Responses of wildlife to acoustic stimuli at railways – implications for wildlife warning systems

Wildlife–train collisions pose a persistent safety and conservation concern. Several mitigation strategies have been tested to reduce risk, and acoustic deterrents have proven successful in case studies. However, systematic evidence regarding the effectiveness of acoustic-based mitigation measures remains scarce. Between November 2021 and September 2024, we used motion-activated sound systems (MASS) at 25 sites along railways in Sweden and Norway to expose free-ranging wildlife to five acoustic stimulus categories (train horn, railway bell, predator growl, human voice, white noise) with silence as a control. From 561 video encounters of seven mammal species, we analysed response intensity and movement direction using Bayesian multilevel models with partial pooling across species and sites. All sounds raised flight probability above silence (0.06), from 0.75 for train horn to 0.95 for white noise. Train horn and railway bell were weakest (immediate flight 0.35 and 0.43); human voice and white noise were strongest (0.67 and 0.78) and white noise most probably ranked highest for immediate flight (posterior probability 0.93). Among animals with a recorded movement direction, active sounds reduced the predicted probability of movement towards the railway from 0.37 to 0.15 (about 60%) across all seven species, and also reduced movement towards the speaker (about 65%). Response intensity declined only weakly with cumulative exposure at a site, and this could not be separated from differences among sites. We conclude that acoustic signals reliably elicit flight and reduce movement towards the track, making acoustic warnings a promising measure to reduce collisions. Conventional railway signals are the weakest option; white noise and human voice were the most effective, and the choice between them may depend on practical considerations, such as whether the signal also warns people. Habituation under operational conditions, where sounds precede real trains, the effect on collision rates and the effect of horn playback at full operational level remain untested.