UD resolution/information content

[Vitek Jirinec]

Hello Chris et al.,

I'm interested in habitat selection analyses using the UD produced by AKDE, such as relating high-use areas to specific habitat features. However, as I understand, the resolution of the UD raster depends on the bandwidth size, which in turn depends on the location sample size as well as the animal's behavior. You superimpose the white grid on the UD plot, but could you tell me how to get the area with distinct information content from the UD object? Is it the DOF of area, bandwidth, or is this obtainable in some other way? I'd like to link the CDF values with landcover, but I want to do this in a sensible way.

Many thanks,

Vitek

--

Vitek Jirinec, PhD

www.vitekology.org

[Christen Fleming]

Hi Vitek,

Sorry for the late reply. I would usually do habitat selection with the location data. There's some new and in-development autocorrelation-weighted & integrated RSF code in the development version of the package on GitHub (see help('rsf.fit')). It also automates numerical convergence.

The grid displayed when plotting the AKDEs is obtained from the bandwidth matrix in the 'H' slot of the UD object. This is kind of like a covariance matrix for the individual kernels that make up the AKDE. The grid is calculated from the matrix-square-root of this matrix, which is like a matrix of standard deviations. I doubt you need this, though.

For making simpler comparisons with the AKDEs, I personally wouldn't do anything much more complicated than calculating expectation values with the distributions, which is easy. If PMF is the exported PMF of the AKDE and R is a raster of covariates, and they are in the same projection with compatible grids, then sum(PMF*R) is the expected value of covariate, E[R], according to that distribution, and VAR[R] is sum(PMF*R^2) - sum(PMF*R)^2.

Best,

Chris