Empirical assessment of route choice impact on emissions over different road types, traffic demands, and driving scenarios

Transcription

1 Empirical assessment of route choice impact on emissions over different road types, traffic demands, and driving scenarios Jorge M. Bandeira a, *, Dário O. Carvalho a, Asad J. Khattak b, Nagui M. Rouphail c, Tânia Fontes a, Paulo Fernandes a, Sérgio R. Pereira and Margarida C. Coelho a a University of Aveiro, Centre for Mechanical Technology and Automation / Department of Mechanical Engineering Campus Universitário de Santiago Aveiro Portugal, Phone: (+351) , Fax: (+351) jorgebandeira@ua.pt, doc@ua.pt, trfontes@ua.pt, paulo.fernandes@ua.pt, sergiofpereira@ua.pt, margarida.coelho@ua.pt b Old Dominion University, Norfolk, Civil & Environmental Engineering Department, 135 Kaufman Hall, VA United States of America, Phone: (757) , Fax: (757) , akhattak@odu.edu c Institute for Transportation Research and Education, Centennial Campus Box 8601, Raleigh, NC United States of America, Phone: (919) , Fax: (919) , rouphail@eos.ncsu.edu * Corresponding author September 2013 Submitted for publication at the International Journal of Sustainable Transportation

2 Abstract Eco-routing has been shown as a promising strategy to reduce emissions. However, during peak-periods, with limited additional capacity, the ecofriendliness of various routes may change. We have explored this issue empirically by covering about 13,300 km, in three different areas, using GPS equipped-vehicles to record second-by-second vehicle dynamics. This study has confirmed the importance of the eco-routing concept since the selection of eco-friendly routes can lead to significant emissions savings. Furthermore, these savings are expected to be practically unchanged during the peak period. However, some potential negative externalities may arise from purely dedicated eco-friendly navigation systems. Keywords Route choice; emissions; eco-routing; congestion; microscopic modelling, Vehicle Specific Power.

3 1. INTRODUCTION AND OBJECTIVES Despite the major importance of innovative vehicle technologies, traffic management plays a major role in reducing automobile emissions. Route selection and vehicle operation choices could lead to about a 45% reduction in the on-road fuel economy per driver (Sivak, and Schoettle, 2011). In the last years, the emergence of connected in-vehicle devices has allowed traffic information providers to efficiently collect road traffic data (European Commission, 2011) but currently, information about eco-friendly routes is not widely available. By reducing the barriers in terms of access and cost of data, the integration of traveller information systems and traffic management strategies may influence travel behaviour since users can adjust their departure times, destinations or route choices in response to more complete information, including eco-friendliness of alternative routes (US-EPA, 1998; Khattak, et al., 1996; Khattak and Khattak, 1998). During non-peak periods, route choice decisions can play a substantial role in terms of fuel consumption and principally local pollutants emissions reduction (Bandeira, et al., 2013). While in non-congested conditions there may be sufficient capacity to accommodate the demand of drivers who want to choose an eco-friendly route, during peak hours with limited capacity the eco-friendly routes may change (Frey et al., 2008). The effect of route choice in reducing emissions has been addressed by several authors. Table 1 lists the most relevant studies about route choice optimization, considering energy and atmospheric emissions. Furthermore, it indicates the methodology for emissions and fuel use estimation (m microscopic models; M Average speed models; f field measurements), the pollutants considered, and the highlights of each work. These studies can be divided into two main groups: in the first one (1-9), mathematical formulations to assign traffic on a virtual network considering atmospheric emissions were developed; and in the second one (10-22), field experiments were conducted and/or a wide range of models were applied to evaluate the impact of route selection in terms of emissions and energy use, over several case-studies. The research of Frey, Barth, Ahn and Rakha should be highlighted. Frey et al. (2008) carried out experiments using a Portable Emission Measurement System (PEMS) under real world driving cycles. Barth and their team developed and patented an environmentally-friendly navigation system (Barth and Boriboonsomsin, 2007; Barth, 2010). Ahn and Rakha (2008) studied two alternative routes using different type of emission models which allowed concluding that average speed models can yield incorrect conclusions. By modelling two large scales areas, the same authors concluded that the market penetration of eco-routing vehicles and the network characteristics are important factors that affect the benefits of eco-routing (Ahn et. al 2012). The majority of studies have concluded that route choice has a significant impact on emissions and energy use. Nevertheless, few studies have addressed the effect of congested periods on emissions (Zhang et al., 2011). The distribution of vehicle speeds and accelerations in traffic vary by type of road facility and amount of traffic volume, generating large discrepancies in emission levels (TUa, 2002). Possibly, this fact has contributed to some inconsistency on literature about this issue. Some studies pointed out that time

4 minimization paths often also minimize energy use and emissions (Barth and Boriboonsomsin, 2007; De Vlieger, 2000; Frey, et al., 2008). On the other hand, research demonstrated that frequently the faster alternatives are not the best from an environmental perspective (Ahn and Rakha, 2008; Bandeira et al, 2013; Minett, et al., 2011; Rilett and Benedek, 1994, Guo et. al 2013). Previous research studies indicate that it is not possible to generalize conclusions, considering limited study areas. Thus, more research is needed to evaluate a wider range of driving patterns conditions, namely at different periods of the day. A more extensive analysis including different scales, and different traffic volumes, as performed here, may better reflect the reality and improve the knowledge to develop further traffic management strategies. In this study we explore the impacts of route choice decision during peak and off peak periods. The information generated in this study can be useful in implementing innovative and sustainable traffic management strategies. This paper intends to focus on the following questions: Is there potential for significant emissions savings during congestion periods and in different contexts? How do traffic volumes affect emissions on different types of roads? How does modal distribution of a microscopic emission model vary over different periods of time and in different contexts? Can recurrent congestion affect the choice of an eco-friendly route? 2. MATERIAL AND METHODS In this section we describe briefly the process of collecting second-by-second vehicle dynamics data, then we present the study area and finally we describe the methodology for estimation of emissions impacts. 2.1 Experimental measurements During experimental tests, vehicles dynamics were recorded second-by-second, using GPS data-logger devices with a resolution of 5 Hz. We collected data for approximately 13,300 km of road coverage, over the course of 222 hours, in different traffic conditions both in Portugal and the USA: a medium-sized city (Aveiro, Portugal), an intercity region (Oporto Aveiro, Portugal), and a metropolitan area (Hampton Roads, VA, USA). In order to ensure realistic options, for all origin/destination pairs (OD pairs), the study routes were selected based on a web trip-planning software (Google-maps) suggestion. Figure 1 and shows the study maps The road tests were performed during weekdays under dry weather conditions during the months of February, March and April of 2010 and According to traffic volume data (Bandeira, et al, 2011; VDOT, 2011), the peak period in the Portugal (PT) site was considered between 7-9 AM and 5-7 PM while in USA the peak period was considered between 6-8 AM and 4-6 PM. According to historical ADT data, each one of these periods represent 13-14% of total ADT.

5 So, all trips whose departure time was within this time range are defined as peak hour tests. The off peak tests occurred between 10 AM-5 PM (PT) and 9 AM-4 PM (USA). The USA tests were performed using the same driver, while in the Portuguese case-studies, different drivers and vehicles (small family vehicles) were used. All drivers tried to avoid hard accelerations and keep the average speed of the traffic flow. 2.2 Description of the study areas To perform this study, three different areas were assessed: Urban (U), Metropolitan area (M) and Intercity (I). For each case different routes were selected: motorways (m), highways (h), urban roads (u) and arterial roads (a). To identify each route the following notation was used: the first capital letter identifies the study area and the second lower case letter identifies the dominant type of road on each route. Table 2 presents the route characteristics. Concerning intercity routes, we evaluated four parallel alternative routes that cross one of the most densely populated areas of Portugal (Figure 1a) between Oporto and Aveiro. Im 1 and Im 2 are based on motorways (m) (with a toll cost of 5). Routes Ih and Iu are mainly performed on a national highway (h) and a national road crossing urbanized areas (u) in 90% of its distance, respectively. Average daily traffic (ADT) on Im 1 and Im 2 is about 40,000 and 25,000, respectively (INIR, 2012). To estimate ADT on Ih and Iu, we collected 7 h of video data at 6 key points of Ih and Iu during the evening rush hour. Using a peak hourly factor of 8.22% (EP, 2010), it was estimated that Iu has an ADT of 11,500 and Ih of 17,500. Both for intercity and urban routes, off peak emissions and a detailed route characterization can be found elsewhere (Bandeira et al., 2013). Regarding urban routes three alternatives were evaluated: routes Um, Ua and Uu (Figure 1b). Route Um is predominantly driven on a motorway (m) (56%), route Ua essentially uses an arterial road (70%), and route Uu is entirely contained in a compact urban environment. In this urban area the ADT never exceeds 25,000 vehicles (Bandeira et al., 2011). All paths connect the city centre (university campus) to a point located in the suburbs (D). With respect to the metropolitan area, we have considered two alternatives routes in Hampton Roads, VA, connecting a university campus in Norfolk, to a large commercial area in Chesapeake, Ma and Mm (Figure 1c). Route Ma is mostly performed on arterial roads crossing downtown Norfolk. Regarding route Mm, 70% of the route distance is travelled through a motorway. Although route Mm presents more intersections, the majority of them are accesses to residential neighbourhoods with little impact on the main road. The ADT are clearly higher on USA routes, reaching values higher than 180,000 cars per day in some sections (VDOT, 2011) (see Figure 1). These routes have high quality data on traffic volumes for all road segments and road characteristics (VDOT, 2011) which allows a detailed analysis of the traffic volume impact on emissions. 2.3 Emissions estimation

6 We used the Vehicle Specific Power (VSP) microscopic model to estimate emissions which is based on the vehicle second-by-second speed, acceleration and slope. This methodology has demonstrated to be very convenient in assessing emissions from both gasoline (Frey et al, 2002) and diesel (Coelho et al. 2009) vehicles. Furthermore, the VSP methodology has proven to be suitable for standardize the comparisons of emission rates for different vehicles and routes (Frey, et al., 2008). The VSP values were categorized in 14 modes and an emission factor for each mode was used to estimate pollutants emissions (CO 2, CO, NO X and HC) from Light Duty Diesel Vehicles (LDDV<1.8 L) and Light Duty Gasoline Vehicles (LDGV<3.5L). Summing up, the calculation of the VSP variable follows the equation (1): where: [ ( ( )) ] (1) VSP = Vehicle Specific Power (kw.ton -1 ); v = speed (m.s -1 ); a = acceleration (m.s -2 ); grade = road grade (decimal fraction). The emissions rates used in this research are available in the references (Coelho, et al., 2009; Frey et.al, 2002). Total pollutants emissions by route were derived based on the average time spent in each VSP mode multiplied by its respective emission factor. The Kolmogorov-Sminorv test (K-S test) was used to assess if the VSP modal distribution between peak and off peak periods differed significantly on all routes performed. Additionally, a comprehensive link-based emissions analysis was performed for peak and off peak hours, using the second-by-second field data for the regional road segments. Using detailed 2009 traffic volumes of winter weekdays and 15- minute raw data, the average speed and total emissions estimated for each link (per lane) was correlated with the corresponding traffic volume reported for the respective period. Therefore, it is assumed that traffic volumes variation during experimental tests were similar to This is an important limitation since vehicle dynamics and traffic volume should be measured simultaneously (using loop sensors, for example). However, the margin of error may be minimal, given that the traffic historical data do not show significant variations of volumes over the last years, and no significant changes in the road network were made (VDOT, 2011). 2.4 Normalization of emissions costs and development of an eco-friendly indicator for route choice There is a lack of knowledge about the environmental impact of route choice, namely if the optimization of different pollutants dictate different routes (Gazis et al, 2012). In this context, an eco-friendly indicator for route choice was developed taking by normalizing the emission impacts of each pollutant. Both for European (Maibach et al, 2008), and USA routes (USDOT, 2012), specific

7 data on damage cost per mass of pollutant emissions, at the national scale were considered (Table 3). The presented values are estimated based on population density, country-specific meteorological conditions, social cost of carbon, price based derivation and contingent valuation (Maibach et al., 2008, USDOT, 2012). Although ideally it would be more accurate to have data based on the effect of each pollutant at a higher spatial resolution, the presented method is a first approach to ponder the health and social impacts of emissions in different countries (Portugal and USA). It should be noted the effect of emissions impacts has to be seen in the context of the full range of uncertainties mentioned in Maibach et al., (2008). A percentile-based indicator to translate the environmental cost impact of route choice over different two different traffic demand scenarios was created. Thus, a hypothetical user could identify either the periods with lower environmental impacts, and within each period, the most environmental friendly alternatives. The indicator colours range from red (> 90 th percentile), to orange (> 70 th percentile), yellow (> 50 th percentile) and green (<20 th percentile). To create an indicator for travel time, a similar approach was followed. 3. RESULTS AND DISCUSSION Firstly, we present an overview of speed data and total emissions per route. Then, an eco-friendly ranking for all routes is presented. In a second step of the analysis, we focus on the impact of differing congestion scenarios on specific links. In order to better understand the previous results a comparison of VSP modal distribution over different periods, is carried out. Finally, we discuss some implications for future eco-navigation systems. 3.1 Speed data and total emissions per route In this section we present the predicted emissions savings rate by pollutant that may occur by choosing an appropriate route. We also examine the influence of the average speed on emissions. Although we verified some slight differences on emissions according to the driver, the route choice was shown to be the main factor that control emissions (Bandeira, et al., 2013). However, it should be noted that driving behaviour plays an important role on total emissions and therefore the following results are only indicative. Table 4 presents the average and standard deviation of speed, the average time spent in different speed intervals and the average number of stops for off peak and peak periods. As expected, average speed decreases for all routes during peak periods. The higher difference is observed at routes Ih, Uh and Mm. Moreover the share of time traveling at lower speeds increase considerably in these routes as well as the number of stops during peak period. Table 5 presents the potential reductions/increases in emissions and environmental costs impact for each OD pair. To address possible trade-offs between travel time and emissions minimization, each route is compared with the fastest route for each OD pair. We present just one direction since the results were relatively similar in both directions. The data are broken down by

8 vehicle type (LDDV and LDGV), OD pair/route, and time period (off peak and peak). In intercity routes, the motorway options (Im 1 and Im 2 ) are clearly less time consuming than the alternative routes (Ih and Iu). Compared to the closer alternatives and during off peak periods, route Ih has a mean travel time 64% higher in relation to route Im 1, and route Iu has a mean travel time 90% higher than Im 2. During peak periods these differences are increased to 85% and 100% respectively. Taking into account the specific OD pair under analyses, it seems that the network is not going toward the user equilibrium. This can be explained by the local and regional traffic with different origin and destinations during peak periods, leading to higher travel time in these routes. Regarding LDGV, CO 2 emissions data show that motorway routes (Im 1 and Im 2 ) lead to less CO 2 emissions and consequently fuel consumption. In this case there is an evident trade-off between CO 2 and local pollutants minimization, since the routes that minimize local pollutants are the slower routes Ih and Iu. For local pollutants, the effect of peak demand is more obvious on the motorway routes Im 1 and Im 2. For Ih and Iu the local pollutants emissions do not change significantly during the peak period. Concerning urban routes, the peak demand effect is more evident on route Uu (centre-suburbs), although it is still the best route considering the minimization of pollutants emissions. This route yields the highest emissions rate per distance but its shorter length leads to a reduction in total emissions of all pollutants. On route Um a high variability in NO X emissions was noticed which can be to a certain extent explained by different driving behaviours (Bandeira et al., 2013). Regarding metropolitan routes, Mm has the highest emissions. Both at off peak and peak, Ma yielded CO 2 emissions saving from 3% up to 5% and CO savings of up to 12%. A more detailed analysis showed that on route Ma, CO emissions are mostly produced during the 6 km motorway section included in this route due to high engine-load conditions. Evaluating the average emission rates per distance, CO 2 and CO emissions rates on Ma were found to be 1.27 and 1.18 times higher than route Mm. However, route Mm is 1.30 times longer than Ma which makes route Ma and thus more environmentally friendly in terms of total emissions produced. Regarding the vehicle type, in terms of environmental damage (ED) costs, LDGV in Europe presents about 50% lower values than LDDV in the same routes. This is mainly explained by the lower NO X emissions levels produced by the gasoline vehicles. In USA, this difference is reduced to approximately 20% because CO emissions (produced principally from LDGV) are more valued than in the European approach for monetization of emissions. For all OD pairs, a slight decrease in the relative differences of environmental damage costs during peak periods is observed. However, both damage costs and emissions savings still to be significant. Figure 2 shows the evolution in terms of emissions per kilometre of a local pollutant CO 2 and CO from LDGV as function of average speed. Peak and off peak tests are displayed in solid and transparent background symbols respectively. Motorway routes are displayed in blue squares and blue rhombus, arterial routes in red triangles, and urban roads in golden circles.

9 For all OD pairs a general trend of decreasing CO 2 emissions (inter route and intra route) with average speed is clear. However, according to the literature (Barth and Boriboonsomsin, 2008) for average speeds values beyond the experimental range (>100 km.h -1 ), CO 2 emissions would tend to increase again. It can be also observed that during the peak period, CO 2 total emissions show a higher increase on the national roads (Ih, Iu), due to more congestion and higher travel times. In the case of intercity routes, the discontinuity between Iu and Ih is caused by variations in the topography of route Ih. This effect is studied in more detail in Bandeira et. al (2013). Although there is a higher variability in comparison to CO 2 emissions, it is visible a general trend to emissions increase with speed in the urban and intercity contexts (Figure 2a and 2b). A higher variability for both pollutants emissions values is observed in metropolitan routes, namely in Ma, as can be confirmed by the dispersion of emissions data points in Figure 2c. This can be explained by the fact that a significant distance of Ma is travelled through downtown and covering more signalized intersections. This contributes to a certain unpredictability in the speed profile. 3.2 Eco-routing indicator A suggestion for an eco-friendly route indicator summarizing the previous findings is presented in Table 6. Regarding intercity routes the most striking factor is that the best eco-friendly route depends on the type of vehicle. While for LDGV, route Im 1 is the best to minimize the environmental damage, for LDDV the most eco-friendly choice is the route Iu. Two main reasons contribute for this: a) in this indicator and in the European context, CO emissions are not valued; b) in terms of CO 2 emissions (and fuel consumption), LDDV are not as penalized by stop and go situations as LDGV. It is also possible to confirm that for all OD pairs, and for each type of vehicle, the eco-friendly route do not change between peak and off peak. However, in the metropolitan area, the most sustainable choice is to travel during off peak hours since during peak period both routes are constantly worse options than traveling during off peak. Finally, for almost cases, the selection of an ecofriendly route may result in an increase in travel time. The only exception to the trade-off (emissions vs. travel time) is for LDGV in the intercity context. 3.3 Link-based emissions Link-based emissions were estimated for peak and off peak hours, using the second-by-second field data for all road segments on metropolitan routes, where detailed traffic data were available (Figure 3). Although not statistically significant, in the majority of the sections, CO 2 emissions during peak are consistently higher than during off peak. Mm-4 and Ma-12 are the segments where the highest increase at peak hour is experienced. This can be explained by the frequent traffic jams that occur at peak hours since these links serve as connector to the belt roads I-64 and I-264/I-464, respectively. Mm-4 and Ma-12 represent approximately 3% and 8% of total CO 2 emissions on Mm and Ma.

10 In I-464 segments (Ma13 to Ma17), there are no significant differences between peak and off peak, because even at peak period, a low volume/capacity ratio is maintained. On the other hand, the I-64 segments (Mm5 to Mm14) are more vulnerable to higher traffic volumes, particularly on the northern sections Mm6 and Mm8. A slight decrease (but not significant) in emissions at peak period is observed on in a small number of sections of route Ma. Furthermore, the average speed during peak is higher than during off peak periods. A detailed analysis of speed data has shown that this occurs due to the coordination of traffic signals, providing more favourable traffic conditions during peak periods. Figure 4 shows CO 2 and NO X emissions under different traffic volumes on an arterial and a motorway. We did a rough calculation of the existing capacity in each lane and we estimated that in each motorway lane, there is a capacity of 1,700 vehicles per hour (vph) and in each arterial lane, 850 vph (considering an average g/c ratio for through traffic of 0.50). The analysed arterial segment is a 4-lane road with 500 m long covering three signalized intersections, one of which is at the interchange with the motorway I- 64 (Mm4). Figure 3 shows that Mm4 is the segment (in route Mm), where a higher difference in CO 2 emissions between peak and off peak is observed. Mm9 segment corresponds to a 3-lane I-64 section which extends over 2000 m. According VDOT data, during the peak period both segments have been classified as operating in Level of Service (LOS) E. For the arterial segment, emissions and average speed remain relatively constant up to a certain traffic volume level. From that point emissions start to increase exponentially. A third-order polynomial was used to fit the data points, shown as solid lines in Figure 4. Although more data points are needed to define a statistically valid trend, these results are consistent with previous research (Sugawara and Niemeier, 2002). A detailed analysis of speed profiles have demonstrated that on Mm4 segment the emissions are strongly dependent of congestion, namely the coordination between the traffic flow with the timing of traffic signals. Under congestion circumstances, an increase of one vehicle per hour generates an average increase in emissions of 0.1%. On the motorway segments, relatively high correlations between CO 2, and traffic volumes were found and the same trend is observed. Thus, for traffic volume close to the capacity estimated, CO 2 emissions start to increase. Moreover, we found a strong correlation between average speed and CO 2 emissions which is consistent with the work of Barth and Boriboonsomsin (2008). However, we did not found correlations for local pollutants. We observed that although the average speed remains relatively constant in free flow situations, CO, HC and NO X, emissions show a higher variability. In fact, slight speed variations produce a significant impact on local pollutants, namely NO X from LDDV. In similar sections the air quality could be improved significantly by minimizing high-emitting driving behaviour. 3.4 VSP modes frequency Since the frequency of occurrences of each VSP mode controls the total of emissions estimated, it is important to understand how VSP modes distribution varies across routes.

11 Modes 1 and 2 represent deceleration modes (negative VSP values), whereas mode 3 represents idling or low speeds situations. Modes 4 to 14 describe different combinations of increasing and positive accelerations. The columns charts shown in Figure 5 display the average time spent in each VSP mode during off peak and peak periods, with the respective standard deviation. The line charts above represent the relative contribution of each VSP mode for the total of CO 2, CO, and HC emissions from LDGV and NO X from LDDV - the major sources of each pollutant. For each OD pair, we considered key route attributes. Ma and Mm have comparable travel times but Ma is 27% shorter. Im 1 and Iu have a similar length, but during peak and off peak Im 1 allows more than 50% of time saving in relation to route Iu. Route Um is the longest urban route but with less travel time. Regarding the VSP modes frequency, the routes Um, Im 1 and Mm (routes which are performed essentially on motorways) have a uniform VSP modes distribution compared with routes Uu, Iu and Ma predominantly driven on builtup areas. In the former cases, the reduction of speed shifts the distribution towards lower VSP modes. For instance, on Iu about 20% (during peak) and 14% (during off peak) of the time is consumed in idling or low speed situations and just 4% of the travel time is spent on VSP modes higher than 7. The greatest difference between peak and off peak and the higher standard deviations intervals occur in VSP mode 3. This is more notorious in route Uu, Iu, and Ma, due to the higher number of intersections, illegal parking, work zones and other incidents leading to idle. For intercity routes, K-S test (95% confidence level) indicated that routes Im 2, Ih and Iu have not the same distribution on the both periods evaluated (p-value of , 0.000, and ). Im 1 did not present significant differences, since during off peak and peak an adequate capacity is available (p=0.999). Regarding the urban scenario, all routes have shown no significant differences (p-values ranged from 0.12 to 0.18). The metropolitan routes showed the highest difference on VSP modes distribution (p-value = 0,000). This can be justified by the higher traffic volumes and congestion situations that occur during peak periods in both routes. Each VSP mode contributes differently to the emission of the various pollutants. The distribution of CO 2 according to the VSP mode follows approximately the same trend of the relative frequency distribution of VSP modes. However, CO emissions are mostly generated during the occurrence of the higher VSP modes. For instance on Ma, more than 55% of CO emissions are generated during the occurrence of the VSP modes which represent nearly 3% of the travel time. The contribution of each VSP mode for the remaining local pollutants (NO X - LDDV, and HC - LDGV) showed a comparable behaviour to CO emissions, but less sensitive to the higher modes. 3.5 Implications for future eco-routing strategies The results have shown that emissions can be considerably reduced if travellers choose eco-friendly routes. In the hypothetical extreme case of everyone choosing an eco-friendly route, a shift to all-or-nothing assignment from user

12 equilibrium (UE) assignment can theoretically occur, producing an opposite result to the desired. Given the heterogeneous behaviour of drivers, it is unlikely that all drivers will follow the same rote. In addition, for the presented casestudies, the empirical results suggest that the route with lowest emissions during peak and off-peak hours is the same and there is room for emissions savings during peak hour. Although some specific links of the analysed networks were found to be close to saturation, the difference between total emissions produced in each route suggests that the designated eco-routes may accommodate a limited number of green routing users. Thus, for realistic market penetration scenarios of eco-friendly navigation systems, they might have sufficient capacity to accommodate demand without increasing emissions in the network. On the other hand, the selection of the eco-friendly route is not always obvious. For example, the intercity routes that yield CO 2 savings might also lead to substantial increases in other pollutants, such as CO and NO X. Furthermore, for all case-studies, the routes that lead to a minimization of local pollutants are those that mainly cross urbanized areas, avoiding motorways. This fact will involve a careful assessment of potential externalities that may arise from a purely dedicated navigation system based on emissions minimization, since higher volumes of traffic crossing urban areas may lead to urban environmental degradation and worse levels of road safety. Further eco-routing algorithms may also focus on pollutant concentration in different areas of the network instead of just focusing on total emission amount. Thus, in addition to the environmental information that can be provided to the drivers, some alternative traffic management strategies may be implemented to improve traffic operations. Moreover, the policies focused on eco-traffic assignment must necessarily be accompanied by efforts to promote eco-driving. The implementation of speed management/harmonization techniques on motorways aiming at reducing excessive high speeds and consequent high emissions levels can be helpful. It can also facilitate the minimization of the trade-off between the minimization of fuel/co 2 emissions and other pollutants, and make less attractive (from the total emissions perspective) the routes that cross the urban centres. When there is no real-time information, pre-trip planning programs can consider the variability of emissions on each link, based on different time periods and estimate the best route for a specific period. Clearly, this is only valid assuming that the impact of these programs will not have a substantial impact on the equilibrium of the network. In a more advanced Intelligent Transportation System (ITS) scenario, in which vehicles are routed dynamically in the network, the changes in traffic volumes between the various routes can be significantly higher. In this scenario, one must consider the road segments capacity and the network configuration, in order to assess the system-wide impacts on emissions. Real-time or historical link based-emissions, such as the information generated in this study, can be incorporated in pre-trip planning software to determine the most eco-friendly route. In future, analytical models should be explored to determine more precisely the additional traffic volume (of eco-routing vehicles) that can be allocated to each

13 route, without significantly compromising the network performance. Further research should also focus on the development of an integrated platform which should be able to simulate various ITS scenarios mentioned above, different methods of traffic assignment, considering the total emissions on the network and externalities (such as road safety and noise) associated with each scenario. Moreover, additional research should assess the trade-off between air quality and human exposure to pollutants, i.e., the issue of diverting traffic to more eco-friendly local streets but the emissions having greater health impacts in densely populated areas. 4. CONCLUSIONS This paper has attempted to assess whether eco-friendly routes change during peak hours, based on three distinct contexts. A total of 222 hours of GPS data were collected and a microscopic emission model was used to generate emissions information during off peak and peak periods. Although the results cannot be generalized, this empirical study has reinforced the relevance of the eco-routing concept. It has been demonstrated that: VSP modal distributions were shown to be significant different at 95% CI during the peak periods on intercity and metropolitan routes; A slight decrease in the differences of total emissions among the various routes during peak periods was observed. However, for each OD pair, the eco-friendliness rating of routes was shown to be constant under different traffic volume levels. This fact suggests that the infrastructures analysed could have sufficient capacity to accommodate a limited extra demand of drivers who would like to select a route with lower emissions levels; Both during off peak and peak periods, the selection of an appropriate route can lead to significant emissions reduction: CO 2 up to 25%, and local pollutants up to 60%. Nevertheless, some limitations must be considered when implementing these systems. Namely, it was observed that the selection of an eco-friendly route selection is not always clear since: The eco-route could depend on the type of vehicle used; In some cases the routes that allow a minimization of pollutants can cross urbanized areas. This fact will involve a careful assessment of potential externalities that may arise from a purely dedicated navigation system based on emissions minimization; In the intercity OD pair, a trade-off between CO 2 vs. local pollutants minimization has been observed. Therefore, it must be emphasized, that the concept of eco-friendly should not be strictly confined to CO 2 /fuel consumption. In similar cases, strategies for assigning relative weights to pollutants (with high spatial resolution) should be considered in order to optimize traffic operations with a maximum environmental net benefit. ACKNOWLEDGMENTS

14 J. Bandeira and P. Fernandes acknowledge the support of the Portuguese Science and Technology Foundation - FCT for the Doctoral grants SFRH/BD/66104/2009 and SFRH/BD/87402/2012, respectively. This work was partially funded by FEDER Funds through the Operational Program Factores de Competitividade COMPETE and by National Funds through FCT within the projects PTDC/SEN-TRA/115117/2009 and PTDC/SEN-TRA/122114/2010. The authors also acknowledge the Strategic Project PEst-C/EME/UI0481/2011 and Toyota Caetano Portugal, which allowed the use of vehicles. The participation of Dr. Khattak in this research was sponsored in part by the Research and Innovative Technology Administration, U.S. Department of Transportation, through the TranLIVE Tier 1 university transportation center. The collaboration between Drs. Coelho and Khattak was under the auspices of the Luso- American Transportations Impacts Study Group (LATIS-G). The contents of this article reflect the views of the authors, who are responsible for the facts and the accuracy of the information presented herein. References Ahn K, Rakha H The effects of route choice decisions on vehicle energy consumption and emissions. Transportation Research Part D: Transport and Environment 133: Ahn K, Rakha H, Moran K. System-wide Impacts of Eco-routing Strategies on Large-scale Networks. Presented at the 91st Transportation Research Board Annual Meeting. Washington DC; Aziz HM, Ukkusuri SV Integration of Environmental Objectives in a System Optimal Dynamic Traffic Assignment Model Computer-Aided.Civil and Infrastructure Engineering 27(7): Bektaş T, Laporte G The Pollution-Routing Problem. Transportation Research Part B: Methodological 4(8): Bandeira JM, Coelho MC, Sá ME, Tavares R, Borrego C Impact of land use on urban mobility patterns emissions and air quality in a Portuguese medium-sized city. Science of the Total Environment 4096: Bandeira JM, Almeida TG, Khattak A, Rouphail NM, Coelho MC Generating emissions information for route selection - Experimental monitoring and routes characterization. Journal of Intelligent Transportations Systems: Technology, Planning, and Operations. Accepted author version posted online: 17 Oct DOI: / Bandeira, JM., Pereira, SR., Fontes, T., Fernandes, P., Khattak, A., & Coelho, M. C. (2013). An Eco-traffic assignment toot. Proceedings of EWGT th Meeting of the EURO Working Group on Transportation, Oporto Barth M, Boriboonsomsin K Environmentally-Friendly Navigation. Proceedings of the 2007 IEEE Intelligent Transportation Systems Conference, Seattle. Barth M, Boriboonsomsin K Real-World Carbon Dioxide Impacts of Traffic Congestion Transportation Research Record: Journal of the

15 Transportation Research Board Transportation Research Board of the National Academies Washington DC 2058: Barth M Environmentally-Friendly Navigation United Sates patent application Coelho MC, Frey HC, Rouphail NM, Zhai H, Pelkmans L Assessing methods for comparing emissions from gasoline and diesel light-duty vehicles based on microscale measurements. Transportation Research Part D: Transport and Environment 14(2): De Vlieger I, De Keukeleere D, Kretzschmar J G Environmental effects of driving behaviour and congestion related to passenger cars. Atmospheric Environment 34(27): Ericsson E, Larsson H, Brundell-Freij K Optimizing route choice for lowest fuel consumption - Potential effects of a new driver support tool. Transportation Research Part C: Emerging Technologies 14(6): Estradas de Portugal (Roads of Portugal) EP 2011 Available at < Accessed in [March 2011] European Commission ITS Action Plan Framework Service contract Available at < Accessed [9 June 2011] Ferguson EM, Duthie J, Waller ST Comparing Delay Minimization and Emissions Minimization in the Network Design Problem. Computer-Aided Civil and Infrastructure Engineering, 27 (4): Figliozzi MA Vehicle Routing Problem for Emissions Minimization. Journal of the Transportation Research Board Transportation Research Board of the National Academies 2197: 1-7. Frey HC, Unal A, Chen J, Li S, Xuan C Methodology for Developing Modal Emission Rates for EPA s Multi-Scale Motor Vehicle & Equipment Emission System Publication EPA420-R Prepared by North Carolina State University for US Environmental Protection. Agency Ann Arbor. Frey HC, Zhang K, Rouphail NM Fuel Use and Emissions Comparisons for Alternative Routes Time of Day Road Grade and Vehicles Based on In-Use Measurements. Environmental Science & Technology 42 (7): Gazis, A, Fontes T, Bandeira J, Pereira S, Coelho MC Integrated Computational Methods for Traffic Emissions Route Assessment. Fifth ACM SIGSPATIAL International Workshop on Computational Transportation Science, Redondo Beach, CA. Guo, L., S. Huang, and A. W. Sadek An Evaluation of Environmental Benefits of Time-Dependent Green Routing in the Greater Buffalo Niagara Region. Journal of Intelligent Transportation Systems 17 (1): Khattak A, Polydoropoulou A, Ben-Akiva M Modeling Revealed and Stated Pretrip Travel Response to Advanced Traveler Information Systems. Transportation Research Record: Journal of the Transportation Research Board 153: Khattak AJ, Khattak AJ Comparative Analysis of Spatial Knowledge and En Route Diversion Behavior in Chicago and San Francisco: Implications for

16 Advanced Traveler Information Systems. Transportation Research Record: Journal of the Transportation Research Board 1621: Kolak, O. İ., O. Feyzioğlu, Ş. İ. Birbil, N. Noyan, and S. Yalçindağ Using emission functions in modeling environmentally sustainable traffic assignment policies. Journal of Industrial and Management Optimization. 9 (2): INIR. Traffic report in the National Network of Highways in from < Accessed in [January 2012]. Maibach M, Schreyer C, Sutter D, Essen HP van, Boon BH, Smokers R, et al. Handbook on estimation of external costs in the transport sector. Solutions. Commissioned by: European Commission DG TREN. Delt, CE Minett C, Salomons M, Daamen W, Arem BV, Kuijpers S Eco-routing: comparing the fuel consumption of different routes between an origin and destination using field test speed profiles and synthetic speed profiles. Proceedings of the IEEEE Forum on Integrated and Sustainable Transportation Systems, Vienna, Austria. Nagurney A, Ramanujam P, Dhanda KK A multimodal traffic network equilibrium model with emission pollution permits: compliance vs. noncompliance.transportation Research Part D: Transport and Environment 3(5): Nie Y, Li Q Ecorouting Model Considering Microscopic Vehicle Operating Conditions Proceedings of the 92nd Transportation Research Board Annual Meeting, Washington DC. Rakha H, Ahn K, and Moran K INTEGUmTION Framework for Modeling Eco-routing Strategies: Logic and Preliminary Results. Proceedings of the 90th Transportation Research Board Annual Meeting,Washington DC. Rilett L, Benedek C Traffic assignment under environmental and equity objectives. Transportation Research Record: Journal of the Transportation Research Board 1443: Sivak M, and B Schoettle Eco-Driving: Strategic Tactical and Operational Decisions of the Driver that Improve Vehicle Fuel Economy. Publication UMTRI University of Michigan.Transportation Research Institute. Sugawara S, Niemeier D How Much Can Vehicle Emissions Be Reduced? Exploratory Analysis of an Upper Boundary Using an Emissions- Optimized Trip Assignment. Transportation Research Record: Journal of the Transportation Research Board 1815: TUa 2002 The Congestion Mitigation and Air Quality Improvement Program [ebook] website < Tzeng GH, Chen CH Multiobjective decision-making for traffic assignment. IEEE Transactions on Engineering Management 40: United States Department of Transportation USDOT 2012 Applications for the Environment: Real-Time Information Synthesis (AERIS) Benefit-Cost Analysis. Available at < Accessed [9 June 2013]

17 United States Environmental Protection Agency - US EPA 1998 Assessing the emissions and fuel consumption impacts of ITS Available at < 8Accessed [9 June 2011]. Virginia Department of Transportation VDOT 2011 Available at < [Accessed 10 July 12011]. Yao, E, and Y. Song Study on Eco-Route Planning Algorithm and Environmental Impact Assessment. Journal of Intelligent Transportation Systems, 17 (1): Zhang K, Batterman S, Dion F Vehicle emissions in congestion: Comparison of work zone rush hour and free-flow conditions. Atmospheric Environment 45(11): Zhang Y, Lv J, and Ying Q Traffic assignment considering air quality. Transportation Research Part D: Transport and Environment 15(8):

18 TABLE 1 Relevant research on the impact of route choice in terms of emissions and fuel use (m Microscopic models, M Average speed models) 1 2 Reference Nagurney et al. (1998) Sugawara and Niemeier (2002) Study location Environmental Goals Virtual Generic NA Virtual CO 3 Figliozzi (2010) Virtual CO Rakha et al. (2011) Ferguson et al. (2012) Bektaşa and Laporte (2011) Aziz and Ukkusuri (2012) Nie and Qianfei (2013) Bandeira et. al. (2013) Emissions Estimation M (f(avg speed)) M (f(avg speed)) Virtual Fuel, CO 2 m (VT-micro) Virtual NO X, VOC, CO Virtual Fuel, CO 2 Virtual CO Virtual CO 2 Virtual Fuel, CO 2, NO X, HC, CO,PM M f(avg speed) M f(avg speed)) M (f(avg speed)) m (CMEM adapted) m(vsp) Tzeng and Chien-Ho (1993) Taipei, Taiwan CO M (Local survey) Rilett and Benedek (1994) Ericsson et al. (2006) Barth et al. (2007) Ottawa, Canada CO Lund, Sweden Fuel, CO 2 Los Angeles CA, USA Fuel, CO 2, NO X, HC, CO M (f(avg speed)) m (VETESS, VETO) m (CMEM) Highlights Multimodal network eq. model with emission pollution permits Trip-assignment model. Emissions savings over UE up to 25% Emissions savings (up to 28%) by following eco-routes Savings in fuel consumption of 15% using the Integration model When a network is designed for minimal travel time, NO x and CO emissions can increase Extension of the Vehicle Routing Problem. Significant CO 2 savings. CO emission with minimized travel time when drivers take longer routes with low speed profiles Microscopic eco-routing model considering delays emissions at intersections Assessment of eco-traffic assignment strategies Development of a trafficassignment method with multipleobjective decision making To minimize CO during peak hours, the system travel time may increase 2% 8.2% fuel savings by using a fueloptimized navigation system. A time minimization path minimizes emissions (CO 2 up to 42%) Frey et al. (2008) Ahn and Rakha (2008) Zhang and Ying (2010) trpd Minett et al. (2011) 18 Ahn et.al (2012) 19 Guo et. al Bandeira et al. (2013) Yao and Song (2013) Kolak et al. (2013) North Carolina, USA Northern Virginia, USA Virtual and College Station, TX, USA Zoetermeer, Holland Cleveland- Columbus,OH USA Greater Buffalo- Niagara, USA Aveiro-Porto, Portugal Beijing, China Sioux Falls, SD, USA Fuel, CO 2, NO X, HC, CO Fuel, CO 2, NO X, HC, CO PEMS & m (VSP) M&m (VTmicro, CMEM, Mobile6) CO M (Mobile 6.2) Fuel Fuel Fuel, CO (others) Fuel, CO 2, NO X, HC, CO Fuel, CO 2, NO X, HC, CO NO X (others) m (VT- CPFEM) m (VT-micro) m (Moves) m (VSP) m (VSP) M (COPERT) NO X savings up to 24% (comparing routes over different periods) Savings over the UE up to: CO 2 7%, NO X 15%, and CO 50% Potential for reduce emissions concentrations with a marginal increase in travel time A provincial route can offer av. time savings of 25% and 45% of fuel. When 20% of eco-routing vehicles are assigned -> vehicles consume higher fuel levels. Assessment of the impact of ecorouting market penetration rates Choosing a fuel/co 2 saving route can increase emissions of CO (58%), NO X (33%) and HC (16%). Eco-route planning algorithm based on mesoscopic emissions models New optimization models to s the sustainable management of

19 Table 2 Characteristics of the analysed routes. Route Length (km) Speed limits (km/h) (% of distance) 50 or Nº of lanes (% of distance) Intersections Ramps Total Tl R On Off Im Im Ih Iu Um-CS Um-SC Ua-CS Ua-SC Uu-CS Uu-SC Ma Mm

20 TABLE 3 Values of damage costs of emissions used in the eco-friendly indicator. Scenario Intercity, Urban ( 2008 /g) Metropolitan ( 2012 USD/g) Emissions cost Source Data CO 2 NO X HC CO Maibach et al, E E E E+00 USDOT, E E E E-03

21 TABLE 4 Observed average speed, % of time spent in specific speed intervals and typical numbers of stops during off peak and peak periods. Speed (km/h) 0-20 km/h Speed range (% of time) Nº. stops Nº. stops km/h km/h km/h per km Im 1 Off peak % 5% 16% 75% Peak % 7% 17% 68% Im 2 Off peak % 7% 20% 70% Peak % 9% 23% 64% Ih Off peak % 22% 48% 22% Peak % 28% 38% 17% Iu Off peak % 39% 46% 1% Peak % 42% 38% 0% Um Off peak % 36% 26% 25% Peak % 32% 21% 19% Uh Off peak % 32% 56% 0% Peak % 30% 41% 0% Uu Off peak % 52% 15% 0% Peak % 46% 12% 0% Ma Off peak % 21% 37% 12% Peak % 22% 31% 11% Mm Off peak % 11% 38% 38% Peak % 17% 45% 17%

22 TABLE 5 Total emissions (CO 2, CO, NO X HC), estimated environmental damage cost (ED) and travel time (TT) per route during off peak and peak periods. Vehicle Route Period CO 2 (g) CO (g) NO X (g) HC (g) ED ($) LDDV LDGV Im 1 Im 2 Ih Iu Um Ua Uu Mm Ma Im1 Im2 Ih Iu Um Ua Uu Mm Ma TT (min) Off peak ,54 73,90 0,928 0,688 48,8 Peak ,55 74,78 0,938 0,697 49,2 Off peak -3% -3% -2% 2% -3% 50,9 Peak -2% -2% 0% 6% -1% 54,3 Off peak 11% 18% -9% 16% 0% 80,1 Peak 12% 22% -10% 22% 0% 90,9 Off peak -3% 10% -34% 13% -21% 97,2 Peak -1% 14% -34% 20% -20% 108,3 Off peak ,17 5,74 0,098 0,098 8,1 Peak ,17 5,47 0,094 0,094 8,1 Off peak -16% -12% -28% -10% -22% 8,2 Peak -5% -1% -17% 7% -11% 9,9 Off peak -25% -21% -35% -12% -30% 8,5 Peak -18% -14% -28% 2% -23% 10,0 Off peak ,636 21,355 0,357 0,842 28,4 Peak ,681 25,426 0,431 0,971 33,7 Off peak -10% -9% -11% 0% -10% 31,6 Peak -8% -8% -7% 2% -7% 38,0 Off peak ,03 12,57 4,014 0,372 48,8 Peak ,20 12,71 4,053 0,377 49,2 Off peak -1% 3% -3% 2% 2% 50,9 Peak 2% 7% -1% 6% 6% 54,3 Off peak 25% -25% 0% -1% 20% 80,1 Peak 30% -29% -1% 1% 24% 90,9 Off peak 23% -59% -22% -18% 14% 97,2 Peak 29% -61% -21% -14% 19% 108,3 Off peak ,60 1,12 0,340 0,042 8,1 Peak ,66 1,05 0,329 0,040 8,1 Off peak -10% -41% -23% -19% -12% 8,2 Peak 4% -29% -12% -3% 2% 9,9 Off peak -15% -47% -31% -22% -17% 8,5 Peak -4% -38% -25% -9% -7% 10,0 Off peak ,08 4,07 1,266 0,660 28,4 Peak ,19 4,68 1,597 0,803 33,7 Off peak -5% -12% -12% -2% -8% 31,6 Peak -3% -6% -10% 0% -5% 38,0

23 TABLE 6 Eco-friendly route indicator based on Environmental Damage costs. Im 1 Im 2 Ih Iu Um Ua Uu Ma Mm NP P NP P NP P NP P NP P NP P NP P NP P NP P Travel time LDDV LDGV NP - Non-peak P - Peak Travel time Faster Slower Emissions cost impact Best Worst

24 a) b) c) Notes: Congestion hotspots. Figure 1 Study routes map and Average Daily Traffic: a) intercity (Oporto-Aveiro): Im1, Im2, Ih and Iu; B) urban routes (Aveiro centre and suburbs): Um, Ua and Uc; and c) Metropolitan (Norfolk-Chesapeake): Ma and Mm with VDOT segments.