Uniting dynamic occupancy and generalized latent variable models to understand the effects of cattle grazing on amphibian communities

Background/Question/Methods

Occupancy models are powerful tools for explaining species distributions while controlling for imperfect detection. However, current methods cannot readily handle correlated predictor variables, and do not provide a general enough framework to evaluate evidence for multiple hypothesized causal pathways. Here, we develop such a framework by combining dynamic community occupancy modeling with elements of generalized latent variable models (GLVMs), examples of which include path analysis and structural equation modeling. Generalized latent variable models are used to assess relationships among unobserved (latent) variables that can represent theoretical constructs, and can include a range of non-normal response variables. As a case study, we use Bayesian inference to evaluate relative support for causal relationships among cattle grazing intensity, habitat structure, and water chemistry that explain amphibian community composition in the San Francisco Bay Area of California, USA.

Results/Conclusions

The GLVM framework can strengthen inferences from occupancy models by facilitating the combination of multiple sources of information and comparing relative support for multiple causal pathways. Application of this framework revealed that cattle grazing intensity, as indicated by cow paddy density and site disturbance, had variable effects on amphibians depending on species identity. We also found evidence that the effects of cattle grazing were indirect, owing to changes in habitat structure and water chemistry, and mixed, depending on species identity. The GLVM framework increases the flexibility of occupancy models by allowing for submodels that represent processes of interest that do not play a direct role in species occupancy or detection probabilities. Occupancy modeling in the context of GLVMs may prove to be useful in wildlife management, where latent variables such as prey availability and breeding site suitability cannot be directly observed, by revealing the strength of relationships between observable indicator variables, latent variables, and species occupancy.