Skip to main content
Dryad

Calving season habitat selection of maternal and non-maternal female moose in Southwest Alaska, U.S.A

Abstract

We predicted habitat selection of maternal and non-maternal female moose according to a use-available design with animal paths as the sampling unit. To make predictions while assessing the uncertainty from randomness in our representation of available habitat, we created a meta-model for each group from 50 independent random forest models each trained on a different set of available paths (but with the same used paths). We trained all models relative to 16 raster covariates representing topographic and plant community characteristics that we expected to be important to moose during calving season. All covariate rasters are provided in this data archive. The resulting habitat selection models successfully distinguished between used and available paths with cross-validation accuracies of 88% for non-maternal moose and 77% for maternal moose. In addition to habitat selection rasters, we provide uncertainty rasters as the per pixel 95% confidence interval width and binary significance rasters where pixels were significant (value = 1) if the 95% confidence interval did not cross 0 (i.e., from selection to avoidance or vice versa). The spatial prediction rasters enabled us to interpret selection patterns relative to landscape structure.