish in a lake:

For soft boundaries, like bathymetry dependence, rsf.fit() can also be used to model that and then fed into akde().

Best,

Chris

[Michael Taylor]

Hi Chris thanks for adding this, I'm looking forward to seeing the publication that comes with it. 

I have been having a play and was wondering if there is any way to account for cases where part of the kernel is orphaned.

In the example below an animal is recorded in water on one side of a peninsula.  When calculating the kernel and enforcing the boundary the area of the peninsula is removed from the kernel, but an area on the far side of the peninsula where the animal has not been recorded is included in the 95% areas. I think I understand why this is happening based on the masking process and underlying code, but it is very unlikely based on the known locations that the animal is using that area in this case. Is this something that could be easily removed or added to the function?

Cheers,

Mike

[Christen Fleming]

Hi Mike,

This issue will be discussed in the upcoming manuscript. RSFs are really IID models and the RSF-kernels are static kernels. When the kernels are large as compared to habitat protrusions like this, you can still get leakage into suitable areas that the individual couldn't reasonably have traveled to (conditional on the data). This will get resolved in the future, when we code up continuous-time habitat selection models, which will generate each kernel via a boundary respecting diffusion process, but before then I'm not sure if there is an easier quick-fix that can be made.

Best,

Chris