*2.2. Bicycle Network*

The municipality of Bologna has made substantial investments in bikeways during the past decade and to date the city offers 129 km bikeways of different types: exclusive access and mixed access with pedestrians or buses [25]. The bicycle network layout is composed of 13 main radial bicycle paths connecting the suburbs to the city center and many other bikeways connecting the radial bike-paths. The bikeway meters per citizen increased by 45% starting with 0.228 m/citizen in the year 2009 and reaching 0.330 m/citizen in 2018 [25]. This is an almost linearly-increasing expansion of the cycling infrastructure. The bike-network map is illustrated in Figure 2.

**Figure 2.** The bike-network map of Bologna [24] and study area (dashed lines).

#### **3. Bicycle Flow Analysis**

#### *3.1. Cyclists' Flows from Traditional Counting Methods*

During the period 2009–2018, manual and instrumental counts of cyclists were carried out by DICAM-Transport of the University of Bologna [30]: the bicycle counts were conducted from September to October of each year within the study area as shown in Figure 2. In recent years, counting has also been performed in May with the aim to evaluate the difference in bicycle flows between different periods of the same year. The locations of bicycle counters have been selected adopting representative and targeted locations: the sites include different geographic areas of the city, different types of bikeways, as well as "pinch points" (i.e., locations where cyclists must converge to cross a barrier) [10]. The 46 (bidirectional) road-sections monitored in 2018 are showed in Figure 3, highlighting the spatial distribution of measurement points. The monitored road-sections included the 13 main radial bicycle paths. Figure 3 also includes images of different typical bikeway types in Bologna.

**Figure 3.** Road sections monitored in 2018—Legend shows sections sorted by flow value.

Manual and instrumental counting was conducted at each road section from 08:30 to 10:30 on weekdays. The trips purpose during this time period is most likely commute trips for the purpose of "work" or "study". It is further assumed that commute trips have a clear destination, with a low occurrence of round-trips or random trips for recreational purposes. The total average flows increased between 2009 and 2018 by approximately 75%, which is significantly greater than the increase in bikeway meters per inhabitant in the same period.

Figure 4 shows the correlation between bikeway meters per inhabitant and the total average bicycle flows: each point represents one year from 2009 to 2018.

**Figure 4.** Regression function between length of cycleways per inhabitant and bike flows.

As shown in Figure 4, the total average bike flows are positively and highly correlated with the length of cycleways per inhabitant (R<sup>2</sup> = 0.96). In the city of Bologna, people use bicycles more often than in the past. Surely, such an increase in cycling is determined, like in other cities, by an integrated package of many di fferent and complementary measures, including infrastructure provision, pro-bicycle programs, supportive land-use planning and restrictions of car use [2]. However, today's bicycle network of Bologna connects the most popular origins and destinations, and the expansion of the cycling network has resulted in an increased level of safety as demonstrated by accident statistics [25]. The increasing bicycle use is also related to an increasing bicycle use of females, growing from a share of below 30% in 2009 to a share of 44% in 2018 [30].

Using the regression function from Figure 4, one can estimate that one additional centimeter of bikeway per inhabitant increases the average bicycle flow by approximately 100 cyclists per hour on the main sections of Bologna's bicycle network. Based on the length increase of the bicycle network, the estimated bicycle mode share is currently almost 10%, following the model proposed by Schweizer and Rupi [31]: their model describes the significant linear relationship between meters of cycling infrastructure per inhabitant and bike mode share (R<sup>2</sup> = 0.81), based on approximately 9000 questionnaires carried out in 14 cities in Central Europe.

#### *3.2. Map Matched Cyclists' Volumes*

A database with GPS traces has been obtained from a data collection initiative called the "European Cycling Challenge" (ECC) [32] which took place in May 2016. In particular, the city of Bologna participated in this initiative among other 51 cities from 18 European countries. In Bologna, 1123 participants, equal to 0.3% of the population, recorded the GPS traces of their bicycle trips during the month of May 2016 by means of a mobile phone application. Participation was on a voluntary basis. The total distance travelled by all participating cyclists was almost 200,000 km and the database contains over 7,998,000 GPS points, with 27,348 individual trips covering the entire road network of Bologna [32] (see Figure 5). There is an area in the southern part of Bologna (encircled in green on Figure 5), with a particularly low density of GPS points, most likely due to the mountains and gardens with bike paths in which the observed bicyclist activity is almost completely absent.

**Figure 5.** ECC 2016: observed cyclist activity in the Bologna network.

The present analysis focuses only on morning trips from 08:30 to 10:30 during work-days in order to be compatible with the manual and instrumental bicycle counts. During this period, 847 trips were

recorded, of which 42% were female and the average age was 38 years. The share of trips carried out by workers with respect to students and the users' gender are very similar to the last trips census survey of Bologna [33]. However, the census is referred to the active people that use all means of transport. In addition, the share of GPS traces recorded by females is very similar to the share of females observed during the manual counts. Consequently, the sample of cyclists recording the GPS traces is representative of the gender of the counted cyclists. Unfortunately, the ECC database contains no information concerning trip purposes.

In order to obtain bicycle flows on network links, the GPS data has been matched to the road-network based on open street map (OSM). The OSM data has been extracted for the Bologna metropolitan area and converted into a SUMO (Simulation of Urban Mobility) network [34] using a software extension called SUMOPy [35] as reported in Rupi and Schweizer [23]. The SUMO network has been manually corrected and enhanced, such that cyclists could potentially pass everywhere, including footpaths and the opposite direction of one-way roads (which is an illegal behavior in Italy). The final network contains 13,959 nodes and 38,324 links. The employed map matching algorithm is part of SUMOPy and based on a method proposed by Marchal et al. [36] and improved by Schweizer et al. [37]. In order to match the GPS points to network links with a high accuracy and to obtain a large number of correctly matched GPS traces, the entire map-matching analysis consists of four phases [23]: (i) an initial filtering process, (ii) the actual map matching process, (iii) a post-filtering process, and (iv) a final analysis of the matched routes. Initially, many GPS traces could not be matched to the network due to missing links or missing access. Successively, the reasons for the failed matching of trips have been analyzed in detail and missing network links or road access attributes have been added. Finally, the map-matching process has been repeated with an increased number of successfully matched trips.

After the map-matching process and a filtering process ensuring a low error rate, 4029 map-matched routes, collected from 842 users, have been used. These traces correspond to 91.6% of all traces recorded during the considered morning period. It is worth mentioning that this percentage is significantly higher than that reported in other studies [23,24]. Starting from these map-matched routes, the bicycle flows (as the number of cyclists passing through each network link per hour) have been evaluated.

#### *3.3. Estimated Cyclists' Volumes*

A linear regression between the cyclists counted with traditional methods and the number of map matched GPS traces with links overlapping the monitored road sections has been carried out. The map matched bicycle volumes have been multiplied by a coe fficient *c* in order to minimize the di fference between the measured flows and flows derived from GPS data.

The regression, shown in Figure 6, is based on the flow-comparison at 23 monitored sections (*c* = 0.91).

The slope of the linear regression function is almost equal to one, highlighting that the average of map-matched cyclist volumes are equal to the average of manually counted cyclist volumes.

The relatively high level of correlation between the measured flows and the flows from the map matched GPS traces is evident.

Given the significant correlation between the GPS dataset and traditional counts, the linear relation between both flow types has been used to determine the flows on all network links where GPS points have been detected. The resulting link flows in cyclists per hour per direction are shown in the Figure 7. This map is particularly useful to quantify the spatial distribution of ridership and provide important cycling exposure data for safety studies. Starting from this map, it is possible to obtain the OD matrix of cyclists, the chosen routes, and the bicycle flow on every link of the network. This is essential information for modelling the cyclists' route choice behavior and for planning the bicycle network.

**Figure 6.** Regression function between manually counted cyclists' volumes and map matched cyclists' volumes (May 2016).

**Figure 7.** Estimated unidirectional bicycle flows in cyclists per hour during workday morning peak hours (from 8:30 to 10:30). Flows only on network links where GPS points have been detected.

In addition, Strava provides the cyclist heatmap of all cities in the world for trips using the Strava app. The heatmap is calculated by counting and normalizing the number of lines connecting recorded GPS points [38]. The Strava app collects mainly recreational trips and in particular sport trips. The Strava density heatmap of recorded GPS points in Bologna is reported in Figure 8. This figure highlights how the recorded trips are also spread in the south of Bologna, in mountain routes as well as in gardens provided with bike paths (encircled in green). Instead, ECC's traces from 8:30 to 10:30

cover the main cycle ways that directly connect different parts of city, but there is an absence of trips in the south part, compared with the encircled area of Figure 5. This is probably due to the difference in trip purposes, supporting the hypothesis that the ECC sample contains few leisure trips.

**Figure 8.** The Strava heat-map of Bologna.
