Supporting Material to NCHRP Report 674 (2011)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

APPENDIX L: Details on Roundabout Signalization Modeling This Appendix was previously published as conference proceedings at the 86th Annual Meeting of the Transportation Research Board, January 21-25, 2007. The citation for this work is: Schroeder, Bastian J., Nagui M. Rouphail and Ron Hughes. Exploratory Analysis of Pedestrian Signalization Treatments at One- and Two-Lane Roundabouts Using VISSIM Microsimulation. 86th Annual Meeting of the TRB, 2007. 200

Schroeder, Rouphail, and Hughes Exploratory Analysis of Pedestrian Signalization Treatments at One- and Two-Lane Roundabouts Using VISSIM Microsimulation By Bastian Jonathan Schroeder, E.I.* Graduate Research Assistant Institute of Transportation Research and Education (ITRE) North Carolina State University Centennial Campus, Box 8601 Raleigh, NC 27695-8601 Tel.: (919) 515-8565 Fax: (919) 515-8898 Email: Bastian_Schroeder@ncsu.edu Nagui M. Rouphail, Ph.D. Director, Institute for Transportation Research and Education (ITRE) Professor of Civil Engineering North Carolina State University Centennial Campus, Box 8601 Raleigh, NC 27695-8601 Tel.: (919) 515-1154 Fax: (919) 515-8898 Email: rouphail@eos.ncsu.edu Ron Hughes, Ph.D. Director, Visual Analytics, Modeling and Simulation (VAMS) Group Institute for Transportation Research and Education (ITRE) North Carolina State University Centennial Campus, Box 8601 Raleigh, NC 27695-8601 Tel.: (919) 515-8523 Fax: (919) 515-8898 Email: rghughes@ncsu.edu November 2006 Submitted for publication and presentation at the 86th Annual Meeting of the Transportation Research Board, January 21-25, 2007 Word Count: 5,551 text words plus 2,000 for figures/tables (8*250) = 7,551 total * Corresponding Author 201

Schroeder, Rouphail, and Hughes ABSTRACT This paper explores the use of signalized pedestrian crossing treatments at one- and two-lane roundabout facilities. Motivated through increasing debate on the safety of roundabouts for pedestrians, this paper assesses the potential for signalization as a means for regulating the interaction of vehicles and pedestrians at these facilities. The use of pedestrian signals at roundabouts is controversial because of the potential for queue spillback into the circulating lane. This paper aims to quantify the effects of different signalization treatments through microsimulation. The paper uses the microscopic modeling tool VISSIM to estimate impacts on pedestrian and vehicle delay for different crossing geometries and signalization schemes. The range of alternate crossing geometries includes ‘proximal’, ‘zig-zag’, and ‘distal’ crossings with varying offset distances of entry and/or exit crosswalk from the circulating lane. The modeled signalization options include one-stage and two-stage pedestrian-actuated control, as well as, the use of HAWK signals. The vehicle models for one- and two-lane roundabouts have been calibrated and will be used to conduct sensitivity analyses for a range of pedestrian and vehicle demands for the different scenarios. The results suggest that the impact of a pedestrian signal at roundabouts is greatest as vehicle volumes approach capacity, but that vehicle delay and queuing can be minimized through innovative signal configurations. The findings are important in light of recent discourse concerning the accessibility of roundabouts to pedestrians with vision impairments that may ultimately move towards a requirement for signalization for certain facility types. 202

Schroeder, Rouphail, and Hughes INTRODUCTION The installation of pedestrian crossing signals at roundabout facilities is a controversial topic in the traffic engineering community. While some US cities have experimented with their use and while their application is more common in Europe and Australia, a common contention is that any form of signalization disrupts the flow of traffic in a roundabout. The attractiveness of a well-designed roundabout is the ability of vehicles to navigate this unsignalized intersection form in a safe and efficient manner. But as more roundabouts are being designed in pedestrian-intensive urban areas, there is a need to evaluate their accessibility for the pedestrian mode. In fact, many downtown revitalization and gateway projects that include roundabouts also focus on a significant pedestrian element. While signals are not the only imaginable treatment to facilitate pedestrian access to modern roundabouts, they are a certain contender in areas of heavy vehicular traffic and at multi-lane facilities. This paper uses a microsimulation approach to assess and compare different alternative signalization treatments at a one-lane and a two-lane roundabout. The analysis includes an evaluation of modified crosswalk geometries and signalization schemes under a range of pedestrian and vehicular volumes. The goal of this effort is to explore these alternatives and provide traffic engineers with a quantitative basis for the discussion of roundabout signalization. BACKGROUND The accessibility of modern roundabouts for pedestrians with vision impairments has received a lot of attention in recent years. At unsignalized facilities, blind pedestrians have to rely on auditory cues when making a crossing decision – a task that is complicated through the ambient noise and uninterrupted flow at roundabouts. Ashmead et al (2002) found that these facilities indeed pose serious difficulties for blind pedestrians. More specifically, researchers have found that crossing becomes increasingly difficult as the conflicting vehicle volume increases and that multi-lane facilities are more challenging than single-lanes (Wall et al. 2005). Guth at al (2005) further showed that crossings at roundabout exit legs are more difficult than at entry legs. The objective of the NCHRP 3-78 research effort is to identify treatments that hold potential for improving access of blind pedestrians to modern roundabouts and channelized right-turn lanes, while maintaining acceptable vehicle levels of service. The research is working to develop a toolbox of treatments to reduce pedestrian risk and delay. The final list of treatments will cover a range of low to high-cost alternatives, distinguish retrofit treatments and guidance for new site construction, and discuss treatments for varying levels of geometry. The authors have submitted a discussion of an evaluation framework of unsignalized facilities in a separate paper for consideration for publication and presentation at the 86th Annual Meeting of the Transportation Research Board. In the mix of treatments, signalized alternatives are attractive in terms of providing accessibility, but fall on the high-cost end of the spectrum. If outfitted with audible pedestrian signals (APS), signals presumably assure equal access to pedestrians with vision impairments. In fact, the ‘Revised Draft Guidelines for Accessible Public Rights-of-Way’ (US Access Board, 2005) call 203

Schroeder, Rouphail, and Hughes for the provisions of “pedestrian activated signals … for each segment of each crosswalk, including the splitter island” at multi-lane facilities. The perceived trade-off of this accessibility from a traffic engineering perspective is an interruption of the intended unsignalized operations of the roundabout. Of concern is especially the increased likelihood of queue spillback into the circulating lane from the exit leg crossing. While any signalization treatment is intuitively associated with some added delay to vehicular traffic, it is unclear as to how much impact a pedestrian signal would actually have on roundabout operations. The abilities of modern microsimulation software offer the unique opportunity to modeling such treatments in a laboratory setting and evaluating the impact of signals prior to implementation. APPROACH The objective of this paper is to evaluate the pedestrian-induced impacts of roundabout signalization on vehicular performance. Using calibrated models of a one-lane and a two-lane roundabout, the authors simulated varying signalization options at one approach to the roundabout and compared performance measures to the no-pedestrian base case. Microsimulation offers a method for unobtrusive evaluation of a range of treatments, implemented at a range of volumes, while minimizing data collection cost. The team used the VISSIM simulation model, because its link-connector structure offers great flexibility in modeling unique roundabout and crosswalk geometries. VISSIM is further able to model user- defined ‘priority rules’ by vehicle or pedestrian class (PTV, 2005) and includes flexible signal control logic to model unconventional signalization schemes. The evaluation of other simulation packages such as CORSIM, Paramics or AIMSUN was beyond the scope of this effort. This paper assesses signalization alternatives at roundabouts in three dimensions: crossing geometry, signal phasing schemes, and traffic/pedestrian intensity. Crossing Geometry The analysis included three alternative crosswalk configurations for roundabouts. The default crossing configuration at most roundabouts in the US is to place the pedestrian crossing at the splitter island. In the following, this will be referred to as the proximal crossing location. Under the premise that the biggest concern for roundabout signalization from a traffic operations perspective is the potential for queue spillback into the circulating lane, the team experimented with two alternate crosswalk configurations that move all or part of the crossing further away from the circle. In the zig-zag crossing configuration, the exit leg component of the pedestrian crossing is ‘off-set’ by a predefined distance to allow for additional queue storage on the exit leg. Assuming that the default proximal crossing location is at a distance of 20 feet (6.1 meters or approximately one car length) from the circulating lane, the zig-zag configuration moves the exit portion of the crosswalk to a distance of 60 feet (18.3 meters) from the circle. Theoretically, this allows for two additional vehicles per lane to be stored before encroaching on circulating traffic. 204

Schroeder, Rouphail, and Hughes Following the same reasoning, the distal crossing configuration moves the entire crosswalk to a distance of 100 feet (30.5 meters) from the circulating lane. In this set-up, both entry and exit leg portions will be moved, to prevent pedestrians from having to travel too far in a longitudinally extended splitter island. The distal crossing theoretically allows for queue storage of 5 vehicles per lane at the exit leg. Figure 1 shows VISSIM screenshots of the proximal, zig-zag and distal crossing configurations for the one-lane roundabout. Figure 1: Roundabout Crosswalk Configurations 1-a: Proximal Crossing located 20’ (6.1m) from circulating lane 1-b: Zig-Zag Crossing located 60’ (18.3m) from circulating lane 1-c: Distal Crossing located 100’ (30.5m) from circulating lane 20’ PROXIMAL CROSSING 100’ DISTAL CROSSING 60’ ZIG-ZAG CROSSING 205

Schroeder, Rouphail, and Hughes Signal Phasing Schemes The analysis compares two different signalization schemes: a conventional pedestrian-actuated (PA) signal, and a pedestrian-actuated ‘High-Intensity Activated crossWalk’ (HAWK) signal. The main characteristic of a HAWK signal that distinguishes it from a PA is that vehicles are allowed to ‘proceed with caution’ during the pedestrian Flashing Don’t Walk (FDW) phase. This is achieved by including a Flashing Red (FR) phase for vehicles in the phasing sequence. Figure 2 illustrates this concept in a side-by-side comparison with a conventional PA signal. Figure 2: Comparing Phasing Sequences of 'Conventional PA' and 'HAWK' signals VE H IC LE S PE D ES TR IA N S VE H IC LE S PE D ES TR IA N S R DW R DW R DW FR DW DW Y Y DWDW G DW blank FDW blank DW FY DW R W FR FDW DWG W R Conventional Signal DWG HAWK Signal Pedestrian Actuation In the absence of a pedestrian actuation, the HAWK signal indication for vehicles is blank, meaning that the signal heads are not illuminated. Once a pedestrian places a call to the signal, the HAWK signal switches to a flashing yellow (FY) indication to alert the driver that a pedestrian is waiting to cross. The HAWK signal then goes through a sequence of yellow (Y), red (R), and pedestrian walk (W) phases, just as a conventional signal would. However, once the pedestrian signal indication switches to flashing don’t walk (FDW) the vehicle indication becomes a flashing red (FR). Similar to the flashing red indication at a signalized intersection in ‘nighttime flashing mode’, driver need to stop and give the right-of-way to the conflicting stream, 206

Schroeder, Rouphail, and Hughes in this case the crossing pedestrians. After the pedestrian has left the crosswalk, vehicles can proceed with caution and do not have to wait for the entire FDW clearance interval to elapse as they would at a conventional signal. It is important to note, that from a pedestrian perspective, the sequences of a conventional PA and a HAWK signal are identical. HAWK signals are currently used in Tucson, AZ (Tucson DOT, 2006) at signalized pedestrian mid-block crossings and may be used in other municipalities in the US. Their proposed benefit from a vehicle operations perspective is a shorter delay to drivers. Especially at multi-lane facilities, the required clearance time for the pedestrian FDW indication can be very long – a function of pedestrian walking speed and the crossing distance. By allowing drivers to proceed with caution as soon as the pedestrian has left the conflict area, the average waiting time for vehicles can presumably be reduced significantly, without sacrificing pedestrian safety or delay. Just as at midblock crossing, pedestrian signals at roundabouts operate independently of any minor street (at a fully signalized intersection with pedestrian phases a HAWK scheme wouldn’t be applicable). It is reasoned that a HAWK signal could provide significant improvements to vehicle delay when compared to a conventional pedestrian-actuated signal; especially at long two-lane roundabout crossings. Population Parameters and Intensities In addition to varying crosswalk geometry and signalization schemes, the analysis included a range of pedestrian and vehicle volumes. Each modeling scenario was analyzed at volumes of zero, 10 and 50 pedestrians per hour. To assess the variability of the pedestrian effect as a function of the frequency of actuations per hour, selected scenarios were tested at an even greater range of pedestrian volumes. All performance measures of interest, including vehicular delay and queuing were analyzed as pedestrian-induced impacts, defined as the difference between a measure at some pedestrian volume compared to the zero-pedestrian case. This form of comparison is possible in a microsimulation environment, if the same random number seeds are used in the two scenarios. With the same random seed, the model will generate the exact same distribution of vehicles, thereby isolating the pedestrian effect. The analysis further evaluated all scenarios at three different vehicle intensities. In the base volume case, the team used actual traffic volumes collected at a one-lane and a two-lane roundabout site during the NCHRP 3-65 research effort (see ‘Model Implementation’ section below). In both cases, the observed volumes were below the theoretical capacity for the respective roundabout size as described in the literature (FHWA, 2000). In order to assess signalization impacts at more congested vehicle operations, the traffic intensities were increased at fixed percentages to get them closer to capacity. Figure 3 shows the approximate volume levels of the one-lane and two-lane roundabout test sites superimposed on the roundabout capacity figure in the FHWA guide. 207

Schroeder, Rouphail, and Hughes Figure 3: Roundabout Entry Volumes Relative to FHWA Theoretical Capacity SOURCE: FHWA (2000), Roundabouts: An Informational Guide In Figure 3, the filled circles indicate the volumes for the four approaches at the one-lane roundabout and the filled rectangles correspond to the two-lane roundabout volumes. These volumes correspond to approximately 1700 vehicles per hour (vph) and 2800 vph, respectively. To investigate signalization impacts at higher volumes, growth rates were applied to each case. The one-lane roundabout volumes were increased by 50% and 100% to get volume scenarios of about 2500 vph and 3400 vph, respectively. For the two-lane roundabout, growth rates of 25% and 35% were used, resulting in 3500 vph and 3800vph. Figure 3 also shows the highest volume cases for the one-lane and two-lane site as hollow circles and rectangles, respectively. Conceptually, the three vehicle volume levels can be described as ‘existing/below capacity’, ‘approaching capacity’ and ‘oversaturated condition’. When modeling the pedestrian signal, it was generally assumed that the crossing at the approach with the highest vehicle volumes would be signalized. TREATMENT MATRIX In developing the treatment test matrix, all three crosswalk configurations were tested in combination with both signalization schemes. In implementing the pedestrian signals, the authors made some assumptions as to whether a particular signal would be more likely to be configured as a one-stage or a two-stage crossing. At a one-stage crossing, the same pedestrian indication is ● – one-lane RAB, 1700 vph ○ – one-lane RAB, 3400 vph ■ – two-lane RAB, 2800 vph □ – two-lane RAB, 3800 vph 208

Schroeder, Rouphail, and Hughes valid for the entire crossing distance and covers both entry and exit lane. Any pedestrian is able to cross the entire roundabout approach and the W and FDW phases are timed accordingly. At a two-stage crossing, it is assumed that the entry and exit lane crossing at a roundabout are timed independently and that it could therefore occur that a pedestrian has to wait on the splitter island. A two-stage crossing generally results in shorter pedestrian phases and therefore less vehicular delay. While this is a reasonable goal from a traffic engineering perspective, its application is only reasonable where the crossing provides adequate and safe pedestrian storage on the splitter island. At a two-stage crossing, a pedestrian is expected to adhere to the signal indication and wait on the splitter island if so directed. The team assumed that this two-stage implementation is only feasible at the zig-zag crossing configuration and at the proximal crossing at of a two-lane roundabout, because it is reasonable to assume a larger splitter island in those cases. All other crossing configurations are coded as one-stage crossings for reasons of safety and for fear of pedestrian non-compliance. In the resulting test matrix, the one-lane roundabout was tested with one-stage proximal and distal, and two-stage zig-zag crossings. The two-lane roundabout was tested with one- and two- stage proximal, two-stage zig-zag, and one-stage distal crossings. For the one-lane roundabout, the team evaluated the 3 crossing configurations and tested each using a conventional pedestrian-actuated (PA) signal and HAWK phasing. Each of the resulting 6 treatment scenarios was modeled for 3 vehicle volume levels and 3 pedestrian intensities, for a total of 54 combinations. The four two-lane roundabout implementations were evaluated accordingly for a total of 72 combinations. Each of the resulting 126 models was replicated 10 times with 10 different random number seeds. MODEL IMPLEMENTATION The team coded two roundabout models, a one-lane and a two-lane, and validated the vehicle operations with data obtained from the NCHRP 3-65 research effort. The two models were coded with observed traffic volumes, turning movements, and lane distributions (for two-lane roundabouts) and with geometric design speeds following the FHWA ‘Roundabout Informational Guide’. The yielding behavior was coded following the ‘priority rule’ concept in VISSIM and was applied consistent with guidance in the software manual. The operational performance of the models was validated by comparing model output travel times and approach queuing to field observations. It was assumed that the pedestrian signal is placed at the busiest approach. Performance Measures The objective of the analysis was to determine the impact of pedestrians crossing at a signalized approach to the roundabout. In order to distinguish this pedestrian effect from the existing vehicle delay, the team evaluated the roundabout in terms of pedestrian-induced impacts as discussed above. For ease of discussion, the team aggregated the delay outputs to the intersection level by defining a data collection node around the entire roundabout. This allowed estimation of average 209

Schroeder, Rouphail, and Hughes pedestrian delay, and average pedestrian-induced vehicle delay for the roundabout system. Results are reported as the mean values and standard errors from 10 replications. The team further extracted data on pedestrian-induced vehicle queues; measured just upstream from the entry and exit crosswalks using the default queue definition in VISSIM. The software provides the average (50th percentile) and the maximum queue observed during the analysis period of one hour. A more in-depth queue analysis including 85th percentile queues and a measure of spillback potential was beyond the scope of this effort. Signalization When coding the varying signalization schemes, the team used an assumed vehicle minimum green time of 45 seconds, measured from the beginning of vehicular green. A pedestrian call at the signal will only be served after this minimum green time has elapsed and in the absence of such call, the signal will default to vehicle green. The team acknowledges that this phasing implementation is possibly overly simplistic, but was used here for ease of discussion and to allow for a clean comparison between alternative treatments. A more detailed analysis of a traffic detection-based signal implementation including minimum green, maximum green and gap extension parameters is left for future research. The team assumed an ‘amber’ phase of 3.0 seconds and an ‘all-red’ clearance time of 1.0 second for vehicles. Pedestrian ‘walk’ phases were assumed at 4.0 seconds for two-stage single-lane crossings and at 7.0 seconds for all others. Pedestrian ‘flashing don’t walk phases’ were timed as a function of the crossing distance and an assumed pedestrian walking speed of 3.5 feet per second. The resulting FDW times were 5.0 seconds for single-lane two-stage crossings, 10.0 seconds for single-lane one-stage crossing, 9.0 seconds for two-lane two-stage crossings, and 19.0 seconds for two-lane one-stage crossings. It was assumed that pedestrians will initiate crossing during the walk phase and the first 1.0 seconds of the FDW interval. In the implementation of the HAWK signal, the flashing red phase for vehicles typically begins after 1.0 seconds of FDW and it is assumed that all vehicles in fact yield to pedestrians in the crosswalk. In the case of a one-stage crossing at a two-lane roundabout, the FR phase is delayed by 10.0 seconds into the FDW phase to assure that any pedestrian who started to cross at the end of W has at least made it past the splitter island. The authors made this assumption, because of a fear of non-compliance of drivers at the far end of the crossing in a real implementation (not yielding to a pedestrian who has yet to reach the splitter island). RESULTS Pedestrian-Induced Vehicle Delay Pedestrian-induced delay is defined as the difference in roundabout system delay at pedestrian volume x, minus the same measure in the zero-pedestrian case. Figures 4a and 4b show the results for the 3 one-lane roundabout scenarios at 10 and 50 pedestrians, respectively. Accordingly, figures 4c and 4d show the four two-lane roundabout scenarios at both signalization schemes and both pedestrian volume levels. The figures further show varying vehicle intensities of 1700, 2500, and 3400 vehicles per hour for the one-lane site and 2800, 210

Schroeder, Rouphail, and Hughes 3500, and 3800 vph for the two-lane roundabout. Note that each sub-figure is presented at a different scale. Figure 4: Pedestrian-Induced Vehicle Delay 4-a) Ped.-Induced Vehicle Delay (sec.) One-Lane RAB, 10 Pedestrians/hour 0 2 4 6 8 10 HAWK ~ 1700 veh/hr PA HAWK ~ 2500 veh/hr PA HAWK ~ 3400 veh/hr PA Pe d- in du ce d de la y (s ec .) Proximal (1-stage) Zig-Zag (2-stage) Distal (2-stage) 4-b) Ped.-Induced Vehicle Delay (sec.) One-Lane RAB, 50 Pedestrians/hour 0 10 20 30 40 50 HAWK ~ 1700 veh/hr PA HAWK ~ 2500 veh/hr PA HAWK ~ 3400 veh/hr PA Pe d- in du ce d de la y (s ec .) Proximal (1-stage) Zig-Zag (2-stage) Distal (1-stage) 4-c) Ped.-Induced Vehicle Delay (sec.) Two-Lane RAB, 10 Pedestrians/hour -5 0 5 10 15 20 25 HAWK ~ 2800 veh/hr PA HAWK ~ 3500 veh/hr PA HAWK ~ 3800 veh/hr PAPe d- in du ce d de la y (s ec .) Proximal (1-stage) Proximal (2-stage) Zig-Zag (2-stage) Distal (2-stage) 4-d) Ped.-Induced Vehicle Delay (sec.) Two-Lane RAB, 50 Pedestrians/hour -20 0 20 40 60 80 100 HAWK ~ 2800 veh/hr PA HAWK ~ 3500 veh/hr PA HAWK ~ 3800 veh/hr PAPe d- in du ce d de la y (s ec .) Proximal (1-stage) Proximal (2-stage) Zig-Zag (2-stage) Distal (1-stage) The one-lane roundabout results suggest that the delay impact of a pedestrian signal on vehicle performance is highest at the 2500 veh/hr scenarios. At lower and higher traffic volumes, the impact is less, due to slow traffic and because of already high vehicle delays at the high-volume case. This suggested non-linear relationship between pedestrian signalization and vehicle volumes is an interesting finding that will be explored more in the future. The proximal crossing location clearly results in the highest vehicle delays across all scenarios. This is explained, because the proximity to the circulating lane results in high queue spillback potential. Also across all pedestrian and vehicle volume levels, the HAWK signal consistently ranks better than the PA signal. Again, the benefits are most predominant at the middle vehicle volume level, but are evident in all cases. When comparing crossing geometry, the zig-zag crossing shows a lot of potential for application at one-lane roundabouts. The combined effects of added queue storage and two-stage phasing resulting in up to 70% delay savings over the proximal alternative. The additional queue storage for the distal crossing does results in some delay savings over the proximal crossing, but does not beat the shorter two-stage phasing in the zig-zag crossing. This suggests that while additional queue storage is important, the impact of shorter vehicle red times is more significant. Both, two- stage crossings and HAWK signalization have that effect. 211

Schroeder, Rouphail, and Hughes The observed trends at the two-lane roundabout are very similar to the one-lane site, with the difference that the distal crossing seems to outperform the zig-zag alternative. Other findings are consistent: two-stage phasing, HAWK implementation and offset exit-leg crosswalk all significantly improve vehicle operations over the one-stage proximal pedestrian-actuated signal. As an interesting caveat, some of the treatment scenarios result in negative pedestrian-induced vehicle delay estimates. In other words, the occurrence of pedestrian actuations actually improves the overall performance of the roundabout. This is explained, because the pedestrian signal acts as a metering signal on the busiest approach and thus facilitates vehicle entry at other (downstream) legs of the roundabout. Vehicle Exit Queues Following the same reporting pattern as for vehicle delay, Figures 5 and 6 show the results for the pedestrian-induced vehicle exit queues for all scenarios. In each case, figures a) and b) show average (50%) queues for 10 and pedestrians, and figures c) and d) show maximum queues for both pedestrian intensities. Each measure represents an average of 10 simulation replications and all values are given in feet (1 foot = 0.305 meter). The authors further attempted to directly compare the observed queues with the available exit lane storage capacity. To recall, the proximal, zig-zag, and distal crossing geometries allow for theoretical queue storage of 20, 60, and 100 feet per lane, respectively. For each figure, the shaded background reflects this capacity for each crossing type. Figure 5: Pedestrian-Induced Exit Queues - One-Lane Roundabout 5-b) Ped.-Induced Average Vehicle Queue - Exit Leg One-Lane Roundabout, 50 Pedestrians/hour 0.0 20.0 40.0 60.0 80.0 100.0 HAWK PA HAWK PA HAWK PA Proximal 1-stage Zig-Zag 2-stage Distal 1-stage A ve ra ge V eh ic le Q ue ue (f ee t) ~1700 veh/hr ~2500 veh/hr ~3400 veh/hr 5-a) Ped.-Induced Average Vehicle Queue - Exit Leg One-Lane Roundabout, 10 Pedestrians/hour 0.0 20.0 40.0 60.0 80.0 100.0 HAWK PA HAWK PA HAWK PA Proximal 1-stage Zig-Zag 2-stage Distal 1-stage A ve ra ge V eh ic le Q ue ue (f ee t) ~1700 veh/hr ~2500 veh/hr ~3400 veh/hr 5-d) Ped.-Induced Maximum Vehicle Queue - Exit Leg One-Lane Roundabout, 50 Pedestrians/hour 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 HAWK PA HAWK PA HAWK PA Proximal 1-stage Zig-Zag 2-stage Distal 1-stage M ax im um V eh ic le Q ue ue (f ee t) ~1700 veh/hr ~2500 veh/hr ~3400 veh/hr 5-c) Ped.-Induced Maximum Vehicle Queue - Exit Leg One-Lane Roundabout, 10 Pedestrians/hour 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 HAWK PA HAWK PA HAWK PA Proximal 1-stage Zig-Zag 2-stage Distal 1-stage M ax im um V eh ic le Q ue ue (f ee t) ~1700 veh/hr ~2500 veh/hr ~3400 veh/hr Results in Figure 5 suggest that the average queues at the one-lane roundabout were mostly well- inside the available queue storage for all scenarios. At 50 pedestrians per hour, the average queue 212

Schroeder, Rouphail, and Hughes at the proximal PA scenario consistently spills over the available storage at higher vehicle demands. Both the zig-zag and distal crossing provide ample storage for the average queue. Looking at the maximum queues, it becomes evident that all scenarios will experience queue spillback into the circulating lane at least once during a one-hour analysis period. The mean maximum queue from 10 replications even extends beyond the 100-foot storage of the distal crossing as pedestrian and vehicle volumes increase. Nonetheless, the short two-stage phasing of the zig-zag crossing consistently results in the lowest queues and it is further evident that the creation of additional vehicle storage at the exit leg is a valuable approach. The HAWK signal queue is less than the PA queue in all cases. Figure 6: Pedestrian-Induced Exit Queues - Two-Lane Roundabout 6-b) Ped.-Induced Average Vehicle Queue - Exit Leg Two-Lane Roundabout, 50 Pedestrians/hour 0.0 50.0 100.0 150.0 200.0 250.0 300.0 HAWK PA HAWK PA HAWK PA HAWK PA Proximal 1-stage Proximal 2-stage Zig-Zag 2-stage Distal 1-stage A ve ra ge V eh ic le Q ue ue (f ee t) ~2500 veh/hr ~3500 veh/hr ~3800 veh/hr 6-a) Ped.-Induced Average Vehicle Queue - Exit Leg Two-Lane Roundabout, 10 Pedestrians/hour 0.0 50.0 100.0 150.0 200.0 250.0 300.0 HAWK PA HAWK PA HAWK PA HAWK PA Proximal 1-stage Proximal 2-stage Zig-Zag 2-stage Distal 1-stage A ve ra ge V eh ic le Q ue ue (f ee t) ~2800 veh/hr ~3500 veh/hr ~3800 veh/hr 6-d) Ped.-Induced Maximum Vehicle Queue - Exit Leg Two-Lane Roundabout, 50 Pedestrians/hour 0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 HAWK PA HAWK PA HAWK PA HAWK PA Proximal 1-stage Proximal 2-stage Zig-Zag 2-stage Distal 1-stage M ax im um V eh ic le Q ue ue (f ee t) ~2500 veh/hr ~3500 veh/hr ~3800 veh/hr 6-c) Ped.-Induced Maximum Vehicle Queue - Exit Leg Two-Lane Roundabout, 10 Pedestrians/hour 0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 HAWK PA HAWK PA HAWK PA HAWK PA Proximal 1-stage Proximal 2-stage Zig-Zag 2-stage Distal 1-stage M ax im um V eh ic le Q ue ue (f ee t) ~2500 veh/hr ~3500 veh/hr ~3800 veh/hr 38 0. 1 93 6. 9 43 6. 3 44 3. 5 1548.0 The results for the two-lane roundabout in Figure 6 generally show much higher pedestrian- induced queues compared to the one-lane site analysis. Judging from the average queue lengths, the benefits of an offset exit crosswalk (zig-zag or distal) are immense. Furthermore, the benefits of the HAWK signal are even more significant than in the one-lane case. At a pedestrian intensity of 50 peds/hour, the average queues at higher vehicle volumes approach 1000’ in the proximal one-stage PA scenario, and are around 400’ for several other proximal scenarios. With added queue storage, the average queues can generally be contained to the exit lane. It is important to point out that the theoretical queue storage is shown for both lanes combined and is thus double to that shown at the one-lane roundabout. The maximum queues paint a similar picture. Roundabout exit queues at the signal can be reduced drastically by creating additional queue storage, by implementing a two-stage crossing and by using the HAWK signalization scheme. 213

Schroeder, Rouphail, and Hughes Pedestrian Delay The pedestrian delay measure is defined as the difference between actual travel time and theoretical travel time (at a randomly distributed walking speed around a mean of 3.5 ft/sec.) through the roundabout node. Given that the signal timing was implemented without vehicle green extension parameters (gap time and max. green) it is expected that this delay is constant for all three volume levels. Similarly, pedestrian delay is the same for the conventional pedestrian-actuated signal and the HAWK scheme. At the one-lane roundabout, the resulting pedestrian delay numbers for the proximal, zig-zag, and distal crossing were 12.3, 21.7, and 11.8 seconds per pedestrians for the 10 peds/hour intensity level and 19.5, 35.2, and 19.5 seconds for the 50 peds/hour level. For the two-lane roundabout, the delay for the proximal 1-stage and 2-stage crossings came out to be 11.2 and 18.2 seconds for 10 peds/hour; and 20.7 and 31.7 seconds for 50 peds/hour. The two-lane roundabout zig-zag and distal crossing pedestrian delay numbers were 17.3 and 18.7 seconds for 10 peds/hour and 30.8 and 31.3 seconds for 50 peds/hour. The one-lane roundabout numbers indicate that the delay at the zig-zag crossing is higher than for the other two, because some pedestrians will likely have to wait at both signals of the two- stage crossing. A comparison of one-stage and two-stage crossings at the two-lane roundabout shows the same results. Furthermore, the delay for 50 peds/hour is consistently higher than for 10 peds/hour, because pedestrians are more likely to arrive during the minimum green constraint. It is expected that this trend will level of as the number of pedestrians increases further (see volume sensitivity section). By coordinating the two phases of a two-stage crossing, it may be possible to partially overcome the apparent disadvantage of the configuration compared to a single-stage crossing. It is important to keep in mind that the different crossing geometries are expected to vary in pedestrian travel time through the crossing due to varying path deflections. These pedestrian travel time numbers are a function of the origin-destination characteristics at a particular site. In that sense, a zig-zag or distal crossing may or may not result in higher travel times depending on the particular pedestrian route. Pedestrian Volume Sensitivity In the final analysis step, the range of pedestrian intensities was varied between zero and 300 pedestrians per hour to perform a sensitivity analysis on the pedestrian-induced effects. It can be reasoned that delay as a function of pedestrian intensity will eventually flatten as the intensity approaches the maximum number of actuations per hour. In a fixed-cycle signal system, the maximum number of pedestrian actuations is given by 3600 seconds divided by the particular cycle length. Figures 7 a) and b) show curves of pedestrian delay, vehicle delay and pedestrian-induced vehicle delay as a function of traffic volume for the one-lane and two-lane roundabout sites. Each figure shows the corresponding curves for the proximal, zig-zag, and distal crossing configuration. The graphs are shown for vehicle volume levels of 2500 and 2800 vehicles per hour for the one-lane and two-lane site, respectively. All numbers shown are for the HAWK signalization scheme, but similar trends are expected for a conventional pedestrian-actuated 214

Schroeder, Rouphail, and Hughes signal, as well as, for other vehicle volumes. Each data point is the average of 10 simulation replications. Figure 7: Pedestrian Volume Sensitivity Analysis 0 10 20 30 40 50 60 0 50 100 150 200 250 300 A ve ra ge D el ay (s ec .) Zig-Zag Crossing Distal Crossing 0 10 20 30 40 50 60 0 50 100 150 200 250 300 Pedestrian Volume (peds/hour) A ve ra ge D el ay (s ec .) Average Vehicle Delay (sec.) Average Pedestrian Delay (sec.) Ped-Induced Vehicle Delay (sec.) 7-a) Effect of Pedestrian Volume on Ped-Induced Delay ONE-LANE ROUNDABOUT ~2500 veh/hr, Error Bars at one standard error Proximal Crossing 0 10 20 30 40 50 60 0 50 100 150 200 250 300 A ve ra ge D el ay (s ec .) Zig-Zag Crossing 0 10 20 30 40 50 60 0 50 100 150 200 250 300 A ve ra ge D el ay (s ec .) Distal Crossing 0 10 20 30 40 50 60 0 50 100 150 200 250 300 Pedestrian Volume (peds/hour) A ve ra ge D el ay (s ec .) Average Vehicle Delay (sec.) Average Pedestrian Delay (sec.) Ped-Induced Vehicle Delay (sec.) 7-b) Effect of Pedestrian Volume on Ped-Induced Delay TWO-LANE ROUNDABOUT ~2800 veh/hr, Error Bars at one standard error Proximal Crossing 0 10 20 30 40 50 60 0 50 100 150 200 250 300 A ve ra ge D el ay (s ec .) The figure shows that as the pedestrian intensity increases, any additional pedestrians will arrive during an existing call for green and won’t further impact vehicle operations. Assuming compliance, a pedestrian signal therefore places a limit on pedestrian-induced vehicle delay even at high volumes, whereas an unsignalized crossing would result in uncontrolled and presumably dangerous situations. To allow for a comparison between crossing geometries, Figures 8 a) and b) aggregate the results for pedestrian-induced vehicle delay for the proximal, zig-zag and distal crossing for the one- lane and two-lane roundabout. 215

Schroeder, Rouphail, and Hughes Figure 8: Pedestrian-Induced Vehicle Delay as a Function of Pedestrian Intensity 8-b) Effect of Pedestrian Volume on Ped.-Induced Veh. Delay, Two-Lane Roundabout ~2800 veh/hr, Error Bars at one standard error 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 10 30 50 100 150 200 300 Pedestrian Volume (peds/hour) A ve ra ge D el ay (s ec .) Proximal Crossing Zig-Zag Crossing Distal Crossing 8-a) Effect of Pedestrian Volume on Ped.-Induced Veh. Delay, One-Lane Roundabout ~2500 veh/hr, Error Bars at one standard error 0 5 10 15 20 25 0 10 30 50 100 150 200 300 Pedestrian Volume (peds/hour) A ve ra ge D el ay (s ec .) Proximal Crossing Zig-Zag Crossing Distal Crossing It is evident from Figure 8-a), that zig-zag and distal crossing configurations provide a clear benefit over the proximal location at the one-lane roundabout. For the two-lane roundabout, the distal crossing again appears to outweigh the zig-zag configurations, although large standard errors of the estimate require additional simulation replications to make this claim statistically significant. CONCLUSION The analysis presented in this paper has provided a quantitative comparison between different options for signalized pedestrian crossings at one-lane and two-lane roundabouts. The results indicate that innovative signalization treatments, including HAWK signals and two-stage 216

Schroeder, Rouphail, and Hughes crossings can significantly decrease vehicle delay. Modified crossing geometries such as a zig- zag or distal crosswalk, can further reduce spillback potential into the circulating lane due to added vehicle storage at the roundabout exit lane. The analysis further suggested a non-linear relationship between the treatments and the levels of vehicle volumes as pedestrian-induced vehicle delays appeared to be greatest as traffic volumes approach roundabout capacity, but not as conditions became oversaturated. The need for innovation in pedestrian signal application is therefore less pronounced at low or very high traffic volumes, but should be a key consideration at busy roundabout junctions. An sensitivity analysis of increasing pedestrian volumes supported the hypothesis that an increase in pedestrian intensities eventually doesn’t add any further delay, as the signal operations approach the limit of ‘maximum number of actuations per hour’. Pedestrian and vehicular delays generally appear to plateau in excess of 200 pedestrians per hour. This suggests an application for signalization as a means of controlling ‘pedestrian interference’ to vehicular operations – an interesting twist to the existing pedestrian signal warrant that evaluates only the available crossing opportunities for pedestrians within a given time interval. LIMITATIONS Due to the nature of existing microsimulation software, some additional consideration ought to be given to the results. For example, the high vehicle volume scenarios are likely underestimating capacity, because they assume unchanged driver gap acceptance behavior compared to the calibrated base case. In reality, it is expected that drivers waiting to enter the roundabout will lower their critical gap and follow-up times as traffic gets heavier, thereby increasing capacity. The current use of ‘priority rules’ in VISSIM does not allow for a decaying critical gap function as a function of waiting time. Also, as discussed above, the queue definitions of maximum and average queues are not as interesting from a traffic engineering perspective as an 85th queue or a measure of ‘percent spillback' would be. In the current configuration of VISSIM, it is possible to extract queue data for shorter time intervals and thus perform a ‘queue study’ for one-minute intervals, but this analysis was beyond the scope of the effort presented here. For a reference on this alternate approach, please refer to Rouphail et al. (2005). The lane distribution at the entry lane of the two-lane roundabout was obtained from NCHRP 3- 65 data and is valid for the base case. The data showed a significant skew towards the right approach lane, suggesting underutilization of the left lane that is probably evident at most two- lane roundabouts in the US. However, as volumes get closer to capacity it is expected that a more even lane distribution is obtained as drivers tend to shift to the lane with the shorter queue. The resulting entry capacity in the simulation models therefore is likely to be somewhat low. Finally, it is debatable whether the measure of ‘pedestrian delay’ is indeed the most appropriate. Assuming that all pedestrian origins and destinations (O/D) are at the roundabout, the additional travel times for zig-zag and distal locations are significant. This assumption clearly is not valid, 217

Schroeder, Rouphail, and Hughes because pedestrian travel paths clearly are dependent on O/D patterns outside the roundabout influence area. Nonetheless, there is always the potential of additional travel time and the related concern of pedestrian compliance to the suggested geometries and signalization schemes. The authors believe that through the use of landscaping features (trees, bushes, walls, fences) and through proper design of pedestrian paths – paths that lead to the crosswalk, not to the intersection – compliance can be maximized. FUTURE RESEARCH For future research, it would be highly interesting to compare the results of pedestrian signalization to capacity reductions from unsignalized pedestrian crossings. This would require a more detailed analysis of pedestrian gap acceptance and driver yielding behavior, as is discussed in a second paper the authors submitted to TRB. In some circumstances, signalization may actually slightly reduce overall vehicle delay and contribute to pedestrian safety. Also, it would be interesting to compare the results for signalized and unsignalized pedestrian crossings obtained in microsimulation to estimates obtained from deterministic equations in the HCM2000 or roundabout analysis software such as aaSIDRA or KREISEL. It would be in the general interest of the traffic engineering community to generalize the results obtained here and to develop improved equations describing pedestrian-vehicle interaction that can be applied independent of microsimulation. ACKNOWLEDGMENTS The research leading up to this document was supported by the NCHRP 3-78 project, ‘Crossing Solutions at Roundabouts and Channelized Turn Lanes for Pedestrians with Visual Disabilities’. The authors would like to thank the National Academies for the opportunity to be involved in the project and for permission to share these results with the Transportation Research Board Community. The authors would also like to thank the members of the project panel, who have provided continuous feedback to the research efforts. Finally, the authors would like to thank the other members of the project team, who have been invaluable in discussing the application of microsimulation models to the project. 218

Schroeder, Rouphail, and Hughes REFERENCES Ashmead, D., Guth, D., Wall, R., Long, R., & Ponchillia, P. Street crossing by Sighted and blind pedestrians at a modern roundabout. In ASCE Journal of Transportation Engineering, Vol. 131, No. 11, November 1, 2005, 812-821 Rouphail, N., Hughes, R., & Chae, K., Exploratory Simulation of Pedestrian Crossings at Roundabouts. In ASCE Journal of Transportation Engineering. Vol. 131, No. 3, March 2005 FHWA, Roundabouts: An Informational Guide. Federal Highway Administration. Turner Fairbank Highway Research Center. FHWA-RD-00-067. McLean, VA. 2000 Guth, D., Ashmead, D., Long, R., Wall, R., & Ponchillia, P. Blind and sighted pedestrians’ judgments of gaps in traffic at roundabouts. In Human Factors, 47, 314- 331. National Cooperative Highway Research Program (NCHRP) Project 3-78. Crossing Solutions at Roundabouts and Channelized Turn Lanes for Pedestrians with Vision Disabilities. Institute of Transportation Research and Education. Raleigh, North Carolina. http://www4.nationalacademies.org Last visited July 26, 2006 PTV (2005), VISSIM 4.10 User Manual. Karlsruhe, Germany. March 2005. Tucson DOT, Pedestrian Traffic Signal Operations. City of Tucson Department of Transportation, http://dot.ci.tucson.az.us/traffic/tspedestrian.cfm. Accessed July 31, 2006 US Access Board, Revised Draft Guidelines for Accessible Public Rights-of-Way. http://www.access-board.gov/prowac/draft.htm. Accessed July 29, 2006 Wall, R., Long, R., Guth, D., Ashmead, D., & Ponchillia, P. Roundabouts: Problems of and strategies for access. International Congress Series 1282, In Proceedings of Vision 2005, April 2005, London 219