www.phidot.org

I am working with a colleague to analyse possum trapping data collected over 7 trapping sessions in 2012. The dataset includes 54 individuals with varying numbers of recaptures. I also have GPS tracking data for 11 individuals, each GPS dataset covers at least a month and 100+ fixes. It is clear from the GPS data that the bush block I was trapping only represents a small portion of the range of the collared possums. All of them were caught multiple times within the block but they spend large amounts of their time in surrounding areas. At the most basic level I am interested in getting a density estimate for the bush block but I am also wondering whether or not there is anything useful that can be done with all the movement data we have.

Thanks in advance,
Jamie MacKay

jwbm wrote:I am working with a colleague to analyse possum trapping data collected over 7 trapping sessions in 2012. The dataset includes 54 individuals with varying numbers of recaptures. I also have GPS tracking data for 11 individuals, each GPS dataset covers at least a month and 100+ fixes. It is clear from the GPS data that the bush block I was trapping only represents a small portion of the range of the collared possums. All of them were caught multiple times within the block but they spend large amounts of their time in surrounding areas. At the most basic level I am interested in getting a density estimate for the bush block but I am also wondering whether or not there is anything useful that can be done with all the movement data we have.

Thanks in advance,
Jamie MacKay

You might want to have a look at the following:

http://www.phidot.org/software/mark/doc ... chap20.pdf

Jake has done a lot of work with combining telemetry and mark-recapture (see his recent Ecology papers).

I am also wondering whether or not there is anything useful that can be done with all the movement data we have

An open question! Let's focus on the data's relevance to density estimation... You can find some simple ways to combine telemetry and capture-recapture data in 'secr' (see ?addtelemetry). However, the fact that it can be done doesn't necessarily make it worthwhile. I assume your possum population was essentially closed during capture-recapture sampling (if we're talking brushtail possums, these guys live several years). The most you can then get from the telemetry data is improved estimates of detection probability, which may have only a small effect on the precision of density estimates (this depends also on the number of individuals caught, which dominates once detection is reasonably well described). There is also a serious question regarding how representative your 11 GPS animals are of the population as a whole: if they are not a random sample they should not be allowed to drive the estimates of detection parameters (lambda0, sigma) that you apply to the population as a whole.

Understanding the pros and cons is something we are working on - expect more on this, enhanced 'secr' capability and better documentation of addTelemetry etc. around the end of the year.
Murray

Hi Murray,
Thank-you for the reply. I am talking about brushtail possums and I'm assuming that the collared possums are reasonably representative, I basically stuck collars on adult possums when I had collars available. However, if you don't think that the data will add much to the precision of the estimate then I will use the data for its original intended purpose of looking at the ranging behaviour of possums in a fragmented landscape.

The ranging data has exposed another potential issue with my data. Preliminary runs of the model have come up with lower-than-expected density estimates for the trapping grid. I've had a look at the recapture data and realised that within-session recaptures are low (mean 20% recaptures within a session, range 6-42%). Just over half the possums were trapped in multiple sessions through the year but they were generally only trapped once during each 2-4 night trapping session. I expect this is due to their ranging behaviour - they are visiting different parts of their ranges on different nights and since their ranges go far beyond the trapping grid they are not always coming into contact with traps. When multiple sessions are analysed together does the model look for recaptures of individuals in later sessions and incorporate this information into the analysis? I have caught quite a few of the possums this year as well so I know they are still alive, is this information useful for improving the precision of the density estimate?

Many thanks,
Jamie

Hi Jamie
This is interesting... The GPS data do give added insight, and I should have acknowledged that in the case of very irregular ranges an approach using 'proportion of time on grid' (PG) can be more robust, given enough data and representativeness (as in Jake Ivan's work; see also the seldom-used radiotelemetry (PG) feature in Density). Representativeness and sample size are still a worry (you would be basing a lot on a few, possibly quirky, home ranges), but no harm in peeking to see what the method produces.

Your explanation for few recaptures within sessions may or may not be right - I would have thought that organisation of activity as you suggest (movement between patches) would lead to positive serial correlation: more detections in a session once the possum was detected. Need first to show that this effect is real (your sessions are short, and we don't know the trap density, so maybe you're getting what you should expect).

When multiple sessions are analysed together does the model look for recaptures of individuals in later sessions and incorporate this information into the analysis?

No. Sessions are defined as independent (capture histories never span more than one session). You have information (from trapping and also GPS) on the presence of known but untrapped animals that is not being used. In the short term, I would be tempted to group adjacent sessions; further out, there will be new methods that use the presence information, including open population models (we don't yet have packaged versions of open SECR models).

Murray

Hi Murray,

I have come on board this analysis to support Jamie. To give you an idea of the project, we will first analyse multi-session trap recapture data only across sites and seasons with sexes, and then use model selection to determine any space or time effects. Straight forward(ish), but need to think about how to best incorporate sex (group/hcov/CL=T, etc) some more.

We then hope to run each session independently with additional GPS data (since GPS data are not yet multi-session) to see if we get more accurate density estimates in each session. Your previous comments regarding sample representativeness are well taken on board, and it may be that the GPS data add no further precision (or counter-productively add bias), but given the data are available we will give it a go and learn from the process! As you suggest a proportion of time on grid may simply be a more straight forward approach, and sometimes keeping it simple is simply best.

I'm having some GPS implementation errors on a platform with a fresh install of R3.0.2 & SECR 2.7.0, but wanted to touch on some philosophical issues first:

How aberrant do you believe incorporating GPS data in to a trapping capture history is? Obviously, GPS locations are not available when the animal is restrained, but they are available 1) on the night the animal is captured before capture and 2) on nights during trapping when an animal isn't captured and 3) on nights outside trapping occasions. One might more scientifically pose this question as "do you think the act of trapping GPS animals substantially biases the value of incorporating their GPS locations". We could discard (1) and I'm not sure how to incorporate (3), but we could use (2).

Regarding (3), is it correct to interpret that SECR can only incorporate telemetry locations on occasions where detectors were used and recorded in the capture file? i.e. we cannot incorporate GPS locations on nights where no detectors were used (all usage = 0). I can enter trap files with occasions where no detectors were used (it does give a valid error), but this doesn't pass the verify test when fitting the model.

Thanks,
James

I think your intention to

incorporat[e] GPS data in to a trapping capture history

is doomed. Trapping and telemetry are different processes, and the telemetry process does not inform us directly on what we most want from capture-recapture analysis, the number and distribution of untagged animals. You may have been misled by the addTelemetry documentation: there we use a notional telemetry 'capthist' as device for holding telemetry data so that it can be modelled jointly with true SECR data. The full documentation pdf was only half written when I had to move on to something else, but I am now coming back to it. The theory is a work in progress, too. Suggest you work through the Examples in addTelemetry.

There is no need to match GPS data to the occasions of a capthist. You can always suppress verify in secr.fit(), but here it is telling you something.

Murray