Update of Highway Capacity Manual: Merge, Diverge, and Weaving Methodologies (2023)

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.

6 Overview This report documents the work conducted by NCHRP Project 07-26, “Update of Highway Capacity Manual: Merge, Diverge, and Weaving Methodologies.” The report is organized as follows: • Chapter 1, “Background,” describes the project background and objectives, synthesizes the results of a review of relevant literature, identifies key knowledge gaps, and summarizes the current (Version 6.0) Highway Capacity Manual (HCM) merge, diverge, and weaving method- ologies and their issues. • Chapter 2, “Research Approach,” describes the data collection effort undertaken by the project. • Chapter 3, “Findings and Applications,” describes the development of new HCM models for evaluating the speed, density, and capacity of merge, diverge, and weaving segments. • Chapter 4, “Conclusions and Suggested Research,” summarizes the improvements made to the HCM merge, diverge, and weaving methods as a result of this research, and suggests future research to address remaining knowledge gaps and to integrate the new segment-based methods into an HCM facility analysis. • Chapter 5, “References,” provides the reference list for Chapters 1 through 4. This final report has a compendium that contains additional information on the data defini- tions, data analysis, and detailed results. These materials can be found in NCHRP Web-Only Document 343: Traffic Modeling Document and include • Appendix A, “Data Dictionary,” defines all variables and parameters collected for all of the study sites; • Appendix B, “Data Analysis Table,” summarizes the capacity, speed, and density estimates for all sites; • Appendix C, “Speed–Flow Plots,” provides side-by-side comparisons of all speed–flow plots for all sites; and • Appendix D, “Detailed Site Summaries,” provides information about each site’s characteristics. In addition, the project developed proposed text for replacement HCM chapters to incor- porate the new merge, diverge, and weaving methodologies developed by this research, and to demonstrate their application through a series of example problems. • Revised HCM, Volume 2, Chapter 13 – “Freeway Weaving Segments.” • Revised HCM, Volume 2, Chapter 14 – “Freeway Merge and Diverge Segments.” • Revised HCM, Volume 4 (online), Chapter 27 – “Freeway Weaving: Supplemental.” • Revised HCM, Volume 4 (online), Chapter 28 – “Freeway Merges and Diverges: Supplemental.” Other products developed during this project include spreadsheet-based, computational engines implementing the proposed methods and a presentation summarizing the research and proposed C H A P T E R 1 Background

Background 7   enhancements to the HCM. NCHRP Web-Only Document 343 and products associated with this report can be found on the National Academies Press website (nap.nationalacademies.org) by searching for NCHRP Research Report 1038. At the time of writing, an update of the HCM (Version 7.0) was in the publication process. The core merge, diverge, and weaving methods contained in Version 7.0 are identical to the Version 6.0 methods described in this report, other than the incorporation of a few small clarifi- cations and errata. Version 7.0 also introduces capacity adjustment factors (CAFs) for the merge, diverge, and weaving methods to account for the presence of connected and automated vehicles (CAVs) in the traffic stream. The new material produced through this project is expected to be incorporated into the HCM in a future update. Research Problem Statement The project team was guided by the background and objectives published in the project’s Request for Proposal and reproduced here. Background Freeway congestion usually occurs at freeway merge, diverge, and weaving segments that have the potential to develop bottlenecks. To alleviate or mitigate the impacts of congestion at these segments, a number of active management operational strategies have been implemented such as ramp metering, hard shoulder running, managed lanes, etc. The current freeway merge and diverge methodologies in Chapter 14 of the HCM 6th Edition were developed more than 25 years ago using limited field-collected data. Although weaving segment analysis was updated more recently, the relationship with the merge and diverge methodologies has not been clearly addressed. In addition to limited data, the methodology does not conform to the fundamental relationship of traffic flow, namely that flow is the product of speed and density. The HCM does not offer any methodology for lane drops or additions, which often occur in the vicinity of freeway merge/ diverge segments. In the past decade, the data available to traffic engineers have expanded expo- nentially with ubiquitous sensor coverage of urban freeways and probe vehicle coverage of entire roadway networks. These new datasets provide a wealth of information to support the develop- ment of updates or changes to the merge, diverge, and weaving segment methodologies, and potentially complement traditional data sources. Objectives The objectives of this research are to (1) develop methodologies to update the HCM related to merge, diverge, and weaving methodologies and (2) pilot the developed methodologies to demonstrate the full range of applicability of the proposed updates to the HCM. Research Approach Accomplishing the project objectives required at least the following 11 tasks outlined within three phases. PHASE I—Planning Task 1. Conduct a literature review of relevant domestic and international research considered to be the state-of-the-practice on analysis of freeway merge, diverge, and weaving segments. The domestic review shall include research conducted through NCHRP, FHWA, and other national, state, and pooled-fund sponsored research. 16497-01_Summary-Ch01-3rdPgs.indd 7 5/4/23 3:01 PM

8 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Task 2. Synthesize the literature review to identify the knowledge gaps and summarize infor- mation on the effectiveness of existing analytical methodologies for freeway merges and diverges, emphasizing any available information on their compatibility with basic freeway and weaving segment methodologies. These gaps should be addressed in the final product or recommended future research as budget permits. Discuss the quantity and quality of data required to achieve the project objectives. Task 3. Propose methodologies to be executed in Phase II, based on currently available data. At a minimum, the methodologies shall address the following. – Data requirements and collection methods. – Different geometric configurations, including those that have not been adequately described in the HCM (for example, lane drops/adds, long acceleration/deceleration lanes for multi- lane ramps, auxiliary lanes, etc.). – Commonly implemented freeway management tools and strategies (for example, managed lanes, ramp metering, etc.). – Compatibility with the basic segment methodology parameters [for example, the free-flow speed (FFS) and CAFs]. Task 4. Identify areas of the HCM that require modification based on the proposed methodologies. Task 5. Prepare Interim Report No. 1 that documents Tasks 1 through 4, includes the data archiving plan, and provides an updated and refined work plan for the remainder of the research no later than 4 months after contract award. The updated plan must describe the process and rationale for the work proposed for Phases II through III. PHASE II—Methodologies Development and Proposed HCM Update Task 6. Execute Task 3 according to the approved Interim Report No.1 and submit a letter report at the end of Task 6 for panel review no later than 9 months after approval of Phase I. Task 7. Develop the proposed update to the HCM with example problems, including a com- parison to the current HCM methodologies. Task 8. Propose at least two states to pilot the proposed updates to the HCM to test and dem- onstrate the full range of their applicability. The pilot includes presenting the proposed HCM updates to the selected states’ DOTs, collecting required data to test the developed method- ologies, performing the analyses, and presenting the results to and collecting feedback from the selected states. NCHRP must approve the state selection. Task 9. Prepare Interim Report No. 2, which documents the results of Tasks 6 through 8 and provides an updated work plan for the remainder of the project. This report is due no later than 15 months after approval of Phase I. The updated plan must describe the work proposed for Phase III. PHASE III—Pilots and Final Products Task 10. Conduct the pilots and revise the HCM updates, including example problems, after consideration of the feedback received from the two states. Task 11. Prepare final deliverables including (1) a final report that documents the entire research effort and (2) the proposed HCM updates with the example problems and imple- mentation plan. Literature Review Introduction The literature review conducted for NCHRP 07-26 starts by first summarizing general consid- erations to keep in mind when comparing results from the literature, including limited dataset

Background 9   sizes and differences in traffic laws, traffic cultures, and road standards between countries. It then presents summaries of relevant literature on freeway merges, freeway diverges, and free- way weaving, respectively, and includes topics within each summary organized in generally decreasing order of previous research activity. A final portion covers potentially relevant recent literature on basic freeway segments. (Citations for the literature described in this review are provided in Chapter 5.) The next section in the chapter synthesizes the key findings from the literature review, including identifying knowledge gaps. The chapter’s final section provides an overview of merge, diverge, and weaving methods in the HCM 6th Edition. An initial set of 166 potential documents were identified by searching TRB’s Transport Research International Documentation (TRID) database, which contains over 1.2 million records on published transportation-related research. Various keywords related to freeway ramp merging, diverging, and weaving were used for this search. The time period covered by this search was limited to 2008 and later, corresponding to the cutoff date of the literature review conducted during the preparation of the HCM 2010. A few older documents were subsequently added to the review based on being classic papers in the field or providing relevant information that had not been subsequently updated. The scope of the literature review was kept broad, covering not only empirical studies of weav- ing, merging, and diverging operations but also papers on traffic flow theory (for example, gap- acceptance models), traffic simulation, and CAVs, with the expectation that papers not directly related to macroscopic modeling could nevertheless provide useful insights to consider when developing the project’s Phase II data collection plan. The initial number of documents for review was reduced by eliminating duplicate papers, papers describing incremental improvements in an author’s theory, papers on freeway work zone merging (for example, merging due to a mainline lane reduction at a work zone), papers on urban street weaving, theoretical papers without at least one comparison of results to field data or current HCM methods, and papers that were not relevant to the project objectives. As a result, this review includes relevant information from more than 70 documents, mostly pub- lished since 2010. Results from the literature are presented using the measurement units from the original paper or report. Approximate values for key conversion factors are 1 meter per second (m/s) = 3.6 km/h and 1 km/h = 0.6 mph. General Considerations Dataset Sizes in the Literature Most of the papers included in the review reported findings based on field data collected at a small number of sites—typically one or two. There are only a handful of larger studies of freeway weaving and merging operations, and even those were not able to cover close to the full range of possible conditions. These studies include: • Weaving – NCHRP Project 03-75, “Analysis of Freeway Weaving Sections” (Roess 2008), the basis for the current HCM weaving method, assembled data from one pilot site, 10 full data collec- tion sites, one site provided by the Ohio Department of Transportation, and two FHWA Next Generation Simulation (NGSIM) sites. – Skabardonis and Mauch (2014) assembled data from studies conducted by Caltrans in the 1980s and 1990s, the NCHRP 03-75 dataset, and the NGSIM dataset (28 sites total), and collected new data at three sites in San Diego. – Xu et al. (2020) collected new data at five sites in North Carolina and used data from one NCHRP 03-75 site.

10 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies – Zhao et al. (2019) studied seven sites in Nanjing, China that had special pavement mark- ings requiring exiting vehicles to weave in the first half of the auxiliary lane and entering vehicles to weave in the second half. • Merging – NCHRP Project 03-37, “Capacity and Level of Service at Ramp-Freeway Junctions,” the basis for the current HCM merge procedure, assembled approximately 1 hour of data each from 57 single-lane merge sites, of which 42 sites could be used for estimating density (Roess and Ulerio 1994). – Bates et al. (2013) developed a model of breakdown probability and queue discharge flow from detector data at 34 freeway on-ramps in the United Kingdom. – Rouphail et al. (2015) studied seven unmetered single-lane on-ramps in Florida, California, and North Carolina. – Asgharzadeh and Kondyli (2018) studied six freeway merges in the Kansas City area. • Diverging – NCHRP 03-37, the basis for the current HCM diverge procedure, assembled approximately 1 hour of data each from 18 single-lane diverge sites, of which 16 sites could be used for estimating density (Roess and Ulerio 1994). • Combinations – Elefteriadou, Kondyli, and St. George (2014) analyzed detector data from 10 urban and four rural sites in Florida, including merge, diverge, and weaving sections that experience breakdowns. – Van Beinum (2018) used a helicopter to collect data at 14 sites (three on-ramps, three off- ramps, and eight weaving segments) in the Netherlands. Considerations When Comparing Findings from Different Countries When comparing findings from different countries, potential differences in how people drive should be considered, whether due to driving rules, driving culture, road standards, or a com- bination of these. This section discusses key issues specific to merging, diverging, and weaving when considering these differences. Traffic Laws or Culture Regarding Right-of-Way at Merges Countries such as Germany, the Netherlands, Spain (Nygaard 1995), and China (Wang and Yang 2012) give absolute priority to mainline vehicles. Denmark gives the right-of-way to the lead vehicle, while Sweden uses a zipper-merging approach with slight priority given to main- line vehicles (Strömgren 2016). In some places, regardless of traffic laws giving priority to freeway traffic, the culture may be to create a gap for a vehicle to merge into, whether by shifting lanes if possible or by slowing down to open a space. For example, in the Netherlands, less traffic has been observed in the right-hand mainline lane just before a freeway on-ramp than on a basic freeway segment, indicating the presence of courtesy lane changing (Knoop et al. 2010). These differences in merging and lane-changing behavior can affect where lane changes take place and the size of the accepted gaps. Traffic Laws Regarding Lane Use Many jurisdictions reserve the left lane for passing on freeways with two directional lanes. When three or more lanes are present, some countries adopt a “slower traffic keep right” approach to minimize lane-changing activity by encouraging drivers to use the lane with a flow that best matches their desired speed. Other countries, such as the Netherlands and Denmark, use a “keep as far right as possible except to pass” approach (van Beinum 2018), which can result in considerable lane changing at lower traffic volumes (for example, moving from the right lane to the middle lane to the left lane and back when a car passes a slower-moving car that is passing

Background 11   an even slower-moving truck). These differences can affect the overall observed turbulence in the freeway traffic stream, particularly when combined with differential passenger car and heavy vehicle speed limits (for example 120 to 130 km/h and 100 km/h, respectively). China segregates lanes by vehicle type; for example, on a freeway with four directional lanes, the left lane (lane 1) is reserved for cars, cars and buses have priority in lane 2, and trucks have priority in lanes 3 and 4. The speed limit is higher in lanes 1 and 2 than in lanes 3 and 4 (120 km/h versus 100 km/h) (Kong et al. 2015). Knoop et al. (2010) studied a three-lane directional freeway section in the Netherlands, using detector data. Under low-volume conditions, nearly all traffic was in the right lane (under a keep- right-except-to-pass driving law). At densities of 30 veh/km and greater, more traffic was in the left two lanes than in the right lane, as motorists avoided being slowed down by heavy vehicles. Above 50 veh/km, the greatest flow was in the left lane, as drivers tried to avoid slower traffic in the right two lanes. After the freeway breakdown (around 75 veh/km), the flow in the left two lanes equalized (the prohibition on passing on the right no longer held) but was still greater than the flow in the right lane. Road Standards Van Beinum (2018) observed an uptick in lane-changing activity about 600 m in advance of an off-ramp in the Netherlands; this is the standard position in the Netherlands for the second upstream exit guide sign. Other potential differences between countries are the use of tapered accel- eration and deceleration lanes at ramps versus parallel (full) lanes (see Figure 1) and the ramp design speed. (These two characteristics can be related—for example, Germany typically uses deceleration lanes but also tends to have low off-ramp design speeds.) Merges Gap Acceptance and Merging Location Daamen, Loot, and Hoogendoorn (2010) studied gap-acceptance behavior at two on-ramps in the Netherlands, using videos shot from a helicopter. They concluded that gap-acceptance theories based on a fixed critical gap do not match the observed behavior. They found that accepted gaps were shorter at the end of the acceleration lane than at the start. Under uncon- gested conditions, 51% of merges occurred within the first 87.5 m of a 200-m acceleration lane. When merge speeds were lower than 60 km/h, merge locations were more spread out along the Source: HCM 6th Edition, Exhibit 14-5. (a) Parallel Acceleration Lane (b) Tapered Acceleration Lane (c) Parallel Deceleration Lane (d) Tapered Deceleration Lane LA LA LD LD Figure 1. Examples of parallel (full) [(a) and (c)] and tapered acceleration and deceleration lanes [(b) and (d)].

12 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies length of the acceleration lane. The smallest accepted gaps in the dataset were between 0.75 and 1.0 seconds, resulting in net headways between two cars of 0.25 seconds. These headways increased downstream to the end of the observable area (about 450 m) although the time/distance to reach normal headways could not be determined. The authors concluded this finding was evidence of the “relaxation” effect that contributes to capacity drop. Deng and Zhang (2015) developed a simulation of a merge based on NGSIM data for east- bound I-80 in Emeryville, California. They found (1) differences in drivers’ choice of merging location (early or late) can lead to oscillations in traffic that can develop into stop-and-go traffic, (2) a consistent choice of merge location eliminates persistent oscillations, and (3) longer merging lanes offer more choices of merging location, which can help smooth out oscillations. Sun et al. (2018) collected video data for a freeway merge in Shanghai and used NGSIM data from US-101 in Los Angeles. The NGSIM site is a weaving segment that has a much higher merge volume than diverge volume; merges that were affected by diverging traffic were not included in the analysis. In both locations, about 55% of drivers rejected at least one gap that was larger than the one they eventually accepted. On average, drivers rejected 2.0 to 2.2 gaps of any size before accepting a gap. [It appears that lane changes occurring before the gore point were not included. Ahmed et al. (2019), using the same NGSIM dataset, found that 50% of weaving lane changes occurred in the first 110 ft of the short length, including in the painted gore area.] In Shanghai, the merging condition (for example, forced merge) worsens as the number of rejected gaps increases, while in Los Angeles it improves (for example, merge at similar speeds) as the number of rejected gaps increases. The auxiliary lane in Los Angeles operates at a higher speed than the acceleration lane in Shanghai. The authors conclude that Shanghai drivers who reject gaps smaller than the one they eventually accept do so to save time; Los Angeles drivers do so to find a better merging condition. The distance to the end of the acceleration/auxiliary lane and the speed differential between the acceleration/auxiliary lane and the mainline lane were found to be predictors of merging behavior. Wan et al. (2017) also studied the Los Angeles US-101 NGSIM site and, like Sun et al. (2018), appear to have only considered lane changes from the auxiliary lane that occurred after the gore point. However, they found that 39% of vehicles rejected the first available gap, compared to 55% for Sun et al. (This seems to be a considerable difference in results, given the two studies used the same source data.) Shen, Qiu, and Zheng (2015) developed a gap-acceptance model for whether a ramp driver will accept a given headway in traffic in the rightmost mainline lane. They tested different driver characteristics (aggressive, average, conservative) and ramp headways, while using a capacity of 1,200 passenger cars per hour (pc/h) in the right lane based on the HCM. They concluded that ramp metering needs to be provided to even out headways when the ramp volume exceeds 148 pc/h. Kong et al. (2015) studied a merge on the Shanghai−Nanjing freeway where pylons are manu- ally installed during high-volume periods to force on-ramp traffic to merge from the second half of the 500-m acceleration lane. They used Aimsun software to compare operations with and without the late merge operation and found that lane changes occurred at higher speeds, with less variation in speed, and in more concentrated locations, resulting in improvements in mainline density and speed. They also found almost no additional benefit of a late merge beyond 200 m. (The ramp has a standard speed limit of 40 km/h, resulting in a significant speed differ- ential with the mainline lanes at the merge point.) Kondyli and Elefteriadou (2010) observed merging driver behavior in Jacksonville, Florida, using a combination of an instrumented car and traffic management center videos. They found

Background 13   that on average, drivers used 41.5% of the acceleration lane length and that very few maneuvers occurred prior to the end of the solid white stripe. The average merging speed and the variation of speed were higher with a tapered merge compared with an acceleration lane. Lwambagaza et al. (2017) observed merging behavior at two on-ramps on I-75 between Fort Myers and Naples, Florida. The ramps had acceleration lane lengths of 1,000 and 1,500 ft respec- tively. The authors compared merging behavior by age group (younger, middle-aged, older). They found that fewer than 10% of drivers in any age group merged in the first 75 to 110 ft at any level of service (LOS) between A and E and that drivers tended to use more of the accelera- tion lane as LOS decreased (that is, vehicle density increased). On freeways, LOS A generally corresponds to low-volume free-flow traffic conditions, while LOS E corresponds to traffic volumes that are close to capacity, but short of resulting in a breakdown of stable flow. Younger and middle-aged drivers tended to merge using the first half of the acceleration lane, while older drivers were more likely to use the last half of the acceleration lane. Merging speeds by age group were similar across LOS levels, except for the taper section of the shorter ramp at LOS E, where older drivers merged more slowly (9 versus 18 mph) than other drivers. Ramp Influence Area Van Beinum (2018) used loop detector data from sites in the Netherlands to compare a free- way’s speed and traffic distributions in the vicinity of on-ramps to those of basic freeway seg- ments. He found that on-ramp influence areas on a three-lane directional freeway started 200 to 300 m in advance of the on-ramp and ended about 90 m downstream. In a follow-up study using video shot from a helicopter at three sites, he determined that discretionary lane changes on the mainline in response to traffic entering from a ramp take place 25 to 100 m upstream of the ramp. In addition, 26% to 41% of merging vehicles performed discretionary lane changes; these took place up to 475 to 575 m downstream of the ramp. Kondyli and Elefteriadou (2010) conducted a study of merging behavior in Jacksonville, Florida. They determined that lane changes due to merging activity started up to 360 ft upstream of the gore and occurred up to 840 ft downstream of the gore. More than 400 merges were observed using an instrumented car and traffic management center videos. Merge Ratios and Traffic Flow in the Right-Hand Mainline Lane Rouphail et al. (2015) noted that four main approaches to allocating flow between an on-ramp and the mainline during periods of breakdown have been used in the literature: • Allocation proportional to the relative demand of the on-ramp and mainline flows. • Allocation proportional to the upstream capacity or number of lanes of the two flows. • Allocation based on a share of the traffic present in the rightmost mainline lane. • Allocation based on multiregime approaches (including the HCM) based on the relative demand between the two flows. Rouphail et al. (2015) studied seven unmetered single-lane on-ramps in Florida, California, and North Carolina, and developed a model to predict the merge ratio (ratio of on-ramp traffic to total traffic). The model inputs are the downstream demand flow per lane and the accelera- tion lane length per 10 mph of the speed limit. The presence of adjacent ramps was not tested, and no information was provided about possible nearby ramps, although this could probably be determined from the site information provided in the paper. Dehman and Drakopoulos (2013) studied four metered on-ramps in Milwaukee. At three of the on-ramps, prebreakdown flow, queue discharge flow, queue discharge speed, and break- down duration worsened with increasing proportion of ramp traffic. At the fourth on-ramp,

14 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies breakdown occurred because of mainline traffic changing lanes to the left in advance of the on-ramp; in this case, prebreakdown flow and queue discharge flow increased with an increas- ing proportion of ramp traffic due to poor mainline utilization of the right-hand lane. Mainline operations did not clearly deteriorate with increasing ramp volume, indicating the benefit of the ramp meters to manage the merging traffic. In a follow-up study of the same ramps, Dehman and Drakopoulos (2016) found that 88% of breakdowns at one site and 43% of breakdowns at a second site occurred at volumes less than 4,600 pc/h in the ramp influence area (RIA); the HCM identifies 4,600 pc/h as the maximum desired flow rate entering the RIA, and the authors sug- gested this value may be too high. They also defined a “ramp volume ratio” as the ramp volume divided by the sum of the ramp volume and the right two mainline lanes volume and found that higher ramp-volume ratios were associated with lower prebreakdown, queue discharge flows, and queue discharge speed, as well as longer congestion durations. Reina and Ahn (2014) reference eight studies showing that flow distributions by mainline lane vary significantly between sites under both congested and uncongested conditions. They studied six congested freeway-to-freeway on-ramps in Southern California: four with full lane adds (sum of upstream mainline + ramp lanes = downstream mainline), one where two ramp lanes ended up as one added lane downstream, and one where three ramp lanes ended up as two added lanes downstream. Different flow distributions were observed at two sites with similar merging configurations; one had a downstream on-ramp, while the other had a downstream off-ramp. They developed models of flow distribution by lane based on upstream mainline flow, the presence of a downstream on-ramp within 0.6 mi, and the presence of a downstream off- ramp within 0.6 mi (only the closest on- or off-ramp is used). The lane flow distributions were then used to predict merge ratios under two scenarios: “fair share” (proportional to mainline versus ramp flow) and “zipper” (1 to 1). Fair share worked better for full lane adds, while zipper worked better where a lane was dropped. The validated models had relatively low errors but did not perform better than the ratio of mainline-to-ramp lanes in many cases. Breakdown Probability Elefteriadou et al. (2011) studied two metered on-ramps, one in Minneapolis and the other in Toronto, and developed models of breakdown probability for different combinations of freeway flow and ramp metering rate. As ramp volume (metering rate) increases, the maximum freeway flow for a given breakdown probability decreases. The authors recommended using a 20% break- down probability as the threshold for activating ramp meters. Bates et al. (2013) developed a model of breakdown probability and queue discharge flow from detector data at 34 motorway on-ramps in the United Kingdom. Model inputs consisted of the combined mainline and ramp volume and the number of downstream mainline lanes (three lanes, or four or more lanes). Other than the number of downstream lanes, different merge configurations did not show statistically significant differences in breakdown probability or queue discharge flow. Han and Ahn (2018) presented a model of breakdown probability and tested it on one free- way merge site on SR-91 in Southern California. The model only considers the right mainline lane of the freeway and not the other freeway lanes. The model indicates that breakdown prob- ability increases with increasing mainline flow, increasing merging flow, decreasing merging speed, and increasing merging spacing. The breakdown probability is 50% when the sum of the right-lane volume and the merge volume is approximately 2,000 vehicles per hour (vph). Below this volume, the breakdown probability is lower for a given volume with more even merging headway distributions; above this volume, the breakdown probability is lower with more uneven headway distributions.

Background 15   Capacity Elefteriadou, Kondyli, and St. George (2014) analyzed detector data from 10 urban and four rural sites in Florida, including three merge, three diverge, and eight weaving sections that expe- rience breakdowns. A “breakdown” was defined as an abrupt speed drop of at least 10 mph at either the upstream or downstream detector. Capacity drops of 5% to 10% were observed, with weaving segments appearing to have higher prebreakdown capacities than merge or diverge seg- ments. Merge and diverge segments had comparable prebreakdown and postbreakdown capaci- ties. Three-lane facilities had higher per-lane capacities than either two-lane facilities or facilities with four or more lanes, which aligns with previous research. The authors recommended pre- breakdown and postbreakdown capacity values (Table 1) for urban and rural facilities with three directional lanes and separate values for segments with two, four, or more lanes. These recom- mended values were to be thought of as planning-level values; field data showed that site-specific capacities varied. Rural capacity values were determined to be lower than urban capacity values; however, fewer rural breakdown events were available for analysis. Kondyli, Gubbala, and Elefteriadou (2016) analyzed detector data from five sites (one in Toronto, two in California, and two in Florida) that experienced breakdowns due to merges. The Florida sites were not metered, while the other sites were metered. In general, the higher the ramp volume, the lower the merge capacity. In addition, per-lane capacity was lower for four- lane mainlines than for three-lane mainlines. Tables are presented with recommended capacity values for different combinations of per-lane upstream demand and the ratio of ramp demand to per-lane upstream demand (see Table 2). Aksoy and Öğüt (2018) studied three unmetered freeway-to-freeway merges in Istanbul, Turkey, two of which involved inside merges (that is, the merging lanes had travel lanes to both Area Type Condition 3 Mainline Lanes 2 & 4+ Mainline Lanes Urban Prebreakdown 2,100 pc/h/ln 2,000 pc/h/ln Queue discharge 1,900 pc/h/ln 1,800 pc/h/ln Rural Prebreakdown 1,900 pc/h/ln 1,700 pc/h/ln Queue discharge 1,800 pc/h/ln 1,600 pc/h/ln NOTE: pc/h/ln = passenger car per hour per lane. SOURCE: Elefteriadou, Kondyli, and St. George (2014). Table 1. Planning-level merge and diverge capacity values. Vup (vph/ln) Vr / Vup 1,600 1,800 2,000 2,200 2,400 0.1 1,845 1,963 2,080 2,197 2,315 0.2 1,827 1,944 2,062 2,179 2,297 0.3 1,809 1,926 2,044 2,161 2,278 0.4 1,791 1,908 2,025 2,143 2,260 0.5 1,772 1,890 2,007 2,125 2,242 0.6 1,754 1,872 1,989 2,106 2,224 0.7 1,736 1,853 1,971 2,088 2,206 0.8 1,718 1,835 1,953 2,070 2,187 0.9 1,700 1,817 1,934 2,052 2,169 1.0 1,681 1,799 1,916 2,034 2,151 NOTE: vph/ln – vehicle per hour per lane; Vr = ramp volume; Vup = upstream mainline volume. SOURCE: Kondyli, Gubbala, and Elefteriadou (2016). Table 2. Merge capacity values (vph/ln) for three-lane freeways.

16 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies the right and the left) and found that the merge section capacity decreased as the merging volume increased. Merge section capacity ranged from 2,138 pc/h/ln when the upstream freeway flow was four times the on-ramp flow to 1,316 pc/h/ln when the two flows were equal. Morris et al. (2011) analyzed detector data for two motorway links (that is, sections between an on-ramp and the next downstream off-ramp) in the United Kingdom. They compared observed throughput to various factors including precipitation, lighting condition (time of day), percent heavy vehicles, percent merging traffic, and percent diverging traffic. At one site, throughput was sensitive to the percent heavy vehicles and the percent merging traffic, while the other site was sensitive to all five factors. Shi et al. (2015) used video cameras installed as part of Beijing’s traffic management system to observe 50 uncongested freeway merges. Based on observations of merge headways, they developed merge CAFs for ramp geometry. These factors reduced the capacity from the Chinese standard merge capacity of 2,100 pc/h. The inputs to the model are merge type (tapered or accel- eration lane), merge length, and ramp angle. Ramp angle was a factor for acceleration lane lengths less than 100 m (including tapered ramps); merge capacity decreased with increasing ramp angle. Merge capacity increased with increasing merge length. For a five-degree ramp angle, a tapered ramp provided higher capacity than acceleration lane lengths of 75 m or less. Ramp Metering and Variable Speed Limits Li et al. (2018) developed calibrated cell transmission models for three merge bottlenecks in California. They noted that the theoretical maximum benefit of ramp metering is dependent on the bottleneck capacity and the capacity drop, while the theoretical maximum benefit of variable speed limits (VSL) is dependent on bottleneck capacity, capacity drop, and the diverge ratio. Both the ramp metering and VSL strategies reduced system travel time in the simulation. Ramp metering was able to achieve the theoretical capacity benefit until such time that the ramp storage was filled, and the queue was released. VSL was about 84% efficient, even with assumed perfect compliance, due to the need to post speeds in multiples of 5 mph. Ramp metering pro- duced a better capacity benefit than VSL when ramp storage was sufficient. The authors cited a previous study that suggests that with limited driver compliance with VSL, the feedback mecha- nism would simply lower the speed further, resulting in only minor impacts to the strategy’s outcomes. Yousif and Al-Obaedi (2011), in a simulation study, also observed the importance of providing sufficient ramp storage to avoid queue flushes from the ramp meter; longer ramp lengths produced higher upstream capacity. Knoop et al. (2010) compared vehicle lane distribution on a freeway with three directional lanes in the Netherlands with and without a 60 km/h VSL in effect. With VSL activated, more traffic used the right-hand freeway lane prior to an on-ramp than without VSL activated, resulting in more efficient use of the mainline lines. At the same time, more traffic in the right lane meant smaller gaps at the merge. Knoop et al. used detector data; therefore, they could not evaluate traffic speeds at the merge. However, they suggested that the result would either be more ramp congestion (waiting for acceptable gaps, given full priority to freeway traffic under the traffic laws) or forced merging at slower speeds, resulting in congestion at the merge point. Average mainline speeds with a 60 km/h VSL were 79 km/h with a standard deviation of 16 km/h. Li and Ranjitkar (2013) simulated a congested section of a motorway in Auckland, New Zealand, that contained five metered on-ramps. They tested two common ramp metering algorithms in combination with the presence or absence of VSL. Compared to a noncontrolled state, the ramp metering algorithms resulted in lower throughput at one or two sites. The combination of ramp metering and VSL produced better throughput than ramp metering alone at three sites. Duan et al. (2013) simulated a section of I-5 in San Diego County, California, and tested the separate effects of ramp metering and VSL on throughput. Both strategies improved throughput,

Background 17   but ramp metering was twice as effective overall (in terms of percentage increase in through- put). Ramp metering was considerably more effective than VSL when on-ramp volumes were low. Zhang et al. (2013), who simulated the same section of I-5, noted that too-aggressive use of VSL merely transfers the delay upstream rather than providing a capacity benefit. However, by spacing out vehicle arrivals in the merge area, VSL has the potential to keep the discharge flow near the bottleneck capacity. In their simulation, VSL improved throughput and average speed. Jin et al. (2017) also identified the need to regulate gaps in the right-hand mainline lane. They proposed a gap metering system in which drivers would be informed of the desired gap in advance of an on-ramp and the distance to their leader. Using simulation, they found that with 20% compliance, a delay savings of 17% could be achieved using gap metering alone, and a delay savings of 27% could be achieved when using gap metering in combination with ramp metering. Texas A&M Transportation Institute (TTI) (2019) described dynamic merge control systems used in the Netherlands and Germany. These systems are used at some two-lane on-ramps with inside merges to close either the right-hand mainline lane or the left-hand ramp lane, depending on the relative volumes on the mainline and the ramp. Overhead lane control signs are used to close the appropriate lane. A pilot study from the Netherlands referenced by TTI found reduc- tions to mainline and ramp delay of 4% to 13% as a result of the system. Alternative Capacity Models Leclercq, Laval, and Chiabaut (2011) discussed classical analytic models of merges (Newell 1982; Deganzo 1995) that have been shown to accurately replicate field conditions when the merge is not an active bottleneck. However, these models are not able to predict the capacity drop at breakdown because capacity is an input to the models. The authors proposed an exten- sion to the classical models to predict the capacity drop. In their model, the capacity drop at breakdown increases nearly linearly with increasing ramp flow while the ramp is in free flow. Once the ramp becomes congested, the capacity drop reaches a maximum value and remains constant thereafter. The relative capacity drop decreases with the length of the acceleration lane; most of the improvement occurs with lengths less than 100 m. The merge ratio does not influ- ence the capacity drop when the mainline has only one lane. The model showed promise when comparing data from one freeway site in the United Kingdom that focused only on the right- hand freeway lane. Leclercq et al. (2016) extended their analytical model to cover multilane freeways. Accord- ing to their model, the local merge ratio has almost no effect on the capacity of the rightmost mainline line: higher merge ratios mean higher traffic and higher speeds on the ramp with corresponding less impact on right-lane speeds, while lower merge ratios mean less traffic but lower speeds on the ramp with corresponding greater impact on right-lane speeds. The capacity of the next-to-the-right lane (C2) was found to be greater than the capacity of the rightmost lane (C1). The length of the area where discretionary lane changes occur had only a small effect on C2, with only a 6% difference between the shortest and longest lengths tested, while the car acceleration rate significantly influenced C1 and C2 (+15% to 17%). Truck percentage was also a significant factor influencing capacity. The model was tested against simulation and one field site in Manchester, United Kingdom, with model results within 9% of the field values. The model requires solving a series of simultaneous equations. Chen and Ahn (2018) developed a numerical model to estimate the capacity drop for a merge with a one-lane mainline (see “Freeway Weaving” later in this chapter for details). They found that the merge capacity drop increases as the merging flow increases (with most of the effect occurring with the first 100 to 200 vph), the merging length decreases, and the merging speed decreases.

18 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Asgharzadeh and Kondyli (2018) compared four freeway merge capacity models—van Aerde, Product Limit Method, Sustainable Flow Index, and HCM 6th Edition—to data collected at six freeway merges in the Kansas City region. The van Aerde method matched the observed average breakdown capacities best, while the HCM model produced the lowest capacity estimates (and lower than the observed capacity) at four of the five sites that had sufficient data available to estimate capacity. They also found significant variability in the HCM’s estimated breakdown prob- ability function, which was greatly impacted by the number of breakdown observations that fell within each volume range. Wu and Lemke (2016) described the German Highway Capacity Manual (HBS) model, which provides a uniform function for all types of merge, diverge, and weaving segments, and which was developed from field data. Traffic quality is assessed in one step for the mainline upstream or downstream of the segment, on the ramps, and in the maneuvering area. All boundary con- ditions are satisfied. The paper gives examples of segment and ramp capacities for a variety of geometric configurations. Inputs required for a diverge are the exiting mainline and off-ramp demands and the geometric configuration. Inputs required for a merge are the entering main- line and on-ramp demands and the geometric configuration. The weaving model applies only to ramp weaves with one or two mainline lanes and requires all entering and exiting demands as inputs. The length of the acceleration lane, deceleration lane, or weaving area is not an input to the model, although the model was calibrated using standard design lengths for these values. Changing the value of this calibration parameter was said to have a minimal effect on capacity. Use of Simulation Sun et al. (2018) found that more than half of merging drivers at sites in Shanghai and Los Angeles rejected a gap that was larger than the one they eventually accepted. They noted that major simulation models (for example, MITSIM, CORSIM, VISSIM, TransModeler) do not account for this behavior. Diverges Prebreakdown Capacity Elefteriadou, Kondyli, and St. George (2014) analyzed diverge capacities experiencing break- downs at three sites in Florida. Diverge capacities were found to be similar to merge capacities but lower than weaving capacities. Table 1, presented previously in the “Merges” section, pro- vides their recommended values for prebreakdown and postbreakdown capacity for urban and rural segments. Capacity Drop Sun, Ma, and Li (2015) studied three freeway off-ramps in Shanghai, China, that experienced breakdowns that were not due to downstream bottlenecks on the off-ramps. They found that when the diverge ratio (that is, the off-ramp volume divided by the total volume) was low (0.16 to 0.40), queue discharge flow was less than prebreakdown flow; however, at a high diverge ratio (0.80), queue discharge flow was greater than prebreakdown flow. In addition, both pre- breakdown and queue discharge capacities decreased as the diverge ratio increased. The freeway had a speed limit of 80 km/h and four mainline lanes, while the study off-ramps had one or two lanes. In two cases, off-ramps were accessed by full deceleration lanes added on the right side of the freeway. In the third case, the right-hand freeway lane was forced to exit, while the lane to its left was an option lane to continue on the mainline or to exit. Chen and Ahn (2018) developed a numerical model to estimate the capacity drop for a diverge with a one-lane mainline (see “Free Weaving” later in this chapter for details). They found that

Background 19   the characteristics of a diverge bottleneck were like those of a merge bottleneck: the capacity drop increases as the diverging flow increases (most of the effect occurs with the first 100−200 vph), the diverging length decreases, and the diverging speed decreases. However, unlike a merge bottle- neck, traffic flow is relatively stable upstream of the bottleneck. Major Diverge Pan et al. (2016) studied a 2-mile section of southbound SR-241 in Orange County, California. This facility had three directional lanes in advance of a major diverge with insufficient peak-hour capacity on the branch leading to SR-261. Detector data indicated that the right-hand lane had a lower FFS and capacity than the left two lanes. Differential speed limits (65 mph for autos and 55 mph for trucks and autos towing trailers) were in effect. Many drivers appeared to be familiar with the presence of congestion; drivers continuing on SR-241 pre-positioned themselves in the left lane to try to avoid congestion in the center (option) lane caused by traffic headed to SR-261. After breakdown occurred, densities in the left and center lanes evened out, as discretionary lane changes occurred. The authors developed a cell transmission model for the site that simultane- ously considered mandatory and discretionary lane changes. Influence Area Van Beinum (2018), using loop detector data from sites in the Netherlands, studied off-ramp influence areas on a three-lane freeway. Influence areas were determined by comparing the freeway’s speed and traffic distributions in the vicinity of off-ramps to those of basic segments. He found that the influence area started 1,000 m in advance of the off-ramp, corresponding to the standard positioning of the first exit guide sign 1,200 m in advance of an off-ramp and ended about 600 m downstream of the off-ramp. In a follow-up study using video shot from a helicopter at three sites, it was determined that 85% of off-ramp vehicles were pre-positioned in the right lane at the start of the observation area. A total of 1,200 to 1,500 m of roadway could be observed; the first exit guide sign at these ramps was located 1,200 m upstream. (The Netherlands’ keep-right-except-to-pass law probably accounts for some of this pre-positioning.) Exiting vehicles needed to perform a lane change to enter the off-ramp deceleration lane; 96% of these lane changes occurred in the first half of the deceleration lane. The off-ramp influence area was determined to start 400 to 600 m in advance of the ramp (600 m corresponding to the standard position of the second exit guide sign) and end 200 to 375 m downstream of the ramp. Freeway Weaving Tests of the HCM 6th Edition Methodologies Xu et al. (2020) compared HCM 6th Edition estimates of space mean speeds to actual speeds measured at six ramp weave sites. The HCM underestimated 75% of speeds, particularly when speeds were in the 50 to 65 mph range; however, the HCM’s speeds fit well at higher and lower speeds. Ahmed et al. (2019) tested the HCM 6th Edition weaving method and found that the HCM’s weaving speeds are sensitive to weaving segment length, while nonweaving traffic speed has no sensitivity to weaving segment length. Because most traffic is nonweaving, changing the segment length has minimal effect on the predicted speed and, therefore, minimal effect on the predicted density and LOS. The authors also compared data from the NGSIM US-101 (Los Angeles) site to the HCM under breakdown conditions. The weaving lane-change rate matched the HCM model’s trend, while the HCM model trend for nonweaving lane changing was counterintuitive. Skabardonis and Mauch (2014) assembled weaving segment data from earlier California studies, the NCHRP 03-75 dataset, and the NGSIM dataset. The data included 10 ramp weaves

20 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies (nine of which had cross sections of five or more lanes), 13 balanced major weaves, and five unbalanced major weaves. The HCM 2010 method was applied to the California datasets that had not been used by NCHRP 03-75 to develop the HCM method. The HCM predicted over- capacity conditions in 15 datasets that operated below capacity; the authors concluded that the method appears to underestimate capacity at sites with high weaving ratios. In eight datasets, volumes were below the HCM-predicted capacity, but the HCM-calculated density exceeded the critical density of 43 pc/mi/ln, which was an inconsistency in the method. Across all datasets, the HCM-predicted density was on average 22% higher than the field-measured density. Skabardonis and Mauch (2014) also collected new data at three ramp weaves in San Diego. At these sites, the HCM method overpredicted densities under low-volume conditions and under- predicted densities when densities were 30 to 40 pc/mi/ln. In four instances, the HCM also predicted densities greater than 43 pc/mi/ln when volumes were below the predicted capacity. Jolovic et al. (2016) calibrated FREEVAL (that is, HCM) and VISSIM models to replicate flows and speeds for a section of I-880 in Fremont, California, consisting of basic, merge, diverg- ing, and weaving segments. VISSIM densities were expressed in vehicles per mile per lane, while FREEVAL densities were expressed in both vehicle miles per lane and passenger cars per lane. In the initial basic segment, VISSIM closely matched field-observed densities, while FREEVAL overestimated density. In the merge segment, both FREEVAL densities (using passenger car equivalents or vehicles) and VISSIM were not significantly different from field densities. Field- measured density could not be determined in the weaving segment due to a lack of detectors, but both VISSIM and FREEVAL vehicle-based densities were similar. Finally, VISSIM and FREEVAL vehicle-based densities in the final basic segment were not significantly different from the field- measured values; VISSIM densities were lower than the FREEVAL densities. Zhou et al. (2015) applied the HCM 2010 method to four example cases and found that in some situations, the HCM predicts that weaving speeds will be greater than nonweaving speeds. They collected data at one ramp weave in Tianjin, China. The HCM overpredicted weaving speed and underpredicted nonweaving speed; the average errors were around 20%. The authors devel- oped a modification to the HCM model to incorporate traffic flow rate and address the identified issue. The average errors of the calibrated model were around 7% to 10%. Stanek (2014) observed that when a weaving section has high proportions of entering or exiting traffic, the HCM weaving method produces worse conditions than the HCM merge or diverge methods. He also commented that guidance is needed for dealing with multiple weaving segments and overlaps between two on-ramps or an on-ramp and a weave, that the distances used to define influence areas seem arbitrary, and that capacity checkpoints for the ramps should be explicitly added to the weaving method, similar to the merge and diverge methods. X. Wang et al. (2014) compared the HCM 2010’s estimated capacities to field-observed capacities at two congested weaving segments in Edmonton, Alberta, using a combination of video camera and detector data. They found that the HCM and field capacities closely matched. Y. Wang et al. (2014) studied three uncongested weaving segments in El Paso using video recordings from the local traffic management center. They found that the HCM’s 2010 speed estimates reasonably matched the field-observed speeds. Location of Weaving Lane Changes Ahmed et al. (2019) analyzed NGSIM data for a weaving segment in Los Angeles with a 698-foot short length. Approximately 350 ft of additional length was available between a physical barrier and the start of the short length, and it was observed that many vehicles made lane changes in this area. About 50% of all lane changes were completed within the first 110 ft of the short length (including lane changes occurring before the start of the short length), 90% of

Background 21   freeway-to-ramp weaves occurred within the first 630 ft of the physical weaving area, and 90% of ramp-to-freeway weaves occurred within the first 1,060 ft of physical weaving area (that is, just past the end of the short length). Lee and Cassidy (2009) collected data using video cameras at two freeway weaving segments (one metered, one unmetered) in Southern California. They found that freeway-to-ramp move- ments were disruptive when they occurred close to the merge point (which tended to occur at low ramp-volumes) or when freeway-to-ramp volumes were high. They concluded that the number of lane changes influences discharge volume in addition to the spatial distribution (concentrations) of those lane changes. They developed a model for predicting where mandatory freeway-to-ramp lane changes would occur. This model includes three variables: the difference in densities between a driver’s lane and the auxiliary lane, the remaining distance to the diverge point, and the number of lanes that must be crossed to finish the weaving maneuver. Pesti et al. (2011) collected data at five weaving segments in Dallas and Houston using a com- bination of pneumatic tubes and video feeds from local traffic management center cameras. They found that lane-changing behavior was not uniform along the length of the auxiliary lane; drivers (both freeway-to-ramp and ramp-to-freeway) became more aggressive as they approached the diverge point. In addition, ramp-to-freeway vehicles accepted shorter gaps in the first 250 ft of a weaving segment than did freeway-to-ramp vehicles. Van Beinum (2018) collected data from a helicopter at eight weaving segments in the Netherlands and found that more than 85% of weaving lane changes took place in the first 50% of the available weaving length and more so under relatively low traffic flows (volume-to-capacity ratio less than 0.6). He and Menendez (2016) studied a three-lane, 535-m weaving segment in Basel, Switzerland. About 80% of all weaving lane changes and 30% of all nonweaving lane changes occurred in the first 100 m during congested conditions, with even higher percentages when noncongested. The maximum sustainable weaving lane-changing rate observed was 35 weaving lane changes per minute and 0.35 weaving lane changes per minute per meter over the first 100 m. When weaving demand exceeded this rate, some of the weaving activity was pushed downstream (see Figure 2). Marczak, Daamen, and Buisson (2014) studied a three-lane, 250-m weaving segment in Grenoble, France, with right-side on- and off-ramps connected by an auxiliary lane, the center lane being a mainline ramp connection to another freeway, and the left lane proceeding straight, eventually transitioning into an urban street. They found that ramp-to-freeway and freeway-to-ramp lane changes tended to occur at the same locations within the weaving segment at a given time (zipper Source: He and Menendez (2016). Figure 2. Distribution of weaving lane changes with different weaving flows.

22 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies effect). However, at higher speeds, freeway-to-ramp drivers changed lanes sooner than ramp-to- freeway drivers. One-half of lane changes occurred • Before or just after the start of the marked weaving area, when speeds were <5 m/s (11 mph); • Within the first 25 m, when speeds were 5−10 m/s (22 mph); • Within the first 50 m, when speeds were 10−15 m/s (34 mph); and • Within the first 75 m, when speeds were 15−20 m/s (45 mph). Under free-flow conditions, 95% of lane changes occurred within the first 150 m. Kusuma et al. (2015) collected lane-changing data in a five-lane (three mainline, two auxil- iary) weaving section on the M1 motorway southeast of Leeds, England. Of the two-lane lane changes observed, 14% were direct (the vehicle moved more or less directly to the desired lane), and 86% were staggered (the vehicle stayed in the new lane for at least 2.5 seconds before chang- ing lanes again). They also observed that lane changes sometimes occurred in groups as platoons (two or more vehicles making the same origin-to-target lane change simultaneously), weaves (two vehicles swapping lanes simultaneously or just after each other), or other (two vehicles from different lanes moving into the same lane just after each other or a following vehicle changing lanes in one direction just after the vehicle previously in front of it changed lanes in the opposite direction). About 15% of all lane changes occurred in groups, about 7% of all lane changes were weaves, and about 76% of lane changes occurred in the first 250 m of the weaving segment (only 320 m of the 1,265-m weaving segment could be observed). The inability to observe the entire weaving segment was not felt to be an issue because a previous study had found that 70% of lane changes occur in the first 250 m. Drivers making multiple lane changes were more aggressive with the first lane change than single-lane-change drivers. X. Wang et al. (2014) studied two weaving segments in Edmonton, Alberta (one of which was a right-to-left side weave involving three lane changes) and found that most initial lane changes occurred near the ramp gore. Tanaka et al. (2017) studied a 48-m weaving segment in Yokohama, Japan, and found that 80% of weaving maneuvers occurred in the first 100 m, leading to inefficient utilization of the remainder of the weaving segment. Sarvi, Zavabeti, and Ejtemai (2011) studied a 900-m weaving segment in Melbourne, Austra- lia, and a merge/diverge combination in Tokyo, Japan, with no auxiliary lane and 500 m between ramps. The behavior (acceleration and deceleration characteristics and gap searching and accep- tance) of a ramp-to-freeway vehicle under congested conditions was found to be influenced by five vehicles: the car in front in the auxiliary lane (if present), the first two cars in front in the freeway lane, and the first two cars behind in the freeway lane. Zhao et al. (2019) studied seven weaving segments in Nanjing, China, with road markings that required freeway-to-ramp traffic to change lanes prior to ramp-to-freeway traffic. A combina- tion of double (one dashed, one solid) or single (dashed) lines and arrows (diagonal or straight) indicated where a given lane was allowed to change lanes (see Figure 3). These segments were compared to standard weaving area markings allowing lane changes anywhere along the seg- ment length. Standard markings were recommended when the volume ratio was greater than 0.7, the diverge ratio was greater than 0.6, or the difference between the merge and diverge ratios was greater than 0.5. The best capacity benefit of the special markings was found to be at volume ratios between 0.4 and 0.6, with the difference between the merge and diverge ratios being around 0.3. Qi et al. (2018) created VISSIM models of two freeway weaving segments in Houston, includ- ing the next downstream on- or off-ramp. They tested the effect of continuing the auxiliary lane beyond the gore point by (1) tapering it immediately back into the right-hand mainline lane and

Background 23   (2) continuing it to the next ramp. They found small (≤1.0 mph) improvements in speeds in both cases but found that ending the auxiliary lane prior to the next ramp was more effective when the ramp had high traffic volumes. Location of Nonweaving Lane Changes Ahmed et  al.’s (2019) analysis of the Los Angeles NGSIM site determined that weaving vehicles made 0.6 discretionary lane changes on average, compared to 0.28 discretionary lane changes per nonweaving vehicle. Vehicles that made more discretionary lane changes experi- enced higher space mean speeds (4 to 8 mph) through the observation area, on average. Wang et al. (2015) evaluated a scenario where the nonweaving freeway lanes were designated for through traffic only, with no lane changes allowed between the nonweaving and weaving lanes within the weaving segment (through traffic could also use the weaving lanes). Under this scenario, the nonweaving freeway lanes could be analyzed as a basic segment, while the weaving lanes could be analyzed as a weaving segment with fewer lanes and a higher volume ratio. They tested different combinations of traffic flow rates, weaving segment short lengths, and volume ratios, and found small improvements in capacity (for example, from 9,000 to 9,200 pc/h on a freeway with four directional mainline lanes and an auxiliary lane). They concluded that this scenario was best suited for weaving segments with relatively low-volume ratios (for example, many mainline lanes relative to ramp lanes). Weaving Segment Length He and Menendez (2016) studied a three-lane, 535-m weaving section in Basel, Switzerland. The on-ramp approach was very long, shrank from two lanes to one, and was fed by three major streets, which could result in a relatively even supply of vehicles. They found that at the merge, densities were similar in the right-hand mainline lane and the auxiliary lane under all p.m. peak- period traffic conditions, which ranged from uncongested to congested. Densities in the left- hand lane were slightly higher; it was postulated that weaving and nonweaving lane-changing activity created voids in the right two lanes that reduced those lanes’ densities. The authors con- cluded that the speed and density homogenization at the merge meant that the weaving section Note: Lw = weaving segment length. Source: Zhao et al. (2019). Figure 3. Lane markings used in Nanjing, China, to regulate weaving maneuver locations.

24 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies length has limited effect on the weaving section’s capacity and operation. Speed and density homogenization caused the weaving section breakdown. At the diverge, in the absence of queue spillback from downstream of the weave, the density of the right-hand mainline lane was lower than at the merge. There was little interaction between the mainline and the auxiliary lane at the diverge, and the capacities of the right two lanes were lower at the diverge than at the merge. The authors suggested that the operation of the weaving section could be improved by encouraging drivers to use the whole weaving section to change lanes. Weaving Volume X. Wang et al. (2014) studied two weaving segments in Edmonton, Alberta, and found that weaving activity reduces capacity when the weaving ratio is moderate or large. Rudjanakanoknad and Akaravorakulchai (2011) studied a weaving section in Bangkok, Thailand, where two on-ramp lanes merged into a narrow, 3.0-m auxiliary lane that fed a two-lane off- ramp. They found that capacity was affected by upstream flow and on-/off-ramp demand, but the effect was indirect. Increases in ramp demand induced lane changes from slow to fast main- line lanes, while increases in mainline demand induced lane changes from fast to slow mainline lanes. The lane changes created changes in weaving capacity. On-Ramp Flow Characteristics Chilukuri, Laval, and Chen (2013) studied the effects of ramp-meter queue flushes on an on-ramp entering a weaving segment in suburban Atlanta. The capacity drop during near- or at-capacity conditions increased as the ratio of on-ramp flow rate during the queue flush to the freeway-flow rate increased. Queue flushes were observed to always produce densities over- capacity in the weaving segment. Marczak, Daamen, and Buisson (2014) found that on average, freeway-to-ramp drivers accepted larger gaps than ramp-to-freeway drivers in their study at a weaving segment; this result was attributed to the presence of larger gaps in the auxiliary lane due to the traffic signal control at the entrance to the ramp. Weaving Segment Influence Area No literature was found specific to the influence area of a weaving segment. See the earlier sections for literature findings on the influence areas of merges and diverges. Capacity Values Elefteriadou, Kondyli, and St. George (2014) recommended the planning values for prebreak- down and postbreakdown weaving capacity shown in Table 3, based on detector observations at eight sites in Florida. In comparison to merge and diverge capacities shown previously in Table 1, only values for urban sites are provided. Use of Simulation Pesti et al. (2011) found that VISSIM most accurately modeled weaving segments when the weaving segment was broken into smaller segments and different driver characteristics (relaxed, Area Type Condition 3 Mainline Lanes 2 & 4+ Mainline Lanes Urban Prebreakdown 2,200 pc/h/ln 2,100 pc/h/ln Queue discharge 2,000 pc/h/ln 1,900 pc/h/ln SOURCE: Elefteriadou, Kondyli, and St. George (2014). Table 3. Planning-level weave capacity values.

Background 25   normal, moderately aggressive, aggressive) were applied to those segments. Freeway-to-ramp movements began as “relaxed” closest to the merge gore point and transitioned to “aggressive” by the end of the weaving segment. Ramp-to-freeway movements began as “normal” closest to the merge gore point and transitioned to “aggressive” by the end of the weaving segment. Van Beinum (2018) compared calibrated VISSIM and MOTUS models to field observations of weaving segments. VISSIM overestimated the number of lane changes, while MOTUS under- estimated them; neither produced a realistic pattern of lane changes. Both models showed reason- able results in terms of gap acceptance. Alternative Models Proposed in the Literature Xu et al. (2020) developed a model to directly estimate space mean speed in ramp weave sections, based on field data from six sites in the United States. The space mean speed could then be used in combination with the fundamental speed–flow diagram to estimate the weaving seg- ment capacity. To do so, the analyst should find the point in the diagram where the density line for capacity (assumed to be 43 pc/h/ln, as per the HCM) intersects the estimated segment space mean speed. The reduction in speed due to weaving turbulence, or speed impedance (SIW), is the difference between the weaving segment speed at capacity and the speed of a basic segment with the same flow. Because capacity for basic segments occurs at a higher density (45 pc/h/ln) than for weaving segments, the basic segment speed will be slightly higher than a basic segment’s speed at capacity. (This characteristic also means that a weaving segment’s capacity will always be lower than a basic segment capacity, regardless of the flow characteristics.) LOS for a given weaving segment flow can then be determined by subtracting the SIW from the basic segment speed for a given flow and identifying the resulting density. The recommended model incor- porates FFS, segment short length, ramp-to-freeway flow rate, freeway-to-ramp flow rate, and total-entering flow rate as inputs. Like the HCM, the model underestimated speeds in the 57 to 63 mph range; however, the authors found that their model better estimated speeds for their data- sets than the HCM model. The authors also found that the model was much more sensitive to the weaving short length than the HCM model. Finally, the authors concluded that the model could be extended to merge and diverge segments. Heikoop and Henkens (2016) described the development of weaving capacity tables in the latest (4th) edition of the Dutch HCM. Field data obtained from license plate matching tech- nology was used to validate the FOSIM simulation software, which was then applied to different combinations of lane configurations, truck percentages, weaving lengths, and weaving traffic percentages by direction (freeway-to-ramp and ramp-to-freeway). In general, the Dutch capacity tables show weaving segment capacity decreasing as weaving percentage or truck percentage increases. Capacity is slightly sensitive to weaving length at low (<25%) weaving percentages and somewhat more so at higher weaving percentages. The Dutch HCM explains that most weaving maneuvers take place at the start of the weaving segment so that the added length has a relatively limited effect on capacity (Dutch Ministry 2015). Skabardonis and Mauch (2014) applied the Caltrans Level D method to three weaving seg- ments in San Diego. This method slightly overpredicted densities under low-volume condi- tions and underpredicted densities when densities were 30 to 40 pc/mi/ln. They also applied the Leisch method and found that it overpredicted densities under low-volume conditions. When densities were 30 to 40 pc/mi/ln, the Leisch method sometimes significantly underpredicted densities during a given 15-minute period and sometimes significantly overpredicted densities; however, the average of the predicted densities was reasonably close to the average of the field- measured densities. Marczak, Leclercq, and Buisson (2015) expanded previous work by Leclercq to develop a macroscopic model of freeway weaving that treats weaving segments as the combination of two merges and two diverges. Ramp and mainline traffic are assumed to share the available capacity

26 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies equally. The model was tested on a single three-lane weaving section in Grenoble, France, with reasonable results. Inputs to the model consisted of the FFS, the jam density, the wave speed in congestion, the length of the anticipation zone (the distance from where a driver decides to make a weaving maneuver and starts to change speed to where the maneuver actually occurs), the vehicle acceleration, and a relaxation factor (accounting for lane-changing vehicles that ini- tially accept shorter headways and gradually adjust their speed to obtain their desired headway from their leader). The capacity drop is most sensitive to the latter two factors. The capacity drop occurs at the start of the anticipation zone. Chen and Ahn (2018) used a combination of analytical modeling to quantify merging and diverging lane-changing effects and numerical simulation to quantify capacity drops for merges, diverges, and weaves with a one-lane mainline. They found that the weave capacity drop is mini- mized when diverging lane changes occur before merging lane changes and maximized when the opposite is true. The more balanced the merging and diverging flows, the lower the weave capacity drop. In the former case, voids left by diverging vehicles are filled by merging vehicles. The weave capacity drop increases as the merging and diverging flows become more imbalanced, the weave length decreases, and the merging lane-changing speed decreases. They compared their model’s results to field data reported by Marczak, Leclercq, and Buisson (2015) for the right- hand mainline lane of a weaving segment in Grenoble, France, and found the maximum error in their capacity-drop estimate to be 15%. Calvert and Minderhoud (2012) simulated several weaving segments with weaving lengths ranging between 100 and 1,000 m, 14 different lane configurations, and five different weaving ratios. Freeway speed limit (100 km/h) and percent heavy vehicles (8%) were held constant. For each weaving lane configuration, a base model was developed from the simulation results for a fixed weaving length of 500 m. CAFs were developed for each model to adjust the capacity for other weaving lengths. Comparing the model to field-measured capacities reported in studies at 14 sites in the Netherlands, the model was found on average to match field capacities but dif- fered by as much as ± 40% in estimating the capacities of specific sites. For two weaving section configurations that the authors felt to be a fair comparison to the HCM 2010, the proposed model differed by no more than 20% at any point from the HCM capacity estimate for a given weaving ratio. X. Wang et al. (2014) developed a lane-changing model to estimate weaving section capacity and tested it with field data from two sites in Edmonton, Alberta. The first site was a weave from the right to the left side with three lane changes, and the second site was a ramp weave. At site #1, the right and center lanes had similar capacities, but the center and left lanes experienced greater capacity drops. At site #2, the capacity drops in the center and left lanes were slightly less than in the right lane; the right lane also had the lowest flow rate and travel speed. Capacity drops between 5% and 20% were observed. Both the proposed model and the HCM 2010 model pro- duced capacities close to those observed in the field, but the proposed model could also estimate discharge capacities. The model’s inputs are the four origin-destination (O-D) flows within the weaving segment from which different demand ratios are determined (freeway-through volume as a percentage of freeway-entering demand, ramp-to-ramp volume as a percentage of ramp- entering demand, and weaving volume as a percentage of total-entering demand). Potentially Relevant Recent Freeway Basic Segment Research Some recent papers and reports on freeway basic segment operation may have relevance to freeway operation in the vicinity of ramps, both because they indicate future directions for basic segment analysis that may also have direct applicability to ramp analysis and because it would be desirable for ramp and basic segment analysis to be compatible in terms of approach.

Background 27   Sasahara, Elefteriadou, and Dong (2019) proposed a lane-by-lane analysis method for basic segments. They studied seven sites in Minnesota, Florida, Utah, and California, and found that lane distribution was sensitive to volume-to-capacity ratio, percent heavy vehicles, upstream and downstream ramp ratios, weekend conditions, and nighttime conditions. The latter two factors were significant in their model but had much lower impacts relative to other factors. They observed maximum throughputs as high as 2,500 vph in the median lane and as low as 1,200 vph in the shoulder lane and suggested capacity values of 2,045 vph/ln for four-lane freeways (that is, two lanes per direction); 1,847 vph/ln for six-lane freeways; and 1,812 vph/ln for eight-lane freeways. Units of vehicles per hour were used instead of passenger cars per hour because measurements were made in vehicles per hour and because percent heavy vehicles were a variable in the model. The largest model errors were for the shoulder lane (mean absolute error of 5.1 to 7.4 mph), which the authors hypothesized was because lane-by-lane heavy vehicle percentages were not used. The work was conducted as part of NCHRP Project 15-57, “Highway Capacity Manual Methodologies for Corridors Involving Freeways and Surface Streets,” which subsequently devel- oped a new Network Analysis chapter for the portion of the HCM that addresses interactions between ramp terminals and freeway mainlines. Other researchers [for example, Pan et al. (2016)] have noted different capacities and FFSs on a lane-by-lane basis in the vicinity of ramps due to dif- ferences in truck lane-by-lane distribution and differential speed limits between autos and trucks as well as the phenomenon of some freeway lanes being congested while others remain uncongested. Zhou, Rilett, and Jones (2019a, 2019b) studied the effect of differential speed limits and high truck percentages on rural freeway operations using a simulation of I-80 in western Nebraska calibrated with detector data. Even when a differential speed limit is not in effect, truck speed limiters installed for fuel economy and safety reasons may create different truck and auto FFSs. On four-lane freeways, a truck passing another at a low speed differential (for example, 1 to 2 mph) may create a moving bottleneck for which the HCM’s passenger car equivalency (PCE) factors do not account, as they were developed from freeways with three directional lanes. The authors found that PCEs, determined on the basis of equal capacity (as used by the HCM), increase as the percentage of tractor trailers increase (but at a decreasing rate); decrease as the heavy vehicle percentage increases up to 35% and increases slightly thereafter; and increase as the speed limit increases, the grade increases, and the percentage of the facility with truck lane restrictions increases. They found that mixed-flow HCM capacity estimates were 6% to 42% lower using the CAFs they developed from empirical Nebraska conditions, compared to HCM Chapter 26 values assuming equal truck and car FFS. The Oregon DOT’s (2019) Analysis Procedures Manual modifies the HCM’s FFS estimation procedure for all types of freeway segments to account for differential truck and car speed limits on rural freeways and portions of some urban freeways, as well as advisory or regulatory truck speeds on steep downgrades. Based on studies in Oregon, Idaho, Indiana, and Michigan, which found that the differences in truck and auto average and 85th percentile speeds were approximately equal to the difference in the posted speed limits, the truck FFS is estimated as the HCM-calculated auto FFS minus the posted speed differential. The FFS used for freeway analysis is then a weighted aver- age of the auto and truck FFS, based on the relative proportions of autos and trucks in the traffic stream. On steep downgrades, the truck FFS is set to the advisory speed or (when the advisory speed is based on truck weight ranges) a weighted average of the truck advisory speeds. Literature Synthesis and Knowledge Gaps This section presents the key findings and gaps in knowledge identified through the literature review, organized by segment type (merge, diverge, weaving, and basic freeway segments). The final subsection summarizes this information in two tables, one on key findings and the other on knowledge gaps.

28 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Merge Segments Prebreakdown Capacity and Breakdown Probability Effect of Ramp Volume There is good international agreement in the literature that prebreakdown merge capacity decreases as ramp demand increases (Aksoy and Öğüt 2018; Bates et al. 2013; Dehman and Drakopoulos 2013, 2016; Elefteriadou et al. 2011; Han and Ahn 2018; Kondyli, Gubbala, and Elefteriadou 2016; Morris et al. 2011; Reina and Ahn 2014; Wu and Lemke 2016). Effect of Number of Downstream Mainline Lanes Four studies (Bates et al. 2013; Elefteriadou, Kondyli, and St. George 2014; Kondyli, Gubbala, and Elefteriadou 2016; Morris et al. 2011) have found that per-lane capacities are lower on freeways with four (and in some studies, more) downstream lanes compared to three lanes. Elefteriadou, Kondyli, and St. George (2014) also found that per-lane capacities with two down- stream lanes were lower than for three lanes and similar to those for four or more lanes. Effect of Ramp Geometry Few studies have compared merge capacity with tapered on-ramps versus acceleration lanes, probably because most jurisdictions use a standard design and due to generally small datasets. Bates et al. (2013), in a study in the United Kingdom, discarded tapered ramps from their analysis and focused only on ramps with acceleration lanes; they found no significant difference in break- down probability due to ramp configuration or acceleration lane length. The German HBS iden- tified different capacity values for different ramp configurations, but the effect of acceleration lane length was small enough that it could be neglected (Wu and Lemke 2016). In a study in China, Shi et al. (2015) found that tapered on-ramps provided better capacity than short (for example, ≤ 75 m) acceleration lanes, that capacity increased as the acceleration lane length increased (up to 300-m total length), and that the capacity of tapered ramps and ramps with short acceleration lanes decreased as the ramp angle increased. Several studies have made observations of merging locations and have found that most drivers merge within the first half of the acceleration lane (Daamen, Loot, and Hoogendoorn 2010; Kondyli and Elefteriadou 2010). Lwambagaza et al. (2017) found that younger and middle-aged drivers tended to merge using the first half of the acceleration lane, while older drivers were more likely to use the last half of the acceleration lane. As merging speeds decrease (that is, as traffic volumes in the right-hand mainline lane increase), drivers use more of the acceleration lane (Daamen, Loot, and Hoogendoorn 2010; Lwambagaza et al. 2017). Overall, there is no consensus in the literature on the effect of on-ramp geometry on prebreak- down merge capacity. Effect of Ramp Flow Patterns and Merging Characteristics Ramp metering evens the arrival rate of merging vehicles. The studies of ramp metering included in the literature review were performed using simulation; these studies generally showed an improvement in prebreakdown throughput due to ramp metering (Duan et al. 2013; Li et al. 2018; Li and Ranjitkar 2013). Han and Ahn’s (2018) modeling work found that the breakdown prob- ability for a given combination of entering demands is lower when the arrival rate of merging traffic is more even. Yousif and Al-Obaedi (2011) found an improvement in upstream mainline capacity with greater ramp storage; more storage meant fewer incidences of queue flushes from the ramp meter. The amount of ramp demand that can merge onto the freeway, relative to mainline demand, may affect capacity. Rouphail et al. (2015) identified four main approaches to allocating flow between an on-ramp and the mainline during periods of breakdown that have been used in the literature and developed a model to predict the merge ratio based on the downstream demand

Background 29   flow per lane and the acceleration lane length per 10 mph of speed limit. Reina and Ahn (2014) predicted merge ratios under two scenarios: “fair share” (proportional to mainline versus ramp flow) and “zipper” (1 to 1). Fair share worked better for full lane adds, while zipper worked better where a lane was dropped. The validated models had relatively low errors but did not perform better than the ratio of mainline-to-ramp lanes in many cases. Given the range of approaches to allocating ramp demand in the literature, more work appears to be needed to either incorporate merging characteristics as a model input or obtain more evidence on which approach works best. Effect of Mainline Flow Characteristics Reina and Ahn (2014) reference eight studies that showed that flow distributions by mainline lane vary significantly between sites, under both congested and uncongested conditions. Their study of six congested freeway-to-freeway on-ramps in Southern California found different flow distributions at two sites with similar merging configurations—one had a downstream on-ramp, while the other had a downstream off-ramp. They developed models of lane flow distribution by lane based on upstream mainline flow and the presence of a downstream on- or off-ramp within 0.6 mi. Dehman and Drakopoulos (2013) also found that one of their four study sites had a much different flow characteristic in the right-hand mainline lane with considerable traffic leaving the lane prior to the on-ramp. Again, given the range of approaches to allocating mainline demand in the literature (per lane, right lane only, right two lanes only), more work appears to be needed. VSL can be used to increase the time headway between vehicles, thereby providing more room for merging vehicles. Knoop et al. (2010) compared vehicle lane distribution on a freeway with three directional lanes in the Netherlands with and without a 60 km/h VSL in effect. With VSL activated, more traffic used the right-hand freeway lane prior to an on-ramp than without VSL activated, resulting in more efficient use of the mainline lines. At the same time, more traffic in the right lane meant smaller gaps at the merge. Knoop et al. (2010) were not able to determine the overall effect on freeway operations due to the limitations of the detector data used in their analysis. Simulation studies (Duan et al. 2013; Zhang et al. 2013) indicate that VSL has the potential to improve throughput but is less efficient than ramp metering because of the need to post speeds in 5 mph increments (and 20 to 30 km/h increments in Europe). Capacity Values Table 4 summarizes empirical prebreakdown capacity values identified through the literature review, identifies whether they are based on 5-minute or 15-minute flows, and identifies the method used for defining capacity. The table also presents the HCM 6th Edition values for com- parison. All the capacities identified in the literature are lower than the HCM values. The HCM also identifies a “maximum desirable flow” in the RIA of 4,600 pc/h. Dehman and Drakopoulos (2016) found that 88% of breakdowns at one site and 43% of breakdowns at a second site occurred at volumes less than 4,600 pc/h in the RIA; they concluded that the HCM’s value may not be conservative enough for design purposes. Source 3-Lane Capacity 4-Lane Capacity Capacity Definition Method HCM 6th Edition 2,250−2,400 pc/h/ln 2,250−2,400 pc/h/ln 15-minute demands; capacity varies by FFS. Elefteriadou, Kondyli, and St. George (2014) 2,100 pc/h/ln 1,950 vph/ln 2,000 pc/h/ln 1,860 vph/ln 15-minute demands; 85th percentile of breakdown flows; 4- lane capacities also apply to 2 and 5+ lanes. Kondyli, Gubbala, and Elefteriadou (2016) 1,920−2,080 vph/ln 1,800−1,960 vph/ln 5-minute demands; regression on breakdown flows; capacity varies by the proportion of ramp volume. Morris et al. (2011) 2,020 vph/ln 1,870 vph/ln 15-minute demands; 95th percentile of breakdown flows; based on one site each. Table 4. Merge capacity values in the literature.

30 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Queue Discharge Flow Factors Affecting the Capacity Drop The developers of numerical models have suggested the following relations between various factors and the size of the capacity drop at a merge bottleneck: • Merging speed (Chen and Ahn 2018; Leclercq et al. 2016). • Merging length (Chen and Ahn 2018; Leclercq, Laval, and Chiabaut 2011). • Merging flow (Chen and Ahn 2018)—the effect being mostly limited to the first 100 to 200 vph of merging demand. In addition, Leclercq et al. (2016) found no relation between the merge ratio and the size of the capacity drop: higher merge ratios mean higher traffic and higher speeds on the ramp with corresponding less impact on right-lane speeds, while lower merge ratios mean less traffic but lower speeds on the ramp with corresponding greater impact on right-lane speeds. Queue Discharge Flow Values Table 5 summarizes empirical queue discharge flow values identified through the literature review and identifies the method used for developing the values. The table also presents HCM 6th Edition values for comparison, based on the average 7% capacity drop suggested in HCM Chapter 10. The queue discharge flow values identified in the literature are lower than the values obtained from the HCM. Elefteriadou, Kondyli, and St. George’s (2014) values represent an approximate 10% capacity drop. Influence Area Elefteriadou and Kondyli (2010) and van Beinum (2018) both studied on-ramp influence areas, and both defined the influence area as the area where lane-changing activity occurred that was influenced by the on-ramp. The two studies found that the influence area started 300 to 330 ft before the gore point and ended 840 ft (Elefteriadou and Kondyli) or 1,550 to 1,900 ft (van Beinum) after the gore point. The HCM defines the merge-influence area as starting at the gore point and ending 1,500 ft downstream. Diverge Segments The literature review found far fewer studies on the capacity and operations of diverge seg- ments compared to merging and weaving segments, possibly indicating that they are less com- mon than bottlenecks and that the factors controlling capacity are simpler than for the other two segment types. Prebreakdown Capacity and Queue Discharge Flow Elefteriadou, Kondyli, and St. George (2014) studied two ramp diverges (one with a lane drop) and a major diverge that experienced breakdown in Florida. They recommended using the same Source 3-Lane Throughput 4-Lane Throughput Definition Method HCM 6th Edition 2,090−2,230 pc/h/ln 2,090−2,230 pc/h/ln Based on 7% capacity drop suggested in Chapter 10; 15- minute demands; throughput varies by FFS. Elefteriadou, Kondyli, and St. George (2014) 1,900 pc/h/ln 1,860 vph/ln 1,800 pc/h/ln 1,760 vph/ln 15-minute demands; 85th percentile of discharge flows; 4- lane values also apply to 2 and 5+ lanes. Bates et al. (2013) 1,800 vph/ln 1,610 vph/ln 15-minute demands; 50th percentile of discharge flow. Table 5. Queue discharge flow values in the literature.

Background 31   prebreakdown capacity and queue discharge flow values as for merge segments (see Table 4 and Table 5 earlier). The German HBS provides different off-ramp and mainline capacities for dif- ferent diverge configurations (Wu and Lemke 2016). Sun, Ma, and Li (2015) studied three diverge bottlenecks in Shanghai, China. At one site, the diverge ratio was very high (0.80), and the queue discharge flow was found to be greater than the prebreakdown flow. At the other two sites with lower diverge ratios, queue discharge flow was less than prebreakdown flow. Both prebreakdown and queue discharge throughput decreased as the diverge ratio increased. Chen and Ahn’s (2018) numerical model of the capacity drop of a diverge from a one-lane mainline found that the capacity drop increases as the diverging flow increases (with most of the effect occurring with the first 100 to 200 vph), as the diverging length decreases, and as the diverging speed decreases. Pan et al. (2016) studied a major diverge in Southern California that experiences breakdowns. They found the need to model operations on a lane-by-lane basis because queue spillback from the right branch of the diverge into the center option lane of the diverge caused traffic headed to the left branch to pre-position themselves in the left lane to try to avoid the congestion. In addition, the capacity and FFS of the right-hand lane were found to be lower than that of the other two lanes, possibly due to differential truck and auto speed limits. The results of these studies indicate a potential need to conduct additional studies of the pre- breakdown and queue discharge throughput of different diverge bottleneck configurations, due to small study sizes in the literature. In addition, the results from Pan et al. (2016), as well as those of weaving studies described in the next section, suggest that using freeway-wide parameters for basic segments (for example, FFS or percent heavy vehicles) may not be appropriate when evaluating the operation of the rightmost one or two lanes. Influence Area The only study of diverge influence area found in the literature review was that of van Beinum (2018) in the Netherlands. Based on helicopter observations at three sites, he found that lane- changing activity in advance of the diverge was correlated to the standard positions of exit guide signs (1,200 and 600 m upstream of the diverge). About 85% of the exiting vehicles were already in the right-hand lane at the start of the observable area (1,200 to 1,500 m upstream of the gore) although this finding may also be due in part to the Netherlands’ keep-right-except-to-pass traffic laws. Based on the region where lane-changing activity was different from that of a basic segment, van Beinum determined that the off-ramp influence area began 400 to 600 m upstream of the ramp and ended 200 to 375 m downstream of the ramp. In comparison, the HCM considers the influence area to begin 455 m (1,500 ft) upstream of the gore and end at the gore. Weaving Segments Issues with the Current HCM Method The following key issues with the current HCM method were identified through the literature review: • The method underestimates speeds in the 50 to 65 mph range (Xu et al. 2020). • The method includes weaving length as an input, but performance measure outputs are insen- sitive to weaving length for practical purposes (Ahmed et al. 2019). • The method underestimates capacities at high weaving ratios (Skabardonis and Mauch 2014). • The method can indicate that volumes are less than capacity but produce densities greater than the density at capacity (Skabardonis and Mauch 2014). • Across the largest U.S. dataset included in the literature review, the method overestimated density by an average of 22% (Skabardonis and Mauch 2014).

32 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies • The method sometimes estimates that weaving speeds are greater than nonweaving speeds (Zhou et al. 2015). • The method can predict worse conditions than the merge or diverge methods with high pro- portions of entering or exiting traffic (Stanek 2014). • Field observations indicate that weaving capacity is higher than merging or diverging capacity, but the HCM predicts lower weaving capacities (Elefteriadou, Kondyli, and St. George 2014). • Guidance or methods are required for analyzing multiple weaving segments and locations with overlaps between weaving and merging/diverging segments (Stanek 2014). Weaving Operations Effect of Weaving Length The ability to model the effect of weaving length on weaving operations has been hampered by the lack of datasets large enough to provide a sufficient range of lengths to analyze. However, most existing methods in wide use (for example, HCM 6th Edition, Caltrans’ Level D, Leisch, Dutch HCM) for estimating weaving segment operations indicate some improvement in seg- ment speed and capacity as the weaving length increases. Recent research by Xu et al. (2020) also indicates that weaving operations are sensitive to weaving length. These methods vary in their degree of sensitivity to weaving length. For example, Ahmed et al. (2019) found that the HCM’s weaving speeds are sensitive to weaving segment length, while nonweaving traffic speeds have no sensitivity to weaving segment length. Because most traffic is typically nonweaving, increas- ing the weaving length has little impact on weaving segment operations, according to the HCM model. The Dutch HCM (Dutch Ministry 2015) shows little effect of weaving length on capacity at volume ratios below 25%. In addition, many of the weaving lengths addressed by the Leisch method fall beyond the range of the dataset on which the method is based (Roess 2008). In contrast, the German equivalent of the HCM (HBS) does not use weaving length as a variable for calculating weaving area capacity, as German field data indicated that weaving length had a minimal effect on capacity (Wu and Lemke 2016). In addition, much of the literature model ing lane-changing behavior in weaving areas (for example, Chen and Ahn 2018; Marczak, Leclercq, and Buisson 2015) indicates that because the great majority of lane changes occur very early along the auxiliary lane, additional weaving length provides little, if any, operations benefits, at least under breakdown conditions. The Dutch HCM (Dutch Ministry 2015) also makes this point. He and Menendez (2016) observed that the speed and density homogenization occurring between the auxiliary lane and the right mainline lane at the start of a weaving area in Basel, Switzerland, was the cause of the breakdown and that lengthening the auxiliary lane would not change this situation. It may also be that weaving length is a proxy for other factors that affect weaving capacity, such as ramp-to-freeway weaving speed or merging volume. For example, He and Menendez (2016) found a maximum sustainable lane-changing rate of 35 lane changes per minute in the first 100 m of the weaving area; once weaving demand exceeded this rate, some of the weaving activity was pushed further downstream. In addition, ramp geometry or ramp metering may cause vehicles to enter the weaving area at a slower speed than on other ramps, creating a need for more room to accelerate to obtain the desired weaving speed. Finally, the definition of weaving length varies from study to study. The HCM uses the “short length,” the distance available for weaving indicated by pavement markings, while other studies use the gore-to-gore length. Data from the Los Angeles NGSIM site indicate that several weav- ing maneuvers occurred prior to the gore point (Ahmed et al. 2019) or in advance of the “short length” as defined in the HCM. Effect of Weaving Volume Similar to several alternatives, primarily simulation-based models (Calvert and Minderhoud 2012; Dutch Ministry 2015; X. Wang et al. 2014), Xu et al.’s (2020) empirical study of six weaving

Background 33   sites found that prebreakdown capacity decreases as the weaving ratio increases. Rudjanaka- noknad and Akaravorakulchai (2011) also found that capacity is related to weaving volumes but that the effect is indirect; higher ramp-volumes induce more mainline lane changing from slow to fast lanes, while higher mainline volumes induce more mainline lane changing from fast to slow lanes. The lane-changing activity causes the capacity decrease. Chen and Ahn (2018) also found a relationship between weaving volume and capacity drop, but the effect was stronger when merging and diverging volumes were imbalanced. Marczak, Leclercq, and Buison (2015) modeled a weaving segment as two combinations of merges and diverges. Given that merge capacity shows similar relationships between ramp volume and flow characteristics as does weaving capacity, one possible approach to updating the HCM model would be to treat merges as a special case of a weaving segment with no diverging flow. Given that diverging lane changes mostly take place close to the merging gore point, the down- stream end of the weaving area could potentially be treated as a simple diverge. He and Menendez (2016) found little interaction between the mainline and the auxiliary lane at the diverge point of a congested weaving segment. Effect of Ramp Flow The findings related to ramp flow (for example, ramp metering) on merge capacity are likely also relevant to weave capacity. Specific to weaving, Chen and Ahn (2018) found a stronger capacity drop as merging speed decreased; a lower merging speed can be caused by high main- line traffic densities, but it can also be created by ramp metering or ramp geometry. Chilukuri, Laval, and Chen (2013) demonstrated the severe effects on weaving segment operation of queue flushes, where a large volume of vehicles is released at the same time from a ramp meter. Marczak, Daamen, and Buisson (2014) found that freeway-to-ramp vehicles accepted larger gaps than did ramp-to-freeway vehicles at their study site; this finding was attributed to traffic signal control at the ramp terminal creating larger gaps. On-ramp flow can be relatively uninterrupted and even (for example, freeway-to-freeway ramps, ramps fed by several sources, ramps with meters in operation) or relatively uneven (for example, ramps fed from traffic signals, queue flushes from ramp meters). The effect of flow evenness on merge and weave segment capacity and operation could explain some of the differ- ences observed between sites and studies. Ramp vehicle speeds at the merge point can be influenced by the ramp length, ramp metering, and ramp geometry. Differences in the ramp and mainline speeds could also explain some of the differences observed between sites and studies. Use of Simulation Both the literature on merging and weaving operations note that standard microsimula- tion models may not accurately model lane-changing behavior, even when calibrated to match average speeds or other parameters. Potential issues include merging/weaving vehicles accept- ing smaller gaps than ones previously rejected (Sun et al. 2018), a number of nonweaving lane changes (van Beinum 2018), and locations where lane changes occur. Pesti et al. (2011) found that VISSIM most accurately modeled weaving segments when the weaving segment was broken into smaller segments with different driver characteristics (relaxed, normal, moderately aggres- sive, aggressive) applied to these segments. Relevant Basic Segment Research Truck and car FFS can differ due to differential speed limits and the use of truck speed limiters (Oregon DOT 2019; Pan et al. 2016; Zhou, Rilett, and Jones 2019a, 2019b). Given that trucks tend to be concentrated in the right-hand lane, the use of an average freeway FFS to represent conditions in the right-hand lane in the vicinity of ramps may not be appropriate. In addition, the

34 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies FFS of the weaving and nonweaving lanes may be substantially different. Sasahara, Elefteriadou, and Dong (2019) propose a method for estimating volume distribution across lanes as part of NCHRP 15-57, which developed a new “Network Analysis” chapter for the HCM. Results of Literature Review The literature review identified several research results relevant to the project objective of updating and improving the intersection type consistency of the freeway merge, diverge, and weaving methodologies in the HCM 6th Edition. Table 6 highlights the relevant results. Table 7 highlights the gaps in the available research. HCM Methodologies This section provides an overview of the HCM 6th Edition merge, diverge, and weaving meth- odologies, highlighting • Data requirements and data collection methods, • Different geometric configurations covered by the methods, • Ability to evaluate freeway management strategies, and • Compatibility with the basic freeway segment method and with each other. Overview of the Current HCM Freeway Analysis Methodology The HCM freeway methodologies output the following performance measures: capacity, FFS, volume-to-capacity (v/c) ratios, space mean speed, average density, travel time [minute per vehicle (min/veh)], vehicle miles traveled, demand and volume-served), vehicle hours of travel, vehicle hours of delay, and automobile LOS for each component freeway segment and for the facility. Table 8 compares the HCM’s methodologies for estimating the performance of basic, merge, diverge, and weaving segments. All the methods rely on the basic freeway segment method in HCM Chapter 12 to estimate FFS. The weaving, merge, and diverge methods use different means to estimate speed. The merge and diverge methods assume the same capacity as a basic segment (subject to entry and exit flow checks), while the weaving method reduces basic segment capacity based on the weaving seg- ment’s traffic and geometric characteristics. The basic and weaving segment methods compute density from the fundamental relationship of volume divided by speed, while the merge and diverge methods compute density from regression equations. Required Data Table 9 lists the data needed to evaluate the full range of performance measures for HCM freeway facility and segment analysis. Individual performance measures may require only a subset of these inputs. FFS estimation using the HCM requires the following information about the facility’s geom- etry: lane widths, right-side lateral clearance, and the number of ramps per mile. Capacity esti- mation requires the FFS plus additional data on heavy vehicles, terrain type, number of lanes, peak-hour factor (the ratio of the average hourly flow to the peak 15-minute flow rate), and the driver population (that is, familiar or unfamiliar drivers). Once FFS and capacity have been calculated, then speed, LOS, and queue lengths can be estimated if additional information about segment lengths and the directional demand (vph) is available.

Capacity On-ramp flow rate, merge ratio, and ramp speed affect merge capacity. Merge capacity is adversely affected by higher ramp flows. There is a 50% probability of breakdown of ramp merge at 2,000 vph in the right lane. Pulsing of on-ramp flows degrades merge capacity; metering improves merge capacity. HCM’s 4,600 vph in the right two lanes at the merge may not be conservative enough for design purposes. Prebreakdown planning capacity is 2,000 pc/h/ln in urban areas, and queue discharge is 1,800 pc/h/ln and may be lower in rural areas (1,800 prebreakdown and 1,600 queue discharge). Fair share and zipper merging explained observed merge capacities during oversaturated conditions; each method is for a specific merging geometry. Diverge capacities are similar to merge capacities. Weaving Segment Geometry Many less common weaving geometries have not been studied in research. Other Inputs 50% of lane changes are completed in the first 110 ft. 90% of freeway-to-ramp lane changes are completed in the first 630 ft. 90% of ramp-to-freeway lane changes are completed in the first 1,060 ft. Ramp-to-freeway vehicles accepted shorter gaps than freeway-to- ramp vehicles in the first 250 ft. Speed HCM underestimates speeds, especially in the 50 to 65 mph range; HCM speed estimates are a good fit at outer ends (for example, <50 mph, FFS). HCM speed estimates are insensitive to weave length for practical purposes, despite weave length being a model parameter. Some studies have found HCM speed estimates to be reasonable for the limited number of sites included in these studies. HCM Methodology Aspect Results of Literature Review Ramp Merge/ Diverge Influence Area The merge-influence area starts 360 ft upstream and ends 840 ft downstream (total length = 1,200 ft) based on one study in Jacksonville, Florida. A Dutch study found a similar upstream distance (330 ft) but a longer downstream distance. A Dutch study found a longer upstream distance for diverges relative to merges (corresponding to guide sign positioning) and shorter downstream distances. There is no recent U.S. study on diverges to compare. Geometry There is little to no research on multilane ramps. The German HBS provides capacity values for a variety of ramp configurations and could at least provide an indication of relative capacities. Additional Inputs to Consider Mainline lane-by-lane flows vary by study site, presence of ramps, mainline, and ramp flows. Right lane flows are heavily influenced by percent trucks and truck speeds relative to auto speeds. Gap acceptance varies by location on the acceleration lane. On-ramp traffic rejects larger gaps initially before eventually accepting shorter gaps. Capacity drop may be due to drivers spreading out after accepting a short gap for merge. Most merges occur in the first 50% of the acceleration lane. The operational benefits of lengthy acceleration lanes decrease rapidly after the first few hundred feet. Speed See “Weaving Segment” later in this table. Density See “Weaving Segment” later in this table. Table 6. Summary results of literature review. (continued on next page)

36 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Other Inputs Lane-by-lane volume and analysis can now be performed. Lane distribution is sensitive to volume-to-capacity ratio, percent heavy vehicles, upstream and downstream ramp ratios, weekend conditions, and nighttime conditions. Speed FFS varies between trucks and passenger cars, even on flat terrain. Capacity Three-lane freeways have higher capacity per lane than freeways with two or more than four directional lanes. Capacities of 1,800 to 2,050 vph/ln, dropping to 1,200 vph/ln for the rightmost lane (effect of trucks and ramp turbulence). Density HCM appears to overpredict densities under low-flow conditions and underpredict densities at high-flow conditions. Four instances were presented in one study where predicted densities before breakdown exceeded 43 pc/mi/ln. Capacity HCM appears to underestimate capacity for high-weave ratios, while other studies have found HCM capacities to be reasonable for the limited number of sites included in those studies. Planning-level capacities: 2,100 pc/h/ln prebreakdown and 1,900 pc/h/ln queue discharge. Basic Segment Geometry No field research on facilities with more than six lanes in each direction. HCM Methodology Aspect Results of Literature Review Table 6. (Continued). HCM Methodology Aspect Gaps in the Literature Ramp Merge/Diverge Influence Area Advance exit guide sign placement is the only potential factor identified affecting the length of the RIA. Geometry Less common ramp configurations and multiple ramps not studied; no new research on multilane ramps. Additional Inputs to Consider Many suggested inputs may be more appropriate for a microanalysis than HCM. Evaluating some measure of the evenness of ramp flow would be potentially useful (both in terms of average ramp vehicle headways and the variability of those headways). Speed Little new research on speed effects of ramp merge/diverge to supplement evaluating models to directly estimate speeds as found in Rouphail et al. (2021). Density No new research found on density effects of ramp merge/diverge. Capacity Extensive research is available, some conflicting. Weaving Segment Geometry No new research found on less common and more complex weaving configurations. Basic Segment Geometry No new research was found on freeways with more than six lanes in each direction. Other Inputs New HCM Chapter 38 lane-by-lane method may be too complicated for a core HCM macroscopic method. However, given that the current HCM method already predicts the volume in the right two lanes, considering the right lane’s FFS and truck percentage separately from the freeway as a whole could improve the results. Lane-by-lane analysis could also be necessary for more complex geometries with higher-than-normal O-D demand in the right lane(s) and for major diverges. Less important when overall volumes are low, heavy vehicle percentage is low, and no truck–car speed differential. Speed Limited truck speed research. Table 7. Gaps in the literature.

Background 37   Performance Measure Basic Segment (HCM Chapter 12) Weaving Segment (HCM Chapter 13) Merge Segment (HCM Chapter 14) Diverge Segment (HCM Chapter 14) Free-Flow Speed (FFS) FFS estimated as function of base FFS, lane widths, right-side clearance, ramp density. FFS is an input. FFS is an input. FFS is an input. Capacity Capacity estimated as function of FFS. Local calibration enabled. Mixed volume converted into passenger car equivalents. Capacity is estimated based on equivalent basic segment capacity, the weave volume ratio, weave length, and number of weave lanes. Capacity may be adjusted for weather and incidents. Same as basic segment capacity, with merge area flow check. Same as basic segment capacity, with diverge area flow check. Speed Speed estimated as function of passenger car volume and FFS. Speed estimated based on the lane- changing rate, FFS, minimum speed. Speed computed from regression equations. Speed computed from regression equations. Density Density computed as ratio of passenger car volume to speed. Density computed as ratio of passenger car volume to speed. Density computed from regression equation (volumes, acceleration lane length). Density computed from regression equation (volumes, deceleration lane length). Table 8. Comparison of HCM basic, merge, diverge, and weaving methods. Performance Measure Input Data (units) FFS Cap Spd LOS Que Rel Default Value Lane Widths and Right- Side Lateral Clearance (ft) • • • • • • 12-ft lanes. 10-ft lateral clearance. Ramp Density (per mile) • • • • • • Must be provided. Percentage Heavy Vehicles • • • • • 12% (rural), 5% (urban). Terrain Type/Specific Grade • • • • • Must be provided. Number of Directional Lanes • • • • • Must be provided. Peak-Hour Factor (decimal) • • • • • 0.94 Driver Population Factor (decimal) • • • • • 1.00 (that is, familiar drivers). Segment Length (mi) • • • • Must be provided. Directional Demand (vph) • • • • Must be provided. Variability of Demand • Must be provided. Incident and Crash Frequencies • Must be provided. Severe Weather Frequencies • Must be provided. Work Zone Frequencies • Must be provided. Note: FFS = free-flow speed (mph), Cap = capacity (vph/ln), Spd = speed (mph), LOS = level of service (A–F), Que = queue (veh), and Rel = travel time reliability (several measures). Source: Dowling et al. (2016). Table 9. Required data for HCM freeway analysis to estimate various performance measures.

38 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Travel time reliability analysis requires the same data needed to estimate speeds, plus infor- mation on the variability of demand; the severity, frequency, and duration of incidents; the frequency of severe weather conditions; and the frequency of work zones by the number of lanes closed by duration. Geometric Congurations Covered e HCM divides the freeway facility into uniform segments that are analyzed to determine capacity, performance, and LOS. • Freeway weaving segments occur when a diverge segment closely follows a merge segment and ramp-to-freeway and freeway-to-ramp trac must cross paths, or when a one-lane o- ramp closely follows a one-lane on-ramp and the two are connected by a continuous auxiliary lane. • Freeway merge and diverge segments occur downstream of on-ramps and upstream of o- ramps unless weaving activity occurs. • Basic freeway segments are those segments that do not fall into the above categories. e beginning and endpoints of the weave, merge, and diverge segments are determined by their inuence areas, as shown in Figure 4. is segmentation method allows the coding of 90% of the geometric congurations present in the United States. e exceptions are • Part-time shoulder lanes, • Mainline toll booths and immigration checkpoints that interfere with the mainline ow, • Dynamic merge lanes that change the lane congurations at the merge, and • Closely spaced ramps (less than 1,500   apart) typical of urban freeway-to-freeway interchanges. Figure 5 illustrates a typical modern freeway-to-freeway interchange in an urban context with closely spaced on- and o-ramps. Challenging Ramp Geometries for the Current HCM Method e merge, diverge, and weaving methods have various limitations on the variety of geomet- ric conditions for which they are applicable, primarily because of the limitations of the original datasets upon which the methods are based. Ramp geometries that are particularly challenging Figure 4. HCM freeway merge, diverge, and weaving segment inuence areas.

Background 39   for the current HCM merge, diverge, and weaving methods are illustrated in Figure 6 through Figure 8 and include the following: 1. Close Merges – Close merges occur where two or more ramps fall within 1,500  of each other. e merge methodology may not adequately account for the concentration of volumes in the right lane(s). One work-around is to combine the volumes on the two ramps and evaluate the merge as a single ramp merge, but it is unclear what accuracy is lost with this approach. 2. Close Diverges – Close diverges occur where two or more o-ramps fall within 1,500  of each other. e current diverge methodology may not adequately account for the concentration of vol- umes in the right lane(s). One work-around is to combine the volumes on the two ramps and evaluate the diverge as a single ramp diverge, but it is unclear what accuracy is lost with this approach. 3. Lane-Add Merge – e current HCM merge method is not designed for the analysis of a long acceleration lane that exceeds 1,500  in length but does not qualify as a weaving segment. Section 4 (“Extensions to the Methodology”) of HCM Chapter 14 recommends evaluating the merge as a basic freeway section. is may be a satisfactory work-around, but there is no published analysis to conrm this approach. 4. Lane-Drop Merge – Same as the lane-add merge, the current HCM diverge method is not designed for decel- eration lengths that exceed 1,500  but do not qualify as a weaving segment. e recom- mended work-around is to evaluate the diverge segment as a basic freeway section. is may be satisfactory, but there is no published analysis conrming this approach. Source: Aerial imagery © 2019 Google. Figure 5. Example freeway-to-freeway interchange in an urban context.

40 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Figure 6. Geometries not well covered by current HCM Chapters 13 and 14 (#1 through #5 in list in text). (j) Merge on Five-Lane Freeways (k) Diverge off Five-Lane Freeways (h) Major Merge (i) Major Diverge 1,500 ft 1,500 ft (f) Two-Lane On-Ramps (g) Two-Lane Off-Ramps Figure 7. Geometries not well covered by current HCM Chapters 13 and 14 (#6 through #11 in list in text).

Background 41   5. Multiple Weaving Segment – Section 4 (“Extensions to the Methodology”) of HCM Chapter 13 recommends that when a series of closely spaced ramps create overlapping weaving movements, the mul- tiple weaving segments be decomposed into separate merge and diverge segments and a simple weaving segment. ere are no published data on the accuracy of this approach. 6. Two-Lane On-Ramps – Section 4 of HCM Chapter 14 describes adaptations to the merge methodology to evalu- ate two-lane on-ramps. It is limited to freeways with four or fewer lanes in the analysis direction. e method does not address lane additions that may be implemented with a two-lane on-ramp. ere are no published data on the accuracy of this approach. 7. Two-Lane O-Ramps – Section 4 of HCM Chapter 14 describes adaptations to the diverge methodology to eval- uate two-lane o-ramps. It is limited to freeways with four or fewer lanes in the analysis direction. e method does not address lane drops that may be implemented with a two- lane o-ramp. ere are no published data on the accuracy of this approach. 8. Major Merge – e HCM does not currently provide a methodology for evaluating major merges, other than checking the capacity of the entry and exit legs. 9. Major Diverge – e HCM does not currently provide a methodology for evaluating major diverges, other than checking the capacity of the entry and exit legs. Section 4 of HCM Chapter 14 pro- vides an equation for estimating the average density of vehicles in the major diverge inuence area. e equation is sensitive to the ow rate per lane and no other factors. 10. Merge On Five-Lane Freeway – Section 4 of HCM Chapter 14 provides a table to estimate the volume in the h lane. e analysis then proceeds with the standard method for the remaining volume in the remaining four lanes. e impact of this approach on the accuracy of the results is unclear. Figure 8. Geometries not well covered by current HCM Chapters 13 and 14 (#12 through #16 in list in text).

42 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies 11. Diverge Off Five-Lane Freeway – Section 4 of HCM Chapter 14 provides a table to estimate the volume in the fifth lane. The analysis then proceeds with the standard method for the remaining volume in the remaining four lanes. The impact of this approach on the accuracy of the results is unclear. 12. Left-Hand On-Ramps – Section 4 of HCM Chapter 14 recommends that left-hand ramps be analyzed as mirror images of right-hand ramps. A table of adjustment factors is provided to increase the left-hand lane flows by 0% to 20% over the equivalent right-hand ramp. There are no published data on the accuracy of this approach. 13. Left-Hand Off-Ramps – Section 4 of HCM Chapter 14 recommends that left-hand ramps be analyzed as mirror images of right-hand ramps. A table of adjustment factors is provided to increase the left- hand lane flows by 0% to 10% over the equivalent right-hand ramp. There are no published data on the accuracy of this approach. 14. Managed Lane Access Points – Section 4 of HCM Chapter 13 provides a method for evaluating the weaving between free- way on- and off-ramps and the access segments of barrier-separated managed lanes. It provides a means to estimate the capacity reduction effect of the weaving on the general- purpose lanes. However, the method does not recognize the operational effects of providing a buffer lane for acceleration and deceleration lanes between the managed lanes and the general-purpose lanes (see Figure 8). 15. Managed Lane Direct Ramps – Section 4 of HCM Chapter 14 suggests how the inputs to the standard ramp merge and diverge methods might be adjusted to reflect conditions unique to managed lanes. It is unclear whether the accuracy of this approach has been validated. 16. Collector–Distributor Roads – The HCM does not provide methods for evaluating merges and weaves within the collector–distributor roads within an interchange (See HCM Chapter 13, Section 4). Internal Consistency There are several situations where it is theoretically (and practically) desirable for the segment analysis methods to converge to agreement on performance: • A weaving segment with no weaving volumes (no on-ramp or off-ramp volumes) should theoretically have nearly the same performance and capacity as a basic freeway section. • A ramp merge segment with no on-ramp volumes should theoretically have nearly the same performance and capacity as a basic freeway section. • A ramp diverge segment with no off-ramp volumes should theoretically have nearly the same performance and capacity as a basic freeway segment. A comparison of the methods in HCM Chapters 12, 13, and 14 resulted in some preliminary conclusions regarding the convergence of the methods at low volumes (see Table 10). In most cases it will be necessary to test the methods with zero ramp-volumes and long weaving dis- tances to confirm these preliminary conclusions around the convergence of the methods at the limits of low ramp-volumes. Calculation Issues As discussed previously in this section, the following key issues with the current HCM weav- ing method have been identified in the literature: • The method underestimates speeds in the 50 to 65 mph range (Xu et al. 2020). • The method includes weaving length as an input, but performance measure outputs are insen- sitive to weaving length for practical purposes (Ahmed et al. 2019).

Background 43   • The method underestimates capacities at high weaving ratios (Skabardonis and Mauch 2014). • The method can indicate that volumes are less than capacity but produce densities greater than the density at capacity (Skabardonis and Mauch 2014). • Across the largest U.S. dataset included in the literature review, the method overestimated density by an average of 22% (Skabardonis and Mauch 2014). • The method sometimes estimates that weaving speeds are greater than nonweaving speeds (Zhou et al. 2015). • The method can predict worse conditions than the merge or diverge methods with high pro- portions of entering or exiting traffic approaching pure merges and diverges (Stanek 2014). • Field observations indicate that weaving capacity is higher than merging or diverging capacity, but the HCM predicts lower weaving capacities (Elefteriadou, Kondyli, and St. George 2014). • Guidance or methods are required for analyzing multiple weaving segments and locations with overlaps between weaving and merging/diverging segments (Stanek 2014). Modeling Freeway Management Strategies There is a wide variety of transportation system management and operations strategies for freeways: • Work zone management. • Traffic incident management. • Special event management. • Road weather management. • Transit management. • Freight management. • Traveler information. • Ramp management. • Congestion pricing. • Active transportation and demand management. • Integrated corridor management. • Access management. • Connected and automated vehicle deployment. Performance Measure Basic Segment (HCM Chapter 12) Weave Segment (HCM Chapter 13) Merge Segment (HCM Chapter 14) Diverge Segment (HCM Chapter 14) Free-Flow Speed Basis for all chapters. Identical to Chapter 12. Identical to Chapter 12. Identical to Chapter 12. Capacity Basis for comparing convergence at zero ramp-volume. Unlikely to converge to basic segment capacities for zero ramp- volume. Identical to Chapter 12. Identical to Chapter 12. Speed Basis for comparing convergence at zero ramp-volume. Unlikely to converge to basic segment speed at zero ramp-volume. Unlikely to converge to basic segment speeds at zero ramp-volume. Unlikely to converge to basic segment speeds at zero ramp-volume. Density Basis for comparing convergence at zero ramp-volume. Unlikely to converge to basic segment densities for zero ramp- volume. Unlikely to converge to basic segment densities at zero ramp- volume. Unlikely to converge to basic segment densities at zero ramp- volume. Table 10. Convergence of basic, weave, merge, diverge methods at zero ramp-volumes.

44 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies Each of these freeway management strategies may employ one or more managed lane tactics. There is a wide variety of these tactics that can be classified by their salient characteristics: • Pricing. • Eligibility. • Access control. • Traffic management technology. More sophisticated managed lane tactics provide facility managers with increasing flexibility and control. The added flexibility comes at the price of increased complexity (see Figure 9). The HCM is a planning and design manual, focused on predicting future facility performance as a function of future traffic demand, facility design, and general facility management strategies that may be enabled by various design options. At this point in time, the HCM is not focused on providing real-time operating control setting recommendations. Consequently, freeway man- agement strategies that focus on real-time control settings for real-time operations were not considered further by this project. Design (as long as design standards are maintained) does not significantly affect real-time management decisions. Management strategies that are related to temporary specialized conditions are not significantly affected by design decisions (as long as design standards are maintained) and are best treated in their own specialized guidebook. Consequently, strategies such as work zone management, traffic incident management, special event management, and road weather management were not considered further by this project. Finally, this project’s objectives focused on the HCM freeway methodologies related to planning and design for recurring congestion. Improvements to the HCM’s reliability analysis methodol- ogy for nonrecurrent congestion was beyond the scope of this project. Source: Federal Highway Administration, Managed Lanes, A Primer, FHWA-HOP-05-031, 2005. Figure 9. Managed lane options.

Background 45   After removing the real-time options, the following freeway management and managed lane tactics remain. It would be desirable for the HCM freeway methodologies to better address: • Ramp Management – Static time of day, local dynamic, system dynamic. • Integrated Corridor Management and Active Transportation and Demand Management – Speed harmonization (dynamic speed limits or advisories.) – Express (toll) lanes, high-occupancy toll (HOT) lanes, and high-occupancy vehicle (HOV) lanes. – CAV lanes. – Bus-only lanes. – Truck-only lanes. – Interchange bypass lanes. – Reversible median lanes. – Part-time shoulder use. Table 11 assesses the suitability and appropriateness of the current HCM freeway analysis methodologies for evaluating the performance of these management strategies and managed lane tactics. Summary Practitioners at state DOTs and their consultants frequently encounter difficulties when attempting to apply the freeway segment analysis methods in the HCM 6th Edition to real world problems. Their problems can be summed up in five points: • The HCM methods are not sensitive to modern freeway operations practices. • There are discontinuities between the HCM segment methods that cause illogical results. Management Tactic Basic Segment Weave Segment Merge/Diverge Segment Ramp Management NA Easy: User inputs metered rate. Easy: User inputs metered rate. Speed Harmonization Easy: User inputs FFS adjustment factor. Easy: User inputs lower FFS. The regression equations are not adaptable. HOV, HOT, Express Lanes Procedures already in HCM. Procedures already in HCM but could use elaboration and testing. Limited procedures in HCM for direct ramps. CAV Lanes No explicit procedures.* No explicit procedures.* No explicit procedures.* Bus-Only Lanes No explicit procedures; TCQSM has bus-only lane procedures. No explicit procedures. No explicit procedures. Truck-Only Lanes No explicit procedures. No explicit procedures. No explicit procedures. Interchange Bypass Lanes No explicit procedures. No explicit procedures. No explicit procedures. Reversible Median Lanes No explicit procedures. No explicit procedures. No explicit procedures. Part-Time Shoulder Use No explicit procedures; recent FHWA research. No explicit procedures. No explicit procedures. Note: *CAFs for CAVs have been added in the HCM 7th Edition. TCQSM = Transit Capacity and Quality of Service Manual. Table 11. Assessment of HCM methods for evaluating freeway management investments.

46 Update of Highway Capacity Manual : Merge, Diverge, and Weaving Methodologies • The HCM’s merge/diverge methods violate the fundamental relationship between flow, density, and speed. • The HCM’s capacity guidance for merge and diverge sections is poor. • The HCM’s methods are silent or inapplicable to some design options, including lane adds and lane drops, interchange types (diamond versus cloverleaf), system interchanges, and collector– distributor systems. The freeway merge and diverge methodologies in HCM Chapter 14 were developed more than 25 years ago and are not sensitive to modern active transportation and demand manage- ment strategies such as ramp metering, dynamic junction control, speed harmonization, and dynamic part-time shoulder use. HCM Chapter 13 weaving segment analysis was updated more recently; however, the relation- ship of the weaving methodology to the merge and diverge methodologies has not been clearly addressed in the HCM 6th Edition. This discontinuity can lead to illogical results, where adding an auxiliary lane between a pair of on- and off-ramps (thereby shifting from a merge/diverge analysis to a weaving analysis) can cause the HCM to predict a worse LOS. In addition, the freeway ramp merge and diverge methodologies, by providing independent and distinct regression equations for predicting flow, speed, and density, do not conform to the fundamental relationship of traffic flow, namely that flow is the product of speed and density. The HCM’s capacity guidance for ramp merge and diverge sections relies on basic segment capacities, not acknowledging the turbulence effects in these sections created by lane changing and merging. Particularly, multiple weaving segments, which the HCM 6th Edition recommends separating into a series of merge and diverge segments, do not account for the turbulence caused by weaving between overlapping O-D pairs. This lack of detail makes the merge/diverge and weav- ing methods challenging to apply because transportation agencies recognize that actual bottleneck capacities within their jurisdiction are often far from an “ideal” capacity of 2,400 pc/h/ln. The HCM freeway merge/diverge methods offer general guidance, but do not offer specific methodology for lane drops or additions, which often occur in the vicinity of freeway merge/diverge segments. HOV bypass lanes on the on-ramps are also not addressed. Finally, the HCM 6th Edition methods are not sensitive to the possible operations impacts of CAVs. The effects of CAVs were the subject of a multistate, pooled-fund study ongoing at the time the literature review was conducted. That project developed CAFs for basic freeway, merge, diverge, and weaving segments that have been incorporated into the HCM 7th Edition. These CAFs are cali- brated to three sets of base capacities: 2,400 pc/h/ln, 2,100 pc/h/ln, and 1,800 pc/h/ln.