Fig. 2: Top-10 variable importance plots ranked by mean gain for spring and fall.

Top-10 variable importance plots ranked by mean gain for A spring and B fall. Blue bars show predictors that had a positive influence on migrant stopover density in >55% of models, red bars show predictors that had a negative influence on migrant stopover density in >55% of models, and gray shows neutral predictors (between 45% and 55% positive). Note that neutral indicates that the variable was not dominantly positive or negative across all models. However, it could hold a negative or positive value. We show the percent of positive instances within models to the right of each bar. Pie charts show the summed percent of gain across non-sampling predictors categories (i.e., we removed distance to radar and year from the summation). For pie charts, “other” was composed of pooled predictors of landcover at 1-km scale (spring = 1.9%, fall = 2.1%), EVI (spring = 4.0%, 3.4%), percent impervious surface (spring = 0.4%, fall = 0.4%,), and percent tree canopy cover (spring = 2.2%, fall = 3.6%); each individual category accounted for less than 5% of the total gain.