Results
We found a diversity of tree and field layer plant species across
different areas of the woodland and the hedgeline sites
(summary in Appendix S2). The complete random forest model with all 76
variables had an OOB error rate of 26.67%, with 25% false positives
and 28.6% false negatives. Variable importance metrics from this full
model revealed 27 variables with significant binomial test p-values, and
among those ten were consistently identified as relevant based on the
other six importance metrics (Fig 3; for display purposes variable
importance is shown based on the three less correlated metrics, Fig S1).
These ten variables included vegetation measurements at different
scales, human impact and nestbox position variables (Fig 3).