Multi-method occupancy without multi-scales

[DEH]

Hi Everyone,
I’m looking for some advice about multiple method occupancy estimation.

Study Summary:
• 168 stations in the study area
• 2-5 avian point counts/station
• point counts occurred both nocturnally and diurnally
• point count radius is 100m

Goal (I’m looking at one species at a time):
I would like to estimate one occupancy rate, but would like to estimate diurnal and nocturnal detection probabilities separately (i.e. treat diurnal and nocturnal surveys as two different detection methods).
EX. Psi= 0.68, p method 1= 0.50, p method 2= 0.60

What I’ve Done:
I’ve run a multi-scale occupancy model (Nichols et al. 2008) with 2 different sampling methods (nocturnal and diurnal surveys), and 15 encounter occasions for each device (total=30 occasions in the encounter history).

This type of model seems to fit the data the best of the ones I've come across, with the exception that what would normally be considered my sample ‘station’ is exactly the same size, spatially, as my sample ‘unit’, so that I’m really only interested in one scale of occupancy. I think this means that when I run a model, my delta estimates should all be 1 (i.e., if the sample unit is occupied, then the station within the unit is always occupied as well).

What I would like to know:
a) Is this interpretation correct?

b) Is it okay to use models that estimate delta, when I’m not really interested in it, and the value of delta will always be 1?

c) Is there another type of model that can provide JUST a single estimate of occupancy with separate estimates of detection probabilities for different methods of sampling that I have missed? ( I REALLY have been looking for one; it seems like it should exist, but try as I might, I’m not finding it).

Any help/advice in this matter would be GREATLY appreciated! Thanks!

[captaingriz]

"What I would like to know:
a) Is this interpretation correct?"

I think your interpretation is fine.

"b) Is it okay to use models that estimate delta, when I’m not really interested in it, and the value of delta will always be 1? "

I think you're better off fixing delta to one, and hence forcing the SE of delta to be 0.


"c) Is there another type of model that can provide JUST a single estimate of occupancy with separate estimates of detection probabilities for different methods of sampling that I have missed? ( I REALLY have been looking for one; it seems like it should exist, but try as I might, I’m not finding it). "

What if you just use the single-season occupancy model with a covariate modeled on p representing the status of the occasion (1=day, 0=night). The point count locations during the night were the same physical locations as during the day, right?

[Lea]

Hi there,
I am in the same boat. We are using two different methods to detect foxes (camera trap and live traps). The locations are the same and only separated in time by a short period so I don't think that closure will be an issue. I can't think of any good reason that the scale would differ among the trap types (both basically point samples as described in Efford & Dawson 2012).

We are interested in seeing how detection varies between the methods and how occupancy varies among years. It seems to me that using a multi-season model and allowing detection probability to vary with the two methods would be the best way to address this analysis but perhaps I am missing something?

Anyone else care to weigh in on this?

thanks,
Lea

[Lea]

So I have done some more reading and I think that because the two detection methods don't occur at exactly the same time that there shouldn't be an issue of non-independence with detection method (e.g. animals in traps would not be "available" for detection on cameras). Please correct me if I am wrong.

[JDJC]

If your cameras are facing your live traps directly and both are operating at the same time, the observations are almost certainly not independent--you could estimate a shared occupancy and availability parameter and a method specific p, but I would think the SE would be over-precise. If cameras were used for a couple of occasions, and then live traps were used for a couple of occasions in the exact same spot, than a method-specific covariate for p is fine, and you aren't able to estimate availability (delta, or theta, or whatever the parameter is) because you don't have contemporaneous sampling.

[Lea]

Hi JDJC, thanks for your response. Our situation was the latter so I will go with a method specific covariate
Cheers,
Lea