Poison-abundance occupancy models in MARK or PRESENCE?

[jlehmkuhl]

I read Royle and Nichols (2003) and other papers, etc. discussing the Poisson-abundance occupancy model, but I cannot figure out how one could run the model in MARK (with which I am familiar) or in PRESENCE (unfamiliar, but explored Help and run windows). The Pledger mixture models don't seem to fit as they don't seem to estimate Poisson lambda, as far as I can understand. MacKenzie et al in their book (p.141) say that the Poisson-abundance model is implemented in Presence 2.0, but I cannot figure out how such a model could be run, unless it is done as a "Predefined model"? But, does that procedure give you Poisson lambda, which I am interested in estimating?

A more general question: Would this model fit my situation? I have encounter histories for occurrence of ungulate pellets in ~100 patches. Visits were "spatial" in the sense of multiple plots visited within patches. I have trial run a conventional single-season occupancy model, but think that detection p could vary with abundance of ungulates by patch habitat type. And, I am interested in abundance estimates, which are estimated by the Poisson method (e.g. the bird example in Royle and Nichols).

[darryl]

I don't believe that model is in MARK at present (but I'll let Gary correct me). To run the model in PRESENCE you select Run>Analysis:single-season-heterogeneity (Royle/Nichols); it's about the 6th option under the Run menu.

On a more general note though, I believe you have to be very careful in how you interpret this lamda parameter with respect to abundance. There's no guarantee in the mathematics or presence/absence data that lamda relates to the number of individual animals; it's more likely to be some sort of relative abundance measure. Other forms of heterogeneity (ie non- abundance induced) could also influence lamda. If you are going to treat it as some measure of abundance, there's also other issues such as what's the effective area being sampled (of course this also applies with occupancy).

If you're primarily interested in occupancy and are looking at ways of accounting for heterogeneity in p, then this is one approach to doing so (where psi = 1-exp(-lamda)), another is to use the Pledger-type models, or include covariates for p (eg you mentioned habitat type). If you're primarily interested in abundance, then pellet counts alone aren't really going to get you there and you likely need to be collecting other types of data that will allow you to estimate abundance directly (ie data based on individual animals).

Cheers
Darryl

[cooch]

I don't believe that model is in MARK at present (but I'll let Gary correct me).

No - it isn't. One of the reasons is that there is a certain amount of 'healthy skepticism' as to the long-term utility of the model (in current form) since it rests on several fairly untenable (in the real world) assumptions about homogeneity of detection probabilities. And, as any experience with abundance estimation will tell you, heterogeneity is invariably true, and the bane of abundance estimation.

The downside to the Royle & Nichols paper is that unless you think about it all very deeply, there is a tendency for folks to think (in part based on Andy and Jim's very important paper) that presence-absence data offers the ultimate panacea - occupancy, plus abundance. Life is rarely that simple - at least, not unless you're willing to make stated and very strong assumptions.

My two cents...

[gwhite]

The Royle/Nichols model will be in the next upload of MARK to the Web, with both the Poisson model and the negative binomial model. But, my skepticism about these models remains. I totally agree with Evan's comments -- these models are not a panacea.

Paul Doherty and I are doing a workshop at Ohio State in September, and having these models available was part of the deal. Still....

Gary

[jlehmkuhl]

Thanks for the good replies darryl et al. Yes, abundance from occupancy data does seem too good to be true, so I will approach with healthy skepticism. I was mainly interested in abundance as a covariate, but the lure of an abundance index (I dare say in a MARK forum) is there to be sure.