When some marked animals are seen and reported outside the official capture season, what is the best way to incorporate such opportunistic encounters in an encounter history format with fixed capture occasions separated by time intervals without recaptures?
Hi. Alberto. in SCR maybe you can use opportunistic encounters out the sampling period, but only to improve sigma calculation. Tenan et al wrote a paper about this issue.
http://biorxiv.org/content/early/2016/09/19/075945
Best,
José
Thanks José,
My question is more focused on how to prepare de data for the analysis. Under a maximum likelihood framework I can think of 2 ways of analysing capture-recapture data with oppportunistic encounters: the 1st is the Barker model, where the encounter history format allows to include live encounters (resights) obtained during the interval between capture occasions (in the same way as recovery data), but the large number of parameters in these models can lead to convergence problems. The second option I have in mind is to code encounters as 2 states (found during or between surveys) and run multistate models, but I don't know if this is mathematically correct, neither I could find any example about this in the literature.
All suggestions are welcome!
Alberto - How much data do you have in your regular sampling, compared to the number of opportunistic sightings you have? Unless you have a lot of opportunistic data, I would guess the added model parameters for trying to incorporate the opportunistic sightings will make it harder to get good model fit/convergence. That's my gut intuition though, perhaps someone with stronger quantitative background will weigh in. As an alternative, you could use opportunistic sightings as a common-sense, qualitative check of models which do not include these sightings.
-Alex
Thanks Alexander,
I am working with 10 years of shark data. On average, 60% of data was collected during sampling season and 40% are opportunistic encounters outside the sampling season. There are few recaptures during sampling so I want to increase recapture probability by including the opportunistic encounters. I like the idea of doing 2 analysis (with and without extra sightings) to double check. However, I am still a bit concerned on the best way to include the extra sightings in my data (Barker LDLD format, multistate codification, artificially increase the sampling period to include more recaptures...).
Al
Hi Albert,
You could consider a robust-design like model with two yearly sessions, i.e. during and outside the study period. For individuals captured during the sampling period, you could force at 0 the recapture probability for the sessions carried out outside the sampling season. By contrast, for individuals captured outside the sampling period, you could force at 0 the recapture probability for the capture sessions carried out during the sampling season. If you are interested in keeping a memory of the capture status of individuals (during vs outside) on survival, the capture status has to be embedded in the states of the model. This type of model can be implemented for instance in e-surge program (see for instance: Cayuela, H., Pradel, R., Joly, P., & Besnard, A. (2017). Analysing movement behaviour and dynamic space‐use strategies among habitats using multi‐event capture‐recapture modelling. Methods in Ecology and Evolution., in press).
Cheers
Hugo
Hi Alberto,
I found your question interesting, so I have run a simple model in e-surge. I have considered a multi-event model with two states (alive or dead) and three possible events (not seen, seen during a sampling session and seen outside a sampling session). The model works... If you want we speak about the model, please contact me by e-mail ([email protected]).
Cheers
Hugo
Hi Hugo,
Thanks! I have also run the same model in E-SURGE, with 2 states, 3 events and 3 matrices in GEPAT (survival matrix, probability of recapture (seen or not) matrix and event probability, given recapture). It works! and I can get estimates for each parameter. But I don't know if treating different temporal encounters as states is mathematically correct as states are usually physical traits or geographic places...Anyway, I will look at the robust design model you commented before and maybe try something like that.
Keep in touch
Al
Hi Al,
Pretty cool if it works ! You could ask Roger Pradel to know if treating different temporal encounters as states is mathematically correct. But I think it is fine. All the parameters seems identifable, even when the survival and recapture are time-specific. Perhaps you do not need to split the event process into two matrices. I think that you can keep a single matrix with the two p parameters on the first row of the matrix.
Good chance for the analyses !
Cheers
Hugo
Yes, maybe there is no need to split the event process in 2 matrices. Thanks a lot for all the suggestions, very helpful.
Cheers!
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology