single season - year covariate

[Dakota12]

I have presence-absence data from multiple species over 4 years (surveyed in October, November, December, March and May – most years); however the same locations were not always surveyed. Based on previous posts I decided to run single season models on each species with year and/or month as covariates, (the null multiple season model would be the same, but I’m not interested in colonisation or extinction rates). Occupancy likely doesn’t change much between seasons as they are small skinks which do not travel far, but detection probability likely does. I’m trying to use covariates to create occupancy maps for my study area, which I can then test with current sampling sessions. I have a basic question though, if year/month isn’t the lowest AIC does it need to be included in all models (as it is in the multiple season model), or can season or year alone or a null model be used for detection. Additionally, a few of the datasets don’t have enough data currently is it be possible to supplement them with the surveys I’m completing this year?
Data Input example for single season by month (and year as a covariate)
O, N, D, M, May
00110
0000-
10111
11- -0

2005 2006 2007 2008
1000
0100
0010
0001

[jhines]

You're free to build any model which makes sense to you, whether you're using the single-season or multi-season models. You can fix colonization and extinction to zero in the multi-season model to force constant occupancy and let AIC tell you if that's the best model. Or, you can do what you've described with the single-season models, essentially treating all years as a single-season and using year as a covariate for detection.

As long as you've included models in your model-set which allow year and month specific occupancy and/or detection, AIC will indicate if year and month are important factors for determining occupancy/detection. It might be the case that occupancy does change a little between years, but you don't have enough data to pick up that difference. In this case, the standard error of the single occupancy estimate will probably be large enough that the confidence interval includes all of the yearly estimates from the year-specific model.

You can use this year's data as another year for those sparse data-sets, but if you don't have enough data to estimate occupancy/detection from the previous years, adding this year won't give you that information (ie., occupancy or detection from the previous years). If you assume occupancy doesn't change, then it possibly could give you estimates of detection from those years. I'd be a little suspicious, though, as you're getting into a longer span of time where you're assuming occupancy doesn't change.

Jim