Guide for Identifying, Classifying, Evaluating, and Mitigating Truck Freight Bottlenecks (2017)

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 4 4.1 Organizing the Speed and Volume Data (Conflation) “Conflation” is the process of matching probe speed data to roadway volume data for subsequent analysis. It is necessary for computing per- formance measures for truck bottleneck analysis when the speed and roadway volume data are provided on different networks. The first step in the conflation process is determining which road- way network will serve as the base network for conflation. The base network is the roadway network, which gets the attributes from the other network loaded on it. Generally, the base network should be the network that more closely aligns with the purpose for the analysis. Because datasets are large and processing time can be lengthy, it is important to consider if any records can be eliminated (i.e., by excluding some functional classes to speed processing time). The process of conflation is facilitated by using GIS information to import and compare the end points of the speed data roadway network with the traffic volume inventory. Quality control is a necessary step to ensure that the data from the speed network aligns with the volume net- work. More information on conflation can be found in recently available research. (12) By combining vehicle speed and truck and passenger car volume data, agencies can compute not only when and where congestion occurs, but the relative size of the delays (in vehicle-hours and truck-hours) that each congestion location causes. It also is possible to track the frequency with which congestion forms. These delay statistics are the primary congestion bottleneck iden- tifiers. By summarizing these data at the location level and using a GIS, it is possible to illustrate on a map the locations of the largest congestion bottlenecks and to develop tabular summaries of the relative sizes of those locations. Figure 4-1 illustrates how a GIS map can be used to illustrate the locations of congestion bottlenecks. Likewise, performing trends over years, agencies also can produce top improvement loca- tions year over year. These locations provide insights into where top delay reductions occur, typically from capacity improvements and/or construction completion. An example is shown in Figure 4-2. Note that red is used in Figure 4-1 to highlight the poor-performing segments, while green is used in Figure 4-2 to accentuate the communication of improved segments. To compute the locations and relative sizes of bottleneck locations, it is necessary to link the available vehicle volume and speed (or travel time by segment) data in a manner that allows the computation of delay statistics. The difficulty of this task is that different data sets tend Organize Data “Conflation” is the process of matching probe speed data to roadway volume data for subsequent analysis.

32 Guide for Identifying, Classifying, evaluating, and Mitigating truck Freight Bottlenecks to use different location referencing systems and time-reporting periods. The specific issues associated with linking databases are: • Point versus segment data. Some data that describe vehicle performance on roads (e.g., vehi- cle speeds) are reported for a point in space. Other data may be reported as travel times over a specified distance (a roadway segment). • Different location referencing systems. Many state highway agencies reference locations on roads by route number (or name) and milepost. Other mapping systems reference location by X/Y coordinates. GISs use a series of defined lines and nodes to describe roads. Other location systems use roadway segment IDs with specific naming conventions. • Direction of travel information. Some highway representations combine both directions of travel into a single road segment description. Other highway representations split the two directions of travel into two separate descriptions, even when those different directions of travel are physically connected. • Different road segment definition. One system might define a road segment as consisting of uniform traffic volume extending from an on-ramp to the next off-ramp. A different system Source: 2013-2014 Indiana Mobility Report: Summary Version, available from http://docs.lib.purdue.edu/imr/ on December 20, 2014 (129). Figure 4-1. Top 20 Interstate segments by total delay.

Organize Data 33 might be based on pavement type, which might change several times within that uniform traffic volume segment and might not have the same end point as the volume-based system. • Different time-referencing approaches. In some instances data describe a specific point in time (e.g., “a vehicle was traveling at 65 mph at 11:07:25 at this location”). In other cases, data are reported as the average of, or total number of, multiple vehicles passing a point during a given time interval (e.g., “the traffic volume at a defined point in the road was 2,300 vehicles from 6:00 a.m. to 7:00 a.m.”) The time interval in which these summarized data are reported can vary from very short, such as 20 seconds for urban freeway systems, to very long, such as the AADT volume reported within HPMS. For analysis purposes, these different referencing systems must be connected during conflation. All the data to be used in the bottleneck analysis must be transformed into a common data structure that describes the conditions to be found on defined road segments during defined time periods. When conducting truck bottleneck analysis, one straightforward choice for a roadway seg- ment and time period data structure is the organizational structure used for the speed data set Source: 2013-2014 Indiana Mobility Report: Summary Version, available from http://docs.lib.purdue.edu/imr/ on December 20, 2014. Figure 4-2. Top 20 Interstate improved segments from 2012 to 2013 by total delay.

34 Guide for Identifying, Classifying, evaluating, and Mitigating truck Freight Bottlenecks an agency plans to use for its bottleneck analysis (e.g., NPMRDS or other vendor). Speed data can be transformed to fit the road segments for which volume data are available or both data sets can be transformed into a third roadway segmentation system. This last option, forming a “composite” segmentation system, is illustrated in Figure 4-3. The “best” of these transforma- tion options will depend on the data available to each state. 4.2 Travel Time (Speed) Data Organization The roadway segmentation system the NPMRDS uses is the TMC protocol, which is com- monly used by Internet mapping companies. TMC segments are directional so that travel times are provided by direction of travel. A GIS shape file that defines each segment is provided along with the NPMRDS travel-time data. The shape file indicates the start and end points of each TMC segment, including the X/Y coordinates of those end points. The NPMRDS dataset also provides a variable that lists the length of each segment. This allows users to convert the reported travel times into estimates of the average speed of the cars, trucks, or “all vehicles” combined as they travel over that TMC, if they desire that statistic for analytical purposes. The result is that the NPMRDS data can be organized into a file structure that looks like the matrix in Table 4-1, where each cell in the matrix contains the travel-time value for that TMC segment for that 5-minute period. The NPMRDS provides data for cars and trucks separately so that separate matrices can be created for car travel times and truck travel times. It also provides an “all vehicle” travel-time estimate. Using the length provided for each roadway segment, it also is possible to transform the travel-time data into average speed or travel rate statistics for each TMC segment and time period. If these different variables (i.e., car travel time, car speed, car travel rate, truck travel time, truck speed, and truck travel rate) are thought of as the third dimension of the above matrix structure, the data structure can be envisioned as a cube, as shown in Figure 4-4, where: • The vertical axis of the cube is time (and date); • The horizontal axis is the roadway segmentation (location) in the order in which a vehicle would drive a given road (the left most column being the first road segment traversed, fol- lowed by the second column, and continuing to additional columns); and • The depth of the cube consists of different variables. Figure 4-3. Illustration for conflating different road segments.

Organize Data 35 Figure 4-4. Example of preliminary cubic data structure. Segment Length Time Period Road Segment 1 Road Segment 2 Road Segment 3 Road Segment 4 … Road Segment n Time 1 Time 2 Time 3 Time 4 Time 5 Time 6 … Time n Note: The travel-time matrices as above can be by car, truck, or “all vehicles” combined when using NPMRDS data. Table 4-1. Example data structure for use of NPMRDS: travel-time data by segment and time period.

36 Guide for Identifying, Classifying, evaluating, and Mitigating truck Freight Bottlenecks A separate cube would exist for each direction of travel for a given roadway. Each cell of the cube describes a specific aspect of what happens on that road segment at that time period. Additional variables (depth to the cube) to be added to this basic cubic structure are discussed later in this report. These additional variables describe other aspects of what occurs on each road segment during each time period. Previous published work by the Texas A&M Transportation Institute illustrated this concept as a “freight box,” which is expandable to mul- tiple freight modes, commodities and associated performance measures. (13) The term “cube” is used in this Guidebook. 4.3 Volume Data Organization The HPMS submittal describes AADT volume on each roadway segment of the NHS, as well as the percentage of trucks using those roadway segments. As described in Section 3.2, there are challenges in using HPMS data in terms of accuracy, especially on lower classification roadways and in the need to develop truck volume data by time of day. Another issue with using HPMS is the need to conflate the HPMS segmentation with the segmentation used by the speed data (e.g., NPMRDS). For example, the HPMS defines new seg- ments differently than the TMC system used by the NPMRDS (and some data providers). As a result, analytical procedures must be developed to convert the HPMS data into traffic and truck volume estimates that apply to the road segments defined by the NPMRDS. At a minimum this includes determining how to split HPMS traffic volumes by direction and time of day. The NPMRDS allows roadway performance to vary every 5 minutes. Ideally, truck and traffic volume data also should be available at this temporal level of disaggregation. How- ever, converting the annual HPMS statistics describing average annual conditions to estimates of conditions for every 5 minutes of the year requires either consider- able amounts of data or the application of a series of assumptions and transformations. Supplemental data available to the roadway agencies that are not included in the HPMS can be very useful in this process. State DOTs and MPOs do not have detailed car and truck volume data at 5-minute aggregations for each TMC or HPMS segment. How- ever, since the primary use for the 5-minute data at the statewide level is to compute delay to identify the major locations where delay is occur- ring, high precision in these 5-minute values is not necessary. What is needed at this point in the truck bottleneck identification process is a reasonable measure of roadway use that can be applied in conjunc- tion with the probe travel time (speed) data to estimate the size of the observed traffic delays. At this stage in the analysis, the focus is the big picture of computing where major delays occur. Of less concern is the precision of those numbers. Therefore, making professionally reason- able, consistent assumptions is sufficient. For those locations selected for bottleneck mitigation, additional truck and traffic volume data should be collected to ensure the reliability of the engineering and operational designs that come from that work, but that detailed level of traffic vol- ume accuracy required for engineering design is not necessary for the majority of miles of roadway in the NHS simply for bottleneck identifi- cation and initial quantification. The most straightforward approach is to assume a time-of-day traffic pattern (preferably a time-of-day pattern that changes by day of week) Case Study Highlight Each year the Texas A&M Transportation Institute develops a “100 Most Congested Roadways” list for the Texas Department of Transportation (TxDOT). Researchers use probe vehicle speeds and volume data from TxDOT’s Roadway Inventory. A number of performance measures are produced, including total delay per mile for ranking the statewide reporting segments. Reporting segments are also ranked by truck delay per mile. Elements of this case study are highlighted throughout this Guidebook. More details are provided in Appendix B about the study, and Appendix D includes detailed calculation procedures, including the use of time-of-day volume profiles used to match with the 15-minute speed data.

Organize Data 37 and apply that pattern to the AADT and truck percentage estimates submitted under HPMS. See the Texas A&M Transportation Institute case study highlight above for reference to the appro- priate appendices for calculation procedures and an application of these methods. A more complex (and better, where the data are available) approach is for the highway agency to develop and apply separate time-of-day patterns for trucks and cars, as well as adjustments for day of week and month of year, to the average annual daily volume and truck percentage esti- mates from HPMS and/or the statewide roadway inventory database, which commonly has these data elements. These adjustment factors are ideally developed so that they apply to roadways on the basis of the function of each road, that road’s location in the state, the rural/urban nature of the traffic on that road, and the observed traffic patterns within that state. It should also be noted that theoretically, probe data can also be used to estimate truck vol- umes. This can be done by estimating the fraction of trucks that are included in the truck probe data set and expanding the number of “pings” at a particular location to a full estimate of truck counts based on this estimated fraction. This technique has not yet been applied to any notable count databases, but has been applied only to specific truck count locations. 4.3.1 Use of Paired Speed-Volume Observations from Detector Data Where permanent, continuous traffic and vehicle classification counters are located on or close to the TMC segments being studied, data from those devices should be used to develop even better traffic and truck volume estimates for nearby analysis segments. For example, many public transportation agencies have roadway ITS detectors to monitor traffic conditions and operate the transportation system. The benefit of these detectors is that they typically can provide very disaggregate data (lane-by-lane, minute-by-minute) for a specific location. If that location is the specific location for which a truck bottleneck is of interest, the analyst benefits from having very good speed and volume information for analysis and decision making. These data are sometimes called “paired speed-volume observations” because the speed and volume data are collected and available over the same time period. With ITS detectors, vehicle classification data is typically available based on vehicle length. Conversion factors are needed to estimate truck volumes and classifications based on vehicle length data. For truck bottleneck analysis (and prioritization), it is preferred to have the “paired speed- volume observations” occur over a representative time period for the locations of interest. This ensures that they will not rank artificially higher (if measured during a highly congested month/ season) or artificially lower (if measured during a relatively low-congestion month/season). Adjustment factors for factor groups and/or representative sites to the data collection site can aid in selection of the “representative” time period to target for analysis. 4.3.2 Assigning Short-Term Volume Count to Continuous Travel-Time Data Another common data scenario is when traffic volumes are available from a short-term vol- ume count (e.g., 48 hours) and continuous travel-time data are available from a commercial source. Continuous means that the travel-time data are available throughout the year (e.g., for each 5-minute period such as the NPMRDS). A short-term volume count typically implies that data are obtained by road tubes or some other means. As discussed, the application here is summarizing annual bottleneck statistics to prioritize truck bottleneck areas. In this case, there is a need to “adjust” the short-term truck volume count

38 Guide for Identifying, Classifying, evaluating, and Mitigating truck Freight Bottlenecks to the same granularity of the travel-time data, which are available throughout the year in this example. The short-term volume count must be adjusted seasonally (hour of day, day of week, and month of year). The following procedure from the AASHTO Guidelines for Traffic Data Programs can be used to convert a short-term volume count (with at least 24 hours of data) into an estimate of AADT. (14) 1. Summarize the count as a set of hourly counts; 2. Divide each hourly count by the appropriate seasonal traffic ratio (or multiply by the appro- priate seasonal traffic factors); and 3. For each hour of the day, average the results of Step 2, producing 24 hourly averages and then sum the 24 hourly averages to produce estimate AADT. This procedure assumes traffic factors are available from continuous monitoring sites that are the reference site for the segment of interest. Traffic volume by vehicle class (e.g., single-unit and combination trucks) is estimated using a similar procedure where the factors used in Step 2 are those developed by vehicle classes of interest. More details about this procedure are available elsewhere. (15, 16) 4.4 Select Roadway Segmentation A key element to successful truck bottleneck analysis is the determination of the appropriate segmentation of the roadway network for the desired analyses. A roadway “analysis segment” is made up of multiple smaller segments. These smaller segments could be TMCs, roadway inventory segments, or some other spatial determination. To assess the regional nature of truck bottlenecks in an urban area, it is desirable to combine short adjacent segments of the roadway network that have similar congestion patterns. By combining short but similar roadway seg- ments, one can identify “big-picture” urban congestion patterns and the most congested loca- tions in the region. When looking at very detailed congestion data on short segments, one can sometimes miss the bigger picture. A more focused, follow-up analysis of the most congested locations will likely analyze these shorter segments to better understand the specific causes of congestion and possible mitigation strategies. Therefore, longer analysis segments (composed of short, adjacent segments) are recommended for the purposes of regional congestion reporting and identifying potential truck bottleneck locations. Traffic levels, congestion patterns, and traffic operation are relatively consis- tent along these congestion reporting segments. A defined segment should not include a mix of free-flowing traffic and congested traffic. Ultimately, the use and context of the congestion measures are the key determining factors in the definition of reporting segments. For exam- ple, a statewide congestion analysis geared to identifying most congested roadways and truck bottlenecks will likely have longer reporting seg- ments than an arterial street facility-based analysis that is geared toward identifying most congested intersections. Table 4-2 provides key steps for roadway segmentation appropriate for truck bottleneck analy- ses in urban areas. Additional information can be found in research on the topic. (17) 4.5 Create Truck Bottleneck Data Analysis Structure Now that there is a basic understanding of where speed and volume data sources originate, the discussion will return to the cube structure introduced in Section 4.2. Not only traffic speed, travel time, and volume data need to be incorporated into the cube-shaped data analysis . . . longer analysis segments (composed of short, adjacent segments) are recommended for the purposes of regional congestion reporting and identifying potential truck bottleneck locations.

Organize Data 39 Roadway/Area Type Key Steps for Roadway Segmentation All Roadways Short segments should be combined into a reporting segment where traffic levels and resulting congestion patterns are relatively consistent. Reporting segments are almost always defined uniquely for each direction of travel. The possible exceptions are where (1) both travel directions have similar congestion patterns or (2) the scale (e.g., statewide or multiregion) of the analysis is conducive to more aggregate reporting. Freeways and Access Controlled Highways In most cases, a freeway reporting segment will include multiple entrance and exit ramps. Freeway segment endpoints are typically entrance or exit ramps from/to another freeway or major cross street, as this is where roadway characteristics, traffic levels, and congestion patterns are most likely to change. Freeway segments in dense, built-up areas typically range from 3 to 5 miles in length. These sections also are likely to have more frequent ramp access points. Freeway segments in less dense, suburban or exurban areas typically range from 5 to 10 miles in length. These sections are likely to have less frequent ramp access. Arterial Streets In most cases, an arterial street segment will include multiple signalized intersections. Arterial street segment endpoints are typically major cross streets, as this is where roadway characteristics, traffic levels, and congestion patterns are most likely to change. Arterial street segments in dense, built-up areas typically range from 1 to 3 miles in length. These sections also are likely to have higher levels of intersection density. Arterial street segments in less dense, suburban, or exurban areas typically range from 3 to 5 miles in length. These sections are likely to have lower levels of intersection density. Rural Areas Longer reporting segmentation is appropriate (e.g., intercity). Table 4-2. Key Steps in roadway segmentation for different roadways/areas.

40 Guide for Identifying, Classifying, evaluating, and Mitigating truck Freight Bottlenecks structure. All data that describe what is happening on the roadway needs to be incorporated into that structure. Thus, the next step in the data organization effort involves expanding the data stored within the cube structure to include data on the events that affect roadway perfor- mance. By obtaining data on these activities and roadway features and placing them within the data analysis cube structure, it is possible to develop automated procedures that allow agencies to not only readily compute the presence, size, and frequency of travel speed bottlenecks, but also to obtain good insight into the causes of those bottlenecks. Under this approach, the cube structure shown in Figure 4-4 expands to include these additional variables, as illustrated in Figure 4-5. These additional data sets also need to be conflated—that is, matched by time and location to the volume and speed/travel-time data—as described in Section 3.3 for volume and speed data. For some data sets—such as the locations of low-height bridges—this is a fairly simple task. For other data sets, this can be a more difficult task. For example, the fact that data show snowfall at a given airport (a location for which weather data can be readily obtained) does not mean that the weather conditions at that airport accurately reflect the weather conditions on a given roadway segment 20 miles away. This weather example also highlights the fact that it can be difficult to determine exactly what data should be placed in the data analysis cube. Continuing the weather example, although it is helpful to know about snowfall at a given time and location, a better statistic would be the amount of snow actually on that pavement section at that time. This is important because if 2 inches of snow have fallen, that snow may well linger on the pavement long after the snow has stopped falling, continuing to cause traffic to slow. Note: NPMRDS is noted in this example, but any travel time (or speed) data source could be used. Figure 4-5. Schematic of data analysis structure with congestion causation factors.

Organize Data 41 Another example is crash data. It is often relatively easy to assign crash data to a specific road segment and time period. However, the cube analysis structure will be more useful if additional information about that crash is available. For example, data could be added to the cube to describe: • The duration of the crash at the scene, • Whether the crash blocked travel lanes or occurred on the side of the roads, and • Whether injuries or fatalities occurred. Many of these variables can be obtained from crash records. Additional data on roadway events can be obtained from incident response databases. Linking and cross-referencing these different databases and placing the appropriate data from them into the cubic data analysis structure are substantial data management tasks. Chapter 6 in this Guidebook illustrates how these causation variables can be used to identify the factors that influence the formation of congestion and delay at each bottleneck location. Note that at the desktop level of analysis, it is only possible to identify potential causes of congestion. It is not possible to directly identify causation, especially because many factors work in concert to cause bottlenecks. That is, rain, a crash, and high volumes may all be present, complicating the task of assigning specific proportions of the observed delay to any one causation factor. Finally, it is important to once again recognize that “perfect data” are not necessary to gain considerable benefit from this desktop analysis process. For example, having access to crash data but not to incident response data will still allow an agency to determine whether vehicle crashes are likely contributing substantially to delay at a particular location. The better the data, the more robust and accurate the out- puts from the initial desktop bottleneck analysis process; but even with limited data sets, considerable insight into bottlenecks can be gained through the use of the cube data structure. Additional insight can be obtained at chosen bottleneck locations by performing detailed, site- specific analyses. 4.6 Data Quality Control Prior to data analysis, it is important that the analyst perform quality control of the datasets to ensure certain specifications are met. The quality control process typically includes one or more of the following actions (18): 1. Reviewing the traffic data format and basic internal consistency; 2. Comparing traffic data values to specified validation criteria; 3. Marking or flagging traffic data values that do not meet the validation criteria; 4. Reviewing marked or flagged traffic data values for final resolution; and 5. Imputing marked, flagged, or missing traffic data values with “best estimates” (while still retaining original data values and labeling imputed values as estimates). The AASHTO Guidelines for Traffic Data Programs (19) describes these quality control pro- cesses in more detail and the interested reader is referred there for further information. Of particular interest are the definitions for traffic data quality measures, including: • Accuracy, • Completeness (also referred to as data availability), • Validity, • Timeliness, • Coverage, and • Accessibility (also referred to as usability). It is important to . . . recognize that “perfect data” are not necessary to gain considerable benefit from this desktop analysis process.

42 Guide for Identifying, Classifying, evaluating, and Mitigating truck Freight Bottlenecks More specifically, AASHTO spells out validation criteria for vehicle count, classification, and weight data from detector sources. In some cases, quality control by visual inspection is valuable. Visual inspection is helpful when it is not easy to automate the quality control with business rules. Sometimes the human eye is more adept at identifying reasonableness in data-time series, for example, graphing speed or volume plots by time for a variety of days in the month on the same graphic or looking at lane-by-lane speed and volume relationships on the same graph. Visual inspection of graphics like this allow the analyst to identify places where more “drill-down” analyses may be warranted if something suspicious is found. More examples are documented elsewhere. (20) As previously discussed, probe speed data are a cost-effective source for systemwide data col- lection. With the increased and widespread use of probe speed data for truck bottleneck analyses, quality control of these data sources is of particular interest. Appendix C of this Guidebook uses the FHWA NPMRDS as an example to illustrate quality control considerations for a probe speed dataset.