A Methodology to Compute Roundabout Corridor Travel Time

Transcription

1 A Methodology to Compute Roundabout Corridor Travel Time By Zachary Bugg, Ph.D* Kittelson & Associates, Inc.** Tel: (443) zbugg@kittelson.com Bastian Schroeder, Ph.D, PE Institute for Transportation Research and Education North Carolina State University Campus Box 8601 Raleigh, NC Tel: (919) bastian_schroeder@ncsu.edu Pete Jenior, PE, PTOE Kittelson & Associates, Inc. Tel: (410) pjenior@kittelson.com Marcus Brewer, PE Texas A&M Transportation Institute 2935 Research Parkway College Station, TX Tel: (979) m-brewer@tamu.edu Lee Rodegerdts, PE Kittelson & Associates, Inc. Tel: (503) lrodegerdts@kittelson.com Submitted for publication and presentation at the 94 th Annual Meeting of the Transportation Research Board, January 11-15, 2015 Word Count: 5,497 text words plus 2,000 for figures/tables (8 x 250) = 7,497 total *Corresponding Author ** Kittelson & Associates, Inc. 36 S Charles St, Suite 1920 Baltimore, MD 21201

2 Bugg, Schroeder, Jenior, Brewer, Rodegerdts ABSTRACT Urban and suburban arterials with roundabouts in series are becoming more prevalent in North America. While the Highway Capacity Manual (HCM) currently provides a methodology for computing the segment travel time of urban streets with various forms of intersection control, it does not provide a similar procedure for corridors with interdependent roundabouts. This methodology is necessary to evaluate roundabout corridor performance, and it allows the practitioner to compare both roundabout and signalized treatments for the same series of intersections. This paper presents a series of models intended to predict arterial travel time for a corridor with roundabouts, including a free-flow speed (FFS) model, a model to predict the length of the roundabout influence area (or RIA, the area where geometric delay is incurred), a model for geometric delay, and models for impeded delay and average travel speed. These models were calibrated with data from seven existing roundabout corridors. The resulting models suggest that while FFS is a function of segment length, posted speed limit, and central island diameter, the RIA length and geometric delay are functions of the FFS itself, as well as other geometric elements. The impeded delay and average travel speed are functions of traffic congestion and FFS. After validating the models with travel time data from two additional roundabout corridors not used in model development, the authors present a framework for incorporating the travel time prediction procedure into the HCM urban streets analysis

3 Bugg, Schroeder, Jenior, Brewer, Rodegerdts INTRODUCTION Traffic on urban and suburban arterials is typically signalized at major intersections, but many transportation agencies are increasingly looking to roundabouts as an alternative to signalized or two-way stop-controlled intersections as a safety improvement and as a means to reduce delay during off-peak periods. This trend has extended to arterials with roundabouts in series, which are becoming more prevalent across the United States. Roundabout Corridor Description As defined in this paper, a roundabout corridor includes a series of three or more roundabouts that function interdependently on an arterial. An example is displayed in Figure FIGURE 1 Roundabout Corridor in Golden, Colorado. (1) Arterials such as the one displayed in Figure 1 would normally be analyzed using the urban streets facility and segment procedures explained in Chapters 16 and 17 of the Highway Capacity Manual (2). However, the HCM currently only allows for the analysis of independent roundabouts (Chapter 21) and does not account for their effect on corridor travel time, platoon dispersion, access management, and other elements. The HCM 2010 also does not incorporate the effects of closely-spaced roundabouts on arterial operational performance. Research Objective The objective of this research is to understand the effect of a series of roundabouts on arterial performance, specifically considering the performance measures of corridor travel time and percent free flow speed. The authors intend to develop a model to predict roundabout corridor performance based on geometric and operational elements and construct a methodology for operational performance evaluation analogous to the urban streets methodology presented in the HCM. The research presented here is also contained in NCHRP Report 772: Evaluating the

4 Bugg, Schroeder, Jenior, Brewer, Rodegerdts Performance of Corridors with Roundabouts (3), which is the result of NCHRP Project While this paper focuses on the model development and implementation of the methodology into the HCM, the reader will be referred to NCHRP Report 772 for details on data collection, model calibration, and model sensitivity analysis. This paper is organized as follows: literature review, methodology, results, implementation, conclusion, and references. Literature Review The transportation profession has an extensive body of knowledge for traffic operations at isolated roundabouts and urban corridors with signalized intersections. Little research has been published on traffic operations of roundabout corridors, but it can be principally divided into safety research and operational research. Emissions, non-motorized transportation modes, constructability, and corridor context are additional considerations that could provide differentiating characteristics between corridor types. Roundabouts have well-documented safety benefits compared to other traffic control types, and these safety benefits are the predominant attractiveness of roundabouts compared to signalized intersections (4, 5, 6, 7). Safety relationships of isolated roundabouts are likely to transfer to roundabout corridors; there are no known specific characteristics of roundabouts in series that diminish the safety performance of the roundabout itself. One key differentiating consideration between corridor types may be safety at midblock access points. Opportunities to use U-turns at roundabouts could potentially eliminate left turns to or from driveways along the corridor, and reduced conflicts at these driveways would positively influence corridor safety. Depending on the spacing of the roundabouts, segment operating speeds could be reduced compared to signalized corridors and, therefore, could reduce crash severity. A crash study of a half-mile roundabout corridor consisting of four roundabouts in Golden, Colorado (Figure 1) indicated a steep decline in crash rates after roundabout installation (8). Between 1996 and 2004, crash rates declined by 88%, from 5.9 to 0.4 crashes per million vehicle miles. Injury crashes were reduced from 31 in the three years prior to installation to one in the 4.5 years after a 93% decline. Isebrands et al. (9) reviewed a corridor in Brown County, Wisconsin, and found that total crashes at one of the Wisconsin roundabouts were reduced by one per year, and injury crashes were nearly eliminated. Roundabout corridors have unique operational characteristics compared to their signalized intersection counterparts. Fundamentally, the notion of moving platoons of vehicles to maximize the performance of signalized intersections is not applicable to roundabouts, where gap acceptance principles allow more dispersed flows to mingle within the intersections. Travel time is a natural performance measure for roundabout and signalized corridors. Roundabouts have greater geometric delay compared to signalized intersections by virtue of their shape; therefore, defining travel time performance measures is of paramount interest. According to Akçelik (10), geometric delay is determined as a function of approach and exit cruise speeds as well as negotiation speeds, which depend on the geometry of the roundabout. Akçelik added that steps could be taken to approximate the value of geometric delay and add it to the control delay computed by the HCM procedure. The Golden, Colorado study also reviewed operating speeds, travel times, and sales tax revenue along the corridor (8). The study concluded that installing the roundabouts resulted in lower speeds between major intersections in the corridor, but travel times also decreased compared to when the corridor was signalized, indicating a decrease in control delay. The

5 Bugg, Schroeder, Jenior, Brewer, Rodegerdts analysis also revealed much less delay at business access points. While adjacent businesses are often concerned about corridor reconfiguration and access management adjustments, the development around the corridor flourished after installation of the roundabouts. This consisted of an additional 75,000 square feet of retail/office space, resulting in a 60% increase in sales tax revenue along the corridor after installation. METHODOLOGY The following sections describe the methodology used to develop the models for roundabout corridor travel time, including site selection, data collection and extraction, and model development (including formulation of the models and model calibration). Data Collection and Extraction At least three methods were available to develop a model framework for roundabout corridor travel time: Analytically, through the use of the HCM models. This would make use of the existing methodology but is limited by the applicability of the current HCM methodology, given that it does not support the interaction of closely-spaced roundabouts. Using data from microsimulation. This technique would supply a heavy amount of data but would also require extensive calibration with field data so that the output from microsimulation yields realistic results. Manually-collected field data. Given the presence of dozens of roundabout corridors in the United States identified as part of the initial efforts of NCHRP Project , the authors concluded that developing models from empirical data is the best approach. GPS devices were used to collect travel time data at these sites, and the reader is referred to NCHRP Report 772 for more details on the data collection procedure (3). To maximize model applicability and range of the dataset, the authors selected sites for data collection representing a variety of characteristics and conditions. Using these considerations, nine roundabout corridors from across the United States were identified. Two of these sites, both in Carmel, Indiana, were reserved for model validation. Table 1 lists each site where data was collected, including the route name, geographic location, number of roundabouts along the corridor, number of lanes on the arterial, number of circulating lanes in the roundabouts, length of the corridor, type of area (commercial, residential, or rural), whether the corridor includes a roundabout (also called raindrop) interchange, year(s) the corridor was constructed, type/presence of restrictive median, and peak hour traffic volume on the arterial.

6 Bugg, Schroeder, Jenior, Brewer, Rodegerdts TABLE 1 Summary of Study Sites Corridor Location # of Roundabouts Arterial # of Lanes Roundabout # of Lanes Length (mi) Area Type Includes Interchange? Built Median Peak Hour Traffic Volume on Arterial (vph) MD-216 Scaggsville, MD Commercial Yes 2002, 2009 Raised 1,600-2,100 La Jolla Blvd San Diego, CA Commercial No Raised 1,000-1,500 Borgen Blvd Gig Harbor, WA Commercial/ Residential Yes Varies 1,000-2,000 SR-539 Whatcom County, WA Rural No 2010 Cable 800 Golden Rd Golden, CO Commercial No , 2004 Avon Rd Avon, CO Commercial Yes 1997 SR-67 Malta, NY Commercial/ Residential Raised with openings Raised with one opening 1,000-1,400 1,300-1,800 Yes Raised 600-1,200 Spring Mill Rd* Carmel, IN Residential No Some 1,100-1,600 Old Meridian St* Carmel, IN * indicates site reserved for model validation Commercial/ Residential No 2006 Mostly raised 500-1,200

7 Bugg, Schroeder, Jenior, Brewer, Rodegerdts During data collection, the authors identified two types of data that were crucial to the development of corridor travel time models. These elements are based on characteristics used in the HCM urban streets analysis procedure. The types of data are as follows: 1. Site-specific data Roundabout geometry (Inscribed Circle Diameter, Central Island Diameter, entry and exit radii) Roundabout spacing Presence of raised median Driveway density Grade Posted speed limit Circulating speed 2. Time-of-day-dependent data Entering volume Circulating volume Volume-to-capacity ratio To be consistent with the HCM urban streets procedure but also to organize the modeling effort to best consider the effects on travel speed of elements within the corridor, the authors proposed a new methodology to partition each corridor into a series of roundabout segments and sub-segments. Essentially, the method distinguishes whether a vehicle is upstream or downstream of a roundabout yield line. Figure 2 illustrates this approach; two roundabouts (RBT1 and RBT2) are separated by (Urban Street) Segment B. For the purpose of analysis, Segment B is divided into sub-segments B1 and B2, where B1 corresponds to the downstream influence of RBT1, and B2 corresponds to the upstream influence of RBT2. The upstream subsegment of RBT1 is consequently labeled A1 (portion of Segment A associated with RBT1), and the downstream sub-segment of RBT labeled C2 (portion of Segment C associated with RBT2).

8 Bugg, Schroeder, Jenior, Brewer, Rodegerdts FIGURE 2 Segment Definitions for Roundabout Corridor Travel Time Modeling. The HCM Chapter 17 defines an urban street segment as the sum of the downstream and upstream sub-segments of two adjacent roundabouts (e.g. B1 plus B2). Similarly, the HCM Chapter 21 defines a roundabout segment as the sum of the upstream and downstream segment of the same roundabout (e.g. A1 plus B1). In developing the models, the upstream and downstream sub-segments are evaluated separately for two primary reasons: (1) this allows aggregation of model results from both HCM Chapter 17 and 21, and (2) the operational effects of the two are hypothesized to be different. To illustrate the latter point, the shaded areas in Figure 2 represent the theoretical areas of geometric delay for upstream and downstream sub-segments. Geometric delay in this case is defined as the difference between segment free-flow speed (FFS) and the unimpeded trajectory speed across the sub-segment length. A trajectory that does not encounter any interaction with other vehicles (e.g. by yielding at the roundabout or braking between intersections) is considered unimpeded. The exhibit makes evident that geometric delay for the upstream roundabout subsegment (gray lines) is arguably much less than for the downstream sub-segment (black crosshatch marks), as the latter includes significant travel distance at the geometricallyconstrained circulating speed. Consistent with these sub-segment definitions, all variables are defined on a sub-segment basis. This includes variables such as the FFS, with each roundabout having a potentially different upstream (A1) and downstream (B1) FFS. Coincidentally, the downstream FFS of RBT1 (B1) is the same as the upstream FFS of RBT2 (B2). 239

9 Bugg, Schroeder, Jenior, Brewer, Rodegerdts Data Exploration The GPS data obtained from the data collection effort allowed the authors to plot the speed profile for each series of travel time trajectories. The speed profiles for each corridor typically resembled the graph displayed in Figure 3. This speed profile, taken from the MD-216 roundabout corridor, displays speed on the vertical axis and distance on the horizontal axis. Each line plotted on the graph represents a different travel time trajectory FIGURE 3 Speed Profile Example. Note that under unimpeded conditions, each driver travels along the corridor at his or her chosen FFS and then slows to a comfortable speed to enter the roundabout. After exiting the roundabout, the driver then accelerates back to FFS. This led to two major observations: 1. The extent of the area in which geometric delay is incurred, dubbed the roundabout influence area (RIA) by the authors, varies by roundabout (Figure 3). 2. The maximum speed which drivers reach and maintain between roundabouts, dubbed the mid-segment running speed, varies by segment and can be as high as the free-flow speed for the entire corridor. Model Development Using the site-specific and time-of-day-dependent elements identified in the previous section, the authors developed models for FFS, RIA length, geometric delay, impeded delay, and average travel speed. The first three models are considered site-level models because they are not sensitive to time-of-day variability in traffic patterns (e.g. volume). In other words, the RIA,

10 Bugg, Schroeder, Jenior, Brewer, Rodegerdts geometric delay, and FFS models are considered to be fixed over time for a specific roundabout approach. On the other hand, the impeded delay and average travel speed models are operational-level models that consider traffic volumes and volume-to-capacity ratios. Conceptually, these latter two models may use one or more of the site-level models as inputs, where, for example, the average travel speed may be a function of the free-flow speed and the RIA estimated in earlier models. In the HCM 2010 Chapter 17, the average travel speed is estimated as a function of the segment running speed and the various sources of delay. The motivation for the formulation of a separate average travel speed model here is based on a desire to verify the applicability of the Chapter 17 method to roundabout corridors. The authors assembled a dataset by calculating geometric and operational parameters for each roundabout approach from seven of the nine roundabout corridors listed in Table 1, leaving the two Carmel corridors out for validation purposes. Each approach was partitioned into subsegments, as shown in Figure 2. The total dataset included 62 roundabout approaches, with each providing an upstream and a downstream sub-segment. Some approaches were excluded due to (a) overlapping influence areas, or (b) a short upstream or downstream sub-segment length at the end of the roundabout corridor. In the case of overlapping influence areas, these approaches were only excluded from the RIA models; these approaches were used in other models like the FFS prediction model. For the short sub-segments, the GPS travel-time vehicles were not able to accelerate to their desired speeds (even in free-flow conditions), because of a nearby intersection or turnaround point. After excluding selected approaches, the remaining dataset included 49 upstream and 52 downstream roundabout sub-segments. The models were developed using the SAS general liner model with a forward selection procedure, and separate models were developed for upstream and downstream segment data. Several individual models were developed for each dependent variable, and then the authors selected a preferred model for each based on goodness-of-fit, simplicity, statistical significance of the independent variables within the model, and general sense of the model within the context of an urban streets analysis. RESULTS This section explains the results of the model calibration and validation efforts. The detailed results for each model (RIA, geometric delay, FFS, average travel speed, and impeded delay) are presented separately, and then the validation results are presented. Roundabout Influence Area After comparing several models obtained from the regression effort, the authors recommend the following sub-models for roundabout influence area length: RIA upstream = * S f 21.1 * S c (1) RIA downstream = * S f 22.5* S c (2) where RIA upstream = upstream roundabout influence area length (ft), RIA downstream = downstream roundabout influence area length (ft), S f = free-flow speed (mph), and S c = circulating speed (mph).

11 Bugg, Schroeder, Jenior, Brewer, Rodegerdts The upstream sub-model has an R 2 of 28.9%, while the downstream sub-model has an R 2 of 71.4%. All terms in the model were statistically significant (p<0.05). The model shows that the RIA length is primarily related to the FFS along the sub-segment as well as the circulating speed (i.e. the minimum unimpeded speed) within the roundabout. For a given sub-segment, as the FFS increases, the RIA length predicted by the model also increases, but this length decreases as the circulating flow in the roundabout adjacent to the segment increases. This is intuitive because it suggests that roundabouts necessitating a greater decrease in speed (as indicated by a large difference between the FFS and circulating speed) should be associated with a longer RIA. Geometric Delay The authors recommended the following sub-models for geometric delay after comparing several models obtained from the regression effort: Delay geom, upstream = *S f 0.21*S c (3) Delay geom, downstream = min{ *S f * ICD *(1/S c -1/S f ),0} (4) where Delay geom, upstream = upstream geometric delay (s/veh), Delay geom, downstream = downstream average geometric delay (s/veh), ICD = inscribed circle diameter (ft), S f = free-flow speed (mph), and S c = circulating speed (mph). The upstream sub-model has an R 2 of 31.5%, while the downstream sub-model has an R 2 of 79.4%. All terms in the model were statistically significant (p<0.05). An interaction term between the FFS, circulating speed, and inscribed circle diameter was also used to model geometric delay with positive results for the downstream geometric delay sub-model. This circulating delay term, shown in Figure 4, is based on the difference between FFS and circulating speed. Intuitively, this term applies to the downstream geometric delay, as it includes the geometric delay within the circle.

12 Bugg, Schroeder, Jenior, Brewer, Rodegerdts FIGURE 4 Illustration of Circulating Delay Term. The geometric delay could be partially explained by the FFS along the segment and the circulating speed within the roundabout. These relationships were similar to those in the RIA model in that a higher FFS along a sub-segment would cause the model to predict a higher geometric delay, but a higher circulating speed within the adjacent roundabout would lower the predicted geometric delay. This is also intuitive, as drivers should experience more geometric delay at a roundabout necessitating a greater decrease in speed. Note that is also necessary to add a boundary condition (minimum value of zero geometric delay) to Equation 4, since low FFS (under 30 mph) on downstream sub-segments will typically result in very low geometric delays. Free Flow Speed The following sub-models were developed for upstream and downstream FFS: S f,upstream = *L *S *CID 4.73*OL (5) S f,downstream = *L *S *CID 4.43*OL (6) where S f,upstream = upstream free-flow speed (mph), S f,downstream = downstream free-flow speed (mph), L = sub-segment length (ft), S = posted speed limit (mph), CID = central island diameter (ft), and OL = dummy variable for overlapping roundabout influence areas, i.e. OL = 1 if the roundabout influence area of the sub-segment is longer than the sub-segment length, 0 otherwise.

13 Bugg, Schroeder, Jenior, Brewer, Rodegerdts The FFS model possesses the strongest fit of any models developed, with an R 2 of 90% for the upstream sub-model and 93% for the downstream sub-model. All terms in the model were significant (p<0.05). FFS was intuitively found to related to the segment length, the posted speed limit along the segment, and the central island diameter of the adjacent roundabout, as well as whether the segment was an overlapping segment. Specifically, the FFS was found to increase with the posted speed, segment length, and central island diameter, but the model assigned a significant penalty (minus 4.5 to 5.0 mph) to the FFS of sub-segments within segments with overlapping RIAs. Average Travel Speed The authors investigated the average speed along the sub-segments of the roundabout corridor, considering traffic characteristics such as the level of congestion. The average travel speed along each sub-segment was calculated for the upstream and downstream segments to calibrate several models. This model is an optional step in the implementation of this procedure in the HCM 2010 Chapter 17, as the existing procedure already contains a step to estimate the average travel speed, considering FFS, delays, and other factors. The following are the sub-models for upstream and downstream average travel speed: S average,upstream = *S f 18.2*v/c (7) S average,downstream = *S f 5.4*v/c (8) where S average,upstream = average travel speed for the upstream sub-segment (mph), S average,downstream = average travel speed for the downstream sub-segment (mph), S f = free-flow speed (mph), and v/c = volume-to-capacity ratio for the downstream roundabout, i.e. the roundabout adjacent to an upstream sub-segment or downstream of (not adjacent to) a downstream sub-segment. The upstream sub-model has an R 2 of 76%, while the downstream sub-model has an R 2 of 83%. All terms in the model were statistically significant (p<0.01). The model indicates that average travel speed is chiefly related to the FFS and volume-tocapacity ratio of the downstream roundabout. Thus, a sub-segment with a higher FFS would experience a higher average speed, but an increase in volume-to-capacity ratio would lower the average speed within the sub-segment. In the application of this model, the FFS would be calculated from the FFS model. The volume-to-capacity ratio is calculated using entering volume and the theoretical roundabout capacity calculated from the equations in the HCM. Impeded Delay Finally, the authors investigated the average impeded delay incurred to each driver at each roundabout by calibrating a delay model for the time-of-day dataset. The impeded delay is defined as the delay experienced due to interaction with other vehicles, yielding at entry, and queuing effects, but it does not include the geometric delay calculated above. The authors determined the impeded delay for each sub-segment by taking the difference between the average travel time and the unimpeded travel time. The impeded delay is a function of these elements: the upstream impeded delay is a function of FFS, the volume-to-capacity ratio, and the

14 Bugg, Schroeder, Jenior, Brewer, Rodegerdts entering flow rate, and the downstream impeded delay is a function of FFS, volume-to-capacity ratio, segment length, median length, and curb length across the segment: Delay impeded,upstream = *S f *v/c 0.03*V ent (9) Delay impeded,downstream = *S f + 3.1*v/c *L 0.001*L m *L c (10) where Delay impeded,upstream = average upstream impeded delay (s/veh), Delay impeded,downstream = average upstream impeded delay (s/veh), V ent = entering volume at the roundabout adjacent to the segment (vph), L = sub-segment length (ft), L m = length of sub-segment with restrictive median (ft), L c = length of sub-segment with curb (ft), S f = free-flow speed (mph), and v/c = volume-to-capacity ratio for the downstream roundabout, i.e. the roundabout adjacent to an upstream sub-segment or downstream of (not adjacent to) a downstream sub-segment. The upstream sub-model has an R 2 of 67%, while the downstream sub-model has an R 2 of 56%. All terms in the model were statistically significant (p<0.1), with the terms in the upstream sub-model being more significant (p<0.01). The upstream sub-model suggests drivers will experience more delay at a roundabout with a high level of congestion (indicated by a high volume-to-capacity ratio), but a roundabout with a higher level of entering flow will incur a lower amount of delay. This latter relationship may seem counterintuitive but should be interpreted along with the volume-to-capacity ratio term. Like the geometric delay model, a segment with a higher FFS is associated with greater impeded delay. For the downstream sub-segments, the FFS, sub-segment length, curb length, and volume-to-capacity ratio all increased the delay, but an increase in median length led to a decrease in the predicted delay. This last relationship may be due to the sample size or range of median lengths. Model Validation This section describes the results of the model validation effort. Two roundabout corridors, both from Carmel, Indiana, were withheld from the calibration dataset and were instead used to validate the models. By applying the HCM 2010 Chapters 16 and 17 methodologies, integrated with modifications for the roundabout nodes, the authors determined the corridor travel time and average speed for the two corridors. Figure 5 displays the results of the validation effort for the first corridor, Old Meridian Street. This corridor is unique to the dataset because it contains an internal signalized intersection between two of the roundabouts.

15 Bugg, Schroeder, Jenior, Brewer, Rodegerdts FIGURE 5 Validation Results for Old Meridian Street. The validation results suggest a close match of the predicted FFS for both northbound and southbound routes, with an error of 0.6 mph and 1.2 mph, respectively. For the average travel speed estimation, the northbound results for the a.m. and p.m. peak match the fieldobserved data with an error of 2.1 mph (8.0%) and 0.2 mph (0.8%) difference. For the southbound route, the average travel speed estimates are lower than the field observed data by 5.4 mph (17.0%) and 2.5 mph (9.2%) for the a.m. and p.m. peak, respectively. This difference in average travel speed translates to a difference in travel times of 0.0 and 0.2 minutes for northbound a.m. and p.m. peak, and 0.6 and 0.5 minutes for the southbound routes.

16 Bugg, Schroeder, Jenior, Brewer, Rodegerdts Figure 6 displays the validation results for the Spring Mill Road corridor FIGURE 6 Validation Results for Spring Mill Road. The Spring Mill corridor validation shows a close match between the model and field estimates. The model slightly overestimated the FFS and the average travel speed by approximately 2 to 5 mph across the four analyzed routes. This translates to a slight underestimation of travel time for three of the routes by 0.2 to 0.8 minutes. However, the resulting percent FFS measure proved to be within 10% of the field-observed data for all routes, which is explained because the error in FFS and average travel speed was in the same direction. HCM IMPLEMENTATION Figure 7 illustrates the calculation framework for applying the models within a given segment. The framework should be applied separately for each upstream and downstream sub-segment before eventually aggregating to the segment level. Further aggregation to the facility level can be performed using the urban streets analysis procedure in HCM 2010 Chapter 17. The framework is divided into computational steps A through L, with reference being made to the corresponding steps in HCM 2010 Chapter 17 for urban street segments at the appropriate time.

17 Bugg, Schroeder, Jenior, Brewer, Rodegerdts FIGURE 7 Roundabout Travel Time Calculation Methodology. First, the analyst gathers input data, and the FFS is calculated based on the posted speed limit, sub-segment length, and central island diameter of the roundabout, with the assumption that overlapping RIAs are not present. On a portion of a roundabout corridor between two roundabouts (i.e. not a first or last segment), the calculation is performed for a downstream subsegment and the upstream sub-segment. Using the model-predicted FFS and the circulating speed within the roundabout, the RIA length is calculated for both sub-segments. The analyst must then check whether the RIAs of the sub-segments overlap. If so, this necessitates a recalculation of the FFS with the overlap (OL) term set equal to 1, which will cause the predicted FFS to decrease. After the final sub-segment FFS is determined, the analyst selects the controlling FFS for each segment. Because the FFS is defined as being measured at the segment mid-point, the same FFS must be used for a downstream sub-segment and the next upstream sub-segment. In this

18 Bugg, Schroeder, Jenior, Brewer, Rodegerdts procedure, the lower of the two free-flow speeds controls and is selected as the FFS value for the entire segment. Next, this procedure uses HCM Chapter 17 Step 2 to estimate the segment running time, followed by roundabout-specific models to estimate geometric delay and impeded delay. Next, the sub-segment performance measures are aggregated to the HCM Chapter 17 segment level (Step J), and Chapter 17 Step 7 is used to determine the average travel speed on the segment. From the average travel speed, the level of service is estimated for the urban street segment with roundabouts. The facility LOS (HCM Chapter 16) and overall corridor travel time are then obtained by aggregating the performance of individual segments, consistent with the procedure for signalized corridors. Model Updates After the models were developed, the authors updated the models to make them more suitable for integration with the HCM urban streets analysis procedure. Specifically, the average travel speed models, while useful for comparison of the results with the existing methodology, were developed independently from the existing HCM Chapter 17 procedure. Thus, they are optional within the framework presented in Figure 7. Additionally, the downstream impeded delay model was updated to the following model: Delay impeded,downstream = *V T *L 0.001*L m *L c (11) where V T = the through-movement volume (vph) along the downstream sub-segment, and all other terms are previously defined. While the first model resulted in a better fit, this model is more intuitive because it removes the term for v/c ratio, an element that, in practice, will be difficult to predict or explain for the downstream portion of the impeded delay. CONCLUSION The modeling efforts described in this paper indicate that a series of simple linear models could be used to successfully describe the relationship between the components of roundabout corridor travel time (roundabout influence area, geometric delay, free-flow speed, average travel speed, and impeded delay) and the geometric and operational elements of a roundabout corridor. The data were modeled separately for upstream sub-segments and downstream sub-segments, and most of the models could explain over 70% of the variability in the data. A few of the models could be used to compute components of other models; e.g. the estimated FFS is an explanatory variable in the other four models. After completing the model development of all predictive models, the authors performed a validation exercise intended to verify that the proposed methodology results in satisfactory performance results. The authors presented an external validation of the methodology through application to two corridors in Carmel, Indiana. Both corridors were excluded from model development and thus represent a true validation of the methodology. The resulting percent FFS estimates matched the field-observed data within 10% for Spring Mill Road (all four routes) and for the northbound routes on Old Meridian Street. The Old Meridian Street southbound routes matched within a 20% difference, but this corridor also contains an internal signalized intersection between two of the roundabouts. After validation, the authors present a step-by-step framework for calculation of roundabout corridor travel time in a manner consistent with the existing HCM 2010 Chapter 17

19 Bugg, Schroeder, Jenior, Brewer, Rodegerdts methodology. This framework is intended to be used to incorporate the models into a future version of the Highway Capacity Manual. ACKNOWLEDGEMENT This paper is based on research conducted during NCHRP Project : Evaluating the Performance of Corridors with Roundabouts. The authors would like to thank the project panel for valuable feedback, as well as the state DOTs for supplying data and participating in interviews during the project. REFERENCES 1. Google Maps < Accessed Transportation Research Board. Highway Capacity Manual Transportation Research Board of the National Academies, Washington, DC, Rodegerdts, L. A., P.M. Jenior, Z. H. Bugg, B. L. Ray, B. J. Schroeder, and M. A. Brewer. NCHRP Report 772: Evaluating the Performance of Corridors with Roundabouts. Transportation Research Board of the National Academies, Washington, DC, Gross, F. C. Lyon, B. Persaud, and R. Srinivasan. Safety Effectiveness of Converting Signalized Intersections to Roundabouts. Presented at the Transportation Research Board 91 st Annual Meeting. Transportation Research Board of the National Academies, Washington, DC, Rodegerdts, L., J. Bansen, C. Tiesler, J. Knudsen, E. Myers, M. Johnson, M. Moule, B. Persaud, C. Lyon, S. Hallmark, H. Isebrands, R. B. Crown, B. Guichet, and A. O Brien. NCHRP Report 672: Roundabouts: An Informational Guide, 2nd ed. Transportation Research Board of the National Academies, Washington, D.C., Rodegerdts, L., M. Blogg, E. Wemple, E. Myers, M. Kyte, M. P. Dixon, G. F. List, A. Flannery, R. Troutbeck, W. Brilon, N. Wu, B. N. Persaud, C. Lyon, D. L. Harkey, and D. Carter. NCHRP Report 572: Roundabouts in the United States. Transportation Research Board of the National Academies, Washington, D.C., Persaud, B., R. Retting, P. Garder, and D. Lord. Safety Effect of Roundabout Conversions in the United States: Empirical Bayes Observational Before-After Study. Transportation Research Record Transportation Research Board of the National Academies, Washington, DC, Ariniello, A.J. Are Roundabouts Good for Business? LSC Transportation Consultants, Inc., Denver, CO. December < Accessed June Isebrands, H., S. Hallmark, E. Fitzsimmons, and J. Stroda. Toolbox to Evaluate the Impacts of Roundabouts on a Corridor or Roadway Network. Report No. MN/RC Minnesota Department of Transportation, St. Paul, MN, Akcelik, R. An Assessment of the Highway Capacity Manual 2010 Roundabout Capacity Model. Proceedings, 2011 Transportation Research Board International Roundabout Conference. Transportation Research Board of the National Academies, Carmel, IN, 2011.