Modelling rare species

[bnewkirk]

Hi,

I am currently working to develop environmental associations of rare fish species using single-season, single-species occupancy models. While running my analysis, I noticed that the models are having a hard time with some of my rarer species. Particularly, it seems I'm having issues when the proportion of sites with detections drops below 0.10. Is there a rule of thumb for the proportion of sites with detections that you should not build these models? I've tried to find some literature and can't find a solid answer.
Also, I've noticed quite a few of my GOF tests providing results with c-hat < 1. From what I have gathered, this could be suggesting that my models are overfit. Am I interpreting those results correctly?

I apologize if these questions have been asked previously on here. I tried searching, but came of short.

Thank you,
Braxton

[jhines]

I don't think I can give a "rule of thumb" for sufficient data to run occupancy models as these models can range from very simple models (single-species, single-season with no covariates) to very complicated models (two-species, multi-state, multi-season, etc.). The more complicated models will require more data. I think the most simple model should work with 10% of sites with detections giving an estimate of occupancy which is better than a naive estimate (10%), but doesn't allow you to address questions about covariates. Depending on the covariate and the number of sites, you may be able to run a model with a covariate on occupancy and/or detection. I recommend starting with the simple model, then work on more complicated models with covariates until the results indicate the model is overparameterized (model not converged and/or estimates at boundaries).