a comparative empirical analysis of eco-friendly...

1 A COMPARATIVE EMPIRICAL ANALYSIS OF ECO-FRIENDLY R OUTES 2

DURING PEAK AND OFF-PEAK HOURS 3 4 5

Jorge Bandeira, MSc. 6 PhD Student, Mechanical Engineering 7

University of Aveiro 8 Centre for Mechanical Technology and Automation / Department of Mechanical Engineering 9

Campus Universitário de Santiago 10 3810-193 Aveiro - Portugal 11 Phone: (+351) 234 370 830 12 Fax: (+351) 234 370 309 13

E-mail: [email protected] 14 15

Dário O. Carvalho, MSc. 16 MSc. Student, Mechanical Engineering 17

University of Aveiro 18 Department of Mechanical Engineering 19



Asad J. Khattak, PhD. 26 Frank Batten Endowed Chair Professor 27

Civil & Environmental Engineering Department 28 135 Kaufman Hall, Old Dominion University 29

Norfolk, VA 23529 30 United States of America 31 Phone: (757) 683-6701 32

E-mail: [email protected] 33

34 Nagui M. Rouphail, PhD. 35

Director, Institute for Transportation Research and Education (ITRE) 36 Centennial Campus Box 8601, Raleigh, NC 27695-8601 37

United States of America 38 Phone: (919) 515-1154 39 Fax: (919) 515-8898 40


Margarida C. Coelho, PhD. 43 Invited Assistant Professor, Mechanical Engineering 44

University of Aveiro 45 Centre for Mechanical Technology and Automation / Department of Mechanical Engineering 46



53 54

November 2011 55 Submitted for consideration for publication and presentation at the 91th Annual Meeting of the Transportation Research 56

Board, January 23-27, 2012 57 Word Count: 5634 text words plus 2000 words for 5 figures/ 3 tables (8×250) =7634 58

59

2 J. M. Bandeira, D.O. Carvalho, A.J. Khattak, N.M. Rouphail, and M.C. Coelho

Abstract 1

2

Automobile emissions reductions can come from several sources that include more efficient vehicles, 3

cleaner fuels, and eco-friendly routing. Providing information about expected emissions on 4

alternative routes can influence route choice of drivers, encouraging some to take more eco-friendly 5

routes. During off peak hours, when substantial capacity is available, the eco-friendly route is likely 6

to be non-changing. However, during peak periods, with limited additional capacity, the eco-7

friendliness of various routes may change, depending on traffic congestion. To explore this issue 8

empirically, more than 13.330 km of data were collected using GPS equipped-vehicles, in the US 9

and Portugal. Specifically, data were collected in diverse locations: a large metropolitan area of 10

Hampton Roads, VA, USA, intercity region of Oporto –Aveiro, Portugal, and the city of Aveiro, 11

Portugal. The results are based on second-by-second vehicle dynamics, using the Vehicle Specific 12

Power (VSP) concept to extract the emissions on various route alternatives between origins and 13

destinations. During peak hours (and off-peak), a selection of eco-friendly routes can lead to 14

significant savings in global pollutants (up to 25% of CO2) and local pollutants (up to 60% for HC, 15

NOX, and CO). Moreover, these savings were practically unchanged during the peak hours for the 16

two case-studies. In some situations the lower emissions routes are those that cut across urban areas, 17

instead of bypassing them. The implications for future eco-routing systems are discussed. 18

19


1. INTRODUCTION AND OBJECTIVES 1

Transportation is heavily dependent on gasoline and related products for 96% of its energy needs in 2

EU (1). In the European Union, oil dependence of transport could represent about 90% of energy 3

needs by 2020 (1). The inefficient use of transport networks is another major problem. In Europe, 4

traffic congestion costs can increase by 50% by 2050. In the US congestion costs have also increased 5

substantially. In 2009 US spent about $115 billion in expenses related to extra time in traffic and 6

wasted fuel (2). 7

It is clear that innovative technologies for vehicles are important but traffic management will 8

play a major role in reducing automobile emissions. The integration of traffic management and 9

traveler information systems may influence travel behavior since users of the system can adjust their 10

departure times, destinations or route choices in response to more complete information, including 11

eco-friendliness of alternative routes (3-5). 12

Route choice decisions can play a substantial role in terms of emissions reduction and energy 13

consumption (6). Currently, widespread information about which routes are more eco-friendly is not 14

widely available. The authors’ earlier efforts were based on generating information about emissions 15

of route alternatives (using a standard vehicle). In non-congested situations there may be sufficient 16

capacity to accommodate the demand of drivers who want to choose an eco-friendly route. 17

However, during peak hours when capacity is limited, eco-friendliness of routes may change (7). 18

This study explores the impacts of route choice decision during peak hours. The information 19

generated in this study can be useful in implementing innovative and sustainable traffic management 20

strategies. This paper intends to focus on the following questions: 21

22

• How can peak hour affect the choice of an eco-friendly route? 23

• How do tailpipe emissions vary during peak hour on different routes and in different 24

environments? 25

• How do traffic volumes affect emissions on different network links? 26

27

28

2. LITERATURE REVIEW 29

The effect of route choice in reducing emissions has been addressed by several authors. Table 1 lists 30

the most relevant studies carried out in the field of route choice optimization, considering energy and 31

air quality. Furthermore, it indicates the methodology for emissions and fuel use estimation (m – 32

microscale models; M–Macro; Me–Meso, f–field measurement) and whether route characteristics 33

(RC) (type of road, incidents, traffic signals, neighborhood, safety) are considered, yes (y) or not (n). 34

These studies can be divided into two groups. In the first (8-13), the authors developed 35

mathematical formulations to assign traffic on a virtual network considering air quality. In a recent 36

study, mathematical programs integrating emission objective into system wide travel time 37

minimization were developed. It was found that in an idealized system optimum (SO) scenario, CO 38

emissions reduction and travel time minimization can occur when drivers chose longer routes with 39

low speed profiles instead of all users selecting the route with the shortest travel time, i.e., user 40

equilibrium. Another study developed a theoretical emissions-optimized (EO) traffic assignment 41

model. First the authors focused on standard user-equilibrium (UE) and system-optimal (SO) 42

assignment methods, and then they derived two new functions to characterize link emissions and 43

vehicle emissions. It was found that under EO conditions the percentage of traffic assigned to 44

freeways is very low since emissions rates are extremely high at freeway free-flow speeds. 45

Moreover, the EO assignment is most effective when the network is under low to moderately 46

congested conditions (12). Other authors employed different solutions such as Genetic Algorithms 47

(8) to solve complicated optimization problems with non-linear terms. Overall, reductions in 48


emissions can be achieved, if travelers choose eco-friendly routes. However, in the extreme case of 1

everyone choosing an eco-friendly route, a shift to all-or-nothing assignment from UE assignment 2

may occur. 3

In the second group (6-7, 14-23) field experiments were conducted and a wide range of 4

models were applied to evaluate the impact of route selection in terms of emissions and energy use, 5

over several case-studies. The research of Rakha, Frey and Barth should be highlighted for different 6

reasons. Ahn and Rakha (19) studied two alternative routes using different type of emission models 7

which allowed concluding that macroscopic emission estimation tools can yield incorrect 8

conclusions. In 2011, Rakha et al. also presented a framework for modeling eco-routing strategies 9

(10). Barth et al. developed and patented an environmentally-friendly navigation system (20, 24) 10

Frey et al. (7) carried out experiments using a portable emission measurement system (PEMS) under 11

real world driving cycles. Then, in order to standardize the comparisons of emission rates for 12

different vehicles and routes, the authors employed the Vehicle Specific Power (VSP) approach to 13

characterize fuel use and emissions (7). VSP variable can explain a significant portion of variability 14

in fuel consumption and emissions (25-26). 15

The majority of studies have concluded that route choice has a significant impact on 16

emissions and energy use. However, few studies have addressed the effect of congested periods on 17

emissions (27). The distribution of vehicle speeds and accelerations in traffic vary by type of road 18

facility and amount of traffic volume, generating large discrepancies in emission levels (28). 19

Possibly, this fact has contributed to some inconsistency on literature about this issue. On the one 20

hand, some studies pointed out that time minimization paths often also minimize energy use and 21

emissions (7, 20, 29). On the other hand, research demonstrated that frequently the faster 22

alternatives are not the best from an environmental perspective (6, 14, 19, 22). 23

What has arisen from literature review is that it is not possible to generalize conclusions, 24

considering the limited study areas, and thus more research is needed to evaluate a wider range of 25

driving patterns conditions, namely at different periods of the day. A more extensive analysis 26

including different scales, and different traffic volumes, as performed here, may better reflect the 27

reality and improve the knowledge to develop further traffic management strategies. 28


1

TABLE 1 Most relevant research on the impact of route choice in terms of emissions and 2

energy consumption 3

4

(m – micro or M – Macro; - Route characteristics (type of road, incidents, traffic signals, neighborhood, etc.) are considered (y) or not (n). 5

6 7 8 9

Reference Study LocationEnvironmental

GoalsEmissions Estimation RC

Nagurney et al. (1998) Virtual Generic NA NA

Sugawara and Niemeier (2002) Virtual CO M (f(av speed)) NA

Figliozzi (2010) Virtual CO2 M (f(av speed)) NA

Rakha et al. (2011) Virtual Fuel, CO2 m (VT-micro) NA

Abdul Aziz and Satish Ukkusuri (2011)

Virtual CO M (f(av speed)) NA

Ferguson et al. (2011) Virtual VOC, NOX CO M f(av speed) NA

Gwo-Hshiung Tzeng and Chien-Ho (1993)

Taipei, Taiwan CO M (Local survey) n

Rilett and Benedek (1994) Ottawa, Canada CO M (f(av speed)) n

Ericsson et al. (2006) Lund, Sweden Fuel, CO2 m (VETESS, VETO) y

Barth et al. (2007) Los Angeles CA, USCO, CO2, NOX,

HC, FUELm (CMEM) n

Frey et al. (2008) North Carolina, USHC, NOX, CO2,

CO, FUEL PEMS & m (VSP) n

Ahn and Rakha (2008) Northern Virginia, US NOX, CO2

FUEL, HC, CO

M&m (VT-micro, CMEM, Mobile6)

n

Feng et al. (2008) Dalian, China CO M (f(av speed)) n

Tavares et al. (2009) Cape Verde Fuel, CO2 M (Copert) y

Zhang and Ying (2010)Virtual and College Station, TX, US

CO M (Mobile 6.2) n

Ganti et al. (2010)Urbana-Champaign, US

FuelMe (f(Speed, traffic lights, slope, car))

n

Bandeira et al. (2011)Aveiro-Porto, Portugal

NOX, CO2, HC,

CO, FUELm (VSP) y

Minett et al. (2011) Zoetermeer, Holland Fuel m (VT-CPFEM) n


3. METHODOLOGY 1

3.1 Experimental Measurements 2

Data for approximately 13.300 km of road coverage over the course of 222 hours were collected, in 3

different traffic conditions both in the USA and Portugal: a metropolitan area (Hampton Roads, VA, 4

USA) an intercity region (Oporto–Aveiro, Portugal) and a medium-sized city (Aveiro, Portugal). 5

For all origin/destination pairs the study routes were based on suggestions from an internet trip-6

planning software (Google-maps). The study maps and route characteristics are shown in Figure 1 7

and Table 2. Regarding urban routes, all paths connect the city centre to a point located in the 8

suburbs. Route RA is predominantly (56%) driven on motorway A25, Route RB essentially uses the 9

arterial road N109. Finally, Route RC is entirely contained in a compact urban environment. In each 10

direction significant length differences are observed due to traffic constrains. Concerning intercity 11

routes, four parallel alternative routes that cross some of the most densely populated areas of 12

Portugal were considered. While R1 and R2 are based on the motorways A1 and A29, respectively 13

(with a toll cost of 5 Euros), R3 and R4 are mainly performed on national roads. Both for intercity 14

and urban routes, off-peak emissions and a detailed route characterization can be found elsewhere 15

(6). 16

In this paper a new pair origin-destination was considered. Thus, two alternative routes with 17

similar travel times were examined both at peak and off-peak hours. Both routes connect the Old 18

Dominion University in Norfolk, to a large commercial area in Chesapeake. Those routes have high 19

quality data on traffic volumes and, road characteristics. RD is mostly performed on an arterial roads 20

(60% of 21.4 km (13.3 mi)), crossing downtown Norfolk. Regarding RE, 70% of the route distance 21

(29.2 km (18.3 mi)) is traveled through the I-64, bypassing the city of Norfolk to the East. Although 22

RE presents more intersections, the majority of them are uncontrolled connections to residential 23

neighborhoods with little impact on the arterial roads. 24

The road tests were performed during weekdays under dry weather conditions during the 25

months of February, March and April of 2010 and 2011. According to traffic volume data (30-31), 26

the peak period in the US site was considered between 6-8 AM and 4-6 PM while in Portugal (PT) 27

the peak period was considered between 7-9 AM and 5-7 PM. So, all trips whose departure time was 28

within this time range are defined as peak hour tests. The off-peak tests occurred between 9 AM-4 29

PM (US) and 10 AM-5 PM (PT). The US tests were performed using the same driver and vehicle 30

(Nissan Versa 1.6 L), while in the Portuguese case-studies, three different drivers and vehicles 31

(Toyota Prius 1.8; VW Polo 1.2; and Opel Corsa 1.2) were used. All drivers tried to avoid hard 32

accelerations and keep the average speed of the traffic flow. When the traffic volume was low, the 33

speed limits were respected. Although some slight differences on emissions according to the driver 34

were verified, the route choice was shown to be the main factor that control emissions (6). 35

During experimental tests, vehicles dynamics were recorded second-by-second, using GPS 36

data-logger devices with a resolution of 5 Hz. Videotaping was also performed to record route 37

characteristics and other incidents that may affect emissions. 38

39

40


TABLE 2 Characteristics of the corridors 1

2

TL – Traffic lights, R - roundabouts 3

4

5 FIGURE 1 Study routes map - a) intercity: R1, R2, R3 and R4, b) metropolitan: RD and RE 6

with VDOT segments, c) urban RA, RB and RC. 7

8

Length Speed limits / (% of distance) Number of Lanes (% of distance)(km) (km/h) Total TL R On Off

RA-CS 6,9 50 (29%), 70 (13%), 120 (58%) 2 (29%), 4 (71%) 11 14 5 6

RA-SC 5,8 50 (32%), 70 (2%), 120 (66%) 2 (32%), 4 (68%) 10 1 33 2

RB-CS 6,4 50 (66%), 70 (34%) 2 (45%), 4(55%) 10 1 5 11 15

RB-SC 5,7 50 (63%), 70 (37%) 2 (39%), 4 (61%) 8 1 3 9 11

RC-CS 4,3 50 (100%) 4 (60%), 2 (40%) 15 3 5 6 5

RC-SC 4,1 50 (100%) 4 (60%), 2 (40%) 15 3 5 8 7

R1 77,1 50 (2%), 90 (7%) 120 (91%) 2 (2%), 4 (82%), 6(7%), 8 (8%) 9 1 3 26 26

R2 77,0 50 (2%), 90 (7%), 100-120 (91%) 2 (2%), 4 (87%), 6 (11%) 9 1 2 35 33

R3 87,2 50 (2%), 50/70 (58%) 90 (7%), 120 (33%) 2 (48%), 3 (12%), 4 (33%), 6 (7%) 135 20 7 48 58

R4 75,7 50 (23%), 50/70, (68%), 90 (6%), 120 (3%)2 (88%); 4 (2%), 6 (10%) 275 46 19 47 45

RD 21,448 (32%), 64 (15%, 80 (34%), 96 (19%)2 (2%), 4 (13%), 6 (47%), 8 (37%) 55 27 0 12 13

RE 29,2 48 (20%), 64 (10%), 96 (70%) 2 (5%), 4 (31%), 6 (59%), 8 (5%) 47 17 0 17 17

RampsIntersections

Urb

an

Inte

rcity

Route

Me

tr

RA

RD

R2 R4

R1

R3

RC

RB

RA

RE

P

A C

s

c

N

a)

b)

c )


1

3.2 Emissions estimation 2

3

For consistency with the previous study, the same methodology based on VSP categorized in 4

14 modes was used to estimate pollutants emissions (global CO2, and local CO, NOX and HC) from 5

Light Diesel Duty Vehicles (LDDV) and Light Gasoline Duty Vehicles (LDGV). Summing up, the 6

calculation of the VSP variable follows the equation (1) but a more detailed explanation on the 7

methodology used can be found elsewhere (6, 25-26). 8

9

�� 1.1 9.81 �� 0.132� 0.000302 � ��(1) 10

11

where: 12

VSP = Vehicle Specific Power (kW.ton-1) 13

v = speed (m/s) 14

a = acceleration (m/s2) 15

Grade = road grade (decimal fraction) 16

17

The LDDV (<3.5 L) and LDGV (<1.8 L) emissions rates used in this research can be found in the 18

references (25-26). Total pollutants emissions by route were derived based on the average time spent 19

in each VSP mode multiplied by its respective emission factor. Besides VSP explains a substantial 20

portion of variability in tailpipe emissions (26), VSP was also shown to be appropriate for the 21

purpose of relative comparisons between alternative routes (7). 22

23

24

4. DISCUSSION AND RESULTS 25

26

4.1 Analysis of VSP modal frequency 27

28

Since it is the frequency of occurrences of each VSP mode that control the total of emissions 29

estimated, it is important to understand how VSP modes distribution varies across routes. A 30

comprehensive analysis of the time spent in each driving VSP mode was carried out. 31

Modes 1 and 2 represent deceleration modes, whereas mode 3 represents idling situation. Modes 4 32

to 14 describe combinations of increasing and positive accelerations. The columns charts shown in 33

Figure 2 display the average time spent in each VSP mode during off-peak and peak periods, with 34

the respective standard deviation. The solid lines represent the relative frequency distribution of the 35

VSP modes, whereas the round dot and the long dash lines indicate the relative contribution of each 36

VSP mode for the total of CO2 and CO emissions, respectively. 37

For each OD pair, key route attributes were considered. R1 and R4 have a similar length but 38

during peak and off-peak R1 allows more than 50% of time saving in relation to R4. RD and RE 39

have comparable travel times but RD is 27 % shorter. Regarding the urban routes, RA is the longest 40

but with less travel time. Thus, the routes R1, RE and RA (the faster routes performed essentially on 41

motorways) have a more uniform VSP modes distribution compared with Routes R4, RD and RC 42

predominantly driven on urbanized areas, where the reduction of speed shift the distribution towards 43

lower VSP modes. For instance, on R4 14% (off-peak) and 20% (peak) of the time is consumed in 44

idling situations and just 4% of the travel time is spent on VSP modes higher than 7. Standard 45

deviations intervals showed a higher variability in Routes R4, RD, and RC, particularly in VSP mode 46

3, due to the natural variability, such as traffic lights, illegal parking, and congestion, which increases 47

idling. 48


The distribution of CO2 according to the VSP mode follows approximately the same trend of 1

the relative frequency distribution of VSP modes. However, CO emissions are mostly generated 2

during the occurrence of the higher VSP modes. For instance on RD, more than 40% of CO 3

emissions are generated during the occurrence of the VSP modes 12, 13 and 14 which represent 4

nearly 2% of the travel time. The contribution of each VSP mode for the remaining local pollutants 5

is not shown. However, a detailed analysis of these pollutants (NOX -- LDDV, and HC - LDGV) 6

showed comparable behavior to CO emissions, but less sensitivity to the higher VSP modes (11-14). 7

8

9 FIGURE 2 VSP mode frequencies (with standard deviation intervals); VSP modes relative 10

frequency distribution at peak and off-peak; and contribution of each VSP mode for CO2 and 11

CO emissions at peak period: Intercity routes (R1, R4); Regional routes (RD, RE); Urban 12

routes (RA, RC). 13

14

15

16

17

18

0%

10%

20%

30%

40%

50%

0

200

400

600

800

1000

1200

1400

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Tim

e (

s)

VSP mode

RD RD off-peak RD peak

RD off-peak % RD peak %

RD peak CO2 % RD peak CO %

0%

5%

10%

15%

20%

25%

30%

35%

40%

0

500

1000

1500

2000

2500

3000

3500

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Tim

e (

s)

VSP mode

R1 R1 off-peak R1 peak

'R1 off-peak % R1 peak %

R1 CO2 peak % R1 CO peak %

0%

5%

10%

15%

20%

25%

30%

35%

40%

0

500

1000

1500

2000

2500

3000

3500

1 2 3 4 5 6 7 8 9 10 11 12 13 14T

ime

(s)

VSP mode

R4 R4 off-peak R4 peak

R4 off-peak % R4 peak%

R4 CO2 peak% R4 CO peak %

0%

5%

10%

15%

20%

25%

30%

35%

40%

0

50

100

150

200

250

300

1 2 3 4 5 6 7 8 9 10 11 12 13 14

VSP mode

Tim

e(s

)

RA RA off-peak RA peak

RA off-peak % RA peak%

RA peak CO2 RA PH CO

0%

5%

10%

15%

20%

25%

30%

35%

40%

0

50

100

150

200

250

300

1 2 3 4 5 6 7 8 9 10 11 12 13 14

VSP mode

Tim

e (

s)

RC RC off-peak RC peak

RC peak CO2 % RC peak CO %

RB off-peak % RB peak %

0%

10%

20%

30%

40%

50%

0

200

400

600

800

1000

1200

1400

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Tim

e (

s)

VSP mode

RE RE off-peak RE peak

RE off-peak % RE peak %

RE peak CO2 % RE peak CO %


4.2 Total emissions per route 1

Table 3 provides descriptive statistics for all study routes. The data are broken down by study-area, 2

route, time period and the difference occurred between peak and off-peak. Once the results are 3

relatively similar in both directions only one direction is presented. The number of samples 4

performed for each situation (N), average speed, and total emissions focusing on CO2, CO, and HC 5

from LDGVs, and NOX from LDDVs (the major sources of each pollutant) are presented. For each 6

OD pair the best route for all parameters is underlined. The column “reduction” shows the potential 7

reductions in emissions by comparing each route with the worst. To address possible trade-offs 8

between travel time (T.T) and emissions minimization, the relative increase in travel time compared 9

with the fastest route is presented in the last column of the table. 10

11

12

TABLE 3 Descriptive statistics of Average speed, Travel Time NOX emissions, from LDDV and 13

CO2 / CO emissions from LDGV (intercity– Aveiro-Oporto; regional – Norfolk-Chesapeake; 14

urban – Centre-Suburbs) 15

16 17

18

Figure 3 shows the evolution of a local pollutant (CO) and a global pollutant (CO2) as 19

function of average speed, being also observable the peak impact on emissions and average speed.20

21

Increase

N Mean P95 Mean P95 Mean P95 Mean P95 Mean P95D NOX G CO2 G CO T.T.

off-peak 6 95 99 49 51 71 84 11991 12327 205 275 4% 22% 8%

peak 13 93 97 50 52 75 81 12225 12435 224 271 0% 25% 1%

-2% -2% 2% 2% 5% -3% 2% 1% 9% -2%

off-peak 7 87 94 53 68 74 84 12489 13924 223 273 --- 23% --- 10%

peak 16 86 95 55 83 75 81 12641 14950 226 256 --- 22% --- 10%

-1% 1% 3% 22% 1% -3% 1% 7% 2% -6%

off-peak 6 64 65 82 88 69 76 15359 15739 166 212 7% --- 26% 69%

peak 12 53 57 99 108 68 74 16256 17697 152 204 9% --- 33% 99%

-16% -12% 20% 23% -2% -3% 6% 12% -8% -4%

off-peak 6 45 47 100 104 49 52 15108 15708 87 96 34% 7% 61% 105%

peak 11 40 47 112 130 50 52 16155 17325 85 92 33% 1% 62% 126%

-10% 0% 12% 25% 3% 1% 7% 10% -2% -4%

off-peak 9 64 68 27 31 21 23 4980 5385 45 52 --- ---

peak 7 55 68 32 40 21 23 5394 6066 40 47 --- --- 8%

-13% 0% 18% 28% -1% 2% 8% 13% -11% -10%

off-peak 9 43 49 30 32 18 22 4655 4937 39 53 12% 7% 14% 10%

peak 7 37 51 33 40 18 26 4660 5476 43 87 10% 14% 0%

-13% 4% 8% 23% 1% 21% 0.0 11% 11% 65%

off-peak 10 57 64 7.2 8.0 5.4 6.6 1323 1398 11.2 15.6 --- --- ---

peak 16 51 56 8.1 10.0 5.5 6.3 1377 1454 10.7 14.2 --- 5% ---

-10% -13% 13% 25% 0% -5% 4% 4% -5% -9%

off-peak 9 47 53 8.1 9.0 4.1 4.5 1287 1367 6.8 8.5 24% 3% 39% 13%

peak 17 39 46 9.9 13.0 4.6 5.3 1447 1732 7.7 9.6 16% --- 28% 22%

-16% -13% 22% 44% 11% 17% 12% 27% 12% 13%

off-peak 9 30 35 8.4 11.0 3.5 3.9 1167 1343 5.7 6.4 36% 12% 49% 17%

peak 17 26 32 10.1 17.0 4.0 5.5 1321 1824 6.6 8.3 28% 9% 38% 25%

-13% -9% 20% 55% 14% 43% 13% 36% 16% 29%

Travel Time (min)

D NOx (g) G CO2 (g) G CO (g)

R4

Ave

iro

- O

port

o

R1

R2

R3

ReductionAv. Speed (km/h)

No

r - C

hes RE

RD

Cen

tre

- S

ubur

bs

RA

RB

RC


1 2

FIGURE 3 CO2 and CO emissions vs. Average speed for; a) intercity routes (dir: Aveiro-3

Porto); b) regional routes (dir: Norfolk-Chesapeake); c) urban routes (dir: centre-suburbs). 4

5

Concerning intercity routes, the motorway options (R1 and R2) are less time consuming than the 6

alternative routes R3 and R4 during off-peak. During peak periods these differences are increased to 7

99% and 104% respectively. In this situation it seems that the network is not going toward the user 8

equilibrium. The fact is that R1 and R2 have toll costs, while R3 and R4 have a lower capacity 9

compared to the freeway routes. The increase of (local and regional) traffic volume during peak 10

periods seems to result in a higher travel time differences. The 95th percentile values, as well as the 11

distribution of data points along average speed axis (Figure 3) suggest that Routes R2, R3 and R4 at 12

peak hours have lower reliability. 13

CO2 emissions data show that motorway routes (R1 and R2) lead to less fuel consumption 14

while R4 shows the opposite trend. This can be confirmed in Figure 3 in which a general trend of 15

decreasing CO2 emissions with average speed is clear. However, according to the literature (32-33) 16

for average speeds values beyond the experimental range (>100 km/h (62 mph)), CO2 emissions 17

would tend to increase again. It can be also observed that during the peak period, CO2 total emissions 18

show a higher increase on the national roads (R3, R4), due to more congestion and higher travel 19

times. Concerning local pollutants the effect of peak demand is more obvious on the freeway routes 20

R1 and R2. For R3 and R4 the local pollutants emissions do not change significantly during the peak 21

3000

4000

5000

6000

7000

8000

10 30 50 70

CO

2(g

)

Av. Speed (km/h)

RD off-peak RE off-peak

RD peak RE peak

b)

0

10

20

30

40

50

60

70

80

10 30 50 70C

O (

g)

Av. Speed (km/h)

RD off-peak RE off-peak

RD peak RE peak

10000

12000

14000

16000

18000

20000

22000

30 50 70 90

CO

2(g

)

Av. Speed (km/h)

R1 off-peak R2 off-peak R3 off-peak R4 off-peak

R1 peak R2 peak R3 peak R4 peak

a)

0

50

100

150

200

250

300

350

400

30 50 70 90

CO

(g

)

Av. Speed (km/h)

R1 off-peak R2 off-peak R3 off-peak R4 off-peak

R1 peak R2 peak R3 peak R4 peak

800

1000

1200

1400

1600

1800

2000

10 20 30 40 50 60 70

CO

2(g

)

Av. Speed (km/h)

RA off-peak RB off-peak

RC off-peak RA peak

RB peak RC peak

3

5

7

9

11

13

15

10 20 30 40 50 60 70

CO

(g

)

Av. Speed (km/h)

RA off-peak RB off-peak

RC off-peak RA peak

RB peak RC peak

c)


period. Although the travel time is higher the decrease of the frequency of the upper VSP modes has 1

contributed to emission reductions. Finally, there is an evident trade-off between CO2 and local 2

pollutants minimization, confirming the findings in (6). 3

Regarding regional routes, RE has the highest emissions. Both at peak and off-peak, RD 4

yielded CO2 emissions saving from 7% up to 14% and CO savings of up to 14%. RE was found to be 5

a better option only in the case of CO emissions during peak period and in the direction Norfolk-6

Chesapeake. A more detailed analysis showed that on RD, the CO emissions are mostly produced 7

during the 6 km freeway section included in this route. 8

Evaluating the average emission rates per distance (km), CO2 and CO emissions rates on RD 9

were found to be 1.27 and 1.18 times higher than RE. However, RE is 1.3 times longer than RD 10

which makes RD more environmentally friendly in terms of total emissions produced. CO2 emissions 11

follow the same trend observed in intercity routes i.e., total emissions decrease inversely to average 12

speed. A higher variability for all pollutants emissions values is observed in RD as can be confirmed 13

by the higher standard deviation and dispersion of data emissions points in Figure 3. This can be 14

explained by the fact that a significant distance of RD is traveled through downtown Norfolk, being 15

mostly performed on arterial roads and covering more signalized intersections, which lead to a 16

certain unpredictability in the speed profile. 17

Regarding travel time, in the Norfolk-Chesapeake direction, RE is likely to be the favorite 18

choice for many drivers, offering an average time reduction of about 10% over RD during off-peak 19

periods. During peak hours RD and RE have similar travel times. 20

Finally, concerning urban routes, the peak demand effect is more evident on RC (centre-21

suburbs), although it is still the best route considering the minimization of pollutants emissions. Once 22

more, this route yields the highest emissions rate per distance but its shorter length leads to a 23

reduction in total emissions. On RA a high variability in CO emissions was noticed which can be to a 24

certain extent explained by different driving behaviors (6). 25

Overall, regarding CO2 emissions (and fuel consumption), a higher correlation between an 26

increase in average speed (inter-trip, the same route) and the reduction of CO2 emissions rates is 27

evident. However, total emissions can be reduced if a considerably shorter path is found, as in the 28

case of RC and RD. With respect to local pollutants, although there is a higher variability in 29

comparison to CO2 emissions, it is evident a general trend to emissions increase with speed. 30

Across US and European routes, the most eco-friendly route during off-peak is also eco-31

friendly during the peak hour. These results suggest that for a limited percentage of drivers diverting 32

to these eco-routes, no significant impacts will be observed on the network. However, for future 33

implementations of eco-routing strategies in different areas it will be important to evaluate the 34

capacity of each segment and their ability to accommodate additional demand. 35

36

37

4.3 Link-Based Emissions 38

39

Link-based emissions were estimated for peak and off-peak hours, using the second-by-second field 40

data for all VDOT road segments on regional routes (Figure 4). In the majority of the sections, CO2 41

emissions during peak are consistently higher than during off-peak. E-D and D-K are the segments 42

where the highest increase at peak hour is experienced. This can be explained by the frequent traffic 43

jams that occur at peak hours since these links serve as connector to the belt roads I-64 and I-264/I-44

464, respectively. In I-464 segments (DL-DQ), there are no significant differences between peak and 45

off-peak, because even at peak a high volume/capacity/ ratio is maintained. On the other hand the I-46

64 segments (E-D to E-O) are more vulnerable to higher traffic volumes, particularly on the northern 47

sections E-G and E-I. A slight decrease (but not significant) in emissions at peak period is observed 48

on E-A, D-A, D-B, D-C. 49


1

FIGURE 4 CO2 emissions and av. speed at off-peak and at peak period for RE and RD 2

segments. 3

4

One method to estimate the energy and emission effects of congestion is to analyze speed 5

patterns of vehicles operating under different levels of congestion. (33). In this section the way 6

emissions change with traffic volume on certain links of the region is discussed. Using detailed 2009 7

traffic volumes of 4 winter weekdays and 15-minute raw data, the average speed and total emissions 8

estimated for each link (per lane) was matched with the corresponding traffic volume reported for the 9

respective 15 minutes period. Therefore, it is assumed that traffic volumes variation during 10

experimental tests were similar to 2009. This is an important limitation since vehicle dynamics and 11

traffic volume should be measured simultaneously (using loop sensors, for example). However, the 12

margin of error may be minimal, given that the traffic historical data do not show significant 13

variations of volumes over the last years, and no significant changes in the road network were made. 14

Figure 5 shows CO2 and NOx emissions under different traffic volumes on an arterial and a freeway 15

segments. A rough calculation of the existing capacity in each lane was done. It was estimated that in 16

each freeway lane, there is a capacity of 1700 vehicles per hour (vph) and in each arterial lane 850 17

vph (assuming an average g/c ratio for through traffic of 0.50). 18 19

0

10

20

30

40

50

60

70

80

90

100

0

100

200

300

400

500

600

700

800

900

1000

E-A E-B

E-C

E-D E-F

E-G

E-H E-I

E-J

E-K E-L

E-M E-N

E-O E-P

Av.

Sp

ee

d (

km

/h)

CO

2(g

/km

)

Segment

CO2 (g/km) off-peak

CO2 (g/km) peak

Av. Speed off peak

Av. Speed Peak

RE

0

10

20

30

40

50

60

70

80

90

100

0

100

200

300

400

500

600

700

800

900

1000

D-A

D-B

D-C

D-D

D-E

D-F

D-G

D-H D-I

D-J

D-K

D-L

D-M D-N

D-O D-P

D-Q D-R

D-S

D-T

D-U

D-V

D-X

Av.

Sp

ee

d (

km

/h)

CO

2(g

/km

)

Segment

CO2 (g/km) off-peak

CO2 (g/km) peak

Av. Speed off peak

Av. Speed peak

RERE

RD


1

2 3

FIGURE 5 NOX and CO2 emissions as function of traffic volume for an arterial route (E-D) 4

and a freeway segment (E-I). 5

6

E-D is a 4-lane arterial segment 500 m long covering 3 signalized intersections, one of which 7

is at the interchange with I-64. Figure 4 shows that this is the road segment of RE where a higher 8

difference in CO2 emissions between peak and off-peak is observed. E-I segment corresponds to a 3-9

lane, I-64 section which extends over 2000 meters. According VDOT data, during the peak both 10

segments have been classified as operating in LOS E. 11

For E-D segment, emissions and average speed remain relatively constant up to a certain 12

traffic volume level. From that point emissions start to increase. A third-order polynomial was used 13

to fit the data points, shown as solid lines in Figure 5. Although more data points are needed to 14

define a statistically valid trend, these results are consistent with previous research (12). A detailed 15

analysis of speed profiles have demonstrated that on E-D segment the emissions are strongly 16

dependent of congestion, namely the coordination between the traffic flow with the timing of traffic 17

lights. 18

On the freeway segments, relatively high correlations between CO2, and traffic volumes were 19

found and the same trend is observed. Thus, for traffic volume higher than the capacity estimated, 20

CO2 emissions start to increase. However, for local pollutants no correlations were found. It should 21

be noted that although the average speed remains relatively constant in free flow situations, NOX, CO 22

and HC show a higher variability, because slight speed variations produce a significant impact on 23

local pollutants. 24

Real time or historical link based-emissions can be incorporated in pre-trip planning software 25

to determine the most eco-friendly route. In the context of dynamic traffic assignment and where 26

real-time information on traffic conditions and emissions is available, new routing algorithms can be 27

developed based on the UE perspective. However, when there is no real-time information, pre-trip 28

y = 4E-07x3 - 0,0003x2 - 0,0228x + 155,2

R² = 0,8511

0

50

100

150

200

250

300

0 200 400 600 800 1000 1200 1400

CO

2(g

)

Traffic Volume per lane (vph)ARTERIAL E-D

y = 1E-09x3 - 8E-07x2 - 0,0004x + 0,5697

R² = 0,8134

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0 200 400 600 800 1000 1200 1400

NO

X(g

)


y = 0,0001x2 - 0,2455x + 398,32

R² = 0,7309

0

50

100

150

200

250

300

350

400

450

0 500 1000 1500 2000

CO

2(g

)

Traffic Volume per lane (vph)FREEWAY E-I

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

1,8

2

0 500 1000 1500 2000

NO

x(g

)

Traffic Volume per lane (vph)FREEWAY E-I

Estimated capacity

y = 4E-07x3 - 0,0003x2 - 0,0228x + 155,2

R² = 0,8511

0

50

100

150

200

250

300

0 200 400 600 800 1000 1200 1400

CO

2(g

)


Estimated capacity

Estimated capacity

Estimated capacity


planning programs can consider the variability of emissions on each link based on different time 1

periods and estimate the best route for a specific period. This is valid assuming that the impact of 2

these programs will not have a substantial impact that dramatically changes the equilibrium in the 3

network. 4

4. CONCLUSIONS AND FUTURE RESEARCH DIRECTIONS 5

6

This paper has attempted to assess whether eco-friendly routes change during peak hours, based on 7

three distinct case-studies on two continents. A total of 222 hours of GPS data were collected and a 8

microscale vehicle-based approach was used to generate emissions information during peak and off-9

peak periods. 10

Although these field tests have been conducted in regions with very different characteristics, 11

the results cannot be generalized to other locations. However, this study has confirmed the potential 12

of the eco-routing concept and has also identified some limitations that must be considered when 13

implementing these systems. 14

In general, it was demonstrated empirically that for the situations examined, the route that is 15

the most eco-friendly during off-peak is also the most eco-friendly during the peak hours. These 16

results suggest that for limited market penetration of eco-friendly navigation systems (using link 17

based emissions data) there can be sufficient capacity to accommodate demand without increasing 18

emissions in the network. In a more advanced ITS scenario, in which vehicles are routed 19

dynamically in the network, the changes in traffic volumes between the various routes can be 20

significantly higher. In this scenario, one must consider each road segment capacity to accommodate 21

more traffic without increasing emissions. 22

Regarding intercity routes both tests conducted during the peak and off-peak hours have 23

shown the faster routes can reduce both CO2 emissions and fuel consumption up to 30%. However, 24

if the main objective is to minimize local pollutants emissions, this case-study has shown that travel 25

time may significantly increase by vehicles taking national roads. On regional and urban routes, the 26

shortest path was shown to be the best option to reduce the emissions. 27

For all case-studies, the routes that lead to a minimization of local pollutants are those that 28

mainly cross urbanized areas, avoiding motorways. This fact will involve a careful assessment of 29

potential externalities that may arise from a purely dedicated navigation system based on emissions 30

minimization, since higher volumes of traffic crossing urban areas may lead to urban environmental 31

degradation and worse levels of road safety. 32

Thus, in addition to the environmental information that can be provided to the drivers, some 33

alternative traffic management strategies may be implemented to improve traffic operations. For 34

instance implementing speed management/harmonization techniques on freeways aiming at the 35

reduction of excessively high speeds and consequent high emissions levels can be helpful. It can also 36

facilitate the minimization of the trade-off between the minimization of fuel/CO2 emissions and other 37

pollutants, and make less attractive (from the total emissions perspective) the routes that cross the 38

urban centers. Simultaneously several traveler information systems, such as variable message 39

signals, or dynamic road pricing schemes, can persuade drivers to avoid the most congested routes or 40

ones with higher emissions levels. 41

Further research will focus on the development of an integrated platform which should be 42

able to simulate various ITS scenarios mentioned above, different methods of traffic assignments, 43

and considering the total emissions on the network and externalities (such as road safety) associated 44

with each scenario. Moreover, further research should assess the tradeoff between air quality and 45

human exposure to pollutants, i.e., the issue of diverting traffic to more eco-friendly local streets but 46

the emissions having greater health impacts in densely populated areas. 47

48

49


ACKNOWLEDGMENTS 1

2

The authors acknowledge to Toyota Caetano Portugal, Municipality of Aveiro, and Virginia 3

Department of Transportation. This work was partially funded by FEDER Funds through the 4

Operational Program “Factores de Competitividade – COMPETE” and by National Funds through 5

FCT – Fundação para a Ciência e Tecnologia within the projects PTDC/SEN-TRA/113499/2009 and 6

PTDC/SEN-TRA/115117/2009, and FLAD – Luso American Foundation (for the Project 91-7

03/2010, within the program FLAD/NSF – “Project-USA: Networks and Partnerships for 8

Research”). M.C. Coelho and J. Bandeira also acknowledge the support of FCT for the Scholarships 9

SFRH/BD/66104/2009 and SFRH/BPD/21317/2005. The collaboration between Drs. Coelho, 10

Khattak and Rouphail was under the auspices of the Luso-American Transportation Impacts Studies 11

Group (LATIS-G). 12

13

REFERENCES 14

15

1. EU. Transportation White Paper - Roadmap to a Single European Transport Area – Towards 16

a competitive and resource efficient transport system. European Commission, 2011. 17

18

2. Urban Mobility Report Powered by INRIX Traffic Data. 2010 Transportation Texas Institute, 19

2010. 20

21

3. US EPA. Assessing the emissions and fuel consumption impacts of ITS. US Environmental 22

Protection Agency, 1998. 23

24

4. Khattak, A.J. and A.J. Khattak. Comparative Analysis of Spatial Knowledge and En Route 25

Diversion Behavior in Chicago and San Francisco: Implications for Advanced Traveler 26

Information Systems. In Transportation Research Record: Journal of the Transportation 27

Research Board, No. 1621,Transportation Research Board of the National Academies, 28

Washington, D.C., 1998, pp.17-35. 29

30

5. Khattak, A., A. Polydoropoulou, and M. Ben-Akiva. Modeling Revealed and Stated Pretrip 31

Travel Response to Advanced Traveler Information Systems. In Transportation Research 32

Record: Journal of the Transportation Research Board, No. 1537, Transportation Research 33

Board of the National Academies, Washington, D.C., 1996, pp. 46-54. 34

35

6. Bandeira, J.M, Almeida T., Rouphail N.M., Khattak, A.J., and M.C. Coelho. Generating 36

emissions information for route selection - Experimental monitoring and routes 37

characterization. Presented at 90th Annual Meeting of the Transportation Research Board, 38

Washington, D.C., 2011. 39

40

7. Frey, H.C., K. Zhang, and N.M. Rouphail. Fuel Use and Emissions Comparisons for 41

Alternative Routes, Time of Day, Road Grade, and Vehicles Based on In-Use Measurements. 42

Journal of Environmental Science & Technology, Vol. 42, No. 7, 2008, pp. 2483-2489. 43

44

8. Ferguson, E.M., J. Duthie, and S.T. Waller. Minimizing Vehicle Emissions through 45

Transportation Road Network Design Incorporating Demand Uncertainty. Presented at 90th 46

Annual Meeting of the Transportation Research Board, Washington, D.C., 2011. 47

48


9. Aziz, H.M.A., and S. Ukkusuri. Environmental Objectives within a Dynamic Traffic 1

Assignment Framework: A Step towards Green Transportation. Presented at 90th Annual 2

Meeting of the Transportation Research Board, Washington, D.C., 2011. 3

4

10. Rakha, H.A., K. Ahn, and K. Moran. INTEGRATION Framework for Modeling Eco-routing 5

Strategies: Logic and Preliminary Results. Presented at 90th Annual Meeting of the 6

Transportation Research Board, Washington, D.C., 2011. 7

8

9

11. Figliozzi, M. Emissions Minimization Vehicle Routing Problem. Presented at 89th Annual 10

Meeting of the Transportation Research Board, Washington, D.C., 2010. 11

12

13

12. Sugawara, S. and D. Niemeier, How Much Can Vehicle Emissions Be Reduced? Exploratory 14

Analysis of an Upper Boundary Using an Emissions-Optimized Trip Assignment. 15

Transportation Research Record: Journal of the Transportation Research Board, Vol. 1815, 16

Transportation Research Board of the National Academies, Washington, D.C., 2002, pp. 29-17

37. 18

19

13. Nagurney, A., P. Ramanujam, and K.K. Danda. A multimodal Traffic Network Equilibrium 20

model emissions pollution permits: Compliance vs Noncompliance. Journal of 21

Transportation Research Part D, Vol. 3, No. 5, 1998. pp. 349-374. 22

23

14. Minett, C. F., M. Salomons, W. Daamen, B. van Arem, and S. Kuijpers. Eco-routing: 24

comparing the fuel consumption of different routes between an origin and destination using 25

field test speed profiles and synthetic speed profiles. Presented at IEEE Forum on Integrated 26

and Sustainable Transportation Systems, Vienna, 2011. 27

28

15. Ganti R.K., N. Peakam, H. Ahmadi, S. Nangia, and T. F. Abdelzaher. GreenGPS: a 29

participatory sensing fuel-efficient maps application. Presented at the 8th international 30

conference on Mobile systems, applications, and services, San Francisco, 2010 31

32

16. Zhang, Y., J. Lv, and Q. Ying. Traffic Assignment Considering Air Quality. Journal of 33

Transportation Research Part D: Transport and Environment, Vol. 15, No. 8, 2010, pp. 497-34

502. 35

36

37

17 Tavares, G., Z. Zsigraiova, V. Semiao, and M.G. Carvalho. Optimisation of MSW collection 38

routes for minimum fuel consumption using 3D GIS modelling. Waste Management, Vol. 29, 39

No. 3, 2009, pp. 1176-1185. 40

41

42

18. Feng, T., J. Zhang, and A. Fujiwara. An Integrated Modeling Framework for 43

Environmentally Efficient Car Ownership and Trip Balance. IATSS Research, Vol. 32, No. 2, 44

2008, pp. 95-108. 45

46

19. Ahn, K., and H. Rakha. The effects of route choice decisions on vehicle energy consumption 47

and emissions. Transportation Research Part D, Transport and Environment, Vol. 13, No. 3, 48

2008, pp. 151-167. 49

50


1

20. Barth, M., K. Boriboonsomsin, and A. Vu. Environmentally-Friendly Navigation. Presented 2

at 2007 IEEE Intelligent Transportation Systems Conference, Seattle, 2007. 3

4

21. Ericsson, E., H. Larsson, and K. Brundell-Freij. Optimizing route choice for lowest fuel 5

consumption - Potential effects of a new driver support tool. Journal of Transportation 6

Research Part C: Emerging Technologies, Vol. 14, No. 6, 2006, pp. 369-383. 7

8

22. Rilett, L., and C. Benedek. Traffic assignment under environmental and equity objectives. 9

Transportation Research Record: Journal of the Transportation Research Board, Vol. 1443, 10

Transportation Research Board of the National Academies, Washington, D.C., 1994, pp. 92-11

99. 12

13

23. Gwo Hshiung, T., and C. Chien-Ho. Multiobjective decision making for traffic assignment. 14

IEEE Transactions on Engineering Management, Vol. 40, No. 2, 1993, pp. 180-187. 15

16

24. Barth, M., Environmentally Friendly Driving Navigation. 2010: Patent nº 20100076675 17

US, 2010. 18

19

20

25. Coelho, M.C, T. Farias, and N.M. Rouphail. Assessing methods for comparing emissions 21

from gasoline and diesel light-duty vehicles based on microscale measurements. Journal of 22

Transportation Research Part D: Transport and Environment, Vol. 14, No. 2, 2009, pp 91-23

99. 24

25

26. NCSU. Methodology for Developing Modal Emission Rates for EPA’s Multi-Scale Motor 26

Vehicle & Equipment Emission System, U. S. Environmental Protection Agency, 2002. 27

28

29

27. Zhang, K., S. Batterman, and F. Dion. Vehicle emissions in congestion: Comparison of work 30

zone, rush hour and free-flow conditions. Atmospheric Environment, Vol. 45, No. 11, 2011 31

pp. 1929-1939. 32

33

28. Committee for the Evaluation of the Congestion Mitigation and Air Quality Improvement 34

Program, National Research Council. The Congestion Mitigation and Air Quality 35

Improvement Program. Special Report 264. Transportation Research Board Academies, 36

Washington D.C., 2002. 37

38

29. De Vlieger, I., D. De Keukeleere, and J.G. Kretzschmar. Environmental effects of driving 39

behaviour and congestion related to passenger cars. Atmospheric Environment, Vol. 34(27), 40

2000, pp. 4649-4655. 41

42

30. Vi rginia Department of Transportation - VDOT. http://www.virginiadot.org/. Acessed Feb 5, 43

2011. 44

45

31. Bandeira, J.M., M. C. Coelho, M. E. Sá, R. Tavares, and C. Borrego. Impact of land use on 46

urban mobility patterns, emissions and air quality in a Portuguese medium-sized city. Science 47

of The Total Environment, Vol. 409, No. 6, 2011, pp. 1154-1163. 48

49


32. Anable, J., B. Boardman, and A. Root. Travel emissions profiles: a tool for strategy 1

development and driver advice, in Energy and Environment Programme Environmental 2

Change Unit, 1997, University of Oxford: Oxford. 3

4

33. Barth, M. and K. Boriboonsomsin. Real-World Carbon Dioxide Impacts of Traffic 5

Congestion. Transportation Research Record: Journal of the Transportation Research Board, 6

Vol. 2058, Transportation Research Board of the National Academies, Washington D.C., 7

2008, pp. 163-171. 8

9

10

a comparative empirical analysis of eco-friendly...

Documents