Pilot Testing of SHRP 2 Reliability Data and Analytical Products: Washington (2014)

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.

62 CHAPTER 6 Pilot Testing and Analysis on SHRP 2 L07 Product 6.1 Tool Introduction and Interface The objective of SHRP 2 L07 is to evaluate the cost-effectiveness of geometric design treatments in reducing nonrecurrent congestion. The L07 products help estimate traffic operational effectiveness and measure economic benefits of various design treatments. In addition to the research report, L07 produced an Excel-based software tool to automate the analysis process. A snapshot of the tool interface is shown in Figure 6.1. Figure 6.1. SHRP 2 L07 product interface. The design treatments considered in the L07 product can be put into four categories as follows:  Shoulder-related treatments o Accessible shoulder (for removal of vehicles) o Alternating shoulder (for work zones) o Drivable shoulder (for diversion of vehicles)  Crash-related treatments o Crash investigation site (urban area)

63 o Emergency pull-off (rural area) o Extra high median barrier (eliminate rubbernecking) o Incident screen (at the roadside)  Emergency treatments o Emergency access (for emergency vehicles) o Emergency crossovers (keep open to all vehicles) o Control (gated) turnarounds (used in emergency for all vehicles)  Treatments for special sites o Runaway truck ramp (used in steep downgrade roads) o Wildlife crash reduction o Anti-icing systems o Snow fence o Blowing sand 6.2 Tool Operability The research team has installed the L07 tool on different operating systems (e.g., 32-bit and 64- bit Windows 7, 64-bit Windows 8, and the OS X 10.6.8 operation system for Mac computers) with different versions of Microsoft Office (e.g., Microsoft Office 2010 and Office 2011 for Mac). The tool can be installed and run successfully for most operating systems. Except for the 64-bit Windows 8 and the OS X 10.6.8, the installation was unsuccessful and a warning textbook popped up as shown in Figure 6.2 and Figure 6.3. In addition, the L07 tool occasionally failed to operate when it was installed on 32-bit and 64-bit Windows 7. The warning message is shown in Figure 6.4. Researchers found that the run- time error ‘1004’ problem can be solved in Excel 2010 by manually selecting “Trust access to the VBA object model” and then choosing “Enable all macros” in the Excel’s trust center. Figure 6.2. Warning dialog for the 64-bit Windows 8 operating system.

64 Figure 6.3. Warning dialog for the OS X 10.6.8 operating system. Figure 6.4. Warning dialog for Windows 7 operating system. 6.3 Tool Usability 6.3.1 User Friendliness In general, the L07 guide can provide meaningful and useful introductions for using the tool, and the tool is found to be easy to understand and use. The interface is user friendly, and most of the icons are shown assisted with useful guides. While using the tool, however, the research team found the following limitations:  The tool interface cannot be moved, minimized, or resized;  If multiple treatments are chosen, only the first treatment can be saved;  Users cannot output results to a separate file; and  Users cannot enlarge the figures or output the source data. These limitations certainly affect the usability of the tool, particularly when an analysis involves lots of data input and similar data can be reused for multiple analyses. 6.3.2 Tool Accuracy The default values of truck ratio and recreation vehicle (RV) ratio are not consistent with the HCM 2010. In the tool, the default values of truck ratio and RV radio are set as 2.0% and 1.0%, respectively; the HCM 2010 recommended values are shown in Figure 6.5 for highways and in

65 Figure 6.6 for freeways. Figure 6.5. HCM 2010 suggested default values for heavy vehicles percentage for highways. Figure 6.6. HCM 2010 suggested default values for trucks and RVs percentage for freeways. The description of treatment “Movable Cable Median Barrier” is confusing. The barrier (see Figure 6.7) is defined as “a special designed wire cable barrier system, which can be removed to allow median crossovers.” A “T” threshold was introduced to indicate the time when the barrier would be moved to allow median crossover. The barrier would not be moved unless

66 the incident duration reaches T. The default values of T can be found in Figure 6.8. Nevertheless, while looking at the default values, the research team found that the T threshold for PDO is smaller than that for major injury or fatality. This confused the research team as most major injury or fatal incidents would last longer than PDO incidents and thus are associated with longer delays. Allowing median crossover sooner in major injury or fatality incident scenarios is certainly beneficial in the research team’s opinion. So, the T threshold for major injury and fatality should be smaller than or equal to that for PDO. Figure 6.7. Example of movable cable median barrier. Figure 6.8. Suggested thresholds for movable cable median barrier treatment. According to the L07 guide, several coefficients for safety effect estimating are provided as in Figure 6.9. But, there is not enough evidence supporting these coefficient values. The

67 L07 team should help provide more details about how they get these values and the reasons for setting up such coefficients so that users can decide whether they need to update these factors regarding different roadway geometries, locations, weather characteristics, culture, and more. Figure 6.9. Suggested default coefficients in L07 guide.

68 6.4 Performance Test Testing of L07 tool performance is conducted in three folds: (1) a comparison study is made with the DRIVE Net system to test the MOE sub-output, (2) a comparison study is made with on-site single-loop detector data to test the tool’s production of the TTI curve, and (3) a case study is conducted to test the benefit–cost sub-output. 6.4.1 Output Comparison with DRIVE Net The key feature of the L07 tool is to estimate the TTI curves both before and after the design treatment. As the DRIVE Net system can also calculate the same MOE for WSDOT’s productions of the Gray Notebooks, the research team compared TTI curves produced by the DRIVE Net and the L07 tool. Gray Notebook capacity analysis includes a travel time analysis method using both loop and INRIX data. The procedure for calculating travel time distribution is quite similar to the methodology recommended by L07. Vehicle average travel time is calculated and updated for each 5-minute period. Then the travel time cumulative distribution is summarized for each time slot in all weekdays throughout the year. The results are more accurate than the travel time estimation results based solely on the output from loop detector, since travel time is calculated using real-time vehicle speed collected from GPS devices when possible. Gray Notebook has been published for many years. The travel time estimates for the selected corridors have been verified through different means in WSDOT. So the Gray Notebook travel time data is a great benchmark data set to compare with calculation results from the SHRP 2 Reliability products. A Gray Notebook data source facility within the L07 test sites is an I-5 segment from milepost 184 to milepost 185.5. DRIVE Net computes two sets of TTI for morning peak (8:20 a.m.) and afternoon peak (5:30 p.m.), respectively. Figure 6.10 shows the outputs of DRIVE Net (a) and the L07 tool (b). Figure 6.10 shows that it is difficult to tell whether the L07 tool gives an accurate estimation of the TTI curve, because the L07 tool does not allow users to resize/enlarge the output graphs nor output the source data. However, when looking at the 50th percentile TTI values for the afternoon period, the research team finds that DRIVE Net reports larger TTI values than those from the L07 tool. Since the selected I-5 facility is very congested during evening peak, and DRIVE Net system is based on daily data over an entire year (workdays), the research team believes that the DRIVE Net output is closer to the ground truth.

69 (a) Output of DRIVE (b) Output of L07 Tool Figure 6.10. Output of DRIVE Net system (a) and L07 tool (b). 6.4.2 Comparison with On-Site Single-Loop Detector Data The TTI curve computed from the on-site single-loop detector measurements is compared with the TTI curve from the L07 tool. Vehicle travel time is calculated using the procedure recommended by the SHRP L02 Guide. Start and end points for single-loop detector data calculation are defined by users thus the method can be easily applied to specific freeway segments. The study site is located on I-5 from milepost 158 to milepost 160. Figure 6.11 shows the traffic volume data detected by single-loop detectors. Each hourly volume data is the 30th highest traffic volume of the year 2012, which is a required input for the L07 tool.

70 Figure 6.12 shows the TTI curves calculated from single-loop data and the L07 tool. Three different hours (3:00 a.m., 8:00 a.m., and 5:00 p.m.) represent low traffic demand, morning and afternoon peaks respectively. For the low traffic demand, the two graphs are similar as they both report a small TTI value. The sudden change in the left graph is because of rounding errors in travel speed calculation. For higher traffic demand periods (8:00 a.m. and 5:00 p.m.), L07 predicts a much smaller TTI value. Again, since the selected I-5 facility is very congested during peak hours, and the single-loop detector result is based on daily data over an entire year (workdays), the research team believes that the output from single-loop detector is closer to the ground truth values. Both DRIVE Net and single-loop detector data suggest that the L07 tool tends to underestimate the travel time during peak hours. Figure 6.11. Traffic volume for the studied site.

71 Figure 6.12. Comparison of outputs from single-loop detector data and L07 tool.

72 6.4.3 Case Study To test the effectiveness of the L07 tool, the research team prefers finding a completed project with the same scope within Washington State. However, as the tool involves only 16 specific design treatments as mentioned in Section 6.1, an effective comparison requires a rigorous selection among previous construction projects. Also, the treatment should start and be completed after January 1, 2009, since data before 2009 were not archived. When looking at all of the 475 projects completed from 2009 to 2013 in Washington State (http://www.wsdot.wa.gov/Projects/completed.htm), only two wildlife projects in rural areas are found to be with the same scope as those listed in the L07 tool. Unfortunately, there is no archived traffic data in the locations of these projects. The research team studied the methodology in L07 and found that the output for L07 benefit–cost analysis was basically determined by the difference of TTI curves and the number of traffic incident reductions. The TTI curves are determined by traffic volume and nonrecurrent events. Thus, the I-5–Marysville to Stillaguamish River–Median Barrier project was selected as the case study project. This project started in June 2009 and was completed in November 2010. Figure 6.13 describes the testing procedure. Choose an Applicable Project Get the Site Data from STAR Lab Database Input the Data Before Treatment into L07 Tool Select the Corresponding Treatment and Modify its Effects Generate Tool-Estimated Treated TTI Curve Input the Data After Treatment into L07 Tool Generate Untreated TTI Curve (Real Treated TTI Curve) Compare the Two Outputs Figure 6.13. Testing procedure for L07. 6.4.3.1 Case Study Project The construction project used for this case study is located on I-5 between Marysville and Stillaguamish River, from milepost 199 to milepost 209. There are three northbound lanes in this location. The segment between milepost 206 and milepost 207 was chosen as the test segment. Figure 6.14 illustrates the case study site location on Google Maps. 6.4.3.2 Test Scenario The project installed a concrete median barrier along a 10-mile stretch of northbound I-5 in the Marysville area and removed the existing low-tension cable median barrier at the same time.

73 Existing southbound cable barrier was left in place to provide redundant protection. The project also widened the median shoulders to 10 feet, bringing them to current standards. Total cost of the construction work was $16.4 million, with $2.5 million of additional funding from the 2009 American Recovery and Reinvestment Act; traffic cameras, electronic message signs, and traffic sensors also were installed along I-5 in Marysville. Map data © 2014 Google Figure 6.14. Test site location and detail for L07. 6.4.3.3 Timeline The 2008 supplemental legislative budget included $26.9 million to install concrete barrier along the 10 miles of northbound I-5 in Marysville. The project was advertised for competitive bidding in April 2009 and awarded to Tri-State Construction in June 2009. Construction began in July 2009 and was completed in November 2010. 6.4.3.4 Traffic Demand Data Loop data from milepost 206 to milepost 207 were used for the testing. Traffic volume data before construction were obtained from January 2009 to June 2009. Data after construction were obtained from January 1, 2011, to December 31, 2011. Following the L07 guide, hourly demand was selected as the 30th highest volume in the year. The hourly traffic demand for the test site is shown in Figure 6.15. When comparing the curves before and after construction, we found that the curves are very similar, only the peak

74 hour demand slightly increased after the construction. Thus, the treatment did not result in a significant increase in traffic demand. Figure 6.15. Hourly traffic demand before and after the treatment. 6.4.3.5 Geometry Data The geometry data inputted into the L07 tool cannot be saved. These data are used to compute free-flow speed for the segment. 6.4.3.6 Traffic Incident Data Incidents for the segment can be found from the WITS database. Numbers of different types of incidents before and after treatment are listed in Table 6.1. Incident numbers before the treatment are estimated as the average number for 2006 through 2008; incident numbers after the treatments are estimated as the average number for 2011 and 2012. 0 500 1000 1500 2000 2500 3000 3500 4000 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 D em an d ( ve h /h ) Timeline (hour) Before Treatment After Treatment

75 Table 6.1. Incident Numbers for I-5 Mileposts 199–209 Before After Decrease Year 2006 2007 2008 Mean 2011 2012 Mean % Property damaged only 17 23 30 23.3 2 8 5 78.6 Minor injury 6 3 7 5.3 3 2 2.5 53.1 Fatality 1 1 2 1.3 0 0 0 100 Non-crash 575 625 627 609 136 130 133 78.2 Total 599 652 666 639 141 140 140.5 78.0 Summarizing Table 6.1, the conclusion can be drawn that the treatment had a significantly positive effect on reducing traffic incidents, especially severe incidents. Looking at the data in Table 6.1, all kinds of crash incidents were reduced by 50% or more after the concrete median barrier was built. For the tool testing purpose, actual incident numbers for the test site are applied to replace default values. For crash costs, the default values suggested by the L07 guide are used. 6.4.3.7 Weather, Event, and Work Zone Data For weather data, defaults provided by the tool are used. The nearest location to provide the weather data is selected as Seattle. No event or work zone happened on the segment during the testing period. 6.4.3.8 Treatment Selection In choosing the proper treatment, the research team tried “Extra High Med Barrier” treatment within the tool first, because the definition of it seems to be the closest to the actual treatment. However, the “Extra High Med Barrier” treatment in the tool only targets gawk-inducing incidents, which contribute only a small proportion to all the incidents. At the same time, if the input value for incident reduction is close to 100%, the software crashes (see Figure 6.16). To make the testing more precise, the research team chose another treatment called “Anti-icing Systems” for the testing. Although the definition of treatment does not come close to the actual median barrier project, the objective of the projects is the same, which is to avoid/reduce traffic incidents. Therefore, the Anti-icing System is selected for the testing.

76 Figure 6.16. L07 tool crash when crash reduction percentage is input at or near 100% (as in box for Minor Injury crashes at left). 6.4.3.9 Tool Outputs 6.4.3.9.1 BENEFIT–COST Figure 6.17 shows the tool output for the benefit–cost analysis. The “Net Present Value of Cost” is set as $16.4 million. The “Net Present Value of Benefits” is about $13 million, and the “Net Present Benefit is –$3.4 million.” Thus, the tool failed to provide positive benefit for this project. 6.4.3.9.2 TRAVEL TIME INDEX The tool generates untreated and treated TTI curves for before-and-after analysis [see Figure 6.18(a)]. To test the software accuracy, the research team inputted the after-treatment demand data as the before-treatment demand and let the tool generate the TTI curve [see Figure 6.18(b)]. Both of the graphs are drawn based on the peak hour data at 4:00 p.m. Theoretically, the treated TTI curve in Figure 6.18 (a) should be the same as the untreated TTI curve in Figure 6.18 (b). However, while comparing the blue curve on the right with the red curve on the left, it is obvious that the 100th percentile TTI values (see the red circles) are different. One is close to 1.4 and the other is close to 1.2. More details cannot be seen from the tool since these output curves cannot be enlarged nor outputted.

77 Figure 6.17. L07 tool output for benefit–cost analysis. (a) (b) Figure 6.18. L07 tool output for TTI analysis: (a) Uses before-treatment demand data as input; (b) Uses after-treatment demand data as input.

78 6.5 Test Conclusions The research team believes that the L07 methodology on computing TTI curves should be further studied and compared. Neither the output comparison between L07 and DRIVE Net nor the software accuracy comparison between L07 before-treatment curve [see Figure 6.18 (b), red curve] and L07 after-treatment curve [see Figure 6.18 (a), blue curve] yields a positive conclusion. At the same time, the research team suggests that the L07 project team help revise the tool and allow the user to obtain more detailed output information from it. In the L07 tool, the treatment “Extra High Med Barrier” only deals with gawk-inducing incidents. However, such treatment in reality can also help prevent other types of incidents. For example, some high concrete median barriers can also prevent vehicles from running into the opposite direction, so that some severe accidents can be prevented. Therefore, more potential effects of the proposed design treatments in L07 are recommended for consideration. In the case study, the test project did not provide meaningful results in the cost–benefit analysis. It may be because of an underestimation of the project effect on preventing major injury and fatal incidents. According to the default values set in the L07 tool, crash cost for fatal and major injury incidents are much more than minor-injury incidents (crash cost for fatal and major injury incident is about 40 times of that for minor incident) and PDO incident (crash cost for fatal and major injury incident is about 200 times of that for PDO incident), reducing the number of fatal and major injury incidents is critical for safety-related benefit. Thus, the change in the number of fatal and major injury incidents is tested. The result can be found in Table 6.2, where the average incident reduction effect is set as 70% (according to Table 6.1). It can be concluded that the net present benefit is sensitive to the number of fatal and major injury incidents. This is consistent with the fact that fatal and major injuries contribute the most to total cost. For most fatal injuries, the cost mostly depends on the number of deaths during the crash; however, the L07 tool suggests using uniform cost values for incidents with the same severity level. Thus, the research team recommends that the L07 tool should allow users to modify the cost of incidents and provide a modification factor for user to input location-specific cost values for different severity levels of incidents. Table 6.2. Effect of Fatal and Major Injury Incident Number on Treatment Benefit Number of Fatal and Major Injury Incidents Per Year 0 1 2 3 Net Present Benefit ($ million) –13.6 –3.4 7.6 18.3