6. Spatial estimates of deer density
We produce spatial estimates of deer density for northern hardwood stands in a section (20 km2) of our study area (Fig. 1b and c) by applying parameter estimates from the ‘best’ and averaged Bayesian linear regression models (from BMA) to raster maps (30 m resolution) of the independent variables.
Confidence intervals for each variable parameter estimate can be used to produce a model ensemble that provides both a mean estimate of deer density and a standard deviation of that estimate for each map pixel .
These standard deviations can be used as an indicator of uncertainty in the pixel estimate (larger standard deviation of predicted values across model realizations
implies greater uncertainty).
The model ensemble is composed of multiple realizations of the Bayesian linear regression models, with parameter values for each realization sampled from the
posterior probability distribution for each independent variable.
We use 1000 realizations for each model ensemble, performing analyses with a bespoke computer program. We examine a section of our study area because our data are spatially incomplete for the entire study area.