2-stage models don't suck: variance propagation for 2-stage models of animal distribution

David Miller (University of St Andrews, UK)

Spatial models of animal distributions need to include information about the detectability of the study species in order for abundance estimates to be useful. Much work has been done to build fancy 1-stage models that fit both detectability and distribution components at the same time. Here we show that a 2-stage model, using regular distance sampling methods to account for detectability and generalized additive models for a spatial model can be used to estimate abundance spatially while reliably including uncertainty from the first stage detection model. The approach is rather general and could be used for other multi-stage modelling approaches. We also propose a model for accounting for spatial variation in group size. Methods are applied to minke whale and harbour porpoise data from SCANS-II.