Road Safety Analysis of Urban Roads: Case Study of an Italian Municipality

: Attention to the most vulnerable road users has grown rapidly over recent decades. The experience gained reveals an important number of fatalities due to accidents in urban branch roads. In this study, an analytical methodology for the calculation of urban branch road safety is proposed. The proposal relies on data collected during road safety inspections; therefore, it can be implemented even when historical data about trafﬁc volume or accidents are not available. It permits us to identify geometric, physical, functional, and transport-related defects, and elements which are causal factors of road accidents, in order to assess the risk of death or serious injuries for users. Trafﬁc volume, average speed, and expected consequences on vulnerable road users in case of an accident allow us to calculate both the level of danger of each homogeneous section which composes the road, and the hazard index of the overall branch. A case study is presented to implement the proposed methodology. The strategy proposed by the authors could have a signiﬁcant impact on the risk management of urban roads, and could be used in decision-making processes to design safer roads and improve the safety of existing roads.


Introduction
User safety is one of the most important issues in road design. The Organization for Economic Co-operation and Development (OECD) demonstrated that in the last two decades, the trend of road accidents is not positive all over the world [1]. The problem is serious, as confirmed by the attention given to it at international level [2]. The United Nations General Assembly declared the years 2011-2020 as a Decade of Action for Road Safety [3]. At the European level, the trend of road fatalities since 2001 appears not to be compliant with the Road Safety Program pursued by the European Commission, which aims at halving road casualties by 2020 [4].
A non-negligible rate of road accidents (73% in Italy according to [5]) occurs in urban areas, where externalities of motorized traffic (e.g., congestion, noise, and pollution) encourage the use of "soft mobility" [6,7]. In such conditions, different users and different vehicles share urban road spaces [8,9]. In the literature, several studies demonstrated that pedestrians, cyclists, and motorcyclists are the most vulnerable users in urban areas [10][11][12].
In order to prevent accidents and reduce their consequences on people, the system composed of drivers, vehicles, the environment, and roads should be investigated [8]. According to several studies available in the literature, most accidents are caused by bad behavior of drivers, i.e., the only ones capable of adapting their behavior towards non-living components [13,14]. However, it has been observed that the incorrect design or management of road infrastructure can induce bad behavior or failure to appreciate risk [15][16][17][18]. Therefore, in order to maximize on-going safety efforts, it is necessary to analyze each component of the road system and evaluate its effect on road safety. have uniform/homogeneous attributes related to physical and operating conditions (i.e., accident rate, geometric layout, composition of cross section, traffic spectrum, average operating speed) [35]. This approach allows for the comparison of "density" of hazardous elements/defects between different (short) sections whose length is suited to the considered urban context. Moreover, it gives good results in terms of fitting the safety performance of road sections, as confirmed by Cafiso et al. [36].
SFR j,r takes into account both the general characteristics of r (i.e., type and frequency of hazardous elements/defects, expected damage on vulnerable road users, and traffic level), and all n elements which affect the dangerousness of j. Equation (1), which complies with an approach recently proposed to analyze and plan maintenance of road safety barriers [32], allows us to calculate SFR j,r : (1) B i is the base value associated to defects i which are along j. K 1i is the priority factor of the category to which the element i belongs. K 2i is the vulnerability factor of users (i.e., pedestrians, cyclists and motorcyclists) along r; and it depends on their volume. K 3 is the motorized traffic factor of r. K 4i is the hazardousness factor; it depends on the consequences of defect i on the most vulnerable road users. K 5i is the extension factor; it depends on continuous or discrete elements/defects i along j.
Given SFR j,r , the corresponding value of SIR can be calculated according to Equation (2) SIR j,r = SFR j,r /SFR max,r × 100 (2) where SIR j,r is the Section Index Risk of the section j belonging to r, and SFR max,r is the highest value of SFR r by attributing the maximum values to defects found on r. Therefore, SFR j,r depends on the detected road elements/defects and the road users of the section j, and SIR j,r depends on the comparison between the attributed and the maximum values of K 1i , K 2i , K 3 , K 4i , and K 5i .
Similarly, the Branch Index Risk BIR r can be calculated according to Equation (3): where R r is the sum of the SIR j,r of m sections which compose the road branch r given by Equation (4) where R max,r is the reference value of the Risk Factor given by Equation (5) R max,r = m × SFR max,r Similar to SIR j,r , BIR r depends on the attributed values of K 1i , K 2i , K 3 , K 4i , and K 5i , and on the defects found along r. Both SIR j,r and BIR r range between 0 and 1.
The implementation of the proposed method requires road inspections to identify and categorize elements and/or defects of infrastructure which could cause accidents. The authors identified 9 categories of elements/defects: geometry (G), cross-section (C); private access (A); pavement (P); lighting (L); road signs (S); intersection (J); urban furniture (F); and stopping (ST). The attribution of possible values of K 1i , K 2i , K 4i , and K 5i , and B i required interviewing technicians from different backgrounds, as well as academic experts in the fields of roads (both geometry and safety), urban planning, transport management, and human health. Eight road engineers, eight urbanists, seven traffic managers, and six traumatologists were interviewed. The authors defined, for each variable, the maximum and minimum value, then each technician attributed the values within the established ranges. Finally, the geometric mean has been used to aggregate individual judgements. Table 1 lists the priority factors K 1 for each category.  (6) and are listed in Table 2. Road layout and geometry were evaluated against the Italian standards and recommendations about the geometric and functional characteristics of roads [37] and intersections [38], road infrastructure safety management [21], and road lighting [39]. These standards were considered as references to identify "not compliant" conditions, and to assign the B i values.
K 2i depends on the observed traffic condition of vulnerable users (i.e., pedestrians, cyclists, and motorcyclists) which are exposed to the risk induced by defect i. The Highway Capacity Manual [40] suggests performing surveys during a 15 min count period in order to obtain the traffic volume data. The authors observed the traffic flow when weather conditions were not an obstacle to traffic, and when all work-and school-related activities were ongoing. Finally, they calculated K 2i according to Equation (7).
where K Pi , K Ci , and K Mi depend on the pedestrian, cyclist, and motorcyclist flow, respectively. Their values satisfy Equations (8)- (10), and depend on the traffic volume observed along the overall examined branches, if the analysis involves a road network.
Particularly, having considered all the examined branches, the average hourly flow of cyclists and motorcyclists (ACF and AMF, respectively), and their standard deviation (DCF and DMF, respectively) are calculated. Finally, these values are compared to the hourly flow of cyclists and motorcyclists (ACF r and AMF r , respectively) and the standard deviations (DCF r and DMF r , respectively) of the examined branch r in order to calculate K Ci and K Mi , respectively (Table 3).  Table 3. Values of the vulnerability factor K 2i .

Users Observed Condition Traffic Flow Condition Code K 2 Values
Pedestrians Pedestrians not exposed to defect i -

K Pi
1.0 Not more than 5 commercial activities and/or any bus stops are in the examined section. Low 1.5 Not more than 10 commercial activities and/or not more than 2 bus stops are in the examined section Medium 2.0 The examined section is very attractive: more than 10 commercial activities and 2 bus stops, schools, public offices, or collective spaces are. Cyclists Cyclists are not exposed to defect i. Cyclists do not use or are not admitted -

Motorcyclists
Motorcyclists are not exposed to defect i. Motorcyclists do not use or are not admitted - The motorized traffic factor K 3 depends on the observed volume of motorized vehicles along the surveyed stretch. According to [41][42][43][44], the authors assumed that the higher their average speed, the greater the danger of exposure for vulnerable users. Therefore, the values of K 3 in Table 4 depend on the typical observed traffic fluidity and the number of daylight hours with congested traffic flow. The term "congested flow" complies with the methodology proposed by [45] to evaluate the level of service of an urban arterial: this condition occurs when the percentage speed reduction from free flow speed appears greater than 50%. The hazardousness factor K 4i depends on the expected fatality induced by the defect i on vulnerable road users. It is calculated according to Equation (11) where K 4Vi and K 4Pi refer to expected consequences on motorized vehicle users (both drivers and passengers), and non-motorized users (pedestrians and cyclists), respectively. Their values satisfy Equations (12) and (13), and are listed in Table 5. The established ranges for K 4Vi and K 4Pi take into account that at the same speed, consequences on vulnerable users are higher than on users of motorized vehicles. On the road, vulnerable users conflict with vehicles with larger dimensions and masses; cyclists and pedestrians clearly constitute the "unprotected" element, and they are more exposed in the event of an accident, as confirmed by [46]. Indeed, Wramborg defined probability curves to represent the fatality risk versus collision speed when a motorized vehicle collides with a vulnerable user, or frontal or hard object collision occurs to motorized vehicles.
K 5i is the extension factor of each element i found along j. According to Table 6, it considers both surveyed continuous and discrete elements/defects. Table 6. Values of the extension factor K 5i .

Continuous
Element/defect i is along less than one-third of section length Low 1.0 Element/defect i is along more than one-third and less than two-thirds of section length Medium 1.5 Element/defect i is along more than two-thirds of section length High 2.0 Discrete 1 element/defect i is along the examined section Low 1.5 2 elements/defects i are along the examined section Medium 2.0 More than 2 elements/defects i are along the examined section High 2.5 Given the above procedures, it is possible to simulate values of BIR s and define appropriate classes of risk. To this end, the authors considered six probabilistic classes of risk level, as usually done for transport infrastructure risk assessment [47][48][49][50][51][52][53]. The definition of ranges for each class requires a significant number of monitored branches. Monte Carlo simulations were used to generate a distribution of simulated BIR s obtained from randomly-assigned values of the proposed variables. Therefore, the Monte Carlo technique gave a virtual sample useful to study the statistical variability of BIR, assuming that the virtual sample was comparable to the real (surveyed) one. According to the Central Limit Theorem [54], as the number of samples from any population increases, the probability distribution of the means will approach a normal distribution. Therefore, it was possible to compare the distribution of simulated BIR s to a Gaussian distribution. Figure 1 shows the results from 3000 simulations, and it compares the simulated and analytical frequency curves: the former derives from the Monte Carlo simulation, the latter represents the Gaussian curve [55]. Figure 2 compares the cumulative frequency curves of the simulated and analytical distributions: their trend confirms the appropriateness of the procedure.         Table 7 lists the six classes of BIR, their minimum and maximum values, and chromatic categorization.    The proposed method could be applied by policy and decision makers when they have to manage urban road safety because it provides results that could be used to critically approach this strategic sector whose impacts are economic, social, and environmental. Indeed, input data are available to a road management body, and the presented method is comprehensive and versatile; therefore, it may be applied to different urban scenarios varying the examined variables and their factors.

Case Study
The proposed methodology has been applied on multiple branches totaling 50 km together in an Italian municipality in order to assess their BIR values. All the roads had the same classification: two-lane urban roads with parking spaces and sidewalks on both sides (Figure 3). Their maximum allowable speed was 50 km/h. The proposed method could be applied by policy and decision makers when they have to manage urban road safety because it provides results that could be used to critically approach this strategic sector whose impacts are economic, social, and environmental. Indeed, input data are available to a road management body, and the presented method is comprehensive and versatile; therefore, it may be applied to different urban scenarios varying the examined variables and their factors.

Case Study
The proposed methodology has been applied on multiple branches totaling 50 km together in an Italian municipality in order to assess their BIR values. All the roads had the same classification: two-lane urban roads with parking spaces and sidewalks on both sides (Figure 3). Their maximum allowable speed was 50 km/h. For the sake of brevity, in this study, the authors present the most interesting and critical roads, where users have to perform crucial and dangerous maneuvers (e.g., intersections without traffic lights, poor visibility, many conflict points). Table 8 summarizes geometric and technical characteristics of the selected roads: they are 7 branches (i.e., ROAD1 to ROAD7) with similar geometric and traffic characteristics.  Figure 4a to f show some of the most frequently-detected defects: irregular pavement, narrow sidewalk, inefficient lighting system, urban furniture occupies shoulder and/or lane, pedestrian crossing without ramps, and not visible traffic light, respectively. For the sake of brevity, in this study, the authors present the most interesting and critical roads, where users have to perform crucial and dangerous maneuvers (e.g., intersections without traffic lights, poor visibility, many conflict points). Table 8 summarizes geometric and technical characteristics of the selected roads: they are 7 branches (i.e., ROAD1 to ROAD7) with similar geometric and traffic characteristics.  Figure 4a to f show some of the most frequently-detected defects: irregular pavement, narrow sidewalk, inefficient lighting system, urban furniture occupies shoulder and/or lane, pedestrian crossing without ramps, and not visible traffic light, respectively.   The bar graph in Figure 5 represents, in descending order, BIR s (i.e., blue bars) to better compare them to the accident density (AD) (i.e., orange bars) retrieved by the authors from the Italian statistical geo-referenced database of occurred accidents [56]. The bar graph in Figure 5 represents, in descending order, BIRs (i.e., blue bars) to better compare them to the accident density (AD) (i.e., orange bars) retrieved by the authors from the Italian statistical geo-referenced database of occurred accidents [56]. In Figure 5, a good correlation between the data from surveys and the statistical data about accidents appears: the higher the accident density (AD), the greater the BIR values. ROAD1 and ROAD2 have the highest values of both BIR and AD. ROAD5 does not comply with this point because AD and BIR values refer to different conditions. Indeed, AD refers to 2015, while surveys to obtain BIR occurred in 2017, after safety works carried out in 2016. Therefore, ROAD5 has been removed from analysis of correlation between road defects and accidents in order to define Equation (14): Moreover, the analysis of SIRj,r allowed for the identification of the most hazardous sections according to Table 7. Figure 6 shows the results obtained for ROAD7. Its risk map is in Figure 7, where it is possible to identify sections 4 and 7 (i.e., orange and yellow bars, respectively).  In Figure 5, a good correlation between the data from surveys and the statistical data about accidents appears: the higher the accident density (AD), the greater the BIR values. ROAD1 and ROAD2 have the highest values of both BIR and AD. ROAD5 does not comply with this point because AD and BIR values refer to different conditions. Indeed, AD refers to 2015, while surveys to obtain BIR occurred in 2017, after safety works carried out in 2016. Therefore, ROAD5 has been removed from analysis of correlation between road defects and accidents in order to define Equation (14): Moreover, the analysis of SIR j,r allowed for the identification of the most hazardous sections according to Table 7. Figure 6 shows the results obtained for ROAD7. Its risk map is in Figure 7, where it is possible to identify sections 4 and 7 (i.e., orange and yellow bars, respectively). The bar graph in Figure 5 represents, in descending order, BIRs (i.e., blue bars) to better compare them to the accident density (AD) (i.e., orange bars) retrieved by the authors from the Italian statistical geo-referenced database of occurred accidents [56]. In Figure 5, a good correlation between the data from surveys and the statistical data about accidents appears: the higher the accident density (AD), the greater the BIR values. ROAD1 and ROAD2 have the highest values of both BIR and AD. ROAD5 does not comply with this point because AD and BIR values refer to different conditions. Indeed, AD refers to 2015, while surveys to obtain BIR occurred in 2017, after safety works carried out in 2016. Therefore, ROAD5 has been removed from analysis of correlation between road defects and accidents in order to define Equation (14): Moreover, the analysis of SIRj,r allowed for the identification of the most hazardous sections according to Table 7. Figure 6 shows the results obtained for ROAD7. Its risk map is in Figure 7, where it is possible to identify sections 4 and 7 (i.e., orange and yellow bars, respectively).  BIR of ROAD7 is equal to 9.15% (i.e., not relevant level of risk); however, according to Figures 6 and 7, it is possible to observe that section n. 4 has a high level of risk (i.e., 31.2%). Table 9 lists data used to perform the risk assessment.  The showed elements could cause rear-end collisions (Figure 8a), and head-on and lateral collisions (Figure 8b) along the examined section. Details about geo-referenced accidents are not available; therefore, it is not possible to correlate type of defects/elements and type of accidents: it is desirable that these data remain open for further investigation. BIR of ROAD7 is equal to 9.15% (i.e., not relevant level of risk); however, according to Figures 6  and 7, it is possible to observe that section n. 4 has a high level of risk (i.e., 31.2%). Table 9 lists data used to perform the risk assessment.  BIR of ROAD7 is equal to 9.15% (i.e., not relevant level of risk); however, according to Figures 6 and 7, it is possible to observe that section n. 4 has a high level of risk (i.e., 31.2%). Table 9 lists data used to perform the risk assessment.  The showed elements could cause rear-end collisions (Figure 8a), and head-on and lateral collisions (Figure 8b) along the examined section. Details about geo-referenced accidents are not available; therefore, it is not possible to correlate type of defects/elements and type of accidents: it is desirable that these data remain open for further investigation. The showed elements could cause rear-end collisions (Figure 8a), and head-on and lateral collisions (Figure 8b) along the examined section. Details about geo-referenced accidents are not available; therefore, it is not possible to correlate type of defects/elements and type of accidents: it is desirable that these data remain open for further investigation.
The implementation of the proposed methodology allows the management body to identify and make decisions about the strategic priorities for interventions of safety improvement at section and branch levels: geometric, functional, and traffic data of the branches contribute to the assessment of the accident risk for vulnerable users. Indeed, the results are numerical and synthetic, and in decision-making processes, they could allow the management body to identify and to determine the strategic priorities about urban road safety.
Moreover, the proposed model could be modified to assess the index risk of urban intersections. Indeed, the proposed procedure could be modified (e.g., modification of the extension factor), and new relevant coefficients could be calibrated. Finally, the implementation of the approach both at section and at intersection levels could permit us to pursue a network-level approach.

Conclusions
The road transport sector is currently adopting growing measures to prevent accidents and reduce their consequences on people, especially on the most vulnerable users (e.g., pedestrians and cyclists).
This paper presents a quantitative risk analysis of deaths and serious injuries caused by urban road accidents. The study proposed a methodology based on the visual inspection to interpret the results from Road Safety Inspections on urban roads, to quantify the safety conditions, and to direct the competent bodies towards the most appropriate interventions.
The method depends on the assumed ranges of variables and risk classes, as well as on the values attributed to the variables used for calculating the hazard index of examined homogeneous road sections and branches. Therefore, both the Section Index Risk (SIR) and the Branch Index Risk (BIR) depend on geometric, functional, physical, and environmental defects or elements which are potential source of road accidents. These factors are then related to the involved vulnerable road users and to existing traffic flows to assess the current levels of risk. The categorization of these values into six levels of risk allows the identification of the most severe conditions and the prioritization of road safety works.
The results from surveying 50 km of roads in an Italian municipality demonstrated the good performance of the proposed tool in identifying, planning, and scheduling all the work required for improving urban road safety, because it is sensitive to improvements of infrastructure.
Moreover, the proposed methodology has both a diagnostic purpose, in order to evaluate whether there may be a correlation between the observed defects and the occurred accidents, and a preventive purpose, in order to correct defects or anomalies that could cause death or serious injuries of road users.