[AniMov] kernelkc questions

Corrie Curtice corrie.curtice at duke.edu
Wed Nov 7 13:52:41 CET 2012

Hello AniMov folks.

I'm using and learning the new kernelkc method implemented in adehabitatHR,
from Keating and Cherry.   I'm hoping someone can help me better understand
the smoothing parameters, so I can successfully use the tool for a project
I'm working on.  I can provide an r script and workspace with a subset of
our data if someone is interested and willing to take a look and provide
guidance. Here is a project summary:

I have data for several humpback whales tagged in Antarctica with Argos
tags.  Duration is several months for each, and they range along the Western
Antarctic Peninsula.  The tags are programmed to transmit for 2 four-hour
periods during the day (00:00 ‹ 4:00 & 12:00 ­ 16:00).  I pre-filtered to
remove duplicates and ran them through the Freitas (2008) SDA filter to
remove unlikely locations.

I'd like to use the kernelkc method to show area use / movement at two time
scales for each whale:
1. Weekly-ish movement over the whole time period and how it changes through
the season - does the amount of area used change weekly? Monthly?
2. Finer scale movement over a single day or transmitting period, for any
periods that have enough relocation points at that scale.
I'm starting w/ a single burst of a single track, to gain experience with
the tool (https://www.dropbox.com/s/qz86jwao8yskfiz/Figure1.pdf). The ltraj
is not "regular" as the locations are not equally spaced in time
(https://www.dropbox.com/s/x9tqf5s7870o8mg/Figure2.pdf), but I don't believe
this should impede the kernelkc process.  The whale can be seen in one area
in March, then moving up the along the coast of the peninsula during April,
then staying in another area in May

I'm having difficulty determining the smoothing parameters when including
the time dimension. When I choose to execute the kernel for each day of
observations, for some days the result is a single color grid and the
getverticeshr fails (95%) with this error:

Error in plot(getverticeshr(uu, 95, standardize = TRUE), lwd = 2, border =
"red",  : 
  error in evaluating the argument 'x' in selecting a method for function
'plot': Error in re[[i]] : subscript out of bounds

I assume it's related to periods of time that are sparse in locations.
Increasing or decreasing the smoothing parameters (x,y, t) did not solve the
issue, so we backed off the sequence of dates to run the kernel on every 2nd
or 3rd day, as below.  Is this an appropriate solution or are there other
more appropriate options?

max(t699_b2[[1]]$date) - min(t699_b2[[1]]$date) # 85 days duration
#so what's the appropriate sliding window time period?
## let's try every 2-3 days,
vv <- seq(min(t699_b2[[1]]$date), max(t699_b2[[1]]$date), length=35)

re <- lapply(1:length(vv), function(i) {
  uu <- kernelkc(t699_b2, h = c(5000,5000,3*3600*24), tcalc= vv[i],

I've read the manual and the vignette (very helpful!) but still have some
* When the kernel is executed at a particular point in time (tcalc), how
does the time smoothing parameter influence the points considered for the
algorithm? In other words, if I pick intervals of 10 days for my tcalc
vector and have a 3 day smoothing parameter, would I be ignoring some days
of data?
* Because our data is so spatially sparse, the resulting UDs have a large
grid extent but very tiny cell areas of density ­ is there any suggestion
for addressing this, to make the densities a bit more visible? We will
eventually plot the density and home range (probably 95% but perhaps 50%
core area if available) over a map so we can show movement in relation to
land or perhaps ice cover. (Series of images here:
One of key things we'd like to learn from this data set is improvements we
could make to future tag programming, to produce data that will more easily
show moving home ranges over time.  Should we turn the tag on for shorter
durations but more times during the day (4 sets of two-hours "on")?  One
nice thing about polar research is good satellite coverage!

Thanks for any help and guidance you may offer.  We are already really
excited in seeing these preliminary results and are eager to apply it to our
whole dataset.


Corrie Curtice
Ari Friedlaender 
Dave Johnston

Corrie Curtice
Research Analyst
Marine Geospatial Ecology Lab
Nicholas School of the Environment, Duke University
em: corrie.curtice at duke.edu
ph: 252-504-7538
cell: 978-857-8266

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.faunalia.it/pipermail/animov/attachments/20121107/0cee37f2/attachment.html>

More information about the AniMov mailing list