Summary
We present a method for modeling multiple species distributions simultaneously using Dirichlet Process random effects to cluster species into guilds. Guilds are ecological groups of species that behave or react similarly to some environmental conditions. By modeling latent guild structure, we capture the cross-correlations in abundance or occurrence of species over surveys. In addition, ecological information about the community structure is obtained as a byproduct of the model. By clustering species into similar functional groups, prediction uncertainty of community structure at additional sites is reduced over treating each species separately. The method is illustrated with a small simulation demonstration, as well as an analysis of a mesopelagic fish survey from the eastern Bering Sea near Alaska. The simulation data analysis shows that guild membership can be extracted as the differences between groups become larger and if guild differences are small the model naturally collapses all the species into a small number of guilds which increases predictive efficiency by reducing the number of parameters to that which is supported by the data.
