Page 1 of 1

Choosing a Final Detector Spacing

PostPosted: Fri Nov 03, 2017 12:01 pm
by ckupferman
Hello,

I have been running preliminary simulations using the package secr for a non-invasive genetic study that will estimate fisher density. Fisher home range varies from 15 km2 for females to 40km2 for males. I will be deploying detectors in a clustered design across the landscape, with two detectors per cluster. I have reached a point in my simulations where all of my least biased cluster and detector spacing results fall within relatively small ranges (spacing between clusters ranges from 3300-3500 m and detector spacing ranges from 250-500 m). All of these outputs have given me similar density estimates. How much of a difference would these few hundred meters make when choosing a final station layout design? Is it better to choose a longer detector spacing per cluster to make each station more independent? Should I choose a cluster spacing that will provide me with the highest number of detectors to potentially increase my sample size? I would appreciate any help you can provide.

Sincerely,
Caitlin Kupferman

Re: Choosing a Final Detector Spacing

PostPosted: Sat Nov 04, 2017 2:31 pm
by murray.efford
Hello Caitlin

I'm glad you're trying clustered designs. In my thinking you should aim for multiple detections of an individual within a cluster, at a scale suitable for estimating the detection function, and few or no re-detections between clusters, leaving them more or less independent of each other. Your design does not seem to achieve that - 2 detectors per cluster will yield few within-cluster redetections, and the spacing is much too small to rely on within-cluster redetections to fit the detection function (independence within clusters is unattainable and unnecessary). (I'm guessing half-normal sigma is about 0.9 km for females and 1.5 km for males).

Answers to your other questions depend on details you don't provide. I suggest you load the new version of secrdesign (2.5.2) and consult http://www.otago.ac.nz/density/pdfs/secrdesign-tools.pdf. First make sure any candidate design gives sensible numbers of individuals and recaptures. The new function Enrm reliable predicts the sample size for any design (expected numbers of individuals and recaptures). The rule-of-thumb RSE(D-hat) is less reliable for clustered designs (I haven't yet investigated), but it still gives you an idea of the what to expect, a lot faster than simulation.

Murray