## Different density results on grouped transects

questions concerning anlysis/theory using program DENSITY and R package secr. Focus on spatially-explicit analysis.

### Different density results on grouped transects

Hi,
I'm running density 4.4.4 on win 7. MRC data from mouse lemurs in n. madagascar.

The question:
We created 6 transects, spaced into 2 groups of 3 transects. Along each transect we placed sherman traps at 20m intervals. Within each group the transects were 200meters apart, and between each group they are 700-1000meters apart.
If i create a trap layout file with both groups of transects (so a wide 700-1000 metre gap in the middle, with 3 transects to each side) i get lower results then if i calculate each group of transects individually (a trap layout file with only three transects, no gaps) and then find the mean of the two groups. Why is this? Is SECR less accurate with larger open spaces and "clusters" of information? And would the second method, of treating the groups individually, give a more accurate representation? The second results, from averaging the density of both groups treated individually seem more probable having been in the field.

Thanking you again!
Sam
sam.meyler

Posts: 2
Joined: Thu Oct 21, 2010 8:02 am

### Re: Different density results on grouped transects

Sam

From what you say I cannot fault your design and the two approaches to analysis should yield comparable results, given adequate sample sizes (I'm assuming there were no recaptures between transects; it would help also to know the number of traps and the recapture distances of the lemurs). Efford et al. Wildlife Society Bulletin (2005) give a comparable example with brushtail possums in New Zealand.

A risk with widely dispersed trap layouts is that the default integration mesh gets quite coarse around any one trap cluster, causing artifacts in the integration of the likelihood, and I suspect this is where things went wrong. You can overcome this in Density by increasing the number of mesh points in Options | Computation (the R package secr gives better control and more options, and the coming version 1.5 as further improvements - beta is on website).

There is a trick that allows you to specify a much finer mesh across a wide area without incurring an impossible processing burden: save the mesh to a file with Tools | Export | Coordinates of mesh and after giving a filename specify a criterion for discarding points far from any trap (default 100 m - should match your intended buffer width). Then go to Options | Computation and choose to import the resulting text file rather than specify a mesh directly.

Using clusters of traps has some big advantages for study design, including probability-based sampling of large areas and the potential for empirical variance computations (empirical.varD in 'secr'), rather than reliance on a Poisson model. Unfortunately, neither your study nor ours really has enough clusters for this to mean a lot!

Hope this moves you forwards
Murray
murray.efford

Posts: 614
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand