4.0 Methodology

This section describes the data and analytical methods used to locate highway truck bottlenecks and calculate truck hours of delay. The analysis involved three steps:

1. Locating highway bottlenecks;
2. Determining truck volumes at the bottlenecks; and
3. Estimating truck hours of delay at the bottlenecks.

4.1 Locating Highway Bottlenecks

The first step was to locate highway truck bottlenecks. The bottlenecks were located by scanning the FHWA Highway Performance Monitoring System (HPMS) database for highway sections that were highly congested as indicated by a high volume of traffic in proportion to the available roadway capacity (the volume-to-capacity ratio).

The information in the HPMS database is submitted by State DOTs and compiled by the FHWA annually. The HPMS database describes physical and traffic conditions for all major roads in the United States. For reporting purposes, the roadways are divided into sections. The average HPMS roadway section in urban areas is 0.7 miles long. In rural areas HPMS roadway sections are longer; they average 2 miles long and can range up to 20 miles or more in length in very isolated areas.

The HPMS has two databases: the Universe database, which reports physical and traffic conditions on all sections on all major roads, providing about 30 data elements describing each highway section; and the Sample database, which covers a limited number of roadway sections, but provides over 100 data elements for each section. These sections are a statistically selected sample, designed so that information reported on traffic volumes and conditions in the sample sections can be extrapolated to represent other similar, but unsampled, sections.

The HPMS 2002 Universe database was used to scan for interchange bottlenecks on urban Interstate highway sections. From prior work with the HPMS, we knew that almost all urban Interstate interchanges or their adjoining sections were represented in the Universe database. The HPMS Universe database reports traffic volumes for each section but not highway capacity. Capacity was calculated from information on the type of roadway, number of lanes, and default values for lane width, shoulder width, and percent trucks.

After the initial scan, these capacity estimates were replaced with more refined estimates of capacity provided by Battelle. These were calculated by identifying the nearest HPMS Sample section, then extrapolating detailed information from the Sample section to the Universe section to more accurately estimate capacity. The refined capacity estimates provided by Battelle were used in all subsequent delay calculations.

The HPMS 2002 Sample database was used to scan for lane-drop, signalized-intersection, and steep-grade bottlenecks on rural Interstate highway sections, rural arterial roads, and urban arterials. The HPMS Sample database was used because it provides more detailed information with which to calculate highway section capacity. The designs of rural Interstate highway sections, rural arterial roads, and urban arterials vary considerably. Using default capacity values and the limited information in the HPMS Universe database, as was done for the more uniform urban Interstate highway sections, does not produce consistently reliable capacity estimates for rural Interstate highway sections, rural arterial roads, and urban arterials. The more detailed HPMS Sample database produces better capacity estimates; however, the HPMS Sample database covers a limited number of highway sections. Therefore, we were able to identify bottlenecks only on those roadway sections that were covered by the HPMS Sample database.¹

The specifics of each scan are as follows:

• Interchange capacity bottlenecks and other roadway capacity bottlenecks on urban freeways. An initial set of urban freeway bottlenecks was identified as part of an earlier study of bottlenecks commissioned by the American Highway Users Alliance (AHUA).² For that study, Cambridge Systematics developed a brief questionnaire that was distributed to all state departments of transportation by the American Association of State Highway and Transportation Officials (AASHTO). The state DOTs were asked to provide information about the worst traffic bottlenecks within their jurisdictions. Twenty-four states responded, identifying about 100 bottlenecks, most of which where urban Interstate interchanges. To supplement the state DOT nominations, an automated scan was run on the HPMS Universe database as part of the AHUA study. That scan identified an additional 150 potential bottlenecks, again primarily interchanges on urban freeways. The HPMS Universe database was rescanned for this white paper using the 2002 HPMS data.
• After the initial scan, Battelle used geographic information system (GIS) technology to map the locations of the sections. With this information, two refinements were made. First, section locations were compared to interchange locations. Where an HPMS Universe section identified in the scan was found to be upstream or downstream of the actual interchange location, Battelle identified the HPMS Universe section closest to the interchange. Second, Battelle identified the HPMS Sample sections closest to the HPMS Universe sections. Battelle used the more detailed HPMS Sample section information to refine the estimates of capacity for the HPMS Universe section. These refined capacity estimates were used in all subsequent delay calculations for the interchange bottlenecks.
• Lane-drop bottlenecks. Roadway capacity bottlenecks on rural Interstate highway sections, rural arterial roads, and urban arterials without signals were identified using the HPMS Sample database. Sections with a volume-to-capacity ratio greater than 0.925 (highly congested) were selected for further analysis.
• Signalized intersection bottlenecks. The HPMS Sample database also was used to scan arterials and locate highly congested signalized intersections. Again, sections with a volume-to-capacity ratio greater than 0.925 (highly congested) were selected for further analysis.
• Steep-grade bottlenecks. Steep grades on interstates and arterials were identified by scanning the HPMS Sample database for roadway sections with grades greater than 4.5 percent and more than a mile long.

Section 5.0 summarizes the findings of the scans. The urban Interstate interchange bottlenecks are summarized in Section 5.0 and listed in Appendix A. The steep-grade, signalized-intersection, and lane-drop bottlenecks are listed in Appendices B, C, and D, respectively. No scans were conducted for rail grade-crossing bottlenecks or regulatory barrier bottlenecks such as those at international border crossings.³

4.2 Determining Truck Volumes at the Bottlenecks

The second step in calculating truck hours of delay was to determine the number of trucks passing through the bottlenecks. The earlier AHUA study did not differentiate automobiles from trucks in calculating the vehicle hours of delay caused by the bottlenecks.

Two sources of truck volume data were used: the FHWA Freight Analysis Framework (FAF) database was used to identify truck volumes for the interchange bottlenecks; and the HPMS Sample database was used to calculate trucks volumes for the roadway capacity, intersection/signal capacity, and steep grade bottlenecks. The next sections describe the databases, their strengths and weaknesses for the purposes of this paper, and how the truck volumes were estimated.

The FAF Database

The FAF is a database of county-to-county freight flows over the national highway, railroad, water, and air freight networks. The FAF is based on public and private surveys and estimates of the tonnage of freight moving into and out of each county. The freight movements are described by commodity type and mode. The commodity tonnage estimates in the FAF are tied to national, regional, and industry economic input-output models so that future year freight flows can be estimated from anticipated industry growth rates.⁴ For commodities shipped by truck, commodity tonnage is divided by the average truck payload for each commodity to estimate the number of truck trips generated or attracted annually by each county.

The current and forecast county-to-county truck trips are then assigned to a FAF highway network. The FAF highway network is a subset of the National Highway Planning Network (NHPN); it includes the Interstate highway system, most major state highways, and many, but not all, urban and rural arterials. The major product of the FAF is an estimate of freight flows—in tons, trucks, and value—over each highway section in the FAF highway network.

Using the FAF database, Battelle identified the volume of "all trucks," "FAF trucks," and "non-FAF trucks" at each of the urban Interstate interchange bottlenecks. While not a precise distinction, the "FAF trucks" represent national and regional, longer-distance truck moves while the "non-FAF trucks" represent metropolitan and local, shorter-distance truck moves. The "FAF trucks" are estimated from the county-to-county commodity flows. The "non-FAF trucks" are estimated by subtracting the "FAF trucks" on each highway from the total of "all trucks" as counted and reported by the state DOT for the HPMS Universe or Sample database section. For the purposes of this white paper, "FAF trucks" are described as "large trucks making longer-distance trips."

The FAF produces reasonably accurate estimates of the number of longer-distance, large-truck trips along major highway corridors, but it cannot estimate accurately the volume of trucks moving on specific roadways, especially on lower-volume roads. The FAF was designed as a national-level freight analysis tool, not a project-level analysis tool. To ensure that data collection and computation were manageable at the national level, the FAF was constructed using county-to-county commodity flow data, which does not include many local, intracounty truck trips, and assumed that all freight shipments originate or terminate at a single central point (centroid) in a county. Using a single centroid in each county as the origin and destination point for truck trips means that trips are routed from the centroid to the nearest major roadway instead of being routed along actual local roads and arterials.

As a result, fewer truck trips are assigned to local roads and arterials and more to major highways. This problem is magnified by the well-known shortcomings of transportation model traffic-assignment procedures. These procedures tend to route longer-distance trips over the most direct major highway when actual truck trips may take parallel and more circuitous routes to pick-up or drop-off shipments, avoid tolls, etc. To address this problem, Battelle checked the FAF assignments against HPMS data and reviewed the results with State DOT staff. Where significant discrepancies were found, the estimated FAF truck volumes were adjusted to correspond to actual on-the-road truck counts.

To compensate for the lack of precision in estimating the number of "FAF trucks" on specific roadways, the number of "FAF trucks" at urban Interstate interchanges were estimated by multiplying the volume of "all trucks" on a bottleneck section by the average percentage of "FAF trucks" in the surrounding urbanized area. The percentage of "FAF trucks" in the urbanized area was calculated by summing "FAF truck" vehicle miles of travel in the urbanized area and dividing by the sum of "all truck" vehicle miles of travel in the urbanized area, as reported in the FAF database.

A similar procedure was used to estimate the percentage of "FAF trucks" making trips longer than 500 miles. The percentage of "FAF trucks" making trips longer than 500 miles was calculated by summing "FAF truck" vehicle miles of travel in the urbanized area for "FAF trucks" making trips longer than 500 miles and dividing by the sum of all "FAF truck" vehicle miles of travel in the urbanized area. The procedure helps identify bottlenecks that delay long-distance freight moves, but does not differentiate between long-distance truck trips that are caught in a bottleneck as they pass though the urbanized area and long-distance truck trips that are caught in a bottleneck because the trip originates or terminates within the urbanized area.⁵

The final step in the analysis process was to estimate the tonnage and value of the commodities moving through the bottlenecks. Battelle identified the commodity tonnage and value for all "FAF trucks" for each of the interchange bottlenecks. These data were used to calculate average commodity tonnage and average commodity value per "FAF truck" and applied to the estimated number of "FAF trucks" and "FAF trucks" making trips longer than 500 miles.

The FAF truck volumes and commodity tonnage and value estimates were based on 1998 data. The FAF estimates were adjusted to 2004 by interpolating the 1998 and 2010 FAF truck volumes.

The HPMS Sample Database

Since most of the lane-drop, signalized-intersection, and steep-grade bottlenecks are on lower-volume highways and arterials, the HPMS Sample database was use to calculate truck volumes for these bottlenecks. The HPMS Sample database provides data on total traffic volume and estimates the percentage of trucks. The HPMS estimates of the percentage of trucks are more consistent than the FAF database estimates for lower-volume highways and arterials; however, the accuracy and reliability of the HPMS estimates vary by state and type of roadway. Some states conduct extensive truck counts and classifications; some conduct infrequent counts and estimate trucks between counts; and yet others apply a statewide "average percentage trucks" to estimate truck volumes for HPMS sections.

The HPMS truck volumes were calculated using 2002 data. The HPMS volumes were adjusted to 2004 using traffic growth factors for each highway section provided by the state DOTs as a part of the HPMS reporting program.

4.3 Estimating Truck Hours of Delay at the Bottlenecks

The third step in the analysis process was to calculate truck hours of delay at each bottleneck. The calculations were based on predictive equations constructed using a simplified queuing-based model, QSIM, developed by Richard Margiotta, Harry Cohen, and Patrick DeCorla-Souza.⁶ QSIM incorporates several features, including:

• Use of queuing analysis to determine delay in oversaturated conditions (queues are carried over into successive time periods and delay in queues is tracked);
• Use of temporal distributions as a basis for developing hourly traffic estimates;
• Estimation of peak spreading;
• Accounting for daily variation in traffic by allowing hourly traffic estimates to vary stochastically;
• For freeways, the inclusion of a capacity drop after flow has broken down (i.e., after the onset of queuing);
• For arterials, considering the effects of signal density and progression;
• Separate functions to estimate speeds in queuing and free-flow conditions based on relationships developed with microscopic traffic simulation models;
• Use of the concept of highway capacity to determine when traffic operates under free-flow and queuing conditions as well as a basis for estimating free-flow speeds and the extent of queuing on the test link; and
• Estimating delay rather than speed as the predictive variable. (Speed is then developed as a function of delay and free-flow speed.)

The model was used to develop a dataset from which a series of predictive equations were developed. The equations use only a few, readily available independent variables for each bottlenecks: annual average daily traffic (AADT), roadway capacity, signal density, and signal progression. The output variable for these equations is "hours of delay per 1,000 vehicle-miles" at each bottleneck. Total truck delay was found by multiplying this value by truck vehicle miles of travel for the bottleneck location. Only the "daily" delay for weekday/weekend combined was considered in this analysis.

The method has been incorporated into the FHWA's Surface Transportation Efficiency Analysis Model (STEAM) and Highway Economic Requirements System (HERS) models. The method is similar in concept to the one used by the Texas Transportation Institute in developing data on national congestion trends, but the development of the method was more detailed for this analysis, particularly with regard to queuing.

Limitations

Delay at Interchange Bottlenecks

At interchanges, the scan identified only the critically congested roadway and the corresponding two-way truck traffic volumes on that roadway. The delay estimation methodology calculates delay resulting from queuing on the critically congested roadway and adjacent highway sections; however, it does not calculate delay on the other roadway at the interchange. This means that truck hours of delay are calculated on only one of the two intersecting highways or two of the four legs on a interchange, probably underreporting total delay at the interchange. The bottleneck delay estimation methodology also does not account for the effects of weaving and merging at interchanges, which aggravate delay, but cannot be calculated from the available HPMS and FAF data.

The Ohio Department of Transportation has commissioned research to develop a more comprehensive delay estimation method based on detailed case studies of congested urban Interstate interchanges in Ohio.⁷ When the results of this research are available, it should be possible to improve the truck-hours-of-delay estimates reported in this white paper.

Incident Delay

The analysis does not account adequately for variability in delay, especially for variability caused by nonrecurring congestion (i.e., congestion caused by incidents and crashes). Much of the delay accruing to trucks, especially in urban areas, is caused by nonrecurring incidents. This type of delay is a major factor in determining the reliability of travel times. Information on the patterns and variability of recurring and nonrecurring urban congestion are being developed under the FHWA's Mobility Monitoring project, but are only available for selected freeways in 29 urban areas and could not be used for this scan.

Truck Exposure to Delay

The calculation of truck hours of delay does not account for actual truck exposure to congestion. The HPMS and FAF databases report annual average daily traffic (AADT) volumes, not hourly traffic volumes. The calculations assume that truck trips are distributed across a 24-hour day much as passenger car trips are; e.g., the highest volume of trips are made in the morning and evening peak periods. However, most motor carriers work aggressively to schedule and route their truck moves outside of peak periods and around known bottlenecks. Truck volumes typically peak during the midday, especially on urban Interstate highways, and are relatively high in the early morning and at night compared to automobile volumes. This suggests that only a portion of the trucks reported in HPMS and FAF may be exposed to the full impact of peak-period congestion; however, the HPMS and FAF do not have information on the distribution of truck trips by time of day. The truck hours of delay reported in this white paper provide a good index to the relative impact of the bottlenecks, but not reliable absolute numbers.

Expansion of Delay Estimates Based on HPMS Sample Data

The statistical-sample framework that underlies the HPMS database is based on volume, mileage, road classification, and state. Volume-related data such as truck hours of delay for the HPMS Sample section bottlenecks can be expanded statistically to estimate total truck hours of delay for all HPMS roadways for all states, but data such as the number of bottlenecks cannot be expanded. This means that while the analysis can identify the number of bottlenecks within the HPMS Sample sections, calculate the truck hours of delay at these bottlenecks, and extrapolate the delay hours to estimate the total, national truck hours of delay for a category of bottlenecks, it cannot identify the total number of bottlenecks or the location of bottlenecks other than those in Sample sections. This also means that the analysis may not have identified the worst lane-drop, signalized intersection, and steep-grade bottlenecks.