Single-season models and multiple years of data

[willard]

I have occupancy data collected over two years, three visits at each survey point. Each year a separate random sample of points was surveyed. The survey points were selected using a generalized random tessalltion stratified (GRTS) design. The issue I have is that in the second year the sample criteria was altered to include more habitat types that appeared suitable to the focal species.
Does it make sense to combine the two years of surveys into one dataset and run a single-season, single-species analysis using year as a covariate or is it better to build separate models for each year. Are there any advantage/disadvantages to doing it one way or the other?

thanks

[jhines]

The advantage of combining the data into one analysis is that you have the potential to share similar parameters between data-sets. It may or may not make sense to do it, but it's possible. With two separate single-season analyses, it's not possible.

In your case, you changed the sampling scheme from one year to the next, so you might think that detection probabilities are different in the 2nd year. If you analyzed as a multi-season model, you can allow detection probs to be different from year to year (or even survey to survey within year). You would also get an estimate of initial occupancy, one colonization rate and one extinction rate. Occupancy in the 2nd year is computed and printed as a derived parameter.

So, as long as the area sampled for occupancy is the same between years, and only the detection method changed, I would suggest the multi-season model. If it turns out that detection probabilities are very similar between years, you have the option of running a model with constant detection probs, giving smaller standard errors.

[willard]

The extent of our study area was the same each year (glaciated region of Ohio), but different discrete wetlands were surveyed each year.
The second year we increased the proportion of semi-permanently flooded wetlands sampled and removed temporarily flooded wetlands from the sampling design.
This was done because the temp. wetlands in Ohio did not seem to fit the marsh bird habitat criteria. So to me it seems like less an issue of detection probability and more
an issue of occupancy.

If I understand the multi-season model correctly, I could not use it because I did not survey the same wetlands. Is that the correct interpretation?

If that is the case, would it be best to combine the years into one single-season model and use year as a covariate for occupancy?

thanks again
I appreciate your help

[jhines]

You can use either the single-season model, with year as a covariate, or the multi-season model. They should both give the same estimates of occupancy in the two years. The fact that occupancy is different in year 2 due to sampling different habitat doesn't matter in the analysis. Since you know that year 1 occupancy applies to one set of wetlands and year 2 occupancy applies to a different set of wetlands, using the multi-season model to get the estimates is still OK. You just wouldn't use the extinction and colonization estimates for anything. You'd only be interested in Psi1, Psi2, and the detection probs.