SECR for Citizen Science

[aleolivieri]

Hello,

I'm looking into comparing the ability of Citizen science data against camera trapping to obtain leopard density estimates within the boundaries of a national park in Tanzania.

I've run DENSITY successfully for the camera trap grid, my only question in this regard is if the DENSITY interface allows me to plot the effective sampling area, the detection function, and the temporal variation (I'm considering the wet vs. dry season) directly. I see that DENSITY should allow me to run the commands in the R interface but I'm not having any luck with R, as R won't run altogether.

My other question is regarding the CS analysis. The data has been collected only along roads by tourist guides so I'm not sure any trap ID accounts for "mobile" detectors.
I'm not sure that dividing the CS area with the same camera-trapping grid spacing I've used is appropriate to run a SECR model as there would be a large amount of grids with no captures recorded simply because there are no roads in that particular grid, not because there aren't any leopards. On the other hand, I was thinking of creating a grid only around the road network (i.e. divide all the road network in sampling areas of 2.2km2 as my camera trapping grid, spacing of 1.4km) and allocate all the recorded sightings to a "central" point to act as my "fixed" detector. I'm not sure however if this approach is appropriate as the animals aren't constrained to the road network and if I should try to obtain linear density which I could then convert into area density by applying a non spatial CMR.

Thank you,

Ale

[murray.efford]

Hi Ale

You have a lot of issues. I'll try to address the key ones.

1. Density is old software and for even mildly complex problems you are right to move to R. The R 'interface' in Density is really only a starting point that will indicate the R commands corresponding to the current settings in Density. You should use them to develop a script to use in R, perhaps via the R-Studio interface. I think you'll find that most satisfying once you get started in R.

2. The effective sampling area in spatially explicit capture-recapture is a number, not something you can map. To plot a detection function in R simply plot the fitted model. Temporal differences (seasons) can be addressed in R by treating the seasons as 'sessions', given that 'session' may appear in a model formula.

3. I think your ad hoc idea for dealing with roadside data is unsafe. I suggest you divide the roads into short (500-m?) segments (like extra fixed camera sites) and include the roadside records as additional 'captures' at these sites. Because effort in each segment will follow different rules to the fixed sites you should include site type as a 2-level trap covariate in any model. The 'snip' function of secr allows you to split a set of linear tracks into equal bits, but some juggling is needed to 'reduce' track data to discrete (proximity) detectors and the process is not well documented (something I will fix in a few weeks) - you may find it easier to assign photos to segments some other way.

Murray

[aleolivieri]

Hi Murray,

Thanks for your comments and help.
Seems like my understanding of the sampling area was right, I just let myself get confused by other sources.

I will take a look at the 2 level trap covariate and see if that works.

Thank you,

Ale