Integrating road traffic externalities through a sustainability indicator

Road traffic poses negative externalities on society and represents a key challenge in sustainable transportation. However, the existing literature about the assessment of traffic externalities drawn on a common measure is scarce. This paper develops a sustainability indicator that integrates traffic-related externalities as means of traffic congestion, noise, greenhouse gases (GHG) and nitrogen oxides emissions, health impacts and road crash related costs, and adjusted to local contexts of vulnerability. Traffic, road crashes, acoustic and vehicle dynamic data were collected from one real-world intercity corridor pair comprising three alternative routes. The site-specific operations were characterized using a modeling platform of traffic, emissions, noise and air quality. A specific methodology is applied for each road traffic externality and translated in a single factor – external cost. The results indicated that road crashes presented the largest share in the partly rural/urban route while GHG emissions had the highest contribution in external costs for the highway routes. Also, the distribution of external cost component varied according to the type of road, mostly due to different levels of exposed inhabitants. This paper offers a line of research that produced a method for decision-makers with a reliable and flexible cost analysis aimed at reducing the negative impacts of road traffic. It also encourages the design of eco-traffic management policies considering the perspective of drivers, commuters and population. Traffic Externalities.


INTRODUCTION AND RESEARCH OBJECTIVES
Road traffic poses negative externalities on society, thereby representing one of the key challenges in sustainable transportation nowadays. In 2016, road transportation accounted for 73% and 83% of transportation greenhouse gases (GHG) emissions in the European Union (EU) (EEA, 2017b) and in the United States (US) (EPA, 2018), respectively. Long term-projections for carbon dioxide (CO2) emissions concerning the passenger transportation in cities of over 300 000 inhabitants show an increase up to 27% in 2050 compared with 2015 levels (Chen and Kauppila, 2017).
Besides GHG emissions, road transportation has long-lasting negative impacts on road safety, human health and wellbeing. Road traffic crashes within EU claimed approximately 25,650 fatalities in 2016 (ERSO, 2018); 54% of these occurred at rural roads (ERSO, 2018). Also, road transportation is one of the major sources of some harmful air pollutants such as particulate matter (PM), nitrogen oxides (NOX) and carbon monoxides (CO) (EEA, 2018a). Around 39% of total NOX came from road transportation (EU member states), which represented the highest share of that gas in 2015 (EEA, 2017a). This sector is, by far, the dominant source of traffic noise in Europe, representing almost 90% of total noise emissions (EEA, 2018b). Approximately 29 million living in main roads outside urban areas in EU-28 were exposed to average day-eveningnight noise levels (Lden) exceeding 55 dBA (EEA, 2018b). Traffic noise causes nuisance, stress reactions, sleep disturbance, and it also has negative effects on health, such as cardiovascular diseases (WHO, 2011).
Understanding the most cost-effective strategies to mitigate both traffic congestion and environmental related costs in road trips has been pointed out as one of the critical issues in transportation for the next 20 years (National Academies of Sciences, 2018). The overall size of transportation external costs is estimated at around 7% of the EU Gross Domestic Product (EC, 2018). In this context, a more efficient use of existing infrastructure is essential to reduce road transportation externalities (EC, 2011).
However, there are some answered questions about the quantification of external costs namely: • What would be the cost of a given route if drivers had to pay for their choices?
• Why would a driver have to choose the route with lowest emissions if local population could be at higher risk exposure to other traffic externalities?
• If drivers shift towards a fast route but with high traffic volumes and resulting pollutant emissions, then what would be the benefits in terms of overall costs compared to slower routes?
For this purpose, a simulation-based approach was conducted combining a methodology for estimating GHG (CO2 and Volatile Organic Compounds -VOCs), NOX and PM emissions, air quality (PM concentrations) and noise using a microscopic traffic simulation tool together with road crashes historical data in a real origin-destination (N-S and S-N) pair between Aveiro and Estarreja, Portugal. The location comprised three alternative routes, as follows: i) partly rural/urban; ii) low-traffic-volume highway with electronic pay tolls; and iii) high-traffic-volume highway with both conventional and electronic pay tolls. The proposed methodology allows to build a link-based sustainability indicator that can be updated in real-time through a set of information sources and translated into a monetary value.
This research intends to contribute for decision making by traffic management entities in the following aspects: • To endow the current navigation platforms with reliable and flexible cost analysis which takes into account local-specific needs; • To include other variables in order to assess their impact on the magnitude and share of traffic externalities according to the type of road; • To encourage the design of eco-traffic management policies considering the perspective of drivers, commuters and population.
The remainder of the paper is organized as follows. Section 2 presents a review of scientific literature regarding the integration of road traffic externalities. In Section 3, the methodology for traffic, vehicular emissions, air quality and noise modeling, and calibration and validation of the simulation platform are presented, as well as the procedure for developing the proposed sustainability indicator. Section 4 describes the real-world intercity corridor, data collection and main modeling tasks. Subsequently, the results are used to assess the sustainable indicator in the candidate case study (Section 5). In all comparisons, the focus will be on range of each cost component value along routes, and potential trade-off among them. The final section outlines the main research findings and contributions and points out some future research needs (Section 6).

LITERATURE REVIEW
Internalizing the external costs of transportation has been an important concern for policy development and transportation research. According to Korzhenevych et al. (2014), internalization of transportation externalities can be based on quantifying in monetary values the associated impacts on society and environment, such as congestion, traffic noise, air pollution, greenhouse effects and road crashes. This degree of damage widely depends on the geographic conditions, intensity of traffic and population exposed (Yeh, 2013).
Negative externalities in the road transportation sector constitute an important development issue with socioeconomic costs (Cecchel et al., 2018) which are known to lead to welfare losses market inefficiencies (Kickhöfer and Kern, 2015). Usually, transportation users only account for marginal private costs, which may lead to welfare losses, since marginal social costs are neglected. To overcome such issues, some authors have been proposed to internalize the difference between generalized prices and marginal social costs by a tool [e.g. (Friesz et al., 2004;Small and Verhoef, 2007)]. However, they focused only on congestion effects. Road vehicles also give rise to side effects such as the productivity losses due to lives lost in road crashes, health costs caused by air or the abatements costs due to climate impacts (Bandeira et al., 2018a;Int Panis et al., 2004;Korzhenevych et al., 2014;Yeh, 2013).
Despite its relevance, the existing literature about the assessment of traffic externalities drawn on a common measure (e.g., sustainability indicator) is scarce El-Rashidy and Grant-Muller, 2015;Kickhöfer and Nagel, 2016;Sdoukopoulos et al., 2019;Torrao et al., 2016) and mostly focused in urban areas (Bandeira et al., 2018b;Sampaio et al., 2019;Tafidis et al., 2017;Yeh, 2013). Torrao et al. (2016) developed a safety, energy efficiency and green indicator based on crash consequences and type, and vehicle characteristics. The models neither accounted with impacts of changes in modal operation, nor included traffic volume as input. Kickhöfer and Nagel (2016) used an agent-based model to internalize air quality costs taking into account both traffic congestion and vehicle characteristics, but they focused only roads in urban areas.
Although rural roads represent 80% of the total road network length in developing countries (Rivera et al., 2015), the development of link-based indicators reflecting traffic-related impacts for this type of roads is little explored. El-Rashidy and Grant-Muller (2015) introduced a fuzzy logic model for assessing the mobility of road transportation networks. The model incorporated a physical connectivity attribute and traffic condition as mobility attributes and was successful tested for different intercity routes. Fernandes et al. (2018) analyzed the impacts of partialmetering strategies at a rural corridor near a shopping mall to reduce emissions, noise and user perspective costs. The proposed system resulted in improvements (up to 13%) compared to the unmetered conditions. Recently, Chang et al. (2018) developed a road pricing model that integrated travel time, CO2 emissions and safety costs by combining them on a green safety indicator for evaluating the level of service in freeway traffic. However, the authors discarded impacts of local pollutants, such as PM.
Link-based indicators can be applied into advanced traffic management systems as vehicle routing problems, but existing literature around this topic is mostly focused on the use of empirical models for route choice optimization in urban areas (Ćirović et al., 2014;Jovanović et al., 2014;Pamučar et al., 2016a;Pamucar and Goran Ćirović, 2018;Pamučar et al., 2016b).
Thus, the following gaps in the literature review were revealed: i) none of the prior studies developed a sustainability indicator for integrating traffic externalities according to the road type, i.e., urban, rural and highway; ii) little is known about the impacts of site operational characteristics on each externality cost value; iii) few studies applied reliable methods for gathering the number of exposure people, who are directly affected to noise, NOX and PM.
The novelty of this research relies in the following aspects: i) To use a simulation-based approach for quantifying and assessing external costs of road traffic at urban, rural and highway scales; ii) To include a trade-off analysis among traffic externalities;

METHODOLOGY
The core idea of the methodology was to use and test a modeling platform to evaluate external costs of road transportation at a segment level. It proceeded in five steps, illustrated in FIGURE 1. The development of the sustainability indicator involved first, collecting traffic volumes, noise, vehicle dynamic (second-by-second speed, acceleration and slope), crash data and population per unit square from one real-world intercity corridor. Second, the modeling platform was calibrated and validated, and then, studied location was divided into multiple sub-segments according to the road type. Finally, external costs of road transportation (Korzhenevych et al., 2014) were computed to obtain the sustainability performance measure in monetary values.  , and geo-processing tolls and air quality models (Borrego et al., 2016;Dias et al., 2018).

Pollutant Emissions
CO2 and NOX generated by Light Duty vehicles -LDV, i.e., PGV, PDV and LCDV were estimated using the VSP-based modeling approach that provides instantaneous vehicle power per unit mass (US EPA, 2002). This regression-based model is sensitive to changes in vehicle dynamic data and offers significant explanatory power for vehicle energy use and emissions rates IOVs (Hu et al., 2016). The use of VSP is justified because a speed-based approach as EMEP/EEA methodology, per se, is less robust to assess emissions of traffic singularities (roundabouts, traffic lights, toll plazas or stop-controlled intersections) and driving behavior states (acceleration, overtaking or gap acceptance) which in turn have impact on GHG and NOX external costs. VSP values are stratified into 14 bins, which in turn correspond to an emission factor on a second-by-second basis (US EPA, 2002). VSP is a function of speed, accelerationdeceleration and slope, as shown in Equation 1 for LDV (US EPA, 2002): where v is the instantaneous speed (m/s); a represents the instantaneous acceleration/deceleration (m/s2), and grade is the road slope (in decimal fraction).
Since VSP accounts for changes in vehicle dynamic with high resolution time, it shows as proper methodology for the quantification of exhaust emissions generated by PGV (Anya et al., 2013), PDV (Coelho et al., 2009), and LCDV (Coelho et al., 2009). A good body of literature has documented the effective use of VSP in assessing vehicular emissions in real-world urban, rural and highway routes (Anya et al., 2013;Coelho et al., 2009;Khan and Frey, 2018).
To obtain emissions estimates for HDV (CO2, NOX, VOCs and PM) and LDV (PM and VOCs), the EMEP/EEA method was used (EEA, 2013). It uses emission factors for diesel HDV from Euro I to VI emission standards and engine capacities as a function of the average speed (EEA, 2013). It must be stressed that EMEP/EEA is less robust to analysis emissions in traffic interruptions (e.g., roundabouts, toll plazas and traffic lights) that are characterized by high stop and go episodes Vicente et al., 2018). For instance, if this methodology was the only used, then the vehicular emissions would be underestimated.
A GUI application in MATLAB was conceived and developed to compute second-by-second LDV and HDV dynamics data from VISSIM output (speed, acceleration and slope). LDV and HDV emissions were summed up and further assigned to a segment. Then, such information incorporated on a GIS platform to assess pollutant concentrations, as described in the following section.

Air Quality
The air quality at the urban scale were evaluated by applying the air quality modeling system URBan AIR (URBAIR) Valente et al., 2014). The URBAIR model is an improved version of the second generation Gaussian model POLARIS developed by Borrego et al. (1997), differing from traditional Gaussian dispersion models in what concerns its dispersion parameters, which have a continuous variation with the atmospheric stability, and it accounts for building-induced dispersion mechanisms.

This steady state atmospheric dispersion model is based on boundary layer scaling parameters
and is suitable to be used for distances up to about 10 km from the source. The URBAIR modelling system is designed to be modular and includes the pre-processing of land use and urban elements geometry (GIS-based), meteorological conditions and air pollutant emissions, coupled with a dispersion module. The system framework is designed in such a way that the inputs/outputs of the different modules are shared and linked along the modeling process.
The meteorological model calculates a set of meteorological parameters, such as atmospheric turbulence characteristics, mixing height, friction velocity, Monin-Obukhov length and surface heat flux, using as initial conditions, or measured data. Since the topography and build-up structure characteristics have a significant influence on the dispersion of atmospheric pollutants, particularly in urban areas, URBAIR also requires characterization of the spatial variation of terrain surface elevation, buildings 3D coordinates and roads 2D coordinates. For simplicity, buildings can be assembled based on proximity and geometry criteria.
URBAIR considers different types of source emissions, namely, area, volume, point (such as industrial facilities and combustion activities for residential and services sectors) and line sources (road traffic emissions). As outputs, URBAIR provides air quality patterns for a given spatial domain (with up to about 50 km from the domain center) and time period (e.g., hourly, daily, one year or multiple years) for different air pollutants, namely: PM10, Nitrogen dioxide (NO2), Sulfur dioxide (SO2) and CO.
URBAIR model has been widely applied and extensively tested, having showed capability to produce robust and realistic results. Recent works showed its usefulness and capability to perform air quality studies at urban scale (Borrego et al., 2016;Dias et al., 2018).
In this study, URBAIR model was selected for two main reasons: 1) it is designed to assess the impact of urban planning and traffic management on air quality; 2) it is an advanced Gaussian model that has been enhanced with several major features, mainly the treatment of road traffic emissions and 3D urban elements.

Noise
The prediction of noise levels was made using a numerical approach developed by Quartieri et al. (2010). This procedure relates directly the acoustical energy sent to a receiver to the number of vehicles, to the source-receiver distance and to the mean traffic speed. The above information is used to assess source power levels and then, equivalent noise levels for a particular segment k (Leq,k), which are obtained at a fixed distance d, according to the distance between the road axis and the receiver. Equation 2 gives the hourly equivalent noise level by segment (Guarnaccia, 2013): where Leq, k is the segment-specific equivalent noise level (dBA); VLDV and VHDV are the hourly LDV and HDV, respectively, volumes (vph); n represents the acoustic equivalent, i.e., the number of LDV that produce the same noise of a HDV; vk is the segment-specific average speed (km.h-1); d -Distance between the road axis and the receiver (m) (Quartieri et al., 2010).
The advantage of this type of semi dynamic noise model is that only information about vehicle speed and traffic volumes for a given segment is needed. This means that there is no need of new noise equation for every other region or country.
To obtain day-evening-night level (Lden,k) on a segment k (dBA), the hourly segment-specific equivalent level (Leq,k) was assumed to be the same during all day. This is a conservative assumption since during the night traffic noise is usually lower than during daytime (EEA, 2018b).

Calibration and Validation
The modeling platform was calibrated and validated using field data collected from the studied location. The data were divided in training (70%) and testing (30%) sets (Liu et al., 2017), randomly selected before calibration procedure. The following strategy was used: • Capacity Calibration -Simulated and observed traffic volumes were compared for each monitoring point. The stopping criterion for this step was: at least 85% must meet the criteria of GEH (acronym for Geoffrey E. Havers) < 4 (Yu and Fan, 2017); • Route Choice and Noise Calibration -Simulated travel time per each route as well as noise were compared against the training data. The procedure stops when the difference in sample mean was not statistically significant within a 95% confidence level (p-value < 0.05); • Route Choice and Noise Validation -Site-specific simulated and testing set of travel time and noise were compared with 10 random seed runs (Winnie et al., 2014).

Sustainability Indicator
The proposed sustainability indicator is intended to account monetary costs per vehicle (€.veh-1) from road transportation activities in terms of: 1) congestion; 2) noise; 3) GHG; 4) NOX; 5) health impacts; and 6) road crashes. The following paragraphs describe in detail each cost component calculations.

Traffic Congestion
For a given segment, depending on the road type, congestion level is represented by the volume- Each segment-specific V/C ratio results in five congestion levels, as follows (Korzhenevych et al., 2014): 1 (free-flow) -V/C < 0.25; 2 -if 0.25 < V/C < 0.50; 3 -0.50 < V/C < 0.75; 4 (near capacity) -0.75 < V/C < 1; 5 (over capacity) V/C > 1. Each level is then, associated to a congestion cost (CCk) on a segment that can be adjusted to the local conditions, road type and vehicle type (Korzhenevych et al., 2014), as given by Equations 4 to 7: where TCk is the traffic congestion cost on a segment k (€.veh-1); cLDV and cHDV are the local congestion costs for LDV and HDV, respectively, depending on the V/C according to the type of road (urban, rural and highway) (€/veh.km); lk is the length of the segment k (km); and Li is the level of congestion, which also depends on the V/C (i = 1,…,5).

Noise
The approach for estimating segment-specific noise costs is based on the cost of noise in €/dBA per exposed person and per hour of the local population potentially exposed to a certain noise range considering the LDV and HDV traffic in kilometers traveled, as given by Equation 8: where Nk is the noise cost on a segment k (€.veh-1); CLden, k is the cost of a given day-eveningnight noise level Lden,k (€/dBA per person and per year) adjusted to the local conditions and type of road (Korzhenevych et al., 2014); popk is the number of individuals potentially exposed to the noise level Lden, k (inhabitants per km of segment length) that is represented by local population; and a and b are equal to 365 (number of days) and 24 (number of hours), respectively.

GHG
In this paper, CO2 and VOCs emissions were considered for the cost quantification related with the impact of GHG on environment, human health and economy. The cost estimation procedure involved three steps: 1) to compute emissions to the overall network according to the share of LDV and HDV; 2) to assign emissions to a segment; 3) to calculate segment-specific emission costs based in the costs provided in using Equation 9: where GHGk is the GHG cost on a segment k (€.veh-1); α1 is the local damage cost of CO2 (Korzhenevych et al., 2014) (€.g-1); vj is the share of the vehicle type j in the LDV vehicle park fleet; efCO2, j, k is the CO2 emission factor vehicle type j in the second of travel i on segment k (g.s-1); ECO2, HDV, k represents the HDV CO2 emissions on a segment k (g.s-1); Nk is the travel time on segment k (s); α 2 is the local damage cost of VOCs (Korzhenevych et al., 2014) (€.g-1); EVOCs, LDV, k represents the LDV VOCs emissions on a segment k (g.s-1); and EVOCs, HDV, k represents the HDV VOCs emissions on a segment k (g.s-1).

NOX
The quantification of NOX costs accounts for the impacts on local population which is represented by the ratio between segment population and national population densities, as given by where NOXk is the NOX cost on a segment k (€.veh-1); β is the local damage cost of NOX (Korzhenevych et al., 2014) (€.g-1); Dk is the number of individuals for segment k per square kilometer; DN is the national population density; efNOX, j, k is the NOX emission factor vehicle type j in the second of travel i on segment k (g.s-1); and ENOX, HDV, k represents the HDV NOX emissions on a segment k (g.s-1).

Health Impacts
Currently, it is well known that air pollution, mainly by the form of particles with an aerodynamic diameter smaller than 10 μm (PM10), is an important incentive for the development and exacerbation of respiratory diseases, such as asthma, chronic obstructive pulmonary disease or lung cancer, as well as a substantial impact on cardiovascular disease (Costa et al., 2014;Rückerl et al., 2011).
where HIk represents the health impacts cost on a segment k (€.veh-1); CRF is the correlation coefficient between the PM10 concentration variation and the probability of experiencing or avoiding a specific health indicator, which was set to 0.0004 YOLL/(person.year.µg.m-3) (EC, 2006); pop30,k is the number of individuals potentially exposed over 30 years (inhabitants per km of segment length); and ck is the average PM10 concentration on a segment k (µg.m-3).

Road Crashes
The level of external crash costs depends not only on the crash severity, but also on the insurance system, i.e., social costs of traffic-related crashes (Korzhenevych et al., 2014). These costs can be obtained by applying an adjusted risk that involves the following cost components: i) death and injury due to an accident for the person exposed to risk; ii) for the relatives and friends of the person exposed to risk; and iii) crash cost for the rest of the society. These considerations are summarized in Equation 12: where RCk is the road crash cost on a segment k (€.veh-1); XF, XSI, XLI are the annual numbers of fatalities, serious and light injury cases, respectively, on a segment k; and SCF, SCSI, SCLI represent the average social accident costs (€) for crashes involving fatalities, serious and light injuries, respectively, adjusted to local conditions.

External Cost by segment and by route
The total external cost on a segment k is defined as the sum of the above cost components for a segment, and denoted as ECk (€.veh-1), as expressed by Equation 13: Lastly, the external cost associated to a route r for a specific travelling direction, here denoted

CASE STUDY
An origin-destination (South to North; North to South) pair, comprising three parallel alternative routes, was sought out for this research. Prior research carried out in this area have shown that road type has impact on pollutant emissions (Bandeira et al., 2013). This intercity corridor provides a direct connection between Aveiro and Estarreja (Portugal) and is near a high-density industrial complex with moderate HDV traffic; hence, the air quality and traffic-related noise can represent an important issue, especially for local population. These routes were chosen based on their different specificities. The routes include urban (with speed (s) limits in the range 0≤ s ≤50 km/h), rural (50≤ s ≤90 km/h) and highway (90≤ s ≤120 km/h) trip sections (FIGURE 2-b). R1 is partly conducted on a rural (63%) and urban (37%) roads, while R2 is mostly a low-trafficvolume section (75%) traversing A29 highway, which has 2 lanes on each direction and an electronic pay toll system. Approximately 65% of R3 is on a high-traffic-volume section along A1 highway, with 2 lanes on each direction, and it includes both conventional and electronic pay toll systems. Average daily traffic (ADT) on A1 and A29 study segments is about 39 950 and 11 700, respectively (IMT, 2019). It must be noted that the classification of roads was based on posted speed limits and also on population density (Korzhenevych et al., 2014).

Data collection
Traffic data were collected in morning (7:00AM-10AM), off-peak (11AM-2PM) and evening peak The database covered a total of 68 crash observations.

Case Study Coding
Posted speed limits along the study domain and gap acceptance (critical and follow-up headways) in roundabout approaches were considered taking into account local driving habits (Vasconcelos et al., 2013). The dwell time distribution at conventional pay tolls was assumed to be same for all gates (6.8-9.6 s) (Coelho et al., 2005). Concerning the EMEP/EEA methodology, the least squares fitting technique was used to find the data best-fitting curve to relate segment-specific average speed and emissions generated by local HDV and LDV taking into account the above car fleet composition (EMISIA, 2017) and considering representatives vehicles and their emission standards, the annual activity (vehicle kilometers traveled per year), and engine size and capacity of the vehicle. Bus activity was also ignored since it represented less than 1% of corridor-specific traffic.

Segments Definition
The study domain was divided into multiple segments to compute each cost component and

Marginal cost factors
The marginal cost factors for the proposed sustainability indicator defined in Section 2.2 are presented in TABLE 2 and TABLE 3 for (Korzhenevych et al., 2014).

RESULTS AND DISCUSSION
In this section, the main results from the field data are analyzed (Section 4.1) followed by the calibration and validation of the modeling platform (Section 4.2), and finally, a representation of the external costs for the studied location is presented (Section 4.3).

Field Data
The analysis of field data suggested the peak hour occurred between 5:30-6:30PM. Thus, such period was selected for the assessment of road transportation external costs.

The hourly traffic volumes distribution (both travelling directions) along the study domain is
shown in FIGURE 5. The number of vehicles in R1 ranged from 922 to 1 108 vph on rural roads.
The difference in the number of vehicles on urban area (from 1 276 to 780 vph) was due to the fact that a portion of traffic diverted from R1 to the downtown city center. Field results suggest that the R3 traffic volumes are three times higher than R2 values. This happens because R3 serves through-traffic between Northbound and Southbound, and it is the main interchange for Eastbound-Westbound traffic. It is worth to notice that HDV represented nearly 3%, 4% and 9% of R1, R2 and R3 traffic composition, respectively, in the studied location.  (47), and it also recorded one fatal crash; and 5) main blackspots were located in influence area of roundabouts and traffic lights along R1 (e.g., segments 2-17, 3-16, 4-15 and 8-9) and R3 highway trip sections (e.g., segments 9-20).

Calibration and Validation
The statistical indicators of the modeling platform showed solid results. For traffic, the calibration target suggested in the literature was accomplished, i.e., GEH was lower than 4 in 39 out of 42 monitoring points (93%) (Yu and Fan, 2017). It should be emphasized that HDV traffic distributions were used in the traffic modeling.
The comparison of simulated and training travel time was performed using 30 floating car runs.
The relative difference in average travel time was lower than 5% (p-value > 0.05, and thus, not statistically significant), as shown in TABLE 4. During calibration, vehicle speed distributions, critical headways at roundabouts, and green times and cycle length at traffic lights were adjusted to fit travel time data. The comparison of testing and estimated travel time sets also demonstrated good degree of consistency (1-6%, depending on the route); no route showed significant differences at a 95% confidence level (p-value between 0.10 and 0.67).

TABLE 4 Summary of Calibration and validation of travel times.
It was also found that the noise estimates using the proposed methodology (Quartieri et al., 2010) matched the field measurements (training test). Under high noise values, the model tends to overestimate experimental data. This happens because field measurements taken at bridges end up being affected by a screening due to the bridge itself, even considering diffraction, i.e., noise emitted by vehicles outside the viewing angle of sound level meter. The predicted coefficient of determination (R2) was almost 80% for simulated Leq using a linear regression analysis (FIGURE 7a-b). An identical trend was observed for noise validation (testing set fit simulated data in 84%).

External Costs
This section presents the main results regarding external costs associated to the road traffic with existing conditions. The sum of each segment costs (EC) along each route confirmed R2 as the best option for the study domain (FIGURE 8 a-f). For instance, if one driver chooses R2 from south to north direction, then one could save 28% and 32% in external costs when compared with R1 and R3, respectively. Since vehicles were subjected to stop-and-go situations at conventional pay tolls (impact on emissions as demonstrated by Coelho et al. (2005) together with moderate traffic volumes in some of its segments, high external costs were observed for R3. For instance, segment with pay tolls accounted for approximately 10% of route external costs.
The analysis of the distribution of cost components along R1 showed the largest share corresponded to the RC-related costs; they represented around 31% and 30% of external costs in south-north and north-south directions, respectively. GHG showed as the largest contributor to external costs (40-45%, depending on travelling direction) in R2. For the latter route, results indicated the share of RC in south-north direction (16%) was higher than in north-south (9%).
This happened because one crash involving a serious injury was recorded in segment 4, resulting thus in high social costs (see Section 3.4 for those details). Almost half of external costs along R3 were based on GHG emissions, and more than 18% based on NOX. This was due to the fact HDV traffic is relevant in that route. In turn, other externalities (HI and TC) had slight impacts. The distribution of cost components differed from the type of road (FIGURE 9 a-c). The highest share of external costs per vehicle, which was about 33% of traffic-related costs in urban sections, was due to noise generated by road traffic. This happened because N is very sensitive to changes in potentially exposed population, which is clearly high in urban segments. Albeit small, NOX and PM10 represented together 35% of costs in urban areas thereby, reflecting its impacts on local population. The findings from rural sections suggested a different trend (GHG accounted for 33% of external costs, followed by RC, with 30%). Concerning the highway, it is interesting to note that GHG represented around 74% of the external costs, while N and NOX had small impacts (⁓10% each). From FIGURE 9, and as expected, traffic congestion had a small expression in external costs regardless of the type of road, which can be explained by the level of congestion along the study domain (Li <4) (Korzhenevych et al., 2014). The learning gained from the test of the proposed sustainability indicator in the real-world case study is promising, which makes possible its integration in current eco-routing systems using the methodology of this paper and apply it to any route. The sustainability indicator was capable of reflecting each externality weight in costs and identifying trade-off concerning the selection of different routes with different purposes. On the one side, if drivers are guided to a route with less GHG emissions, they can be guided to roads with higher noise or air quality levels, confirming thus, the relevance for a quantification of potential population exposure. On the other side, a faster route (e.g., R3) may not represent lower external costs when compared to a slower one, emission and road crashes costs could be significant in some of its stretches when levels of traffic flows are significant. In these circumstances, the eco-routing information should be provided for ensuring both marginal private and social costs.

CONCLUSIONS
.The integration of road traffic impacts in one single indicator was one major drawback for the use of advance traffic management systems for estimating external costs. This paper developed a sustainability indicator for quantifying traffic externalities as means of traffic congestion, noise, GHG, NOX, health impacts and road crash related costs. The proposed methodology was tested in a commuting corridor with three main alternative routes.
Low-traffic-volume highway yielded 28% and 32% lower external costs than other routes. Road crash costs presented the largest share along the partly rural/urban route while GHG costs were most significant in routes with highway trip sections. For the road-level analysis, some differences in the distribution of external costs can be highlighted. The share of noise and NOX in external costs were only significant in urban roads mostly due to higher potentially exposed population in those areas.