Study design question fro occupancy estimation

[mspinola]

Dear list members,

I am estimating occupancy rate for an amphibian species.
I have 4 forest patches and I am conducting a series of transects allocated in a systematic way in each forest patch. Each transect consists in 17 square plots where I survey 5 of them chosen randomly 5 times looking for the amphibian species.
I want to compare the occupancy rate among forest patches, so I will end with 4 occupancy rates (I am trying to avoid pseudoreplication by not having an overall estimate).
I am considering the forest patch my primary units, the transects my secondary units, and the square plots my tertiary units. Is this correct?
Do I have the risk of autocorrelation in each transect? Are plots in a transect always producing autocorrelation?
Do I need to allocate my tertiary units randomly without consider a transect to avoid autocorrelation?
Thank you very much in advance.
Best,

Manuel Spínola

[darryl]

Hi Manuel
How many transects per patch? Were you thinking of using the plots as your repeat surveys or will each plot be surveyed multiple times? Autocorrelation may or may not be a problem. What sizes are you plots going to be and how does that compare to the species biology?
Darryl

[darryl]

Manuel responded to me directly so I've quoted his reply below for prosperity, then I respond.

The plots are 3 x 3 m (9 square meters), based on the home range of the species. The home range for the species is about 9 or 16 square meters.
The number of transects per patch is to be decided on the size of the patch but we did a sample size calculations and gave us 110 plots (3 x 3 plots) and there is 17 plots per transects, but only 5 plots will be surveyed. That gives us 22 transects that should be distributed in the 4 forest patches according to their size (proportional allocation).
The plots are the repeat surveys (5 times each plot).
May be is better not to use transects at all? I thought it will be more easy to lay the square plots in the field.

If the plot is the sample unit at which you're trying to establish species presence/absence, then by using the transects you have a type of cluster sampling essentially. If you don't take account of this additional level of sampling then estimates may be biased, or at least SE's too small. I can think of 1 way using WinBUGS that may allow you do this (not entirely sure that it would work at the moment), but my preference at this stage would be to consider simple random sampling of plots within the patches (possibly after overlaying a grid on each patch).

How big are the patches? It's good you've done some initial sample size calculations, but if you're interested in comparing occupancy between patches, then I'm guessing you're going to be wanting at least 50-ish within each patch to get a good estimate of that difference.

Cheers
Darryl