Road and landscape features affecting the aggregation of ungulate vehicle collisions in southern Sweden

Abstract: Ungulate-vehicle collisions (UVC) are increasing in Sweden and raises concern to traffic safety, socio-economics and wildlife management. Accident numbers are steadily growing but the trends are not well related to the changes in ungulate population sizes or hunting bag statistics. Authorities ask for more efficient mitigation strategies, but this require a good understanding of where and why UVC occur more frequently in some areas compared to others and which factors that affect these aggregated patterns. To find out which factors that are crucial to the emergence of UVC we studied a selection of roads stretches where UVC were frequent and compared road and landscape features with stretches with lower frequency of accidents. I used UVC records during 2010 - 2014 provided by hunters who have been called by the police to the accident site. In contrast to the official police records, these hunter reports contain exact location data as well as correct species identification. A total of 189,733 UVCs has been reported during the 5-year period, of which most involved roe deer (77%), fewer involved moose (11%), wild boar (9%), fallow deer (3%) and red deer (1%). While roe deer and moose occur broadly across Sweden, the other ungulates have more restricted but expanding distributional ranges. For my study, I therefore selected southern and south-central Sweden where all five species occur and where road density, human population and UVC frequencies are highest. I further focused on primary and secondary roads, excluding the more comprehensive tertiary and private road network where about only 15% of reported UVC occur. I studied the summed UVC pattern in general and did not distinguish between the involved species. To distinguish road stretches with high density of UVC (clusters) from stretches with low UVC density I used a modified kernel density estimation approach (KDE+; Bil et al. 2013) where a high density UVC road stretch have a minimum number of UVC (≥ 5 accidents within the cluster road section) I identified a total of 1596 UVC clusters. From these, we randomly selected 474 clusters, which we compared to 429 random and non-aggregated UVC sites outside the identified clusters. Due to the spatial error and uncertainty in UVC positioning, we considered each UVC location (in and outside cluster) to represent a 500 m road segment. At each site, we measured 15 road related factors (ocular evaluation of Google Street ViewTM imagery) and 17 landscape related factors (derived from topographic map data and GIS data bases within 1 km radius around the site). We used a generalized logistic regression approach to identify themost important factor combinations explaining the clustering of UVC. According to our results, the clustering of UVC tends to occur in areas where the road corridor is attractive, accessible and open for wildlife. Such areas are characterized by diverse landscapes with forest patches and with many leading structures such as watercourses, other roads, lakes etc. These features, in combination with traffic and road related data (speed, traffic volume, absence of wildlife fences) provide a powerful explanation of UVC clustering.