concerned about violating closure

[mtbikernate]

I am concerned about violating the closure assumption in my analysis. In the past few months, I've been working with my detection history data getting a feel for PRESENCE and how to work with things.

Study Background:

I obtained detection histories using motion-sensing trail cameras. My goal was to survey several private properties intensively with the goal of detecting a handful of species of interest. I detected a great deal more species (more than a dozen) than I sought to detect initially. I arranged sites in a regular grid arrangement on several private properties. I had about 30 sites at a time (with two cameras at each site), and rotated my cameras between properties in order to survey the entire area. Cameras were active at each site for 10-14 days before being moved. I had a winter season (Dec-March) and a summer season (May-Aug). Each site was visited twice during each season.

Due to my intensive surveying efforts, several species were detected at most sites at least once during each 10-14 day interval. I am confident I can assume closure for that amount of time. That said, if I use each 10-14 day interval as an "occasion", then my detection probabilities for most species are 1 and that doesn't give me much useful information about the covariates. I have considered using each day the cameras are active as an "occasion" because for each 10-14 day interval, I detected most species on fewer than half of the days. This should give me some useful information about the covariates since my detection probabilities would be less than 1.

I am concerned about violating the closure assumption for each of my seasons. Can I assume closure for each entire season (Dec-Mar and May-Aug) and therefore use a 2 season model, or should I break up those seasons into smaller time periods?

[bacollier]

mtibikernate,
To answer your question, yes, you can consider each day as an 'occasion' if you want and model whatever covariates you are interested in. You probably can assume closure for each within season, but not the between period. Are you 'occurrence' measures at a scale that is demographically reasonable (space use, closure, etc) for your species of interest for that time frame?

Having said that, I guess I am kind of a pragmatist <begin soap box> on occurrence modeling, your data indicate that you don't have to estimate detection when you survey 10-14 days, all sites were occupied based on your data so I would argue since detection is typically considered a nuisance parameter in CMR type models to better inform the occupancy process, what information are you going to gain by modeling detection covariates? I suppose you might find some variable that 'looks' like it drives detection, or maybe 10 days is too long and critters move around alot (see scale thought above), but I would not trade occurrence certainty without a model for less certainty and a model any time </end soapbox> Granted you won't get to have a 'detection model' to talk about, but that's not always a bad thing...

Not knowing what your surveying for, you might (if you have not already) see OConnell et al. 2006. J. Wildlife Management 70:1625 as it seems roughly similar to your approach with cameras.

Bret

[mtbikernate]

Bret-
Thanks for your response.

Are you 'occurrence' measures at a scale that is demographically reasonable (space use, closure, etc) for your species of interest for that time frame?

Forgive my slowness (brain damage). What do you mean by this? Do you mean spatial scale or temporal scale? Each site is approximately 200m from the nearest 4 other sites.

what information are you going to gain by modeling detection covariates

I hope to gain some information about what landscape features are important to the species I detected. I am also working to model landscape change in order to develop a model to predict the effects of that change on the different species I detected.

maybe 10 days is too long and critters move around alot (see scale thought above)

that's a possibility. I know that for a number of species (like bobcats and coyotes, for example), their home ranges included several sites. Since one of the goals of the project is to learn about the importance of different landscape features, I wanted the more intensive sampling.

Not knowing what your surveying for

Mesopredator detections (primary interest for study) included:
raccoons
opossums
striped skunks
(1) spotted skunk
grey foxes
coyotes
bobcats
river otters
domestic cats
domestic dogs

other species included (I'm not terribly interested in these, but may test some of them):
deer
feral hogs
cottontail rabbits
squirrels
various smaller rodents including cotton rats and mice
armadillos
various birds ranging from birds of prey to songbirds

OConnell et al. 2006. J. Wildlife Management 70:1625

I can't remember if I've read that one yet or not, but I'll take a look. Thanks.

[bacollier]

Bret-
Thanks for your response.

Are you 'occurrence' measures at a scale that is demographically reasonable (space use, closure, etc) for your species of interest for that time frame?

Forgive my slowness (brain damage). What do you mean by this? Do you mean spatial scale or temporal scale? Each site is approximately 200m from the nearest 4 other sites.

what information are you going to gain by modeling detection covariates

For occupancy models to be appropriately used to address occurrence rates (imho), the spatial scale at which occurrence is measured has to be relevant to the demography of the species. For instance, do 100m by 100m grid cells in a forest represent a scale at which we should monitor occurrence if the species can transition between cells easily due to large range sizes, or from an avian perspective what does occupancy rate or colonization/extinction rates at a random point in a patch of woods from a point count survey tell us-should we perhaps be monitoring at the 'patch of woods' scale.

I hope to gain some information about what landscape features are important to the species I detected. I am also working to model landscape change in order to develop a model to predict the effects of that change on the different species I detected.

That may be difficult if all your sites are occupied to tease any relationships out on what impacts landscape has on occurrence, regardless of of how you model the detection histories.

maybe 10 days is too long and critters move around alot (see scale thought above)

that's a possibility. I know that for a number of species (like bobcats and coyotes, for example), their home ranges included several sites. Since one of the goals of the project is to learn about the importance of different landscape features, I wanted the more intensive sampling.

Not knowing what your surveying for


Mesopredator detections (primary interest for study) included:
raccoons
opossums
striped skunks
(1) spotted skunk
grey foxes
coyotes
bobcats
river otters
domestic cats
domestic dogs

other species included (I'm not terribly interested in these, but may test some of them):
deer
feral hogs
cottontail rabbits
squirrels
various smaller rodents including cotton rats and mice
armadillos
various birds ranging from birds of prey to songbirds

Ok, for some of these I could see closure being ok, perhaps not for a dog or coyote but whatever.

OConnell et al. 2006. J. Wildlife Management 70:1625

I can't remember if I've read that one yet or not, but I'll take a look. Thanks.

No sweat.