Extended Data Fig. 5: Varied relationships between covariates and habitat selection across species.

Scatter plot smoothed curves (black lines) of the relationship between predictions of the species habitat-selection models (boosted regression trees) (vertical axis) and the values of covariates used as predictors in our boosted regression tree models (horizontal axis). The smoothed curves were drawn by fitting generalized additive models for large datasets with a thin plate regression spline basis, as LOESS (locally estimated scatter plot) smoothing was not computationally feasible. Full covariate names and units are provided in Supplementary Table 2. Higher habitat-selection values indicate higher probabilities of use, irrespective of availability in this case. A smooth curve is shown for each species. Because each species had one to five predictions, for different life-history stages, we took the mean habitat-selection estimate per cell for each species. Rug marks on the horizontal axis indicate the distributions of the data points.