Spatial Modeling of Bird Population Using Citizen Science Data
Observation count data from eBird can be used to model the relative abundance of bird species. We found that such data is generally overdispersed compared to a Poisson distribution and that a quasi-Poisson generalized additive model is appropriate for the data. Expanding on previous research for eBird data, we incorporated spatial dependence into the modeling task by performing hierarchical generalized additive modeling with a spatial conditional autoregressive structure for random effects. We found that our data contains moderate spatial dependence and that models that account for spatial dependence have superior predictive performance to those that do not. We conclude that quasi-Poisson hierarchical generalized additive models with spatial random effects provide the best representation of the relative abundance of bird populations. Moreover, our spatially explicit models are more realistic based on domain knowledge when regarding the impact of environmental covariates, which is important when considering conservation implications and future projections.