Population size under closure violoations: POPAN vs alternat

[mondulkiri]

Dear Forum,

I am currently analysing data from a fecal-DNA survey of Asian Elephant in a protected area in eastern Cambodia and would appreciate advice. We conducted five sampling sessions during which elephant dung samples were collected across a study area of 130-km2. Four survey teams, targeting different areas, were used to try to evenly survey the study area with each team visiting different elephant ‘hot-spots’. Each sampling session was between 8 and 10 days (mean 8.2) with between 10 and 13 days (mean 12) between sampling sessions. Individual elephants were identified from genetic analysis of the dung. A total of 78 individual elephants were identified (55 caught in one session; 18 two sessions; 5 three sessions). However in addition to these between sessions captures there were 44 in session captures (from 24 individual elephants).

The initial plan was to analyse the data using closed capture models to estimate population size (the parameter of interest in this study). Therefore the overall survey period (89 days) was sufficiently short for demographic closure to be likely and surveys were conducted in the dry-season when elephant distribution is limited to areas with water and animal movements therefore not extensive (our a priori hypothesis was during this period elephants were largely restricted to the 130-km2 study area as this was where most water remained in the dry-season). However examination of data suggests geographic closure may be violated. Evidence from several tests (detailed below) suggests the population was not completely closed....

1. genetic evidence of one elephant (sampled in sessions 1, 2 & 3) moving to a second surveyed study area approximately 60-km away in session 4
2. in closed capture models in MARK Mb is selected as the second best model (after M0; based on AICc) with lower recapture (0,15) than capture probabilities (0.21). However due to non-invasive sampling (with limited disturbance to study areas) Mb seems biologically unrealistic. Instead it seems likely that low recapture probabilities are due to individuals leaving study area.
3. Stanley & Burnham Closure Test (p=0.3) & Otis et al. Closure Test (p-0.04) run in CloseTest suggest borderline closure violation
4. Pradel survival/recruitment test (with no time effects) in MARK gives Phi=0.79 and f=0.15 suggesting some movement in and out of the study area. However if these two parameters are constrained to 1 and 0 respectively the AICc score of the constrained model is given more support (delta AICc 0,84).

Therefore evidence that the population is almost closed but not quite....I would therefore appreciate advice regarding the best approach to estimate population size from this data...should I just assume closure and justify with some of the info above or would it be best to explore some open models.... POPAN has been suggested to me but I’ve read a number of comments on this forum suggesting this isn’t recommended for estimating population size? What is the reason for this – if it is because it is difficult to interpret ‘super-population’ that is less of an issue here as the elephant ‘super-population’ using the protected area (which is adjacent to two others with elephant populations) would be valuable conservation information.

Given the number of within session captures would a robust design be suitable? However I am a little worried that the data-set may be too sparse for this given the number of parameters likely to be included in such a model & (from my limited reading) of robust models they don’t seem to give an estimate of overall ‘super-population’ abundance. Finally I’ve seen papers on software CAPWIRE which does seem to be able to account for multiple within session re-captures (but also assumes population closure) what are the forums opinions of using this approach to analyse the data?

Apologies for the long message....

Thanks
Tom Gray

[murray.efford]

Tom
This sounds like a spatially explicit capture-recapture problem, not a non-spatial N-estimation problem. The relevant papers are Royle & Young (2008) in Ecology and one I have in-press with Ecology. The methods you need are provided in the R package 'secr'. This would be an interesting first (to my knowledge) application of that package to DNA samples from area searches. Follow up on the Density|secr forum if you're interested.
Murray

[mondulkiri]

Murray,

Thanks for your rapid response. However as the parameter of interest density per se isn’t particularly useful for Asian elephants. Unlike say large cats or bears elephants aren’t territorial and instead occur in groups with large, over-lapping (and seasonally very variable) home-ranges. Elephant density would be a constantly changing parameter depending upon the size of home-ranges which in turn is dependent upon water availability / disturbance etc.

Instead for conservation purposes the population or super-population size using the protected area is much more relevant and informative for trying to monitor this species....

Tom

[murray.efford]

Tom
phidot just timed out and binned my beautifully constructed reply . Tersely: you should still try spatially explicit modelling if there is any localisation of animals within your study area. If not, then you have problems. Spatially explicit modelling is not just for density: in a wide range of conditions it also gives better estimates of population size. The POPAN superpopulation is not relevant in my judgement (why would you allow an elephant only one visit?), but I'm still hoping for expert input on that question from a previous post.
Murray

[egc]

Tom
phidot just timed out and binned my beautifully constructed reply .

The forum is set to time out an individual session after 25 minutes. This time horizon was set based on analysis of the distribution of 'session times', and is in place (in the first instance) as a back-end security measure (technical details and reasons won't be of general interest). For the vast majority of posts (>97%), this won't pose a problem.

However for longer posts (or, more typically, for users who are logged in (as opposed to browsing without logging in) for significant -- >25 minutes -- amounts of time), getting 'timed out' will happen periodically. For now, I'm going to leave the default time-out interval as is, with apologies to those who run into the problem Murray had.

[egc]

Tom
phidot just timed out and binned my beautifully constructed reply .

The forum is set to time out an individual session after 25 minutes. This time horizon was set based on analysis of the distribution of 'session times', and is in place (in the first instance) as a back-end security measure (technical details and reasons won't be of general interest). For the vast majority of posts (>97%), this won't pose a problem.

However for longer posts (or, more typically, for users who are logged in (as opposed to browsing without logging in) for significant -- >25 minutes -- amounts of time), getting 'timed out' will happen periodically. For now, I'm going to leave the default time-out interval as is, with apologies to those who run into the problem Murray had.

Since it was not my intent to 'frustrate' people who spend a fair bit of their time helping others on the forum (people like Murray), by having their sessions time-out before they hit the submit button, I've made a few 'behind-the-scenes' changes which I think should significantly reduce the frequency of these sorts of issues.

I'm afraid the only test of whether or not the changes have 'worked' is if the problem never arises again.