Multiple data types and abundance

[AWalpole]

Hello all,

I am looking for advice or suggestions about how to estimate abundance with my data. I am new with program Mark and have been working through the Gentle Guide and some of the cited literature. I have multiple types of data for a population of flying squirrels; including pit tag (resight data), live-recapture and also some dead tag recoveries (from telemetry). This data seems suitable for use with Barker’s Model. This model can only estimate survival but abundance estimates are also of interest to me.

Due to the design of the study and the way in which the sampling occurred (originally geared towards social behaviour) I am confronted with tricky decisions about how to calculate abundance and survival estimate s. First, there is potentially severe heterogeneity in individual encounter probabilities due to non-random timing and placement of the pit tag readers throughout the study area. I feel I can deal with this issue by including individual covariates. Also, the fates of many individuals are unknown because radio collars were not found (right censored). The data could be organized into a robust design type (Poisson log normal mixed effects) but it seems that this type of model would preclude the use of multiple data types.

Are there any thoughts that jump out at anyone about how I can estimate survival and abundance from multiple data types?

Thanks,
Aaron

[dhewitt]

We have a similar issue with trying to use remote detections of PIT tags in models to estimate lambda (Pradel types). The severe heterogeneity in the encounter probs. is bad. We don't try to estimate population size, however.

Out of curiosity, how do you plan to use individual covariates to deal with this?

[murray.efford]

This is an interesting challenge. Thinking just about abundance, which here probably means density, the data have spatial structure so spatially explicit methods are likely to be needed. It's tempting to think in terms of spatial mark-resight, but of course PIT tag readers don't count unmarked animals, so that is out. One possibility is to collapse each closed session to a single 'occasion' and use the approach in Efford, Dawson & Borchers Ecology (2009) 90: 2676–2682. The data could be binary (detection/non detection at each reader location) or count (number of detections per reader location) [I can't remember if the current version of the 'secr' software provides for counts; if not I can fix this]. Differential monitoring of different reader sites could be allowed for with a site-specific covariate.
Murray

[John Boulanger]

Hi Aaron.

We have encountered similar challenges with DNA mark-recapture of estimation of population size of grizzly bears. I suggest collapsing down sessions or using closed mixture models to deal with extreme heterogeneity. How you combine data types depends somewhat on how correlated the capture probabilities are between the data types. Multiple data types can be a powerful way to confront heterogeneity found in each of the data types. Some manuscripts that deal with combining multiple data types that we have published are below. These deal mainly with estimation of N.

Boulanger, J., K. C. Kendall, J. B. Stetz, D. A. Roon, L. P. Waits, and D. Paetkau. 2008. Multiple data sources improve DNA-based mark-recapture population estimates of grizzly bears. Ecological Applications 18:577-589.

Kendall, K. C., J. B. Stetz, J. Boulanger, A. C. Macleoud, D. Paetkau, and G. C. White. 2009. Demography and genetic structure of a recovering grizzly bear population. Journal of Wildlife Management 73:3-17.

Hope this helps.

[bmcclintock]

I agree with Murray's point about abundance (or density), and will clarify one point about mark-resight. Aaron previously implied that all individuals in the "population" were marked. If one believes all individuals are marked, then mark-resight methods (with binary or count data) can be used to estimate the number of marked individuals (i.e., population size) when there are no unmarked individuals to count. However, the marked individual sighting process needs to be adequately explained (e.g., via covariates). As Murray and others have suggested, this would likely be difficult in your case because of the spatial component.

Cheers,
Brett

[murray.efford]

I want to qualify what I said yesterday (really, I want to erase an error). If there are unmarked animals (in general there will be) then existing spatially explicit methods won't work reliably with PIT tag readers when data are collapsed to one interval as I recommended. An unmarked animal that first encounters a reader will not be recognized when it encounters another reader or trap. However, it should be easy to come up with a method that combines the different types of detector. (Not sure when I'll get a chance to code this).
Sorry for the confusion.
Murray

[AWalpole]

Hello all,

Thanks for all your replies. Lately, I‘ve been steering towards using only the pit tag data since it accounts for almost all of our recapture/resights. The SECR method is very interesting, I’ll have to do a bit more reading on that subject (thanks for the papers). For example, how to account for lack of independently distributed non-territorial animals (group nesting) with covariates.

I’m currently leaning towards a Jolly-Seber Model with a staggered entry of animals. First encounter would be capture and all successive encounters of the marked animals would be with the readers. We are also continuously marking new animals and adding them to the sample. The readers are running constantly (for almost 3 years) so we’ll have daily encounter occasions (open popn).

From telemetry, we’re pretty in sync with their habitual cavities so we know when we’ve got a reader on a cavity frequented by an individual. We’d account for uneven sampling by having daily covariates (1 if one or more of that squirrels cavities were monitored that day and 0 otherwise). They move between the cavities quite a bit each night so there is a good
chance we’ll detect one if we’re monitoring one of the cavities. Other covariates would include sex and age at capture. All this is tentative since I’m still organizing the data for input into mark.

Sorry for the long winded response and thanks a lot for the interest. Any thoughts?

Aaron