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.
27 As a fi rst step in reliability monitoring data collection, agencies need to thoroughly evaluate the existing data sources in their region and determine how they can be lev- eraged to support travel time computations. After this step, agencies can take appro- priate measures to determine how these sources can be integrated into the reliability monitoring system and identify where existing infrastructure should be supplemented with additional sensors or data sources. The fi rst part of this chapter describes the potential sources for traffi c data that can support the computation of route-level travel times and the potential sources of nontraffi c data related to the sources of congestion. Figure 2.1 summarizes the data fl ow from various data sources into route-level travel time probability density functions (TT-PDFs). This chapter discusses the traffi c data sources and data types shown in the two lower portions of the fi gure. The pro- cessing steps for converting the various data sources into route-level TT-PDFs (upper portion of Figure 2.1) are presented in Chapter 3. The two broad categories of traffi c data sources shown in Figure 2.1, infrastruc- ture-based sources (devices mounted along the roadside) and vehicle-based sources, can be further separated into four categories of traffi c data sources: 1. Infrastructure-based detectors that can sense volume, occupancy, and other mea- sures, but not speeds; 2. Infrastructure-based detectors that can sense speeds, as well as other measures; 3. Automated vehicle identifi cation (AVI) systems; and 4. Automated vehicle location (AVL) systems. 2 DATA COLLECTION AND MANAGEMENT
28 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Public agencies typically own and operate the infrastructure-based detectors and the AVI systems (used for tolls), and private, third-party sources often own and oper- ate the AVL systems (although transit agencies are increasingly using AVL systems). This chapter outlines the differences in capabilities between these traffic data collection types and discusses the benefits and drawbacks of available technologies. Figures 2.2 and 2.3 show the categories and types of infrastructure-based and vehicle-based sen- sors described in this chapter. INFRASTRUCTURE-BASED SOURCES Infrastructure-based detectors, which include loop and radar detectors, are already a common component of traffic management systems in many regions. Some can measure vehicle speeds directly, and others use postprocessing algorithms to estimate speeds based on counts and occupancy. The detectors that can directly measure speeds are more valuable. These sensors have a limited ability to monitor travel time reliabil- ity. Although very prevalent, these technologies only provide data values at one fixed location along the roadway, which means they can only report spot speeds. Conse- quently, they cannot directly inform on an individual vehicleâs route or time of travel between two points. As a result, the data they transmit require some processing and extrapolation before travel times can be calculated. This limitation also means that the accuracy of the travel time measures they produce is a function of how closely detectors are spaced along the roadway. If existing deployments have detectors spaced Figure 2.1. Data flow from various sources into route-level TT-PDFs. AVI = automatic vehicle identification and AVL = automatic vehicle location.
29 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Values across a screenline FullOrigin Time stamp Figure 2.2. Types of infrastructure-based data collection sources. Figure 2.3. Types of vehicle-based data collection sources.
30 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY at a frequency of one-half mile or less, they are suggested for inclusion in a reliability monitoring system. If detectors are placed less frequently on key routes, agencies may want to consider either installing more detectors or supplementing the existing detec- tion with AVI sensors. Data from infrastructure-based sources need to be reviewed for quality before calculating travel time reliability metrics. An example of this review is presented in Chapter 4 in the Northern Virginia case study, in which a process is illustrated for identifying detectors that are stuck or not collecting data for inductive loops and radar-based sensors. The Northern Virginia case study also describes challenges and mitigations for integrating the data feed from infrastructure-based detectors with a reliability monitoring system. The following sections discuss technologies that are considered to be infrastruc- ture-based sources. Loop Detectors Loop detectors are located in-pavement on many roadway facilities. They have histori- cally been the most common traffic monitoring tool due to their relatively low instal- lation cost and high performance. Coverage, however, varies greatly among cities and states. In many urban locations, they are common on freeway facilities. Many arterials also use loop detectors to control actuated and adaptive traffic signals. However, it should be noted that loop detectors used in traffic-responsive signal systems are usu- ally not well adapted to providing the data required to support reliability monitoring. First, due to memory constraints in the traffic signal controller, data are usually not collected for each individual lane and are usually discarded rather than being sent to a control center for archiving and analysis. Even when data are sent to a control center, they are usually aggregated up to levels that are too high (e.g., hourly flow and occupancy values) to enable accurate travel time calculations. However, in some cases it will be possible for agencies to modify the existing signal system sensors to col- lect additional data and transmit them to a centralized location to support reliability monitoring. Loop detectors typically measure traffic volumes and occupancies and send data to a centralized location every 20 to 60 seconds. From these data, spot speeds can be calculated with reasonable accuracy, at least in freeway applications, and can be used to extrapolate travel times. Loop detectors in a dual configuration can directly report speed values. However, loop detectors have two significant drawbacks: their intrusive installation and their significant maintenance requirements stemming from their vul- nerability to damage or destruction. Wireless Magnetometer Detectors Like loop detectors, wireless magnetometer detectors are located in the roadway sur- face. However, wireless magnetometer detectors can be installed simply by drilling a hole into the pavement, eliminating the need for cutting pavement during installation and reducing maintenance requirements. These sensors use radio signals to communi- cate with access points located on the roadside, usually on poles, preventing the need to hardwire a detector to a controller cabinet. Like loop detectors, they report volume
31 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY and occupancy data with a granularity that depends on the sensorâs setting. Sensors in a dual configuration can also directly report speed values. Because the data accuracy of wireless magnetometer detectors is similar to that of inductive loops (1), wireless magnetometer sensors are a good alternative to loop detectors for agencies that wish to install additional in-road infrastructure detectors. Recent developments have also adapted some wireless magnetometer detectors to reidentify vehicles at a second detec- tor, giving them AVI capabilities that are described later in this chapter. Video Image Processors Many agencies have begun installing video image processors, especially on arterial facilities, as an alternative to loop detection. Video image processing can retrieve vol- ume, occupancy, and speed data from cameras on the roadway. This technology usu- ally requires users to manually set up detection zones on a computer that are in the field of view of each camera, meaning that it is important that the cameras not be moved and the detection zones be set up correctly. Some specialized systems can also reidentify vehicles detected at two cameras (2), giving them AVI capabilities (discussed below). This technology could be a viable method for travel time reliability monitoring for agencies that already have video cameras installed. Video-based detection generally works best in locations with relatively mild weather, as they are sensitive to challeng- ing weather conditions such as snow, fog, or temperature change. Radar Detectors To overcome the intrusive installation and maintenance of loop detectors, many re- gions have deployed microwave radar detectors, which are placed overhead or road- side and measure volume and speed data. One drawback to radar detectors is that they can lose their speed calibrations. Nevertheless, radar detectors are a viable option for agencies that want to increase the frequency of data collection infrastructure along a roadway without installing additional loop detectors. Other Infrastructure-Based Sources Other overhead vehicle detection technologies have capabilities similar to those of microwave radar detectors. These can be considered on a site-specific basis or used for travel time reliability monitoring in locations where they have already been deployed. These technologies include passive infrared sensors, ultrasonic sensors, passive acous- tic array sensors, and combinations of these technologies (2). AUTOMATED VEHICLE IDENTIFICATION SOURCES AVI data collection sources, which include Bluetooth readers and license plate readers (LPRs), detect a passing vehicle at one sensor and reidentify the vehicle at a second sensor, allowing the vehicleâs trip time between two points to be directly computed. The drawback of AVI technologies is that although they provide the trip time between two points, they cannot identify the route taken by individual vehicles or whether the trip included any stops. Since there are often multiple ways to travel between two points, especially in urban areas, and since the travel time without stops is the value
32 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY of interest in reliability calculations, processing and filtering is required to ensure that reliability computations are based on representative travel times for a given route. Techniques to perform this data processing are described in Chapter 3, and an ex- ample is illustrated in the SacramentoâLake Tahoe, California, case study in Chapter 4. Inaccuracies can also be reduced by deploying sensor readers at frequent intervals to decrease the likelihood that a vehicle took a different route than the one assumed in the computation. Appendix B provides guidance about sensor spacing and sampling. The technologies discussed below are sources for AVI trip time data. Bluetooth Bluetooth receiver technology has only recently been applied to traffic data collection, but appears promising for measuring trip times, especially on arterials. Bluetooth de- tectors record the public media access control address of a driverâs mobile phone or other consumer electronic device as the vehicle passes a point. This recorded ID num- ber (or a curtailed version of it, to reduce privacy concerns) can then be matched as the vehicle passes subsequent detectors, allowing travel times between points to be cal- culated. Some vendors are also developing Wi-Fi detectors based on similar principles. This technology is advantageous in that it is accurate, low cost, and portable. A drawback, however, is that only a small percentage of drivers have Bluetooth-enabled devices in their vehicles; estimates range from 5% in the Washington, D.C., metro- politan area to 1% outside Indianapolis (3, 4). Still, it can be assumed that these per- centages will grow as commercial Bluetooth applications, particularly smartphones, become more prevalent, making Bluetooth an important data collection alternative for future projects. Several issues with Bluetooth measurements need to be accounted for in the data filtration process. First, Bluetooth readers frequently record the same wireless network ID more than once as a vehicle passes, especially when vehicles are traveling slowly. These duplicate addresses have to be removed to avoid counting a vehicleâs travel time more than once. Second, Bluetooth readers have a wide detection range that could col- lect travel times that do not reflect actual conditions. For example, a Bluetooth sensor station on a freeway might detect a vehicle waiting in a queue on an entrance ramp, thus generating a travel time that is not representative of mainline freeway traffic flow. These unrepresentative trip times have to be filtered out during data processing. Additionally, on arterial streets, Bluetooth readers report trip times from nonvehicular modes like walking or cycling; these times have to be removed in the data cleaning process. License Plate Readers LPRs employ cameras that capture a digital image of a vehicleâs license plate and use optical character recognition to read the plate number. Although primarily used for toll enforcement, LPR can also be used to calculate trip times for vehicles that pass by two or more cameras. LPR has been used by the Florida Department of Transporta- tion (DOT) to distribute travel times to changeable message signs in various Florida cities and has also been used on the Arizona SR-68 work zone project near the Hoover Dam (5).
33 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY The advantage of LPR is that it can collect trip time samples from vehicles without requiring the presence of any specific device within the vehicle. However, this method has several drawbacks. It is not well suited for data collection on high-speed freeways. Plate matching is not always accurate, especially during adverse weather conditions (6). The equipment needed is costly, and there are privacy concerns that come with tracking a vehicle by its license plate number. The FHWA Travel Time Data Collection Handbook includes the following guide- lines for using LPR (6): ⢠Place cameras at midroute checkpoints such as major interchanges, intersections, jurisdictional boundaries, and transition points between different roadway cross sections or land uses with the following spacing (which can vary depending on the roadway network and the desired detail of study): â Freeways and expressways, high access frequency: 1 to 3 mi; â Freeways and expressways, low access frequency: 3 to 5 mi; â Arterial streets, high cross-street frequency: 0.5 to 1 mi; and â Arterial streets, low cross-street frequency: 1 to 2 mi. ⢠Fifty license plate matches is the target sample size for a given roadway segment and time period to accurately represent travel time variability. The percentage of successful matches is about 5% to 20% in a given period. Because of LPRâs accuracy issues and high cost, it is recommended that only those locations that have already installed LPR infrastructure use it as a primary method of data collection for reliability monitoring. Radio-Frequency Identification Radio-frequency identification (RFID) technology is employed in electronic toll collec- tion and can be used to reidentify vehicles for travel time purposes. RFID is embedded in toll tags such as E-ZPass on the East Coast and FasTrak in the San Francisco Bay Area, California. More than 20 states have locations that use RFID toll tags, including some that have collected travel time data from their systems, such as the San Francisco Bay Area; New York and New Jersey; Houston, Texas; and Orlando, Florida (7, 8). The iFlorida toll tag travel time project found that toll tag penetration is high in urban areas with toll roads, but much lower in other areas. This means that this data collec- tion option is best suited for urban areas with a high toll tag saturation rate. The study found comparable rates of saturation between urban freeways and urban arterials; however, the percentage of vehicles that could be reidentified at a second sensor was lower for arterials because more vehicles enter and exit the facility between sensor stations. As a result, in Orlando, toll tag readers usually only generated between 10 and 20 travel time estimates per hour (8). Agencies should thoroughly evaluate their regional saturation rate of RFID toll tags to determine whether this technology can supply the number of travel time samples needed to robustly estimate reliability mea- sures over time.
34 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY RFID is an alternative for areas that do not have electronic toll collection if agen- cies distribute RFID tags to volunteer drivers. This was done in San Antonio, Texas, as part of the TransGuide traffic monitoring program, where 30,000 tags were dis- tributed in the first 10 months of operation (9). To increase the number of RFID tags in the regional vehicle fleet, FHWAâs Travel Time Data Collection Handbook suggests offering the following incentives: ⢠Priority passage through toll facilities; ⢠Priority passage through weigh and inspection stations (for trucks); ⢠Ability to receive in-vehicle travel information; ⢠Free electronic tag; ⢠Voucher for merchandise or fuel; ⢠Registration for roadside maintenance service; and ⢠Some type of user incentive payment. Aside from sample size concerns, privacy issues are raised because RFID trans- mits data that identifies an individual vehicle. Therefore, if RFID is used to collect travel times, the system will need to encrypt data to remove personal information. The iFlorida deployment does this by sending the DOT database an encrypted hash that represents the toll tag number, rather than the actual toll tag number itself. Vehicle Signature Matching Vehicle signature matching refers to methods that match the unique magnetic signa- ture of a vehicle as it passes over a detector to the same signature from an upstream detector. Single-loop, double-loop, and wireless magnetometer detectors all have this capability. Although loops are not capable of matching every vehicle, research and testing of this method have shown that it can match enough vehicles to provide accu- rate travel time distributions for both freeways and arterials (10). One advantage of vehicle signature matching is that it can use existing detectors in new ways that improve travel time data accuracy. For arterials, it is advantageous over traditional detector data because it estimates travel times without the need for signal phase information. It also offers an additional benefit over other AVI technologies in that it avoids potential privacy concerns through anonymity. This technology has only seen limited use in practice thus far, with projects in a few locations in California, but it appears promising for measuring travel times on both freeways and arterials (10). AUTOMATED VEHICLE LOCATION SOURCES AVL refers to technologies that track a vehicle along its entire path of travel. These methods provide the most accurate and direct measurements of travel times, but they have not yet seen deployment sufficient to provide reliable data on a regional scale. This situation will change as more vehicles become equipped with AVL technologies and agencies become more accustomed to using them for real-time data collection. An
35 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY example of using AVL equipment to analyze transit performance is illustrated in the San Diego, California, case study in Chapter 4. Global Positioning Satellite Any vehicle equipped with a global positioning satellite (GPS) receiver can be tracked along its path of travel to calculate route-based travel times and other traffic data. GPS technology is well suited for accurate travel time calculations because it can pinpoint a carâs location within a few meters and its speed within 3 mph (11). GPS has traditionally been used to calculate travel times through test probe vehi- cles equipped with GPS receivers. The value of these data is limited because of the small number of test probe vehicles typically deployed, and the method does not pro- vide real-time data on a permanent basis. However, even in a more advanced system that monitors all GPS-equipped vehicles in real time, the low market penetration rate of GPS technology will be a constraint on the ability to accurately represent travel time variations. However, it can be reasonably assumed that more vehicles and devices will have GPS capabilities in the future. GPS is also used by many transit agencies to monitor bus locations and schedule adherence in real time. Another alternative for agencies that wish to monitor reliability is to use equipped buses as travel time probes. By identifying and separating out bus- specific activities, such as dwell times and different acceleration rates, arterial travel times can be estimated from bus AVL data (12). Connected Vehicle The connected vehicle research program, sponsored by the U.S. Department of Trans- portation (DOT), is focused on leveraging wireless technology to allow vehicles and roadway facilities to communicate with one another with the aim of improving safety, monitoring conditions, and providing traveler information (13). The majority of con- nected vehicle research will be completed by 2013. Although the full scope of the contributions that the research will make to reliability monitoring efforts is as yet unknown, it seems that connected vehicle technologies could provide a rich source of travel time information, because the vehicle-to-infrastructure communication channels implemented through the connected vehicle program could be used to send collected vehicle-specific location data to a central data server for travel time processing. Cellular Networks Cellular telephone networks provide cell phone coverage to travelers by shifting the signal to different base stations during travel; this process generates tracking data that can be used to calculate travel times. The precision of location data increases with the number of cellular towers for which a phone is in range. In urban areas, location accuracy can be within a hundred feet (14), which in some cases is too large to assign vehicles to a specific link, especially in dense urban networks. In rural areas, location accuracy can be off by more than a mile, which would negate the value of travel times estimated in this manner. To obtain cellular travel times for reliability monitoring, agencies would either have to partner with cell phone companies or buy data from a third-party provider.
36 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY PRIVATE SECTORâBASED SOURCES Interview results from this project established that many public transportation agen- cies are interested in obtaining data from private sources in order to save time and money on data collection and processing. Although these private sources can provide data for facilities that are otherwise unmonitored (such as arterials), the lack of trans- parency in their proprietary methods of data collection could present challenges for agencies seeking to monitor reliability. The data disseminated by these sources vary. In most cases, the commercial offer- ings provide a speed range (i.e., 30 to 40 mph) for stretches of roadway defined by traffic message channel IDs. These data are, by their very nature, opaque to agencies. For example, it is not clear from the data where on that stretch of roadway the speeds were observed, or when during the time period they were observed. More importantly, little information is given on the methods used to calculate the speeds. For example, the speeds may have been calculated from multiple GPS probe readings on the road- way, and thus be highly accurate, or they may have been interpolated entirely from historical data because no real-time samples were collected during the time period. Transportation Data Private source firms collect data from a variety of sources, including GPS probes, road sensors, toll tags, and cell phones. Many of these firms also collect incident and event data. Typically, the business model of these firms revolves largely around the traveler information market, and they sell data to public agencies as a sideline to their core business. As a result, the data largely reflect the needs of traveler information systems and not performance measurement. The simplest data these firms collect are from fixed roadway sensors. These are largely the result of a series of publicâprivate partnerships, stretching back to the mid 1990s, in which firms were allowed to install and maintain fixed detectors on public roadways, usually in exchange for an exclusive concession to sell the traffic data to another market, such as the local media market. Typically, these data are available to the public agency as part of the concession. In some cases, the agency might procure these data, or additional rights to data they already receive (e.g., as part of a new travel time reliability system). Increasingly, private firms are collecting probe data. Probe data have historically been the purview of freight companies, which were the only firms that had the neces- sary cost incentives to equip their vehicles with GPS. For example, freight companies can rent or purchase tracking devices to place on vehicles and pay a flat communica- tion fee to receive web access and real-time alerts on vehicle locations. Thus, the first data sources for private providers were primarily freight carriers. However, in a world of cheaper GPS and ubiquitous smartphones, this situation is rapidly changing. An estimated 51% of drivers have smartphones, many of whom use the devicesâ GPS capabilities in-vehicle for navigation assistance (15). Firms are increasingly acquiring data directly from consumers as part of the growing personal navigation market. As a consequence, the size and diversity of the probe data sets are exploding.
37 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Data Transparency Although some providers may supply metadata on the data quality (e.g., a ranking scale), the methods for quality assessment remain opaque. For the most part, these limitations are inherent to the business model of the data provider. Private source data providers tend to base their competitive advantage on their network of data sources and data fusion methodologies. Because of this, they are unlikely to reveal the underly- ing sources and methodologies to the agencies that may wish to know. This opacity must be considered by agencies interested in using private source data to produce or supplement reliability information. Generally speaking, an agency must ask itself two questions: ⢠Can we perform the necessary (travel time reliability) calculations with this form of data? ⢠Do we believe that the data are sufficiently accurate? If the answer to both questions is yes, then the black box approach to data acqui- sition may be entirely appropriate for a given agency. The questions generally must be answered on an application-by-application basis. For example, to answer the first question, an agency might attempt to build travel time distributions out of data binned by speed to examine if these simplified distributions were adequate to its needs. To answer the second question, an agency could purchase a data sample from a firm and independently test its accuracy. DATA MANAGEMENT FOR RECURRING CONDITIONS Travel time reliability is often monitored through a combination of real-time and off- line systems. Offline, regimes are developed that characterize the manner in which seg- ments and routes operate. In real time, tests are performed to determine what regime is operative for a segment or route at any given point in time and what regimes are likely to be operative in the short-term future. The regimes are described by PDFs that portray the distribution of travel times likely to be experienced by travelers. The best database design for this type of situation is a data warehouse model in which data are stored in the database in a nonnormalized format at multiple levels of aggregation and imputation. This technique ensures that the system can support report generation at several levels of temporal and geographic granularity. A data warehouse is a type of relational database that is composed of two- dimensional storage elements (tables). Each table has fields (columns) and records (rows). Tables relate to one another through common fields. An example relation- ship between tables in a travel time database is shown in Table 2.1. The top table of Table 2.1 contains background information on routes in the system, and the bottom table contains 5-minute travel times. The two tables relate to one another through the Route ID field.
38 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Five types of tables are commonly used in reliability monitoring systems: 1. Configuration information. This is information that is not in real time and does not change in the field. Examples of configuration information include the defi- nitions of the routes for which reliability will be monitored and the locations of detectors and sensors on freeways and arterials. 2. Raw data. These data are directly received by devices in the field and are not pro- cessed in any significant way. The form of the raw data is dependent on the sensor technology; for example, data from loop detectors contain different fields from data from a GPS device. As a result, different raw data tables are needed for each of the four data sources: infrastructure-based detectors that can sense volume, occu pancy, and other measures, but not speeds; infrastructure-based detectors that can sense speeds, as well as other measures; AVI; and AVL. Examples of these are included in Appendix A. 3. Travel time information by sensor type. Because the computational process for deriving travel time information from the raw data is dependent on the technology used in data collection, different database tables are needed to store the travel time information derived from the different source types. Where routes are monitored by multiple technologies, data from the different sources can be fused to improve the travel time information. In these cases, a final database table is needed to store the fused travel time data, which ultimately are used to update the travel time density functions and regimes and facilitate the reliability computations. 4. Travel time density functions. These density functions are derived offline from the travel times kept in the data tables described above. They are commonly separated by segment and route regimes that have been identified by the travel time reliabil- ity monitoring methodology described in Chapter 3. 5. Reliability summaries. These tables contain information that portrays the manner in which segments, routes, and special areas have been operating in specific spatial and temporal time frames. TABLE 2.1. EXAMPLE DATABASE TABLE RELATIONSHIP Route ID Route Name Length (mi) 009 Golden Gate Bridge to Bay Bridge 6 020 Walnut Creek to Oakland Coliseum 15 116 Downtown Berkeley to Rockridge 6 Time ID Route ID Travel Time (min) Operative Regime 11/25/200917:00:00 009 25 1 11/25/200917:00:00 020 32 3 11/25/200917:00:00 116 15 2
39 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Database design usually takes the form of a schema that formally describes the database structure, including the tables, their relationships, and constraints on data value types and lengths. Appendix A presents sample tables that can store information generated during all steps of the reliability monitoring computation process, from the raw data to the travel time density functions and reliability metrics. The exact tables, fields, and relationships are flexible to the needs of the agency, the data available, and the desired reporting capabilities. DATA MANAGEMENT FOR NONRECURRING EVENTS Additional information beyond the already-discussed traffic data is required to under- stand the impact of nonrecurring events on travel time reliability. The primary non- recurring events that affect reliability are incidents, weather, construction, and special events. The ability of agencies to collect data on these events, and the types of data they can collect, will vary between locations. Because of this variation, the discussion in this section is meant to be flexible, so that agencies can tailor recommendations to fit the data that they have at their disposal. Data Characterization The four major sources of nonrecurrent congestion (incidents, weather, construction, and special events) are described below. Transportation Incidents Traffic incidents can reduce available roadway capacity. Incidents include accidents, vehicle disablements, debris on the roadway, and vehicle abandonments. Higher fre- quency of incidents on specific roadway segments is typically associated with greater travel time variability. There are many viable sources for collecting incident data. Most state (and some local) emergency response agencies use computer-aided dispatch sys- tems to respond to incidents; these systems have feeds that can be used by transporta- tion agencies. The benefit of this data source is that it is in real time, but the drawback is that the data have not been cleaned (e.g., locations may not be clearly specified and durations may be inaccurate). Many state DOTs have databases with cleaned-up inci- dent records for state highways (e.g., the Caltrans Accident Surveillance and Analysis System) for the purpose of performing detailed analyses. These sources can also be leveraged for reliability monitoring. Another potential source for incident data is the local traffic management center, where operators usually enter incident information into their management software. Finally, private sources such as Traffic.com often col- lect incident data at a high level of specificity from various sources, including video, mobile (patrol) units, and emergency communication frequencies. Although many potential sources for incident data exist, these data are often incomplete, many times lacking severity indicators, clearance times, and exact incident locations. The following variables can be used to relate traffic incidents with travel time vari- ability: location, date, type, starting time and duration, full time to clearance, severity, and lanes affected.
40 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Transit incidents, such as bus collisions or disablements, can disrupt the opera- tions of a transit system and cause major delays. Such incidents are increasingly being detected by the AVL systems used by transit agencies. Weather Severe weather events include rain, snow, extreme temperatures, strong winds, blow- ing dust, and wildfires. Adverse weather typically results in lower vehicular speeds and higher incident frequencies, increasing travel time variability. One source for weather data is existing weather stations operated by various governmental organizations or research bodies. The most accurate sources of weather information are the Automated Surface Observing System and Automated Weather Observing System stations main- tained and used for real-time airport operations by the Federal Aviation Administra- tion. Another good source is an online interface from the National Climatic Data Center of the National Oceanic and Atmospheric Administration, which provides hourly, daily, and monthly weather summaries for 1,600 U.S. locations. For moun- tainous rural areas, the major sources of weather-related delay are closures and chain- control stations. These data are frequently available from rural traffic management centers, although collecting feeds of such data is rare and problematic. One of the richer sources of this data may be highway advisory radio networks, which broadcast closure and chain-control locations and are frequently available via statewide feed. Any weather data obtained from sources not directly on a monitored route will have to be associated with nearby routes in the system. Another option for collecting weather data is to install environmental sensor sta- tions at key roadway locations. Many states have used these to build road weather information systems that archive weather data and use them in roadway-related deci- sion making. The following variables can be used to relate weather to travel time variability: air temperature, type of precipitation, amount of precipitation, visibility, wind speed, pavement temperature, and surface condition. Transit agencies can use similar methods to monitor weather conditions and develop operational plans to help them deal with potential disruptions in service and variability in travel times during a variety of adverse weather events. As of this writing, the process for connecting weather data to travel time monitor- ing systems is largely manual. Examples from the San Diego and Atlanta, Georgia, case studies are presented in Chapter 4. Work Zones Work zones typically reduce the capacity of roadways, increase driver distractions, increase incident frequency, and cause major delays in the transportation network. They are typically associated with higher travel time variability. There are a few dif- ferent sources of information for construction-related lane closures. Many states have lane-closure systems that serve as a communication interface between contractors and state agencies to facilitate lane-closure management. This data source can be obtained in real time. Private sources are another option; for example, Traffic.com reports both
41 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY scheduled and unscheduled construction events. Another option is to manually obtain construction-related information from changeable message sign logs or feeds. The following variables can be used to relate work zones to travel time variability: start time and duration, start and end locations, and lanes affected. Special Events Because there are no standardized, centralized sources for special event data, this in- formation frequently must be collected by individuals rather than by systems. One option is to manually review calendars for major event venues near a route. Another option is to obtain event data from TMCs, many of which collect event logs to know when and where to activate event-based signal-timing plans. The following variables can be used to relate special events to travel time variability: location, routes affected, duration, type of event, and attendance. Data Storage The data storage regime for nonrecurring events is dependent on exactly which vari- ables are collected, and at what granularity. The spatial and temporal resolution of nonrecurring events data is an important consideration that affects the strength of the relationships developed with travel time variability. Data on nonrecurring events, to some degree, need to be aggregated to the same temporal and spatial resolution, in that all the data must be spatially collected by route and temporally collected for each day in the analysis period. Collecting data on some of the sources at higher spatial and temporal resolutions would lead to more accurate analysis. The case studies and use case analyses (Chapter 4) suggest that a granularity of 5 minutes is sufficient to link the effects of nonrecurring events like weather to their travel time impacts. Granularities greater than 15 minutes are too coarse. Finer granu- larities, such as 1 minute or 30 seconds, are helpful but not necessary. The data on nonrecurrent events do not need to be stored in the same tables as the route travel times, as the analysis to link travel time variability with its causes is typically a manual exercise. Thus, the database for nonrecurring events can be uniquely designed to store the data that each agency is able to collect. Examples of typical data tables are pro- vided in Appendix A. The most useful way to store event data is in separate files tailored to the event type and in which beginning and ending times can be recorded along with descriptive information. Time of event data are needed because events have no specific synchro- nization with traffic sampling intervals and should not be forced to be synchronous. Moreover, events can have temporal and spatial impacts outside the bounds of their duration and physical location. For example, as the Lake Tahoe analyses show, the effects of heavy snowstorms extended well beyond the time when the snow stopped falling. In Sacramento, accidents on intersecting freeways, near the interchanges with the subject facility, affected the travel times observed. An important corollary to this is that queries seeking explanatory information for high travel rates or times at a specific location and time must look outside the bounds of that location and time. For example, although the weather data may indicate that the weather event is over, the query must seek causal explanations from earlier weather
42 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY events. Similarly, although there may be no event at the location being examined, the query must look for events on intersecting facilities, as well as downstream of the loca- tion being examined. A significant and unfortunate issue is that event data frequently lack quality infor- mation about duration. Moreover, the end of the interval during which an incident has an impact on travel times tends to be larger than the duration of the incident event itself; that is, increased travel times persist for some time after an incident has been recorded as having concluded. A final note is that AVI and AVL data show travel time (and rate) transients with a granularity dictated by the frequency with which such observations are obtained. The case studies and use cases tend to suggest that with todayâs penetration rate for such devices, keeping all the raw data makes sense (sanitized to protect privacy). Summariz- ing the data based on some granularity (e.g., 5 minutes) has no value and can obscure useful information about travel time and travel rate transients. SUMMARY The final database design for a reliability monitoring system must reflect the technolo- gies used to collect data and the processes used to derive travel time estimates from the raw data. In general, four types of tables are needed to fully describe the monitoring system and its outputs. Configuration tables are needed to define the routes and route segments for which travel times are to be computed, including their starting and end- ing point, the location of their detectors, and the type of detection. Raw data tables are needed to store the unaltered inputs from the various detection types. Travel time tables are needed to store the representative travel times calculated for each route segment, time period, and detection type, as well as the intermediary information generated dur- ing the computation. Finally, reliability tables can be used to store high-level monthly, quarterly, and yearly summaries of reliability statistics for individual routes or higher spatial aggregations. One-way (irreversible) encryption can be applied to protect the privacy of personally identifiable information such as license plate numbers. REFERENCES 1. California Center for Innovative Transportation. Evaluation of Wireless Traffic Sensors by Sensys Networks, Inc. University of California, Berkeley, October 2006. 2. Klein, L., M. Mills, and D. Gibson. Traffic Detector Handbook: Third Editionâ Volume 1. Report No. FHWA-HRT-06-108. TurnerâFairbank Highway Research Center, Federal Highway Administration, McLean, Va., Oct. 2006. 3. Venere, E. Method Uses âBluetoothâ to Track Travel Time for Vehicles, Pedestrians. Purdue University News, May 27, 2008. 4. Traffax, Inc. http://www.traffaxinc.com. Accessed Oct. 16, 2009. 5. PBS&J. License Plate Reader Operations. Florida Department of Transportation, Tallahassee, 2006.
43 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 6. Texas Transportation Institute. Travel Time Data Collection Handbook. Report FHWA-PL-98-035. Federal Highway Administration, Washington, D.C., 1998. 7. Dahlgren, J., and J. Wright. Using Vehicles Equipped with Toll Tags As Probes for Providing Travel Times. California Partners for Advanced Transit and Highways, Berkeley, 2001. 8. Haas, R., M. Carter, E. Perry, J. Trombly, E. Bedsole, and R. Margiotta. iFlorida Model Deployment Final Evaluation Report. FHWA Report No FHWA- HOP-08-050. Federal Highway Administration, Washington, D.C., 2009. 9. Riley, J. D. Evaluation of Travel Time Estimates Derived from Automatic Vehicle Identification Tags in San Antonio, TX. MS thesis. Virginia Polytechnic Institute and State University, Blacksburg, 1999. 10. Kavaler, R., K. Kwong, R. Rajagopal, and P. Varaiya. Arterial Travel Time Estima- tion Based on Vehicle Re-Identification Using Wireless Magnetic Sensors. Transpor- tation Research Part C: Emerging Technologies, Vol. 17, No. 6, 2009, pp. 586â606. 11. New Research Project Captures Traffic Data Using GPS-Enabled Cell Phones. University of California, Berkeley. Feb. 10, 2008. http://www.physorg.com/ news121845452.html. Accessed Oct. 19, 2009. 12. Berkow, M., J. Chee, R. Bertini, and C. Monsere. Transit Performance Measure- ment and Arterial Travel Time Estimation Using Archived AVL Data. Presented at ITE District 6 Annual Meeting, Portland, Ore., 2007. 13. Research and Innovative Technology Administration. Achieving the Vision: From VII to Intellidrive. http://www.its.dot.gov/press/2010/vii2intellidrive.htm. Accessed Sept. 2, 2012. 14. Wunnava, S., K. Yen, T. Babij, R. Zavaleta, R. Romero, and C. Archilla. Travel Time Estimation Using Cell Phones (TTECP) for Highways and Roadways. Florida Department of Transportation, Tallahassee, 2007. 15. J.D. Power and Associates. Smartphone Ownership Drives Increased Vehicle Owner Interest in Communication- and Connectivity-Related Features. Press release. http://businesscenter.jdpower.com/news/pressrelease.aspx?ID=2010095. Accessed Sept. 2, 2012.
