LAUSR

Abstract Trains on railways collide with and kill wildlife, incurring economic costs for railway operators and impacting species of conservation concern. We proposed to address this problem with train-triggered warning signals, consisting of flashing lights and bell sounds emitted in the 30 s leading up to train arrival, that animals could learn to associate with train arrival. We installed our warning systems at four sites on an active railway where train-triggered cameras recorded the responses of animals to trains. We contrasted the observed flight initiation times when warning signals were active (treatment) and inactive (control) and when trains approached from straight versus curved track, predicting that animals would be more responsive to warning signals on curved approaches because those trains would be harder for animals to detect. When warning signals were provided, animals left the track earlier. For animals larger than and including coyotes (Canis latrans), flight responses were 62% earlier (17.0 s, SE = 1.5 vs. 10.5 s, SE = 1.2 ); smaller animals initiated flight responses 29% earlier (14.6 s, SE = 1.9 vs. 11.3 s, SE = 1.6 ). Contrary to our prediction, animals were more responsive to warning signals when trains approached from straightaways (large animals 10.3 s earlier for treatment, SE = 2.6 ; small animals 4.2 s earlier for treatment, SE = 2.4 ) than when trains approached from curves (large animals 2.7 s earlier for treatment, SE = 2.8 ; small animals 2.4 s earlier for treatment, SE = 1.6 ). Non-aversive warning systems could reduce train collisions by providing additional time for animals to adopt an effective escape trajectory.