Abstract
The Asiatic wild ass or khulan (Equus hemionus) is a highly mobile ungulate species inhabiting the arid regions of Central and East Asia. Khulan play a vital ecological role in the Gobi ecosystem, and their nomadic movement patterns are key to thriving in these unpredictable environments. However, expanding infrastructure poses a serious threat by fragmenting their habitat and restricting their movement.
In this study, I used a dataset of hourly GPS locations, 5-minute accelerometer data, and 30-minute imagery collected via camera collars from free-ranging khulan over the course of one year. I aimed to assess the impact of linear infrastructure on movement patterns, activity levels, and group size. I tested how behaviour and crossing frequency vary by different infrastructure types, hypothesizing that fenced railways and paved roads act as the strongest barriers. Additionally, we examined whether proximity to infrastructure alters activity levels and group size, with the expectation of increased activity near fences and decreased activity near high-disturbance paved roads.
I applied the Barrier Behaviour Analysis (BaBA) framework to hourly GPS tracking data to classify movement trajectories into categories such as bounce, quick cross, and back-and-forth behaviour. To examine activity changes near infrastructure, I used high-resolution tri-axial accelerometer data averaged to 5-minute intervals and modelled activity responses with Generalized Linear Mixed Models (GLMMs), accounting for overdispersion and zero inflation. In addition, I explored potential changes in group size by coding camera collar imagery for the number of khulan visible within 5000 m of linear infrastructure.
The results show variation in behavioural and spatial responses depending on the type of infrastructure. Fenced structures exhibited the strongest barrier effect, while paved roads were associated with lower crossing rates. Activity analysis revealed weak negative effect along all types of infrastructure except fenced railroad. Group size observation did not revealed specific pattern through simple visualization.
We suggest to double check classified events from BaBA, with visual inspection to achieve better precision. For a future analysis of khulan activity levels as well as potential modelling of picture data for group sizes we recommend to use non linear models with additional explanatory variables.
