A simple framework for a complex problem? Predicting wildlife-vehicle collisions.

Area of Science:

• Ecological modeling
• Road ecology
• Conservation science

Background:

• Wildlife-vehicle collisions pose significant risks to animals and human safety.
• Traditional methods for identifying collision hotspots are often costly and logistically challenging.
• Existing computer models for predicting collisions can be difficult for wildlife managers to interpret and apply.

Purpose of the Study:

• To introduce a novel spatial modeling framework for predicting wildlife-vehicle collision risk.
• To disentangle natural (animal presence) and anthropogenic (vehicle movement) factors influencing collision likelihood.
• To provide a practical tool for wildlife managers to reduce collision risks using existing data.

Main Methods:

• Modeled vehicle hazard by predicting traffic volume and speed using human demographic variables.
• Modeled animal exposure by predicting suitable habitat for Eastern Grey Kangaroos using survey and environmental data.
• Integrated hazard and exposure models to predict collision risk across geographic space.

Main Results:

• The species occurrence (exposure) model showed good predictive performance, reducing null deviance by 30.4%.
• Vehicle (hazard) models explained substantial variance in traffic volume (54.7%) and speed (58.7%).
• The integrated collision risk model explained 23.7% of the deviance in reported collisions and demonstrated good discrimination ability on independent data.

Conclusions:

• Collision risks can be effectively modeled across geographic space using a novel analytical framework with existing data sources.
• This approach reduces the need for expensive and time-consuming field data collection.
• The framework's novelty lies in its ability to separately model and integrate hazard and exposure components, offering tunable submodels for practical application.