Assistance with results interpretation (PRESENCE)

[dann]

Hi There,

I am looking for some help with the interpretation of outputs given by PRESENCE.

I am studying a rainforest snake species. The species is thought to live in 3 seperate populations and I surveyed each using repeated transect surveys. Each transect was surveyed 16 times and transects were unevenly divided between 3 habitat types (site covariate #1; habitat). In addition, due to logistical constraints the lengths of each transect were different (covariate #2; transect length).

First problem; one of the populations yielded not a single snake in any habitat even after >200 person hours search effort. Further, the snake has never been seen by aboriginal people or by professional herpetologists – the records from this particular site were anecdotal and, I believe, a case of mis-identification. This species has an average detection probability of 0.6 in other populations and is very conspicuous… Therefore I genuinely believe it to be absent from this site… please don’t flame for that..

Not-surprisingly, the most favoured model I ran for this population was the constant psi(.)p(.) model (due to no presences; AIC of 4) The model gives psi = 0 with SE = 0, however it gives p as 0.242 with a very large standard error.

First question; is there any application for this detection probability or is it just a rubbish result? Is it possible to use this value to say that, given such a probability of detection the species should have been found in the area after N visits?

I analysed each population separately. Would I do better to analyse all the data together with population (i.e pop1, 2 & 3) as a site covariate?


Second problem; In another population (where snakes do occur) for the best model (psi(habitat)p(length)) I have complete data separation because snakes were never once found in woodland. Therefore, the first site covariate (habitat) perfectly predicts occupancy. However, for woodland habitats, where no snakes were observed, I still get a detection probability of around 0.5 for each site (p at rainforest sites, where the species was found, ranges between 0.55-0.94 using this model). Is this because transect length was shorter for woodand sites, on average, than rainforests sites?

Question 2; can somebody please offer me an biological interpretation for why I may get this p value and what it means for the study.

What my study aims to find out is what is the probability of occupancy at each population and each habitat within each population. The program does this well but I am still confused as to why detection probability can be the same in habitats where I found 0 animals as habitats where I found many.

Any help you can give would be much appreciated.

Cheers,
Dan

[darryl]

Hi Dan,
1. These methods are statistical, not magical. With no detections at all in population 1, the p estimate is meaningless.
2. One reason for combining data from all populations is if you suspect some of the parameter estimates might be similar so you share information by analyzing all at once. Another reason is if you want to directly compare populations and assess whether they are doing the same or different things.
3. By using a p(length) model you've forced p to be the same in both habitats, with only transect length being important. You could include a habitat effect, but if you have no detections at all in woodland then you have a similar problem to population 1.
4. Detection probability is conditional upon presence. Hence you may have a similar detection, but different level of presence. Also remember that these are only estimates that come from the model that you decide to fit to the data. You therefore need to think about what types of models make sense before sitting down at the computer.

Good luck.
Darryl

[dann]

Hi Darryl,

Thanks for your help. After playing with some more models I have decided to run with a straight psi(habitat)p(.) (which was the model that fit best in the first place).
With more than 140 surveys in Population 1 I will just have to let the search effort speak for itself.

Thanks again,
Dan