Modeling suggestions

[Erin Netoskie]

Hi Jeff and all,

I'm embarking on my first attempts at building occupancy models and I think spOccupancy will be the best fit! Apologies in advance if this is rudimentary.

For context, I am looking to build a multi-species, single-season stacked occupancy model. My detection/non-detection data is coming from passive acoustic surveys of units deployed for 5 weeks at at time. I am focusing on 8 species that were surveyed across about 1500 sites each survey year for 4 years and my secondary sampling periods are 7-day periods within those 5 weeks. 

I initially used non-spatial msPGOcc(). The detection covariates converged perfectly, but my occurrence covariates have really high Rhat and super low ESS. I ran under two different scenarios 30,000 samples, 10,000 burn in, 50 thin, and 3 chains (second scenario I just doubled everything, but kept 3 chains).

Now I'm wondering if maybe a joint species distribution model is more appropriate, but I'm not working with 100+ species, so maybe it's not appropriate? Any advice on data checks I should do or other considerations would be appreciated! Happy to provide more info as well.

Thanks!
Erin (PhD student)

[Jeffrey Doser]

Hi Erin, 

Thanks for the message. Here are some potential suggestions for you to consider: 

• The first thing to consider is the covariates themselves in the occupancy portion of the model. Have you made sure there are not large correlations between covariates? That can lead to convergence problems. I also recommend making sure to standardize all continuous covariates to have a mean of 0 and standard deviation of 1, otherwise the model may not be able to converge if covariates are on drastically different scales. One way to troubleshoot this would be to simplify the occupancy model covariates (e.g., don't have any covariates on occupancy), fit that model, and see if you get better convergence. Then you could build your way up until you encounter a problem. 
• There is nothing prohibiting you from fitting a JSDM with 8 species. While the algorithm for that can efficiently accommodate larger groups of species, there is no problem with using it for 8 species. However, I don't think that this is what would be causing your convergence issue, so that in and of itself would not be a great reason to pursue a JSDM-type model. 
• If you do go to fit a JSDM-type model or a spatially-explicit model, you are going to want to switch the way you have formatted the data for the stacked model. If you intend to fit a spatial model, your current way of formatting the stacked data for use in msPGOcc() will result in an error. Instead, the way you can fit a stacked spatial model would be to format data for use with multi-season, multi-species occupancy models (tMsPGOcc and the corresponding spatial version stMsPGOcc). While there isn't a specific vignette on those functions, you might want to take a look at the vignette on single-species occupancy models, which could help with formatting and/or understanding how spOccupancy handles multi-season data. 

Hope that helps!

Jeff

--
You received this message because you are subscribed to the Google Groups "spOccupancy and spAbundance users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spocc-spabund-u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/spocc-spabund-users/7dae31a9-e22a-492e-939c-21efa60d00a4n%40googlegroups.com.

--

Jeffrey W. Doser, Ph.D.

Assistant Professor

Department of Forestry and Environmental Resources

North Carolina State University

Statistical Ecology and Forest Science Lab

Pronouns: he/him/his