# [tlocoh-info] tspan

Andy Lyons ajlyons at berkeley.edu
Mon Mar 3 17:26:07 CET 2014

```Hi Horst,

Thanks for your question about the hull metric 'tspan'. You are correct
that it isn't obvious (or well-documented) what this hull metric is all
about or the units. Basically, it is the time-span of the nearest
neighbors of a hull, expressed as a multiple of the median sampling
interval. So for example if the median sampling interval for a given
trajectory is 3600 seconds (1 hour), and a particular hull selected from
that dataset was constructed from the parent point and 10 nearest
neighbors, then the tspan value for that hull would be duration of time
in seconds between the earliest and last location used to construct the
hull (including the parent point), divided by 3600. If there are no time
stamps associated with the locations, tspan will be NULL.

When / whether this is a useful hull metric depends on the data and the
question. It could be a measure of the duration of time an individual
was present in a given hull. This could be useful, for example, to see
which areas of the habitat were used for a short period vs. which were
used over a longer period of time. This will of course be sensitive to
the length of data collection as well. A related hull metric that
measures duration of time in an area, but which takes into account the
notion of a visit, is the mean number of locations per visit (mnlv),
where a 'visit' is defined by an additional parameter, the
inter-visit-gap (ivg) period.

tspan can be used to see the effect of incorporating time into nearest
neighbor selection. When time is included (e.g., s>0), you would expect
the distribution of tspan values to shift to the left (i.e., get
smaller), because points further away in time get bypassed as nearest
neighbors. That is essentially what the function lxy.plot.tspan shows.

You can check the hull metric value of tspan for an arbitrary hull using
the code below. This code illustrates how to extract information from a
version 1.15, updated about two weeks ago, has a vignette about the data
structure of a locoh hullset object). Given a hullset object named lhs,
you could compute tspan of the ith hull in the first set of hulls by:

i <- 5

## Get the indices of the enclosed points of the ith hull
ep.idx <- lhs[[1]]\$enc.pts\$idx[[i]]

## Get those enclosed points that are also nearest neighbors
nn.idx <- ep.idx[lhs[[1]]\$enc.pts\$nn[[i]]]

## Get the time in seconds of the date stamps of the nearest neighbors
of the ith hull
dt_secs <- as.numeric(lhs[[1]]\$pts\$dt[nn.idx])

## Compute max - min
dt_secs_range <- diff(range(dt_secs))

## Get the median sampling interval(also saved in seconds)
tau <- lhs[[1]]\$rw.params\$time.step.median

## Express the time span as a multiple of the median
dt_secs_range / tau

## See if this is the same as the saved value for tspan
lhs[[1]]\$hulls at data\$tspan[i]

In general, the function hm.expr() will display a list of the hull
metrics that are supported in the package, and a short description of
each. Please keep sending questions like this which highlight where the
documentation needs some more work. I am also hoping to write a vignette
on how to use hull metrics and code up new metrics, because its the
hulls and their metrics which gives tlocoh the ability to generate
analytical insights beyond traditional utilization distributions.

Best,

Andy

On 2/28/2014 12:36 AM, Horst Reinecke wrote:
> Hello list,
>
> I'm working with tlocoh  and data of red deer. I'm uncertain how to
> interpret the tspan data because I don't know neither the unit nor how
> tspan is caluclated.
> Any suggests?
>
> Horst
```