This study was conducted on the Canadian Pacific main line railway, which bisects the Bow River valley within Banff National Park, Alberta in the Canadian Rocky Mountains. Several dozen train-animal collisions occur annually in this area.
The warning system comprised of four types of self-contained electronic devices connected through a wireless radio-frequency network, which were termed train detectors, warning devices, camera controllers, and signal repeaters. These devices were deployed along the railway track to coordinate the activation of warning signals and cameras (for observing animal responses) with the arrival of a train. A warning time of 30 ± 5 seconds was targeted, in contrast to previous work that targeted 20 ± 5 seconds, based on the desire to ensure that conditioned (warning) stimuli were typically presented to animals before unconditioned (train) stimuli.
At each site, a test zone was designated for the study, consisting of a 200-meter straight length of track exiting a curve, where trains were more likely to be obscured by vegetation and topography. Train detectors were mounted on the track 40 seconds away at previously measured mean train speed in both directions from the test zone center. When a train moving towards the test zone passed a detector, this device sent a radio signal to all other devices in the network. This signal was received by camera controllers (two per site) mounted in trees on either side of the test zone, where they triggered trail cameras facing the test zone to take 90 photographs at up to two frames per second, yielding at least 45 seconds of footage. The radio signal was also received by warning devices (four per site) within the test zone, which emitted the warning signals (flashing amber lights and bell sounds) after a 10 seconds delay for a period of 35 seconds (30 seconds before and 5 seconds after train arrival at the test zone center). Signal repeaters were placed as needed (one to four per site) between the train detectors and camera controllers to ensure network connectivity. All devices were programmed to wait six minutes after an activation while the train passed.
Two pairs of sites were selected where animal observations were expected to occur most frequently. In each pair, one site was equipped with the warning devices (the treatment condition) while the other was not (control condition). Treatment and control locations were swapped every 2-4 weeks to control for seasonal effects. A team of observers reviewed the recorded images and identified timestamps for moments where an animal initiated a flight behavior, and timestamps for the arrival of the train. Of 711 unique events where animals were visible, 280 were interpretable based on the classification of a flight response.
When warning signals were provided, both large and small animals retreated earlier from approaching trains than when signals were not provided. Averaged across trains approaching from curves and straightaways, large animals retreated 6.5 seconds earlier and small animals retreated 3.3 seconds earlier. Although apparently small, these differences represent 62 percent and 29 percent increases over the mean flight initiation time with no warning signals, respectively. At the mean train speed in the analyzed sample (60.5 km/h), these time differences correspond to increased separations between animal and train of 110 meters (large animals) and 55 meters (small animals) at the moment of flight initiation.
Volume 87, October 2020