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.
80 C h a p t e r 5 Introduction The objective of this chapter is to describe in detail the factors that cause congestion, with the specific intent of helping agencies respond cost-effectively to reduce the formation of congestion. The results of a series of analyses that examined the causes of freeway congestion, first in Atlanta, then in greater detail in the Seattle metropolitan region, are dis- cussed. The analyses were based on an entire yearâs worth of freeway operations data that covered a significant portion of freeways in the two regions. The freeway performance infor- mation was combined with data that described when inci- dents, accidents, and construction activity occurred and tracked the effects of weather. The effects of a variety of spe- cial events in Seattle were also tracked. The analyses did not include an examination of ramp delays, either entering (ramp meters) or exiting (queuing due to inadequate ramp intersec- tion capacity) the roadway. Many analyses have been performed over the years to examine the causes of roadway delay (Table 2.5). Tradition- ally those studies have been based on (a) queuing analysis of specific incidents; (b) simulation of specific roadway corri- dors, given a limited set of volume conditions and incident and nonincident conditions; and (c) national scale estimates based on base roadway volumes and reported incident and crash rates. preliminary Look at Congestion by Source: atlanta A simple analysis was undertaken in Atlanta to develop a point of comparison for the detailed Seattle analysis. The times and locations of incidents and weather conditions dur- ing the Atlanta study section peak periods were merged with the traffic data. Any incident that started 15 minutes before the peak start or lane-blocking incidents that started an hour before the peak start were assumed to influence the traffic flow and were counted. Each peak period was assigned an influencing cause: incidents, weather, or both. No attempt was made to track incident-caused queues in time and space; if an incident occurred at any time or location during the peak, the entire peak was described as incident influenced. This assumption will overstate the importance of incidents as a contributor to total congestion. Overall, the recurringânonrecurring split was roughly 50â50 (Table 5.1). A breakdown of nonrecurring incidents appears in Table 5.2; the significance of incidents is clear, as roughly a third of the congestion occurred on days when inci- dents occurred. Figure 5.1 examines congestion causes for the 50 worst congestion peak periods on these sections (i.e., those with the highest Travel Time Index [TTI]). Another potential source of congestion, high demand, was added to incidents and weather (high demand was defined as days with demand vol- umes higher than the average, plus 5%). For simplicity, the three sources were placed in a hierarchy, and only one source was assigned responsibility: incident, weather, or high vol- ume, in that order. For example, if a day had at least one inci- dent and high volumes, the cause was assigned as incident. Even with the addition of high demand, 21% of the days could not be assigned to a source. Several potential sources may explain these conditions: ⢠Congestion that forms off section and spills back into the study section, which could be from a downstream section or an exit ramp to either a surface street or an intersecting freeway; and ⢠Minor perturbations in traffic flow at a microlevel, which could be brief surges in demand or variations in driver behavior that cause flow breakdown when volumes are operating very near to physical capacity. Estimating Congestion by Source: The Cause of Congestion
81 a Closer Look at Congestion by Source: Seattle Background Analysis Overview To examine some of the issues raised in the preliminary Atlanta analysis, a detailed analysis was conducted using data from Seattle. This effort used measured roadway perfor- mance data (volumes and travel times taken every 5 minutes) for an entire year on approximately 120 centerline miles of urban freeway. These data included all crashes that occurred on those roadway segments, all noncrash incidents to which Washington State Department of Transportation (WSDOT) personnel responded, and National Oceanic and Atmo- spheric Administration (NOAA) weather data for the region. Based on these data, the analysis examined how a wide variety of factors affected travel times experienced by travelers on different freeway sections throughout the Seattle metropoli- tan region. Unlike traditional queuing analysis, using segment- based travel times over defined roadway segments as the dependent variable allowed the research team to explore the upstream and downstream impacts of a wide variety of dis- ruptions, as well as to examine the effect of those disruptions on travel time reliability. The primary intent of this section is to explore the causes of congestion on the instrumented Seattle freeway system and summarize those findings in a generalized manner so that the results are applicable elsewhere. Table 5.1. Recurring Versus Nonrecurring Congestion During Peak Period in Atlanta (2008) Congestion Type Nonrecurring Recurring Section No. of Incidents Congestion (%) No. of Incidents Congestion (%) I-75 northbound from I-285 to Roswell Road 128 52.0 118 48.0 I-75 southbound from I-285 to Roswell Road 81 41.8 113 58.2 I-285 eastbound from GA 400 to I-75 89 46.8 101 53.2 I-285 westbound from GA 400 to I-75 126 56.5 97 43.5 I-285 eastbound from GA 400 to I-85 159 64.6 87 35.4 I-285 westbound from GA 400 to I-85 134 56.5 103 43.5 I-75 northbound from Roswell Road to Barrett Parkway 121 49.2 125 50.8 I-75 southbound from Roswell Road to Barrett Parkway 100 42.3 136 57.6 Total 938 51.6 880 48.4 Table 5.2. Congestion by Source During Peak Period in Atlanta (2008) Source Congestion (%) Recurring (bottleneck) 48.4 Incidents 32.8 Weather 11.1 Incidents and weather 7.7 Figure 5.1. Congestion causes for the 50 worst congested peak periods in Atlanta (2008).
82 Factors Affecting Congestion Given that congestion occurs when there is too much vol- ume and too little roadway capacity, it can be said that all congestion is caused by having too much traffic volume. In some cases, too much volume is associated with routine tem- poral fluctuations in demand, such as peak period commute congestion in urban areas. In other cases, congestion is asso- ciated with demand associated with special events, such as sports or cultural activities. In still other cases, analysis sug- gests that microscale variations in demand during periods of already high demand can cause congestion even when hourly volumes would not indicate that capacity has been reached. However, traffic engineers know that roadway capacity is not a constant. A variety of factors reduce effective or opera- tional roadway capacity from the normal capacity figures that are computed with Highway Capacity Manual procedures. These factors can cause congestion even when volumes are lower than normal, theoretical roadway capacity. It is commonly accepted that there are a limited number of basic factors that cause congestion to form; these are usually referred to as the seven sources of congestion: 1. Traffic incidents; 2. Weather; 3. Work zones; 4. Fluctuations in demand; 5. Special events; 6. Traffic control devices; and 7. Bottlenecks or inadequate base capacity. Traffic incidents (including crashes, debris on the roadway, and other types of incidents) decrease effective capacity either by physically blocking lanes or by producing visual distractions that cause motorists to slow, resulting in lowered roadway throughput. Weather has similar effects on effective roadway capacity. Poor weather causes drivers to drive more cautiously, slowing down and leaving more space between vehicles to maintain safety, thus reducing effective roadway throughput. Work zones narrow lanes or reduce the total number of lanes available. They also can reduce speed limits and fre- quently include right- or left-lane shifts. All these physical changes decrease available or effective roadway capacity. Fluctuations in demand cause congestion because demand that exceeds roadway capacity causes queuing to occur, and that queuing reduces effective vehicle throughput. Thus, the arrival rates (timing) with which vehicles access a roadway segment is another cause of congestion. In a simple example, a two-lane (one-direction) freeway has a capacity of 4,000 vehicles per hour (vph). In a 3-hour period, 11,000 vehicles need to use that facility. If that demand is uniformly distributed, no congestion occurs, as volume never exceeds 4,000 vph. However, if demand arrives at the roadway section in the form of 2,200 vehicles in the first hour, 5,000 in the second hour, and 3,800 in the third hour, congestion will occur in the second hour. That congestion will cause queuing that will, effectively, further reduce roadway capacity, creating delays even in the third hour, despite the fact that demand is then lower than theoretical capacity. Special events cause congestion because they create signifi- cant fluctuations in demand. The starting and ending times of major events create surges in traffic demand that overwhelm roadway capacity near the event venue, causing congestion. Traffic control devices (e.g., traffic signals) delay some vehicles to allow other vehicles to move safely. Therefore, by definition, traffic control devices create (control) delay. When optimally timed, traffic control delays minimize congestion. When not optimally timed, traffic control devices create unnecessary delays to vehicles. Inadequate base capacity and bottlenecks create delay in the same way that traffic volume fluctuations cause delay. Inade- quate base capacity (i.e., not enough roadway capacity for nor- mal traffic flows) most frequently manifests itself at points along a segment of roadway where effective capacity is routinely lowestâa bottleneck. Bottlenecks are a decrease in effective roadway capacity that occur as a result of some physical change in roadway geometry or environment (e.g., a lane drop, a weav- ing section). That geographic location becomes the initial point at which traffic demand first exceeds effective capacity, causing queuing, which further decreases effective capacity. As the above discussion indicates, two of the causes of con- gestion (fluctuations in demand and special events) influence the demand side of the volumeâcapacity relationship, which ultimately determines formation of congestion, and the other five influence the actual volume-carrying capacity of the road- way. The cause of congestion has significance to transporta- tion agencies, in part, because it describes the level of control the agency has over that measure, and consequently the level to which it can anticipate and mitigate congestion formation. For example, the agency has no control over weather; it can only react to weather events. But the agency can directly influence other causes, such as the operation of traffic control devices or the design and timing of work zones. Data Description Traffic Incidents Data on traffic incidents were obtained from WITS, the WSDOT incident response program resource management system data- base, and the State of Washingtonâs accident reports. The more detailed and useful data source is the WITS data- base, which was created to track the work performed by WSDOTâs freeway service patrol personnel. Key variables for each task performed by WITS field staff are recorded, giving
83 WSDOT a record of when an incident was reported (used as an estimate of when that event occurred), as well as when the incident respondent declared the site of the incident cleared. The location (route, milepost, and direction) of the incident and whether a lane of traffic was blocked by the incident are also reported. Although these data allow detailed analysis of different incident types, this project limited the analysis to (a) when and where an incident occurred, (b) how long that incident lasted (in seconds), and (c) whether that incident closed a lane. In 2006 WITS reported only WSDOTâs incident response team actions, so no records exist for incidents to which WITS personnel did not respond. Because most WITS staff work during the peak commute periods, many incidents occurring on weekends or at night are not reported in WITS. This is a limitation of this analysis database. Accident records were used to supplement the WITS data. Accident records should be present for all significant acci- dents that occurred within the study area. During peak peri- ods, accident records generally match with WITS records, as WITS members are usually called to the scene of accidents when they are on duty. In a number of instances accident records and WITS records appeared to reference the same event but listed slightly different starting times. This project did not try to identify which of these times were correct, but kept both, and used the time related to a particular kind of event. That is, for an analysis of crash effects, the time from the accident record was used. If the analysis concerned the effects of all incidents, then the time noted in the WITS data- base was used. During times when WITS was not actively patrolling or on the rare occasions when WITS staff were busy on other calls and did not respond to an accident scene, the accident records indicate the occurrence of the accident but not the duration of the disruption. This is another limitation of this analysis database. Weather The weather data used for these analyses were obtained from publicly available records collected from the NOAA weather station at Sea-Tac International Airport. The analytic data- base created for this study tracked the major statistics reported by NOAA, including the following weather information: ⢠Visibility 44 Up to 10 miles; ⢠Temperature 44 Dry bulb; ⢠Wind speed 44 Average speed, and 44 Gust speed (highest gust speed that hour); ⢠Precipitation 44 Inches; and ⢠Weather type 44 Rain, 44 Mist, 44 Thunderstorm, 44 Drizzle, 44 Haze, 44 Snow, 44 Freezing, 44 Small hail, 44 Hail, 44 Ice pellets, 44 Squall, and 44 Fog. These data were too detailed for the basic analyses intended for this study. Consequently, the project team performed an extensive analysis to determine the types of summary weather statistics that would effectively indicate whether weather con- ditions contributed to congestion. A summary of these tests is given in Appendix E, and findings from the most important tests are presented later in this chapter. The outcome of the analysis was to define the indicator of bad weather as any time period in which any measurable precipitation had fallen at some time in the previous hour. Importantly, the use of this indicator discounts several weather effects, including wind, fog, snow, and rainfall intensity. An analysis of the effects of wind on roadway performance indicated that on the two roadways (I-90 and SR 520) that cross Lake Washington on floating bridges, high winds (gusts above 20 mph) had an observable effect in moderate volume conditions, especially eastbound when the winds caused waves to crash against the bridge, creating significant spray. (Winds are generally from the south, so the spray affects the eastbound traffic more than westbound traffic.) However, wind appeared to have little observable effect on the other freeway corridors in the region. The analysis of the effects of fog was problematic, as fog tends to be localized. Thus, while the airport could be very foggy (to the point that landings and take-offs are restricted for lack of visibility), at the same time I-5, passing within 2 miles of Sea-Tac, could have clear visibility. As a result, a fog variable was not useful in identifying specific fog-related delays. The examination of fog as a weather variable highlighted the problems associated with using weather data from a sin- gle point to represent weather experienced around a fairly large geographic region. That is, although the Sea-Tac weather records accurately reflect conditions at the airport, the weather experienced simultaneously in other areas of the metropolitan region can be different. For example, a storm moving south to north that affects Sea-Tac at 5:00 p.m. will
84 have occurred in the southernmost roadway sections before 5:00 p.m. and in the northern part of the city some time after 5:00 p.m. In addition, that storm may have dropped exactly 0.25 inch of rain at the airport, but it may have deposited only 0.1 inch south of the airport, and 0.5 inch in areas north of the airport. Although these rain data provide a reasonable estimate of weather conditions, they cannot be used as a pre- cise, highly accurate measure of the actual weather occurring on any given segment of roadway during a specific 5-minute interval. In addition to the basic time and geographic problems noted above, the snow and rainfall intensity variables pre- sented a second problem in that many of the effects of heavy rain (i.e., heavy rain short of intense thundershowers, which rarely happen in Seattle) occur after the precipitation has fallen. This is especially true for snowfall, as the effects of fall- ing snow are not nearly as significant as the effects of snow accumulations on the ground, depending on the amount remaining on the roadway. For example, snow flurries have little effect on driving, but 4 inches of snow on the ground 2 hours after the snow has stopped falling has a major impact on roadway performance. Another issue associated with snowfall in the Seattle area arose from the combination of how rarely snow falls in the region and how travel times are computed. When snow falls (and sticks), Seattleites tend to avoid driving whenever pos- sible. The region does not use salt; agencies do not clear snow as effectively as those in regions of the country that routinely experience snowfall; and snow is frequently turned into sheet ice on the roadways by cars that do travel, making the areaâs hilly terrain dangerous. The result is that a large percentage of travelers simply avoid going out. Therefore, after snow falls, volume and lane occupancy are frequently low on the free- ways despite the slow speed of those cars that are present. However, the loop detector system only sees low volumes and occupancy values, and can thus overestimate the speeds at which the vehicles are moving. Fortunately for this study, the number of days on which snow fell during the analysis year was small. Work Zones To identify work zones, variable messages sign (VMS) logs were examined. From the VMS logs, it was possible to iden- tify where, when, and for what period work zone messages were posted. It also was possible to determine from the logs when lanes were closed, but the number of lanes closed for a given construction lane closure was not incorporated into the analysis database. The closure times recorded in the VMS logs are approximate (e.g., 9:00 p.m. to 5:00 a.m.) and do not represent the exact time when lanes were actually closed or open to traffic. Long-term construction changes (e.g., narrowed lanes dur- ing lengthy construction projects or the presence of construc- tion barrels on shoulders in and approaching a work zone) that are likely to also cause minor disruptions in normal traffic flows are not included in the VMS database. However, because the freeways examined were major urban highways, all work zones had nighttime and weekend closures. No lanes were closed during normal weekday business hours. Fluctuations in Demand Volume data for the study were obtained from FLOW, the WSDOT Northwest Regionâs traffic management center data- base system. All traffic volume data used in the study were collected with permanent inductive loops that are part of that system. Loops are located roughly every half mile on the free- ways analyzed. Each loop reports total volume every 5 min- utes, as well as average lane occupancy for that location. Because 5 minutes is the basic WSDOT data-reporting period, the analyses for this report were based on these 5-minute periods. Traffic volumes were available every 5 min- utes, for every roadway study segment, for all 365 days for 2006. Some corridors were missing specific days or times of data because of equipment malfunction. Because volumes varied over the course of the roadway study segments, several volume statistics were used to describe each 5-minute period for each roadway segment. These are ⢠The maximum volume observed for the roadway segment in that 5-minute interval; ⢠The minimum volume observed for the roadway segment in that 5-minute interval; ⢠The average volume over the length of the segment; ⢠The vehicle miles traveled for the segment; and ⢠The vehicle hours traveled for the segment. Volumes were reported in units of vehicles per hour. Special Events Some special event data were collected by manually reviewing calendars for major regional venues (e.g., the Seattle Mari- nersâ game schedule allowed the researchers to identify the dates and start times of Mariner baseball games in 2006). However, it quickly became apparent that collecting uniform special event data would not be possible. In part, this was because there is no uniform definition of how big an event must be to be classified as a special event. Major league base- ball games with 30,000 people attending undoubtedly qualify, but do major college basketball games with 8,000 people attending? What about games with 2,500 people? Although all major sporting events have known start times, many
85 (e.g., baseball) do not have consistent durations, and their ending times are not easily determined. The lack of a defi- nite duration complicated the analysis of postevent traffic, in many cases beyond what could be addressed in this project. Although there is little argument that major sporting events are special events, what about community events? Large events such as July 4 fireworks displays are obviously special events from a traffic perspective, but what about parades or conventions? Not only are the sizes of these events difficult to obtain, but their start and end times are far less consistent, especially in terms of when traffic volumes going to and from those events affect roadway performance. A final consideration in developing the analysis data set was that special event traffic generally only affects roadway performance near the event venue. That is, when a major col- lege or professional football game takes place, traffic near the stadium is bad, but traffic farther from the stadium is often light (because a large percentage of the population is at the game or watching it on television). Previous work for WSDOT showed that while special event (professional baseball and basketball) traffic had statistically significant effects on major freeways leading to the event locations, roadway performance in the opposite direction before the game began was generally not statistically significantly different (1). Consequently, special event data need to be applied on a site-specific basis; descriptive information (time, location, and size) and local knowledge of the likely routes of travel affected by the event are required. These site-specific require- ments made attempting to analyze 21 roadway corridors on five freeways covering approximately 120 centerline miles of roadway problematic. In the end, the project team decided to simply use the volume data from the freeway and to analyze the effects of special events as case studies to illustrate the relative size and significance of their impacts. Traffic Control Devices This study did not collect data on traffic control devices. All sections of freeway under study operate under ramp meter- ing control. The fuzzy, neutral ramp-metering algorithm used by WSDOT changes ramp metering rates dynamically in response to a combination of inputs, including mainline vol- umes and lane occupancy values at the ramp, upstream of the ramp, and downstream from the ramp, as well as the presence of ramp queues and the determination of whether those queues are long enough to affect arterial operations. Ramps are metered whenever congestion routinely forms. This includes all commute periods and most weekend after- noons for freeways near the downtown core areas. Metering is only applied in the direction in which congestion is (or has) formed. Because only 1 year of data were analyzed in this study, it was not possible to determine the effects of the ramp-metering algorithm on congestion. A case study is presented below that describes the benefits obtained from meters. Other than that case study, traffic control devices are not examined in this report. Bottlenecks and Inadequate Base Capacity No specific data were collected relative to the base capacity of the roadways being studied. Several major bottlenecks are represented in the data set. In most cases, bottlenecks are located at the ends of study sections. One type of bottleneck is a ramp terminal at the end of a roadway. Two examples of this occur: the eastern end of SR 520 (affecting SR 520 Red- mond eastbound) and the western end of I-90 (affecting I-90 Seattle westbound). A second type of bottleneck is a freeway- to-freeway ramp interchange, where ramp volumes overwhelm the interchange capacity. One example is the interchange between northbound SR 167 and I-405 (both directions). This bottleneck affects SR 167 Renton northbound, I-405 Kennydale southbound, and I-405 South northbound. Other freeway-to-freeway ramps also contribute to congestion, usu- ally because the mainlines to which they lead experience rou- tine backups. Although these may not be classic bottlenecks, ramp queues can cause congestion. Freeway-to-freeway ramps that exhibit these conditions fairly frequently include SR 520 Redmond going westbound to I-405 Kirkland north- bound and I-405 Bellevue central business district (CBD) southbound; SR 520 Seattle westbound to I-5 Seattle North northbound; and I-5 Seattle CBD southbound. Both the northbound Seattle CBD and southbound Seattle North sec- tions of I-5 can be affected by queues extending from the eastbound SR 520 Seattle study section. Similarly, both direc- tions of the Seattle CBD sections of I-5 are affected by queues on the westbound I-90 Seattle section. Finally, the I-90/I-405 ramps cause delays primarily to four movements: to west- bound I-90 from the northbound (Eastgate) and southbound (Bellevue CBD) sections of I-405, and to westbound I-90 from northbound I-405. The ramp to southbound I-405 also backs up, but the queues to that ramp rarely affect I-90 per- formance because of the storage available on the ramps. The I-5 Seattle CBD sections in both directions contain several bottlenecks. In addition to the freeway interchanges, this section of freeway is affected by several C-class weaving movements, a variety of lane drops and adds, and the north- bound entrance and southbound exit from the I-5 express lanes. (The performances of the I-90 and I-5 express lanes were not included in this study.) The southbound entrance and exit to the express lanes also affect traffic on I-5 south- bound on the North King study section and northbound on the Seattle North section. The I-90 express lane entrances and
86 exits have less of an impact (the westbound on-ramp mod- estly affects the I-90 bridge section in both directions). The other major bottlenecks of special significance are the two Lake Washington floating bridges (SR 520 and I-90). The entrances to the SR 520 bridge, in particular, are major bottle- necks, as they both involve a combination of narrow lanes, strong visual impacts, and ramp entrances. In both cases, the bridge bottlenecks are located in the middle of the study sec- tion. The affected sections are the two Seattle sections of SR 520 and the two bridge sections of I-90. No attempt was made to quantify the specific capacity reduc- tions caused by these bottlenecks. However, as the results presented later in this report show, these sections all experi- ence considerably more delay than freeway sections without bottlenecks. Computed Variables Used for Tracking the Influence of Disruptions on Travel Times and Delays The interaction of all of the factors discussed above is very complex. All analytic methodologies have limitations when trying to determine how each factor of a given set of factors affects the delays experienced by a traveler using the roadway system. To decrease the effects of these limitations, the research team developed additional variables to help associate travel times and delays with specific disruptions. To understand the need for these variables, consider the following example incident. A major traffic accident occurs early in the morning, before the start of the morning commute, in the outer extent of the metropolitan region. The accident blocks most of the freeway and lasts 2 hours, forming a significant queue despite the early hour. Because traffic from the outlying areas is blocked, inbound commute travel times downstream of the accident start off better than normal. The accident is cleared after the morning commute peak begins. Once the accident has been cleared, a major pulse of traffic flows downstream from the accident location because the roadway clearance releases the large queue of vehicles stored upstream of the accident scene. That pulse of traffic nearly equals roadway capacity. When normal on-ramp volumes are added to that flow, congestion forms in unusual locations. The result is significant travel time delay that continues well after the accident has been cleared from the roadway, with the congestion occurring well downstream of the accident location. If a queuing analysis is performed for the accident location, only the delay computed upstream of the accident location is attributed to the accident, as the downstream congestion occurs both after the accident has been cleared and at loca- tions that are geographically removed from the accident site. Thus, the delays associated with the accident are computed to be smaller than the real congestion caused by the accident, which should include the delays occurring downstream of the accident site. At the same time, some of that congestion should rightly be attributed to routine peak period morning traffic, which always causes congestion. Therefore, not all the delays in the corridor should be attributed to the accident. The delays are influenced by the accident, but high volumes also contributed to the measured delay. With the above scenario in mind, the project team devel- oped a set of variables to help relate the measured performance of the roadway (travel times, volumes, and delays) to known disruptions. A value was assigned for each of these new, computed variables for every 5-minute time interval in the analysis data set (i.e., all of 2006). These additional variables included the following: ⢠Travel delays were computed by corridor segment so that all delay (any travel less than 60 mph, in units of vehicle seconds) was computed. ⢠The times when potential disruptions took place were identified for each type of disruption event, and variables identifying that a disruption was active or not present were created for each 5-minute interval for the year. ⢠Binary influence variables were computed for which influ- ence was defined as occurring when either (a) the potential disruption event was active during a given 5-minute period or (b) travel times for the corridor were observed to be slower than any observed during the observed disruption. This definition of influence essentially means that slow- downs occurring in the corridor during the period of active disruption are at least partially caused by that dis- ruption; that is, travel times are influenced by a given dis- ruption. In the analysis, the binary influence tag stays on until travel times in the corridor return to values equal to or faster than the fastest travel time observed during the duration of the event itself. That is, if a crash or incident occurs at the beginning shoulder of a peak period and some congestion forms (even if the majority of that con- gestion is caused by the increasing peak period volumes), then the influence tag will likely stay on until after the peak period congestion eases. This is an intended outcome. It signals that the disruption (crash or incident) may have caused congestion to be worse and last longer than it other- wise would have. The influence tag is turned off once travel times return to predisruption levels, indicating that any queues present in the corridor are no larger than those that existed before the effects of the disruption. (A more complete discussion of the influence variables is found in Appendix D.) ⢠Influence variables were computed for (a) all incidents, (b) only those incidents that involved lane closures, (c) vehi- cle crashes, (d) active construction events, (e) bad weather,
87 and (f) rubbernecking, where rubbernecking was defined as a time during which a crash or incident was active on the roadway section being studied, but in the opposite direc- tion of travel. A variety of influence variable calculations were computed and tested. Variables were developed that would allow off-segment congestion influences to be related to the segment under study. (A detailed description of the variable codes or categories used to indicate the influence of congestion from off-study segments on the study sec- tion of interest is found in Appendix C.) These variables were activated when the first detector (mainline or ramp) downstream of the study section had an occupancy value of greater than 35% for the 5-minute period of interest. When that occurred, these variables were set to a categori- cal value that described the influences on the congestion of that downstream segment. Variables were created for the downstream mainline roadway sections, for freeway-to- freeway ramps known to experience backups, and for major off-ramps known to spill back on the mainline roadway during peak commute periods. These variables were designed to allow transfer of the effects of a downstream disruption to an upstream roadway study segment when queues from that disruption extended off the end of the downstream segment. For example, if a crash on the roadway section just north of the CBD caused a queue on I-5 northbound that reached the detector just downstream of the north- bound CBD roadway study section, the variable represent- ing the mainline roadway section downstream of the CBD section would be set to crash-influenced congestion so that analyses of the CBD roadway section would include the fact that an off-segment crash was influencing the perfor- mance of the roadway segment. ⢠The regime variable was developed to describe the worst condition found in the test segment during each 5-minute interval. (A detailed description of the regime variable is found in Appendix C.) Regime is a categorical variable in which 1 = free-flow traffic, low volumes; 2 = free-flow traf- fic, less than one lane of capacity remains; 3 = constrained flow, very high volumes; 4 = congestion exists; and 5 = recovery. Regime, which is illustrated in Figure 5.2, was used to define the basic operating condition of the road- way study section. ⢠Six binary variables were defined to indicate whether a road- way section moved from a free-flowing regime to a con- gested regime within a given time frame. These variables allowed an estimate of the probability that a specific event resulted in congestion formation when the period was compared with similar time periods on other days when operating conditions were similar. Three binary variables described whether roadway operation moved from Regime 2 to Regime 4 within 5, 10, or 15 minutes. The other three variables described whether roadway operation moved from Regime 3 to Regime 4 within 5, 10, or 15 minutes. ⢠The time when congestion ended was computed for both the a.m. and p.m. peak periods. This time was defined as the first 5-minute period after the start of the peak period (7:00 a.m. or 4:00 p.m.) when travel times were no more than 5% above travel at the speed limit. For example, if travel at the speed limit required 300 seconds, congestion ended for the peak period on any given day when four con- secutive travel times were observed to be below 315 sec- onds. (A more complete discussion of this variable is included in Appendix C.) On 11 of the 42 study sections, this definition created mean congestion ending times for Figure 5.2. Illustrations of roadway operating regimes.
88 the a.m. peak period that were later than noon because of various volume and bottleneck conditions that caused midday traffic to routinely travel below the speed limit. For some specific analyses, congestion was defined on these sec- tions only as being when travel time dropped to within 10% or 20% of travel at the speed limit. Findings from Seattle The findings are divided into four major subsections: 1. Congestion by source; 2. The effect of weather; 3. The effects of crashes and incidents on travel times; and 4. The effects of crashes and noncrash incidents on the extent of congestion. The first subsection examines, at an aggregated annual level, how delay changes with different types of disruptions to the fixed infrastructure. Congestion sources examined include weather, crashes, other noncrash incidents, and construction activities. The second subsection looks specifically at how weather, primarily rain, affects travel times and congestion formation. The third subsection examines how travel times change given the occurrence of incidents and the queues that result from those incidents. As part of this analysis, the specific effects of vehicle crashes are examined, both independent of noncrash incidents and in combination with noncrash incidents. The fourth subsection examines how the duration of peak periodârelated congestion changes as a result of crashes and noncrash incidents. The intent of this analysis was to put into context how crashes and incidents change the travel experi- ences of commuters in a congested urban area. Congestion by Source This analysis examined how different types of disruptions influence the formation of congestion and the degree of delay experienced by travelers. It covered only general-purpose travel lanes (no high-occupancy vehicle [HOV] or high- occupancy toll lanes) and used units of vehicle hours of delay, not person hours, as the available data did not account for changes in vehicle occupancy during different days of the week, times of day, or types of facilities (e.g., weekends having much higher vehicle occupancy rates than weekdays, com- mute hours having generally higher occupancy rates than the middle of the day on weekdays, and HOV lanes having much higher occupancy rates than general-purpose lanes). The analysis covered only urban freeways in the Seattle metro- politan region. The analysis did not attempt to differentiate among relative causes when two or more causative factors were present. That is, when a crash happened in the rain dur- ing the peak period in the peak direction, the analysis did not attempt to determine how much of the delay was caused by the crash, how much was caused by rain, and how much was caused by high peak period volumes. Methodology The congestion by source analysis computed delay per 5- minute period for all 5-minute periods in the year (2006) and assigned that delay on the basis of the influence variables associated with each of those 5-minute periods. (See Appen- dix C for a description of the influence variables.) Delay was computed with the following equation: delay actual travel time travel time at the= â speed limit roadway segment volume ( ) ( ) where roadway segment volume was the maximum volume observed in the study section for that 5-minute period. Actual volume counts tend to underestimate the number of vehicles queued within a section during times of heavy congestion; consequently, this equation slightly overstated delay in lower- volume periods but better estimated the number of vehicles actually in the roadway section during times of peak conges- tion. When study section travel times were faster than the speed limit, conditions were assumed to be operating at the speed limit. A categorical variable was developed that allowed any combination of influences to be maintained simultaneously. The following influences were tracked: ⢠No cause indicated; ⢠Only incident-influenced queues are present; ⢠Only crash-influenced queues are present; ⢠Only rain is present; ⢠Both a crash and an incident have influenced queues that are present; ⢠Both rain and an incident have influenced queues that are present; ⢠Both rain and a crash have influenced queues that are present; ⢠Rain, a crash, and an incident have influenced queues that are present; ⢠Queues from a ramp have influenced mainline queues, but the ramp delays have no identified influence factor; ⢠Construction activity has influenced queues; ⢠Construction and queues from a ramp (cause unknown) have influenced mainline queues; ⢠Construction and an incident have influenced queues that are present; ⢠Construction and a crash have influenced queues that are present;
89 ⢠Construction and rain have influenced queues that are present; ⢠Construction, a crash, and an incident have influenced queues that are present; ⢠Construction, rain, and an incident have influenced queues that are present; ⢠Construction, rain, and a crash have influenced queues that are present; and ⢠Construction, rain, a crash, and an incident have influ- enced queues that are present. Delay statistics were then aggregated by type of influence present. Traffic volume, whether it was routine volume or an unusual surge in volume associated with something like a special event, was not explicitly tracked in this analysis. Un explained congestion was assumed to be caused exclu- sively by the presence of too much traffic volume. Results Table 5.3 summarizes the amount of delay influenced by each type of disruption tracked in this study. Percentage of delay was computed by totaling all vehicle hours of delay in the region associated with each type of disruptions, and then dividing by the sum of all measured delays. This computa- tion automatically weighted the delays experienced by each roadway on the basis of the relative number of vehicle hours traveled on that roadway section. Of interest is the fact that rain had almost as much influ- ence on congestion as vehicle crashes. Not surprisingly, con- struction (defined as lane closures during active construction or maintenance activity) had the least influence on conges- tion formation. The percentage of delay associated with construction is small mainly because construction closures are only allowed on urban area freeways during the late-night hours, when volumes are low. Thus, even when congestion (measured in terms of either the queue length or the amount of time an individual spends in that queue) is significant as a result of construction lane closures, total vehicle delay (vehi- cle hours) is small relative to the amount of delay experienced in the peak periods, when volumes are high. One type of construction delay not included in Table 5.3 is delay caused by the temporary geometric changes (narrowed lane widths, lane shifts) that are commonly required by many urban freeway construction activities. These geometric restric- tions are likely to cause congestion to form earlier and last longer than it would with the roadwayâs normal geometry. The project team did not attempt to establish when these semipermanent geometric conditions were implemented, nor did the team attempt to associate delays with these changes during nonclosure hours (e.g., a.m. and p.m. peak periods). âNo cause indicatedâ in Table 5.3 means that no cause of congestion was reported other than high traffic volume levels. The team examined a number of these conditions as case studies. It was clear from that review that a variety of disrup- tions occur that affect traffic flow but that are not recorded within conventional traffic operations databases. Many of these disruptions are visual distractions (e.g., boats on the lake, slowdowns due to sunglare) that cause measurable delays only when traffic volumes are relatively high. In some of the case study investigations, traffic volumes on the study corridor were abnormally high because of disruptions on parallel roadways. This analysis did not attempt to track route diversion onto parallel roadways and, therefore, was not able to associate congestion on one roadway with disruptions occurring on a second roadway. This subject is discussed in more detail later in this section. Table 5.4 shows a more disaggregated version of Table 5.3 in that it tracks multiple disruptions occurring at the same time. Table 5.4 also illustrates the wide variation among the 42 study sections in the percentage of delay influenced by any given cause (e.g., incident-influenced queues may have been much more prevalent at one study site than at another) by presenting the maximum and minimum values observed for each combination of delay causes. Table 5.5 shows the total number of vehicle hours of delay measured. Note that the northbound I-405 data sets are miss- ing about 1.5 months of data (mostly from November and December); other corridors periodically missed days or weeks of data as a result of various data quality and availability issues. These missing data mean that the total measured delay was not the true annual delay for the regionâs freeways. How- ever, the missing data should have only a marginal effect on the percentages of delay associated with different types of dis- ruptions. In general, the roadway corridors with the highest Table 5.3. Percentage of Delay by Type of Disruption Influencing Congestion Type of Disruption Delaya (%) Incidents 38.5 Crashes 19.5 Bad weather (rain) 17.7 Constructionb 1.2 No cause indicated (mostly volume) 42.2 a Delays that occurred when more than one type of disruption influenced the size and scope of that delay were counted in each of the categories of disruption and, therefore, the per- centages total to more than 100%. b Construction delays do not include any delays caused because general roadway capacity was reduced as a result of temporarily narrowed or reconfigured lanes. Construction delay was computed only when construction activity actively took place along the roadway.
90 percentage of delay attributed to unknown causes tended to be those roadway sections with the least absolute vehicle delay. That is, nine of the 10 sections with the highest per- centage of delay not caused, at least in part, by a known traffic disruption were among the 13 sections with the lowest total vehicle delay for the year. The converse of this statement was not true. Although the two test sections with the most vehicle hours of delay did have fairly low percentages of delay not associated with known dis- ruptions, only half of the 10 test sections with the highest vehicle delay were among the 10 sections with the lowest per- centage of congestion influenced by an unspecified disrup- tion. The sections with very large amounts of total vehicle delay and large amounts of delay caused by unknown disrup- tions were all segments where frequent, significant peak period delays occurred. The westbound segment of the SR 520 Seattle bridge has a large bottleneck at the eastern end of the 2-mile-long floating bridge. Both SR 520 and I-405 Kennydale (both directions for both corridors) operate near or above capacity for 10 to 14 hours per day. The two I-5 sec- tions (the South section northbound and North King section southbound) experience routine a.m. peak congestion. Con- sequently, it is reasonable to assume that large amounts of the delay in these corridors are simply caused by too much peak period volume. The percentage of delay occurring with no reported dis- ruption also was compared with the a.m. and p.m. peak period travel rates (defined as the mean travel time for the peak period converted to units of minutes per mile) for each corridor. No correlation between these values was apparent. This lack of correlation between different measures of con- gestion and the amount of delay without a known disruption Table 5.4. Percentage of Delay by Type of Disruption Influencing That Congestion Type of Disruption Delay (%) Maximum Percentage Within a Corridor (%) Minimum Percentage Within a Corridor (%) No cause indicated 37.1 74.2 14.3 Incident-influenced queues are present 23.9 48.2 1.0 Crash-influenced queues are present 6.0 25.3 1.7 Rain is present 8.4 25.8 2.0 Both a crash and an incident have influenced queues that are present 9.2 23.9 0.5 Both rain and an incident have influenced queues that are present 5.0 8.9 0.0 Both rain and a crash have influenced queues that are present 1.6 8.7 0.2 Rain, a crash, and an incident have influenced queues that are present 2.4 13.6 0.0 Queues from a ramp (cause unknown) have influenced mainline queues 5.1 37.3 0.0 Construction activity has influenced queues 0.6 16.2 0.0 Construction and queues from a ramp (cause unknown) have influenced mainline queues 0.0 0.2 0.0 Construction and an incident have influenced queues that are present 0.2 2.6 0.0 Construction and a crash have influenced queues that are present 0.1 1.4 0.0 Construction and rain have influenced queues that are present 0.1 4.6 0.0 A crash, an incident, and construction have influenced queues that are present 0.1 1.2 0.0 Construction, rain, and an incident have influenced queues that are present 0.0 0.5 0.0 Construction, rain, and a crash have influenced queues that are present 0.0 0.7 0.0
91Table 5.5. Hours of Delay Versus Percentage of Delay Without a Known Type of Disruption Corridor Vehicle Delay (h) Delay Not Associated with a Disruption (%) I-5 Seattle CBD northbound 28,689,099 14.3 I-5 Seattle North southbound 19,828,935 23.1 I-5 South southbound 14,063,546 27.7 I-5 Seattle CBD southbound 12,997,924 21.5 SR 520 Seattle bridge westbound 12,901,102 43.3 I-405 Kennydale northbound 11,531,897 55.3 I-405 Bellevue southbound 11,345,712 20.8 I-405 Kennydale southbound 11,077,760 56.9 I-5 North King southbound 10,782,330 45.2 I-5 South northbound 10,441,430 41.6 I-405 Kirkland southbound 9,655,929 34.0 I-405 Kirkland northbound 9,651,791 24.4 I-405 North southbound 9,116,178 44.2 I-5 Lynnwood southbound 8,517,553 39.8 I-5 Lynnwood northbound 7,733,702 53.5 SR 520 Seattle bridge eastbound 6,445,475 29.6 I-5 North King northbound 6,020,659 22.6 I-5 Tukwila northbound 5,997,528 42.5 I-90 Bridge westbound 5,310,825 57.3 SR 167 Renton northbound 4,980,431 28.0 SR 167 Renton southbound 4,582,608 58.3 I-5 Seattle North northbound 4,399,711 35.9 I-405 North northbound 4,327,382 56.4 I-405 South northbound 4,091,618 61.8 I-5 Tukwila southbound 3,863,679 45.1 I-5 Everett northbound 3,838,909 33.0 I-405 Bellevue northbound 3,773,393 52.0 I-90 Bridge eastbound 3,744,002 17.2 SR 520 Redmond eastbound 3,307,029 36.2 SR 167 Auburn southbound 3,305,901 59.9 I-90 Issaquah westbound 3,229,088 73.4 I-405 Eastgate southbound 2,861,851 64.8 I-405 South southbound 2,740,581 74.2 SR 167 Auburn northbound 2,167,614 73.0 I-90 Seattle eastbound 1,738,429 65.6 I-405 Eastgate northbound 1,715,306 64.4 I-90 Bellevue westbound 1,705,939 30.6 SR 520 Redmond westbound 1,399,767 19.7 I-5 Everett southbound 915,200 41.2 I-90 Bellevue eastbound 519,902 66.1 I-90 Seattle westbound 454,026 40.8 I-90 Issaquah eastbound 256,341 63.5
92 was not expected at the outset of this analysis. It had been assumed that most of the delay without an observable cause was primarily due to too much traffic volume. The expecta- tion was that highly congested locations, especially those with well-known geographic bottlenecks, would have the most delay with unspecified causes because the congestion would be caused by a combination of volume and roadway geometryâ based capacity limitations. Test sections with lower levels of routine delay were expected to have higher percentages of delays with identified disruptions, as delay would exist on those road segments primarily when unusual events occurred. Instead of simple volume and capacity issues being the pri- mary cause of high levels of delay unrelated to observable disruptions, further analysis of the study corridors identified at least three major reasons for delay occurring without known disruptions being present: 1. Operating agencies simply do not record many of the dis- ruptions that occur, especially on less congested corridors and during less congested periods (weekends, at night); 2. In several cases, the research teamâs analytic approaches did not adequately track all of the disruptions that occurred, given the data available to indicate when and where those disruptions actually happened; and 3. Even on Seattleâs less congested urban freeway segments that do not have major geometric bottlenecks, volume is fre- quently sufficient to cause at least modest amounts of delay. When total delay values are small, these types of no-cause delays can represent a fairly high percentage of total annual delay. These conclusions were supported by several case study examinations of the various study corridors. One case study was performed on the I-90 Issaquah eastbound section, which had the lowest measured annual delay of all 42 seg- ments studied for this project. Only 256,000 vehicle hours of delay were measured in 2006, and 63.5% of that delay was not associated with an identified disruption. This roadway seg- ment experienced two major delay-causing events in Novem- ber 2006 that were not identified by the analysis methods described above. One of those events was a snow storm; the second was a major truck accident. A special analysis of the snow event determined that roughly 5.9% of all delay mea- sured for the year for this section of roadway occurred during that event. Yet because the snow stopped falling (at least at the weather station from which data were obtained) several hours before congestion started on this freeway segment, the con- gestion delays recorded were not associated with that weather phenomenon. A review of newspaper stories published the next morning confirmed that massive snow-related problems occurred that night on that roadway section. Additional dis- cussion of the difficulty in analyzing snow-related delays is presented later in this report. On a second day in November 2006, an accident involving a truck killed the driver of a passenger car on I-90. That acci- dent was not listed in either the state accident database or the WSDOT WITS database. Newspaper accounts indicated that the crash occurred in the westbound lanes of I-90 at 10:38 a.m. west of Front Street, which is on the eastern end (but within the boundaries) of the I-90 Issaquah test section. Although the crash occurred in the direction opposite the I-90 Issaquah eastbound section examined in the case study, the eastbound section reported far longer delays than the westbound section after 10:30 a.m. The longer delay may have been due to the location of the crash, which likely caused much of the west- bound queue to form east of the monitored portion of the roadway. In addition, the eastbound delays were likely pri- marily rubbernecking delays, although some response equip- ment may have been parked on the eastbound section of the roadway. The exact reasons are not clear, but it was clear from the database that travel times were significantly affected, as would be expected with an accident involving a truck and with the time and lane closures required to investigate a fatal accident. Although some delays on that day were associated with rain, the majority of delay was not associated with any disruption. Thus, another 5.1% of all annual delay (8.1% of delay not associated with a disruption) was erroneously attrib- uted to no cause other than volume. Consequently, for this roadway section, of the 63.5% of delay ânot associated with a disruption,â 11% was actually associated with just two events, leaving at most 53% caused only by too much traffic volume. Similar case study analyses of significant but unexplained delays were undertaken on road segments with greater con- gestion. One of the most congested segments in the region is the westbound section of SR 520 as it crosses Lake Washing- ton from Bellevue to Seattle. This segment experiences over 50 times the annual delay experienced on the I-90 Issaquah section discussed above. The SR 520 bridge operates near or over capacity for 13 to 14 hours every weekday. It is parallel to another cross-lake bridge (the I-90 bridge, located to the south of SR 520), which is close enough so that motorists can easily divert between the two when one of them experiences heavy congestion. Each August, a major hydroplane race takes place on Lake Washington south of the I-90 bridge. During the weekend of the race, the Navyâs Blue Angels flying team also performs an air show in between hydroplane race heats. The Blue Angels practice their routine during the day on the Thursday and Friday preceding the air show. During the times when the Blue Angels are practicing or performing their show, the I-90 bridge is closed to traffic. Not surprisingly, considerable delay occurs that week crossing the two bridges. Much of that delay is caused by the visual distraction of pleasure boats on the lake going to and
93 from the race course and by airplanes flying low overhead. In addition, because the I-90 bridge is closed to traffic during the Blue Angel flights, considerable traffic diverts to the SR 520 bridge. All this activity results in the perfect storm for creating congestion on SR 520, much of which is not related to a specific disruption on SR 520. The disruption (as noted in VMS records) is on I-90. In 2006, on the Thursday before the hydroplane races, westbound SR 520 did not experience any major disruptions (i.e., recorded construction, lane closures, crashes, or rain). However, it did experience 117,000 vehicle hours of delay (roughly half the total annual delay of the I-90 Issaquah east- bound test section). About half of that delay was not associ- ated with a disruption in the analysis database, and that value was over 2.5 times the usual noninfluenced Thursday delay. It is obvious from a manual review of the data that these delays were caused by excessive demand resulting from the 2-hour closure of the I-90 bridge combined with a high level of visual distraction for motorists crossing the lake. However, because the delays routinely experienced on this section of roadway are so high, this very bad day for travel on this section only contributed 0.9% of the total annual delay for this test sec- tion, and thus the large not-influenced delay for that day was less than 0.5% of the annual total. Taken together, these case studies illustrate that a large per- centage of the congestion in the analysis data set without a cause can be traced back to some type of unusual occurrence. However, because of limitations in both the analysis data set and the methodology used to associate delays with specific events, this analysis was unable to reliably identify all these congestion sources. Consequently, three conclusions were drawn from the above examples: 1. The statistics presented in this report should be assumed to be a very conservative estimate of the amount of delay caused by the various types of disruptions; 2. The percentage of delay caused by any given factor can be a misleading statistic about the importance of that factor, since it is highly correlated to the total amount of delay on a given roadway; and 3. In the presence of moderately heavy volumes, a large number of factors that are not tracked by operating agen- cies may be the cause of congestion. Effects of Weather The case study of delays on I-90 when snow fell illustrates the difficulties in determining the effects of weather on roadway performance. The largest roadway performance effects caused by the snowfall did not occur while the snow was falling at the weather station. Instead, they occurred as a result of snow accumulation on the roadway and the conversion of that snow into sheet ice on some roadway sections. The latter of these events took place well after the snow had stopped falling at the weather station. In addition, the analysis of that case study reveals that delays did not happen similarly on all roadway sections that evening (although the newspaper reported long delays on several corridors). In fact, the eastbound and westbound sec- tions of I-90 (presumed to experience the same level of snow- fall) experienced very different roadway performance (delay) conditions during and after the snow storm. While the west- bound direction showed modest delays in the evening, with moderate delays occurring between 6:00 and 9:00 p.m., the eastbound section experienced an unusually heavy day of congestion before the snowfall, and then a major additional pulse of congestion starting at 8:00 p.m. and lasting well into the morning hours. Exacerbating the eastbound congestion was the traffic volume added because of a professional foot- ball game that occurred that night in downtown Seattle. The Seahawks played the Packers on Monday Night football, add- ing 65,000 fans, divided across multiple freeways, to the out- bound traffic beginning at about 8:30 p.m. Methodology The snowfall case study revealed a number of the analytic problems associated with an analysis of the effects of bad weather. The first major problem is defining, in analytic terms, bad weather. As discussed previously, the key region- wide weather variable used to indicate bad weather was whether measurable rain had fallen in the past hour. This variable was then used as an independent variable to predict the probability that any given roadway section was operating in a given regime (essentially, level of service). The analysis computed the probability that a given test sec- tion of roadway was operating in each regime for each time slice of a day. These probabilities were computed for days when rain occurred within the past hour and were then compared with probabilities on days when the same roadway was dry at that same time of day. The mean, median, 80th percentile, and 95th percentile travel times for each corridor and time period also could be computed for wet and dry conditions. One limitation with the travel time analysis is best explained with an example. Rain falls between 3:00 and 4:00 p.m. The time periods between 3:00 and 5:00 p.m. are assumed to be rain affected (within 1 hour of when measurable rain has fallen). Travel times occurring at 4:55 p.m. that day are rain affected, but travel times at 5:05 p.m. are considered dry trips. The limitation with this analysis is that the rain may have cre- ated a queue that affects the 5:05 p.m. dry trip. For the analy- sis results in the discussion below, such a possibility was ignored, thus slightly underestimating the potential impacts of rain on travel time.
94 Sensitivity tests were performed with various definitions of rain (e.g., requiring different fractions of an inch of rain fall- ing within the previous hour for the pavement to be consid- ered wet) and with different time periods within which rain had to have fallen (e.g., within the past hour or 2, 4, or 8 hours for the pavement to be considered wet) to test how sensitive the results were, given different definitions of wet. In general, any measurable rain falling within the past hour had the greatest effect on congestion formation and the resulting travel time. Other values showed slightly lower effects. The effects of wind on roadway performance were ana- lyzed differently from the effects of rain. This is partly because, other than the lasting effects of any queues being formed, wind does not have a lasting effect similar to that of rain. Once wind stops, its direct effects stop. That is, wind does not have a lasting effect equivalent to spray from wet roadways caused by rain. The lack of this effect also limited the teamâs confidence in the use of the available NOAA wind data for specific roadway sections. As a consequence, the wind gust variable produced by NOAA was not used. The project team had little confidence that this variable was effectively applicable to geographically removed locations. Similarly, the wind speed variable that was used was assumed to be only a reasonable surrogate for windy conditions, and not a definitive statistic indicating the precise wind speed at which travel might be affected. To test the effects of wind on travel times, the data set was divided into wind-affected and not-wind-affected groups on the basis of the wind speed variable present in each 5-minute time slice. The travel times for these two groups were then compared within specific time intervals with both traditional t-tests, which assumed normally distributed travel times within those time periods, and nonparametric tests of the sample means. Tests were performed only for nonholiday Tuesdays, Wednesdays, and Thursdays (combined). Sensitivity tests were performed with different values of the wind speed variable to determine the sensitivity of the analysis results to the breakpoint selected for identifying windy versus not-windy conditions. The performance of dif- ferent roadway corridors was found to be sensitive to differ- ent wind speeds. The authors believe that this is due in part to differences between actual wind speeds within the study corridor and those measured at the airport, and in part to the way that site-specific roadway geometry affects how drivers respond to wind. For example, travel times over the SR 520 floating bridge, which has narrow lanes, no shoulders, and physically moves when struck by wind-blown waves, are affected at much lower wind speeds than travel times on I-5 in the northern reaches of the metropolitan region, where lanes are wider, full-width shoulders exist, and wind does not cause the roadway to move. In the end, sustained wind speeds of 16 mph were used as the primary split between windy and not-windy conditions. Adopting a different definition would marginally change the travel times associated with windy and not-windy conditions for some corridors but would not change the ultimate conclusions of the study. Results Not surprisingly, the results uniformly showed that the occur- rence of rain led to a statistically significant increase in the amount of congestion, but only during periods of moderately high traffic volume. That is, rain does not cause congestion uniformly throughout the day. The probability of congestion forming as a result of rain is a function of the underlying level of vehicular demand. And given the time series nature of traf- fic flow, time of day and day of week can be used as surrogates for vehicular demand when estimating the probability of congestion forming. Rain causes the roadway to operate just a little less effi- ciently than it would otherwise (2, pp. 1â14; 3, pp. 8â18). The result, as observed in the data set, is that given a normal com- mute period, the roadway is likely to break down a little ear- lier than it would otherwise under conditions of similar demand on dry roadways. The amount of rainfall likely deter- mines the degree to which roadway efficiency declines, but an analysis confirming this was not completed for this study. Because the roadway breaks down earlier than it would if rain had not occurred, the queues grow larger than they otherwise would, and consequently last longer. The moderate rate at which rain falls in Seattle (or more accurately, the regionâs frequently wet roadways) does not cause congestion; it simply lowers the amount of traffic volume that a given roadway can handle before it becomes congested. Therefore, the roadway breaks down earlier in the commute period than it would otherwise. Figure 5.3 illustrates this trend for SR 520 Seattle west- bound crossing the Lake Washington floating bridge. The gray line shows the probability of a traveler experiencing con- gestion on this corridor on a dry day. The black line illustrates the probability of being in congestion if rain has fallen within the past hour. SR 520 westbound into Seattle is one of the more congested roadway segments in the region. It experi- ences congestion during both the a.m. and p.m. peaks, as well as periodically in the middle of the day. Figure 5.4 shows one of the less congested roadway sec- tions in the region. In this case, only the a.m. peak period routinely experiences congestion. Therefore, in the morning when volumes are high, if rain falls, the probability of conges- tion forming in the next hour increases. However, after the peak period ends, the fact that rain has fallen has no discern- ible impact on the formation of congestion. Yes, falling rain may increase accident rates during off-peak times (see the discussion below on accident rates and the presence of rain),
95 but congestion caused by that increase in accident rates is no more likely to occur than congestion from other sources. The greater probability of congestion early in the peak period and the longer queues that result from that early start to congestion also mean longer travel times on rainy days. Figure 5.5 illustrates how mean travel times increase along with the increased probability of being in congestion. This graphic shows the probability of congestion having formed by time of day when the roadway is dry (gray line) or has been rained on in the past hour (black line). It also shows the change in mean travel time when rain has fallen (dashed line), where the travel time increase is shown on the right- hand axis. As Figure 5.5 shows, at no time does the mean travel time decrease with statistical significance when rain is present. Interestingly, this figure also shows that the declining volumes at the end of the commute period quickly moderate the travel time effects of the congestion developed as a result of early queue formation in the rain. That is, even though the queues are longer and the travel times worse in the peak period, the mean travel time for a trip starting at the end of Figure 5.3. Probability of being in congestion: rain versus no rain on SR 520 westbound from Bellevue to Seattle. Figure 5.4. Probability of being in congestion: rain versus no rain on I-90 westbound from Issaquah to Bellevue.
96 the commute period is only marginally worse than normal, and by the end of the peak period, travel times are nearly the same as normal, regardless of whether rain has fallen. While the effects shown in Figure 5.5 were observed fairly universally for all roadway segments studied, further analysis of the 42 study segments revealed two significant differences in the effects of rain between less congested and more con- gested roadway segments. First, on the more congested seg- ments, enough volume exists during the middle of the day that rain causes an increased likelihood of congestion form- ing during midday periods. On less congested roadway seg- ments this is not the case. The project team believes that on road segments that operate near capacity during midday, the decreasing roadway efficiency caused by wet pavement is sufficient to create congestion, regardless of increases in crash rates caused by the wet pavement. Additional analysis is required to determine the effects of the increased acci- dent rates versus the simple effect of wet pavements. On less heavi ly traveled (and thus less congested) roadway segments, the modest loss of efficiency caused by wet pavement does not create conditions that result in congestion, except on rare occasions when major crashes occur. The second significant difference between heavily con- gested and less heavily congested roadway segments is that on the most congested segments, the probability of congestion during the heart of the peak period approaches 100%. As a result, rain does not increase the probability of congestion forming during those periods. On less congested roadways, there are lower-volume commute periods (e.g., the workdays near major holidays) when congestion may not form. Rainfall on those lower-volume work days may decrease roadway per- formance to a degree sufficient for congestion to form. Figure 5.5 illustrates the effects of rain on a moderately congested roadway segment (there are no uncongested free- way segments in the Seattle region). Figure 5.6 illustrates how rain affects a heavily congested segment. In this figure, it is easy to see that the probability that congestion will form does not change significantly during the core of the p.m. peak period. However, during the early portion of the p.m. peak, travel times do increase when rain falls. This is because queues form earlier than normal and are, therefore, longer than nor- mal at later points in the day. Interestingly, in Figure 5.6 the travel time increases in the rain are briefly moderated just after the midpoint of the p.m. peak period. The increases in travel time caused by rain approach zero shortly before 6:00 p.m. (18 on the x-axis of the graph), only to rebound by 6:30 p.m. This outcome does not represent a lack of effect from the rain on commute times. Instead, it is an artifact of the roadway segmentation used for this specific analysis. On this particular roadway segment, the normal queue extends roughly to the end of the roadway analysis segment at the peak of the p.m. peak period. This maximum queue length occurs at roughly 6:00 p.m. Because the section already is fully congested, estimated travel times for the segment do not increase on the study section when it rains, and thus travel times do not increase. Instead, travel times increase on the upstream section of the roadway (in this case the SR 520 Red- mond westbound study section) because the queue from the Figure 5.5. Correspondence of increase in mean travel times with increase in probability of congestion due to rain on I-90 westbound from Issaquah.
97 first section has extended back onto the second section. Thus, travelers do experience slower trip times, but the reported travel time on this section is not worse. As the extra-long queue moderates toward the end of the peak period, travel times on the Seattle test section again increase, simply because the normal queue is once again shorter than the length of the entire roadway section. Research (and most driversâ personal experience) has shown that high winds frequently cause motorists to drive more slowly and carefully, as wind can affect vehicle handling. Under high winds, many drivers slow slightly (4; 5, pp. 24â30). As with rain, this more cautious approach to driving under heavy wind conditions can negatively affect the relationship of vehicle volume and speed, causing the roadway to operate less efficiently. Given high enough traffic volumes, this loss of efficiency results in congestion, although under normal cir- cumstances it would not form. Under these conditions, wind will result in statistically significant increases in travel time. An analysis of roadway performance and wind data in the Seattle region supported these basic findings. However, the analytic tests performed on the Seattle test corridors showed that travel times in all test corridors were not equally affected by wind. In fact, in many corridors, wind did not have any statistically significant effect on travel times. In other corri- dors, wind had a very high impact on roadway performance. Table 5.6 gives examples of how wind affected various corri- dors differently, even though the corridors are directly con- nected. Table 5.6 also gives examples of the results of the sensitivity tests performed with different wind speeds to sep- arate windy from not-windy conditions. As can be seen in Table 5.6, the SR 520 bridge is affected by relatively moderate winds (10 mph sustained wind speeds). The bridge is a 2-mile-long floating span with a roadway two lanes in each direction with no shoulders. In even moderate wind, a driver can feel the bridge sway. The wind also can cre- ate some spray when wind-driven waves break against the bridge, causing drivers to slow down. Because the bridge operates near capacity 12 to 14 hours each weekday, these wind effects are sufficient to cause congestion. The I-90 bridge, located nearby to the south, also is affected by wind, but to a lesser extent than the SR 520 bridge. This is most likely due to a combination of factors: the I-90 bridge is more modern, has full shoulders, and sits higher off the water (and, therefore, experiences less wind-driven spray). Interest- ingly, the evening commute across the I-90 bridge is affected by wind but the morning commute is not, even though traffic volumes are similar in both periods. This difference is partly because the test section that included the I-90 bridge also included a large segment of nonbridge travel across Mercer Island. Backups on the bridge affecting eastbound traffic actually create some free-flow conditions on the island itself, decreasing the travel time impact of the wind. However, wind-caused backups significantly affect the upstream sec- tion of eastbound I-90 (the Seattle section is also shown in Table 5.6). This explains why the I-90 Seattle section is statis- tically affected by wind in the morning, even though it does Figure 5.6. Correspondence of increase in mean travel times with increase in probability of congestion due to rain on SR 520 westbound, Bellevue toward Seattle.
98 not include the bridge itself. At more moderate wind speeds (e.g., 10 mph sustained winds), none of the I-90 segments show a statistically significant change in expected travel time. Looking at the I-5 segments included in Table 5.6, it can be seen that wind affects some corridors in some peak periods, but not all corridors or all peak periods within all corridors. In general, high peak period volumes relative to capacity make roadway segments more likely to be affected by high winds. Other reasons that a roadway may be susceptible to winds are that the road segment is exposed to high levels of wind (e.g., the I-5 North Seattle segment crosses the Ship Canal Bridge, an exposed portion of road where wind is often felt) or that the segment is immediately upstream of another segment that is wind affected. The I-5 North King segment is upstream of the I-5 North Seattle segment. The I-5 Everett segment is considerably farther north and does not experi- ence spillback from North King or North Seattle segments, except in very extreme cases. Figure 5.7 illustrates how wind affects the SR 520 bridge westbound, and Figure 5.8 illustrates the I-90 eastbound bridge section. In both figures, it can be seen that the primary effects of wind are in the peak periods when traffic volumes are highest. If the same graphic were presented with a higher wind speed, more impacts would be seen in the middle of the day, especially on SR 520. In Figure 5.8, wind appears to have a significant effect on expected travel times during the later portion of the a.m. peak period, but not on the earlier portion of the peak. This helps explain why the difference in mean travel times shown in Table 5.6 is not statistically significant. Given Seattleâs relatively benign climate, it can be said that most weather impacts in the Seattle region are small, at least in terms of the changes in vehicle speed and throughput that they directly cause. During most parts of the day, on most roadway segments, the travel time changes that these small differences in speed create are not statistically significant. However, when those small changes occur in combination with large traffic volumes, especially during the beginning shoulder of a peak period, those small changes can result in congestion that will, in turn, generate much more significant increases in expected travel times. The use of rain variables that account for the continuing presence of spray from wet roadways suggests that spray has as much of an impact on roadway performance as moderate rain- fall itself. Similarly, except in the case of heavy snowfall (when low visibility affects driversâ behavior), the major impacts of snow are the result of snow accumulation, not the snowfall itself. Anecdotal evidence of this same effect also was apparent for ice formation in Seattle. The project team attempted to com- pute times when black ice formation might be present by using humidity and temperature data from the Sea-Tac weather sta- tion. However, these factors did not result in successful identifi- cation of ice formation in the informal tests conducted during the winter of 2008. Therefore, the team concluded that using regional weather station data is not an effective way to accu- rately determine the presence of snow and ice on roadways. Table 5.6. Example Effects of Wind on Travel Times by Corridor Route Mean Travel Time A.M. Peak P.M. Peak With Winda (s) Without Wind (s) Difference (s) Statistically Significant? With Wind (s) Without Wind (s) Difference (s) Statistically Significant? I-5 Everett southbound 190 207 -17 No 191 209 -18 No I-5 North King southbound 759 690 68 Yes 400 422 -22 No I-5 North Seattle southbound 751 606 145 Yes 926 686 239 Yes I-5 South northbound 1,671 1,073 598 Yes 649 649 0 No SR 520 Seattle westbound 1,020 638 382 Yes 1,548 1,052 495 Yes I-90 Bridge Eastbound 425 410 15 No 543 437 106 Yes I-90 Seattle eastbound 198 169 29 Yes 151 115 36 Yes SR 520 Seattle westbound, 10 mph wind speed 781 626 154 Yes 1,093 1,049 44 Yes I-90 Bridge eastbound, 10 mph wind speed 434 407 27 No 431 441 -10 No I-90 Seattle eastbound, 10 mph wind speed 174 169 5 No 107 118 -12 No a Sustained wind speed is greater than 16 mph. b Sustained wind speed is less than or equal to 16 mph.
99 The effects of wind are similar to those of rain. High winds cause motorists to drive more cautiously. The degree to which they adjust their behavior for a given wind condition is a function of the roadway section: How wide are the lanes? Are there shoulders? How exposed is the roadway section to wind? This in turn reduces the functional capacity of the roadway during high-wind conditions. These effects do not appear to be as uniform as the effects of rain, since geographic differences in terrain and geometric differences in roadway right-of-way appear to play bigger roles in determining the effects of wind on roadway performance than they do in the case of rain. When wind is significant and traffic volumes are light, travel times increase only marginally, in direct proportion to Figure 5.7. Mean travel times by time of day in wind and no-wind conditions on SR 520 westbound, Bellevue toward Seattle. Figure 5.8. Mean travel times by time of day in wind and no-wind conditions on I-90 bridge section eastbound, Seattle toward Bellevue.
100 the slowing that individual vehicles exhibit under windy con- ditions. However, when volumes are high, the reduced func- tional roadway capacity resulting from motoristsâ voluntary slowing can create congestion that would not occur under aver- age weather conditions. That congestion frequently becomes self-sustaining during peak periods; that is, the queue itself creates a further decrease in functional roadway capacity, which further increases the length of the queue and increases travel times on the roadway section. In summary, the analysis of the impacts of bad weather on congestion formation on Seattle freeways identified the fol- lowing major conclusions: ⢠Small disruptions, such as those caused by moderate amounts of rain or even spray from wet pavements, only cause con- gestion when they occur in combination with sufficient volume relative to the available capacity; ⢠Precipitation can affect roadway performance as long as the roadway remains wet; ⢠The probability that bad weather will significantly affect roadway performance on any given roadway section is a function of the expected demand and capacity condition of that road section and the significance of the weather event (e.g., light rain versus a heavy thundershower); and ⢠Bad weather also increases the probability of crashes occurring, which further increases the probability of sig- nificantly increased travel times. Effects of Incidents and Crashes The effects of crashes and other kinds of traffic disruptions are of significant interest both because they are common causes of travel delay and because they are disruptions over which operating agencies have some level of control. That is, highway agencies cannot prevent rain, but they can design roadways to minimize the number and severity of crashes, and they can respond effectively and efficiently to crashes to limit their duration. Consequently, the project team looked at the effects of both crashes and noncrash incidents. Incidents and crashes differ from weather in three signifi- cant ways. First, incidents and crashes are highly correlated with traffic volume, while weather is not. More crashes occur when volumes increase, but increasing volumes do not affect rainfall. Therefore, crashes and incidents are not evenly dis- tributed over time, but bad weather (at least in Seattle) is much more evenly distributed throughout the day. Second, incidents and crashes have small footprints in comparison to weather. A crash or incident occurs at a spe- cific location, which has a relatively small geographic scope (this does not include any queues that may form), but the same weather generally occurs over a larger geographic area. This small footprint can have considerable impact on segment-based analysis procedures. This impact is discussed below in the subsection on methodology. Finally, crashes and incidents are, in many ways, even more variable than weather. Incidents can be anything from minor debris in the roadway (e.g., pieces of a blown truck tire), to a distraction on the side of the road (e.g., a stalled car), to a fatal crash. Methodology Considerable research has been conducted to explore the impacts of incidents on roadway performance, especially in terms of vehicle throughput, queue formation, and roadway recovery at the incident scene. Much of this work has involved the use of queuing theory to explore the size and speed of queue formation, given incoming and exiting traffic volumes, along with descriptors of specific incidents (duration, num- ber of lanes closed). The intended result of most of these efforts has been to determine the benefits that can be gained from improvements in incident response efforts. One limitation in these studies has been the fact that once queues form during peak periods, the queue itself can become its own self-sustaining bottleneck. Thus, even after the inci- dent has been cleared, the back of the queue may become the point at which congestion forms, effectively replacing the incident scene that started the congestion. A second limita- tion is that a bottleneck at one point of a roadway segment has implications on the performance of the rest of that road- way segment, as well as the segments upstream and down- stream from that segment. Consequently, this project used two approaches to examine the larger, corridor-long effects of incidents and crashes. The first approach examined the travel times that occur under incident or crash conditions. This analysis took advantage of influence variables (these are discussed above and in Appen- dix A). As described, the influence of every crash and incident was noted in the 5-minute travel time records for each road- way test segment. It was possible, for any definition of disrup- tion, to segregate the travel time records for a given test section into two groups: those influenced by a specific type of disrup- tion and those not influenced by that type of disruption. Statistical tests could then be performed on those two groups. Because of the time series nature of travel times, combined with the time-lagged nature of the effects of inci- dents, these statistical comparisons were somewhat complex. That is, traffic conditions at 7:00 a.m. on a Monday are differ- ent than those at 8:00 a.m. for that same stretch of road, so travel times at these two times should not be directly com- pared. Similarly, a crash that happens at 7:00 a.m. has a differ- ent effect on travel time at 7:05 a.m. than it has at 7:15 a.m. Because disruptions happen at different times during the day, the aggregated effects of these disruptions are complex.
101 The primary statistical test used to compare influenced and noninfluenced travel times was an independent sample t-test. The majority of tests involved only data for Tuesdays, Wednes- days, and Thursdays to limit the effects that variations in day- of-week traffic volumes would have on the statistical results. This test was originally applied independently for each 5-min- ute period. That is, influenced travel time data for the 7:00 to 7:05 a.m. period for all Tuesdays through Thursdays were compared with noninfluenced travel times for that one period. Because each 5-minute time period occurred on a different day, each sample was truly independent of all other samples; that is, the 7:00 a.m. travel time today has no influence on the 7:00 a.m. travel time tomorrow. Because travel times were taken from only one 5-minute period, the time-dependent effects of travel also were removed. The difficulty with this approach is that it required per- forming 288 statistical tests to examine the daily differences in incident-influenced and noninfluenced travel times. To reduce the analytic load, the project team grouped the 5-minute average travel times by 30-minute increments, with the statis- tical tests performed for each 30-minute interval. In this approach, the six 5-minute travel times were treated as independent travel time estimates within that 30-minute period. For example, assume that no incident happens on a study corridor on March 7 until the 7:15 a.m. period. That incident influences the rest of the morning commute. The average 5-minute travel times stored in the 7:00, 7:05, and 7:10 a.m. analysis time periods are reported as not incident influenced. All three 5-minute average travel times are included in the computation of the travel time distribution for the not incidentâinfluenced 30-minute period covering 7:00 to 7:29 a.m., and the three 5-minute periods from 7:15 to 7:25 a.m. are included in the influenced travel time distri- bution for that same 7:00 to 7:29 a.m. period. There were two advantages to the 30-minute approach. One was the reduction in the number of statistical tests that had to be performed and summarized. The second was the increase in the sample size for each test. The downside of the 30-minute test was that the six travel times were no longer truly independent samples, as the 7:05 a.m. travel time would be highly correlated to the 7:00 a.m. travel time. When the results of tests conducted with both levels of aggregation were analyzed, little difference was found between the statistical outcomes of the 5- and 30-minute comparisons, so most analysis results in this report are pre- sented in the 30-minute format to make the results more readable. When the results of the 5- and 30-minute analyses were compared, the most significant differences were found in the shoulders of the peak period. These differences did not change any of the basic conclusions of this report. Statistical comparisons between influenced and noninflu- enced travel times were made in a number of ways. Various comparisons were possible because of the multiple ways that influence was calculated in the project database. Influence was examined for crashes (only crashes reported in the state accident records), for incidents (any incident reported by WSDOTâs service patrol), for any incident reported by WITS that involved lane closures, or for any one of these types of disruptions. Travel times associated with these disruptions could then be compared with either all other travel times or only travel times when no disruption influenced travel. This flexibility allowed a very thorough comparison of incident-influenced conditions. In most cases, the best com- parison was with no known disruption currently influencing conditions, but in some cases it was important to make a comparison with all other travel times (e.g., comparing travel times when crashes had influenced travel versus noncrash- influenced travel). In most cases, nonholiday Tuesday through Thursday travel times were used as the population for which travel times were compared. Some analyses also were performed for weekends and for all weekdays combined. Although these analyses were useful for describing total delay in a year caused by a specific type of disruption, they were not as useful in describing the effects of disruptions on travel times compared with normal conditions. Therefore, most results presented in this report involve Tuesday through Thursday (nonholiday) comparisons. One difficulty with these comparisons is that they were not measures of what would have happened if the disrup- tion had not taken place. They were simply comparisons of the expected conditions when a specific type of disruption occurred versus expected conditions when those types of events had not taken place. The research team hoped that by combining an entire yearâs worth of data, the number of events included in the database would limit the biases in travel time impacts that could be associated with specific incidents occurring at specific times and locations. To make a direct comparison of actual conditions versus what would have happened would require a carefully calibrated microscale simulation model. Such an effort was well beyond the scope of this project. Because they are not direct measures of what would have happened, the resulting graphs and computed statistics must be used carefully. They describe the differences in expected conditions if a specific type of event has occurred and its influ- ences are still being felt. That second clause is important. One problem with not using a simulation to make this comparison is answering the question, when does the influence of an event end? The travel time comparisons assumed that the effects of any disruption ended once conditions returned to what they were at the time the disruption took place, not the condition that would normally be present at that time. This definition was selected because a review of the project data set found
102 many cases in which predisruption travel times were much faster than normal; when the disruption occurred, travel times slowed, but they never degraded to the point of normal con- ditions. Moreover, travel times returned to the faster-than- normal conditions that existed before the disruption. If normal travel times had been used as the measure of influ- ence, these events would have had no influence. But they obvi- ously caused delay. As a result, the definition of influence was based on travel times returning to preexisting conditions. A second limitation with the corridor-based analysis process described above was caused by the site-specific nature of crash and incident impacts relative to the roadway segmentation used for the analysis. The disadvantage of using travel times is that travel time is a function of selected segment end points, and those defined segments may or may not include all the effects (e.g., slow-moving vehicles) caused by a given incident. Figures 5.9 through 5.11 illustrate this problem. Taken together, they show how the location of a crash or inci- dent within a corridor can influence how effectively the mea- sured travel times in a test section reflect the delays caused by that crash or incident. In Figure 5.9, the crash occurs near the downstream end of the roadway segment. In this case, travel times measured in the corridor capture all the delays occurring in the test section, Figure 5.9. Illustration of a crash at the downstream end of a test corridor. Figure 5.10. Illustration of a crash in the middle of a test corridor. Figure 5.11. Illustration of a crash at the upstream end of a test corridor.
103 unless the queue is longer than the test section. This situation did happen on the test sections, but given the 2-mile mini- mum length of those test sections, it was unusual. In Figure 5.10, the crash occurs in the middle of the test section. In this case, if the queue is minor, the entire queue and the travel time influences of that queue are contained in the test corridor. However, if the queue is long, it will extend back into the upstream roadway segment, creating delays on that segment that are not explained by an incident or crash within that segment. Thus, the study segment that contains the crash will see some, but not all, of the delays associated with the crash, while the upstream segment will see unexplained congestion. In Figure 5.11, the crash occurs near the upstream end of the study segment. In this case, the study segment will not experience the majority of the delays caused by the crash. Those delays will occur on the test section upstream of the study section. The study section is likely to show good travel times because in this study, travel times are based on multiple- point speed measurements, and the queues at the upstream end will allow the majority of the study segment to operate in a free-flow condition. The moderately long roadway segments and the careful selection of the breakpoints between those segments in this study limited the frequency with which congestion crossed segment boundaries, but there were still many occasions when this happened. The travel time analyses presented in the following section do not effectively account for these cross- segment boundary occurrences. When they occurred, the slower travel times these extended queues caused were associ- ated with normal (or nonincident) conditions. As a result, the comparisons between incident-influenced and nonincident- influenced conditions described below should be considered conservative measures of the effects of incidents on travel times and travel time reliability, as many off-segment effects of crashes and incidents were not accounted for. Although they are useful in describing the effects that dif- ferent types of disruptions have on travel time, the definition of influence described above and the statistical travel time comparison based on that definition have significant explan- atory limitations. In particular, analysis using this definition of influence does not do a good job of answering questions such as, âWhat impact does a crash have on my commute?â The different times and locations of such a disruption will result in different outcomes, and it cannot be known when the individual asking that question makes his trip. Conse- quently, a second type of analysis was performed that exam- ined changes in congestion from a different perspective. In this second set of analyses, the study team defined when congestion ends at the end of both the a.m. and p.m. peak periods. The idea came from two observations noted in the development of the influence variables: (a) once congestion starts (often as a result of a disruption) during the peak period, that congestion tends to last until the end of the peak; and (b) although the previously described analysis can predict how much longer a given trip will last once a disruption has occurred, it does not estimate how long the congestion effect will last. Determining how much longer congestion lasts would provide insight into that missing piece of information. To perform the required analysis, end of congestion was defined as the time when 20 consecutive minutes (four 5-minute periods) of travel time were less than travel time at the speed limit plus 5%. The 20-minute interval was selected to account for modest fluctuations in travel times (vehicle speeds) caused by unstable traffic flow occurring as conges- tion eases. The 5% value was selected as a result of sensitivity tests; while it represents a fairly small increase in travel time, it does appear to identify the effects of modest congestion that occur at a single location within a longer corridor. Once the end of congestion was identified for each peak period for each day, three sets of travel time statistics were computed for all nonholiday Tuesdays, Wednesdays, and Thursdays describing the time that congestion ended for days when (a) any crash occurred (the crash must have occurred after 4:00 a.m. for the morning peak period test or after 3:00 p.m. for the evening peak period test), (b) any noncrash incident occurred, or (c) no incident occurred. Only one end of congestion time was assigned for each peak period for each day; that is, the first time period that met the selected criteria was the end of congestion for that peak period. Occasionally disruptions of one type occurred after congestion ended, cre- ating a second congestion period within the traditional hours associated with the peak period. These cases were treated as occurring after the peak period had ended. These statistics were compared by using both normal and nonparametric statistical tests to determine the extent to which crashes and other types of traffic disruptions can be expected to extend peak period congestion. A problem arose in that the definition for end of congestion proved too strict for some segments. The mean time when the a.m. peak period congestion ended was well after noon on 11 test sections, and frequently it did not end until after 6:00 p.m. on these corridors. A review of the travel times routinely expe- rienced on these routes showed that a variety of traffic flow conditions (e.g., excessive merging at bottlenecks near the end of the corridor, large volumes of heavy trucks) frequently kept these road segments operating slightly below the speed limit even during late morning and midday periods. These routes all operated at or above the speed limit during late-night hours and during many midday hours. But they routinely operated at speeds lower than the speed limit during the middle of the day for reasons other than traffic disruptions. This normal condition limited the benefit of the intended analysis. As a result, for the a.m. peak period on these 11 routes,
104 end of congestion was redefined as being sustained speeds within either 10% or 20% of the speed limit, depending on the corridor. The intent of this new, corridor-specific definition was simply to allow better examination of how crashes and other disruptions affect when slow travel associated with peak period volumes ends. These lowered expectations were tested on other corridors and for other periods. The results were generally not good. Using lowered average travel speeds to define end of peak period congestion frequently caused the end of congestion flag to be set during obviously congested conditions on these other routes. This was particularly true in the afternoon peak period, when all routes reached travel times within 5% of that achieved at the speed limit by a reasonable time of day. Con- sequently, the slower speed that was required to allow this approach was used only for those 11 roadway segments and only for the a.m. peak period. Results: Travel Time Effects of Incidents and Crashes In general, the effects of crashes and incidents on travel times were similar to each other and to the expected travel times that resulted from rainfall. That is, the shape of the expected (mean) travel time patterns by time of day when incidents and crashes occurred was similar in shape to the expected travel times when rain fell. These similarities are illustrated in Figures 5.12 through 5.14. Figures 5.12 through 5.14 illustrate the mean travel time for nonholiday Tuesdays, Wednesdays, and Thursdays for all of 2006 (a) under nonrain conditions (regardless of incident conditions), (b) when rain had fallen within the past hour (regardless of incident conditions), (c) when a crash on the study section was influencing traffic conditions, and (d) when any traffic incident was reported as occurring on the study section by WSDOTâs service patrols. Thus, the four expected travel time conditions are not fully independent of each other. But each gives an excellent understanding of expected conditions. For example, the rain travel time line answers the travelerâs question, âHow long should I expect my commute trip to last on this corridor if it is raining?â The response includes days when crashes occur and others when they do not occur. Note that during any given period the crash and incident travel time curves drop to zero when there were no reported crashes or incidents. As the curves show, when free-flow conditions are the rou- tine condition, incidents and rain have little effect on mean travel times. In some cases, crashes create sufficient disrup- tion that travel times increase in lower-volume periods. In the figures, the relative size of the travel time changes measured during incident and crash conditions (e.g., com- pared with the no-rain condition) is not consistent from cor- ridor to corridor. These differences are caused by a variety of factors, including differences in (a) the sizes of the incidents and crashes occurring on each study segment during 2006, (b) the locations of the incidents and crashes relative to the end points of each study segment, and (c) the volume-to- capacity ratio occurring on the study section at the time of the traffic disruption. Perhaps even more importantly, the travel time statistics do not account for off-segment traffic Figure 5.12. Mean travel times under rain, crash, or noncrash traffic incident conditions on I-5 northbound, South corridor.
105 disruptions. That is, case studies of a number of specific days in 2006 showed that congestion on one roadway segment can frequently grow to the point that it affects the upstream road segment. While roadway segment boundaries can be chosen to minimize the effects of known geometric bottlenecks, major traffic disruptions often create temporary bottlenecks that are not located at known bottleneck locations. The con- gestion on study segments caused by these off-segment events increased the noninfluenced travel times against which study outcomes were compared. The combined result of these various factors is that the relative importance of any specific type of traffic disruption varies from study segment to study segment. In the northbound I-5 South segment (Figure 5.12), rain had a more substantial effect on the a.m. peak period travel times than did crashes. Late at night (midnight to 2:00 a.m.), Figure 5.13. Mean travel times under rain, crash, or noncrash traffic incident conditions on I-5 southbound, Lynnwood corridor. Figure 5.14. Mean travel times under rain, crash, or noncrash traffic incident conditions on I-5 southbound, North Seattle corridor.
106 the disruption imposed on the traffic stream. Therefore, crashes frequently have more significant effects during times of lower volume. But during peak conditions, the simple cre- ation of congestion, which can occur given a much smaller disruption, may be as significant as the size of the disruption itself. That is, once the roadway congests, a large disruption adds only a marginal increase to the delay, whereas a smaller disruption occurring before congestion forms can create an even larger change in expected travel times during the course of the peak period because of the growth of the queue associ- ated with the initial congestion point. Results: Effects of Crashes or Noncrash Incidents on Peak Period Travel Time and Travel Reliability The previous section illustrates that traffic volumes during Seattleâs peak periods are sufficient on many corridors to cre- ate congestion, and that congestion may result in a variety of travel times. When the effects of disruptions were added to those traffic volumes, travel times generally increased, as illustrated in Figures 5.12 through 5.14. When computing incident-influenced travel times, only incidents that had a still-active effect on roadway performance were considered (still active means that travel times in the test section were slower than measured when the disruption was actually in place). One difficulty with this approach is that it is hard to explain. It also does not generalize well. For a different approach to looking at the effects of traffic disruptions on travel times, this study computed the expected mean, 80th percentile, and 95th percentile peak period travel times for each study corridor, accounting for whether a dis- ruption (crash or noncrash reported incident) had taken place. This approach basically answers the travelerâs question, âIf a crash (or other noncrash disruption) occurs today, how much worse will my commute be?â To analytically answer this question, each nonholiday Tuesday through Thursday, 5-minute travel time was placed in one of three categories: (a) not influenced, (b) influenced by a crash, or (c) influenced by a reported noncrash incident. Once a disruption had occurred during a peak period, all remaining 5-minute travel times for the rest of that peak period were assumed to be influenced by that event. The a.m. peak was assumed to occur between 6:30 and 9:30 a.m. Any disruption that occurred after 4:00 a.m. was included in the analysis. The p.m. peak was assumed to occur between 3:00 and 7:00 p.m. Only traffic disruptions that occurred after 2:00 p.m. were included in the analysis. If a crash occurred at 5:00 p.m., the 5-minute travel times before 5:00 p.m. were classified as noninfluenced, and those after 5:00 p.m. were crash influenced. If both a crash and a noncrash incident occurred, all time periods after the crash were considered crash influenced. Because the mean, 80th percentile, and 95th incidents were seen to have a significant impact. A review of these data indicated that the incidents in question occurred during a planned construction lane closure, resulting in a large roadway capacity reduction during that maintenance activity, with substantial congestion being the result. In the middle of the afternoon and during the p.m. peak period (when the northbound I-5 South corridor operates in the reverse of the peak direction and is, therefore, not usually congested), crashes were the primary causes of travel time delays. The Lynnwood corridor (Figure 5.13) presented the most normal effects of both weather and traffic disruptions. No late-night congestion is apparent in the figure, although some late-evening delays (~9:00 p.m.) are evident as a result of vehicle crashes. In the a.m. peak period, rain had the greatest effect in terms of increasing expected travel times. Both rain and vehicle crashes tended to cause travel delays slightly ear- lier in the a.m. peak period than did incidents, which tracked more closely to the normal peak period travel times until almost the peak of the a.m. travel time curve, when the effects of incidents caused substantial additional travel time. In the p.m. peak period (again, on this corridor the p.m. peak is a reverse-direction commute), only modest increases in travel times due to rain, incidents, or crashes occurred, with crashes having the most significant impact. On the North Seattle southbound corridor (Figure 5.14), travel times routinely degrade in both peak periods. This cor- ridor differed from the other two examples in that crashes had a more significant impact on mean travel time in the a.m. peak than did rain. This is partly due to the fact that this corridor ends in two back-to-back C-class weaving sections that constitute both a major routine bottleneck and a high- accident location. The result is that most of the causes of congestion in this section occurred within this section. Con- gestion spillback from downstream roadway segments on rainy days was not as significant a factor on this section as it was on the Lynnwood section. Consequently, crashes were more often a factor, especially in the morning. A comparison of the three figures indicates that Figure 5.14 shows more off-peak congestion than Figures 5.12 and 5.13. The southbound I-5 North Seattle roadway corridor carries considerable traffic volume relative to the roadwayâs capacity even in off-peak periods. This large traffic volume frequently results in moderate southbound congestion, even in the mid- dle of the day. As a result, relatively minor traffic incidents or bad weather can start with a moderate situation in the middle of the day and make it considerably worse. In contrast, Fig- ures 5.12 and 5.13 show that traffic disruptions have relatively little impact on midday and evening roadway performance on the other example roadway segments. Thus, the impacts of any disruption are a function of the underlying traffic volume condition during which that dis- ruption occurs. The next most important factor is the size of
107 percentile travel times were computed from the entire pool of travel times within each classification of trips, this approach did create a minor bias toward lower travel times in the nonin- fluenced category, as a disproportionate number of travel times for that category were taken from the early (least congested) portion of the peak periods. This bias was somewhat balanced by the inability of this analysis to account for the effects of con- gestion spillback from one roadway segment to another. The results of this analysis are shown in Table 5.7, which describes the impacts of crashes and noncrash incidents on the mean travel times computed for the a.m. peak period. For each study corridor, the mean travel time increase (in seconds) Table 5.7. Effects of Incidents and Crashes on A.M. Peak Period Travel Times Study Corridor A.M. Peak Travel Rate Mean Travel Time Increase from All Traffic Incidents (s) Increase over Nonincident Conditions (%) Mean Median Noncrash Incident Crash I-405 Kennydale northbound 3.66 3.4 179 11 17 I-405 North southbound 2.82 2.4 347 35 45 I-5 North King southbound 2.07 1.8 139 22 43 I-5 Seattle CBD northbound 1.91 1.8 361 51 57 I-405 Kirkland southbound 1.76 1.8 80 9 14 SR 520 Seattle eastbound 1.70 1.8 98 13 32 I-5 Lynnwood southbound 1.89 1.6 251 31 60 I-5 South northbound 1.75 1.6 364 43 58 SR 167 Auburn northbound 1.68 1.6 21 8 15 I-405 Eastgate northbound 1.66 1.6 17 8 24 I-5 Seattle North southbound 2.15 1.4 232 47 84 I-405 Kennydale southbound 1.54 1.4 96 15 34 I-405 South southbound 1.45 1.4 50 28 14 SR 167 Renton northbound 1.62 1.2 390 75 76 SR 520 Seattle westbound 1.51 1.2 183 30 19 I-5 Tukwila northbound 1.50 1.2 254 57 76 I-90 Issaquah westbound 1.46 1.2 169 29 60 I-90 Bellevue westbound 1.30 1.2 73 24 22 I-405 Bellevue northbound 1.27 1.2 39 12 27 I-405 South northbound 1.24 1.2 30 19 15 I-90 Seattle eastbound 1.96 1 50 27 36 I-90 Seattle westbound 1.20 1 20 21 27 I-90 Bridge Eastbound 1.18 1 40 9 38 I-405 Bellevue southbound 1.16 1 25 11 62 I-5 Everett southbound 1.15 1 39 20 90 I-90 Bridge westbound 1.15 1 40 11 22 I-5 Seattle CBD southbound 1.10 1 48 9 24 SR 167 Auburn southbound 1.06 1 -2 -1 -6 I-405 Eastgate southbound 1.05 1 9 7 76 (continued on next page)
108 caused by noncrash traffic incidents is presented. This increase is then shown as a percentage change in study section travel time in comparison with the mean travel time with no disrup- tion. The percentage increase in travel time associated with a crash is shown to illustrate the relative significance of crashes and noncrash traffic disruptions. The 42 study segments are sorted from most congested to least congested on the basis of their median and mean travel rates for all weekdays. As Table 5.7 shows, the mean travel time increased when traffic disruptions occurred for all corridor study segments that had a mean travel rate greater than 1.0. (A travel rate equal to 1.0 indicates that vehicles can operate at the speed limit [60 mph].) For all but four of those corridors, the occur- rence of a crash had a greater impact on expected travel times than a reported noncrash incident. A more mixed effect of both crashes and noncrash incidents is evident for corridors that did not routinely exhibit at least a modest level of con- gestion. No direct correlation is observable between the delays that occurred in response to traffic incidents and either the mean or median travel rates. The p.m. peak period version of Table 5.7 is shown in Table 5.8. As with the a.m. peak results, all the p.m. corridors with a median travel rate greater than 1.0 showed increases in mean travel time when any kind of traffic disruption occurred. Crashes resulted in a greater increase in the mean travel time than noncrash incidents on all but four of the study corridors. Because p.m. peak travel is different from a.m. peak travel, the corridors in Table 5.8 do not match those in Table 5.7. Other than the basic, if obvious, conclusion that traffic dis- ruptions can be expected to increase travel times for moder- ately to heavily congested travel corridors, there are relatively few patterns in the data contained in Tables 5.7 and 5.8. There appears to be no consistent relationship between the percent- age change in travel time and the base statistics that describe mean peak period travel conditions (either mean travel rate or median travel rate). On some heavily congested corridors (e.g., I-405 Bellevue southbound p.m. peak, I-5 North Seattle south- bound p.m. peak, I-5 South northbound a.m. peak), crashes and other incidents caused dramatic increases in expected travel times, even doubling the expected time to traverse the study section. On other heavily congested corridors (e.g., I-405 Eastgate southbound p.m. peak, I-405 Kennydale northbound a.m. peak), the travel time effects were considerably smaller, in the range of a 10% to 25% increase in expected travel times. When looked at more comprehensively, noncrash inci- dents increased travel times an average of 17% in the morn- ing and 21% in the evening on corridors that had mean peak period travel rates above 1.10. However, mean travel time changes ranged from 9% to 75% in the morning. In the eve- ning, travel times changes ranged from 6% to 119%. If only crashes are considered, the a.m. peak changes ranged from 14% to 90%, with an average of 40%. The p.m. changes ranged from 9% to 176%, with an average of 41%. SR 167 Renton southbound 1.04 1 -1 0 -6 I-5 Tukwila southbound 1.02 1 14 3 -2 SR 520 Redmond westbound 1.02 1 29 8 7 I-405 North northbound 1.02 1 8 2 4 I-5 Everett northbound 1.01 1 5 3 51 I-5 Lynnwood northbound 1.01 1 16 3 52 I-5 Seattle North northbound 1.01 1 3 1 0 I-90 Bellevue eastbound 1.01 1 -1 -1 0 I-5 South southbound 1.00 1 23 4 3 I-405 Kirkland northbound 1.00 1 7 1 1 SR 520 Redmond eastbound 1.00 1 1 0 0 I-90 Issaquah eastbound 1.00 1 -1 0 0 I-5 North King northbound 1.00 1 1 0 0 Table 5.7. Effects of Incidents and Crashes on A.M. Peak Period Travel Times (continued) Study Corridor A.M. Peak Travel Rate Mean Travel Time Increase from All Traffic Incidents (s) Increase over Nonincident Conditions (%) Mean Median Noncrash Incident Crash
109Table 5.8. Effects of Incidents and Crashes on P.M. Peak Period Travel Times Study Corridor P.M. Peak Travel Rate Mean Travel Time Increase from All Traffic Incidents (s) Increase over Nonincident Conditions (%) Mean Median Noncrash Incident Crash I-405 Bellevue southbound 3.73 3.6 400 88 102 I-405 Eastgate southbound 2.73 2.6 29 10 25 SR 520 Seattle westbound 2.72 2.6 230 23 18 I-405 South northbound 2.58 2.6 47 17 14 I-5 Seattle North southbound 2.56 2 410 119 138 I-405 Kirkland northbound 1.99 2 127 14 26 I-5 Seattle CBD northbound 1.96 1.8 350 52 60 I-405 Kennydale southbound 1.90 1.8 109 15 23 I-5 North King northbound 1.79 1.8 92 17 24 I-5 Seattle CBD southbound 1.72 1.8 153 22 30 SR 167 Auburn southbound 1.96 1.6 90 29 33 SR 520 Redmond eastbound 1.87 1.6 83 14 34 I-5 South southbound 1.76 1.6 265 30 46 I-405 North northbound 1.61 1.6 37 6 29 I-5 Everett northbound 1.87 1.4 128 50 55 I-5 Seattle North northbound 1.74 1.4 73 18 29 SR 167 Renton southbound 1.63 1.4 180 31 57 I-405 South southbound 1.52 1.4 79 43 26 I-90 Bridge westbound 1.73 1.2 122 25 12 SR 520 Seattle eastbound 1.49 1.2 115 20 34 I-5 Lynnwood northbound 1.38 1.2 101 17 45 I-405 Bellevue northbound 1.34 1.2 89 35 68 I-90 Seattle westbound 1.13 1.2 7 8 9 SR 520 Redmond westbound 1.49 1 168 38 40 I-90 Seattle eastbound 1.43 1 84 72 54 I-90 Bridge eastbound 1.40 1 111 27 35 I-5 North King southbound 1.33 1 232 67 176 I-90 Bellevue westbound 1.30 1 154 63 96 I-5 Tukwila southbound 1.19 1 102 22 66 SR 167 Renton northbound 1.17 1 151 37 40 I-405 Kennydale northbound 1.17 1 84 17 56 I-90 Bellevue eastbound 1.11 1 48 19 50 I-5 Everett southbound 1.10 1 16 8 63 I-5 Lynnwood southbound 1.10 1 59 11 44 I-405 North southbound 1.09 1 64 16 37 I-405 Kirkland southbound 1.09 1 101 19 27 I-5 Tukwila northbound 1.07 1 96 24 81 SR 167 Auburn northbound 1.05 1 13 6 16 I-405 Eastgate northbound 1.04 1 19 16 18 I-90 Issaquah eastbound 1.01 1 0 0 -1 I-5 South northbound 1.01 1 14 2 6 I-90 Issaquah westbound 1.00 1 17 4 5
110 A review of base data for a sample of these corridors sug- gested that two factors contributed to this variation. In some cases, the noninfluenced annual mean travel time was signifi- cantly affected by downstream congestion when that down- stream congestion was caused both by routine conditions and by traffic disruptions on the downstream roadway segments. The result of this downstream congestion backing up on the study section was that an abnormally high mean travel time for nondisruption-influenced travel times occurred on the study section. This result, in turn, decreased both the absolute and percentage differences in crash-influenced travel times. The second factor was simply the number and variety of incidents or crashes occurring in the different test sections. Some traffic disruptions were more significant in terms of the number of lanes they blocked and the time at which they occurred. A modest number of very bad traffic disruptions can cause a fairly high increase in the mean travel time because of the modest number of data points in each sample. To further explore the effects of incidents and crashes on travel time reliability, Tables 5.9 and 5.10 describe the mea- sured changes in the 80th and 95th percentile travel times when crashes and noncrash incidents occur. Similar to Tables Table 5.9. Effects of Crashes and Noncrash Incidents on A.M. Peak Period 80th and 95th Percentile Travel Times Study Corridor Mean A.M. Peak Travel Rate Increase in Travel Time (%) Noncrash Incident Crash 80th Percentile 95th Percentile 80th Percentile 95th Percentile I-405 Kennydale northbound 3.66 6.3 10.9 9.4 8.2 I-405 North southbound 2.82 -6.4 9.6 2.1 13.5 I-5 North King southbound 2.07 0.0 -5.3 16.8 25.8 I-5 Seattle CBD northbound 1.91 15.4 17.1 40.5 35.4 I-405 Kirkland southbound 1.76 -2.2 -4.6 1.1 2.3 SR 520 Seattle eastbound 1.70 5.6 12.0 20.5 52.5 I-5 Lynnwood southbound 1.89 -16.0 -15.4 9.5 15.0 I-5 South northbound 1.75 -18.6 -16.8 -6.1 -0.1 SR 167 Auburn northbound 1.68 2.3 7.5 18.3 37.9 I-405 Eastgate northbound 1.66 5.5 6.7 6.8 30.9 I-5 Seattle North southbound 2.15 2.5 -0.1 31.3 24.7 I-405 Kennydale southbound 1.54 -1.7 10.1 21.7 27.8 I-405 South southbound 1.45 2.7 20.3 -0.5 17.0 SR 167 Renton northbound 1.62 -4.2 -14.7 84.8 61.8 SR 520 Seattle westbound 1.51 -0.2 -4.1 17.9 12.7 I-5 Tukwila northbound 1.50 -5.4 -13.4 16.8 12.3 I-90 Issaquah westbound 1.46 21.0 0.5 28.3 36.2 I-90 Bellevue westbound 1.30 -0.5 30.2 12.6 10.8 I-405 Bellevue northbound 1.27 18.5 23.9 33.4 35.2 I-405 South northbound 1.24 0.6 -0.5 2.7 9.7 I-90 Seattle eastbound 1.96 -2.1 -9.7 23.3 45.8 I-90 Seattle westbound 1.20 9.2 -5.9 35.0 11.8 I-90 Bridge eastbound 1.18 34.1 12.7 51.2 41.5 I-405 Bellevue southbound 1.16 0.3 -5.1 93.2 114.6 I-5 Everett southbound 1.15 -2.6 -29.2 4.1 17.6 (continued on next page)
111 I-90 Bridge westbound 1.15 0.0 -1.0 56.2 162.0 I-5 Seattle CBD southbound 1.10 1.8 -1.3 13.0 28.6 SR 167 Auburn southbound 1.06 2.7 8.8 No crashes a.m. peak I-405 Eastgate southbound 1.05 0.3 -0.1 6.5 76.5 SR 167 Renton southbound 1.04 2.5 1.3 0.8 0.1 I-5 Tukwila southbound 1.02 0.0 -0.4 0.4 0.4 SR 520 Redmond westbound 1.02 0.6 12.7 29.3 76.6 I-405 North northbound 1.02 -0.2 1.1 1.2 6.2 I-5 Everett northbound 1.01 -0.1 -0.1 1.2 38.4 I-5 Lynnwood northbound 1.01 0.1 0.1 0.6 195.6 I-5 Seattle North northbound 1.01 2.3 2.9 5.9 5.3 I-90 Bellevue eastbound 1.01 0.0 0.9 0.0 0.7 I-5 South southbound 1.00 0.0 0.0 0.0 19.6 I-405 Kirkland northbound 1.00 0.2 2.4 0.2 0.2 SR 520 Redmond eastbound 1.00 -10.5 -14.9 -9.3 -16.5 I-90 Issaquah eastbound 1.00 0.0 0.0 0.0 0.0 I-5 North King northbound 1.00 0.0 0.0 0.0 0.0 Table 5.9. Effects of Crashes and Noncrash Incidents on A.M. Peak Period 80th and 95th Percentile Travel Times (continued) Study Corridor Mean A.M. Peak Travel Rate Increase in Travel Time (%) Noncrash Incident Crash 80th Percentile 95th Percentile 80th Percentile 95th Percentile Table 5.10. Effects of Crashes and Noncrash Incidents on P.M. Peak Period 80th and 95th Percentile Travel Times Study Corridor Mean P.M. Peak Travel Rate Increase in Travel Time (%) Noncrash Incident Crash 80th Percentile 95th Percentile 80th Percentile 95th Percentile I-405 Bellevue southbound 3.73 9.8 -3.4 10.4 0.3 I-405 Eastgate southbound 2.73 3.0 -5.8 6.4 27.3 SR 520 Seattle westbound 2.72 21.4 4.1 25.7 1.7 I-405 South northbound 2.58 12.4 7.8 7.0 8.4 I-5 Seattle North southbound 2.56 5.2 1.3 21.8 14.9 I-405 Kirkland northbound 1.99 4.6 3.6 18.7 27.6 I-5 Seattle CBD northbound 1.96 29.5 26.0 52.4 30.8 I-405 Kennydale southbound 1.90 -11.6 9.0 -6.4 -0.1 I-5 North King northbound 1.79 12.6 10.9 11.3 17.1 (continued on next page)
112 I-5 Seattle CBD southbound 1.72 2.4 -3.1 5.2 12.7 SR 167 Auburn southbound 1.96 0.6 22.5 29.4 11.2 SR 520 Redmond eastbound 1.87 -10.5 -14.9 -9.3 -16.5 I-5 South southbound 1.76 10.3 9.8 16.8 34.9 I-405 North northbound 1.61 8.4 30.5 27.4 59.5 I-5 Everett northbound 1.87 -6.8 -0.2 -3.4 2.5 I-5 Seattle North northbound 1.74 9.4 11.4 18.3 0.1 SR 167 Renton southbound 1.63 15.9 16.3 48.2 67.7 I-405 South southbound 1.52 5.1 6.9 7.8 22.6 I-90 Bridge westbound 1.73 48.8 29.6 47.5 13.4 SR 520 Seattle eastbound 1.49 24.4 21.3 32.1 42.9 I-5 Lynnwood northbound 1.38 12.8 2.4 43.1 60.9 I-405 Bellevue northbound 1.34 7.3 -8.0 54.5 68.1 I-90 Seattle westbound 1.13 0.6 7.7 1.7 9.5 SR 520 Redmond westbound 1.49 169.2 23.4 171.8 50.9 I-90 Seattle eastbound 1.43 25.9 -31.4 33.4 13.2 I-90 Bridge eastbound 1.40 51.3 15.4 83.5 21.5 I-5 North King southbound 1.33 9.9 86.8 151.5 114.6 I-90 Bellevue westbound 1.30 494.3 244.6 213.0 107.2 I-5 Tukwila southbound 1.19 8.4 7.9 48.6 18.0 SR 167 Renton northbound 1.17 6.3 23.1 26.3 44.4 I-405 Kennydale northbound 1.17 7.1 -3.9 40.4 98.1 I-90 Bellevue eastbound 1.11 3.2 -0.2 19.0 419.0 I-5 Everett southbound 1.10 5.9 8.6 57.5 149.3 I-5 Lynnwood southbound 1.10 -0.4 -7.2 20.6 41.3 I-405 North southbound 1.09 0.3 45.0 58.1 41.6 I-405 Kirkland southbound 1.09 6.6 19.3 20.9 29.1 I-5 Tukwila northbound 1.07 2.1 -6.7 113.7 146.3 SR 167 Auburn northbound 1.05 5.4 139.6 61.0 58.6 I-405 Eastgate northbound 1.04 -2.0 -6.8 26.7 142.6 I-90 Issaquah eastbound 1.01 1.2 1.0 0.1 -6.0 I-5 South northbound 1.01 -0.2 -0.1 4.7 77.8 I-90 Issaquah westbound 1.00 0.1 3.5 -0.4 -0.5 Table 5.10. Effects of Crashes and Noncrash Incidents on P.M. Peak Period 80th and 95th Percentile Travel Times (continued) Study Corridor Mean P.M. Peak Travel Rate Increase in Travel Time (%) Noncrash Incident Crash 80th Percentile 95th Percentile 80th Percentile 95th Percentile
113 5.7 and 5.8, these two tables are sorted from most congested to least congested study corridor. Table 5.9 presents the changes to a.m. peak period travel times, and Table 5.10 pre- sents the p.m. peak period results. As the tables show, in most cases, crashes had a greater impact than noncrash traffic incidents in both the a.m. and p.m. peak periods. In addition, the least congested corridors in both peak periods generally showed the least change in the measured 80th and 95th percentile travel times when crashes and other traffic incidents occurred. The most significant difference was that all corridors with median peak period travel rates for all weekdays above 1.0 or mean weekday travel rates above 1.10 showed an increase in mean travel times on days when either a crash or noncrash incident occurred. However, many corridors did not show increased 80th or 95th percentile travel times under those same incident conditions, especially for noncrash incidents. The effects of noncrash incidents were particularly mixed. Eleven of 27 corridors in the a.m. peak period and four of 34 corridors in the p.m. peak period did not have increased 80th percentile travel times due to noncrash incidents. Only two corridors in the morning and three corridors in the afternoon among these moderately to heavily congested corridors had peak periods in which the 80th percentile travel times did not increase under crash conditions. Similarly, 15 of these corri- dors in the morning and 10 of them in the afternoon did not show an increased 95th percentile travel time. Only one corri- dor in the a.m. peak and two in the p.m. peak had 95th percen- tile travel times that did not increase when crashes occurred. In all cases, several additional corridors showed only mar- ginal changes in these statistics. If the results for the corridors with average weekday mean travel rates above 1.10 are simply averaged, then ⢠Noncrash incidents increase the 80th percentile travel times only 2% in the a.m. peak and 29% in the p.m. peak; ⢠Noncrash incidents increase the 95th percentile travel times only 1% in the a.m. peak and 16% in the p.m. peak; ⢠Crashes increase the 80th percentile travel times 24% in the a.m. peak and 39% in the p.m. peak; and ⢠Crashes increase the 95th percentile travel times 33% in the a.m. peak and 47% in the p.m. peak. Taken together, these results indicate that noncrash incidents were mostly responsible for modest changes in travel times. Those changes were more pronounced during periods of higher traffic volume and were thus generally more significant in the p.m. peak than in the a.m. peak. Noncrash incidents generally had very modest impacts on the worst travel days. In contrast, crashes had more substantial impacts on both the a.m. and p.m. peak periods. The fact that an accident occurred could be expected to add 20% to 40% to the travel times in much of the travel time distribution curve, whether that was the mean, 80th percentile, or 95th percentile travel time, with some crashes being responsible for much larger increases. Results: Incident-Related Changes in When Peak Period Congestion Ends In Figures 5.12 through 5.14, only those travel times influ- enced by an incident or crash were included in the computa- tion of the mean travel time associated with incidents and crashes. The problem with this (or any) approach to defining the influence of disruptions on travel times is understanding when those influences end. That is, the definition of incident influence used in the previous section means that only inci- dents that had a still-active effect on roadway performance were considered when computing incident-influenced travel time (where still active means that travel times in the test sec- tion were slower than those measured when the disruption was occurring). If an incident is quickly cleared and the dis- ruption is minimized, how does that event affect the travel time experienced? To better understand the effects of incidents and crashes, an entirely different examination of the impacts of those dis- ruptions is discussed below that examines when congestion, as part of the normal peak period increase in travel demand, can be expected to end. An examination of Figures 5.12 through 5.14 shows that mean travel times slow earlier in the day and last longer into the day whenever traffic disruptions occur. From the motoristsâ perspective, this means not only that their trip during the heart of the commute is longer, but that even if they have delayed their trip until after the normal peak period, they may still be stuck in congestion. To examine this phenomenon, the project team computed when the a.m. and p.m. peak periods normally ended for each study corridor. The team then examined whether the ending time of the peak period changed as a result of the occurrence of crashes or noncrash incidents. The resulting summary sta- tistics for these analyses are shown in Tables 5.11 and 5.12. (All the statistics generated from this analysis are shown in Appendix D.) The tables are sorted so that the study sections with the slowest, most congested corridors (as defined by their peak period median travel rate in minutes per mile) are at the top of the table, and the fastest, least congested corri- dors are at the bottom. Within a given travel rate, routes are sorted by their mean travel rate. Both tables show the mean time of day when congestion ended on days that did not experience reported incidents or crashes, and the mean dif- ference (in minutes) in the time of day for the end of conges- tion for each corridor when at least one crash or incident was reported within the study section in the indicated direction of travel. If both a crash and a noncrash incident occurred,
114 Table 5.11. Effects of Incidents and Crashes on Ending Time of P.M. Peak Period Congestion Study Corridor P.M. Peak Travel Rate Normal Time When Congestion Ended Additional Congestion Time (min) Mean Median Noncrash Incident Crash I-405 Bellevue southbound 3.73 3.6 19:44 0:00 0:20 I-405 Eastgate southbound 2.73 2.6 19:12 0:00 0:15 SR 520 Seattle westbound 2.72 2.6 20:00 0:00 0:12 I-405 South northbound 2.58 2.6 20:41 0:00 0:00 I-5 Seattle North southbound 2.56 2 18:49 0:00 0:31 I-405 Kirkland northbound 1.99 2 19:03 0:00 0:11 I-5 Seattle CBD northbound 1.96 1.8 18:53 0:00 0:00 I-405 Kennydale southbound 1.90 1.8 19:27 0:00 0:00 I-5 North King northbound 1.79 1.8 18:55 0:00 0:12 I-5 Seattle CBD southbound 1.72 1.8 18:20 0:00 0:00 SR 167 Auburn southbound 1.96 1.6 18:47 0:00 0:08 SR 520 Redmond eastbound 1.87 1.6 19:09 0:00 0:00 I-5 South southbound 1.76 1.6 18:08 0:00 0:00 I-405 North northbound 1.61 1.6 19:18 0:00 0:14 I-5 Everett northbound 1.87 1.4 17:08 0:28 0:58 I-5 Seattle North northbound 1.74 1.4 18:34 0:00 0:00 SR 167 Renton southbound 1.63 1.4 18:47 0:00 0:00 I-405 South southbound 1.52 1.4 19:36 0:00 0:00 I-90 Bridge westbound 1.73 1.2 18:25 0:34 0:48 SR 520 Seattle eastbound 1.49 1.2 18:52 0:00 0:22 I-5 Lynnwood northbound 1.38 1.2 19:00 0:00 0:00 I-405 Bellevue northbound 1.34 1.2 18:09 0:00 0:27 I-90 Seattle westbound 1.13 1.2 17:29 0:00 0:00 SR 520 Redmond westbound 1.49 1 16:51 1:24 1:53 I-90 Seattle eastbound 1.43 1 17:07 0:00 1:05 I-90 Bridge eastbound 1.40 1 18:18 0:22 0:35 I-5 North King southbound 1.33 1 16:47 0:29 1:57 I-90 Bellevue westbound 1.30 1 16:13 1:21 2:10 I-5 Tukwila southbound 1.19 1 17:18 0:21 0:51 SR 167 Renton northbound 1.17 1 17:22 0:27 0:57 I-405 Kennydale northbound 1.17 1 18:05 0:00 0:23 I-90 Bellevue eastbound 1.11 1 16:35 0:00 1:02 I-5 Everett southbound 1.10 1 16:35 0:24 0:57 I-5 Lynnwood southbound 1.10 1 17:21 0:00 1:09 I-405 North southbound 1.09 1 17:40 0:00 0:46 I-405 Kirkland southbound 1.09 1 16:55 1:00 1:21 I-5 Tukwila northbound 1.07 1 16:23 0:23 1:56 SR 167 Auburn northbound 1.05 1 17:31 0:00 0:00 I-405 Eastgate northbound 1.04 1 16:24 0:00 0:47 I-90 Issaquah eastbound 1.01 1 16:10 0:00 0:00 I-5 South northbound 1.01 1 16:05 0:00 0:45 I-90 Issaquah westbound 1.00 1 16:05 0:00 0:00
115 Table 5.12. Effects of Incidents and Crashes on Ending Time of A.M. Peak Period Congestion Study Corridor A.M. Peak Travel Rate Normal Time When Congestion Ended Additional Congestion Time (min) Adjusted End of Congestion Mean Median Noncrash Incident Crash Travel Time Valuea I-405 Kennydale northbound 3.66 3.4 11:47 0:00 1:33 10% I-405 North southbound 2.82 2.4 9:56 1:27 2:09 NA I-5 North King southbound 2.07 1.8 11:06 0:48 1:29 10% I-5 Seattle CBD northbound 1.91 1.8 12:15 0:00 0:00 No disruption- free days I-405 Kirkland southbound 1.76 1.8 10:16 0:56 1:14 NA SR 520 Seattle eastbound 1.70 1.8 11:54 6:02 6:53 NA I-5 Lynnwood southbound 1.89 1.6 10:06 1:57 1:39 NA I-5 South northbound 1.75 1.6 9:16 0:00 0:22 NA SR 167 Auburn northbound 1.68 1.6 11:40 0:00 0:00 20% I-405 Eastgate northbound 1.66 1.6 11:38 0:00 1:04 10% I-5 Seattle North southbound 2.15 1.4 9:38 0:00 4:58 NA I-405 Kennydale southbound 1.54 1.4 9:08 1:19 1:23 20% I-405 South southbound 1.45 1.4 12:46 3:12 2:17 20% SR 167 Renton northbound 1.62 1.2 9:13 1:47 1:22 20% SR 520 Seattle westbound 1.51 1.2 9:51 0:00 2:54 10% I-5 Tukwila northbound 1.50 1.2 10:06 0:00 0:32 NA I-90 Issaquah westbound 1.46 1.2 9:10 0:00 0:33 NA I-90 Bellevue westbound 1.30 1.2 9:26 0:13 0:00 NA I-405 Bellevue northbound 1.27 1.2 11:01 3:34 5:00 10% I-405 South northbound 1.24 1.2 8:21 4:49 7:47 20% I-90 Seattle eastbound 1.96 1 8:45 0:00 1:05 NA I-90 Seattle westbound 1.20 1 7:35 0:42 1:52 NA I-90 Bridge eastbound 1.18 1 9:23 0:45 1:04 NA I-405 Bellevue southbound 1.16 1 8:27 7:56 11:07 10% I-5 Everett southbound 1.15 1 7:08 0:00 1:06 NA I-90 Bridge westbound 1.15 1 8:04 0:26 1:30 NA I-5 Seattle CBD southbound 1.10 1 9:28 0:00 4:57 NA SR 167 Auburn southbound 1.06 1 8:58 7:29 9:59 NA I-405 Eastgate southbound 1.05 1 7:22 0:00 0:32 NA SR 167 Renton southbound 1.04 1 9:42 7:33 7:30 NA I-5 Tukwila southbound 1.02 1 7:08 0:00 7:51 NA SR 520 Redmond westbound 1.02 1 7:09 0:56 2:10 NA I-405 North northbound 1.02 1 7:56 0:12 1:47 NA I-5 Everett northbound 1.01 1 7:05 0:00 0:14 NA I-5 Lynnwood northbound 1.01 1 7:13 0:00 0:00 NA I-5 Seattle North northbound 1.01 1 7:07 0:00 0:00 NA (continued on next page)
116 equal to the speed limit, the end of congestion time was extended when incidents occurred. Several significant differences were observed between the effects of incidents and crashes in the morning peak period described in Table 5.12 and those shown for the evening peak period in Table 5.11. The most significant difference is that the heavily congested a.m. corridors are much more sensitive to incidents than their p.m. peak period counterparts. None of the 14 corridors with p.m. peak median travel rates above 1.4 had congestion durations that showed sensitivity to non- crash incidents, but five of the 10 morning peak corridors operating at this level of congestion were sensitive to non- crash incidents. One corridor, I-5 Seattle CBD northbound, had so many disruptions that no comparison could be made. This study segment had only one day among all nonholiday Tuesdays, Wednesdays, and Thursdays in 2006 that did not contain either a crash or a WITS-reported incident. Clearly, one day is not sufficient to make a statistically significant comparison. A second difference between the morning and evening periods was the size of the change when incidents and crashes affected the end of congestion. When incidents and crashes had an effect in the evening, the mean change in the duration of the peak period tended to be between 15 minutes and 1 hour, at most (35 of 45 statistically significant differences were less than 1 hour). In contrast, morning peak period cor- ridors affected by crashes and other incidents routinely saw congestion extend for more than an hour, and in many cases, multiple hours. However, at the less congested end of the congestion distri- bution, the morning peak period was similar to the evening peak period. More than half of the study corridors with a the day was classified as being affected by a crash. For the a.m. peak, the crash or incident must have taken place after 4:00 a.m. and before the end of congestion was reached. For the p.m. peak, the crash or incident must have taken place after 3:00 p.m. and before the end of congestion was reached. Statistical comparisons were performed by using the non- parametric AndersonâDarling k-sample test, with p-values of less than .01 being used to determine statistically significant end of congestion times. Statistically insignificant differences are set to zero in Tables 5.11 and 5.12. While the nature (size, duration, and specific location) of incidents affects exactly how much disruption each incident causes, and these differences in incident size and duration are not directly accounted for, some generalizations can be made from these tables. Among these are the following: ⢠Incidents that occur in the evening peak period have little measurable effect on the time that peak period congestion abates for (a) very heavily congested roadway sections or (b) very lightly congested sections; ⢠Crashes extend the evening commute periodâs congestion more significantly than noncrash incidents, and they are more likely to affect roadway performance than other kinds of incidents; and ⢠The duration of congestion on a surprising number of cor- ridors is not significantly affected by a crash occurring on that section. Of the 18 corridors with a median p.m. peak period travel rate of 1.4 or greater, the end of congestion was extended by noncrash incidents in a statistically significant manner for only one. For nine of 19 corridors with a median travel rate I-90 Bellevue eastbound 1.01 1 7:05 0:00 0:00 NA I-5 South southbound 1.00 1 7:07 0:08 0:00 NA I-405 Kirkland northbound 1.00 1 7:05 0:05 0:00 NA SR 520 Redmond eastbound 1.00 1 7:05 0:00 0:00 NA I-90 Issaquah eastbound 1.00 1 7:05 0:00 0:00 NA I-5 North King northbound 1.00 1 7:05 0:00 0:00 NA a On some study corridors, for the end of congestion to occur before noon after the a.m. peak period on days without incidents or crashes, it was necessary to change the definition of congestion from 20 consecutive minutes of average travel times being faster than 1.05 times the travel time at the speed limit to either 1.10 times the travel times at the speed limit (indicated by the value of 10%) or 1.20 times for travel time at the speed limit (indicated by 20%). Table 5.12. Effects of Incidents and Crashes on Ending Time of A.M. Peak Period Congestion (continued) Study Corridor A.M. Peak Travel Rate Normal Time When Congestion Ended Additional Congestion Time (min) Adjusted End of Congestion Mean Median Noncrash Incident Crash Travel Time Valuea
117 the normal operations of a roadway. This combination of supply and demand effects are generally categorized into the seven sources of congestion. These factors interact in the for- mation of congestion, and the relative importance of any one of these factors varies from location to location. In many rural areas, demand is routinely low relative to roadway capacity. Consequently, delay only happens when major disruptions occur, usually as a result of bad weather (e.g., snow), a major traffic incident, or reductions in roadway capacity due to road construction and maintenance activities. In other rural areas, especially those that experience recre- ational traffic flows, large and somewhat predictable surges of traffic demand create traffic congestion during times of peak demand. Similarly, in suburban and urban areas, traffic flows associated with work and other common activities often reach levels that typically push traffic demand beyond avail- able roadway capacity, creating routine congestion. In both of these cases, a large percentage increase in congestion can occur on top of the existing base congestion as a result of a disruption in roadway operations, especially when that dis- ruption occurs during times of high traffic volumes. Lastly, in larger urban areas, traffic can routinely exceed roadway capacity for many hours each work day. In these areas, numerous roads operate near capacity for many addi- tional hours of the day. Disruptions on these roads can add large amounts of delay, but that added delay may be only a modest percentage increase in total annual delay. In simple terms, routine congestion already may have slowed traffic, so that a fender-bender in the existing queue slows vehicles only a little more because they already are moving slowly. The 42 directional roadway sections studied in this analysis all experienced at least some routine congestion in either the a.m. or p.m. peak periods. Many sections experienced routine congestion during only one of the peak periods, but a num- ber of the sections experienced significant congestion in both peaks, as well as periodic congestion in the middle of the day. Table 5.13 summarizes the amount of delay influenced by each type of disruption tracked in this study. Delay was cal- culated for each 5-minute time interval of 2006 for each road- way segment in units of vehicle seconds as follows: delay actual travel time travel time at the= â speed limit roadway segment volume ( ) ( ) The percentage of delay was computed by totaling all vehicle hours of delay in the region associated with each type of dis- ruption, and then dividing by the sum of all measured delays. When more than one disruption occurred simultaneously, the resulting delay was credited to all of the associated causes. Thus the sum of the percentages in Table 5.13 exceeds 100%. Taken at face value, this simple summary table supports the commonly heard statement that âincidents and crashes cause between 40% and 60% of all delay.â In reality, the median travel rate equal to the speed limit (1.0) had conges- tion ending times that were not statistically affected by inci- dents. The majority of these corridors also had a mean travel rate of less than 1.02 and were reasonably insensitive to con- gestion caused by crashes. These results indicate that if traffic volume relative to capacity is low enough to not produce even light routine congestion, then only very large incidents and crashes will create congestion. These observed differences further strengthen the primary finding of this study: the overriding factor affecting travel time reliability is the back- ground traffic volume. Although there were many differences in the a.m. and p.m. peak periods, one of the key differences was that the morning leading (early) shoulder had very low traffic volumes. There- fore, as noted earlier, incidents tended to have little impact early in the a.m. peak period. In the evening, traffic volume dropped off rapidly at the end of the peak period, and con- gestion frequently abated rapidly simply because traffic vol- umes were low enough for queues to clear. At the end of the morning peak, however, traffic volumes remained modest because of the addition of noncommute trips to the traffic stream. Thus, incident congestion formed during the a.m. peak tended to last much longer than incident congestion formed in the p.m. peak. Conversely, significant incidents that occurred well before the start of the p.m. peak period had the potential to cause the entire p.m. peak period to be congested if they were not cleared quickly, but incidents occurring an hour before the start of the a.m. peak, if they were cleared with even modest speed, were far less likely to affect the morning commute. Summary: Causes of Congestion Congestion occurs when there is too much volume and too little roadway capacity. This can occur because ⢠Traffic demand is too great for the designed roadway capacity; or ⢠Some disruption reduces functional roadway capacity (supply) to levels below demand. Demand varies because of repeating travel patterns (e.g., time of day, day of week, seasonal patterns) and as a result of unusual activity that causes more travelers than typical to use a roadway at a given time. These unusual activities can be planned events, such as a major sporting event, or unplanned events, such as vehicles diverting to one roadway to avoid congestion on another. Functional roadway capacity (supply) can vary as a result of numerous factors, including weather, traffic management strategies (work zones, the application of different traffic control plans), and a variety of traffic incidents that disrupt
118 bad weather, and traffic volumes on travel times on I-5 north- bound heading toward downtown Seattle. This graphic shows that congestion formed only as traffic volumes peaked. It also shows that the resulting congestion reduced observed throughput while increasing travel times. In addition, it illustrates how all types of disruptions to normal roadway performance (rain, crashes, noncrash incidents) caused congestion to start earlier and last longer during the peak period, while increasing travel times during the normally congested times. Incidents and other disruptions also can cause congestion to form during times of the day that are normally free from congestion, but only when the disruption lowers functional capacity below traffic demand. Thus, as seen in Figure 5.15, minor disruptions such as rain or noncrash incidents on this section of I-5 did not cause congestion in the midday or the evening peak period (in the off-peak direction). For this four- lane freeway section, enough unused capacity exists during those periods that modest disruptions to roadway capacity did not cause congestion, although some crashes caused suf- ficient disruption to create congestion during these off-peak periods. Late at night, because construction activity was tak- ing place along this roadway segment, even smaller incidents (combined with those construction lane closures) caused congestion to form. Thus volume, relative to roadway capacity, is a key compo- nent of congestion formation, and in urban areas it is likely to be the primary source of congestion. Disruptions then sig- nificantly increase the delay that the basic volume condition creates. amount of delay caused by incidents was actually less than that indicated in Table 5.13 because a considerable portion of the incident- and crash-associated delay was caused by large traffic volumes. There were numerous examples in the analy- sis data set of significant crashes and other incidents that caused little or no congestion because of when they occurred. These examples showed that without sufficient volume, an incident causes no measurable change in delay. Travel Time Impacts Caused by Disruptions In the Seattle area, many incidents take place during peak periods, causing already existing congestion to grow worse. Figure 5.15 illustrates the interwoven effects of incidents, Table 5.13. Percentage of Delay by Type of Disruption Influencing Congestion Duration and Severity Type of Disruption Delay (%) Incidents 38.5 Crashes 19.5 Bad weather (rain) 17.7 Constructiona 1.2 No cause indicated (mostly volume) 42.2 a Construction delay was computed only when construction work actively took place along the roadway and did not include any delays caused because general roadway capacity was reduced as a result of temporarily narrowed or reconfigured lanes. Figure 5.15. Effect of disruptions and traffic volume on travel time on I-5, northbound South section.
119 a crash occurring during the a.m. peak period adds an average of 2 hours and 17 minutes to the duration of the morningâs peak period congestion. In the p.m. peak, a crash adds only adds 33 minutes to the time when congestion normally can be expected to clear. Similarly, a noncrash incident adds 1 hour and 14 minutes to the morning peak, but in the p.m. only 10 minutes are added to the time that congestion can be expected to last. As seen in Figure 5.15, travel times also generally increased within the peak period when disruptions occurred to normal freeway flow. If the peak period is held constant (6:30 to 9:30 a.m. for the morning peak and 3:00 to 7:00 p.m. for the eve- ning peak), average travel times during those periods increased when a crash or noncrash incident occurred on a roadway segment. Morning travel times increased by 17% in corridors that experienced even modest a.m. peak period congestion when noncrash incidents occurred. Noncrash incidents increased p.m. travel times an average of 21% on corridors that experienced any routine increase in p.m. peak travel. In both the a.m. and p.m. peaks, crashes added roughly 40% to the expected travel times. These effects varied significantly from corridor to corri- dor, depending on the nature of the traffic volumes and rou- tine congestion patterns. They also changed dramatically within any given corridor on the basis of the size, duration, and timing of the disruption. Interestingly, 80th and 95th percentile travel times were less affected by noncrash inci- dents, but crashes generally had significant impacts on both of these performance measures. This is not surprising because Not only does traffic volume affect whether an incident causes congestion, but it affects how long that congestion lasts once the primary incident has been removed. The Seattle data showed that in the morning peaks, disruptions had a more noticeable effect on the timing of the end of the peak period, while in the evening the opposite was true. In the afternoon, as Figure 5.16 shows, disruptions began to cause greater travel time changes well before the start of the traditional peak period. However, most congestion ended very close to when congestion under no rain-no disruption conditions would have occurred. The effects of late-night crashes can be seen in the graph. The volume lines in Figures 5.15 and 5.16 explain the dis- crepancies in the end times of the a.m. and p.m. peak conges- tion. Very early in the a.m. peak period, insufficient volume exists to cause congestion to form. Once volumes grow and congestion occurs, disruptions (incidents or rain) make that congestion worse. Because midday volumes are still fairly high, residual queues can take a long time to clear. In the p.m., those same fairly high midday volumes (espe- cially for corridors experiencing peak direction movements) mean that even small disruptions are likely to cause conges- tion before the normal start of the p.m. peak period. How- ever, even though queues grow larger than usual during those peak periods, the sharp decline in traffic volumes at the end of the p.m. peak means that as long as the disruption has been cleared, those queues tend to dissipate quickly at the end of the peak period. Although results varied dramatically between study sec- tions, if the results of all 42 study sections are simply averaged, Figure 5.16. Effect of disruptions and traffic volume on travel time on I-5, northbound Lynnwood section.
120 references 1. Kopf, J. M., J. Nee, J. M. Ishimaru, and M. E. Hallenbeck. Measure- ment of Recurring and Non-Recurring Congestion: Phase II. Final research report. Washington State Transportation Center, Seattle, 2005. www.wsdot.wa.gov/research/reports/fullreports/619.1.pdf. 2. Edwards, J. B. Speed Adjustment of Motorway Commuter Traffic to Inclement Weather. Transportation Research Part F, Vol. 2, No. 1, 1999, pp. 1â14. 3. Rakha, H. A., M. Farzaneh, M. Arafeh, and E. Sterzin. Inclement Weather Impacts on Freeway Traffic Stream Behavior. In Transporta- tion Research Record: Journal of the Transportation Research Board, No. 2071, Transportation Research Board of the National Academies, Washington, D.C., 2008, pp. 8â18. 4. Cools, M., E. Moons, and G. Wets. Assessing the Impact of Weather on Traffic Intensity. Presented at 87th Annual Meeting of the Trans- portation Research Board, Washington, D.C., 2008. 5. Unrau, D., and J. Andrey. Driver Response to Rainfall on Urban Expressways. In Transportation Research Record: Journal of the Trans- portation Research Board, No. 1980, Transportation Research Board of the National Academies, Washington, D.C., 2006, pp. 24â30. noncrash incidents tend to be smaller disruptions, and con- sequently have less of an impact on those very bad days when congestion is at its worst. Crashes, however, are often one of the contributing factors to very bad commute days. Summary Analysis of 42 roadway segments in the Seattle area showed that a majority of travel delay in the region is the direct result of traffic volume demand exceeding available roadway capacity. Whenever they occur, incidents, crashes, and bad weather add significantly to the delays that can be otherwise expected. The largest of these disruptions plays a significant role in the worst travel times that travelers experience on these roadways. However, the relative importance of any one type of disruption can vary considerably from corridor to corridor.
