Linear habitat and secr

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

Linear habitat and secr

Postby CHWarbington » Fri Aug 09, 2019 7:27 pm

Hello -
I am struggling with how to salvage an analysis!
I am trying to get a baseline density estimate of sitatunga. My study animal is a wetland specialist, that seldom forage on dry land habitats adjacent to the wetlands. I expect their activity centres to lie within wetlands, thus have created a habitat mask for available habitat. There is no information about their movement rates, thus I included in the mask a 5km extension from the ends of the trapping array to hopefully close the state space (negligible chance to capture those outside of the area). However, these are riverine wetlands, and thus are linear in nature (our camera trapping array was about 10 km long x 0.5 km wide) surrounded by dry lands. In order to maximize detection, the trapping array we used was oriented along the river, which I understand now can bias density and sigma estimates.
I have fit a null SECR model, which (according to plotting the predictDsurface) shows no variation in density in the habitat mask, and a very high density estimate of 16 / km^2. As I understand it, if the (mostly)linear trapping array and the home ranges are elongated in the same direction, then we risk severe bias in the parameter estimates. Now I am wondering what the next step should be. Is there a way to check if the trapping array is two dimensional "enough" to use isotropic detection models? Since sitatunga can use non-wetland habitat, should I use a buffer around the traps instead of the strict habitat mask? Are anisotropic detection models worth pursuing, since the bias may not be removed? Is there another option?
Thanks in advance for any advice!


**EDITED to add (in case it matters): I ran a model using a buffer of 3500 instead of using the mask, and the density estimate is much more reasonable (4.5 / km^2). The suggest.buffer command on both models is very similar (3319 for the version with the habitat mask, and 3371 for the buffered version).
CHWarbington
 
Posts: 11
Joined: Fri Aug 09, 2019 5:40 pm

Re: Linear habitat and secr

Postby murray.efford » Sat Aug 17, 2019 7:09 pm

Trying to think this through...
You seem to have an intermediate case where neither 2-D nor 1-D model is entirely appropriate. Given the linearity of the habitat and the large extent of movements compared to its width, I'm tending to think it should be seen as linear, and density should be expressed per km of river. This leads to the linear models in 'secrlinear', and avoids most of the other headaches. You would project the camera locations onto the centreline.
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Linear habitat and secr

Postby CHWarbington » Wed Aug 28, 2019 4:15 pm

Hello again!
Thank you so much for your reply.
I have modified my mask and traps file to be linear, and am working through the secrlinear vignette.
I am encountering errors when I run secr.fit, and not sure how to address it :

> fit1DFull <- secr.fit(fulldata, mask = rivermask, trace = FALSE,
+ details = list(userdist = networkdistance))
There were 50 or more warnings (use warnings() to see the first 50)
> warnings()
Warning messages:
1: In valid.userdist(details$userdist, detector(session.traps), ... :
replacing infinite, negative and NA userdist values with 1e10

All of the warnings are the same. I checked the traps and mask files, and there does not seem to be any reason why this error crops up more than 50 times. I have tried a more specific userdist:
details = list(userdist = networkdistance (traps, rivermask))
With the same error warnings.

Could this be a spacing issue? I loaded the mask as follows:
riverSLDF <- rgdal::readOGR(dsn = "D:/Student Workspaces/Warbington/GIS", layer = "UGRivers_Clip")
rivermask <- read.linearmask(data = riverSLDF, spacing = 50)

But the traps are not limited by the same spacing. I have 29 "traps", so I understand that 29 userdist values could be zero, but not 50+

Could the mask be too large? I included all line segments within 20km from any trap point due to the potentially large movement of the species.

thank you for your time any any assistance you could provide.
CHWarbington
 
Posts: 11
Joined: Fri Aug 09, 2019 5:40 pm

Re: Linear habitat and secr

Postby murray.efford » Wed Aug 28, 2019 6:39 pm

Computation of distances through the network is going haywire, presumably because the network is misspecified in some way (including some gaps or disjunctions, it seems). It's a while since I did this myself. Thoughts: look at the representation of the network to find the problem, try the showpath function and other aids in the vignette, precompute the distance matrix, examine that, and use it as input in details$userdist.

Points on a linear mask can be closely spaced without inflating computation time much.

If all else fails, send me the data and I'll diagnose, but that may not be the fastest!
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Linear habitat and secr

Postby CHWarbington » Thu Aug 29, 2019 4:42 pm

After much closer inspection, there were a lot of oxbows and river fragments in the mask that were not connected to the main channel where the traps were located. Once I deleted these, the errors have also gone.
Thanks for all of your help!
CHWarbington
 
Posts: 11
Joined: Fri Aug 09, 2019 5:40 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 10 guests