Trap shyness and learning

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

Trap shyness and learning

Postby MWhitehead » Wed Nov 23, 2011 12:52 am

Hi all,

I was recently spurred to try out secr in R after finding a response to my initial query in this thread: viewtopic.php?f=1&t=1903

Just to reiterate on my sampling and study, I have carried out a mark recapture study of solitary wasps using pheromone baits to attract individuals. In 18 trapping days (occasions), the first 9 occasions were mark and recapture at 9 fixed baiting points (detectors). The following 9 occasions were random spot samples throughout the site to recover already marked individuals.

The reason for the dual sampling strategy is connected to my two main objectives for secr analysis:
- To derive population density estimates
- To look for evidence of learned avoidance to traps post capture
(Home range size would be a bonus and looks like secr can help me here too)


I am interested to garner some thoughts on how I might make a more nuanced model of learning than the
g0 ~ b model which I have tried.

I understand the g0 ~ b models a step change in detection probability of individuals after first detection. However a more realistic model of wasp learning would be one where the step change in detection probability only reduces recapture for the specific trap location of first detection, with subsequent detection probabilities remaining constant for all other traps.

Is there a way that sessions or covariates could be used to achieve this type of model?

Thank you in advance for any help!


***

In case you're interested I've run with and without learning on the full dataset as well as just the first 9 occasions. There is no difference in models for the first 9 occasions. This leads me to believe that the better fit of the learning model for the entire study is confounded by switching sampling strategy.

FIRST 9 OCCASIONS ONLY
model detectfn npar logLik AIC AICc dAICc AICwt
Mark0 D~1 g0~1 sigma~1 halfnormal 3 -607.3535 1220.707 1220.757 0.000 0.701
Markb D~1 g0~b sigma~1 halfnormal 4 -607.1889 1222.378 1222.461 1.704 0.299

ENTIRE 18 OCCASIONS
secrb D~1 g0~b sigma~1 halfnormal 4 -1191.533 2391.066 2391.147 0.000 1
secr0 D~1 g0~1 sigma~1 halfnormal 3 -1325.065 2656.130 2656.178 265.031 0
MWhitehead
 
Posts: 5
Joined: Mon Jun 27, 2011 9:35 pm

Re: Trap shyness and learning

Postby murray.efford » Wed Nov 23, 2011 4:16 pm

Michael

1. The software does not have a model with a trap-specific learned response, interesting as that would be. In fact, I put one in originally but pulled it early on when I came to suspect it was not identifiable. It at least needs work before it goes out to innocent users. A trap-specific learned response implicitly specifies g0 as a joint function of individual, occasion and site: this is not something that can be done using individual, occasion and site covariates separately.

2. The difference between your 9-day and 18-day model comparisons is large, which leads me to suspect an error in the formulation. This might happen if you entered each of the sites sampled on days 10-18 as if it was active for the whole 9 days, and in fact each was sampled on only one day. Selective use of sites can be coded via the 'usage' attribute of the traps object, which can be entered along with the detector coordinates. For example

A 123 456 111111111000000000
B 234 567 111111111000000000
.
.
.
a 123 567 000000000100000000
b 234 567 000000000010000000
c 123 678 000000000001000000
etc.

Here sites with ID in caps were used for the first 9 days and no more; sites with ID in lower case were each used for a single day.

3. As I understand it, no marking was done in the second half of the study, so this really is a mark-resight design as I think you said originally. I expect the error from fitting a mark-recapture model might be quite small in your case; this could be determined by simulation. There is a mark-resight formulation of SECR waiting in the wings for a good data set. Contact me offline if you want to follow up on this.

Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Trap shyness and learning

Postby murray.efford » Thu Dec 22, 2011 3:26 pm

1. The software does not have a model with a trap-specific learned response, interesting as that would be. In fact, I put one in originally but pulled it early on when I came to suspect it was not identifiable. It at least needs work before it goes out to innocent users.


Updating my earlier response: Models 'bk' and Bk' for permanent and transient trap-specific behavioural response are included in secr 2.3.1. My doubts about identifiability were misplaced - the implementation for competing detectors (i.e. traps, detector = 'multi') just demanded more time and thought than I had committed to my first attempt.
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand


Return to analysis help

Who is online

Users browsing this forum: No registered users and 12 guests

cron