ASSESSING HUMAN DISTURBANCE OF BREEDING BALD EAGLES WITH CLASSIFICATION TREE MODELS

We recorded 4,188 events of human activity and associated bald eagle (Haliaeetus leucocephalus) response in the vicinity of 13 central Arizona nest sites during 1983-85. A hierarchical classification of 9 dependent and 3 independent parameters was developed to quantify pedestrian, aquatic, vehicle, noise (gunshot/sonic boom), and aircraft disturbance groups. Type and frequency of response varied inversely with the distance from an eagle to the disturbance. Bald eagles were more often flushed from perches than nests and were most easily disturbed when foraging. Pedestrian was the most disturbing human activity, whereas aircraft was the least. We developed classification tree (CART) models for pooled and group disturbances to evaluate response severity and to formulate disturbance-specific management criteria. Response frequencies and critical distances for pooled disturbance were 64% at less-than-or-equal-to 215 m, 45% between 216 and 583 m, and 24% at > 583 m. Frequencies, distances, and the influence of secondary characteristics varied among disturbance groups. The CART models ranked distance to disturbance as the most important classifier of eagle response, followed in decreasing order of discriminatory value by duration of disturbance, visibility, number of units per event, position relative to affected eagle, and sound. This procedure offers improved specificity in human disturbance assessment.