Diverse behavioural responses of vampire bats to human disturbance revealed by state-space modelling of sparse GPS data.

BackgroundGeomorphologically complex landscapes constrain animal movement and, therefore, the spatial spread of host-associated pathogens. Monitoring animal movement in such landscapes is required to understand these processes but remains challenging because structural barriers to animal movement (forests, mountains) also disrupt the visibility and signal quality of satellites used by animal-borne Global Positioning System (GPS) tags. In Latin America, rabies transmitted by vampire bats (Desmodus rotundus) epitomises this challenge: viral spread is strongly constrained by mountains and may be exacerbated by bat dispersal responses to human disturbance. MethodsWe used GPS tags to track the post-disturbance (i.e., capture, handling and tagging) movement of 93 vampire bats from five roosts in the highly complex landscape of inter-Andean valleys in Peru, where signal degradation reduced usable data by 89%. To overcome this data sparsity, the information gap needed to be filled by modelling and data imputation. Therefore, we developed a probabilistic Bayesian state-space model to infer the movement dynamics of vampire bats and to identify correlates of departure time and distance from the roost, with a focus on disturbance effects. ResultsImmediately following the disturbance of capture, 43% of bats, particularly males and stronger flyers, disappeared from their roosts for the remainder of the study period, suggesting an acute disturbance response. For the 55 remaining GPS-tagged vampire bats, we observed longer flight distances and high individual heterogeneity on the night of disturbance, with bats gradually remaining closer to the roost over the 9-day study period. ConclusionsRoost abandonment and expanded foraging areas are expected to exacerbate rabies spread, demonstrating how human disturbances that alter animal movement may have unintended consequences for disease transmission. More generally, our Bayesian model accounting for GPS uncertainty and missing data extends the utility of animal-borne tracking to species living in complex landscapes with poor satellite visibility.