Analytical Procedures for Determining the Impacts of Reliability Mitigation Strategies (2012)

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.

31 C h a p t e r 3 Introduction The research team decided at the beginning of the project that an empirical approach would be used to develop predic- tive relationships for reliability. The alternative would have been to conduct a large number of simulation-based experi- ments. However, team members had conducted several previ- ous projects using empirical data and were confident that these data could be used successfully. In addition, the large amount of empirical data that would be assembled could not only be used in this project but would have value for future research. Such an approach is not without risk. Real-world data can be subject to measurement error, and it was clear that the extremely large amount of data that would be needed could not be uniquely collected by the project; that is, the data collection itself was outside of the team’s control. Neverthe- less, continuous travel time data collected for a sufficiently long period of time is an absolute requirement for empirical studies of reliability, as reliability is defined by how travel times vary over a considerable time span. Given the myriad of factors that influence reliability, the team estimated that a complete year of data would be needed. The majority of the project’s effort was the creation of analysis data sets. Data set creation involved obtaining, clean- ing, and integrating data collected primarily by public agen- cies, but also private vendors. The research team selected agencies that had a long history of data collection and (based on the team’s experience) had data of the coverage and qual- ity required to undertake the research. The challenges in this approach were twofold: (a) processing, reviewing, and reduc- ing the raw data to summary measurements for the analysis; and (b) matching the different types of data geographically. Assembling empirical data from locations around the country proved to be challenging, but manageable. Traffic data are relatively consistent from location to location, but incident and work zone data do not seem to follow any stan- dard definitions. Fusion of the event data with the traffic data also posed problems; in some cases these had to be matched manually. traffic and travel time Data Urban Freeways The project team assembled urban freeway data from traffic management centers (TMCs) that were considered to be at the forefront of maintaining quality traffic data. Other con- siderations in selecting the cities were the availability of inci- dent data from the TMCs, the presence of before-and-after improvement situations, and a fairly long history of archiving data. Table 3.1 summarizes the cities and Table 3.2 summa- rizes the study sections. The locations of the sections appear in Figures 3.1 through 3.7. All these sections were considered in the exploratory analyses (Chapter 4) and the statistical modeling (Chapter 7), and several sections were used in the before-and-after analysis (see boldface portion of Table 3.2). A separate data set of urban freeways was compiled for the Seattle area for the congestion by source analysis. Seattle freeways are not included in Table 3.2. Seattle data were used in the congestion by source analysis and before- and-after studies (described further in Chapter 5). The urban freeway data set was the most complete of all the data sets assembled for the project. In addition to traffic data, all the sections had incident and weather data available. Signalized Arterials Table 3.3 shows the data assembled for signalized arterials. Data were derived from both public and private sources and several technologies. The privately provided data were pur- chased from Inrix, which has nationwide agreements with private fleets to capture travel time information. Inrix sells these data primarily for real-time traveler information to both private and public entities (such as the I-95 Corridor Data Collection, Assembly, and Fusion

32 Coalition), but it also archives the data for other uses. In late 2007, the research team asked Inrix to review their data qual- ity and to provide suggestions for arterial sections they felt had the best quality of data and the highest sample sizes. However, upon review of the data, it was determined that the Inrix data for signalized arterials had an insufficient number of samples to define reliability for the research. Although the sources of the travel time measurements are proprietary, the small number of measurements during traditional peak peri- ods, at least during the 2006 to 2007 period, led the team to surmise that most of the Inrix measurements were derived from fleet vehicles. The team was also cautious about the use of Inrix measurements for signalized highways, as they are not distance-based measurements like those taken from toll tags. They may have been adequate, but given the sample size problems, they were not tested. The net result was that only the two arterials in Orlando could be used for the analysis. Table 3.1. Urban Freeway Study Section Summary City Number of Directional Study Sections Total Directional Mileage Houston, Texas 13 58.80 Minneapolis, Minnesota 16 62.63 Los Angeles, California 3 50.27 San Francisco Bay Area, California 4 19.98 San Diego, California 6 28.04 Atlanta, Georgia 10 54.66 Jacksonville, Florida 8 17.71 Total 60 292.09 Table 3.2. Urban Freeway Study Sections Number City Route Directions Covered Beginning Landmark Ending Landmark Length (mi) Time Period Covered 1 Houston U.S. 290 Northwest Eastbound Barker Cypress FM 1960 4.05 1/1/2006–12/31/2007 2 Houston U.S. 290 Northwest Eastbound FM 1960 Sam Houston 5.10 1/1/2006–12/31/2007 3 Houston U.S. 290 Northwest Eastbound Fairbanks–N Houston W 34th 5.35 1/1/2006–12/31/2007 4 Houston U.S. 290 Northwest Westbound Pinemont Sam Houston 4.45 1/1/2006–12/31/2007 5 Houston U.S. 290 Northwest Westbound Sam Houston FM 1960 4.25 1/1/2006–12/31/2007 6 Houston U.S. 290 Northwest Westbound FM 1960 Barker Cypress 4.90 1/1/2006–12/31/2007 7 Houston I-45 Gulf Northbound Nasa Road 1 Dixie Farm Road 5.10 1/1/2006–12/31/2007 8 Houston I-45 Gulf Northbound Dixie Farm Road Fuqua 2.80 1/1/2006–12/31/2007 9 Houston I-45 Gulf Northbound Edgebrook Broadway 4.70 1/1/2006–12/31/2007 10 Houston I-45 Gulf Northbound Woodridge Scott Street 4.15 1/1/2006–12/31/2007 11 Houston I-45 Gulf Southbound Scott Street Woodridge 4.15 1/1/2006–12/31/2007 12 Houston I-45 Gulf Southbound Broadway Edgebrook 4.70 1/1/2006–12/31/2007 13 Houston I-45 Gulf Southbound Dixie Farm Road Nasa Road 1 5.10 1/1/2006–12/31/2007 14 Minneapolis– St. Paul I-35 W Northbound W 106th Street South of I-494 3.47 1/1/2006–12/31/2007 15 Minneapolis– St. Paul I-35 W Southbound South of I-494 W 106th Street 3.64 1/1/2006–12/31/2007 16 Minneapolis– St. Paul I-35 W Northbound T.H. 36 I-694 3.37 1/1/2006–12/31/2007 17 Minneapolis– St. Paul I-35 W Southbound I-694 T.H. 36 3.29 1/1/2006–12/31/2007 (continued on next page)

33 Table 3.2. Urban Freeway Study Sections (continued) Number City Route Directions Covered Beginning Landmark Ending Landmark Length (mi) Time Period Covered 18 Minneapolis– St. Paul T.H. 36 Eastbound Fairview Avenue I-35 E 4.41 1/1/2006–12/31/2007 19 Minneapolis– St. Paul T.H. 36 Westbound I-35 East Fairview Avenue 4.35 1/1/2006–12/31/2007 20 Minneapolis– St. Paul I-35 E Northbound W 7th Street I-94 3.48 1/1/2006–12/31/2007 21 Minneapolis– St. Paul I-35 E Southbound I-94 W 7th Street 3.59 1/1/2006–12/31/2007 22 Minneapolis– St. Paul T.H. 77 Northbound T.H. 13 I-494 3.43 1/1/2006–12/31/2007 23 Minneapolis– St. Paul T.H. 77 Southbound I-494 T.H. 13 3.43 1/1/2006–12/31/2007 24 Minneapolis– St. Paul I-94 Eastbound Highway 100 I-494 7.00 09/2000–09/2001 and 11/2004–11/2005 25 Minneapolis– St. Paul I-94 Westbound I-494 Highway 100 7.00 09/2000–09/2001 and 11/2004–11/2005 26 Minneapolis– St. Paul I-494 Eastbound Highway 100 Highway 5/312 4.00 05/2002-05/2003 and 11/2005–11/2006 27 Minneapolis– St. Paul I-394 Westbound Highway 100 Highway 169 3.17 07/2004–07/2005 and 11/2005–11/2006 28 Minneapolis– St. Paul Highway 169 Southbound T.H. 62 I-494 2.00 06/2005–06/2006 and 11/2006–11/2007 29 Minneapolis– St. Paul Highway 100 Northbound 36th Street I-394 2.80 04/2005–04/2006 and 11/2006–11/2007 30 Los Angeles I-210 Westbound Foothill Highway and Ventura Freeway Interchange S. Asuza Avenue and Foothill Freeway Interchange 13.63 10/2000–12/2002 31 Los Angeles I-710 Northbound Interchange: I-710 and I-5 I-710 and W. Ocean Boulevard 18.32 04/2004–06/2006 32 Los Angeles I-710 Southbound Interchange: I-710 and I-5 I-710 and W. Ocean Boulevard 18.32 04/2004–06/2006 33 Bay Area I-880 Northbound Oak Street Ramps I-980 Ramps 1.35 01/2008–12/2008 34 Bay Area I-880 Southbound Oak Street Ramps I-980 Ramps 1.35 01/2008–12/2008 35 Bay Area I-580 Eastbound Eden Canyon Ramps 1st Street and I-580 Interchange, Livermore 8.64 06/2002–07/2004 36 Bay Area I-580 Westbound Eden Canyon Ramps 1st Street and I-580 Interchange, Livermore 8.64 06/2002–07/2004 37 San Diego SR 52 Eastbound Santo Road Ramps SR 52 and SR 125 Interchange 5.96 06/2004–12/2006 38 San Diego SR 52 Westbound Santo Road Ramps SR 52 and SR 125 Interchange 5.96 06/2004–12/2006 39 San Diego I-5 Northbound Del Mar Heights Road Ramps Carmel Valley Road Interchange 3.38 06/2001–08/2006 40 San Diego I-5 Southbound Del Mar Heights Road Ramps Carmel Valley Road Interchange 3.38 06/2001–08/2006 (continued on next page)

34 Table 3.2. Urban Freeway Study Sections (continued) Number City Route Directions Covered Beginning Landmark Ending Landmark Length (mi) Time Period Covered 41 San Diego I-8 Northbound North 2nd Street Interchange Lake Jennings Park Interchange 4.68 06/2004–08/2006 42 San Diego I-8 Southbound North 2nd Street Interchange Lake Jennings Park Interchange 4.68 06/2004–08/2006 43 Atlanta I-75 Northbound I-285 Roswell Road 5.19 01/2006–12/2008 44 Atlanta I-75 Southbound I-285 Roswell Road 5.19 01/2006–12/2008 45 Atlanta I-75 Northbound I-20 I-85 4.43 01/2006–12/2008 46 Atlanta I-75 Southbound I-20 I-85 4.43 01/2006–12/2008 47 Atlanta I-285 Eastbound I-75 GA 400 6.50 01/2006–12/2008 48 Atlanta I-285 Westbound I-75 GA 400 6.50 01/2006–12/2008 49 Atlanta I-285 Eastbound GA 400 I-85 6.03 01/2006–12/2008 50 Atlanta I-285 Westbound GA 400 I-85 6.03 01/2006–12/2008 51 Atlanta I-75 Northbound Roswell Road Barrett Parkway 5.18 01/2006–12/2008 52 Atlanta I-75 Southbound Roswell Road Barrett Parkway 5.18 01/2006–12/2008 53 Seattle SR 520 Eastbound– westbound I-5 I-405 7.00 01/2006–12/2008 54 Seattle SR 520 Eastbound– westbound I-405 SR 202 5.50 01/2006–12/2008 55 Seattle I-90 Eastbound– westbound I-5 West End Floating Bridge 1.24 01/2006–12/2008 56 Seattle I-90 Eastbound– westbound West End Floating Bridge I-405 4.76 01/2006–12/2008 57 Seattle I-90 Eastbound– westbound I-405 West Lake Sammamish 4.00 01/2006–12/2008 58 Seattle I-90 Eastbound– westbound West Lake Sammamish West of High Point Road 6.37 01/2006–12/2008 59 Seattle SR 167 Northbound– southbound 15th Street NW SR 516 3.70 01/2006–12/2008 60 Seattle SR 167 Northbound– southbound SR 516 I-405 6.10 01/2006–12/2008 61 Seattle I-405 Northbound– southbound I-5 Tukwila SR 167 2.30 01/2006–12/2008 62 Seattle I-405 Northbound– southbound SR 167 112th Avenue SE 7.70 01/2006–12/2008 63 Seattle I-405 Northbound– southbound 112th Avenue S.E. I-90 2.20 01/2006–12/2008 64 Seattle I-405 Northbound– southbound I-90 SR 520 3.40 01/2006–12/2008 65 Seattle I-405 Northbound– southbound SR 520 SR 522 8.40 01/2006–12/2008 66 Seattle I-405 Northbound– southbound SR 522 I-5 Lynnwood 6.50 01/2006–12/2008 67 Seattle I-5 Northbound– southbound South 320th Street I-405 Tukwila 10.40 01/2006–12/2008 68 Seattle I-5 Northbound– southbound I-405 Tukwila Albro Place 6.60 01/2006–12/2008 (continued on next page)

35 Table 3.2. Urban Freeway Study Sections (continued) Number City Route Directions Covered Beginning Landmark Ending Landmark Length (mi) Time Period Covered 69 Seattle I-5 Northbound– southbound Albro Place SR 520 7.80 01/2006–12/2008 70 Seattle I-5 Northbound– southbound SR 520 Northgate 4.10 01/2006–12/2008 71 Seattle I-5 Northbound– southbound Northgate Snohomish/King County Line 5.40 01/2006–12/2008 72 Seattle I-5 Northbound– southbound Snohomish–King County Line 128th SW 8.10 01/2006–12/2008 73 Seattle I-5 Northbound– southbound 128th SW Marine View Drive 8.40 01/2006–12/2008 74 Jacksonville I-95 Northbound Phillips Highway SR 202 5.16 03/2008–12/2008 75 Jacksonville I-95 Southbound Phillips Highway SR 202 5.16 03/2008–12/2008 76 Jacksonville I-95 Northbound SR 202 Atlantic Boulevard 4.56 03/2008–12/2008 77 Jacksonville I-95 Southbound SR 202 Atlantic Boulevard 4.56 03/2008–12/2008 78 Jacksonville I-95 Northbound U.S. 23 SR 111 (Edgewood) 3.85 03/2008–12/2008 79 Jacksonville I-95 Southbound U.S. 23 SR 111 3.85 03/2008–12/2008 80 Jacksonville I-95 Northbound SR 111 I-295 4.13 03/2008–12/2008 81 Jacksonville I-95 Southbound SR 111 I-295 4.13 03/2008–12/2008 Note: Houston data are based on toll tag–equipped probe vehicles and comprise direct travel time measurements. The remaining locations’ data comprise roadway- based sensor measurements of volume, speed, and lane occupancy. Sections in boldface were used in the before-and-after analysis. All sections were considered by the statistical modeling. Figure 3.1. Atlanta base map.

36 Figure 3.2. Houston base map. Figure 3.3. Minneapolis–St. Paul base map.

37 Figure 3.4. San Francisco Bay Area base map. Figure 3.5. Los Angeles base map.

38 Figure 3.6. San Diego base map. Figure 3.7. Jacksonville base map.

39 Rural Freeways Table 3.4 presents the sections for which rural freeway travel time data were assembled. The Inrix data were deemed to have sufficient sample sizes for the two locations indicated. Incident and Work Zone Data Incident and Work Zone Characteristics Data on the basic characteristics of incidents were available from three sources and were used to varying degrees, depend- ing on the team’s assessment of the data sources for each city. Incident and event data were provided at no cost to the proj- ect team by the private vendor Traffic.com from their traveler information management system (TIMS). The TIMS data provided a standardized source of information for traffic incidents, events, scheduled and unscheduled construction, and other events that could affect traffic conditions (such as severe weather or transit delays). Incident data are gathered directly by Traffic.com observ- ers using numerous sources of information, such as video images, aircraft, mobile units, and police and emergency communication frequencies. In some cities, Traffic.com observ- ers are stationed on the floor of the regional TMC. In other cities, Traffic.com observers are mobile or are stationed in a connected operations center. The incident data from Traffic.com were chosen for this study because it has several unique attributes: • All reported incidents are entered. Traffic.com does attempt to confirm reports, and will indicate in their sys- tem when the reported incident has been confirmed. Thus, they provide both reported incidents, as well as confirmed incidents; • Traffic.com incident data are collected by an independent entity that is not involved in the traffic or emergency man- agement process. It was reasoned that Traffic.com staff could gather more complete and accurate data because Table 3.3. Signalized Arterial Study Sections City Arterial From To Length (mi) Travel Time Data Data Technology Period Orlando Section 1: SR 50 eastbound Florida Turnpike SR 408W 6.85 Tag-based probe March 2008+ Section 2: SR 50 westbound SR 408W Florida Turnpike 6.85 Tag-based probe March 2008+ Section 3: U.S. 441 northbound SR 417 SR 408 10.67 Tag-based probe March 2008+ Section 4: U.S. 441 southbound SR 408 SR 417 10.67 Tag-based probe March 2008+ Section 5: U.S. 441 northbound SR 408 N John Young Parkway 4.35 Tag-based probe March 2008+ Section 6: U.S. 441 southbound N John Young Parkway SR 408 4.35 Tag-based probe March 2008+ Los Angeles Santa Monica Boulevard I-405 N Gardner Street 6.9 GPS probe (Inrix) 2006–2007 Phoenix E Camelback Road 44th Street Highway 51 4.2 GPS probe (Inrix) 2006–2007 Minneapolis– St. Paul Washington Avenue County Highway 153 U.S. 65 3.4 GPS probe (Inrix) 2006–2007 Miami U.S. 1 17th Avenue Le Jeune Road 3.8 GPS probe (Inrix) 2006–2007 Houston Westheimer Road W Sam Houston I-610 6.9 GPS probe (Inrix) 2006–2007 Note: Probe-tag technology provided direct estimates of travel time over the segment. Inrix-provided data were supplied as speed estimates by link (approximately 0.5 to 1 mile long). Only the Orlando sections were used in the analysis because of sample size limitations on the other sections. Table 3.4. Rural Freeway Study Sections State Route From To Length Travel Time Data Data Technology Period Texas I-45 Exit 213, Navarro County Exit 267, Ellis County 54.1 GPS probe (Inrix) 2006–2007 South Carolina I-95 South Carolina–Georgia Border SR 68, Hampton County 38.2 GPS probe (Inrix) 2006–2007

40 information gathering and reporting were their sole focus (in contrast, public agency traffic managers typically must manage incidents and crises and record relevant informa- tion at the same time). Additionally, the Traffic.com incident data are routinely reviewed to ensure quality data entry by Traffic.com observers; • Traffic.com incident data contain the sequence of events as an incident is reported, responded to, and cleared. For example, an incident record is updated and appended whenever the status or conditions of the incident change. This information provides more specificity about the inci- dent; and • Traffic.com incident data provided consistent data attri- butes in all of this study’s cities and also used unambiguous location referencing. The following incident attributes were used in this study: • Unique traffic item identifier—A unique identifier for each record or observation; • Unique original traffic item identifier—A unique identi- fier for the original traffic incident that did not change as information about the same incident was updated; • Metropolitan area—Unique city identifier; • Roadway and location identifier—Unique combination of identifiers for the location. • Type of traffic item—Possible entries include: 44 Accident; 44 Alert; 44 Congestion; 44 Disabled vehicle; 44 Mass transit; 44 Miscellaneous; 44 Other news; 44 Planned event; 44 Road hazard; 44 Scheduled construction; 44 Unscheduled construction; and 44 Weather. • Verification—An indication of whether the incident or event was verified; • Number of lanes blocked—Numeric entry for number of travel lanes that were blocked; and • Start and ending times—The combination of these attri- butes provided incident duration. The start time was the time when the lane or shoulder blockage began; the end time was when the blockage was cleared. Data collected by TMC operators and entered into consoles at the TMC and/or entered by freeway service patrols were also available for some cities. The type of data collected by these entities varies, but they generally correspond to Traffic.com data; the key items of location, duration, and lane blockage are the same. The sources of incident data used in the urban freeway analysis were as follows: • Atlanta—TMC data were the primary source (this included work zones and special events), checked against both Traffic.com and Georgia DOT crash data; • Houston—Traffic.com data were found to match TMC (Transtar) incident data very well, and since Traffic.com contains work zones and special events, was the source of incident information; • Minneapolis—Traffic.com data; • San Diego, Los Angeles, and San Francisco Bay Area— Traffic.com data; • Seattle—Special data set, a fusion of TMC and police computer-aided dispatch data; and • Jacksonville—TMC data. Incident Activity Data Areas with incident management differ substantially in the institutional arrangements and policies that govern their day- to-day operations. Many of these incident management approaches are subjective and did not lend themselves to the quantification that was needed for the statistical modeling. Initially, it was thought that the approach taken in SHRP 2 Project L06, Institutional Architectures to Advance Opera- tional Strategies, could be used. The L06 approach is based on adapting the capability maturity model developed for soft- ware engineering to operations activities in transportation agencies. The capability maturity model in software engi- neering is a model of the maturity of the capability of certain business processes. A maturity model can be described as a structured collection of elements that describe certain aspects of maturity in an organization; the model aids in defining and understanding an organization’s processes. It was hoped that the resulting levels could be used as indicators of the degree of sophistication in policies and institutional arrangements with regard to incident management. Unfortunately, Project L06’s capability maturity model was too general and nonspecific to incident management to be of use in the statistical modeling for this project. Instead, the team used the Traffic Incident Management (TIM) Self-Assessment procedure developed by FHWA to indicate the sophistication of incident management arrangements for modeling purposes. This procedure has the advantages of capturing a wide range of activities in a single numeric score and being widespread among operators to facil- itate application of the final models. The TIM Self-Assessment consists of three primary assessment areas: 1. Program and institutional issues; 2. Operational issues; and 3. Communications and technology issues.

41 Composite scores are given for each of these areas (there are multiple attributes in each area), as well as a single overall score. The team explored using both the individual scores, as well as the overall score, in the modeling. Unfortunately, self- assessment scores were only available for three cities, which were not enough for model development. Nonetheless, pre- liminary (but inconclusive) results are presented. Weather Data Overview Weather data were obtained from the National Climatic Data Center (NCDC) of the National Oceanic and Atmospheric Administration. NCDC, the world’s largest active archive of weather data, produces climate publications and responds to data requests from all over the world. NCDC offers a wide range of climate products and services, including a surface cli- mate product that provides local climatological data, as well as marine and upper-air data. The local climatological data product consists of hourly, daily, and monthly climatological summaries for approxi- mately 1,600 U.S. locations (daily summary forms are not available for all stations). Since the end of January 2005, the local climatological data have been processed through auto- mated quality control processing. About 480 first-order sta- tions also undergo additional quality control after the end of the month. Data Access Similar to other NCDC products and services, the local cli- matological data are available through a variety of media, including online access, annual subscriptions, CD-ROMs, DVDs, computer tabulations, maps, and publications (1). Free access to NCDC data is granted to certain users, such as academic and educational users, using reverse domain lookup. The local climatological data for specific locations and specific time frames are available for download. Final data loads occur on a monthly basis, usually overnight. Data gaps may exist during the time frame of previous and current final data loads. Data Format and Description of Hourly Data The hourly data files used for the research contained the fol- lowing basic weather elements: • Sky condition—Cloud height and amount (clear, scat- tered, broken, and overcast) up to 12,000 feet; • Visibility (to at least 10 statute miles); • Basic present weather information—Type and intensity for rain, snow, and freezing rain; • Obstructions to vision—Fog, haze; • Pressure—Sea-level pressure, altimeter setting; • Ambient temperature and dew point temperature; • Wind—Direction, speed, and character (gusts, squalls); • Precipitation accumulation; and • Selected significant remarks, including variable visibility, precipitation beginning and ending times, rapid pres- sure changes, pressure change tendency, wind shift, and peak wind. Geometric, Operating, and Improvement Data Geometric data were obtained from satellite imagery (lane configurations) and the 2007 Highway Performance Moni- toring data. Operating and improvement data were obtained directly from state DOTs. The most important data in this category were those elements related to calculating capacity for each individual link. Data processing procedures Urban Freeway Data Processing Data for all urban freeway sections were centrally processed to ensure consistency. The processing began with quality control of the data as received from the TMCs. The data qual- ity checks used were those developed for FHWA (2). Data were aggregated to 5-minute by-lane summaries. Aggrega- tion rules followed those in a forthcoming ASTM standard (ASTM E2665-08). Two levels of spatial aggregation were done on the 5-minute-interval data: 1. Station level—Data were aggregated laterally over all lanes in a direction; and 2. Section level—Station-level data were aggregated longi- tudinally for all stations on a study section. The aggregation process is shown in Figure 3.8. From the section-level data, a procedure for estimating the start and end times of the peak hour and peak period was applied; this procedure is detailed in Chapter 4 under the section “Defin- ing Peak Hour and Peak Period.” Analysts then reviewed the start and end times and made adjustments based on local knowledge. Section-level statistics were computed for each time slice to be used in the analysis: • Peak hour (weekdays only); • Peak period (weekdays only); • Counterpeak hour (weekdays only; the opposite time slice from the peak hour; that is, if the peak hour is in the morn- ing, then the counterpeak is in the afternoon);

42 • Midday; • Week day (all hours of the day); and • Weekend and holiday (all hours of the day). Signalized Arterial Data Processing Calculating travel time and reliability statistics from toll tag– equipped probe vehicles is straightforward—travel times are measured directly so there is no need for transformations as shown in Figure 3.8. Data quality control is different, how- ever. Because of the opportunities for vehicles to make incom- plete trips through a section of arterial (such as stopping at adjacent land uses), some travel times will be detected as being excessively high. As a result, probe data quality controls have been focused on eliminating outliers. In FHWA’s Mobil- ity Monitoring and Urban Congestion Report Project (3), the quality control criterion for probe data states that two consecutive travel times cannot change by more than 40%. Another method proposed by researchers at the University of Washington is that a travel time cannot be more than one standard deviation above or below the moving average of the 10 previous entries. These methods work well for freeway data, for which probe data coverage is high. However, probe data on arterials are considerably sparser. Many of the outliers in arterial data will pass through this method undetected because there are not enough immediate adjacent observations. Instead of relying on continuous observations, arterial data quality Figure 3.8. Study sections mapped to original experimental design matrix (3).

43 control focused more on the overall spread of the data. Examination of the arterial data led the team to develop the following quality control processing rules, all of which were applied to the data: 1. Any days with extremely low or high travel times are removed by visual inspection. 2. Rank all travel time for a section, and treat any value greater than the 75th percentile plus 1.5 times the interquartile distance, or less than the 25th percentile minus 1.5 times the interquartile distance, as an outlier. This technique is robust because it uses the quartile values instead of variance to describe the spread of the data. 3. Two consecutive travel times cannot change more than 40%. 4. A travel time cannot be more than one standard deviation above or below the moving average of the 10 previous entries. These 10 previous entries must be continuous and valid data. Rural Freeway Data Processing The rural freeway portion of the study relied on speed data supplied by Inrix on Traffic Message Channel links. From a processing standpoint, Inrix data were treated in the same way as detector data. However, because the Inrix data were provided by relatively short links, and many links comprised the very long rural segments used in the research, a trajectory- based method was used to estimate travel times for the entire segment. The vehicle trajectory method traces the vehicle trip in time and applies the link travel time corre- sponding to the precise time in which a vehicle is expected to traverse the link. For example, a section travel time that begins at 7:00 a.m. will use a link travel time for 7:00 a.m. to 7:05 a.m. at the trip origin, but could use a link travel time from 7:05 a.m. to 7:10 a.m., or from 7:10 a.m. to 7:15 a.m., at the trip destination. The vehicle trajectory method attempts to more closely model the actual link travel times experienced by motorists as they traverse the freeway system. Figure 3.9 shows how the vehicle trajectory method works compared with the snapshot method used for the much shorter urban freeway sections. In the trajectory method, the vehicle stair- steps through the time–distance matrix (rows are time and columns are distance along the route, as indicated by detector location); these are shown as the gray arrows moving up from right to left. Thus, the travel time speed at any given location depends on when the vehicle is at that location. The snapshot method simply takes all the travel times and speeds for a time slice along the entire route (black arrows moving straight across from left to right); that is, speeds are not considered to be time dependent. Calculation of Free-Flow Speed The distribution statistics for the Travel Time Index (TTI) depend on measuring travel time relative to an ideal or free- flow speed. Deviation from the free-flow speed indicates that congestion has occurred. For urban freeways, the research team used a constant value of 60 mph for all sections. This is a well-established threshold for measuring congestion on urban freeways. For signalized highways and rural freeways, the situation is more complex due to variations in speed lim- its and signal-influenced delay, even at very low volumes. For these sections, the 85th percentile speed was used as the free- flow speed. In all cases, if section speeds were higher than the free-flow speed, the TTI was set to 1.0; no credit was given for going faster than the free-flow speed. Final Data Set for Statistical analyses As the preceding discussion and figures show, a large array of data sets at various levels of spatial and temporal aggregation was created. The end result of the processing and fusing was the data set used in the statistical analyses. This data set was highly summarized, which was necessary because reliability is defined over a long period of time to allow all the factors to exert influence on it. Each observation in the statistical analy- sis data set was for an individual section for an entire year for each of the daily time slices studied: peak hour, peak period, midday, weekday, and weekend and holiday. The data set con- tained the data types described in the following subsections; the data were intended to capture characteristics for an entire year on the study section. Appendix A shows the variables in the final data set. Reliability Metrics • Mean, standard deviation, median, mode, minimum, and percentiles (10th, 80th, 95th, and 99th) for both the travel time and the TTI; • Buffer indices (based on mean and median), Planning Time Index, skew statistic, and Misery Index; and • On-time percentages for thresholds of median plus 10% and median plus 25%; and average speeds of 30, 45, and 50 mph. Operations Characteristics • Areawide and section-level service patrol trucks (average number of patrol trucks per day); • Areawide and section-level service patrol trucks per mile (average number of patrol trucks per day divided by center- line mile); • Traffic Incident Management Self-Assessment scores;

44 • Quick clearance law (yes or no); • Property damage only move-to-shoulder law (yes or no); • Able to move fatalities without medical examiner (yes or no); • TMC staff per mile covered; and • Number of ramp meters, dynamic message signs, and CCTVs. Capacity and Volume Characteristics • Start and end times for the peak hour and peak period; • Calculated and imputed vehicle miles traveled; • Demand-to-capacity and annual average daily traffic-to- capacity ratios: 44 Average of all links on the section, and 44 Highest for all links on the section; and • Annual average daily traffic-to-capacity ratios for down- stream bottlenecks by ramp merge area. Incident Characteristics • Number of incidents (annual); • Incident rate per 100 million vehicle miles; • Incident lane hours lost (annual); Figure 3.9. Snapshot and vehicle trajectory methods of estimating travel times from spot speeds (4).

45 • Incident shoulder hours lost (annual); and • Mean, standard deviation, and 95th percentile of incident duration. Work Zone Characteristics • Number of work zones (annual); • Work zone lane hours lost (annual); • Work zone shoulder hours lost (annual); • Mean, standard deviation, and 95th percentile of work zone duration. Weather Characteristics • Number of (annual) hours with precipitation amounts greater than or equal to 0.01, 0.05, 0.10, 0.25, and 0.50 inches; • Number of (annual) hours with measurable snow; • Number of (annual) hours with frozen precipitation; and • Number of (annual) hours with fog present. Assigning Incidents to Study Sections Incidents were assigned spatially to the study sections based on the linear referencing in the traffic and incident data sets. Only incidents that actually occurred on the section were included. Flow on a study section is influenced by incidents that occur immediately downstream of that section, and incidents that occur just beyond the extreme upstream end of the study sec- tion will influence the downstream study section. The decision to include only on-section incidents was based on the appli- cation of the statistical models: it is far easier for practitioners to develop section-specific data than to have to compile off- section data, as well. Also, the goal of the statistical modeling is not to build a deterministic model of traffic flow but to try to capture the cumulative, annual flow characteristics of a section. For the peak hour, peak period, and midday time slices, an incident was assigned to the time slice if it began in the time slice, ended in the time slice, or spanned the time slice. To capture the effect of incidents that occurred immediately before the start of a time slice, a 15-minute window was allowed. The lane hours lost calculation was based on those that were lost solely within the period. For example, consider a peak hour of 7:30 to 8:30 a.m. If an incident began at 8:00 a.m. and lasted until 9:15 a.m. (a total of 75 minutes), only the lane blockages from 8:00 to 8:30 a.m. would count. Description of Seattle Study area This section briefly describes the portions of the Seattle free- way system included in the congestion by source analyses; more detail is provided in Appendices C and D. Figure 3.10 illustrates the 21 study sections. Each of these roadway seg- ments was studied by direction, leading to an analysis of 42 study sections. Five freeways were included in the analysis: I-5, I-405, I-90, SR 167, and SR 520. They were broken into multiple segments based on a combination of geometric and travel patterns. The segmentation of each roadway is described briefly below. Freeway I-5 I-5 was divided into six segments, named (from south to north) South, Tukwila, Seattle central business district (CBD), Seattle North, North King, Lynnwood, and Everett. The basic attributes of these six segments are discussed below. South is the longest segment. It is primarily four lanes wide, with a high-occupancy vehicle (HOV) lane on the left side, and travels from the southern edge of WSDOT’s instru- mented roadway system to the southern I-5/I-405/SR 518 interchange. Its traffic is heavily directional (relative to its capacity). It contains a very large hill located at the northern end of the segment. The hill can affect congestion south- bound in the evening peak period due to the slow speeds of buses and trucks climbing the grade, especially those entering I-5 from I-405 and SR 518. Both directions of traffic can be affected by downstream congestion. Tukwila, the next segment to the north, goes from the southern I-5/I-405 interchange to just north of Boeing Field; it is also mostly four general-purpose (GP) lanes wide, with one inside HOV lane. The northern end of Boeing Field is the approximate end of the backup from much of the recurring congestion that occurs in the a.m. and many p.m. peak periods. Much of that congestion stems from bottlenecks occurring in the next roadway section to the south. In the southbound direction of travel this segment tends to be relatively conges- tion free (the congestion tends to be bottlenecked to the north in the downtown sections). It occasionally suffers from back- ups in the downstream segment, when very severe conges- tion entering I-405 northbound combined with queuing on the South Center Hill can interfere with traffic flow. Other- wise, most congestion is commonly caused by disruptions of some kind. The Seattle CBD section, the next section to the north, contains a significant number of bottlenecks in closely spaced succession. Unfortunately, they are so closely spaced that it was not practical to divide them into separate roadway sec- tions. At its southern end, this is a four-lane GP, one-lane HOV roadway. One of the GP lanes is dropped at the West Seattle freeway interchange. This is followed by the inter- change with I-90, which includes a collector–distributor lane that also serves as a mechanism for separating downtown ramps from some of the mainline traffic flows. Immediately north of the I-90 interchange is the southern terminus of the I-5 express lanes, a reversible roadway that operates primarily

46 southbound in the a.m. and northbound in the p.m. In this stretch of highway, the left-hand HOV lane first becomes a GP lane, and then becomes an exit-only lane to the north- bound express lanes. When the express lanes are operating southbound, these lanes become part of a left-hand exit to downtown. Another of the through lanes also becomes an exit- only ramp to downtown, leaving only two GP through lanes. (One additional lane exists as part of the collector–distributor to I-90 and other downtown ramps.) This is another bottleneck location. This area is followed by a series of on- and off-ramps, including the on-ramp from the collector–distributor, which provides the on-ramps from I-90, to downtown. This section of the freeway also moves underneath the Washington State Con- vention Center, as part of a short tunnel segment, with modest visibility and sight distances. The collector–distributor becomes the third lane when it rejoins the main roadway underneath the convention center, and then a fourth lane is added part way through downtown as an add lane from one of the downtown ramps. No HOV lane exists on this stretch of freeway. Finally, as the roadway exits the downtown Seattle area, it reaches the end of this roadway segment at the SR 520 interchange. The right two lanes become exit-only lanes to SR 520. These lanes are often stop-and-go during both peak periods due to congestion on SR 520. One final bottleneck appears in the last ramp from downtown (Mercer Street), a left-hand on- ramp that sets up a C-class weave, as many vehicles entering at Mercer wish to be in the right-hand lanes in order to exit to SR 520. All these features exist in the southbound direction. The only difference is that the express lanes terminus is an add lane located just south of the downtown core. Consequently, it has less impact on the overall freeway performance than the northbound terminus does. However, the C-class weave from SR 520 to Mercer (again, a left-hand on-ramp followed by a right-hand exit lane) is a bottleneck, as are the effects of the downtown exit- and on-ramps. The North Seattle roadway section is the next section to the north. This section starts at the I-5/SR 520 interchange, goes Figure 3.10. Map illustrating Seattle study sections.

47 across the Ship Canal Bridge, and continues to the northern terminus of the express lanes. This section of roadway has only modest routine northbound congestion. However, southbound, it is affected by a C-class weave from the NE 45th Street and NE 50th Street entrances to the SR 520 inter- change. In addition, the Ship Canal Bridge is exposed to wind, adding to the factors that affect throughput on this roadway. This roadway is four GP lanes wide in the southern section, and becomes three lanes wide with an add–drop lane to Lake City Way (about half way through the study segment). No HOV lane exists in this section of the roadway. Note that this study does not include the express lanes themselves, which serve as the HOV facility (and as additional GP capacity) dur- ing the peak directional movements. The North King section of the roadway starts with the northern entrance of the express lanes and continues to the King County–Snohomish County line. It is four lanes wide, with an HOV lane on the left. The HOV lane starts (ends) at the express lanes terminus. This roadway experiences routine congestion associated with that terminus in both directions. When the express lanes are operating northbound, consider- able weaving takes place into and out of the left-hand HOV lane. Northbound, modest volumes of vehicles move from the left-hand entrance to the GP exits on the right side of the free- way. Southbound, this section of roadway has minor merge- related slowdowns, both when vehicles enter the express lanes, and when the express lanes are closed, as I-5 loses two lanes of capacity at that time (one GP lane and the HOV lane). Lynnwood is the next section of I-5 to the north. This sec- tion of roadway goes from the King County–Snohomish County line to SE 128th Street, and it includes the northern I-5/I-405 interchange. This section of roadway has four GP lanes and one HOV lane. Additional lanes exist at the I-405 interchange to smooth flow between the freeways. Everett, the final I-5 section, is primarily three GP lanes wide with an HOV lane on the left side. Of greatest signifi- cance for this study is that in 2006, north of the instrumented roadway, a major construction project was underway. This project included the extension of the HOV lanes and signifi- cant redesign of the ramps in the city of Everett. These con- struction activities created some backups that extended back onto the Everett study section, mostly late at night, but occa- sionally on weekends. Freeway I-405 The I-405 freeway was divided into six sections. From the south they are • South; • Kennydale; • Eastlake; • Bellevue CBD; • Kirkland–Redmond; and • North. The South section contains two GP lanes and one left-hand HOV lane. This section extends from the I-405/I-5 interchange to the SR 167 interchange. Bottlenecks occur at both of these interchanges, with the most significant of those being the northbound movement. The southern end of this study seg- ment is also significantly affected by on- and off-ramps leading to and from the South Center Mall. (Short ramp lengths and the narrow freeway lead to difficulty merging and the com- mensurate increase in traffic disruption from these ramps.) The Kennydale section is among the most routinely con- gested sections in the region. It stretches from the SR 167 inter- change to 2 miles south of the I-90 interchange. This stretch of road includes the merge (northbound) from SR 167 and diverge (southbound) from I-405 to SR 167. Both of these movements cause major bottlenecks because they are rou- tinely over capacity. North of the SR 167 interchange on I-405 are a series of ramps to and from the city of Renton, which creates considerable ramp disruptions. The freeway then goes up and over a major hill (the Kennydale Hill) which can slow heavy trucks, and there is significant heavy truck traffic on this route as it is the primary route for travel from the region’s major distribution centers to I-90 and all points east. Because the roadway is only two GP lanes and one HOV lane through most of this entire section (there are some add–drop lanes), any slow-moving vehicle is likely to create minor congestion. The roadway is severely over capacity, especially northbound in the morning and southbound in the evening. The Eastlake section of the freeway is a short 2-mile seg- ment designed to examine the effects of I-90 interchange congestion. In the peak directions, this segment is very con- gested; in the off-peak directions it flows well. The Bellevue CBD section stretches from the I-90 inter- change just south of the Bellevue CBD to the SR 520 inter- change just north of the Bellevue CBD. Bellevue is the second largest city in the region and a significant urban center. Con- siderable traffic uses I-405 to reach Bellevue, and the freeway serves a considerable pass-through movement, as well. For traffic coming from the north (including SR 520, which serves the Microsoft headquarters complex), I-405 is the primary connection to I-90 and the other bridge across Lake Washing- ton. As a result of the combination of through movements, large Bellevue-based ramp movements, and the congestion that occurs at the I-90 and SR 520 interchanges, this section of roadway is routinely congested during peak periods. The Kirkland–Redmond roadway section has a southern boundary at the SR 520 interchange and travels north to the SR 522 interchange. Unlike I-405 south of Bellevue (which although directional has a strong reverse direction movement),

48 the Kirkland–Redmond section is very directional, southbound in the morning, northbound in the evening. The roadway changes width from three GP lanes and one HOV lane north of the NE 80th Street interchange to four GP lanes and one HOV lane between SR 520 and Kirkland. In addition to severe demand- related congestion at most of the major on-ramps, the roadway study segment has a very steep hill (uphill southbound) just south of the SR 522 interchange. The North study segment is the last of the I-405 roadway segments. It is a two-GP, one-HOV lane section that extends from SR 522 to the northern I-5/I-405 interchange. This sec- tion has no significant bottleneck points, but it does have some simple capacity issues, primarily southbound in the morning. Freeway I-90 The I-90 roadway was divided into four segments from Issaquah to downtown Seattle. These are (moving from west to east) Issaquah, Bellevue, Bridge, and Seattle. The Issaquah segment is a three-GP-, one-HOV-lane road- way section that travels 6 miles from the city of Issaquah toward Bellevue. While there are no significant geometric bottlenecks on this study segment, it does contain three very- high-volume ramps. The result is routine a.m. congestion westbound. In the evening, some off-ramp queuing can cause delays in the right-hand lanes of the roadway eastbound. The Bellevue study segment covers the remaining distance between Issaquah and the I-405 interchange. Two additional on-ramps add traffic, although an additional lane is added in this section, before becoming a drop lane at the I-90 inter- change. As with the Issaquah eastbound p.m. movement, this roadway section can be affected by significant off-ramp queu- ing to I-405, in this case, in the westbound a.m. peak period. On very bad days queues on I-90 from the downstream sec- tion of I-90 can also reach the western portions of this seg- ment during the a.m. peak period. The Bridge study section contains both I-90’s Lake Wash- ington floating bridge and the stretch of I-90 that crosses Mercer Island, which also contains a short tunnel. A revers- ible express lane also sits in the middle of this study section. (The express lane section is not included in this analysis.) The eastern end of the express lane is located just to the west of I-405. The eastbound exits from the express lanes cause little disruption because of direct ramps from that facility to the I-405 interchange and an add lane to the I-90 mainline. West- bound it causes congestion only when the express lane is east- bound, in which case the HOV lane must merge into the three GP lanes, causing a merge bottleneck. In addition to the ramps from Mercer Island to I-90, several other locations on this section of roadway can become bottlenecks under spe- cific conditions. The most significant are the exit from the tunnel section (which leads to the bridge, and creates some visibility issues when the sun is at certain angles) and the bridge itself (where drivers can also suffer from considerable visual distraction). The Seattle section is the last section on I-90. It covers from the western end of the I-90 floating bridge, through tunnels underneath Capitol Hill, and to I-5, where I-90 ends. West- bound travelers can exit to downtown Seattle or turn north or south on I-5. All three of these ramps can experience queues that extend back onto I-90 depending on the time of day, the types of events occurring in downtown Seattle, and the congestion found on I-5. Eastbound, this roadway section has only one entrance ramp, other than the ramps from I-5 or downtown. Merge congestion is therefore modest. How- ever, backups from the Bridge section of I-90 can easily extend back onto this section, creating congestion. Freeway SR 167 This roadway is east of I-5, and travels in a north–south direc- tion through the region’s primary warehouse and distribu- tion centers. It also serves manufacturing areas and a growing residential population, especially to the far south. This road- way was divided into two study sections for this project, Auburn and Renton. The entire roadway contains two GP lanes and one HOV lane. The HOV lane is now a high-occupancy toll lane, but in 2006 it was still a traditional HOV lane. The Auburn section extends from the SR 18 interchange (the southern end of the surveillance equipment, although not the end of the SR 167 freeway), to the city of Kent. This stretch of roadway has no major geometric bottlenecks northbound, but it does suffer from on-ramp merge congestion due to high traffic volumes northbound in the a.m. Southbound in the p.m., it has a bottleneck at the southern terminus to the study section, where the HOV lane ends (becoming a GP lane), and one of the GP lanes becomes an exit-only lane to SR 18. In addition, due to the restricted number of lanes, traffic south of this bottleneck can move very slowly in the p.m. peak, further worsening the queues observed southbound on the study section. The Renton study section travels from Kent to the I-405 interchange, which is a significant bottleneck. The ramp queues from northbound SR 167 to I-405 frequently back up onto SR 167 in both peak periods (although the a.m. peak is the pri- mary movement), as I-405 simply does not have the capacity to accept the SR 167 traffic volumes. Southbound the SR 167 section also congests because of very high traffic volumes. There are no significant geometric causes for those delays. Freeway SR 520 The final roadway in the study section, SR 520, was divided into two sections, called Seattle and Redmond.

49 The Seattle section goes from I-5 across the Lake Washing- ton floating bridge to I-405. This section is two GP lanes. There is an HOV lane only in the westbound direction; that HOV lane ends in a lane drop at the approach to the bridge itself. The bridge has no shoulders. The lack of shoulders means any incident occurring on the bridge approaches or on the bridge itself blocks a lane. On the western end of the study section are two ramps, one of which leads to the University of Washington. This roadway operates near capacity in both directions over 13 hours each weekday. Because both direc- tions are capacity constrained, the directional volumes are roughly equal throughout the day. The primary difference in the measured performance of the two directions for this roadway is the location of the bridge relative to the entire study section. Eastbound, the study section travels a little over 1 mile from I-5 to the bridge itself, and all of this distance is a two-lane roadway. This means that the measured queue east- bound is never larger than roughly 1 mile. Once the queue grows larger than 1 mile, it extends onto I-5, where its effects are felt in the southbound Seattle North study section or the northbound Seattle CBD study section. Conversely, in the westbound direction, the study section allows for the mea- sured queue from the bridge deck to extend for more than 3 miles. In the heart of the p.m. peak period, this entire road- way section is routinely stop-and-go congestion. The Redmond study section includes that section of SR 520 from I-405 east to the end of the freeway, a signalized intersection with SR 202 and other local roads. The freeway branches into two parts as it ends, each of which ends at a signal. The freeway passes by the Microsoft headquarters campus. Consequently, significant traffic volumes move toward the center of this study section in the a.m. peak period and away from the center of the study section in the p.m. peak period. In addition, the eastern end of the roadway serves a large residential population that travels to both Bellevue and Seattle. Thus the a.m. peak also contains a large westbound home-to-work movement that extends the length of the study section, while the p.m. peak contains a large work-to- home movement. The signalized intersections at the eastern end of the facility create this section’s only major bottleneck. The signals cause significant congestion to extend back from the eastern end of the facility during the p.m. peak period. In the morning, these lights simply serve to meter traffic entering the roadway, allowing the roadway to operate fairly well. The only other bottlenecks that occur are minor ramp delays leading to Microsoft (these can add considerable delay to travelers headed to Microsoft, but they do not significantly affect the main freeway lanes) and queues that originate on the Seattle section of SR 520 but extend back onto the Red- mond section. This happens, on average, at least once a week, usually as a result of crashes or other major traffic incidents on the Seattle section of the roadway. references 1. NCDC Climate Data Online. National Climatic Data Center, National Oceanic and Atmospheric Administration, U.S. Depart- ment of Commerce. http://www7.ncdc.noaa.gov/CDO/cdo. 2. Texas Transportation Institute. Quality Control Procedures for Archived Operations Traffic Data: Synthesis of Practice and Recom- mendations—Final Report. Office of Highway Policy Information, Federal Highway Administration, U.S. Department of Transporta- tion, 2007. www.fhwa.dot.gov/policy/ohpi/travel/qc/index.cfm. 3. Turner, S., R. Margiotta, and T. Lomax. Final Report—Monitoring Urban Freeways in 2003: Current Conditions and Trends from Archived Operations Data. Report FHWA-HOP-05-018. Federal Highway Administration, U.S. Department of Transportation, 2004. http://mobility.tamu.edu/mmp/FHWA-HOP-05-018/. 4. Cambridge Systematics, Inc., Texas Transportation Institute, Univer- sity of Washington, and Dowling Associates. Guide to Effective Free- way Performance Measurement. NCHRP Web-Only Document No. 97. Transportation Research Board of the National Academies, Washington, D.C., 2006. www.trb.org/TRBNet/ProjectDisplay.asp? ProjectID=822. Accessed May 16, 2012.