Two surveys per transect in single season.

[Diego.Pavon]

Hi,

I would like to run occupancy models (single species, single season) from count data that has been gathered in 150 sites surveyed (transect) ONLY TWICE (same year).
I was wondering if this is enough of the surveys per site should be > 3?

Thank you very much

Diego

[dhewitt]

If you have covariates that can be used in the effort, some would argue that you only need one visit. PDF available here under Publications #3:

http://www.math.ualberta.ca/~slele/

- Dave Hewitt

[darryl]

And others would argue that with only 1 survey:
1. the effects of covariates that influence both detection and occupancy are confounded (ie you can't have a psi(X)p(X) model)
2. while solutions exist, they're not unique. The model psi(X)p(Y) has the same likelihood result as psi(Y)p(X)
3. if you don't have a covariate, you can't use those methods
4. you still have to make some form of de-facto closure assumption for the results to be biologically interpretable, otherwise you have no basis for extrapolating whether the species was present/absent at a site beyond that actual survey period. ie if you're doing single 5-minute point counts, without some form of closure assumption your inference is restricted to those 5 minutes at that point. If you detect your bird species in your point count and want to assume that detection indicates the bird species is at that point for some longer timeframe than the survey itself (eg a day, 2 weeks or breeding season), you're making a closure assumption.
5. You can often get replicate surveys from only a single visit. It's about thinking hard before you go in the field.
6. if p<1, you're going to get a more precise occupancy estimate by going to 50 sites twice than 100 sites once (and 25 sites 4 times may be even better if p is low)
7. with Lele et al's approach you are more model dependent for your inferences

As for the original question, 2 might be enough if p is high (eg >0.7), but more will be better if it's lower. How many more depends on how low p might be.

Cheers
Darryl