Standard error of probability contours in relation to lscv Kernel Density Estimation

[Sarah Woodward]

Hi,

I have been using OpenJUMP HoRAE to calculate lscv kernel densities for some bird telemetry data.  I used my KDEs to create probability contours to estimate home range at 50%, 75% and 95% probabilities.  I have had very good success with this program and I really enjoy its ease of use.  My question is regarding standard error.  Is there a way to display standard error in relation to the size of the polygons that were created by my probability contours?

I appreciate any advice you might be able to provide!

Thanks,

Sarah

[Stefan Steiniger]

Hi Sarah,

nice to hear from you and that you like OJ HoRAE.

With respect to your question: I am thinking about your email since
reading it yesterday (well.. almost ;). And I am not having an answer.

The KDE methods deliver the density as a "terrain"/raster image from
which we can create a contour for every height-value we want, i.e.
translated: for every type of density or probability value.

Problem is: I am not sure where to get the error from and if creating
confidence values makes sense. It could be done for sure if the true
distribution would be know, but we don't have this.

I also haven't read one home range paper in which uncertainty boundaries
are displayed/calculated ... maybe in some older one from the 90ies? If
you can find one, please let me know so I can think about it further.

If you have further ideas/insights on the topic, let me know.
cheers,
stefan

Am 03.12.13 17:39, schrieb Sarah Woodward:

> --
> You received this message because you are subscribed to the Google
> Groups "OpenJUMP HoRAE Users" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ojhorae-user...@googlegroups.com.
> To post to this group, send email to ojhora...@googlegroups.com.
> Visit this group at http://groups.google.com/group/ojhorae-users.
> For more options, visit https://groups.google.com/groups/opt_out.

[Andrew Hunter]

Hi Sarah, Stefan,

The simplest way to do this might be to add functionality to the tool that allows CI of kernels to be determined. There are a couple of functions in R, asciker and bsciker, that use asymptotic and bootstrap approaches, respectively, to do this.

I have attached a paper by Fiorio that outlines how this can be done using R.

Cheers

Andrew

[Stefan Steiniger]

Thanks Andrew,

Error estimation via bootstrapping/resampling is an idea. Not sure
though, how much time it would take to implement it. Interestingly the
point KDE uses kind of re-sampling when asymptotes are calculated, so
its a bit like changing to a different sample set...

will think about,
stefan

Am 04.12.13 18:17, schrieb Andrew Hunter:

> Andrew Hunter PEng RPSurv PhD
> Assistant Professor & Associate Head (Undergraduate Studies)
> Department of Geomatics Engineering
> Schulich School of Engineering
> University of Calgary
>
> T: +403.220.7377
> F: +403.284.1980
> E: ahunter (at) ucalgary (dot) ca
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
>
>
>
>
>
>