Diagnostics for occuMulti two-species occupancy models in unmarked

[johann...@mac.com]

Hi everyone!

I am fitting two-species occupancy models (occuMulti) in unmarked using Rota et al.'s (2016) approach for wolves and deer with 2 years of camera trap data, and I am looking for advice on model diagnostics / goodness-of-fit.

I chose the Rota et al. framework instead of a traditional two-species model such as Richmond et al. (2010) because my main question is about how covariates influence deer-wolf co-occurrence itself, rather than just how occupancy of one species changes conditional on the other. In particular, we are interested in how forest age may influence deer-wolf co-occurrence, while also estimating what influences the marginal occupancy of each species individually.

Model structure:

• I am fitting two separate MSOMs, one for summer (April-September) and one for winter (October-March) using bi-weekly sampling occasions (13-14 occasions per season)

• I'm using a stepwise approach in building my models, selecting models using AIC: first selecting the top detection model holding all state formulas at 1, then identifying the top marginal occupancy models for each species (including the detection formula), and then testing single covariates for the pairwise interaction parameter. (My top detection model includes 3 covariates, with the same detection structure for both species and the top marginal occupancy models include about 3-4 covariates per species

My top model summary seems reasonable: the optim convergence code is 0 (model successfully convered), parameter estimates and SEs don't have any extreme values (and SEs < estimates). I checked for correlation among covariates ahead of running models and I've been using AIC for model selection, but I'm curious what additional diagnostics could I implement for MSOMs in unmarked to assess model fit, overdispersion, etc.? My understanding is that the Mackenzie-Bailey GOF test which I've used previously for single-species models doesn't worked with these models.

Any suggestions or references would be much appreciated.

Thanks very much,

Johanna Griggs