single survey avain point counts

[Valerie]

Hello,
I have a dataset where 150 sites were visited once a year for 5 years. Each site had one avifaunal point count broken down into the first 3 minutes, 2 min, and 5 minutes. I know there are ways to use these time periods as "surveys" to estimate detection probability, but how do you account for the lack of independence (i.e. if an individual bird was heard in the first 3 min and in the 2 min period, it was only recorded in the 3 min period to avoid double counting)? For setting up the matrix would you assume that once a species was detected, it was present in all remaining time periods? And, if so, would this inflate the detection probability?
Any advice would be appreciated.
Thanks,
Valerie

[darryl]

You should use the detection/detection data as collected and either 1) define a survey-specific covariate that =0 up to and including the survey when the species was detected for the first time in a year, and =1 afterwards (then reset each year); or 2) analyse the data for each year separately and use the single-season, custom model with 'spatial' correlation which is really just accounting for a form of dependence between surveys, although it may not work very successfully with only 3 surveys per year (unless Jim Hines has extended it to multi-season models now too).

Other approach is to just use the time to first detection by changing the data to missing values (-) after first detection within a year, although by doing so you get a bit less flexibility in terms of modelling detection.

Other things to keep in mind

1. these are species-level models so while there may be some concerns about independence at a finer scale, they may get largely aggregated away at the species level. Good thing is you can actually test that by comparing models with and without the covariate, for example.

2. with all surveys in 1 day (and actually back-back in 10 minutes) you want to think really hard about what occupancy actually means. It's where were birds present on that particularly day (or whose, during the actual time of 10 mins of surveying), so you want to extrapolate out to a longer timeframe you need to make some assumptions about movement (in particular, the lack of it).

Darryl