Hello Martin,
I'm working on Ring-Billed Gull habitat selection with GPS data logger technology. I read about "adehabitat" and I'm really interesting to work with this package. But there's my problem: we have some kind of irregular or "hierarchical" time lag. We get relocation every second during 1 min (60 relocations), each 4 min, 24 hours a day for 3 days. Is there a way to work with all our relocations or we must work with a 4 min time lag and loss some information? I would prefer to work with a 20-30 sec regular time lag, but it was not possible with our technology last spring. Besides, for next spring fieldwork, we are thinking about turning off our data loggers during the night. Is there a problem again with this kind of design, time lag and analysis with
I think that you have first to define precisely the aim of your study.
The best approach depends on many different things, including your issue
(including your preliminary scientific hypotheses), the spatial and
temporal scale at which you want to draw your conclusions, the
resolution of your habitat information, whether you want to perform an
exploratory or confirmatory analysis, etc. You did not give enough
information. There are many many ways to study habitat selection (which
is a rather vague objective: habitat selection occurs at multiple
spatial scales, and your data are also collected at multiple temporal
scales; etc.). For example:
First, you may choose to ignore the time dependence of the successive
relocations, and to compare the habitat use (e.g. measured by the
proportion of relocations in each habitat type) and habitat availability
(e.g. habitat proportion in the home range). Of course, this is only
exploratory (the unaccounted time dependence between successive
relocations prevents any inference relying on the independence between
successive relocations), but can return interesting results (and again,
there are many different ways to carry out such an exploratory analysis,
depending on what you do want to see).
Or/and you may want to take into account the trajectory structure. For
example, if your aim is to study the small scale moving behavior of the
gull, that you have habitat maps with a very fine resolution, it may be
sensible to define regular "bursts" of relocations, each burst covering
one minute with one relocation every second, and then to relate e.g. the
speed of the animal, the turning angles between successive moves, etc.
with environmental variables. Or, if there is temporal dependence
between successive moves of 1 second, you may also partition each
trajectory (i.e. each burst of relocations) into "types of behaviour"
(e.g. using modpartltraj, or any partitioning method existing in the
literature), and relating these types of behaviors to the habitat (with
some kind of discriminant analysis).
Or, if you want to study the movements of the animals at the scale of
the day and if you do not want to consider the timing of the
relocations, but just the shape of the trajectory of the animal, one
solution could be to define an "ltraj" object with each burst covering
24 hours, and then to rediscretize the trajectory into regular steps
(see the function redisltraj), and finally to analyse how the turning
angles and rediscretized relocations are related to the habitat. Or, if
you want to take into account the time, you may consider the first
relocation of each burst of 1 min in defining bursts of 24 hours with
one relocation every 4 min, to consider the timing of your relocations.
Or, you may...
What I am trying to say by giving these untidy examples is that there
are many many possible different ways to analyse habitat selection with
your GPS data, and that the best depends on your issue. But, above all,
that the most sensible analysis is not necessarily the method that will
allow you to take into account all your relocations. The analysis
approach depends on your question. You can use all your data or not, but
this is not, IMHO, the most important point. If you delete some
relocations, it will result into a loss of information, but this
information may not be relevant for your question. First define your
question, and then, given your data, define the most sensible analysis
approach. The chosen approach may imply the loss of 80% of the original
information to build a relevant dataset. Keeping information not
relevant to the question adds noise to the data and should be avoided.
HTH,
Clément Calenge
--
Clément CALENGE
Cellule d'appui à l'analyse de données
Office national de la chasse et de la faune sauvage
Saint Benoist - 78610 Auffargis
tel. (33) 01.30.46.54.14
_______________________________________________
AniMov mailing list
[email protected]
http://lists.faunalia.it/cgi-bin/mailman/listinfo/animov
