AKDE for a central place foraging seabird species

[Thomas B]

Hi Chris, hi everybody, 

I want to get population-level autocorrelated kernel density home-range estimates for a seabird species. I have around 20 individuals tracked for several days. I joined an example for one individual.

They usually make trips for several hours during the day, go back to the colony, and rest during the night. When considering all GPS locations, the variograms seem ok, and I'm able to get AKDE. However, more than half of the GPS points are located at the colony, and I suspect this result is not reflecting what I want. I would like an area reflecting the population's 'foraging home range' if that makes any sense. 

To do that, I'm considering subsets of GPS locations: 

1) GPS locations of foraging trips (removing all the locations too close to the colony)

2) Foraging locations. I have two different definitions of foraging locations here: locations of dives (pressure sensor) and locations corresponding to one of the states from an HMM I used to segment data. 

In any case, my variogram never reaches an asymptote. I read in previous posts that I may need to expand the time. I'm not sure to understand what that means and how to do it in practice. I have periods where the speed is null (night at the colony), whereas other periods have a speed higher than during foraging segments (flying transit). Is there any way to address my problem?

Best, 

Thomas

[Christen Fleming]

Hi Thomas,

I'm curious as to how the empirical variograms compare to the fitted model, as the all-data variogram looks good before segmentation. But this example also looks like two very distinct foraging areas, where the seabird spent a few days foraging in one area and then switched to a second area. This would make the variogram increase on that scale, like you see. You could consider those areas separately, though the sample size is getting fairly small with about 1 DOF per trip.

Best,

Chris

[Thomas B]

Hi Chris, 

Thank you very much for your answer. Most of my individuals have these type of behaviour on the considered period, going to a foraging area once and then to another one. The effective sample size is critically low for many birds, especially after segmenting data. Since I'm interested only in the colony foraging area of each year and all the trips seem independent even within a same individual, I considered a slightly different approach. I created a "newdate" variable adding a sufficient time between each trip (one week for example) to avoid overlapping and considered all data as if it was a single individual. Is this approach correct? Do I need to be careful about specific things when doing it? (I tried different order of magnitude for the time gap between each trip and I had almost the same results each time). I had to play with the dt argument to visualize model fit but I always have a strange "peak" at the beginning of the variogram. When zooming in the beginning of the variogram it's like it's not at the same scale. Is it just a visualization artifact or it could indicate a problem? 

Best, 

Thomas

[Christen Fleming]

Hi Thomas,

You definitely don't want to merge multiple individuals into one track when the data are sampled around the same time. This causes the artifactual decorrelation that you see in the initial peaks of the variogram. You could add a long time in between individuals and merge their tracks that way.

Best,

Chris