Assessing model fit with no 'global model'

[pennyb]

Hi all,

I am very new to PRESENCE, so please forgive me if this is a simple/easy question or if I'm doing it completely wrong! - I am learning myself and don't have anyone to ask things like this.

So a bit of background to my project - I am conducting single-season single-species occupancy modelling on a species of frog. My objective overall is to determine the probability of occupancy + detection of the species at each of my survey sites (15 in total), and to investigate what variables may be 'important' in doing so.

I firstly explored the importance of survey specific covariates by holding occupancy constant while allowing detection probability to vary with time and each survey-specific covariate separately. The 'top' models from this process were then used as my candidate set of models from which I investigated site-specific variable effects on occupancy.

I want to be able to assess my model fit, and I know usually this can be done on the global model and c-hat adjusted for to account for overdispersed data. However, because I have combined a set of models that don't have common covariates in each (e.g. some are not reduced versions of the other) I am not sure how to assess model fit- do I assess the fit on the most parameterised model?

To make this a bit clearer..here is a small sample of my candidate models..

psi(.)p(air temp)
psi(.)p(rainfall on survey day)
psi(.)p(area water + water depth)
psi(.)p(air temp + humidity + sky illumination)

just say they were the only models i was running- as the p(air temp + humidity + sky illumination) model has the most parameters- would i assess model fit on this? or would each model have to include all covariates in the analysis? I'm afraid this would lead to over-parameterisation if I did that...

Thanks in advance for any help, it would be greatly appreciated!!