I'm currently using the R
latticeDensity package to estimate home ranges of animals.
Within the documentation,
I've tried reading the cited Sain, Baggerly and Scott (1994) paper, but unfortunately it is above and beyond my understanding. Similarly, any searches for "cross validating" a model seem are largely applicable to people who know what they're doing with custom built models (e.g., here or here).
From my very basic understanding (largely from this post), cross validation ensures you do not overfit your model. In order to not overfit this home range density model, one must choose the lowest
k value that is spit out by this UVC function.
The model inputs in this particular package are lat/long animal locations within a lattice grid, the distance between each grid point, the number of 'steps' an animal is making across that grid, and a probability value that the animal will jump to the next grid point vs. staying in place. From the original
This would be analogous to adding a quantity of dye to each location where a fish or animal was observed, then allowing the dye to diffuse outward. A density map based on the concentration of the dye at different times would result in a density estimator that is faithful to the boundaries of the region.
Unfortunately, running the cross validation function within the
latticeDensity package takes hours to run on my computer, and is not feasible given the number of home ranges I need to calculate. I understand the underlying ecological principle of animals diffusing across a grid, but I do not actually understand what the purpose of cross validation is within the context of calculating home range size.
My questions are:
- What is this UVC method actually doing?
- How important is running my model with the lowest
kvalue to produce an ecologically meaningful result? Can I just choose a
kvalue and run with it for every one of my species home range models, or will this
kvalue likely change drastically between species?