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.

132 CHAPTER 9 Conclusions and Potential Improvements 9.1 Summary and Conclusions In sum, the research team has tested and evaluated the analytical products from the SHRP 2 projects. The major conclusions for each product are summarized as follows: The L02 methodology builds a strong foundation for travel time reliability monitoring. In this project, travel time calculations and congestion data were acquired from single-loop detectors at 5-minute intervals. Nonrecurring condition data for incidents and weather were taken from the WITS and local weather stations. Plotting these data with cumulative distribution functions provided a clear diagnosis for each route by analyzing performance under congestion and nonrecurring conditions and provides a strong framework for comparison between routes. For example, comparing distributions for the alternative routes of I-5 and I-405 in the Seattle Metro Area clearly highlighted that I-405 was more reliable across various levels of congestion and nonrecurring conditions. The use of L02 to analyze reliability performance of roadway improvements was also tested and found to be quite effective. However, this analysis was found to be most effective at a smaller scale than at the route level since these improvements often affect a much smaller portion of roadway. For example, the I-405 Braided Ramps Project that was tested modified approximately 1 mile of roadway. Therefore, reliability performance measurement was scaled down to a 3-mile segment, where improvement in reliability across most conditions was clearly observed. Additionally, research revealed that the cumulative distribution charts provided primarily qualitative reliability information. The use of pie charts to show regime breakdown, and standard deviation of TTI to measure reliability improvements, were helpful in converting reliability information to quantitative results. The most practical application for the L02 methodology and results was to upload them to the DRIVE Net platform. DRIVE Net is an online tool where transportation agencies and everyday commuters can view travel time reliability information for any route or combination of routes. This accessible information can aid roadway improvement planning and evaluation and help drivers find the best commute routes. For the pilot test of L07, various traffic data have been used, including WSDOT DRIVE Net Gray Notebook capacity analysis, single-loop detector data, traffic accident data, and WSDOT projects information. This study compared the measure of effectiveness, TTI curve, and the benefit–cost analysis with the results computed based on empirical data. The test results suggest that the tool tends to underestimate travel time under high traffic volumes and to generate overoptimistic measure of effectiveness and TTI curves. The major findings are (1) the classification of treatment types is trivial and inefficient, and the 15 types of very specific treatments are unable to address actual projects; (2) it is difficult to define some parameters for the treatment (e.g., the reduction of average accident clearance time) for the benefit–cost analysis; (3) travel time reliability improvement only takes up a small portion of the total treatment benefit; (4) the major benefits result from the reduction of number of accidents, and the

133 accuracy in estimating the future accident number is the key factor influencing the benefit–cost analysis results; and (5) the detailed results and TTI curves are inaccessible, which limits further comparison. For FREEVAL, tests were conducted to verify tool accuracy for two different study sites in Seattle, Washington: an urban section of I-5 with a high ramp density and a less urban section of I-405 with zero ramps. Ground truth travel times for each study site were calculated from spot speed data collected from dual-loop detectors. The Gray Notebook procedure was used to calculate segment-level travel times from spot speeds. The results obtained from this study by comparing the predicted travel time distribution outputted from FREEVAL to the ground truth travel times show that FREEVAL tends to be overoptimistic in its predictions of travel times. A second test comparing results between different seed days showed that the seed day does have an influence on the effect of the results. This suggests that multiple trial runs using several different seed days may be necessary in order to be confident in the test results. In sum, based on the testing results, FREEVAL does provide a decent ballpark estimation of the actual distribution on travel times and hints that the main sources and factors influencing travel time reliability have been accounted for by the tool. In order to assess the accuracy of the STREETVAL software, a test was performed on an urban arterial in Seattle, Washington. Results from the test were obtained by comparing the predicted travel times for the study facility outputted by the tool, to the actual travel times obtained from ALPR data. The results show that the tool tends to underpredict the dispersion level of the travel time distribution. The predicted travel time distribution is less dispersed than the actual travel time distribution from the ALPR data, although the tool can reasonably predict the mean travel time. The discrepancy in travel times suggests that some other factors (not accounted for) are influencing the vehicle travel times. A few possible unaccounted factors are (1) vehicle speeds may be different than the posted speed limit and need to be properly calibrated for in the model; (2) vehicles slowing down or speeding up to catch traffic lights; and (3) vehicles may be blinded by the sun during the sunrise and sunset hours, and this could have an influence on the driver speed and segment travel times. C11 accounts for travel time reliability as well as reoccurring congestion. It requires minimal data for performing assessment of impacts of highway investments, and thus allows users to perform quick assessment of the effects of highway investments. The tool comes with simple and easy scenario management features. It facilitates analyses of multiple scenarios by allowing, creating, and saving new scenarios with relative ease. The tool was tested to assess if it needs any further improvements for enhancing its potential for use by transportation agencies. After extensive testing on different improvement options, the project team developed a set of recommendations for further improvement of the tool. Detailed suggestions and potential improvements for each tool can be found in Section 9.2.

134 9.2 Suggestions and Potential Improvements 9.2.1 Potential Improvements on SHRP 2 L02 Product In general, the L02 is useful for outlining specifications for the data needed to create a TTRMS system, guiding how to organize different conditions for the CDF, helping understand how to read the CDF for impacts on delay, and identifying congestion sources for different corridors. By testing the L02 procedure, the research team finds that there are limitations within the guide.  The events classified in the guide are listed as either weather or incident. However, there is no category for “weather+incident” events. Because sometimes the cause of incidents can be attributed to and exacerbated by adverse weather conditions, the addition of a third “weather+incident” category is necessary. Guidance should also be provided for when an event should be considered a combined “weather+incident” and when these events should be considered separately.  The unique impact of each incident and weather event on travel time is hard to show by grouping large amount of data into the CDF curves. It is certainly possible to make a large number of curves and more specific nonrecurring conditions, such as collisions versus disabled vehicles and light rain versus snow versus fog. However, the data can only provide meaningful curves if there are sufficient data points to plot for each regime. Thus, the guide should help provide guides on when and how to establish TTRMS for different weather/incident severities. The recommendations on the minimum sample size for drawing meaningful curves are also needed.  The guide does not provide guides on the determination of route ends. For example, if traffic design treatments are implemented on a segment, how should engineers choose the length/boundary of the corridor for travel time reliability monitoring/analyzing relevant to the design treatments?  The guide may consider including recommended methods to analyze the duration of the impact of incidents, weather events (especially winter storm events), and other nonrecurring conditions and recognize that their impacts on travel reliability can extend past the duration of the condition.  The guide should recommend using additional charts beyond the CDF for evaluating reliability, especially where they can provide clearer quantitative information and help guide policy makers in planning future roadway improvements.  The guide suggests analyzing for improvements at the route level; however, improvements are not generally implemented along the entire route, but rather in hot spots or bottlenecks. Therefore, it is also necessary to analyze segment CDFs in addition to route-level CDFs when considering roadway modifications to improve reliability. Recommended methods for TTRMS at the segment level would help identify areas contributing the most to unreliability so that improvements can be targeted more precisely.

135 As a final note, the guide assumes the existence of a highly intelligent data collection system to synthesize the data and make a TTRMS work effectively. For example, the I-5 facility could be much better analyzed with a more extensive network of weather stations, especially those closer to the roadway. Then this weather needs to be efficiently paired to each loop observation. Weather conditions, such as brief downpours, can be very local in nature, and investing in a higher resolution of weather data would make this system much more effective. Additionally, a system with traffic detector data and incident data temporally and spatially connected can make it much easier to analyze the true impact of incidents. The research team expects that regions having data collection systems with these (or similar) features will have the easiest time implementing the L02 methodology and derive the greatest benefit from its results. Nevertheless, the team has found it to be an effective guiding tool for examining the travel time reliability in a greater detail of a region’s transportation network. 9.2.2 Potential Improvements on SHRP 2 L07 Product The L07 tool has friendly interface and is easy to use. However, the software currently only considers less commonly used design treatments for roadway segments. Based on the testing results, the research team suggests the following potential tool/guide refinements for L07:  Add a “Compute” button to allow the user to choose when to start the computation, so that the software does not need to spend time computing every time the user changes a single value.  Make the interface fit different computer resolutions. For example, if an 800*600 resolution screen is used (for most projectors), only the rows on the right and in the middle can be shown.  Be able to predict travel time during peak hours more precisely, as the tool tends to underestimate the effect of congestion.  Enable software to save results to a separate file and include more details about the results.  Consider the effect of combining multiple design treatments, because in some instances two or more treatments may be implemented on the same site.  Present more detailed guidance for some default values such as event and work zone characteristics, treatment effects.  Investigate further about the treatment effects, including potential effects, and make the coefficients in Figure 6.9 more open for modification.  Further consider effects of ramp metering on mainline flow. Because of its definition of solutions, L07 may not be an ideal tool to estimate the effect of ramp metering. However, it is possible for L07 to provide MOEs for these situations: o Whether and how mid-interchange off-ramps will affect traffic. o How on-ramp design features will affect traffic flow. For example, different ramp

136 lengths and lane numbers will have different effects on mainline traffic condition. o Effects of ramp spacing and interchange type on mainline flow. 9.2.3 Potential Improvements on SHRP 2 L08 Product In general, the FREEVAL tool is a powerful simulation tool for evaluating different reliability alternatives in association with various nonrecurrent traffic events. However, because the tool intends to cover as many aspects as possible, it requires multiple data sources, and the input procedure is complex. Below are potential improvements the research team found to be critical for improving the FREEVAL tool.  Put all the tool guide information together for user reference. For now, users need to refer to multiple reference documents that L08 provided to make sure all the steps are correctly followed.  Disable the unnecessary options for the selection of the number of HCM segments and disable the option of selecting nonbasic segment types for the beginning and ending segments.  Show alerts when steps are missing. For example, the software will keep working if the user fails to choose the ramp metering method. Another alternative is to show data input summary, the model run will not be executed until the user has confirmed the data entry is complete.  Allow more flexible data input. Though using “seed day demand + demand multiplier table” would save the user a lot of time inputting the data, it is time consuming for most engineers to get the demand multiplier table.  Because the urban and rural defaults for the selection of demand ratios in the freeway scenario generator are based on data from I-40, it is not accurate to apply these values to other study locations because demand patterns are location-specific. Either this default data option should be removed, or it should be clearly noted that these values might not be valid because they are based on one particular study location.  Most national holidays are on Mondays and Fridays. When we researchers calculated the demand multiplier they found a large travel demand variation on these days. The research team is not sure whether to use the holiday data to compute the multipliers or to consider these days as outliers and exclude them for the multiplier computing. Because of this issue, there is uncertainty about whether it will still be useful to include Mondays and Fridays. A potential improvement to the software would be allowing users to select which workdays are included in the analysis. To make the tool easier to use, there are a few aspects that could be improved for STREETVAL.  STREETVAL requires a large range of data input; researchers were unable to meet the necessary data requirements demanded from using multiple sources of loop and camera

137 data. Even if a complete set of demand data is available (most likely provided by imbedded loop detectors) for each approach, and at each intersection along the study site, additional access point demand data are still required to complete an analysis, and this probably means collecting data manually, which is a time-consuming and costly procedure. To avoid this costly manual data collection procedure, the tool should offer a method to estimate access point demand data and seed demand data.  Other improvements could be made to the procedure itself since this can be confusing for a first-time user. Providing the user with steps with clearly defined tasks would make this tool much easier and friendly to the user. The FREEVAL software is good in this respect; each task was a specific task that the user could follow consecutively in order to complete an analysis. Also, the aesthetics of the interface require some touch-ups, and there are a few glitches, such as the malfunctioning buttons and floating spreadsheet numbers. 9.2.4 Potential Improvements on SHRP 2 C11 Product The travel time reliability estimation tool was tested to assess if the tool needs any further improvements for enhancing its potential for use by transportation agencies. After extensive testing on different improvement options, a set of recommendations has been developed for further improvement of the tool. These are:  All three sub-tools—the travel time reliability, market access, and intermodal connectivity tools—could be designed as a coordinated suite with provisions to use them individually, if desired. This would allow easily combining the benefits from all these tools for use in further economic analyses. It would be more useful if the tool performs benefit–cost analysis by taking necessary information from a user about project’s capital and operation and maintenance costs and other benefits calculated outside this tool.  The tool is found to underestimate TTI values. Researchers recommend revisiting the calculation methodology and assumptions and modifying the tool to provide TTI and other performance metrics by direction of travel and time of day.  The tool takes input for incident reduction frequency and duration, instead of helping estimate or suggesting values for these inputs. The tool does not suggest which tools/methodologies to use to estimate incident reduction frequency and duration. The study team recommends adding some suggestions about what tool can be used to generate these inputs or providing a set of default values to choose from depending on improvement types being analyzed.  The input to the tool does not distinguish between types of trucks (e.g., light, medium, and heavy trucks). Instead of using proportion of different truck types, the tool uses an overall percentage of trucks in the vehicle mix. To capture travel impacts more accurately, the study team recommends performing analysis by taking truck classification into accounts. It is also recommended to use the values of time for light, medium, and

138 heavy trucks. These modifications would improve quality of assessment of travel time reliability and congestion costs.  For all multilane and signalized highways, the tool derives two-way capacity from one- way capacity (input by users) by assuming symmetrical geometry on both directions of travel. Two directions of a highway segment are not always similar in terms of geometry and other characteristics affecting capacity. Therefore, it may not be always appropriate to derive two-way capacity from one-way data. The research team recommends modifying the tool to accommodate input for both directions of travel and perform calculations by directions.  The study team recommends allowing input of hourly traffic volume in addition to AADT to facilitate calculation of travel delay and its economic impacts for any desired time of day (e.g., a.m. or p.m. peak hour). This will help assess travel impacts for any time period of a day.  Hourly traffic volume plays an important role in calculating 24-hour delay and associated costs to travelers. The temporal distribution of traffic varies by corridor (and even by specific locations within a corridor) based on land use type, employment, etc. The study team suggests modifying the tool to allow making changes to the default hourly factors that comes with the tool. Thus, users would have two options: either use the default values or enter project-specific temporal distribution data (if available).  The tool provides an option to select an analysis period (i.e., time of day) from four exclusive options (6:00 a.m. to 9:00 a.m., 9:00 a.m. to 3:00 p.m., 3:00 p.m. to 7:00 p.m., and 6:00 a.m. to 7:00 p.m.). It does not include night in the analysis. Also it does not allow selecting two or more time periods (for example both a.m. and p.m. peak periods) for analysis. To analyze peak demand periods, the tool needs to be run separately for each of the peak periods (e.g., a.m. peak or p.m. peak periods). The study team suggests expanding the list of analysis periods to include “Night” and “Daily” as options as well as allowing selecting multiple time periods for a single run.  The tool provides options to either directly enter capacity calculated based on HCM methodology or simply selecting a terrain type (e.g., flat, rolling, or mountainous) representing the project. When terrain is selected, the algorithm in the tool estimates peak capacity assuming values for other parameters needed for calculations. This capacity calculation could be made more rigorous by taking lane width, shoulder width, and other necessary data from users.  The tool comes with analysis capability of only a uniform segment of a roadway between two interchanges or signals. It would be more useful if the scope of the tool were expanded to include multiple segments containing interchanges/signals in-between or network of roadways with different geometric and traffic conditions.  For a relatively long stretch of a roadway, the tool’s architecture requires dividing the roadway into a number of segments within the scope of a scenario, because the tool analyzes only segments between two adjacent interchanges and/or signal controls. In such

139 cases, the tool takes inputs and produces outputs for each segment separately. It would be helpful if the tool summarizes the outputs by combining the data from all the segments.  The current version of the tool provides annual weekday delays and congestion costs. The project team recommends modifying the tool to provide annual output for weekdays and weekends. It is also recommended to produce output by hour of day. This will allow performing analyses by time of day (peak hour, peak period, daily, etc.), if necessary.  The tool comes with default values of reliability ratios (i.e., value of reliability over value of travel time) for personal and commercial travel. These ratios may vary by geographic location (e.g., state, region, county, city, or a subarea) of the project. It is suggested to provide links to references (if any research materials are available) with a possible range of default values so that a user can choose values appropriate for the geographic location of the project to analyze.  The tool does not take any input to specify which the base year is; instead the tool assumes the current year as the base year. This assumption may not hold for all cases. The study team recommends modifying the input screen to allow users to enter the base year of analysis. 9.3 Future Works After completing this project, the research team has found that there are some opportunities for future testing and work on SHRP 2 reliability products. The future works are listed below.  Evaluate alternative sources of travel time data such as INRIX and Bluetooth tracking. Other accurate sources of travel time data (e.g., INRIX and Bluetooth detection data) can be used as alternatives of the travel times generated from single-loop detectors, although these new travel time data are not as readily available for L02. INRIX provides travel time data collected from motorists that are using its navigation services. Bluetooth detection technology also has the ability to measure travel times by tracking cell phones and other devices. Although detectors are currently not widespread enough for network- level travel time calculation, this is an excellent emerging technology that can be applied for reliability research.  Apply L02 methodology to signalized highways and arterials to evaluate travel time reliability. Travel time data from single-loop detectors does not transfer well from freeways to signalized highways and arterials, as it uses two point speeds and assumes an average speed to calculate segment travel times. This assumption is invalid for the signalized highways and arterials. However by using INRIX or Bluetooth data for travel time calculation, travel time reliability can easily be measured for roadways other than freeways.  Expand access to travel time reliability information by advancing the DRIVE Net platform. Access to reliability information for transportation agencies and drivers can be expanded by increasing the quality and quantity of the data provided on online platforms

140 such as DRIVE Net. By acquiring travel time data from Bluetooth detectors and/or INRIX, the data might be more accurate, reliable, and available for many more roadways. This will enable much more personalized reliability data. Making this additional data available on DRIVE Net and expanding the reliability visualization tools available to users will help create a more reliable, efficient transportation network.  The testing of L07 tool mainly focuses on freeways since the loop detector data are available for calculating travel time reliability. Many roadway treatments provided in the L07 tool are designed for highways, where the required traffic data are not available for this project. Thus, the findings and results generated from the analysis for freeway systems are not directly applicable to highways. By acquiring appropriate traffic data, the benefit–cost analysis of roadway treatments for highways can be conducted. Moreover, if L07 can provide more details about the tool results, the effectiveness of the algorithm can be also examined.  For testing of FREEVAL, ground truth travel times were calculated from spot speed data generated from loop detector sensors. Travel times collected from ALPR cameras were used as the source of ground truth data for STREETVAL. For the future work, other sources of data might also be used for the same purpose, such as dedicated short-range communication device data like Wi-Fi and Bluetooth as well as a probe vehicle data source.