Hello all,
I am new to both occupancy modelling, and the unmarked package. I would like to
run several questions past those more knowledgeable for your feedback.
Firstly, the
study question:
I am interested in investigating how the occupancy of a snake species varies in
response to burn intensity post bushfires.
The data:
I have a detection history across two distinct seasons ( ~ 1 year post fire,
and ~2 year post fire) for 70 sites that are high intensity burn, low intensity
burn or unburnt. These sites are across four separate national parks. Detection
histories were constructed from standardised active searches to locate the
snakes. In the first year due to logistical constraints (both site access
post-fire and covid), we only had 3-4 four surveys per site, the second season
we sampled sites between 6-8 times. For one park (9 sites in total) we only
have occupancy for the second season as we were not able to access it the first
year.
To account for the different sampling lengths within years and across years I’ve added ‘NA’ values to the detection history so that all sites had equal secondary sampling of 8 (maximum number of visits that occurred at a single site in a season) instances, for both seasons.
The code:
Given we sampled across years I ran a dynamic model using the colext() function
of unmarked. The data frame was the detection history combined with site
covariates of burn intensity (factor) and park/reserve(factor). Year was
included as a yearly covariate.
dynamic_occ_mustard <- colext(
psiformula= ~ intensity + park ,
gammaformula = intensity,
epsilonformula = intensity,
pformula = ~ intensity + park + year,
data = mustarddata,
method = "BFGS")
Model selection indicated to drop all factors, expect park and year from the detection formula.
I checked model fit with mb.gof.test and the chi squared statistic was >0.05. Chat was = 0 though.
The concerns:
1) Is
the inclusion of NAs to account for different secondary sampling lengths
appropriate?
2) Can I run an colext model given for one park (nine sites) there was no sampling in the first year?
3) Why is chat = 0?
3) Are there any general recommendations or further reading you could suggest?
The reason I am seeking feedback is that when I account for different detection across parks and between years, I see a significant drop in the mean among site occupancy between year 1 and 2. From ~0.50 to 0.26. I also see a very high among site extinction probability 0.74. I should note I determined occupancy for each year using the smoothed function, and extracted among site probabilities using predict.
I just want to confirm that I have not made an error in the model as these results, while possible, seem extreme.
Thanks for your time.
Cheers,
Mitch