Milan is working towards promoting cycling as a major mode of transportation. This article is the second part of a comprehensive study on the city’s cyclability. While the first part focused on identifying conflict areas and their relationship to accidents, the second part aims to deepen our understanding of the quality of cycling infrastructure, accident frequency, and their connections within Milan. The overall goal of these analyses is to support Milan’s transformation into a cycling-friendly city, empower residents to exercise their Right to Cycle, and foster evidence-based urban development.
Introduction
Cycling is an increasingly popular and efficient mode of transportation, praised for its environmental and health benefits. However, it remains a vulnerable means of travel, causing numerous injuries and fatalities in Milan (Bozzuto, Manfredini, Guastamacchia, 2024), highlighting the need for high-quality infrastructure to ensure rider safety. This study aims to analyse the historical development of Milan’s cycling network, assess the trends and locations of crashes from 2019 to 2023, and explore the relationship between the network quality and accident occurrences.
Enabling Data and Methodology
This study relies on five key geo-spatial and statistical datasets (see Table 1). Cycle counts are collected through Strava Metro, a GPS dataset gathering cycle trajectories of its users, that quantifies monthly cycle volumes on street segments. Vehicular counts are obtained from a proprietary traffic model, that estimates the peak hour flows of cars and heavy vehicles on street segments. Geo-localized bicycle crashes are sourced from AREU-Agenzia Regionale Emergenza Urgenza, these include all crashes that involved a cyclist between 2019 and 2023, at multiple severity levels. The historical cycling network since 2019 has been obtained using the Open Street Map (OSM) database. This is an open, collaborative database containing geographical information on the streets. Finally, we have also used the shapes of the urban roadway (streets and intersections) using the Database Topografico (DBT) of 2020.
The research methodology is structured into three distinct phases. The first phase involves an evaluation of the historical development of Milan’s cycling network using the OpenStreetMap dataset. This includes all bicycle-specific infrastructures such as cycleways, cycle lanes, pedestrian areas, and 30 km/h zones. The second phase focuses on assessing the evolution and locations of crashes that occurred between 2019 and 2023. This phase aims to identify accident hotspots, emphasizing that intersections are disproportionately represented in accident statistics. The final phase proposes metrics to predict the likelihood of an intersection being hazardous. To achieve this, each intersection is normalized with an index that considers surface area, bicycle, and car flows. Then, we introduce indicators such as the Level of Traffic Stress, intersection complexity, traffic lights density, the presence of crossings, and correlate these factors with the accident rate.
Results
Cycling network in Milan
A consistent increase in the length of cycling infrastructure has been observed, largely driven by the recent development of 30 km/h areas, the introduction of cycle lanes, and pedestrian areas (see Chart 1 and Figure 1). When compared to the Milan’s cycling development plan 2030, it is found that the completion rate has increased from 40% to 55% between 2019 and 2024 (see Figure 2) when expected to reach 100% by 2030.
Where do crashes occur in Milan
Map displaying the locations of accidents in Milan between 2019 and 2023.
Figure 3 Trend in the number of crashes involving cyclists
A significant majority of bicycle-car accidents, more than 95%, happen within 5 meters of street facilities, implying that these incidents predominantly occur in public street areas. Additionally, intersections are overrepresented in these crashes, accounting for 60% of them despite representing only 29% of the area, indicating a 3.7-fold increased risk (see Table 2). This highlights the importance of focusing the analysis on intersections, where the risks are greatest due to the frequent crossing paths of cyclists and cars (Isaksson-Hellman et al., 2017). The rest of the study will focus on intersections and seek to uncover connections between infrastructure and crash rates.
Crashes hotspots in crossings
The number of crashes occurrences at each intersection serves as the primary input. We observe an overrepresentation of crashes at larger intersections (see Figure 4), as their greater surface area and number of roads make them more susceptible to crashes.
Figure 5 Average Traffic Flow of Cars and Bikes at Intersections
Index = Surface x Car Flux x Bike Flux
We defined an index to normalize the results based on the surface area, the number of bicycles, and the number of cars transiting through each intersection (see Figure 5), as these elements are linked to the number of crashes. We then calculated the accident rate for each intersection relative to this index, defined as the number of crashes that occurred at the intersection divided by the index. The intersections were divided into four categories: Low (no crash since 2019), and the remaining were divided into three quantiles, classified by accident rate (see Figure 6).
Table 3 Number of intersections in each accident rate category and number of crashes covered by each
Level of traffic stress
The Level of Traffic Stress (LTS), developed by The Institute of Mineta (Furth et al., 2016), classifies streets based on their infrastructure safety and is a reliable indicator of a street’s danger when other information is unavailable (Chen et al., 2017). This indicator considers various factors, such as the number of car lanes, the presence of bike lanes, the road classification (primary, secondary, tertiary, etc.), and the speed limit. Streets are categorized into four stress levels, from lowest to highest (see Table 4). While typically assessed visually, some research suggests that Open Street Map’s data can efficiently perform this task (Wasserman et al., 2019).
We used the algorithm provided by the BikeOttawa association (see Figure 7), to compute the distribution of LTS categories. At each intersection, we summarized the average distributions by accident rate categories. We found that the proportion of LTS2 is positively correlated with accident rates, while the other three LTS categories are negatively correlated (see Figure 8). This indicates that there are no linear effects between different types of LTS and overall empirical safety. An intersection with both safe and dangerous infrastructures tends to be safer than one with more intermediate infrastructures.
Intersections complexity
The complexity of intersections represents the number of streets converging per hectare for each intersection. This provides an indication of the complexity of each intersection. We can observe that intersections with high accident rate have 22% more intersections per hectare than Mid-Low ones (see Figure 9).
Map displaying the density of crossing streets in each intersection in Milan.
Figure 9 Spatial density of crossings streets in each intersection category
Traffic Lights
A correlation is observed between the geographical density of traffic lights and the safety of intersections, with a 34% higher signalization density per hectare for mid-low accident rate intersections compared to high accident rate intersections (see Figure 10).
Map displaying the density of traffic lights in each intersection in Milan.
Figure 10 Spatial density of traffic lights in each intersection category
Crossings
A correlation is also observed between the length of zebra crossings per hectare and the safety of intersections, with a 61% higher density per hectare for High accident rate intersections compared to Mid-Low accident rate intersections (see Figure 11).
Map displaying the density of zebra crossings in each intersection in Milan.
Figure 11 Spatial density of zebra crossings in each intersection category
Conclusion and future work
The study emphasizes Milan’s continuous expansion of cycling infrastructure. Through identifying high-risk intersections and examining factors like Level of Traffic Stress and intersection complexity, this research provides practical insights to improve cyclist safety. Future studies could investigate variables such as lighting and weather conditions. A thorough quantitative analysis could model focused interventions and forecast their impact on enhancing Milan’s cycling infrastructure and lowering accident rates.
Acknowledgments
We thank AREU (Agenzia Regionale Emergenza Urgenza) for their fruitful collaboration and for sharing data. The analyzed data were treated according to the GDPR-General Data Protection Regulation (EU, 2016/679). This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
References
Bozzuto, P., Manfredini, F., Guastamacchia, E. (2024). THE ATLAS OF THE DEAD (and badly injured) CYCLISTS IN ITALY. Accessed July 9, 2024. Retrieved from: https://www.maudlab.polimi.it/2024/04/18/atlante-dei-morti-e-feriti-in-bici/
Chen, C., Anderson, J. C., Wang, H., Wang, Y., Vogt, R., & Hernandez, S. (2017). How bicycle level of traffic stress correlate with reported cyclist accidents injury severities: A geospatial and mixed logit analysis. Accident Analysis & Prevention, 108, 234–244. https://doi.org/10.1016/j.aap.2017.09.001
Furth, P. G., Mekuria, M. C., & Nixon, H. (2016). Network Connectivity for Low-Stress Bicycling. Transportation Research Record, 2587(1), 41–49. https://doi.org/10.3141/2587-06
Isaksson-Hellman I., Werneke J. (2017). Detailed description of bicycle and passenger car collisions based on insurance claims, Safety Science, Volume 92, 2017, Pages 330-337, ISSN 0925-7535, https://doi.org/10.1016/j.ssci.2016.02.008
Wasserman, D., Rixey, A., Zhou, X. (Elynor), Levitt, D., & Benjamin, M. (2019). Evaluating OpenStreetMap’s Performance Potential for Level of Traffic Stress Analysis. Transportation Research Record, 2673(4), 284–294. https://doi.org/10.1177/0361198119836772