CMR and Occupancy data

[sixtystrat]

I’m involved in trying to develop statewide monitoring programs for black bears in 2 states. We’ve been collecting hair sampling data on 3 areas in Louisiana for several years now and I want to help them put together a long term monitoring program. In Florida, they are just on the cusp of starting a new mark-recapture program (last one was 10 years ago) to monitor their statewide population.

I am interested in the paper by Conroy et al. 2008 (Ecology) on 2-phase sampling because I want to monitor abundance in the primary areas inhabited by bears but I also want to track abundance and range expansion in the peripheral areas. This method was interesting because I thought I might be able to use bait-stations (baits suspended from trees with claw marks on the trees as evidence of detection) distributed in primary and secondary areas to estimate occupancy. I would use CMR based on hair DNA to estimate abundance on the primary areas. I would then take advantage of the relationship between detection probability and abundance on the primary areas to extrapolate abundance to the peripheral areas as described in the paper. (My thought was that, at some point, maybe only bait-stations could be used without CMR being necessary in the primary areas every year).

At any rate, the Microtus example used in the paper is different from my situation. First, I wanted to do the 2 phases in reverse, since I already know where the bears are and are not (CMR already done in LA and we know where the bears are in FL from work 10 years ago). Second, I will not be able to do all the CMR sampling at one time in Florida. There, I will be limited to CMR sampling at 1 area at a time and will probably have to use robust design methods because closure probably isn’t met. So it is going to take many years to estimate abundance on all the primary areas (I was thinking I would refine the relationship between detection and abundance as the primary areas are surveyed). However, I think that annual statewide bait-station surveys are do-able in both states. Lastly, the bears are probably distributed more patchily than the Microtus were and we will not be able to conduct many CMR studies. Do you think the Conroy et al.'s approach would work in my 2 situations?

Joe Clark

[murray.efford]

Joe
A couple of comments
- I think if you read the Conroy et al paper closely you'll find it's not actually about occupancy
- their approach neglected the problem of spatial sampling, and the problem probably should be addressed with spatially explicit models
- efficient composite designs (single points + multiple point arrays) are quite feasible with spatially explicit capture recapture, but not yet published in detail
Murray

[sixtystrat]

Hi Murray. Yes, I thought about spatially explicit sampling also. My hope is that if I could estimate N (or as you suggest spatially explicit density), and collected detection data at the same time, we could use that relationship to predict density in outlying areas where only detection data are collected. Can you suggest some literature? Thanks!
Joe

[murray.efford]

Joe

The basic idea of estimating a detection function with one set of samples and applying to other samples taken using the same devices can be implemented in existing software ('DENSITY', 'secr') using the conditional likelihood options. It's been used in distance sampling by e.g. Buckland et al. J Appl Ecol 2006 43:377-384 and Marques et al. 2009 J. Acoust. Soc. Am. 125:1982-1994. The idea of combining data from different detector types is more tricky: you can probably assume a similar scale of detection for the two (so long as baits do not attract from a distance), but with existing code you might need some sampling overlap for calibration of the intercept (g0).

It's also perfectly feasible in the existing software to fit a full-likelihood model simultaneously to both the intensive and extensive components, which has the advantage of allowing covariates on density.

(Being pedantic again: I'm trying to get away from the idea that spatially explicit capture-recapture is just for density: it also gives better estimates of N).

Perhaps we need to get this off the Presence page!
Murray

[sixtystrat]

Makes sense Murray. One last question: For the analysis you suggest, does it matter that the extensive detectors will not be identifying individuals by DNA (like the intensive detectors will), just recording a bear visit or not?

I'll get off the Presence site! (we're probably driving Darryl nuts).
joe

[murray.efford]

Mixing presence/absence data and SECR data is slightly novel i.e. unpublished. There's some undocumented code in 'secr' for this, and it works. I just need a little time...
Murray

[darryl]

Hi Joe
It takes a lot to drive me crazy.

Murray's suggestion of considering SECR is a good one as in your case you want to estimate how many individual home ranges overlap a bait station, and also how many bait stations get encompassed within an individuals home range. Is the sampling intensity (ie bait stations per ha) going to be the same in the primary and secondary areas? Are you going to be using hair-traps in addition to bait stations in the primary areas?

You also have to assume that any relationships/patterns that you observe in the primary areas also holds in the secondary areas (ie the 2 aren't systematically different in, for example, bear home range size). Have the primary areas been selected because they represent 'good' bear habitat or is is safe to assume that they are representative of the wider area of interest?

Cheers
Darryl

[murray.efford]

...in your case you want to estimate how many individual home ranges overlap a bait station, and also how many bait stations get encompassed within an individuals home range.

I think the aim is to estimate density or population size, which can be done bypassing these slightly fuzzy questions.

Is the sampling intensity (ie bait stations per ha) going to be the same in the primary and secondary areas?

In case you take Darryl's comment to imply otherwise: this is not necessary.

You also have to assume that any relationships/patterns that you observe in the primary areas also holds in the secondary areas (ie the 2 aren't systematically different in, for example, bear home range size). Have the primary areas been selected because they represent 'good' bear habitat or is is safe to assume that they are representative of the wider area of interest?

I hope at this stage it is possible to implement a rigorous spatial sampling design - perhaps a random systematic grid of bait stations with hair snags at a GRTS subsample of sites?

Murray

[darryl]

...in your case you want to estimate how many individual home ranges overlap a bait station, and also how many bait stations get encompassed within an individuals home range.

I think the aim is to estimate density or population size, which can be done bypassing these slightly fuzzy questions.

Even when combining 2 data sets, 1 with individual information and 1 that is just detection? Can't think of a reliable way that you could try to back out density or population size from detection data without some basic idea of home range size and overlap. My unwritten thought was that using SECR with the primary area data essentially gives you this information (maybe not directly) for the detection data so that you 'know' how many bears might have been around a bait station in order to give you a detection. SECR also gives information on home range size so you don't end up double counting the same individual that gets detected at multiple bait stations. Just estimating N within a primary area (that I presumed would include many bait stations that could each be visited by different bears, and each bear could visit more than one station) using standard closed-population methods wouldn't really help you when trying to combine it with the broader scale detection data as Joe was initially suggesting.

[murray.efford]

Even when combining 2 data sets, 1 with individual information and 1 that is just detection?

Yes. Observations of presence/absence at a point may be predicted from the SECR detection function without inferring anything about home range boundaries and home range 'size' in the sense you are using it. Most of the information for estimating the detection function comes from the capture-recapture part of the dataset. The presence/absence data contribute information akin to what distance people call the 'encounter rate', and may help when fitting an uneven density surface.

Murray