multi-criteria assessment of crosswalk ...transportes-tema.web.ua.pt/publications/177.pdf38 in...

MULTI-CRITERIA ASSESSMENT OF CROSSWALK LOCATION IN URBAN 1 ROUNDABOUT CORRIDORS 2

3 Paulo Fernandes, MSc. 4

Graduate Student, Mechanical Engineering 5 University of Aveiro 6

Dept. Mechanical Engineering / Centre for Mechanical Technology and Automation (TEMA) 7 Campus Universitário de Santiago, 3810-193 Aveiro - Portugal 8

Phone: (+351) 234 370 830, E-mail: [email protected] 9 (corresponding author) 10

11 Tânia Fontes, PhD. 12

Research Assistant, Mechanical Engineering 13 University of Aveiro 14

Dept. Mechanical Engineering / Centre for Mechanical Technology and Automation (TEMA) 15 Campus Universitário de Santiago, 3810-193 Aveiro - Portugal 16

Phone: (+351) 234 370 830, E-mail: [email protected] 17 18

Sérgio Ramos Pereira, MSc. 19 Research Assistant, Mechanical Engineering 20

University of Aveiro 21 Dept. Mechanical Engineering / Centre for Mechanical Technology and Automation (TEMA) 22

Campus Universitário de Santiago, 3810-193 Aveiro - Portugal 23 Phone: (+351) 234 370 830, E-mail: [email protected] 24

25 Nagui M. Rouphail, PhD. 26

Director, Institute for Transportation Research and Education 27 North Carolina State University 28

NCSU Campus Box 8601, Raleigh, NC 27695-8601 29 Phone: (919) 515-1154, E-mail: [email protected] 30

31 Margarida C. Coelho, PhD. 32

Assistant Professor, Mechanical Engineering 33 University of Aveiro 34

Dept. Mechanical Engineering / Centre for Mechanical Technology and Automation (TEMA) 35 Campus Universitário de Santiago, 3810-193 Aveiro - Portugal 36 Phone: (+351) 234 370 830, E-mail: [email protected] 37

38 39 40 41 42 43 44 45 46

July 2014 47 Submitted for consideration for publication and presentation at the 94th Annual Meeting of 48

the Transportation Research Board, January 11-15, 2015. 49 50

Total number of words (excluding references): 4,748 (text) plus 2,250 for figures/tables 51 (9*250) = 6,998 words (Max 7000 words). 52

53

mailto:[email protected]





Fernandes, Fontes, Pereira, Rouphail, Coelho 2

ABSTRACT 1 Mid-block pedestrian crossing areas between closely-spaced roundabouts can have an 2 effect on traffic operations and could result in a trade-off among capacity, environment, 3 and safety. Although research on the impacts of traffic performance on pedestrian 4 crosswalks located at isolated roundabouts has been conducted, very few studies have 5 focused on how traffic operations are impacted by pedestrian crosswalks between adjacent 6 roundabouts in close proximity. 7

This study examined the integrated effect of a pedestrian crosswalk at different 8 locations between closely-spaced two-lane roundabouts on traffic delay, CO2 emissions 9 and relative speed between vehicles and pedestrians by using a microsimulation approach. 10 The main purpose of the research was to develop a simulation platform of traffic 11 (VISSIM), emissions (Vehicle Specific Power - VSP) and safety (Surrogate Safety 12 Assessment Methodology - SSAM) in order to optimize such variables. The Fast Non-13 Dominated Sorting Genetic Algorithm (NSGA-II) was mobilized to identify a set of 14 optimized pedestrian crosswalk locations for the roundabout exit section along the mid-15 block segment. 16

The results indicated that locating the crosswalk at 15, 20 and 30 meters from the 17 exit section seemed to be an acceptable solution, providing a good balance among traffic 18 performance, emissions and pedestrians’ safety. It was also observed that even at low 19 pedestrian demands, the effectiveness of the crosswalk, in terms of capacity and 20 environment, gradually decreased when located near the circulatory ring delimitation (<10 21 meters). The findings suggested that crosswalks at mid-block segment (55/60 meters from 22 the exit section) must be also taken into account, especially under high traffic demands. 23

24 Keywords: Pedestrians crosswalks, Roundabouts corridors, Microscale modeling, Multi-25 objective optimization 26


1. INTRODUCTION AND OBJECTIVES 1 Roundabouts can provide a safe environment for non-motorized users such as pedestrians 2 and bicycles (1). Considering the benefits to pedestrians, roundabouts induce travel at 3 slower speeds, provide shorter crossing distances, and allow pedestrians to only cross one 4 direction of travel at a time (2). Operational information, energy and safety impacts of 5 roundabouts are usually collected at isolated intersections. Nevertheless, the impact of 6 roundabouts in corridors is much different. The problem of pedestrian crosswalks on 7 roundabouts capacity may arise in conditions of intense pedestrian or vehicular flows (2, 8 3). 9

A great deal of research has been conducted in the United States and Europe in the 10 past decades to study the effect of pedestrian crosswalks at isolated roundabouts. In most 11 design manuals, it is suggested that the crosswalks be located 10 to 15 meters downstream 12 of the exit junction in order to avoid affecting the traffic flow in the circulatory ring. 13 However, this range is empirical and supported by few scientific studies (2, 4-5). The 14 influence of crosswalks near roundabouts at isolated intersections is usually explored at 15 two levels: capacity/delays and safety. 16

In terms of capacity, the Highway Capacity Manual (HCM) provides some 17 relationships used to determine the reduction of capacity at roundabouts as a result of the 18 influence of pedestrian streams on traffic. Nonetheless, it does not include the case of a 19 roundabout in close proximity to one or more other roundabouts (6). Additionally, several 20 authors developed analytical models addressing the vehicle-pedestrian interaction in 21 roundabouts installed at isolated intersections in the United States (7, 8) and Europe (9). 22 Concurrently, the use of simulation tools has become widespread for analyzing 23 pedestrian’s operations on roundabouts installed at isolated intersections (3, 10). 24

Regarding safety, the assessment of pedestrian accessibility requirements has been 25 the main focus of the current literature. Several studies have demonstrated that 26 roundabouts bring accessibility challenges to pedestrians, especially those who are visually 27 impaired (11, 12). As such, there has been an interest in testing different treatments at 28 roundabouts to increase pedestrian safety (13). However, these studies did not include the 29 analysis of crosswalks at mid-block areas between adjacent roundabouts. 30

Albeit extensive, the research on emissions or fuel consumption in roundabouts did 31 not consider the influence of pedestrian crosswalks either at isolated intersections or at a 32 corridor level (14-17). Similarly, the few studies that were carried out at roundabout 33 corridors did not examine the influence of pedestrians on traffic operations (18, 19). Bak 34 and Kiec (20) evaluated the influence of various types of mid-block pedestrian crossing in 35 terms of the overall delay for both vehicles and pedestrians. However, this investigation 36 did not include crosswalks in mid-block segments between adjacent roundabouts. 37

In summary, the literature review showed that some studies were concerned with 38 the influence of pedestrian crosswalks on the available capacity in isolated roundabouts. 39 Others focused on the accessibility of the crosswalks with the main aim of improving 40 pedestrian’s safety. None addressed the influence of mid-block pedestrian crossing 41 between adjacent roundabouts in close proximity on traffic operations, namely, on 42 capacity/delays, vehicular emissions, and pedestrian safety. Basically, the literature lacks a 43 methodology that integrates all the concerns above. 44

The motivation of this research is to assess the impact of pedestrian crosswalks in 45 roundabout corridors on traffic delay, emissions, and pedestrian safety. Emissions and 46 safety, in conjunction with a multi-objective genetic algorithm, will be used to study the 47 impact of crosswalk locations in a microsimulation platform of traffic. It was hypothesized 48 that the effects of pedestrian crosswalk locations lead to a trade-off analysis among the 49 selected variables. It was expected that, under high traffic and pedestrian demands, a 50


crosswalk near a roundabout will have a negative impact on emissions and delays and at 1 the same time would be safe for pedestrians, since vehicles drive at slower speeds. In 2 contrast, far crosswalks (close to mid-block) can improve capacity and emissions, but are 3 less safe for pedestrians because vehicles can experience higher speeds at that location. 4

The present study examines the influence of crosswalks on the different traffic 5 commuters at distinct levels simultaneously, i.e., crosswalk location vs. traffic delay, 6 crosswalk location vs. CO2 emissions, and crosswalk location vs. pedestrians’ safety. This 7 study is also intended to demonstrate that the spacing between roundabouts constrains the 8 pedestrian crosswalk location along the mid-block segment. Therefore the main research 9 questions addressed in this paper are: 10

11 What is the impact of crosswalk location along the mid-block segment of a 12 roundabout corridor with variations in traffic and pedestrian demand on vehicles delay, 13 CO2 emissions, and pedestrian’s safety? 14 Considering the same criteria, what are the best locations to build a pedestrian 15 crosswalk in a roundabout corridor? 16 17

2. METHODOLOGY 18 The core idea of the proposed methodology was to develop a microsimulation framework 19 to assess pedestrian crosswalks’ impacts on vehicle’s delay and emissions, as well as, 20 pedestrian safety. The methodology was divided into five steps. First, data was collected in 21 the study domain (Section 2.2). Then, the network was modelled and evaluated using a 22 microscopic traffic model for the baseline scenario (Section 2.3.1). After that, several 23 scenarios were defined and evaluated (Section 2.4). For each scenario, emissions and 24 safety were evaluated using the VSP methodology and SSAM model (Sections 2.3.2 and 25 2.3.3, respectively). The steps four and five were related with the calibration/validation of 26 traffic model (Section 2.3.4) and multi-objective procedure, respectively (Sections 2.3.5). 27 Figure 1 illustrates the modelling framework. 28


1 FIGURE 1 Methodological framework. 2

3 2.1. Baseline site 4 The case study consisted of an urban roundabout corridor in the city of Chaves (Portugal), 5 a European medium-sized city with 41,243 inhabitants with a population density of 2,000 6 inhabitants/km2 in its downtown area (21). As shown in Figure 2, the corridor was 480 7 meters long (measured from the mid-block of one roundabout to the mid-block of the 8 adjacent roundabout), comprising two two-lane roundabouts, one with three legs (RBT1) 9 and another with five legs (RBT2). The spacing between roundabouts (measured from the 10 upstream yield lanes) is about 150 meters. An arterial with one lane in each direction 11 connects both roundabouts. Due to its central location and roundabout spacing, the arterial 12 has a limited capacity (≈750 vph/pl). The posted speed limit in the study area was 50 km/h. 13

Traffic and pedestrian counts;

Origin/ Destination (O/D) matrices;

Gap acceptance and rejection times;

Vehicle dynamics (speed,

acceleration/deceleration, road grade);

Road configuration.

BASELINE

SCENARIO SCENARIOS

Data collection

Microscopic traffic

model

Microscopic traffic

model

Emission

model

Safety

model

Calibration

Traffic volumes by link;

Vehicle speeds by link.

Validation

O/D matrices (based on traffic volumes);

Travel times;

VSP cumulative probability distributions.

Data outputs

Evaluation of different pedestrians crosswalks locations (PC);

Delays (Delay);

CO2 emissions;

Relative difference between vehicles and pedestrians speed (DeltaS).

Regression models

Multi-objective optimization

Delay = f(PC);

CO2 Emissions = f(PC);

DeltaS = f(PC).

Safety model

Emission

model


The actual location of the pedestrian crosswalk is at the upstream end of RBT2. 1 Several explanations have justified this choice: a) higher pedestrian traffic flow (≈ 160 2 p/h), and b) proximity from the roundabout exit section (≈ 10 meters). 3

Sites’ characteristics such as location, circulating width, inscribed circle diameters 4 and traffic data for each entry and exit legs were summarized in Table 1. 5

6

7 FIGURE 2 Aerial view of the selected corridor with the roundabouts identification 8

(RBT1 and RBT2), legs (L), location of the pedestrian crosswalk (PC) (Chaves, 9 Portugal) and input of pedestrians/centroids (IP). 10

11 TABLE 1 Key characteristics of the selected corridor 12

Roundabout

#

circulating

lanes

Circulating

width

[m]

Inscribed

circle

diameter

[m]

Central

island

[m]

Leg

#

approach

lanes

Entry

traffic

[vph]

*

Exit

traffic

[vph]

*

RBT1 2 7 16 40

L1 1 457 347

L2 1 138 187

L3 1 445 507

RBT2 2 8 41 76

L1 1 509 455

L2 1 358 285

L3 2 243 243

L4 1 130 217

L5 1 419 547

*During the evening peak periods (4:00 p.m. – 6:00 p.m.) 13 14 2.2. Data collection 15 Traffic and pedestrian volumes, as well as time-dependent Origin-Destination (O/D) 16 matrices, were gathered from two video cameras installed at strategic points in the selected 17 corridor. Time-gap distributions data (gap-acceptance and gap-rejection) for all turning 18 maneuvers was also extracted from the videotapes. Data was collected at evening peak 19 (4:00 p.m. to 6:00 p.m.) during three typical weekdays (Tuesday to Thursday) in February 20 2014 under dry weather conditions. 21

For vehicular activity estimation, second-by-second vehicle dynamics data was 22 recorded. A Light Duty Vehicle (LDV) equipped with a GPS Travel recorder was used to 23 perform all feasible movements. 200 GPS travel runs for each movement were extracted 24 and identified for this research (approximately 200 km of road coverage over the course of 25 12 hours). 26

2

1

© 2014 Nokia © 2014 Microsoft Corporation

RBT2

Exit section

L2

L1 L3

L1

L2

L3

L4 L5

Roadway

Crosswalk

45 m

PC

RBT1

RBT2

IP

IP


2.3. Microsimulation platform for traffic, emissions, and safety 1 2 2.3.1. Traffic modelling 3 VISSIM software package was selected to simulate traffic operations (22). VISSIM is 4 widely recognized as a powerful tool for roundabout operational analysis, since it can be 5 calibrated to match deterministic capacity relationships (3). It also has been previously 6 used to model pedestrian-vehicle interaction at roundabouts (10). VISSIM also allows 7 exporting full disaggregated vehicle and pedestrian trajectory files that can be used by 8 external applications to assess environmental and safety impacts, as described on the 9 following sections. 10

The simulation model was run for 90 minutes (4:30-6:00 p.m.) with the first 30 11 minutes used for a warm-up period, and data was extracted only for the remaining 60 12 minutes. The following distribution fleet composition was considered (23): 44.7% of Light 13 Duty Gasoline Vehicles (LDGV), 34.3% of Light Duty Diesel Vehicles (LDDV) and 14 21.0% of Light Commercial Diesel Vehicles (LCDV). Since remaining categories 15 represented only 1.0% of traffic composition, they were excluded from this analysis. The 16 coded network in VISSIM is depicted in Figure 2. Links volumes (traffic and pedestrians) 17 and speeds were available for all of these links. An average pedestrian walking speed value 18 of 1.2 m/s was adopted (6). 19

20 2.3.2. CO2 emissions 21 To estimate vehicular emissions the VSP methodology was employed (24). Several 22 motivations have supported the use of this methodology: a) VSP allows estimating 23 instantaneous emissions based on a second-by-second vehicle’s dynamics, taking as input 24 the trajectory files given by VISSIM and b) VSP can be applied to the European car fleet 25 because it includes a wide range of engine displacement values (24). VSP is a function of 26 the instantaneous speed, acceleration/deceleration, and the road grade. The VSP values are 27 categorized in 14 modes, and an emission factor for each mode is used to estimate, among 28 others, the CO2 emissions from LDGV<1.4L (24) LDDV<1.9 L (25) and LCDV<2.5L 29 (26). Previous research has documented the effectiveness of the VSP approach in 30 analyzing emission impacts of different roundabouts’ layouts (e.g. 14, 17). Due to the flat 31 terrain, the effect of the grade was ignored. 32 33 2.3.3. Safety 34 For the safety assessment approach, the software package developed by the Federal 35 Highway Administration - FHWA (Surrogate Safety Assessment Model – SSAM) (27) 36 was used, which automates traffic conflict analysis by processing vehicle and pedestrian 37 trajectories (*.trj file) on a second-by-second basis. This approach has all the common 38 advantages of simulation (e.g. safety assessment of new facilities before the occurrence of 39 crashes), but also has some drawbacks: current microscopic traffic models are not able to 40 model some crash types such as sideswipe, head-on or U-turn related collisions (27). 41 Vasconcelos et al. (28), recognize that despite some limitations regarding the nature of 42 traffic models, SSAM is capable of evaluating the relative safety of different roundabouts 43 layouts. 44

For each interaction, SSAM stores the trajectories of vehicles (or pedestrians) from 45 the traffic model and records surrogate measures of safety and determines whether or not 46 that interaction satisfies the condition to be deemed a conflict. Time-to-Collision (TTC) 47 was used as a threshold to define if a given vehicle-pedestrian interaction is a conflict; the 48 Relative Speed (DeltaS) was used as a proxy for the crash severity (27). TTC is the 49 minimum time-to-collision value observed during the interaction of two vehicles (or 50


pedestrians) on collision route. If at any time the TTC drops below a given threshold (2 1 seconds, as suggested for vehicle-pedestrian events (29)), the interaction is tagged as a 2 conflict. DeltaS is the difference in vehicles’ (or pedestrians) speeds as observed at the 3 instant of the minimum TTC (27). 4

SSAM classifies resulting conflicts into three categories based on a conflict angle 5 (from -180° to +180°). The angle is expressed in the perspective of the first vehicle (or 6 pedestrian), which is arriving at the conflict point, and indicates the direction from which 7 the second vehicle is approaching relatively to the first vehicle (or pedestrian). The type is 8 classified as rear end if 0º < conflict angle < 30°, a crossing conflict if 85º < conflict angle 9 < 180°, or is otherwise a lane change conflict (27). 10

11 2.3.4. Calibration and Validation 12 Calibration of VISSIM parameters was made by modifying driver behavior and vehicle 13 performance parameters in the traffic model, and by examining their effect on traffic 14 volumes and speed for each link. The main driver behavior parameters are car-following 15 parameters (average standstill distance, additive and multiple part of safety distance), lane-16 change parameters, gap acceptance parameters (minimal gap time and minimal headway) 17 and simulation resolution (22). To optimize the aforementioned parameters, a procedure 18 based on the Simultaneous Perturbation Stochastic Approximation (SPSA) genetic 19 algorithm was used. The objective function, the minimization of Normalized Root Mean 20 Square (NRMS), is denoted by Eq. (1). The normalization enables the consideration of 21 multiple performance measures, in this case, link volumes and speeds. The calibration 22 procedure was formulated as follows (30): 23

24

2 2

1 11

1Min 1

TI Ii ii i

i it i i

v v s sNRMS W W

v sN

(1) 25

26 Subject to: Lower bound ≤ θ ≤ Upper bound 27 28 Where: 29 vi = Observed link volumes for link i; 30 v (θ)i = Estimated link volumes for link i; 31 si = Observed speeds for link i; 32 s (θ)i = Estimated speeds for link i; 33 N = Total number of links in the coded network; 34 T = Total number of time periods t; 35 W = Weight used to assign more or less value to volumes or speeds. 36 37

Considering the calibration criteria, the current accepted practice, as recommended 38 by the FHWA is to rely on the Geoffrey E. Havers (GEH) statistic for assessing goodness-39 of-fit. The difference between observed and estimated link volumes (traffic or pedestrians) 40 should be less than 5% for at least 85% of the coded links (31). 41

The model validation was focused on the comparison between estimated and 42 observed O/D matrices and travel times for a preliminary number of runs (between 10 and 43 20, as suggested by Hale (32)). The GEH statistic was also used as a measure of the 44 goodness-of-fit. Along the validation of previous outputs, the observed and estimated 45 Vehicle Specific Power (VSP) cumulative probability distributions were also compared. 46 This validation step allowed assessing the differences in the acceleration/deceleration 47


profiles between field measurements and simulation. Crash data records were not available 1 in the studied location. Thus, the validation procedure did not include any comparison 2 between conflicts from SSAM and crash data. About 80% of the data was used for 3 calibration and the remaining data for validation. 4 5

2.3.5. Multi-objective optimization 6 The Fast Non-Dominated Sorting Genetic Algorithm (NSGA-II) was adopted in this 7 research (33). Several motivations have justified its use: a) diversity in optimal solutions 8 by incorporating the crowding distance into the fitness function and b) a binary tournament 9 approach that accommodates the selection process (33). NSGA-II has been reported as an 10 effective algorithm to find a good approximation of an optimal Pareto front (34). 11

Figure 3 displays the NSGA-II main steps, which was implemented in Matlab. A 12 user pre-specified maximum number of generations was defined as the stopping 13 (convergence) criteria of the NSGA-II procedure. Multi-objective optimization results 14 must ensure both the convergence to Pareto Optimal Front (POF) and the diversity of the 15 solutions (34). The convergence to POF is based on the comparison among the sets of non-16 dominated solutions from various generations. The convergence is better if the number of 17 dominated solutions is smaller. The diversity of the solutions is measured by the Spread 18 and the Uniformity Measure metrics estimation (35). For purpose of analysis, the delay, 19 the CO2, and the DeltaS variables are considered to have the same weight during the 20 optimization procedure. 21

Initially, the maximum number of generations was set to 2000 for all the test 22 instances, while the crossover and mutation rates were set at 90% and 10%, respectively. 23 Each scenario was then run 10 times in the NSGA-II code. For each repetition, the outputs 24 concerning the number of dominated solutions, Spread and Uniformity Measure, were 25 computed. Once the convergence to POF was guaranteed and diversity of the solutions in 26 all scenarios was accomplished, an equal maximum number of generations was used. 27 28


1 FIGURE 3 Basic structure of NSGA-II algorithm. 2

3

2.4. Scenarios 4 In this research, the baseline scenario was the validated simulation model with the 5 observed pedestrian (160 p/h) and traffic demands (100% of demand factor). A 6 preliminary analysis performed in the simulation demonstrated that pedestrian demands of 7 330 p/h and 135% of the traffic demand initiated traffic congestion in the coded network. 8 Thus, the effects of both the uniform pedestrian flows and traffic growth (only motor 9 vehicles) were explored at two levels: 10 11

Pedestrian demands: S1) 240 p/h, and S2) 320 p/h; 12

Stopping Criteria

Pareto-optimal Solution

Initiation of population with size N

t = 0

Rank the population according to non-dominating

criteria

Calculate all the objective functions

Binary Tournament Selection

Parents – Pt

Intermediate Crossover

Gaussian Mutation

Offsprings – Qt

Calculate the objective function of the new

population

Combine old and new population

Non-dominating ranking on the combined

population

Calculate the crowding distance of all the

solutions

Get the N from the combined population

Replace population by the better members of the

combined population

YES

NO

R t = Pt U Qt

Parents of new generation – Pt+1


Traffic growth: S3) 115%, and S4) 130% demand factors. 1 2

For each level, two main demand scenarios were defined: the first level evaluates 3 how vehicle’s delay, emissions, and pedestrian’s safety change with the increasing of 4 pedestrian traffic demand, assuming no changes on the traffic flows (S1 and S2); the 5 second level analyzes the performance of different pedestrian locations under different 6 traffic flows, assuming no changes on the pedestrians flows (S3 and S4). For all 7 crosswalks locations, the authors modeled the centroids where pedestrians enter and leave 8 in the coded network in the same place as the baseline scenario. Also, pedestrians only 9 walked to the crosswalk independently of its location. 10

Only crossing conflicts at the selected pedestrian crosswalk were taken into 11 account. It should be mentioned that SSAM identifies not only vehicle-to-pedestrian 12 conflicts, but also pedestrian-to-pedestrian conflicts. In fact, all traffic (vehicles and 13 pedestrians) appear in the *.trj file that is used by SSAM. To address this problem, the 14 authors filtered out any conflict where the maximum speed was lower than 2.2 m/s (which 15 is beyond the natural pedestrian walking speed). The delay and CO2 emissions per unit 16 distance were given from the vehicle record evaluation. Considering the safety analysis, 17 the conflicts’ classification was made according to the FHWA criteria (27). 18

These five scenarios (baseline, S1-4) were then applied, assuming several possible 19 locations for the pedestrian crosswalk (PC) from 5 to 60 meters in increments of 5 meters 20 (relatively to the RBT2 exit section). For all these scenarios, PC was measured from the 21 circulatory ring delimitation of RBT2 to the limit of crosswalk, as illustrated in Figure 2. 22 Three objective functions were optimized for each scenario: delay, CO2 emissions, and 23 DeltaS. 24

These functions were used as decision variables of the PC location subject to: 5 25 meters ≤ PC ≤ 60 meters. The regression functions were PC vs delay, PC vs CO2 26 emissions, and PC vs. DeltaS. A set of 10 optimal solutions was considered for this 27 analysis. 28

29 3. RESULTS AND DISCUSSION 30 31 3.1. Calibration and Validation 32 Figure 4 a and b exhibit the observed and estimated traffic volumes and vehicle speeds 33 before (with VISSIM default parameters) and after the calibration of the traffic model. The 34 results yielded larger improvements for vehicle speed counts while traffic volumes were 35 slightly modified. After the calibration, speeds improved for 75% (n=29) of the links. The 36 remaining speeds were close to the initial speed values. The analysis of the calibration 37 procedure also demonstrated that speed values were more sensitive than links’ values to 38 changes in the model parameters. 39

Table 2 summarizes the traffic calibration results obtained for NRMS, the GEH 40 statistic, as well as the total link volumes before (default) and after the calibration of the 41 model. Both lane-change parameters and simulation resolution were unaffected by the 42 calibration. The results confirmed that the calibrated model parameters improved the GEH 43 statistic. Every link achieved a GEH statistic less than 5, thereby satisfying the calibration 44 criteria. The NRSM went from 0.97 to 0.34. The total difference between observed and 45 estimated link volumes was approximately 1% for all links in the coded network. 46 47


a) Default model b) Calibrated model

FIGURE 4 Observed vs. Estimated traffic volumes and vehicle’s speed: a) default 1

model; b) calibrated model. 2 3

TABLE 2 Summary of calibration for the traffic model 4

Parameter Value NRMS GEH

Total Link

Volumes

[vph]

Model

Default

Average standstill distance (m) 2.5

0.970

< 5 for

86 % of

the cases

15,443

Additive part of safety distance 2.0

Multiple part of safety distance 3.0

Minimal gap time (s) 3.0

Minimal headway (m) 5.0

Calibrated

Model

Average standstill distance (m) 0.9

0.336

< 5 for 100

% of the

cases

15,957

Additive part of safety distance 1.3

Multiple part of safety distance 2.2

Minimal gap time (s) 2.8

Minimal headway (m) 4.9

Note: The observed total link volumes was 16,112 vph; 5 The weight factor (W) was set to 0.5. 6

7 Concerning validation, a comparison of observed and estimated O/D traffic flows at the 8 roundabouts was conducted using 15 random seed runs (26). It showed that 88% of the 34 9 loop detectors (all feasible movements for RBT1 and RBT2) reached GEH values below 5. 10 These validation results suggested a very good degree of consistency for all cases (25). 11 The analysis of the travel times (using 200 floating car runs) pointed out small differences 12 between observed and estimated data (1-3% for all movements). In the case of VSP 13 cumulative probability distributions, the two-sample K-S test at a 5% significance level 14 (D-value) for East-West and West-East movements were 0.036 (p-value = 0.155) and 15 0.030 (p-value = 0.364), respectively. Similar results were achieved in the remaining 16 movements. Therefore, there was evidence to suggest that the two VSP modes 17 distributions were the same for the observed and estimated routes. 18 19 3.2. Regression models 20 Figure 5 depicts the results for each of the scenarios by varying the location of the 21 pedestrian crosswalk from 5 and 60 meters to the RBT2 exit section. Various models were 22

0

150

300

450

600

0 150 300 450 600

Est

imat

ed v

olu

mes

[vp

h]

Observed volumes [vph]

0

150

300

450

600

0 150 300 450 600

Est

imat

ed v

olu

mes

[vp

h]

Observed volumes [vph]

0

15

30

45

60

0 15 30 45 60

Est

imat

ed s

pee

ds

[km

/h]

Observed speeds [km/h]

0

15

30

45

60

0 15 30 45 60

Est

imat

ed s

pee

ds

[km

/h]

Observed speeds [km/h]

R2 = 0.91 R2 = 0.94

R2 = 0.94 R2 = 0.63


tested to identify whether the predictive regression models were a good fit for the analyzed 1 data. 2

The results indicated that, regardless of the pedestrian demand (baseline, S1 and 3 S2), both delays and CO2 emissions per unit distance were meaningful for pedestrian 4 crosswalks placed less than 10 meters of the RBT2 exit section (Figure 5 a-b, d-e). On 5 average, delays and CO2 emissions at those locations were higher by 25% and 10%, 6 respectively over the average values recorded in the furthest crosswalks (PC>20 meters). 7 Similarly, the difference between vehicle and pedestrian speeds (DeltaS) was found to be 8 small for the crosswalks near the circulatory ring (PC<15 meters), increasing gradually for 9 the crosswalks close to the mid-block segment (PC>30 meters). From that location, it did 10 not vary significantly (~30 km/h). This result is possibly due to lower speeds that vehicles 11 experienced, since they did not reach the cruise speed over the mid-block section between 12 RBT1 and RBT2. 13

When the levels of traffic demand increased (S3 and S4), the impacts associated 14 with the location of the pedestrian crosswalk assumed a greater influence for locations next 15 to the limit of the RBT2 circulatory carriageway (Figure 5 j-m). In real time, the 16 visualization of the simulation demonstrated that the locations that provides stocking 17 capacities of one (PC=5 meters) and two vehicles (PC=10 meters) tend to generate queue 18 in the exit zone, whose extension reaches the circulation ring of RBT2. Subsequently, the 19 traffic from RBT2 almost backed up to RBT1. The values of delay and CO2 emissions per 20 unit distance on those locations confirmed these findings. Nonetheless, vehicles were 21 stopping at crosswalks close to the RBT2 exit section driving at low speeds, which 22 contributed to improving safety for pedestrians. 23

These results led to the conclusion that crosswalks close to a roundabout exit 24 section have a negative influence both on the entry capacity and CO2 emissions, and 25 induce congestion on the second roundabout, especially under high traffic demands. 26 Clearly, the trade-off among performance, environment, and safety was observed as the 27 pedestrian crosswalk is moved along the mid-block section. Rather than the resulting 28 outputs being read from the graphs, the regression equations were employed in the NSGA-29 II algorithm to identify possible optimal solutions. 30 31


Base

lin

e a) b) c)

S1

d) e) f)

S2

g) h) i)

S3

j) k) l)

S4

m) n) o)

FIGURE 5 Regression models results for scenario PC vs. Delay, PC vs. CO2 and PC 1 vs. DeltaS: baseline scenario (a, b, c), S1 (d, e, f), S2 (g, h, i), S3 (j, k, l) and S4 (m, n, 2

o). 3 4

6

8

10

12

0 10 20 30 40 50 60 70

Del

ay [

s]

PC [m]

delay = -7.3×10-5PC3

+9.4×10-3PC2-0.4PC+12.9

adjusted R2=0.93

0

50

100

150

200

0 10 20 30 40 50 60 70

CO

2E

mis

sion

s [g

/km

]

PC [m]

CO2 =1.1×10-5PC4-

1.7×10-3PC3+0.1PC2-

3.2PC+204.3

adjusted R2=0.98

0

8

16

24

32

40

0 10 20 30 40 50 60 70

Del

taS

[k

m/h

]

PC [m]

DeltaS = 1.4×10-5PC4-

1.6×10-3PC3+5.4×10-2PC2

+16.0

adjusted R2=0.93

6

8

10

12

14

0 10 20 30 40 50 60 70

Del

ay [

s]

PC [m]

delay =2.7×10-6PC4 -

4.8×10-4PC3+3.0×10-2PC2 -

0.8PC+16.2

adjusted R2=0.97

0

50

100

150

200

0 10 20 30 40 50 60 70

CO

2E

mis

sion

s [g

/km

]

PC [m]

CO2 =8.9×10-6PC4-

1.5×10-3PC3+0.1PC2-

2.8PC+201.4

adjusted R2=0.96

0

8

16

24

32

40

0 10 20 30 40 50 60 70

Del

taS

[k

m/h

]

PC [m]

DeltaS = 1.7×10-5PC4-

2.0×10-3PC3+6.4×10-2PC2

+14.5

adjusted R2=0.90

6

8

10

12

14

16

0 10 20 30 40 50 60 70

Del

ay [

s]

PC [m]

delay =9.3×10-7PC4 -

2.7×10-4PC3+2.4×10-2PC2

-0.8PC+18.8

adjusted R2=0.90

0

35

70

105

140

175

210

0 10 20 30 40 50 60 70

CO

2E

mis

sion

s [g

/km

]

PC [m]

CO2 =2.1×10-5PC4 -

3.3×10-3PC3 +0.2PC2-

5.0PC+220.7

adjusted R2=0.97

0

8

16

24

32

40

0 10 20 30 40 50 60 70

Del

taS

[k

m/h

]

PC [m]

DeltaS = 1.5×10-5PC4-

1.8×10-3PC3 +5.9×10-2PC2

+14.4

adjusted R2=0.89

0

5

10

15

20

25

30

0 10 20 30 40 50 60 70

Del

ay [

s]

PC [m]

delay =7.3×10-6PC4 -

1.3×10-3PC3 +7.8×10-2PC2

-2.1PC+34.9

adjusted R2=0.96

0

60

120

180

240

0 10 20 30 40 50 60 70

CO

2E

mis

sion

s [g

/km

]

PC [m]

CO2 =3.3×10-5PC4-

5.2×10-3PC3+0.3PC2-

6.6PC+250.5

adjusted R2=0.96

0

8

16

24

32

40

0 10 20 30 40 50 60 70

Del

taS

[k

m/h

]

PC [m]

DeltaS = 1.1×10-5PC4-

1.3×10-3PC3 +4.4×10-2PC2

+11.7

adjusted R2=0.94

0

10

20

30

40

50

0 10 20 30 40 50 60 70

Del

ay [

s]

PC [m]

delay =-4.2×10-4PC3

+5.4×10-2PC2 -2.2PC+55.4

adjusted R2=0.940

70

140

210

280

0 10 20 30 40 50 60 70

CO

2E

mis

sion

s [g

/km

]

PC [m]

CO2 =-6.0×10-3PC3

+7.7×10-2PC2-

3.3PC+282.3

adjusted R2=0.90

0

8

16

24

32

40

0 10 20 30 40 50 60 70

Del

taS

[k

m/h

]

PC [m]

DeltaS = 1.2×10-5PC4-

1.5×10-3PC3+4.7×10-2PC2

+12.7

adjusted R2=0.89


3.3. Multi-objective optimization 1 This section presents the main results of the multi-objective optimization of delay, CO2 2 emissions and DeltaS as a result of the crosswalk (PC) location along the mid-block 3 section, for each one of the scenarios previously defined (baseline, S1-S4).The analysis of 4 the convergence to POF and the diversity of solutions indicated that a maximum of 300 5 iterations were sufficient to reach convergence. With reference to crossover and mutation, 6 each solution of the final POF did not vary much with different values of different rates. 7 Thus, the crossover rate was set at 90% and the mutation rate was set at 10%. 8

Solutions resulting from the evaluated scenarios are summarized in Table 3. Figure 9 6 illustrates the Pareto fronts estimated from the initial (1st iteration) and final (300th 10 iteration) populations. For each scenario, a 3-D scatter plot with the three objective 11 functions – CO2 Emissions (x-axis), delay (y-axis), DeltaS (z-axis) – as a function of PC is 12 shown. 13

It was confirmed that the approximate Pareto front moves markedly in the lower-14 left direction (Figure 6-a) in the baseline scenario, which means that crosswalks placed 15 more than 32 meters to the circulatory ring are not good solutions considering the current 16 traffic conditions. It was clear that adopting a crosswalk placed at 15 meters away from the 17 RBT2 section (Solution 6), average delay and CO2 emissions per unit distance decreased 18 by 19% and 7% respectively, while the difference between vehicles and safety speed was 19 increased by no less than 37% in comparison with the lower-level solution (Solution 1). 20 Crosswalks located at 20 and 31 meters away from RBT2 are also proved as optimal 21 solutions on the selected case study (Table 3). Similar results were found in S1 and S2 22 scenarios. 23

Regarding the high traffic demand scenarios (S3 and S4), the final Pareto front 24 moved in the upper-left and lower-right directions, that is, the solutions associated to the 25 faraway (PC = 60 meters) and nearest crosswalks (PC = 5 meters) respectively from the 26 RBT2 section. In the first case, this was explained by the lower values of DeltaS (~24 27 km/h) on those locations in comparison to the lower demand scenarios, i.e., baseline, S1 28 and S2 (~30 km/h). Also, the average values of Delay and CO2 emissions decreased as the 29 crosswalk was placed farther away from the RBT2 exit section (Figure 5 j-k, n-m). For 30 instance, if one adopted a crosswalk 60 meters away from the exit section (assuming the 31 traffic conditions of S4), then one could save up to 43% and 13% in average delay and 32 CO2 emissions, respectively, while DeltaS increases by 74% (Table 3). In the second case, 33 high congestion levels resulted in low DeltaS values near to RBT2 exit section (~14 km/h), 34 and therefore such location was pointed out as a solution to be implemented at the arterial. 35

These findings suggested that the location of the crosswalk far away from the 36 circulatory carriageway can also be used (PC=55/60 meters), especially under high traffic 37 conditions. In such cases, safety for pedestrians was perceived as slightly negatively 38 impacted, since the spacing of the roundabouts constrains the vehicles speeds at the mid-39 block segments. However, locating the crosswalk in further distances from the downstream 40 roundabout (e.g. PC=80 meters) affected the traffic operations in the upstream roundabout. 41 Although the crosswalks near to the exit section were undoubtedly safe for pedestrians, 42 their practical implementation would not provide any global competition in capacity and 43 environmental point of views. Previous research demonstrated that for high pedestrian 44 demand (P>400 p/h) at 70% of the saturation rate, a crosswalk was ineffective on traffic 45 operations when located less than 15 meters away from the roundabout delimitation (5). 46 47


TABLE 3 Solution lists of the pedestrian crosswalk locations (after 300 iterations) 1

Scenario Solution

No.

PC

[m]

Delay

[s]

CO2

[g/km]

DeltaS

[km/h]

Baseline

1 5.000 11.310 191.307 17.147

2 5.000 11.310 191.307 17.147

3 7.226 10.710 187.133 18.236

4 9.033 10.280 184.380 19.289

5 12.338 9.618 180.653 21.465

6 15.238 9.160 178.606 23.486

7 15.993 9.057 178.240 24.012

8 20.438 8.587 177.297 26.956

9 31.303 8.188 181.696 31.694

10 31.303 8.188 181.696 31.694

S1

1 5.000 12.918 190.021 15.856

2 5.000 12.918 190.021 15.856

3 5.413 12.701 189.273 16.067

4 7.358 11.786 186.102 17.210

5 11.001 10.477 181.544 19.811

6 14.086 9.715 178.865 22.238

7 18.541 9.047 176.479 25.670

8 27.115 8.678 175.072 30.550

9 27.627 8.679 175.069 30.735

10 27.627 8.679 175.069 30.735

S2

1 5.000 15.154 200.267 15.643

2 5.000 15.154 200.267 15.643

3 8.010 13.459 191.774 17.280

4 11.347 11.962 184.996 19.522

5 14.496 10.871 180.647 21.811

6 18.544 9.862 177.284 24.682

7 20.875 9.451 176.198 26.181

8 27.412 8.819 175.213 29.339

9 32.253 8.710 175.449 30.443

10 32.253 8.710 175.449 30.443

S3

1 5.000 26.223 223.778 12.674

2 5.000 26.223 223.778 12.674

3 5.388 25.684 222.197 12.813

4 8.920 21.541 210.476 14.371

5 11.272 19.463 205.001 15.621

6 15.223 16.962 199.061 17.888

7 22.958 14.620 195.158 22.069

8 55.666 13.919 192.001 24.878

9 56.744 13.925 191.953 25.006

10 56.744 13.925 191.953 25.006

S4

1 5.000 45.628 267.678 13.721

2 5.000 45.628 267.678 13.721

3 8.275 40.557 259.975 14.146

4 10.311 37.881 255.866 15.208

5 14.226 33.670 249.299 17.415

6 17.756 30.823 244.731 19.381

7 21.918 28.465 240.772 21.420

8 27.500 26.694 237.448 23.360

9 60.000 25.899 231.841 23.937

10 60.000 25.899 231.841 23.937

2


a) Baseline

b) S1

c) S2

d) S3

e) S4

FIGURE 6 The approximate initial (left) and final (right) Pareto front: a) Baseline; 1

b) S1; c) S2; d) S3 and e) S4. 2

810

12 170

180

1900

20

40

CO2 Emissions (g/km)

Generation 28 / 28

Delay (s)

Del

taS

(k

m/h

)Optimal solutions on Pareto Front

810

12 160180

20010

20

30

40


Generation 300 / 300

Delay (s)

Del

taS

(k

m/h

)

Optimal solutions on Pareto Front

810

12 175

180

18510

20

30

40


Generation 4 / 4

Delay (s)

Del

taS

(k

m/h

)


510

15 160

180

20010

20

30

40



Delay (s)

Del

taS

(k

m/h

)


5

10

15 160

180

200

10

20

30

40


Generation 4 / 4

Delay (s)

Del

taS

(k

m/h

)


0

10

20 150

200

25010

20

30

40

CO2 Emissions (g(km)


Delay (s)

Del

taS

(k

m/h

)


20

40

60 200

250

30010

15

20

25


Generation 5 / 5

Delay (s)

Del

taS

(k

m/h

)


20

40

60 200

250

30010

15

20

25



Delay (s)

Del

taS

(k

m/h

)


2626.5

2727.5 232

234

236

238

22

23

24

25


Generation 4 / 4

Delay (s)

Del

taS

(k

m/h

)


2030

4050 200

250

30010

15

20

25



Delay (s)

Del

taS

(k

m/h

)


1

2 3

4 5

6 7 8

9

10

1

2

3 4 5 6

7 8

9 10

1 2

3 4 5 6

7 8 9

10

1 2

3 4 5

6

7 8 9

10

1

2 3

4 5

6 7

8 9

10

Initial Final

Initial Final

Initial Final

Initial

Initial

Final

Final


4. CONCLUSIONS 1 This paper explored different pedestrian crosswalk locations between two closely spaced 2 roundabouts in terms of capacity/delays, the environment, and pedestrian’s safety. The 3 analysis was based on a microsimulation approach, using a traffic model combined with 4 emission and safety models. Three regression models were established to express the 5 trade-off among delay, CO2 emissions per unit distance, and the difference between 6 vehicles’ and pedestrians’ speeds as a function of several pedestrian crosswalk locations 7 along the mid-block sub-segment. As a solution algorithm for the models, NSGA-II was 8 used to search the optimal solutions for the proposed problem. 9

Crosswalks near the circulating carriageway (< 10 meters) were associated with 10 weak performance levels and high CO2 emissions rates. The distances of 15, 20 and 30 11 meters were predicted to be appropriate in this an interrupted traffic facility, regardless of 12 the pedestrian and traffic demands. It was also found that for high traffic flows the location 13 of the crosswalks near the mid-block (55/60 meters) improved the capacity and emissions 14 use without affecting significantly the safety of pedestrians. 15

This methodology can be tailored to analyze other arterials with closely-spaced 16 roundabouts in which crosswalks are located at the mid-block segments, and whose 17 impacts on traffic are not thoroughly evaluated. Obviously, vehicular capacity or emissions 18 used are not the only considerations for the assessment of crosswalks locations at 19 roundabouts. The improvement of the safety for pedestrians remains the most important 20 “selling point” of any good crosswalk location. Still, the findings in this study provide 21 relevant information for local authorities for the balanced implementation of a pedestrian 22 crosswalk by taking into account the trade-off among capacity, environment, and safety. A 23 site’s specific operational conditions may favor other pedestrian crosswalks locations and, 24 as a result, the models should be calibrated for each location (with its own traffic, 25 pedestrian demands, and driving patterns). 26

There were three main limitations that must be outlined. The first was that only the 27 impacts on the pedestrian crosswalk were taken into account; the remaining crosswalks and 28 the pedestrian patterns were excluded from this analysis. The second was the absence of 29 specific measures that can reflect the pedestrian performance, as delay. The third limitation 30 was that pedestrians only walked to the crosswalk. 31

Therefore, future work is needed to enhance and calibrate the pedestrians patterns 32 along the coding network as well as to include those additional measures that can reflect the 33 impact on pedestrians by moving a crosswalk along the arterial. A relationship between the 34 crosswalk location and speed could also be examined. The assessment of additional 35 objective environmental variables (such as carbon monoxide, hydrocarbons or nitrogen 36 oxides) to the optimization procedure would be explored in future work. 37 38 ACKOWLEDGEMENTS 39 This work was partially funded by FEDER Funds through the Operational Program 40 “Factores de Competitividade COMPETE” and by National Funds through FCT within the 41 project PTDC/SEN-TRA/122114/2010 and the Strategic Project PEst-42 C/EME/UI0481/2013. P. Fernandes acknowledges the support of FCT for the scholarship 43 SFRH/BD/87402/2012. 44 45 REFERENCES 46

1. Isebrands, H., S. Hallmark, E. Fitzsimmons, and J. Stroda. Toolbox to Evaluate the 47 Impacts of Roundabouts on a Corridor or Roadway Network. Publication MN/RC 48 2008-24, Minnesota Department of Transportation Research Services Section, 2008. 49


2. Rodegerdts, L.; et al; Roundabouts: An Informational Guide Second Edition. NCHRP 1 Report, Issue 672, 2010, 407 pp. 2

3. Schroeder, B., N. M. Rouphail, K. Salamati, and Z. Bugg. Effect of Pedestrian 3 Impedance on Vehicular Capacity at Multilane Roundabouts with Consideration of 4 Crossing Treatments. In Transportation Research Record: Journal of the 5 Transportation Research Board, No. 2312, Transportation Research Board of the 6 National Academies, Washington, D.C., 2012, pp. 14-24. 7

4. DfT. Geometric Design of Roundabouts. Design Manual for Roads and Bridges - Road 8 Geometry Junctions. Vol. 6, Section 2, Part 3. Report TD16/2007, Department of 9 Transport, United Kingdom, 2007. 10

5. Silva, A. B., J. Cunha, T. Relvão, and J. P. Silva. Evaluation of Effect of Pedestrian 11 Crossings on Roundabout Performance Using Microsimulation. Presented at 92nd 12 Annual Meeting of the Transportation Research Board, Washington, D.C., 2013. 13

6. HCM. Highway Capacity Manual. Transportation Research Board, Washington, D.C., 14 2010. 15

7. Hellinga, B., and A. Sindi. Analytical Method for Estimating Delays to Vehicles 16 Traversing Single-Lane Roundabouts as a Function of Vehicle and Pedestrian 17 Volumes. In Transportation Research Record: Journal of the Transportation Research 18 Board, No. 2312, Transportation Research Board of the National Academies, 19 Washington, D.C., 2012, pp. 56-66. 20

8. Salamati, K., B. Schroeder, D. Geruschat, and N. M. Rouphail. Event-Based Modeling 21 of Driver Yielding Behavior to Pedestrians at Two-Lane Roundabout Approaches. In 22 Transportation Research Record: Journal of the Transportation Research Board, No. 23 2389, Transportation Research Board of the National Academies, Washington, D.C., 24 2013, pp. 1-11. 25

9. Tollazzi, T., T. Lerher, and M. Šraml. Simulation of the Pedestrians Influence to the 26 Capacity of Motorised Vehicles in a Roundabout. American Journal of Applied 27 Sciences, Vol. 5, No. 1, 2008, pp. 34-41. 28

10. Schroeder, B., N. M. Rouphail, and R. Hughes. Toward Roundabout Accessibility—29 Exploring the Operational Impact of Pedestrian Signalization Options at Modern 30 Roundabouts. Journal of Transportation Engineering, Vol. 134, No. 6, 2008, pp. 262-31 271. 32

11. Ashmead, D., D. Guth, R. Wall, R. Long, and P. Ponchillia. Street Crossing by Sighted 33 and Blind Pedestrians at a Modern Roundabout. Journal of Transportation 34 Engineering, Vol. 131, No. 11, 2005, pp. 812-821. 35

12. Guth D., D. Ashmead, R. Long, R. Wall, and P. Ponchillia. Blind and Sighted 36 Pedestrians' Judgments of Gaps in Traffic at Roundabouts. Human Factors, Vol. 47, 37 No. 2, 2005, pp. 314-342. 38

13. Salamati, K., B. Schroeder, N. M. Rouphail, C. Cunningham, Y. Zhang, and D. Kaber. 39 Simulator Study of Driver Responses to Pedestrian Treatments at Multilane 40 Roundabouts. In Transportation Research Record: Journal of the Transportation 41 Research Board, No. 2312, Transportation Research Board of the National Academies, 42 Washington, D.C., 2012, pp. 67-75. 43


14. Coelho, M. C., T. L. Farias, and N. M. Rouphail. Effect of roundabout operations on 1 pollutant emissions. Transportation Research Part D: Transport and Environment, 2 Vol. 11, No. 5, 2006, pp. 333-343. 3

15. Ahn, K., N. Kronprasert, and H. A. Rakha. Energy and Environmental Assessment of 4 High-Speed Roundabouts. In Transportation Research Record: Journal of the 5 Transportation Research Board, No. 2123, Transportation Research Board of the 6 National Academies, Washington, D.C., 2009, pp. 54-65. 7

16. Rakha, H. A., and M. Jackson. Are Roundabout Environmentally Friendly? An 8 Evaluation for Uniform Approach Demands. Presented at 91st Annual Meeting of the 9 Transportation Research Board, Washington, D.C., 2012. 10

17. Salamati, K., M. C. Coelho, P. J. Fernandes, N. M. Rouphail, H. C. Frey, and J. 11 Bandeira. Emission Estimation at Multilane Roundabouts: Effect of Movement and 12 Approach Lane. In Transportation Research Record: Journal of the Transportation 13 Research Board, No. 2266, Transportation Research Board of the National Academies, 14 Washington, D.C., 2013, pp. 12-21. 15

18. Hallmark, S., E. Fitzsimmons, H. Isebrands, and K. Giese. Roundabouts in Signalized 16 Corridors: Evaluation of Traffic Flow Impacts. In Transportation Research Record: 17 Journal of the Transportation Research Board, No. 2182, Transportation Research 18 Board of the National Academies, Washington, D.C., 2010, pp. 139-147. 19

19. Krogscheepers, J., and M. Watters. Roundabouts along Rural Arterials in South Africa. 20 Presented at 93rd Annual Meeting of the Transportation Research Board, Washington, 21 D.C., 2014. 22

20. Bak, R., and M. Kiec. Influence of Midblock Pedestrian Crossings on Urban Street 23 Capacity. In Transportation Research Record: Journal of the Transportation Research 24 Board, No. 2316, Transportation Research Board of the National Academies, 25 Washington, D.C., 2012, pp. 76-83. 26

21. Statistics Portugal. Censos 2011 [In Portuguese] http://mapas.ine.pt/map.phtml, 27 Accessed June 24, 2014. 28

22. PTV AG. VISSIM 5.30-05 user manual. Planung Transport Verkehr AG, Karlsruhe, 29 Germany, 2011. 30

23. ACAP. Automobile Industry Statistics 2011 Edition. ACAP – Automobile Association 31 of Portugal, 2012. 32

24. US EPA. Methodology for developing modal emission rates for EPA’s multi-scale 33 motor vehicle & equipment emission system. Prepared by North Carolina State 34 University for US EPA, Ann Arbor, MI, August, EPA420, 2002. 35

25. Frey, H. C., K. Zhang, and N. M. Rouphail. Fuel Use and Emissions Comparisons for 36 Alternative Routes, Time of Day, Road Grade, and Vehicles Based on In-Use 37 Measurements. Environmental Science & Technology, Vol. 42, No. 7, 2008, pp. 2483-38 2489. 39

26. Coelho, M. C., H. C. Frey, N. M. Rouphail, H. Zhai, and L. Pelkmans. Assessing 40 methods for comparing emissions from gasoline and diesel light-duty vehicles based 41 on microscale measurements. Transportation Research Part D: Transport and 42 Environment, Vol. 14, No. 2, 2009, pp. 91-99. 43

http://mapas.ine.pt/map.phtml


27. Gettman, D., L. Pu, T. Sayed, and S. G. Shelby. Surrogate safety assessment model and 1 validation: final report. Publication FHWA-HRT-08-051, FHWA, U.S. Department of 2 Transportation, 2008. 3

28. Vasconcelos, L., L. F. Cavaco, A. Seco, and A. B. Silva. Validation of Surrogate 4 Safety Analysis Module Technique for Assessment of Intersection Safety. Presented at 5 93rd Annual Meeting of the Transportation Research Board, Washington, D.C., 2014. 6

29. Salamati, K., B. Schroeder, N. M. Rouphail, C. Cunningham, R. Long, and J. Barlow. 7 Development and Implementation of Conflict-Based Assessment of Pedestrian Safety 8 to Evaluate Accessibility of Complex Intersections. In Transportation Research 9 Record: Journal of the Transportation Research Board, No. 2264, Transportation 10 Research Board of the National Academies, Washington, D.C., 2011, pp. 148-155. 11

30. Paz, A., V. Molano, and A. Khan. Calibration of Microscopic Traffic Flow Models 12 Considering all Parameters Simultaneously. Presented at 93rd Annual Meeting of the 13 Transportation Research Board, Washington, D.C., 2014. 14

31. Dowling, R., A. Skabadonis, and V. Alexiadis. Traffic analysis toolbox, volume III: 15 Guidelines for applying traffic microsimulation software. Publication FHWA-HRT-04-16 040, FHWA, U.S. Department of Transportation, 2004. 17

32. Hale, D. “How many netsim runs are enough?”. McTrans, Vol. 11, No. 3, 1997, pp. 1-18 9. 19

33. Deb, K., A. Pratap, S. Agarwal, and T. Meyarivan. A fast and elitist multiobjective 20 genetic algorithm: NSGA-II. Evolutionary Computation, IEEE Transactions on 21 Evolutionary Computation, Vol. 6, No. 2, 2002, pp.182-197. 22

34. Konak, A., D. W. Coit, and A. E. Smith. Multi-objective optimization using genetic 23 algorithms: A tutorial. Reliability Engineering & System Safety, Vol. 91, No. 9, 2006, 24 pp. 992-1007. 25

35. Deb, K. Multi-Objective Optimization Using Evolutionary Algorithms. John Wiley & 26 Sons, 2001. 27

multi-criteria assessment of crosswalk ...transportes-tema.web.ua.pt/publications/177.pdf38 in...

Documents