Animal Movement - Mobility Data - page 278

allows one to reduce the sample size, which, however, should be representative
of the variability within the population of interest. In case of GPS devices, there
exists a trade-off (given a certain load allowed by organism's size) between the
number of spatio-temporal positions and the duration of the survey. A com-
promise is often used, adopting a standard low frequency of spatio-temporal
positions (typically one per 4 hours) but programming a denser collection of
spatio-temporal positions during a specific period of interest (breeding time,
natal dispersal, etc.).
There is a basic difference in sampling with VHF and GPS. In the former case
one tends to use independent spatio-temporal positions, while in the second case
there is a specific aim to exploit the autocorrelation among positions to deduce
the movement tactics used by the organism. In the case of VHF telemetry the
interval among spatio-temporal positions is large (e.g., at least 24 hours) to
guarantee a certain degree of statistical independence, and the time of location
shifts from an occasion to the next one (usually one or two hours) to cover the
whole 24 hours (often sampling is stratified by month or season). However, this
sampling approach makes difficult the analysis of trajectories. On the contrary,
the usual sampling with GPS is to compute a location at fixed times of the day
(e.g., midnight, 2, 4, ... , 22) with an interval between spatio-temporal positions
usually ranging between 2 and 6 hours. In general it is preferable to sample
at fixed time so as to yield a good estimate of animal's speed. Modern collars,
which are going to be on sale soon, have internal capabilities for analysis of
geographical data. This would allow for adaptive sampling schedules, the merits
and potentialities of which are not yet well understood.
Many methods are available for uncertainty reduction and to get a precise
trajectory reconstruction . The causes of imprecision are reviewed in Chapter 5 .
In animal movement studies, fuzzy sets are not used for correcting spatial uncer-
tainty, and state-space Bayesian models are becoming popular in the ecological
literature. Independently of the method used, a filtering of the trajectory is usu-
ally necessary. It is useful to think to a hierarchical analysis: first-order analyses
are intraindividual and so include spatio-temporal position correction, path inter-
polation, computation of angular resultants, mean speed of displacement, and so
forth. The second-order analysis refers to the sample and in this case statistics
are relative to the population on which to make inference, whether individuals
are sampled independently.
13.2.5 Analysis of Animal Trajectories
Turchin (1998) represents the reference text in this field. Let us suppose we
are recording the true trajectory or path of an animal. We may represent the
path as a set of vectors connecting spatio-temporal positions. In the first case
(Figure 13.1 a) our sample overlaps the walked path at a large extent, while in
the second case (Figure 13.1 b) our representation is much coarser and important