Analysing and predicting wildlife–vehicle collision hotspots for the Swiss road network

Laube, Patrick; Ratnaweera, Nils; Wróbel, Anna; Kaelin, Ivo; Stephani, Annette; Reifler-Baechtiger, Martina; Graf, Roland Felix; Suter, Stefan M. · 2023 · OpenAlex-citations

DOI: 10.1007/s10980-023-01655-5

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the significant ecological and economic impacts of wildlife–vehicle collisions (WVCs) in Switzerland, aiming to identify key environmental risk factors and model collision risk on a nationwide scale. Motivated by the high frequency of accidents involving roe deer, red deer, wild boar, and chamois, the research seeks to improve collision prevention measures by leveraging fine-grained spatial data science. The authors specifically investigate differences between midland and mountainous landscapes and evaluate the efficacy of different road segmentation and modeling techniques. The methodology utilized 43,000 collision records from 2010 to 2015 across three Swiss cantons representing distinct landscape types: Zurich and Fribourg (midlands) and Grisons (mountainous). The researchers compared two road segmentation approaches: a fixed-length segmentation (200 m segments) and a data-driven segmentation using Kernel Density Estimation (KDE) with narrow and wide kernels. Environmental variables were derived from high-resolution geospatial datasets, including topographic models, land-cover data, traffic noise, and vegetation height. These variables were categorized into attribute, form, distance, areal neighborhood, and complex modeled factors (such as leading structures and feeding ground availability). Statistical analysis employed multivariate logistic regression and random forest classifiers to rank predictors and predict collision risk, using KDE-derived hotspots and carefully selected coldspots as controls. The results identified road sinuosity as a primary predictor for WVC hotspots, alongside two composite factors representing browsing/forage availability and traffic noise (a proxy for traffic flow). The best-performing models achieved sensitivities between 82.5% and 88.6%, with misclassification rates of 20.14% and 27.03%. The study found that prediction accuracy was higher in forested areas, while performance in open landscapes was limited by a lack of up-to-date data on annual crop changes. The comparison of segmentation methods revealed that KDE-based segmentation provided a more balanced dataset for statistical analysis compared to the highly unbalanced fixed-length approach. The significance of this work lies in demonstrating the added value of using fine-grained, semantically detailed land-cover data for WVC modeling. The authors illustrate how spatial neighborhood functions can effectively annotate road segments with landscape data from national mapping agencies. They conclude that incorporating annual crop data is essential for improving model accuracy in open landscapes. By validating workflows across different ecosystem types, the study provides a robust framework for identifying hotspots and implementing targeted prevention measures to reduce socio-economic costs and enhance motorist safety.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• naturalistic crash near crash