Data formatting and session versus occasion

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

Data formatting and session versus occasion

Postby martenmore » Fri Nov 18, 2011 11:18 am

Dear All,

I am hoping to use Density to calculate density/abundance of pine marten in a study area that has several discrete habitat parcels. The basic study design has 125 regularly (500m) spaced hair tubes which are sampled 6 weekly intervals between June and November. During each sampling period bait is put in every tube on day 1 and on day 10 hair samples are collected for DNA analysis of individual identity.

Just a few basic enquiries with some issues I am having, particularly related to capture data formatting. In terms of sessions and occasions it appears that for the majority of studies both are different - occasions often refer to daily sample collection. In my study, there are 4 sampling periods (sessions?) and 4 hair collection periods (occasions?), each 10 days long. Therefore, in my case both are the same and should be coded 1-4?

The second issue is related to capture data format. My data includes recaptures during each collection period (e.g. 1 animal can be found in as many as 5 tubes). Density seems to drop duplicate capture records which I assume corresponds to these within session recaptures. I have had a look at some example datasets and many appear not to contain such data - do these recapture records need to modified prior to analysis in Density or I am doing something wrong here.

Any help anyone can provide would be very much appreciated.
martenmore
 
Posts: 3
Joined: Thu Nov 17, 2011 1:06 pm

Re: Data formatting and session versus accasion

Postby murray.efford » Fri Nov 18, 2011 3:18 pm

The terminology gets confusing alright. As I understand it you have 4 samples (each corresponding to a 10-day period) collected over 6 months. The real question concerns pine marten biology, and whether the population was closed to births, deaths and movement of home ranges over the entire 6-month study: I presume not. This requires you to treat the samples as separate ‘sessions’ in Density. Sessions may be numbered 1-4 or given names. Even if the composition of the population changed, it may be reasonable to assume _parameter_ constancy over sessions in a between-session model; this should improve precision.

The next problem is how to estimate detection parameters from a single sample (=occasion) per session. Estimating detection parameters by capture-recapture would ordinarily require repeat samples (multiple occasions in each session). However, with spatial sampling using ‘proximity’ detectors such as yours, the multiple records for some individuals within a session can do the job (see e.g. Efford, Dawson & Borchers 2009 Ecology 90:2676). The catch is that Density throws away repeat records at the same place and time, even for ‘proximity’ detectors (you will get a message). You can still estimate density, but it’s painful to lose that fraction of the data.

The way around this would be to use the ‘count’ detector type in the R package ‘secr’. This models the number of detections per animal per detector per occasion (0, 1, 2, …) rather than just whether an animal appeared or did not appear (0, 1). You can expect somewhat tighter confidence limits if there are many repeats at a site. [As an aside, this is not always good news: ferrets seem to become trap-happy once they learn about a particular baited site, and maybe pine martens are similar. I think this is difficult to model, although Beth Gardner has attempted it. It may be that ‘proximity’ detection is more robust].

In summary, I think you’re on the right track, but maybe need to set the detector type to ‘proximity’ in the Trap layout box.

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

Re: Data formatting and session versus accasion

Postby martenmore » Wed Nov 23, 2011 7:17 am

Hi Murray,

Thanks for getting back to me on this, your input is very useful. In terms of pine marten biology and potential closure - it could be argued that this was the case. This years juvenile marten would have been present in the population during June (perhaps not very 'available' to enter tubes in first survey due to size), any deaths are likely to have been very small (total population size is likely to be small), pine marten home ranges are relatively stable during this time of year (from radio-tracking). If this is the case then may it be possible to treat samples as a single session in Density and generate an estimate?

Your help in terms of Density throwing away repeat records at the same place and time was also very useful. To address this I have actually perhaps beat Density to this and coducted an analysis that does not include repeat records at same place and time (painful as you suggested) and got reasonable (if relatively high) density esitmates of 1.2/km2.

Unforunately I have not deleved much in R (something I need to address) and am hoping to use Density 4.4 for calculations even though R has more capabilities.

To sum up then if I use Proximity as a detector type, perhaps loose repeat records at same place/time then Density 4.4 should produce a relatively reliable estimate?

Thanks again for the help and input.
martenmore
 
Posts: 3
Joined: Thu Nov 17, 2011 1:06 pm

Re: Data formatting and session versus accasion

Postby murray.efford » Wed Nov 23, 2011 4:39 pm

I can see there is a case for treating the population as closed over that time, and I would be tempted, too, but it will ring alarm bells for some readers. At least you should check the more rigorous alternative to see if it gives adequate precision (fitting a model with constant density across the 4 sessions, but doing single-occasion analysis within each session).

Regarding R and count detectors: I don't think you're missing much. 'Density' automatically discards extra detections at a site, so you don't need to change the data file. Is the end result reliable? You will have fitted a model that from this distance seems appropriate, estimates from these models are fairly robust, and bias from misspecification of the model is probably an order of magnitude less than sampling error. [Erroneous high density estimates can result when you forget to increase the buffer from the default 100 m to something appropriate to your species - certainly >1000 m]

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

Re: Data formatting and session versus accasion

Postby martenmore » Thu Dec 01, 2011 7:42 am

Murray,

Thanks very much for this information and sorry for the disjointed nature of this chat - I am in the middle of fieldwork and out of the office most of the time.

So following your last advice i have used all data (with Density dumping repeat records at same place and time) with proximity detectors and 1000m buffer in an ML SECR within session model using default settings. Individual session mean density estimates range from 0.37 -0.57 marten per km2 which intuitively seems reasonable based knowledge of pine marten biology here.I also ran a between session model and individual session estimates were also good.

Finally a bit of help would be appreciated in terms of fitting a model with constant density across all sessions - I have had a look at this and am not sure how to initialise - maybe its staring me in the face!!!

Thanks again for all your help and advice.
martenmore
 
Posts: 3
Joined: Thu Nov 17, 2011 1:06 pm

Re: Data formatting and session versus accasion

Postby murray.efford » Thu Dec 01, 2011 11:25 pm

Go to the 'Between-session model' tab on the ML SECR options page. The default 'Constraint' for Density is 'Session', meaning a distinct level of density will be fitted for each session. Double click on the cell to cycle through options until you get 'Constant'. (You should also get a drop-down menu by right-clicking on the cell, but that requires the whole cell to have focus). Make sure you remember to select 'Use between-session model'.
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