Occupancy with single occasion for one level of factor

[jCeradini]

Hi Analyzers,

In a single-season occupancy model, I have 10 occasions of detections from camera traps (each camera occ = 2 wks) and a single occasion of detections from a scat survey (with species ID based on genetics). So, each site has 11 occasions (10 camera, 1 scat) and n = 64 sites. Scat surveys were just a single survey at the site (scat was collected for DNA lab analysis). No attempts were made to have multiple scat surveys (occasions) per site. Scat search time and other relevant covariate data were collected per site.

Can I estimate p ~ survey method (camera vs. scat)? I have a binary time-varying covariate for survey method for each occasion. Models run fine, and the estimates make sense (way higher for scat than camera...for multiple reasons, like representing animal use over different time periods). But, there's only one scat occasion per site. So, can I estimate p ~ scat? If so, does it change the interpretation of the beta in any way?

These papers demonstrate methods for single-occasion occupancy models, so that gives me some hope:

https://academic.oup.com/jpe/article/5/1/22/1296465
"In general, site occupancy and detection probability parameters can be estimated using a single survey provided (i) probability of occupancy and probability of detection depend on covariates and (ii) the set of covariates that affect occupancy and the set of covariates that affect detection differby at least one variable."

https://besjournals.onlinelibrary.wiley ... 2664.12925

Thanks!

[sbonner]

Hi J,

Fitting the model with p ~ survey method. Essentially, you have variation in the detection probability with is modelled through the covariate (type of survey). Heuristically, the model uses the information from sites where the species was detected (and so known to exist) to estimate the probability of detection for either method. These can then be used to estimate the probability of occupancy at the sites where the species wasn't detected. This is all fairly standard.

Regarding the p ~ scat model, I think you are suggesting limiting to using only the scat data. Is that correct? In that case, there would be only one detection probability so it is not clear to me what the covariate is. Alternatively, you mentioned scat search time. Is that the covariate? Are there covariates on the occupancy probability as well?

Regardless of this, you should also look at the 2014 paper by Jonas Knape and Fränzi Korner-Nievergelt (https://besjournals.onlinelibrary.wiley ... 210X.12329). Knape and Korner-Nievergelt showed that the identifiability of this model depends on specific assumptions about the relationship between the covariate and the detection probability. Essentially, it is impossible to distinguish between alternative models that would give you the exact same fit to the data, but very different estimates of occupancy. This should make you very wary of using the models with only a single survey.

I hope that helps,

Simon

[jCeradini]

Hi Simon. Thanks for the reply and the paper!

I think I muddied the waters by writing "p ~ scat". The question is just if, given the data structure/survey design, I can estimate p ~ survey method (camera vs. scat)? I was concerned because each site has 10 camera occasions and only a single scat occasion. I wasn't sure if there were some important assumptions to be aware of that may influence how the coefficients are interpreted. But sounds like I was overthinking it and the standard occupancy model framework applies for p ~ survey method (as you describe in your 1st paragraph).

I am not planning to run a model with just the single occasion of scat data (if I did it would just be logistic regression).

[sbonner]

Thanks for clarifying. Yes, you can absolutely estimate the probability of detecting occupancy via the scat surveys using the model with the combined camera + scat survey. In that case, the paper by Lele et al is not relevant. They consider experiments in which each site has only a single survey.

Cheers,

Simon