Dispersive and Reproductive KDE results

[Unai Ormazabal]

Good morning Chris:

First of all, my apologies if this question has already been posed and answered, I haven't been able to find it. I am currently working with GPS data from large birds of prey at quite high resolution (fix every 5 minutes) and I have done KDE, AKDE and wAKDE-s for the subsample of the data I normally use to test new functions and packages. In this subsample I have several reproductive individuals of different species and a dispersal individual from one species. I have looked at the tutorials in the AniMove workshops and from what I gather the more autocorrelated the dataset is the bigger the difference will be in the estimate of KDE vs AKDE & wAKDE. From that logic I would infer that the dispersing individuals who are not bound to a territory and therefore can move more freely and in a "more Brownian" way should have less autocorrelated positions and thus that the KDE-s should be very similar to AKDE or wAKDE-s. Nevertheless, the opposite is true. In all of the territorial individuals I have compared different types of KDE-s differences are pretty negligible (see L_2023_wAKDE.tif and L_2023_KDE.tif as an example). In the other hand, the KDE-s for the dispersive individual are really different if the autocorrelation is taken into account, furthermore, the level of uncertainty in the plots and the differences in KDE plots seem to reduce as the animal turns more and more territorial (see C_2022KDE through C_2024_wAKDE). Why is that? It seems to be that I have understood everything backwards and I am not sure how positional and velocity autocorrelation works. Furthermore, is it that normal for KDE and AKDE-s to be that similar in reproductive individuals even with such high fix resolution? Is it because large birds of prey are able to move through their territory a lot quicker than other animals (like the buffalo or the gazzelle in the examples) and thus the position autocorrelation is lost in a shorter timespan? I've been trying to answer this questions with the help of AI but I am not sure about the responses and it would be really helpful if you could help me interpret these results.

               Kind regards:

                          Unai

PS: I have tried to attach images in text but I could only attach them this way sorry.

[Christen Fleming]

Hi Unai,

Dispersing individuals will have more autocorrelated locations (with the same sampling), because their autocorrelation timescales are longer, because it takes longer for a dispersal event to occur than a home-range crossing event to occur.

If you can sample a large number of range crossings, which is more possible when the individuals can cross their home-range faster, then KDEc method in ctmm will converge to the AKDEc method in ctmm with on the order of 100 range crossings observed, which you can see in the DOF[area] parameter estimate.

Best,

Chris