Network Screening Approach for Cyclist Safety in Quebec City
BACKGROUND
The objective of this study is to demonstrate a simple approach for estimating bicycle activity across an entire network, to perform network screening by merging bicycle activity estimates with bicycle collision data, and to apply the methodology in a case study of the City of Québec. Only a few studies have combined GPS bicycle trip data with screenline bicycle counts to generate network-wide cycling activity estimates [1-2] and only one know study has used network-wide bicycle exposure data to estimate risk in a network screening process [1]. The results from the study in Québec revealed high-risk intersections in need of prioritization for safety treatments and cycling desire lines.

PURPOSE
This research demonstrates a relatively simple, practice-ready approach for estimating bicycle activity across all network elements (roads, bike paths and intersections) by combining screenline bicycle counts from automated bicycle counters with GPS bicycle trip traces. The resulting estimates can be merged with bicycle-vehicle crash data to perform network screening at intersections, and with existing bicycle infrastructure maps to reveal desire lines.

METHODOLGY
Long-term, automatic, bicycle counters have been used as references to estimate daily averages using short-duration counts [3-5]. However, few studies have demonstrated that long-term screenline counting sites can be used as references to estimate daily averages from GPS trace cyclist data [1]. The merging of bicycle screenline counts and network-wide GPS trace data allows for bicycle activity to be estimated throughout the entire road network. The bicycle counts at each segment of the network are expanded to average daily bicyclists using a factor, which is determined through linear regression. The dependent variable being the average daily bicyclists derived from the cleaned and validated [6] long-term and short-term screenline counting data. The network-wide bicycle activity, used as an exposure measure, is combined with reported collision data to estimate risk at intersections (defined as collisions per million cyclists) for the purpose of network screening.

The methodology was applied in a case study of Québec City. The study uses data from three long-term bicycle counters, 20 short-term counting sites and GPS traces from 6100 bike trips generated by 650 citizens in 2015 as part of a campaign to identify desired travel routes through the city using a mobile application called Mon Trajet Vélo.

RESULTS
All intersections with bicycle traffic exceeding 1000 trips per day are located along bicycle facilities, confirming that Québec City cyclists prefer traveling on cycling facilities. Figure 1 illustrates the raw data from the Mon Trajet Vélo smartphone application (red traces) and average daily bike counts at point locations from automatic bicycle counters (white circles with average daily bike counts given).
Intersections with the most collisions involving cyclists occur along the most used cycling facilities near the city center. The number of reported collisions involving a cyclist, by intersection, are plotted in Figure 2.
The cyclist collision rates at intersections throughout the city are illustrated in Figure 3. Most of the high-risk intersections are not located along cycling facilities. A total of nine intersections had an estimated collision rate of more than ten collisions per million cyclist trips, eight of which are not located on cycling facilities. The highest risk intersections appear along several corridors that form a connection between the suburban areas and the city center.

CONCLUSION
The estimated network-wide cycling activity has several applications. Firstly, it provides a cycling heatmap that can help identify where road treatments and maintenance should be prioritized and where bicycle parking is needed. Secondly, cycling activity is required as a level of exposure to estimate cycling risk. Maps are generated that identify the most high-risk intersections (defined as collisions per million cyclists) in the city. Lastly, desire lines are determined by identifying the streets and corridors with heavy cyclist activity, despite having no cycling infrastructure.

This research demonstrates a practice-ready approach for estimating bicycle activity across a network. When paired with collision data and the existing bicycle network, the results can help inform transportation departments in planning for and designing safe infrastructure.

Presenter: DAVID BEITEL
Agency Affiliation: Eco-Counter

Presenter Biographical Statement: [biography]