Content uploaded by Michael D. Fontaine
Author content
All content in this area was uploaded by Michael D. Fontaine on Jul 29, 2016
Content may be subject to copyright.
Fontaine, Chun, and Cottrell 1
1
2
USING PRIVATE SECTOR TRAVEL TIME DATA FOR PROJECT-LEVEL 3
WORK ZONE MOBILITY PERFORMANCE MEASUREMENT 4
5
6
7
8
9
Michael D. Fontaine, P.E., Ph.D. 10
Associate Principal Research Scientist 11
Virginia Center for Transportation Innovation and Research 12
530 Edgemont Rd 13
Charlottesville, VA 22903 14
E-mail: Michael.Fontaine@VDOT.Virginia.Gov 15
Phone: 434-293-1980 16
Fax: 434-293-1990 17
18
PilJin Chun 19
Undergraduate Student 20
University of Virginia 21
E-mail: pc8jx@virginia.edu 22
23
Benjamin H. Cottrell, Jr., P.E. 24
Associate Principal Research Scientist 25
Virginia Center for Transportation Innovation and Research 26
530 Edgemont Rd 27
Charlottesville, VA 22903 28
E-mail: Ben.Cottrell@VDOT.Virginia.Gov 29
Phone: 434-293-1932 30
Fax: 434-293-1990 31
32
Word Count: 33
Figures: 6 × 250 = 1500 34
Tables: 2 × 250 = 500 35
Words: 5485 36
Total: 7485 37
38
39
40
41
42
43
44
45
46
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 2
ABSTRACT 1
2
The Federal Highway Administration (FHWA) has been encouraging states to better monitor and 3
track work zone operational performance. The use of mobility performance measures will 4
enable agencies to better assess the contribution of work zones to network congestion, identify 5
specific projects that are in need of remedial action, and potentially assess penalties to 6
contractors creating excessive impacts. A major challenge in implementing work zone mobility 7
performance measures has been the availability of traffic condition data. States have become 8
increasingly interested in using travel time data from private sector vendors to generate this 9
information since this data set offers the ability to obtain condition information over a wide area 10
without deploying any sensor infrastructure. 11
This paper summarizes lessons learned about using private sector data to develop project-12
level work zone mobility performance measures based on experiences in Virginia. A series of 13
case studies are used to show considerations in using private sector data to develop delay and 14
queue length performance measures at four sites. Issues related to the spatial and temporal 15
granularity of the data are discussed, as well as the ability of the data to reflect performance at 16
urban and rural sites. The experience and insight s shown in this paper can help guide agencies 17
to better construct new mobility performance measurement programs using this data source. 18
19
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 3
INTRODUCTION 1
2
The Federal Highway Administration (FHWA) has been encouraging states to better monitor and 3
track work zone impacts by creating performance measurement programs that cover a broad 4
range of exposure, safety, and mobility effects (1). The accurate and consistent tracking of work 5
zone mobility impacts has been particularly challenging, however, as states develop performance 6
measurement programs. Consistent and accurate mobility data would allow state departments of 7
transportation (DOTs) to better assess the overall contribution of work zones to network 8
congestion, identify specific projects that are in need of remedial action, and potentially assess 9
penalties to contractors creating excessive delays. This data could also be useful when 10
evaluating contractor requests to work outside pre-defined allowable work hours. 11
A major challenge in implementing work zone mobility performance measures has been 12
the availability of traffic condition data. Outside of major urban areas, traditional point detector 13
systems like inductive loops or side fire radar are often located at wide spacings. This makes it 14
unlikely that existing sensors would be available to provide data on many operational metrics 15
used by DOTs. In urban areas, traditional sensor coverage is denser, but sensors are often taken 16
off line during construction. Installing new, temporary sensors specifically to monitor work zone 17
mobility is an option, but it is often only cost effective for long-term, major projects. Even then, 18
construction activities could require that sensors be relocated several times during the course of 19
the project, resulting in additional expenses to the DOT. 20
One option that may overcome traditional point sensor limitations is the use of private 21
sector travel time data generated from probe vehicles. Many states have begun utilizing travel 22
time and speed data that has been purchased from private sector companies like INRIX, 23
TomTom, and NAVTEQ for performance measurement and real-time traveler information. For 24
example, the Texas A&M Transportation Institute Urban Mobility Report (2) currently uses 25
INRIX data to quantify congestion in U.S. cities, and the Virginia DOT and Maryland State 26
Highway Administration use INRIX data to provide real-time traveler information on overhead 27
variable message signs (3,4). Private sector companies typically create their travel time 28
estimates using global positioning system (GPS) data obtained from commercial fleet 29
management systems and private vehicles using navigation or traveler information smart phone 30
applications. This location data is then processed to estimate travel times on roadways. Several 31
studies have shown that this data is generally accurate on freeways, and can be used for real-time 32
traveler information and freeway performance measurement (5,6). 33
Private sector data has been used to generate work zone project-level performance 34
measures in several states. One recent study developed work zone travel time reliability 35
measures for 15 projects in Virginia (7). The Ohio DOT has also developed a systematic 36
program to track mobility performance measures on a project-level basis using private sector 37
data from INRIX. Figure 1 shows an example of performance measures generated for one 38
project in Ohio (8). The performance measure shown is the number of hours operating under 25 39
mph. Monthly performance is shown and contrasted to preconstruction performance and speeds 40
observed during the prior calendar year. 41
As DOTs begin to use private sector data to assess work zone mobility, it is extremely 42
important that they have a clear understanding of the capabilities and limitations of this data 43
source. This paper discusses lessons learned from the development of work zone mobility 44
performance measures in Virginia using private sector data, with the goal of identifying key 45
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 4
decisions that DOTs must make before mainstreaming private sector data into their work zone 1
mobility program. 2
3
4
FIGURE 1 Example of work zone performance measures from Ohio DOT (8). 5
6
OBJECTIVES AND SCOPE 7
8
The Virginia DOT has purchased statewide real-time access to private sector travel time and 9
speed data, and has used this information to investigate potential project-level work zone 10
performance measures. The paper discusses several key lessons learned from this investigation, 11
and illustrates some of the capabilities and limitations of this data set using a series of case 12
studies. 13
The objective of this paper is to identify critical issues in the use of private sector data 14
streams for work zone mobility performance measurement. The ability of the private sector data 15
to produce commonly used work zone performance measures is illustrated using data from 16
INRIX for a sample of Virginia work zones. Although the data was provided by INRIX, the 17
lessons learned from this investigation should apply across other private sector providers as well. 18
Some specific issues that were assessed included: 19
20
What performance measures can be calculated reliably using the private sector data? 21
What is the impact of the temporal level of aggregation on the performance measures? 22
How does the spatial granularity of the private sector data impact performance measures? 23
24
The focus of this paper is on project-level performance measures, although many of the findings 25
can be aggregated upward to be applicable to programmatic performance measures. The 26
accuracy of the data has been previously established in other research (5, 6), so that is not 27
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 5
discussed in detail in this paper. Prior work in Virginia validated the data quality of the INRIX 1
estimates by comparing them to Bluetooth reidentification travel time estimates on 615 2
directional miles of freeway between June 2011 and June 2013 (9). Those results showed that 3
INRIX data summarized in 5-minute intervals had a mean bias of +0.45 mph and a mean 4
absolute error of 4.7 mph relative to the Bluetooth data, which was within the error tolerances 5
specified by Virginia DOT. These evaluations covered both recurring and non-recurring 6
congestion. As a result, this paper does not present further detailed analysis of the quality of the 7
data. Rather, the focus of this paper is on how the data is summarized and manipulated to 8
develop performance measures. 9
10
WORK ZONE MOBILITY PERFORMANCE MEASURES 11
12
First, work zone mobility performance measures used by DOTs were reviewed to identify the 13
most common potential application areas for private sector data. A recent domestic scan report 14
(10) summarized common work zone mobility performance measures used by DOTs. Table 1 15
provides a brief summary of the findings of that report. Delay and queue length were the most 16
common performance measures in use by DOTs, although other unlisted measures (such as 17
travel time reliability metrics (7)), are also possible. It is unclear from the literature, however, 18
whether performance measures were selected because they could be easily measured or because 19
they best represent work zone mobility. 20
21
TABLE 1 Performance Measures and Thresholds (adapted from (10)) 22
Agency Performance Measure Performance Threshold
California DOT Delay 0 to 20 minute delay depending on location and complexity
of project
Florida DOT Queue length 2 mile maximum queue on interstates or highways with speed
> 55 mph
Indiana DOT Queue length Queues cannot be present for > 6 continuous hours, or for
more than 12 hours/day. Queues > 1.5 miles are not
permitted, and queues between 0.5 and 1.5 miles have are
limited to between 2 and 4 hours, depending on length
Maryland DOT Queue length
Delay
Level of service (LOS)
Freeways: queues > 2 miles are not acceptable, queues < 1
mile are permitted, and queues between 1 and 1.5 miles are
limited to 2 hours
Delays < 15 minutes on arterials
LOS requirements are set separately for signalized and
unsignalized intersections, depending on initial LOS
Michigan DOT Delay
Volume/Capacity (V/C)
Ratio
Level of Service
Delays < 10 minutes
V/C < 0.8
Drop in LOS < 2 levels, no LOS worse than D
Missouri DOT Delay Delays > 15 minutes are considered excessive
New Hampshire DOT Delay Delays > 10 minutes undesirable
New Jersey DOT Delay Delays < 15 minutes
Ohio DOT Queue Length Queues > 1.5 miles are not acceptable
Oregon DOT Delay Delays < 10% of the peak travel time
Pennsylvania DOT Delay Delays between 15 - 30 minutes limited to 2 consecutive
hours
Wisconsin DOT Delay Maximum of 15 minutes of added delays between major city
nodes
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 6
1
Given the popularity of delay and queue length measures, this paper focuses on the ability of 2
private sector data to produce these two metrics. Issues with applying private sector data to the 3
calculation of each of these two performance measures are discussed below. 4
5
Using Private Sector Data to Calculate Delay Metrics 6
7
Delay is a measure of the additional time incurred by travelers as a result of the work zone. 8
Delay can be summarized in several ways, including: 9
10
Average delay per vehicle 11
Average delay per person 12
Total vehicle-hours of delay 13
Total person-hours of delay 14
15
While private sector data directly provides the speed or travel time information required 16
to determine average vehicle delay values, additional information is required to determine other 17
metrics. The total vehicle-hours of delay measure requires that consistent traffic volume data be 18
available. The per person measures also require that data on vehicle occupancy be available. 19
In order to calculate all of these delay metrics, the analyst must have at least two pieces 20
of information: 21
22
Data on travel time or speed in the work zone 23
A benchmark travel time or speed for comparison purposes 24
25
These values are compared to determine whether the work zone has created a negative 26
mobility impact. As a result, the selection of the benchmark travel time/speed is a critical part of 27
the calculation of delay. Common benchmarks for delay and their advantages/disadvantages are 28
shown in Table 2. The different benchmarks have tradeoffs in terms of ease of 29
calculation/collection versus their ability to separate work zone impacts from background 30
conditions. 31
32
TABLE 2 Advantages and Disadvantages of Different Benchmarks for Delay Calculations 33
Benchmark Advantages Disadvantages
Posted speed limit Easily defined and understood, often
readily available in DOT databases,
provides constant benchmark for site
Posted speed limit may not be realistically
attained during higher volume periods of the
day, especially on arterial routes; may cause
delay attributed to the work zone to appear
higher than it should be
Free flow speed Theoretical upper maximum of travel
speed, easily understood, provided by
some private sector companies,
provides constant benchmark for site.
Much of the daytime period will be determined
to have delay even if there are no readily
apparent operational issues; delay attributed to
work zone may appear higher than reality
Historic average speed Allows for separation of work zone
impacts from pre-construction
recurring congestion
Benchmark varies by time of day, data
availability could be problematic (although it is
provided by some companies)
34
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 7
Overall, the private sector data can certainly serve as a viable data source for delay 1
measures. Generally speaking, private sector data lends itself most directly to measuring delay 2
in terms of average delay per vehicle. Often accurate volume data from point detectors is not 3
available at work zones, limiting the ability of a DOT to determine total delay. Depending on 4
the vendor selected, each of the benchmarks listed in Table 2 may be available, so the DOT 5
would need to determine which measure would be most widely accepted by the agency. 6
Performance measures involving total delay or per person measures can be partially fed by the 7
private sector data stream, but would require fusion with other data streams to create the final 8
measure. 9
10
Using Private Sector Data to Measure Queue Length 11
12
While private sector data directly reports travel times and speeds, queue lengths are not 13
measured directly. When using private sector data to develop this measure, queued traffic is 14
typically identified based on when the reported travel speeds drop below a predefined threshold. 15
Thus, selection of this threshold will have a direct impact on the estimated length and duration of 16
queues. For example, one study in North Carolina defined a queue as being present from when 17
traffic is either stopped or slowed more than 25 mph below the posted speed limit until it has 18
reached average speed of 45 mph or more (11). If a DOT used different thresholds, then 19
obviously the estimated queue length and duration would change.20
Another concern when using private sector data to assess queue length relates to the way 21
in which speed/travel time data is reported by the vendor. Many vendors report travel times and 22
speeds using Traffic Message Channel (TMC) links. TMCs were defined by mapping companies 23
as a consistent way to report traveler information on digital mapping devices (12). TMCs have 24
been typically defined as homogeneous segments between major interchanges or intersections. 25
If a queue is defined as when a TMC falls below a certain speed threshold, this will cause the 26
entire TMC to be categorized as either queued or not queued. 27
The interaction of the TMC length and the speed threshold plays an important role when 28
using private sector data to estimate queue length. TMC lengths can vary considerably 29
depending on roadway functional class and setting. Urban TMCs are often very short, allowing 30
more precision in the estimation of queue lengths. In rural areas, however, TMCs can be much 31
longer, which can obscure the impact of the work zone in a local area. This can be particularly 32
problematic in work zones since project boundaries or impacts may not align well with the 33
TMCs. One prior study of 15 work zones found that, on average, an additional 1.8 miles of non-34
work zone roadway was included in the TMCs that contained the work zone (7). To illustrate the 35
impact of this spatial mismatch, consider an 8-mile long TMC that has a 2-mile work zone in the 36
middle of it. In this case, the private sector data may never detect any localized queuing at the 37
work zone since impacts would be “washed out” by the conditions on the other 6 miles of the 38
TMC. 39
40
CASE STUDY METHODOLOGY 41
42
Four case studies from Virginia are presented to illustrate the abilities, limitations, and key 43
tradeoffs that must be made when using private sector data to develop project-level delay and 44
queue length estimates. The Regional Integrated Transportation Information System (RITIS) 45
developed by the University of Maryland was used to acquire real-time private sector data for 46
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 8
four work zone sites in Virginia with a range of traffic and site characteristics. Information from 1
the Virginia Department of Transportation was used to identify study locations, define the time 2
periods when the work zone was active, as well as specific work zone activities that were 3
occurring. 4
While all four case studies occurred on the interstate system, each had diverse traits that 5
could pose a challenge when using private sector data. The four case studies evaluated were: 6
7
1. I-81 Northbound, Milepost (MP) 191-200: This site involved a lane closure in a rural, 8
mountainous area of the state. This site is used to illustrate whether performance 9
measures could be generated under lower volume conditions during overnight hours in 10
rural areas. 11
12
2. I-95 Southbound, MP 74-84: This site was located in an urban downtown area with 13
densely spaced interchanges, but construction occurred overnight, when private sector 14
data is often more limited. 15
16
3. I-95 Southbound, MP 158-161: This case study examined a one-day project occurring 17
during overnight hours in a suburban area, and served to illustrate whether short-term 18
work zone impacts could be captured accurately. 19
20
4. I-81 Southbound, MP 118-140, and US Route 460/11: This case study involved a full 21
freeway closure in a rural area, and a subsequent detour onto a parallel arterial route. It 22
serves to illustrate whether impacts due to the work zone could be captured on 23
surrounding facilities. 24
25
In all four cases, field observations noted that queuing and congestion were present during the 26
work activities. Case studies 1, 2, and 4 present a snapshot of data from a single day within a 27
long term work zone, while case study 3 shows an example of analysis from a short term work 28
zone. Performance measures for long term work zones could be aggregated using these daily 29
analyses, but are not illustrated in this paper. 30
31
Data Collection and Performance Measure Calculation 32
33
First, information on the spatial extent and duration of the work zones were collected for each 34
site from VDOT project logs. Specific information acquired included the location of the work 35
zone (route mileposts) and direction of travel, time when work zone was active, nature of the 36
work zone, and traffic volumes. Dates and times when lane closures or detours were present 37
were also noted. Since the focus of this paper is on project level performance measures, data for 38
one day in which congestion was present at each site is reviewed to illustrate some tradeoffs and 39
issues that must be confronted by DOTs when calculating performance measures using private 40
sector data. 41
Next, speed and travel time data were acquired based on the project information. The 42
two metrics selected for investigation were average delay per vehicle and queue length. Data 43
was downloaded using the RITIS Massive Raw Data Downloader based on each site’s location 44
and the time when work zone impacts were observed. For the purposes of the work zone impact 45
analysis, it is important to download data that extends farther than the work zone limits in order 46
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 9
to capture any impacts that may extend past the advance warning area. For example, if the work 1
zone limits were from MP 50-60, data was initially queried for at least from MP 45-65 to ensure 2
that the full extent of congestion and queuing was collected. Data were then aggregated using 3
both 1-hour and 5-minute intervals to compare and contrast the impacts of temporal aggregation 4
on the resulting performance measures. These time intervals were selected to illustrate a range 5
of values of aggregation, one very disaggregate and one very aggregate. Intermediate levels of 6
time aggregation would produce results that vary between these values. 7
The procedure for performance measure calculation was as follows: 8
9
1. First, the maximum spatial extent of the queuing and congestion was determined. 10
Historic speeds at the site when no work zone was present were provided by the private 11
sector provider as a reference point to compare against real-time data while the work 12
zone was active. For each TMC and each time interval, the historic average speed was 13
compared to the real-time speed when the work zone was present. If real-time speeds 14
were less than 90 percent of what was observed historically on that TMC at that time of 15
day, it was assumed that the TMC was impacted by the work zone. The 90 percent 16
threshold was set to try to eliminate periods where minor fluctuations in speed were due 17
to random variation unrelated to the work zone effects, but still try to capture the work 18
zone impacts as fully as possible. To be conservative, if a TMC is determined to be 19
affected by the work zone at any given interval, that TMC was retained throughout the 20
analysis. Any TMCs that never dropped below this threshold were assumed to not be 21
impacted by the work zone and were removed from further analysis. 22
23
2. Next, the TMCs were examined to determine whether traffic was queued. RITIS uses a 24
threshold of 60 percent of the historic average speed to determine bottleneck locations 25
(13), and this was adopted as the threshold for determining queued traffic. If a TMC 26
speed drops below 60 percent of the historic average speed, it was marked as queued. 27
The sum of the lengths of the TMCs falling below this threshold during each time 28
interval was summed to determine the queue length. The duration when any link was 29
marked as queued was used to determine the overall queue duration at the work zone. 30
The use of the historic data threshold allows for the separation of work zone impacts 31
from pre-existing congestion at the site. 32
33
3. To determine the delay caused by the work zone, the sum of the historic travel times 34
across contiguous impacted TMCs was subtracted from the observed travel time during 35
work zone operations on those same TMCs. This provides a measure of the average 36
delay per vehicle. 37
38
The performance measures were then compared across sites to help illustrate variations in the 39
quality of performance measures that could be generated from private sector data. 40
41
42
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 10
CASE STUDY RESULTS 1
2
Case Study 1: I-81 Northbound, MP 191 – 200 3
4
The first case study occurred on I-81 Northbound from MP 195 to 197 in Rockbridge County, 5
VA. The right lane of a two lane directional segment was closed as part of ongoing work on a 6
truck climbing lane project. The section was located in a rural area and had a grade of 7
approximately +2.9 percent. The 2012 directional annual average daily traffic (AADT) was 8
approximately 20,000 veh/day. This work zone project began in February 2009 and had an 9
estimated completion date in late 2013, but this case study focuses on data from 6/19/12. 10
Although the work zone was only present from MP 195 to 197, the private sector data revealed 11
traffic impacts that extended from MP 191 to 200. A key consideration at this site is that heavy 12
vehicles composed approximately 50 percent of the traffic stream during these hours. Since the 13
work zone was on an uphill grade, the high truck volumes caused traffic capacity to be 14
substantially lower than would be observed on flat, level terrain. Furthermore, the trucks often 15
drove side-by-side until they reached the taper to deter queue jumping, which further reduced 16
capacity and increased queuing. 17
Figure 2a shows the average delay experienced between MP 191 and 200 between 7:00 18
PM and 11:30 PM on 6/19/12 using both 1-hour and 5-minute aggregation intervals. Figure 2a 19
shows that there is a marked difference in the performance measures calculated depending on the 20
aggregation interval used. Using a 1-hour interval serves to dampen variation in the data at this 21
site. The maximum delay using the 1-hour interval was 20.56 minutes, while it was 31.52 22
minutes using the 5-minute summary interval. This difference could create significant impacts, 23
depending on how the performance measures are being used at the project level. If they are 24
being used to assess penalties to contractors or determine compliance with work zone 25
performance measures, an hourly aggregation interval would be less likely to detect sub-hourly 26
intervals where the contractor exceeds allowable thresholds. Thus, while an hourly aggregation 27
interval may reduce analytical workload for the development of monthly or annual programmatic 28
work zone performance measures, they may be very conservative if they are being used to assess 29
real-time contractor compliance with operational targets. 30
Figure 2b shows the queue length that was estimated at this site using the private sector 31
data. The queuing diagram exhibits far less variation than the delay figure due to the influence 32
of the TMC size. In this case, the 9.3 mile analysis length was composed of only 4 TMCs which 33
ranged in length from 0.58 and 5.04 miles. Since each of these four TMCs were either identified 34
as queued or not queued based on the speed threshold noted earlier, this produced a queuing 35
figure that resembles a step function with very sudden changes. In this case, long TMCs (like 36
the 5.04 mile section) create impediments to using private sector data for queuing performance 37
measures in rural areas. These issues are not present in the delay calculations at this site, 38
however. This indicates that queue measures should be viewed with caution in rural areas due to 39
the influence of average TMC size on the queue length estimates. 40
41
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 11
1
2
3
(a) 4
5
6
7
(b) 8
9
FIGURE 2 (a) Average delay and (b) Queue length for case study 1 on 6/19/12 between 7 PM and 11:30 PM. 10
11
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 12
1
Case Study 2: I-95 Southbound, MP 74 - 84 2
3
The second case study involved closure of 2 out of 3 southbound lanes on I-95 in downtown 4
Richmond, VA on Sunday through Thursday evenings (9/30/2012 – 10/5/2012) from 8:00 PM to 5
6:00 AM. I-95 was reduced to one travel lane in each direction between the Lombardy Street 6
Bridge (MP 77) and Laburnum Avenue (MP 79). Although drivers were advised to follow 7
posted detour routes, significant congestion was still observed on I-95. The 2012 AADT of this 8
location was approximately 65,000 veh/day. Based on the private sector data, it was found that 9
the work zone impacts extended from MP 74 to 84. 10
Figures 3a and 3b show the results of the delay and queue performance measure 11
calculations at this site on 9/30/12 between 8 PM and Midnight. Figure 3a shows the delay 12
performance measure calculations for the site for the 1-hour and 5-minute aggregation intervals. 13
In this case, there are smaller differences between the two aggregation intervals than in case 14
study 1, but the hourly aggregation interval still serves to dampen the impact of the work zone 15
peaks. It also fails to capture the onset and dissipation of congestion at the site as accurately 16
since the start/end of congestion both happened approximately midway through the hour. 17
Figure 3b shows the queuing profile for the site, and serves as a contrast to case study 1. 18
The total analysis length of this site was similar to case study 1 (9.3 miles for case study 1 vs. 9.6 19
miles for case study 2), but many more TMCs were present for case study 2. A total of 4 TMCs 20
were available for case study 1 vs. 17 TMCs for case study 2. The 17 TMCs for case study 2 21
ranged in length from 0.08 to 1.89 miles. The shorter mean TMC length for case study 2 is 22
representative of what is often seen in urban areas where TMCs have been created based on 23
complex, closely spaced interchanges. This granularity in turns permits much more accurate 24
estimates of queue length to be developed in comparison to more rural cases with longer TMCs. 25
The 5-minute aggregation interval still reflects changing queues more rapidly, as well as the sub- 26
hourly variation, but differences between 1-hour and 5-minute results are generally not as large 27
as in case study 1. 28
29
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 13
1
2
3
(a) 4
5
6
7
(b) 8
9
FIGURE 3 (a) Average delay and (b) Queue length for case study 2 on 9/30/12 between 8 PM and 12:00 AM. 10
11
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 14
1
Case Study 3: I-95 Southbound, MP 158-161 2
3
The third case study examined a short-term work zone on I-95 SB in the suburbs of Washington, 4
D.C. At 9:00 PM on 2/17/2012, VDOT removed a 30 foot tall cantilevered sign structure located 5
on southbound I-95 at the interchange with the Prince William Parkway in Woodbridge. The 6
2012 directional AADT of this section of road was approximately 80,000 veh/day. Two of three 7
lanes were closed, but motorists could avoid delays by using the parallel high occupancy vehicle 8
(HOV) lanes. The private sector data showed that the area impacted by the work zone extended 9
for approximately 3.5 miles from MP 158 to 161. 10
This case study served to examine whether performance measures could be determined 11
for a one-time, short term work zone that occurred during overnight hours. Real-time data was 12
available for this site, and measured impacts were corroborated by field observations. Figure 4a 13
shows the delay measurements at the site. Generally speaking, this site showed greater 14
consistency between the 5-minute and 1-hour aggregation intervals than other case studies. In 15
general, the 5-minute results were within ± 5 minute of the 1-hour averages. Once again, the 5-16
minuteintervals showed more variability in results. 17
Figure 4b shows the queues experienced at the site. Since this was an urban area, once 18
again TMC sizes were relatively short. The 3.56 mile analysis length was composed of 5 TMCs 19
with lengths between 0.36 and 1.42 miles. Similar to the delay calculations, the results were 20
relatively similar between the 5-minute and 1-hour aggregation intervals. The small average 21
TMC length allowed for reasonably detailed queue length estimates to be generated as well. 22
Estimated queue length durations were longer for the 1-hour aggregation interval, however, since 23
the onset and dissipation of congestion were not captured to as great of a temporal resolution. 24
25
26
27
28
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 15
1
2
3
(a) 4
5
6
7
(b) 8
9
FIGURE 4 (a) Average delay and (b) Queue length for case study 3 on 2/17/12 between 9:00 PM and 3:00 10
AM. 11
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 16
1
2
Case Study 4: I-81 Southbound, MP 118 – 140, and US 11/460 3
4
The final case study involved construction of a new truck climbing lane on I-81 SB between 5
Salem, VA and Christiansburg, VA. This project began in 2010 and has an estimated completion 6
in late 2013. The work was located on an existing two lane directional segment with a +3.7 7
percent grade sustained for 2.2 miles and a 2012 directional AADT of 25,000 veh/day. On July 8
18, 2012, a full interstate closure was conducted due to blasting operations on I-81. Traffic was 9
detoured onto US 11/460 at I-81 exit 132 and back onto I-81 at exit 118. The detour route had a 10
2012 AADT (without detour traffic) of approximately 18,000 for both directions of travel 11
combined. The detour route was a four lane divided arterial. 12
On the day studied, the detour was implemented between 10:30 AM and 1:00 PM. A 13
right lane closure was also in place on westbound Route 11/460 during the detour. The private 14
sector data showed work zone impacts extended for approximately 8 miles from MP 132 to 140 15
on I-81, as well as throughout the length of the detour route. This case study serves to illustrate 16
how the private sector data can be used to assess system-wide impacts of work zones on parallel 17
facilities, as well as issues related to the use of the data on arterial roads. 18
Figures 5a and 5b show the delay and queuing impacts that occurred on I-81 as a result of 19
the freeway closure. The section of road monitored was 16.8 miles long, and consisted of 6 20
TMCs. The TMCs ranged in length from 0.17 to 8.64 miles. Figure 5a shows that using a 1-21
hour aggregation interval masks many sub-hourly variations, sometimes causing an under-22
reporting of delay by almost 15 minutes/veh. The onset and dissipation of congestion is also not 23
captured adequately. Figure 5b shows the estimated queue that was determined. The maximum 24
queue is significantly lower using the 1-hour interval, and the duration of queuing is also 25
underestimated by approximately 1 hour. In this case, the TMC located closest to the diversion 26
point was 4.42 miles long. As a result, the initial queue quickly changed from 0to 4.42 miles. 27
The relatively coarse spatial granularity on the section of rural interstate makes the queuing 28
estimates less reliable than what was seen in case studies 2 and 3. 29
Figures 6a and 6b show conditions along the arterial detour route during the freeway 30
closure. The detour route was 10.73 miles long, and composed of just 2 TMCs which were 4.35 31
and 6.38 miles long. Figure 5a shows the delay that was experienced on the detour route. The 32
figure shows that the hourly aggregation significantly deviates from the 5-minute aggregation 33
during the onset and dissipation of congestion. When the data in Figure 6a is combined with the 34
data from Figure 5a, it should be possible to determine a combined user delay impact for the 35
entire work zone. Figure 6b shows the estimated queue from the private sector data. Given that 36
only 2 TMCs were available over the entire 10.73 mile detour route, the estimated queues are 37
very coarse and cannot be assumed to reliably show the spatial extents of queuing. There is also 38
a significant difference in the estimated queue duration of about 1 hour between the 1-hour and 39
5-minute aggregation intervals. Again, use of the hourly data causes the averages to be much 40
lower than when shorter durations are used. 41
42
43
44
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 17
1
2
(a) 3
4
(b) 5
6
FIGURE 5. (a) Average delay and (b) Queue length for I-81 case study 4 on 7/18/12 between 10 AM and 2 7
PM. 8
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 18
1
2
3
(a) 4
5
6
7
(b) 8
9
FIGURE 6. (a) Average delay and (b) Queue length for US 11/460 case study 4 on 7/18/12 between 10 AM 10
and 2 PM. 11
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 19
While this case study shows that it is possible to generate system-wide estimates of work 1
zone mobility impacts, it also shows some of the difficulties than can arise as a result of long 2
aggregation intervals and large TMC lengths. The hourly data significantly underestimated the 3
duration of queuing on both the mainline and arterial route, and the large TMC lengths on these 4
rural sections also made estimates of queue length inherently imprecise. These queue length 5
measures would not appear to be suitable for project level performance measurement, 6
particularly if contractor performance would be subject to penalties or remedial actions. 7
8
LESSONS LEARNED 9
10
This research attempted to define specific lessons learned that would impact how a DOT sets up 11
a work zone mobility performance measurement program using private sector travel time and 12
speed data. Major findings from the cases studies are discussed below. 13
14
Applicability to Different Performance Measures 15
16
Private sector travel time data were used to determine two commonly used work zone 17
performance measures: delay and queue length. Delay could generally be calculated easily from 18
the data sets that were used. Historic average speeds were available for pre-construction 19
conditions which allowed calculation relative to that baseline measure. Alternatively, posted 20
speed limits or other thresholds could be easily used. 21
Queue length was a more problematic measure for the private sector data. First, a speed 22
threshold had to be set below which traffic on the link was determined to be queued. Thus, the 23
amount of queuing determined will be sensitive to the threshold value selected by the DOT. A 24
more significant issue is the influence of TMC granularity on the determination of queue length 25
and queue duration. The case studies in this paper serve to illustrate the large variability in TMC 26
size that can be present in rural vs. urban facilities. The rural facilities of case studies 1 and 4 27
had mean TMC lengths of between 2.3 and 5.4 miles, while the urban/suburban work zones of 28
case studies 2 and 3 had average TMC lengths of 0.56 and 0.71 miles. Thus, using private sector 29
data to define queue lengths should be viewed cautiously, especially in rural areas. This concern 30
is heighted for project-level performance measures where there could be project cost 31
implications if queues are not measured correctly. 32
33
Spatial Aggregation Impacts 34
35
The spatial granularity at which the data is reported should be closely considered when defining 36
a work zone performance measurement program. TMCs are typically defined based on the 37
locations of intersections or interchanges. Since work zones may occur at the midpoint of TMCs, 38
long TMCs could mask the impact of the work zone by including a significant amount of non-39
work zone travel data. Given the differences in rural and urban TMC lengths, it currently 40
appears that urban sites could be most effectively monitored using the private sector data. 41
Analysts should closely examine the TMC sizes before proceeding with performance measure 42
calculations, especially in rural locations. If the spatial aggregation intervals are too large, the 43
agency may need to install temporary point detectors to adequately monitor conditions at a work 44
zone. 45
46
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 20
Temporal Aggregation Impacts 1
2
Another issue that could impact performance measure outputs is the level of temporal 3
aggregation. This is especially important for project-level assessments that could be used to ask 4
contractors to make changes to the traffic control plans or to assess performance penalties. As 5
the time over which performance measures are created increases, the severity of congestion tends 6
to be masked. For example, if the DOT is interested in the amount of time delay exceeds 15 7
minutes at a work zone, the a 1-hour aggregation interval will result in fewer violations than if 8
the performance measures are calculated every 5 minutes. Thus, the DOT should carefully 9
consider and specify the time over which any performance measures are aggregated 10
11
FUTURE DIRECTIONS FOR PRIVATE SECTOR DATA 12
13
It should be emphasized that the findings of this research represent current conditions as of the 14
writing of this paper. Private sector companies are continually refining their methodologies and 15
attempting to improve their sample sizes. Future developments may allow for dynamic 16
segmentation of TMCs or other improvements that could remove some of the current limitations 17
of using private sector data in rural areas. 18
Likewise, DOTs will need to have a computerized system that can perform project 19
monitoring automatically if private sector data is to be mainstreamed into work zone 20
performance measurement. In this paper, the RITIS software developed by the University of 21
Maryland was used. Several vendors are making software modifications that can fuse together 22
DOT work zone records with probe vehicle records to generate performance measure figures and 23
charts in an automated manner. Provision of these tools will help enhance and expand the use of 24
this data stream for mobility measurement in work zones. 25
26
REFERENCES 27
1. Ullman, G.L., T.J. Lomax, and T. Scriba. A Primer on Work Zone Safety and Mobility
Performance Measurement. Report FHWA-HOP-11-033. FHWA, U.S. Department of
Transportation, 2011.
2. Schrank, D., Eisele, B., and T. Lomax. TTI’s 2012 Urban Mobility Report Powered by
INRIX Traffic Data. Texas A&M Transportation Institute, College Station, TX, 2012.
3. Maryland rolls out travel time message signs. Baltimore Sun, March 9, 2010.
http://weblogs.baltimoresun.com/news/traffic/2010/03/maryland_rollls_out_travel_tim.ht
ml. Accessed July 19, 2013.
4. Virginia Department of Transportation. I-95 Richmond Bridge restorations Electronic
Message Signs Travel Time FAQs.
http://www.virginiadot.org/VDOT/Newsroom/Richmond/2013/asset_upload_file186_653
68.pdf. Accessed July 19, 2013.
5. Haghani, A., M. Hamedi, and K.F. Sadabadi. I-95 Corridor Coalition Vehicle Probe
Project: Validation of INRIX Data July-September 2008. University of Maryland,
College Park, MD., 2009.
TRB 2014 Annual Meeting Paper revised from original submittal.
Fontaine, Chun, and Cottrell 21
6. Fontaine, M.D. Evaluating Travel Time Data Quality form a Private Sector Data
Provider: A Case Study of I-66 in Northern Virginia. North American Traffic
Monitoring Exhibition and Conference, Dallas, TX, June 6, 2012.
7. Edwards, M.B. and M.D. Fontaine. Investigation of Work Zone Travel Time Reliability
Using Private Sector Data. In Transportation Research Record: Journal of the
Transportation Research Board, No. 2272, Transportation Research Board of the
National Academies, Washington, D.C., 2012, pp. 9-18.
8. Stargell, R. Ohio DOT Performance Measures. Work Zone Performance Measurement
Peer Exchange Workshop, Atlanta, GA, May 8, 2012.
9. Fontaine, M.D. Executive Summary: Evaluation of INRIX Travel Time Data in Virginia.
Virginia Center for Transportation Innovation and Research, Virginia Department of
Transportation, Charlottesville, VA 2013.
10. Bourne, J.S., T.A. Scriba, C. Eng, R.D. Lipps, G.L. Ullman, D.L. Markow, D. Gomez,
K.C. Matthews, D.L. Holstein, B. Zimmerman, and R. Stargell. Best Practices in Work
Zone Assessment, Data Collection, and Performance Evaluation. Domestic Scan Report
08-04, National Cooperative Highway Research Program, Washington, D.C., 2010.
11. Chandler, B., Beasley, K., and Rephlo, J. National Evaluation of the SafeTrip-21
Initiative: I-95 Corridor Coalition Test Bed, Final Evaluation Report: North Carolina
Deployment of Portable Traffic Monitoring Devices. Report FHWA-JPO-10-058,
FHWA, U.S. Department of Transportation, 2010.
12. NAVTEQ. NAVTEQ and Tele Atlas Collaborate to Develop Standard Traffic Codes for
Digital Maps of the United States. March 31, 2004.
http://press.navteq.com/index.php?s=4260&item=5690. Accessed July 19, 2013.
13. University of Maryland Center for Advanced Transportation Technology Laboratory.
Vehicle Probe Project Suite Frequent Asked Questions.
http://vpp.ritis.org/suite/faq/#/how-are-bottleneck-conditions-tracked. Accessed July 19,
2013.
TRB 2014 Annual Meeting Paper revised from original submittal.