Guide to Establishing Monitoring Programs for Travel Time Reliability (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.

499 INTRODUCTION This appendix shows how a travel time reliability monitoring system (TTRMS) can be used to address questions about network reliability. Early on, the study team recog- nized that these use cases provided a perspective on how the monitoring system should be designed. They helped shape the system’s functional specifi cations. The narrative for each use case includes a description of the question being asked, the reason it would be posed, the steps involved in obtaining the answer, the inputs needed (including external data) to answer the question, and the results that would be developed. Travel time and travel rate probability density functions (TT-PDFs and TR-PDFs, respectively) are always at the heart of addressing the questions. Although popu- lar metrics like the travel time index, planning time index, and buffer time are not employed here, the values of those metrics can be derived from the PDFs. Effectively the PDFs are interpreted in the context of the question asked. SHRP 2 Project L14 is tasked with determining which of these measures, or others, would be most effective in conveying reliability information. The intent in Project L02 is to create the right PDF. At several places in this appendix it seems helpful to mention one or more of these common measures, but no endorsement of any specifi c measure is intended. In a few instances, the ideas of disutility and risk are used to portray the results. D USE CASE ANALYSES

500 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY OVERVIEW OF USE CASES The use cases fit the template shown in Table D.1. The template calls for a definition of the type of person asking the question (user), the question being posed, the steps involved in answering the question, the inputs needed to answer the question, and the results expected. Three additional comments about the use cases are helpful. First, the term on- time is an easy concept to grasp but difficult to define in technical terms. Even though people think about on-time as meaning not missed, there is no guarantee about being on time. Here on-time means arriving with a certain probability of not being late (or possibly early, as is often the case for freight shipments). Second, anywhere the acronym TT-PDF is used (or TT-CDF, the travel time cumulative density function), it refers to the PDF for individual vehicle travel times unless the text says otherwise. Third, fairly technical information is presented for the results (e.g., TT-PDFs for the routes that might be selected). This does not mean that such information is the only way to convey the results. Rather, this implies that such information is the basis for the answer; however, the communication paradigm might be simpler, as in a single number (e.g., from SHRP 2 Project L14). The use cases are clustered around types of TTRMS users most likely to make the inquiry. They are also broken down into providers and consumers (i.e., the supply and demand sides of system use). The stakeholders, shown in Table D.2, come from four categories: • Policy and planning support. Agency administrators and planners who have re- sponsibility for and make capital investment and operational decisions about the system; • Overall highway system. Operators of the roadway system (supply), including its freeways, arterials, collectors, and local streets; and drivers of private autos, trucks, and transit vehicles (demand); • Transit subsystem. Operators of transit systems that operate on the highway net- work, primarily buses and light rail (supply) and riders (demand); and • Freight subsystem. Freight service suppliers (supply) and shippers and receivers that make use of those services (demand). TABLE D.1. USE CASE TEMPLATE User The type of TTRMS user posing the question. Question A description of the inquiry and why it would be posed. Steps A list of the actions that have to be performed to answer the question. Inputs The data and information needed to answer the question. This description helps users understand the inputs required and helps programmers understand the data inputs that must be assembled. Result The system output at the completion of the use case.

501 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Agency administrators and planners typically want summary information about system performance. They want to know how various factors (e.g., growing demand or inclement weather) affect reliability so they can make investment decisions or for- mulate policies that ensure system reliability will be acceptable. System operators, transit operators, and freight service providers think in terms of service provided: whether trips take longer or shorter than they ought to or than they promised they would. These inquirers want technical, quantitative information, both near-term, real-time data for operations management and archived historical trend data for strategic and investment planning. Drivers, transit riders, and shippers want qualitative, anecdotal information and objective, quantitative information about reliability. They think in terms of deviations in trip time relative to the total trip time or how often they or their shipments are able to arrive within a particular time window. What they experience affects departure times, mode choice, route choice, and even destination and location choices. More- over, they make location decisions based on expected network reliability. Factors that affect reliability are clearly of interest to all system users. Some fac- tors are internal to the system, such as its operational control (e.g., signal timing), base capacity, and maintenance (e.g., work zones); other factors relate to the users, like incidents, unusually high demand, and special events; and still others are related to exogenous factors like weather and the performance of complementary and compet- ing modes. The use cases are listed in Table D.3. They are grouped into categories that pertain to agency administrators and planners, system operators and users, transit passengers, schedulers or operators, and freight customers or operators. TABLE D.2. USER TYPES AND THEIR CLASSIFICATION System User Type Service Provider (Supply) User (Demand) Policy and planning support Administrators and planners N/A Overall highway system Highway system operators (public or private) Privately owned vehicle drivers, taxi drivers, limousine drivers Transit subsystem Transit operators, transit vehicle operators Transit passengers Freight subsystem Carriers, freight movers, truck drivers Freight customers (including both shippers and receivers) Note: N/A = not applicable.

502 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.3. USE CASES FOR A TRAVEL TIME RELIABILITY MONITORING SYSTEM Category Subgroup Use Case System administrators and planners Administrators AE1: See what factors affect reliability AE2: Assess the contributions of the factors AE3: View the travel time reliability of a subarea AE4: Assist planning and programming decisions AE5: Document agency accomplishments AE6: Assess progress toward long-term reliability goals AE7: Assess the reliability impact of a specific investment Planners AP1: Find the facilities with highest variability AP2: Assess the reliability trends over time for a route AP3: Assess changes in the hours of unreliability for a route AP4: Assess the sources of unreliability for a route AP5: Determine when a route is unreliable AP6: Assist rural freight operations decisions Roadway network managers and users Managers MM1: View historical reliability impacts of adverse conditions MM2: Be alerted when the system is struggling with reliability MM3: Compare a recent adverse condition with prior ones MM4: Gauge the impacts of new arterial management strategies MM5: Gauge the impacts of new freeway management strategies MM6: Determine pricing levels using reliability data Drivers— constrained trips MC1: Understand departure times and routes for a trip MC2: Determine a departure time and route just before a trip MC3: Understand the extra time needed for a trip MC4: Decide how to compensate for an adverse condition MC5: Decide en route whether to change routes Drivers— unconstrained trips MU1: Determine the best time of day to make trip MU2: Determine how much extra time is needed Transit system Transit planners TP1: Determine routes with the least travel time variability TP2: Compare exclusive bus lanes with mixed-traffic operations Transit schedulers TS1: Acquire reliability data for building schedules TS2: Choose departure times to minimize arrival uncertainty Transit operators TO1: Identify routes with the poorest reliability TO2: Review reliability for a route TO3: Examine the potential impacts of bus priority on a route TO4: Assess a mitigating action for an adverse condition Transit passengers TC1: Determine the on-time performance of a trip TC2: Determine an arrival time just before a trip TC3: Determine a friend’s arrival time TC4: Understand a trip with a transfer Freight system Freight service providers FP1: Identify the most reliable delivery time FP2: Estimate a delivery window FP3: Identify how to maximize the probability of an on-time delivery FP4: Assess the on-time probability for a scheduled shipment FP5: Assess the impacts of adverse highway conditions FP6: Determine the start time for a delivery route FP7: Find the departure time and routing for a set of deliveries FP8: Solve the multiple vehicle routing problem under uncertainty FP9: Alter delivery schedules in real time Freight customers FC1: Minimize shipping costs due to unreliability FC2: Determine storage space for just-in-time deliveries FC3: Find the lowest-cost reliable origin FC4: Find the warehouse site with the best distribution reliability

503 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY AGENCY ADMINISTRATORS AND PLANNERS This section describes the use cases that agency administrators and planners might employ to learn about system reliability, determine what factors cause the system to be unreliable, and track reliability performance over time. Agency Administrators Agency administrators are responsible for making decisions about how to expand the network and how to operate it. Hence, they want to ask questions about what fac- tors affect reliability, to what extent such factors cause impacts, what trends can be observed across time, and what effects prior actions have had. See What Factors Affect Reliability (AE1) In this use case, the agency administrator wants to see what factors affect the reliability of the segments and routes in the system. That is, he or she wants to know to what extent system reliability is affected by incidents, weather, work zones, special events, traffic control devices, fluctuations in demand, and demand exceeding capacity. For example, if the analysis shows that the system is experiencing unreliability largely due to incidents, the administrator might want to choose to increase spending on inci- dent management systems or roadway safety improvements. The analysis might also help administrators set benchmarks against which they can test future improvements. Table D.4 presents the key question, steps, inputs, and the result for this use case. Step 1 is to select the system of interest; often, this is a region or set of facilities. In this instance the system selected is three freeway routes from the I-5/I-805 junction north of downtown San Diego, California, to the I-5/SR-15 junction in downtown San Diego. Figure D.1 shows the three routes: I-5 (Route 1), I-805/SR-15/I-5 (Route 2), and I-805/SR-163/I-5 (Route 3). In subsequent text, these three routes are identified more succinctly as I-5, SR-15, and SR-163. TABLE D.4. SEE WHAT FACTORS AFFECT RELIABILITY (AE1) User Agency administrator Question What factors affect reliability? Steps 1. Select the system of interest (e.g., a region or set of facilities). 2. Select the time frame for the analysis: the date range, days of the week, and times of day. 3. Assemble travel time (travel rate) observations for the system for the time frame of interest. 4. Label each observation in terms of the regime that was operative at the time the observation was made (i.e., each combination of nominal congestion and nonrecurring event, including none). 5. Prepare TR-PDFs for each regime identified. 6. Analyze the contributions of the various factors so that the differences in impacts can be assessed. Inputs Travel times and rates for the system and date range of interest plus information about the nominal system loading that would have been expected and any nonrecurring events. Result A set of TR-PDFs that portray the impacts of various factors on travel time reliability. Note: TR-PDF = travel rate PDF.

504 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 10 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx shows the three routes: I-5 (Route 1), I-805/SR-15/I-5 (Route 2), and I-805/SR-163/I-5 (Route 3). In subsequent text, these three routes are identified more succinctly as I-5, SR-15, and SR- 163. [Insert Figure D.1] [caption] Figure D.1. Subarea examined for Use Case AE1. [source] Source: © 2012 Google. Step 2 is to select the time frame of interest. In this instance it is 2011, all weekdays, and all 24 hours during those days. Figure D.1. Subarea examined for Use Case AE1. Map data © 2012 Google. Step 2 is to select the time frame of interest. In this instance it is 2011, all week- days, and all 24 hours during those days. Step 3 is to focus on assembling travel rate data. In this case the data are average travel rates from A to B for each route based on system detector data obtained by walking the time–space matrix for hypothetical trips that start every 5 minutes during the day on all three routes. The travel rates are displayed in Figure D.2 plotted against time of day and in Figure D.3 plotted against vehicle miles traveled (VMT) per hour. Since the data for the entire year are shown, there are 72,000 values for each route. Hence, there are 216,000 data points in the combined graphs. Travel rates are needed because normalizing by the distance makes it possible to compare the performance of one route with the others without having the differences in length confound the analysis. Step 4 is to label each observation—all 216,000 in this case—in terms of the regime that was operative for each observation. A regime is a combination of nominal system loading (e.g., VMT per hour) and a nonrecurring event, as explained in the Guide. The technique for adding these labels involves two substeps. The first substep is to add a nonrecurring event designation, if any. If these events have been tracked in real time and fields that describe them are already in the database, then this substep is

505 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY done automatically. (The Guide and case studies illustrate how this can be done and the challenges involved.) If not, they have to be identified by looking for outliers. The data for each route is plotted against time of day and VMT per hour (system loading) as shown in Figures D.2 and D.3, respectively. Hourly VMT data (effectively VMT per hour) were obtained from the Performance Monitoring System (PeMS). The actual hourly values were assigned to the sixth 5-minute observation in each hour (25 min- utes), and the other 5-minute values were generated by interpolating between these values. Starting with the most extreme (largest) outliers first, web-based databases are queried to see if explanatory nonrecurring events can be identified for the dates and times when the unusual travel rates occurred. For this particular system, the types of nonrecurring events were incidents, weather, special events, and demand. An incident was an accident or some other disruptive traffic event recorded in the PeMS database or some other source; weather was an inclement weather event; special event was an unusual event, often sports related; and demand was a condition when the VMT (implicitly, the traffic flows) was higher than normal for the time of day at which the high travel rate arose. Data points not falling into any one of these categories remained in a category designated normal. A weakness of this approach is that nonrecurring events that do not create outliers might be missed. Comments about the demand category are useful before proceeding. First, a demand designation was always the last one added. That is, explanations were sought related to weather, special events, or incidents before using demand as the explanation. The former three categories always trumped the demand designation. Hence, values in the demand category were extracted from those remaining in the normal category after those explained by weather, special events, incidents, or other nonrecurring events (e.g., work zones) are removed. This removal process was iterative. There was nothing permanent about the demand designation, unlike the other three. Second, the identification of the demand category data points had two facets. The first involved comparing the value of VMT per hour for a given 5-minute observa- tion with the average for that 5-minute time period. If the value was more than two standard deviations above the mean, it was given a demand designation. Because this technique did not work during the highly congested time periods when VMT per hour was constrained by capacity—because the VMT per hour could not be higher—a sec- ond analysis was conducted. Sequences of 5-minute time periods were sought when the VMT per hour and the travel rate were both high. (Effectively these were con- ditions when the demand-to-capacity ratio was higher than the volume-to-capacity ratio, implying there were standing queues in the system.) In this particular instance the values used were 75,000 VMT/h, 80 s/mi, and 30 minutes. The use of these values means that 5-minute time periods were labeled as being in the demand category if their VMT per hour exceeded 70,000 VMT/h, their travel rate was greater than 80 s/mi, and at least the next five 5-minute time periods (30 minutes total) were in the same condition.

506 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.2. Five-minute average weekday travel rates for three routes in San Diego. 40 60 80 100 120 140 160 180 200 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 Tr av el R at e (s ec /m i) Time of Day (hh:mm) Trends in Travel Rates by Time of Day for the I-5 Route 40 60 80 100 120 140 160 180 200 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 Tr av el R at e (s ec /m i) Time of Day (hh:mm) Trends in Travel Rates by Time of Day for the SR 15 Route L02  Guide   Inserts  for  2nd  pages   2014.08.07     For  Figure  4.9:  Bottom  of  three  graphs:               40 60 80 100 120 140 160 180 200 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 Tr av el  R at e   (s ec /m i) Time  of  Day  (hh:mm) Trends  in  Travel  Rates  by  Time  of  Day  for  the  SR  163  Route

507 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.3. Five-minute average weekday travel rates plotted against VMT per hour for three routes in San Diego.

508 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Obviously, changing these criteria affects the selection process. Basically, it changes the separation between observations that are considered “normal, high congestion” and those that are attributed to high demand on top of high congestion. The values were chosen because 70,000 VMT/h, especially for the SR-163 route, was the point at which there was a step change in the variability of the travel rates; 80 s/mi is the same as 45 mph, which is often the speed that arises when freeways are operating at capacity; and 30 minutes was deemed to be a reasonable system recovery time. It is effectively how long one assumes it takes the system to recover from normal high demand and return to a status at which the travel rate is less than 80 s/mi. Higher values imply that it is acceptable for the system to take longer to recover; shorter values assume it should take less time. Setting it at zero, for example, would suggest that the system should be able to recover from travel rates above 80 s/mi in 5 minutes. The second substep (in Step 4) involves labeling each observation based on the nominal loading of the system expected for each observation. This is done by analyz- ing the observations that remain once the nonrecurring events have been removed. The purpose of the congestion level designations is to differentiate the observa- tions based on the reliability performance to be expected based on system loading (e.g., congestion). As explained in Chapter 3, the semivariance (SV) was used because it is sensitive to how the data are distributed above the minimum value. In this instance, SV values were computed for every 5-minute interval for each of the three routes. Fig- ure D.4 presents the results. The value of r employed for each route was the minimum Figure D.4. Semivariances by 5-minute time period for the normal condition for three routes in San Diego. 0 20 40 60 80 100 120 140 160 180 200 0:00 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 0:00 Se m i-V ar ia nc e of th e Tr av el R at e pe r O bs er va ti on ([ se c/ m i]^ 2/ n) Time of Day (hh:mm:ss) I-5 SR 15 SR 163

509 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY travel rate observed for the entire year. Because the number of observations varied from one 5-minute period to another, the SVs were divided by the number of observa- tions n (effectively creating an average per observation so that the results would be comparable among the 5-minute time periods). As shown in Figure D.4, reliability becomes worse as the traffic levels increase. This should be expected: reliability should be best when the traffic volumes are low, like late at night or early in the morning. It should be poorer when the traffic volumes are higher and vehicles interact more, like during the midday, and it should be poorest when traffic volumes are highest, as in the p.m. peak, when the varying lengths of the standing queues have an impact. The maximum SV values, which are not shown in the figure, reach about 1,000. Although no right answer exists for the number of congestion categories to use, four were selected here: uncongested, low, moderate, and high. Uncongested meant the SV was below 20; low meant 20 to 40; moderate meant 40 to 120; and high meant above 120. Thus, the I-5 route was classified as uncongested all day, except high from 2:15 to 6:50 p.m. The SR-15 route was classified as follows: • Uncongested from midnight to 2:10 a.m.; • Low from 2:15 to 6:45 a.m.; • Uncongested from 6:50 to 8:15 a.m.; • Low from 8:20 to 9:05 a.m.; • Moderate from 9:10 a.m. to 2:10 p.m.; • High from 2:15 to 7:20 p.m.; and • Uncongested from 7:25 p.m. to midnight. The SR-163 route was classified as follows: • Uncongested from midnight to 6:45 a.m.; • Moderate from 6:50 a.m. to 2:15 p.m.; • High from 2:20 to 7:20 p.m.; and • Uncongested from 7:25 p.m. to midnight. Step 5 is to develop TR-CDFs for each regime, that is, each combination of nomi- nal loading (from the analysis above) and nonrecurring event (from the first categori- cal analysis), including none. The TR-CDFs are created by appropriately binning the 5-minute travel time observations. Figure D.5 presents the results. Step 6 is to interpret the results in terms of the effects on reliability of the various factors. This step overlaps with the following use case, so the results are presented there.

510 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.5. CDFs by regime for the three routes in San Diego. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 40 50 60 70 80 90 100 110 120 130 140 Cu m ul ati ve P ro ba bi lit y Travel Rate (sec/mi) TR-CDFs by Regime for the SR 15 Route Normal Uncong Normal Low Normal Mod Normal High Demand Uncong Demand Low Demand Mod Demand High Weather Uncong Weather Low Weather Mod Weather High Special Events Uncong Special Events High Incident Uncong Incident Low Incident Mod Incident High 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 40 50 60 70 80 90 100 110 120 130 140 Cu m ul ati ve P ro ba bi lit y Travel Rate (sec/mi) TR-CDFs by Regime for the SR 163 Route Normal Uncong Normal Mod Normal High Demand Uncong Demand Mod Demand High Weather Uncong Weather Mod Weather High Special Events Uncong Special Events Mod Special Events High Incident Uncong Incident Mod Incident High 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 40 50 60 70 80 90 100 110 120 130 140 Cu m ul ati ve P ro ba bi lit y Travel Rate (sec/mi) TR-CDFs by Regime for the I-5 Route Normal Uncong Normal High Demand Uncong Demand High Weather Uncong Weather High Special Events Uncong Special Events High Incident Uncong Incident High

511 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Assess the Contributions of the Factors (AE2) The objective in this use case is to determine how various factors affect system reli- ability. Such information helps inform decisions about how to improve performance: geometric treatments, capacity enhancements, operational changes, better signage, im- proved roadway striping, resurfacing, or better lighting. It can also help managers de- termine which facilities need better real-time traveler information (such as changeable message signs displaying alternate routes and travel times). Table D.5 presents the key question, steps, inputs, and the result for this use case. Step 1 involves selecting the system of interest. In this case, it is the same three routes in San Diego that were studied in the previous use case. Step 2 involves selecting the time frame for which the analysis is to be conducted. Again, consistent with the previous use case, for this analysis the time frame is 2011, weekdays, all hours. Steps 3 and 4 involve assembling the data and creating the TR-PDFs for the system in each regime. (In this case, that means for each route and regime.) Travel rates need to be used to create comparability because the facilities are of different lengths. Step 5 aims to determine the extent to which the facilities are affected by various factors. Figures D.4 and D.5 can be studied to develop these insights. Figure D.4 shows that the three routes have somewhat different daily patterns of reliability. The I-5 route has high reliability (a low SV value) throughout the day except during the p.m. peak. In contrast, the SR-15 route has an increase in its SV (a drop in reliability) across the midday (a higher SV). The SR-163 route has an even more dramatic increase in its SV across the midday but a lower SV during the early morning hours. In addition, the SR-163 route has a discernible a.m. peak, but the other two routes do not. From an interpretation standpoint, this means the I-5 route is probably the most reliable. It is still challenged during the peak, but consistently has the lowest SV values except for a few 5-minute periods from around 7 to 9 p.m. Interestingly, this means TABLE D.5. ASSESS THE CONTRIBUTIONS OF THE FACTORS (AE2) User Agency administrator Question How do various factors affect system reliability? Steps 1. Select the system of interest (e.g., facilities, routes). 2. Select the time frame for which the analysis is to be conducted. 3. Assemble travel rate data for each facility. 4. Create TR-PDFs (rates) for each facility and regime (i.e., combinations of system loading and nonrecurring event). 5. Study the TR-PDFs and determine the extent to which the facilities are affected by the various factors. 6. Rank order the facilities based on the relative impacts so that those most affected can receive mitigating treatments. Inputs A database of TR-PDFs with each observation labeled based on the regime to which it belongs (i.e., system loading and nonrecurring event). Result A rank-ordered list of the facilities based on the TR-PDFs by regime.

512 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY that even though Figure D.5 suggests the SR-15 route may have the lowest average travel rates most of the day, the most reliable route is I-5. Figure D.4 suggests that SR-163 is the least reliable route. It has the highest SV dur- ing the day (except in the early morning, when the SR-15 route has higher values), and its SV is significantly higher, especially during the morning and midday time periods. Figure D.5 provides additional insights. Although the plots are rather dense, they tell a story about the performance of these three routes. Looking at I-5 first, the TR-CDF for the uncongested or normal condition is at the far left and is almost ver- tical. This means it has very reliable travel rates during this condition. During these uncongested conditions, even the nonrecurring events affect only the top 30% of the 5-minute periods, and in the worst case they double the travel rate at the 100th percen- tile from about 50 s/mi up to 100 s/mi (seen fourth from the left and the most jagged of the group related to incidents). The performance of I-5 during congested conditions is quite different. Even when there are no identifiable nonrecurring events, higher travel rates are involved, as seen by the smooth red-colored CDF having travel rates from about 50 to 100 s/mi. More- over, when nonrecurring events occur during high congestion, the impacts are severe: the travel rates are substantially higher than for normal, high-congestion conditions. The TR-CDFs for three of these conditions largely overlap for incidents, special events, and weather, and no one CDF dominates the other. However, the TR-CDF for the demand condition under high congestion is strikingly different. It has much higher travel rates even at low percentiles, a kink at about 82 s/mi when the demand events during the high-congestion condition begin to affect the CDF, and a maximum value that is substantially smaller than that for the other three nonrecurring categories. The implication is that demand needs to be a cause for concern, and reducing the rates for low-percentile values may be possible through geometric improvements. Reducing the tail may not be that important; rather, it may be more important to focus on the tail for the three other conditions that involve much higher travel rates, even above the 50th percentile or so. The story for the SR-15 route is similar. Almost all of the regimes involving no or low congestion have similar TR-CDFs. There is some spread between 50 and 60 s/mi, but the TR-CDFs are all nearly vertical; not much variation in the travel rate occurs. The one notable exception is the TR-CDF for uncongested conditions with incidents. As with the I-5 TR-CDFs, incidents produce a major shift for the travel rates at the higher-percentile values, in this case above about 90%. The TR-CDF for high conges- tion during normal conditions is the very smooth curve on the right-hand edge of the large cluster. Like I-5, it involves a much larger range of travel rates, from 50 to 85 s/mi, and more change in the travel rate as the percentiles increase. The four TR-CDFs that are strikingly different are those for incidents, special events, weather, and demand during periods that would normally involve high conges- tion. This is not surprising, but it does reinforce the importance of taking actions to

513 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY help manage the severity of these events when they occur during congested operation. In this case, for the demand conditions there is a significant shift in the travel rates from 50 to 80 s/mi even at the zero percentile. The story for the SR-163 route is quite different. It obviously has problems. Its TR-CDFs are widely scattered, and nonrecurring events have an impact under all levels of congestion. The most important details to notice are that (a) the most significant impacts (the CDFs farthest to the right, all during high congestion) come from (right to left) weather, special events, and incidents; (b) the next two (light blue and dark red) are for weather under moderate congestion and demand during high congestion; and (c) the next three (right to left) are incidents, special events, and demand under low, not moderate, congestion conditions. With these differences noted, the reliability performance of SR-163 is otherwise similar to the other two routes. More specifically, it has a travel rate performance very similar to the other two routes under uncongested, normal conditions, but it struggles to maintain that performance when the congestion levels get higher or nonrecurring events occur. The fact that the SR-163 route has more significant shifts in the TR-PDFs for vari- ous conditions leads to a conclusion that there are problems with this route between I-805 and I-5. It is not too difficult to see why by driving the route, either physically or virtually, and observing its physical features and congestion. The highway has many curves, its geometry is tight, and there are closely spaced interchanges. More specifi- cally, SR-163 between I-8 and I-5 has the tight geometry common to older freeway facilities, and it is only two lanes wide in each direction. Although it is not the purpose of Project L02 to determine what geometric and other treatments would help alleviate reliability problems—that is the focus of other SHRP 2 projects like L07—it is obvi- ous that this section of SR-163 is one where geometric improvements and expedient response to incidents would be likely to have a significant impact on reliability. Step 6 involves rank ordering the facilities based on the relative impacts so that those facilities most affected can receive mitigating treatments. Table D.6 provides a way to develop the rankings. Columns 3 to 12 report the average SV values for each regime and the frequency (n) with which that regime occurs. The second column from the right shows the SV totals (in thousands) for each congestion condition (e.g., 573,000 for I-5 during uncongested conditions and 4,705,000 during congested con- ditions). These values are based on the sum–product of the SV and n values. The far- right column (Facility Total) in Table D.6 reports the total SV in the travel rate for the year. Table D.7 provides percentages for each regime and condition for each facility.

514 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.6. SEMIVARIANCES FOR EACH REGIME FOR THREE ROUTES IN SAN DIEGO Route Condition Normal Demand Weather Special Events Incidents ∑(SV*n) (1,000) Facility Total (1,000)SV n SV n SV n SV n SV n I-5 Uncongested 7 55,533 60 1,250 46 797 111 135 172 285 573 5,278 High 205 12,783 1,415 472 2,563 175 1,399 104 1,769 466 4,705 SR-15 Uncongested 15 24,491 47 147 68 229 29 77 139 55 400 9,465 Low 27 15,931 118 102 106 193 0 0 97 25 457 Moderate 46 14,863 127 13 151 271 0 0 93 103 740 High 241 13,918 2,415 665 3,751 162 3,113 168 3,032 587 7,868 SR-163 Uncongested 11 32,823 13 1,019 61 277 21 29 54 102 386 9,561 Moderate 56 20,950 169 519 399 333 601 344 684 354 1,841 High 261 12,764 1,789 1,028 1,924 254 1,424 243 1,385 961 7,333 Note: n = number of observations. TABLE D.7. PERCENTAGES FOR SEMIVARIANCES FOR EACH REGIME FOR THREE ROUTES IN SAN DIEGO Route Condition Normal (%) Demand (%) Weather (%) Special Events (%) Incidents (%) Total (%) Facility Total (%) I-5 Uncongested 8 1 1 0 1 11 100 High 50 13 8 3 16 89 SR-15 Uncongested 4 0 0 0 0 4 100 Low 4 0 0 0 0 5 Moderate 7 0 0 0 0 8 High 35 17 6 6 19 83 SR-163 Uncongested 4 0 0 0 0 4 100 Moderate 12 1 1 2 3 19 High 35 19 5 4 14 77 Inspection of the facility totals suggests that the least reliable facility is SR-163, which is consistent with the impression one gains from the scatterplots shown in Fig- ures D.2 and D.3. The SR-15 route is the next most unreliable (9,465 versus 9,561), but its distribution of the SV is slightly different. As Table D.7 shows, a higher percent- age can be attributed to incidents and special events during nominally high-congestion conditions. A summary of this analysis might be as follows: all three routes exhibit variations in reliability depending on the recurring congestion condition and nonrecurring event. Evidence of these differences is most significant for the SR-163 route, and it seems apparent the problems it has are due to the geometric conditions on the section of SR-163 from I-805 to I-5. All three routes are significantly affected by high conges- tion, even under normal conditions; the TR-CDF for that condition is dramatically

515 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY different from the CDFs for normal operation under less-congested conditions. Inci- dents, weather, special events, and higher-than-normal demand all have a significant effect on reliability during highly congested conditions. Finally, it is clear that these TR-CDFs provide guidance about actions that might be useful to help alleviate the reliability problems. View the Travel Time Reliability of a Subarea (AE3) In this use case, the agency administrator wants to review the travel time reliability performance of a subarea of the network. Subarea aggregations support transporta- tion network planning and operations decisions for large-scale metropolitan networks. Table D.8 presents the key question, steps, inputs, and the result for this use case. As shown in Figure D.6, two spatial aggregation approaches can provide users with subarea travel time reliability statistics: 1. A windowing approach, which isolates the subarea and focuses only on routes entirely within the subarea’s boundary. This approach allows for the evaluation of the reliability impacts of policies enacted within a specific subarea and the analysis of subarea boundary-to-boundary travel time reliability measures. 2. A focusing approach, which is aimed at reliability measures for all of the routes that pass through the subarea. This approach allows linkages and relationships to be maintained between a subarea and its surrounding districts. It can generate statistics on the reduced subarea networks without losing reliability information at the origin–destination (O–D) level for long-distance trips. TABLE D.8. VIEW THE TRAVEL TIME RELIABILITY PERFORMANCE OF A SUBAREA (AE3) User Agency administrator Question What is the reliability performance of a subarea? Steps 1. Define the boundary of the subarea of interest. 2. Choose the spatial aggregation method: windowing or focusing approach. 3. Select the date range, days of the week, and times of day over which to aggregate data. 4. Assemble TR-PDFs (rate) for the subarea differentiated by facility type, operating condition, time of day, and so forth. 5. Develop a picture of the reliability of the region that reflects the importance of each of the facility types, operating conditions, and times of day. Inputs TR-PDFs (rates) for the subarea differentiated by facility type, operating condition, time of day, and so forth. Result TR-PDFs for the region and selected routes (as shown in Figure D.6) that reflect the importance of each of the facility types, operating conditions, and times of day.

516 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 is to select the subarea of interest. In this case it is the same portion of the San Diego metropolitan area. Step 2 is to choose the aggregation method. In this particular instance, a win- dowing approach is employed, studying the reliability performance of the same three routes from A to B considered in the previous use case: I-5, SR-15, and SR-163. Step 3 is to select the date range, days of the week, and times of day over which to conduct the analysis. As with the prior use case, 2011 has been chosen, all weekdays, and all 24 hours. Step 4 is to assemble the TR-PDFs differentiated by facility type, operating condi- tion, time of day, and so forth. In this case the same regimes used in the previous use case are used: the nominal loading (uncongested or low, moderate, or high conges- tion) and nonrecurring event (incident, special event, weather) including normal (no unusual nonrecurring condition). Hence, the TR-CDFs presented in Figure D.5 still pertain. Step 5 is to focus on developing a picture of the reliability of the region that reflects the importance of each of the facility types, operating conditions, and times of day. For this purpose, the data presented in Table D.6 can be used. Assume that the three routes represent all the significant facilities in the region and that a picture of their combined contributions to unreliability is to be developed. The average SV values can be converted into totals and then summed to gain a sense of the answer. Table D.9 presents the results. The table shows that the regime involving high congestion and no nonrecurring event (i.e., normal conditions) contributes most to the total SV. High congestion, nonrecurring event is followed by the regimes for high congestion, demand and high congestion, incidents. The contributions from the regimes are individually under 10%. High congestion, weather contributes 6%; and normal, uncongested, 5%. Figure D.6. Subarea aggregation approaches.

517 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Returning to Table D.6 and reexamining the average SVs provides a more detailed sense of the similarities and differences among the regimes. For example, the SR-15 route has many of the highest average SV values (e.g., 3,751 for weather under high congestion; 3,032 for incidents under high congestion), and SVs for these regimes con- tribute substantially to the total (17% and 19%, respectively). Digging deeper leads to Figure D.5, where one can see that the CDFs for those two situations are still climbing through the 70th and 80th percentiles at the maximum travel rates (130 to 140 s/mi) plotted. Clearly, these are the conditions under which these factors have the greatest impacts on reliability. Assist Planning and Programming Decisions (AE4) In this use case, the user wants to make planning and programming decisions based on inputs from a TTRMS. Table D.10 presents the key question, steps, inputs, and the result for this use case. TABLE D.9. CONTRIBUTION BY REGIME TO TOTAL SEMIVARIANCE FOR THREE ROUTES IN SAN DIEGO Condition Normal (%) Demand (%) Weather (%) Special Events (%) Incidents (%) Total (%) Uncongested 5 0 0 0 0 6 Low 2 0 0 0 0 2 Moderate 8 0 1 1 1 11 High 38 17 6 4 16 82 Total 52 18 7 5 17 100 TABLE D.10. ASSIST PLANNING AND PROGRAMMING DECISIONS (AE4) User Agency administrator Question What agency actions are having significant impacts on travel time reliability? Steps 1. Select routes and subareas for analysis. 2. Assemble TR-PDFs for routes and areas for before-and-after traffic conditions under equivalent operating conditions. 3. Analyze the before-and-after changes in reliability along those routes and in those subareas, and relate those changes to the actions taken as part of the transportation improvement projects. 4. Find the trends in the efficacy of different types of transportation improvement projects. 5. Use the results as input into decision making associated with future agency planning and programming decisions. Inputs TR-PDFs for each route and area under the before-and-after conditions for similar network operating conditions associated with the travel rate observations about the transportation improvement project actions that were taken. Result Results of the before-and-after cause-and-effect analysis of the improvement actions taken and a process for conducting the assessment.

518 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY A description of the use case can build off AE1. Imagine a hypothetical situation in which the conditions portrayed for SR-163 are the status of that route at the end of 2009; the SR-15 conditions are for that same route (SR-163) at the end of 2010; and the conditions for I-5 are the status of that route (SR-163) at the end of 2011. Admittedly, such change would be remarkable progress, but that is actually useful here because it makes the differences clear. Step 1 is to select the routes and subareas of interest. In this case it is the same por- tion of the San Diego metropolitan area shown in Figure D.1, and the route of interest is the one involving SR-163. Step 2 is to assemble TR-PDFs for routes and areas for before-and-after traffic conditions under equivalent operating conditions. In the context of the hypothetical situation described above, this has already been done and the results are presented in Figure D.5. (However, those results must be interpreted as reflecting the SR-163 route performance in reverse chronological order, with the most recent performance presented first.) Step 3 is to analyze the before-and-after changes in reliability along the routes in the subarea and relating those changes to the actions taken as part of transportation improvement projects. Hence, in the context of the hypothetical situation described in Step 2, there are remarkable changes to assess. Step 4 is to find trends in the efficacy of different types of transportation improve- ment projects. Table D.11 presents the findings consistent with the hypothetical construct pre- sented earlier. It shows that the agency actions are improving the reliability of the facility under normal conditions, dropping the SV under uncongested conditions from 11 to 7, eliminating the existence of a moderate-congestion condition, and reducing the SV during high congestion from 261 to 205. The SV trend under the other condi- tions is less clear. In some cases there is improvement, but in other cases there is not. Generating clear trends in these other categories is somewhat difficult because the severity of the nonrecurring events can differ in one year versus another. TABLE D.11. CHANGES IN RELIABILITY OVER TIME FOR A HYPOTHETICAL ROUTE Condition Year Normal Demand Weather Special Events Incidents SV n SV n SV n SV n SV n Uncongested X 11 32,823 13 1,019 61 277 21 29 54 102 Y 15 24,491 47 147 68 229 29 77 139 55 Z 7 55,533 60 1,250 46 797 111 135 172 285 Moderate X 56 20,950 169 519 399 333 601 344 684 354 Y 73 30,794 244 115 257 464 0 0 190 128 Z 0 0 0 0 0 0 0 0 0 0 High X 261 12,764 1,789 1,028 1,924 254 1,424 243 1,385 961 Y 241 13,918 2,415 665 3,751 162 3,113 168 3,032 587 Z 205 12,783 1,415 472 2,563 175 1,399 104 1,769 466 Note: SV = semivariance; n = number of observations.

519 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Even though the agency may be doing a better job of managing the consequences of the events, the distribution of the severity of the events may make it difficult to see the impacts in a simple measure like the SV. Note in Table D.11 that the SV values for the special event and incident events show increases in the SV values from Year X to Year Y and then decreases from Year Y to Year Z. Whether progress has been made is unclear. However, in Figure D.7, the progress with special events becomes clearer. Notice that the TR-CDF for Year Z has upper percentile values that are better than in either Year X or Year Y; so, although the lower percentiles are not particularly improved, the higher percentiles are. The same is true for incidents. Although the performance for lower percentiles in Year Z is not better than in Year Y (but better than or the same as in Year X), it is better for the higher percentiles, at about the 88th percentile and above. Hence, an improvement in reliability has been accomplished. Of course, although improvements have been made in the higher per- centiles, further improvement can be made in the lower ones (for the high-congestion condition). Putting more resources into special event management and incident clear- ance would help improve performance further. Step 5 is to focus on using the results as input into decision making associated with future agency planning and programming decisions. The aggregate SVs do not portray the picture as completely as the CDFs, but they are more succinct and convey a general sense of the situation. For example, the normal condition occurs frequently, as would be expected, but the SVs are all quite small. The largest values occur during high congestion and range from 200 to 300. By comparison, the SVs for the less-frequent conditions range up to almost 4,000. Figure D.8 plots the SVs against the occurrence frequencies for events that occur 1,500 times per year or less. The normal data points are not present because their frequencies of occurrence are much larger, but as the figure shows, their SVs are relatively low, as well. The figure also shows that a few conditions merit significant attention—those located on the outer boundary of the plot—in terms of mitigating low- probability, high-consequence events. As Figure D.8 shows, demand events on SR-15 and SR-163 under nominally high-congestion conditions, incident events on I-5 and SR-15 under nominally high- congestion conditions, and weather events on SR-15 under nominally high-conges- tion conditions are situations for which mitigating strategies would have a significant payoff. Interestingly, and perhaps unexpectedly, it is the SR-15 route rather than the SR-163 route that may deserve the most attention in terms of managing significant consequences of nonrecurring events. That is, although the SR-163 route certainly has reliability problems, it does not surface as being the route that produces the most significant reliability problems. Document Agency Accomplishments (AE5) In this use case, an agency administrator wants to document recent accomplishments. These can be helpful for comparing past years with the current one. Table D.12 pres- ents the key question, steps, inputs, and the result for this use case.

520 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 37 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx incident events show increases in the SV values from Year X to Year Y and then decreases from Year Y to Year Z. Whether progress has been made is unclear. However, in Figure D.7, the progress with special events becomes clearer. Notice that the TR-CDF for Year Z has upper percentile values that are better than in either Year X or Year Y; so, although the lower percentiles are not particularly improved, the higher percentiles are. The same is true for incidents. [Insert Figure D.7] [caption] 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 40 50 60 70 80 90 100 110 120 130 140 Cu mu lat ive Pr ob ab ilit y Travel Rate (sec/mi) TR-CDFs for the I-5 Route for Special Events, High Congestion Year X Year Y Year Z 38 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.7. Using TR-CDFs to analyze performance changes. Although the performance for lower percentiles in Year Z is not better than in Year Y (but better than or the same as in Year X), it is better for the higher percentiles, at about the 88th percentile and above. Hence, an improvement in reliability has been accomplished. Of course, although improvements have been made in the higher percentiles, further improvement can be made in the lower ones (for the high-congestion condition). Putting more resources into special event management and incident clearance would help improve performance further. Step 5 is to focus on using the results as input into decision making associated with future agency planning and programming decisions. The aggregate SVs do not portray the picture as completely as the CDFs, but they are more succinct and convey a general sense of the situation. For example, the normal condition occurs frequently, as would be expected, but the SVs are all quite small. The largest values occur during high congestion and range from 200 to 300. By comparison, the SVs for the less-frequent conditions range up to almost 4,000. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 40 50 60 70 80 90 100 110 120 130 140 Cu mu lat ive Pr ob ab ilit y Travel Rate (sec/mi) TR-CDFs for the I-5 Route for Incidents, High Congestion Year X Year Y Year Z Figure D.7. Using TR-CDFs to analyze performance changes.

521 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.12. DOCUMENT AGENCY ACCOMPLISHMENTS (AE5) User Agency administrator Question What has the agency accomplished in terms of travel time reliability improvements? Steps 1. Select routes and subareas for analysis. 2. Assemble TR-PDFs for routes and areas for before-and-after traffic conditions under equivalent operating conditions. 3. Analyze the before-and-after changes in reliability along those routes and in those subareas, and relate those changes to the actions taken as part of the transportation improvement projects. 4. Find the trends in the efficacy of different types of transportation improvement projects. 5. Use the results as input into decision making associated with future agency planning and programming decisions. Inputs TR-PDFs for each route and area under the before-and-after conditions for similar network operating conditions associated with the travel rate observations about the transportation improvement project actions that were taken. Result Results of the before-and-after–based cause-and-effect analysis of the agency actions taken, as well as a process for conducting the assessment. Figure D.8. Relative importance of different conditions. 0 500 1000 1500 2000 2500 3000 3500 4000 0 200 400 600 800 1000 1200 1400 Av er ag e Tr av el R at e Se m i-V ar ia nc e of th e 5- m in ut e Pe rio ds (s ec /m i)^ 2 Frequency of Occurrence (5-minute periods/year) Factor Impacts Normal Demand Weather Special Events Incidents I-15, High I-15, High SR 15, High SR 15, High SR 163, High I-5, High SR 15, High

522 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Table D.11 and Figure D.7 can be used again for this purpose. Assuming the same hypothetical situation previously described, these two exhibits show how the performance of this hypothetical facility has changed in response to agency treatments. Table D.11 uses the SV values to show the changes. Figure D.7 uses the TR-CDFs. Fig- ure D.7, which provides added detail by showing how the performance has changed (improved) for all percentiles of the TR-CDF, can be used for detailed presentations and analyses. Table D.11 is suitable for broad-brush analyses and for gaining a general sense of how reliability has changed. Inasmuch as this use case focuses on determining the sources of unreliability for a given set of facilities and time frame of interest, the performance of several facilities needs to be assessed. There is no one right way to do this combined analysis, but a logical one is to focus on the delays caused by unreliability for each of the facilities individually and then examine the combined results. In principle, a facility or system’s performance would be completely reliable if all the travel times were the same (i.e., at the minimum travel time). Clearly, this is an ideal condition, but it is a standard or goal against which the facility or system’s performance can be measured. The steps involved in doing this analysis for a set of facilities or a region is as fol- lows. These steps are different from those shown in Table D.12. Step 1 involves selecting a region or facility for analysis. As with the prior use cases, San Diego is selected using the same three routes. Step 2 involves selecting the date range over which to view the data and the days of the week and times of day to include. As before, 2011 is examined for all weekdays and all hours. Step 3 involves assembling travel time or travel rate (or both) and traffic volume information. In this instance, travel rate and VMT data are employed. Step 4 involves analyzing the contributions to unreliability from the various sources. In this instance the following procedure was used. First, vehicle hours of delay were computed for every 5 minutes by multiplying the difference between the travel rate and the minimum rate by the VMT per hour and then dividing by 12 (to obtain VMT per 5 minutes). These results were summed for each combination of nominal congestion condition (uncongested and low, moderate, and high congestion) and non- recurring event (incident, demand, special event, weather) including normal. Finally, the percentage breakdown was computed among these conditions. Table D.13 shows the results for each of the three routes. As can be seen, for this “system,” the normal conditions are the main contributors to unreliability for the high-congestion condition on SR-15 and SR-163 and the uncongested condition for I-5 (I-5 does not spend time in either low- or moderate-congestion conditions). Clearly, focusing on improving performance during normal operation will reduce the vehicle hours of delay due to unreliability. In fact, the vehicle hours of delay in all the other categories are no more than 3% of the total for any facility.

523 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.13. CONTRIBUTORS TO UNRELIABILITY Congestion Level Nonrecurring Event All Vehicle Hours of Delay (%) I-5 SR-15 SR-163 High Normal 31.7 35.8 30.0 Demand 1.2 1.9 2.6 Incidents 1.2 1.5 2.2 Weather 0.0 0.4 0.5 Special events 0.3 0.4 0.6 Moderate Normal 0.0 25.7 39.1 Demand 0.0 0.0 1.1 Incidents 0.0 0.2 0.7 Weather 0.0 0.4 0.6 Special events 0.0 0.0 0.6 Low Normal 0.0 10.0 0.0 Demand 0.0 0.0 0.0 Incidents 0.0 0.0 0.0 Weather 0.0 0.1 0.0 Special events 0.0 0.0 0.0 Uncongested Normal 62.7 23.0 20.9 Demand 1.2 0.1 0.7 Incidents 0.4 0.1 0.1 Weather 0.8 0.2 0.2 Special events 0.0 0.0 0.0 For the nonrecurring conditions other than normal, the next most significant source of delay is during demand conditions when the facility would normally be in high congestion. The second-most significant category is incidents during high con- gestion. Hence, to have the most significant impacts on delay due to unreliability, the priorities would be (a) normal conditions during high congestion, (b) other normal conditions during other levels of congestion, (c) demand during high congestion, and (d) incidents during high congestion. This prioritization covers the conditions that contribute more than 1%, except for demand during uncongested operation for I-5 and demand during moderate congestion for SR-163. Assess Progress Toward Long-Term Reliability Goals (AE6) An administrator wants to determine if the agency is meeting its long-term reliability improvement goals by viewing changes to travel time variability over time. The agency administrator might use this information to see how well the agency is meeting the established goals and report the progress observed to the public. Table D.14 presents the key question, steps, inputs, and the result for this use case.

524 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY This use case is identical to AE4 or AE5 except that the focus is on a region or system instead of a route. The steps in the analysis are effectively the same, and the results would be interpreted in the same manner. Assess Reliability Impact of a Specific Investment (AE7) The agency administrator wants to see if a specific agency investment, improvement action, or policy has had a positive impact on travel time reliability. He or she wants to evaluate the impacts of past decisions to decide if the action has been effective and to report this information to the public. Table D.15 presents the key question, steps, inputs, and the result for this use case. This use case is similar to AE4 and AE6 except that the focus is on the impacts of a specific investment. Of course, compound effects like changes in flow patterns result- ing from a capacity enhancement or increase can cloud this analysis, but if the change is substantial enough that the impact can be seen in spite of these other factors, then TABLE D.14. ASSESS PROGRESS TOWARD LONG-TERM RELIABILITY GOALS (AE6) User Agency administrator Question What progress has been made toward achieving long-term reliability goals? Steps 1. Select a region for analysis. 2. Select the date range over which to view the data, as well as the days of the week and times of day to include in the analysis. 3. Select a granularity for the analysis (e.g., year, quarters, seasons). 4. Assemble TR-PDFs for routes and subareas and the region so that trends in the TR-PDFs can be observed. 5. Determine what changes and rates of change in reliability have been occurring by examining the rates at which various percentiles of the TR-PDFs are changing. Inputs TR-PDFs for each route and area across time for the date range and other specifications desired. Result An assessment of the trends in changes of the percentile values of the TR-PDFs. TABLE D.15. ASSESS RELIABILITY IMPACT OF A SPECIFIC INVESTMENT (AE7) User Agency administrator Question What has been the impact on reliability of a specific investment? Steps 1. Select the area or routes on which the change was implemented. 2. Select the date the change was implemented, as well as the date ranges for the prechange and postchange analyses. 3. Select the conditions under which the change will be assessed (e.g., peak hours, weekdays). 4. Assemble TR-PDFs for the areas and routes where the change was implemented for the prechange and postchange date ranges for the operating conditions of interest. 5. See if significant changes in the TR-PDFs have occurred in one or more instances. Inputs TR-PDFs for each route and area across time for the date ranges of interest and other specifications desired. Result An assessment of the effect on the TR-PDFs caused by the change.

525 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY the impact can be assessed. The procedure involved is similar to that presented in AE4. As described in Step 4, assemble TR-PDFs for the areas and routes where the change was implemented for the prechange and postchange date ranges (safety treatment impact analyses use a similar procedure) to see what improvement in reliability has occurred. If the changes from year to year are due to specific actions being taken, then the changes in reliability are attributable at least in part to those actions. Microscopic traffic simulation is a useful tool here in that it allows the impacts of such changes to be assessed one at a time, removing the compound effects. Agency Planners This group of use cases shows how a TTRMS can be used by highway agency planners to understand the reliability performance of their system. Find the Facilities with Highest Variability (AP1) In this use case, the agency planner wants to determine which regional facilities experi- ence the highest travel time variability. She or he can use this information to determine which facilities need further reliability analysis and to develop appropriate policy or investment actions. Table D.16 presents the key question, steps, inputs, and the result for this use case. The question posed is effectively the same as in AE2. It seeks a rank ordering of selected facilities based on reliability. The aim is to identify appropriate corrective treatments. Table D.6, presented as part of AE2, responds directly to the question. As the table shows, the SR-163 route has the greatest variability in travel times. It should be ranked first. It is followed closely by SR-15 and distantly by I-5 (the latter has a variability that is half as large). Digging further, and examining individual regimes, one can see that focusing on SR-15 might be a better choice than SR-163, even though it is not top ranked. The SR-15 route has the highest average SVs during high conges- tion and weather, high congestion and special events, and high congestion and inci- dents. For the SR-163 route, the most important target seems to be high congestion and weather, which might be addressed by improvements on the section of SR-163 between I-8 and I-5. TABLE D.16. FIND THE FACILITIES WITH HIGHEST VARIABILITY (AP1) User Agency planner Question What facilities have the most travel time variability? Steps 1. Select the facilities of interest (could be all of them). 2. Determine the metric by which the variability will be assessed (e.g., the SV). 3. Assemble TR-PDFs (rates) for each facility under equivalent operating conditions (could be more than one, or all together, or the same number of observations of time periods spent in each condition). 4. Rank order the facilities based on the variability metric. Inputs A database of TR-PDFs (rate) for each facility under equivalent operating conditions (could be more than one, or all together, or the same number of observations of time periods spent in each condition). Result A rank-ordered list of the facilities based on the variability in travel times.

526 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Assess the Reliability Trends over Time for a Route (AP2) An agency planner wants to see how travel time variability is changing over time for a route. The planner could use this information to monitor reliability trends for a route. If the travel time variability is increasing over time, the planner could monitor the situation to determine when a policy or improvement intervention would be valuable. Table D.17 presents the key question, steps, inputs, and the result for this use case. This use case is very similar to AE4. Reflecting back to its discussion, Table D.11 shows changes in reliability over time for a hypothetical route. That same information would be developed to answer the question posed here. The analysis would be the same, as would the conclusions drawn. Assess Changes in the Hours of Unreliability for a Route (AP3) This is a different type of use case that focuses on how long (for how many hours) a route’s performance is unacceptable and allows the agency planner to see how this per- centage has changed over time. This information would help the planner determine if travel time variability is increasing, which would mean that periods of high variability are lasting longer and are affecting more travelers. For routes with lengthening periods of variability, the planner might want to perform further analysis to find the cause and determine what mitigating actions can be taken. Table D.18 presents the key question, steps, inputs, and the result for this use case. TABLE D.17. ASSESS THE RELIABILITY TRENDS OVER TIME FOR A ROUTE (AP2) User Agency planner Question What is the trend in reliability over time for a route? Steps 1. Select a route for analysis. 2. Select the date range over which to view the data, as well as the days of the week and times of day to include in the analysis. 3. Select a granularity for the analysis (e.g., year, quarters, seasons). 4. Assemble TR-PDFs for the route for the days, times of day, and so forth, of interest. 5. Determine what changes in reliability have occurred by examining the changes in the various percentiles of the TR-PDFs. Inputs TR-PDFs for the route across time for the date range and other specifications desired. Result An assessment of the trends in changes of the percentile values of the TR-PDFs.

527 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY The data used in AE4 can be repurposed here to answer the question posed. As indicated there, Table D.11 shows the average SV values by regime for each of three years for a hypothetical facility. What is not shown explicitly is the number of hours that the facility operated in each regime; instead, the number of 5-minute time periods is shown. Table D.19 presents the data from Table D.11 in a slightly different format. The average SV values are still shown, but the counts of 5-minute time periods have been replaced by the number of hours that the facility was operating in regimes when the SV value was 100 or greater. TABLE D.18. ASSESS CHANGES IN THE HOURS OF UNRELIABILITY FOR A ROUTE (AP3) User Agency planner Question What is the change in the number of hours for which the route has an unreliable travel time? Steps 1. Select the route for which the change assessment is desired. 2. Determine how unreliable performance is defined (e.g., the spread between the 80th and 20th percentile travel rates divided by the 50th percentile travel rate). 3. Select a value of the metric that is considered representative of reliable travel times. 4. Determine the period of time for which the hours will be counted (e.g., a year, a quarter, a season, only weekdays). 5. Select the date ranges for the before-and-after conditions. 6. Select the operating conditions under which the change will be assessed (e.g., peak hours, weekdays, all times). 7. Assemble TR-PDFs for the route for the before-and-after date ranges and for the system operating conditions of interest. 8. Determine the number of hours that the travel rate is unreliable (per unit time) for the before-and- after conditions. 9. Determine the change in the number of hours of unreliable travel time (per unit time). Inputs TR-PDFs for the route and across time for the date ranges of interest and other specifications desired. Result An assessment of the extent to which the number of hours of unreliable operation has changed. TABLE D.19. CHANGES IN HOURS OF UNRELIABILITY OVER TIME FOR A HYPOTHETICAL ROUTE Condition Year Normal Demand Weather Special Events Incidents TotalSV Hours SV Hours SV Hours SV Hours SV Hours Uncongested X 11 0 13 0 61 0 21 0 54 0 0 Y 15 0 47 0 68 0 29 0 139 5 5 Z 7 0 60 0 46 0 111 11 172 24 35 Moderate X 56 0 169 43 399 28 601 29 684 30 129 Y 73 0 244 10 257 39 0 0 190 11 59 Z 0 0 0 0 0 0 0 0 0 0 0 High X 261 1,064 1,789 86 1,924 21 1,424 20 1,385 80 1,271 Y 241 1,160 2,415 55 3,751 14 3,113 14 3,032 49 1,292 Z 205 1,065 1,415 39 2,563 15 1,399 9 1,769 39 1,167

528 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Notice that almost all of the hours of unreliable operation are during the high-con- gestion, normal regime. Moreover, those hours are largely unchanged across the years. However, the hours spent in the other unreliable regimes decline. The only exception is the uncongested, incidents regime, in which the number of hours increases from zero to five and then to 24. In all other regimes, the total declines. For example, the hours drop from 86 to 55 and then to 39 for the high-congestion, demand regime; and for the high-congestion, incidents regime, from 80 to 49 and then to 39. Clearly, the facil- ity’s reliability performance has improved. Assess the Sources of Unreliability for a Route (AP4) In this use case, an agency planner wants to determine what factors affect the reli- ability of a route. She or he wants to investigate this to prioritize long-term system improvements. For example, if incidents are creating high and variable travel times along a route, the planner might want to stress safety improvements, such as deploy- ing freeway service patrols at certain corridor locations to clear disabled vehicles more quickly. Table D.20 presents the key question, steps, inputs, and the result for this use case. The three routes in San Diego employed earlier could be employed to illustrate this use case. However, a new example will be used. Step 1 is to select the route of interest. Westbound I-8 in San Diego has been cho- sen. As shown in Figure D.9, the section being used for analysis lies between La Mesa on the eastern end of the study area and Morena on its western end, or more specifi- cally from Baltimore Drive to the interchange with I-5. The data being used to conduct the analysis are observations of average weekday travel times. Step 2 is to select the data range over which to review the data, as well as the days of the week and times of day to include in the analysis. In this instance the date range is November 3, 2008, until February 27, 2009, for weekdays and all hours. TABLE D.20. ASSESS THE SOURCES OF UNRELIABILITY FOR A ROUTE (AP4) User Agency planner Question What are the sources of unreliability for a route? Steps 1. Select the route of interest. 2. Select the date range over which to view the data, as well as the days of the week and times of day to include in the analysis. 3. Label each observation in terms of the regime to which it belongs: its nominal loading condition (congestion level) and nonrecurring event (including none). 4. Create TR-PDFs so that the impacts of various factors can be assessed. 5. Analyze the changes in reliability caused by these factors so that the differences in impact severity can be assessed. Inputs TT-PDFs for the route for the date range, and so forth, for which the data are desired. Result An assessment of the impacts that various factors have on travel time reliability.

529 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 3 involves labeling each of the data points based on the regime to which it belongs: combinations of nominal loading condition (demand-to-capacity ratio) and nonrecurring event (including none). The procedure used is identical to that described for Use Case AE1. Plots of the data versus time of day and VMT per hour were pre- pared, and outliers were identified. Explanations were sought for the outliers, and the remaining data points (in the none category in terms of nonrecurring events) were analyzed to establish categories of operation in terms of system loading (congestion). The results are shown in Figures D.10 and D.11. Figure D.10 shows plots of the data against time of day and VMT per hour. Figure D.11 shows the SV trends. The SV plot in Figure D.11 shows that the route has fairly reliable travel times except during the a.m. peak (about 7:00 to 10:00 a.m.) and during the p.m. peak (about 4:30 to 8:00 p.m.), but to a much lesser extent. The average travel rate follows a similar trend. Figure D.12 shows the CDFs by regime. The CDFs with the most vari- ation in travel rates are (from right to left) high congestion, weather; high congestion, incidents; moderate congestion, special events; and moderate congestion, incidents. The next three (right to left) are low congestion, weather; moderate congestion, weather; and high congestion, normal. The CDFs for the remaining regimes are all nearly identical. It is important to note the number of instances in which weather is a significant, nonrecurring factor. Table D.21 shows the average SV values and the number of 5-minute time periods by regime, and Table D.22 shows the number of hours for those regimes in which the SVs exceed 100. Figure D.9. I-8 study area, San Diego. Map data © 2012 Google. 57 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx The three routes in San Diego employed earlier could be employed to illustrate this use case. However, a new example will be used. Step 1 is to select the route of interest. Westbound I-8 in San Diego has been chosen. As shown in Figure D.9, the section being used for analysis lies between La Mesa on the eastern end of the study area and Morena on its western end, or more specifically from Baltimore Drive to the interchange with I-5. The data being used to conduct the analysis are observations of average weekday travel times. [Insert Figure D.9] [caption] Figure D.9. I-8 study area, San Diego. [source] Source: © 2012 Google.

530 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 59 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.10. Plots of average travel time against time of day and VMT per hour, I-8 westbound, San Diego. [Insert Figure D.11] [caption] Figure D.10. Plots of verage travel time against time of a d VMT per hour, I-8 westbound, San Diego.

531 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 60 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.11. Average semivariance (SV) trends, I-8 westbound, San Diego. The SV plot in Figure D.11 shows that the route has fairly reliable travel times except during the a.m. peak (about 7:00 to 10:00 a.m.) and during the p.m. peak (about 4:30 to 8:00 p.m.), but to a much lesser extent. The average travel rate follows a similar trend. Figure D.12 shows the CDFs by regime. The CDFs with the most variation in travel rates are (from right to left) high congestion, weather; high congestion, incidents; moderate congestion, special events; and moderate congestion, incidents. [Insert Figure D.12] [caption] Figure D.11. Average semivariance (SV) trends, I-8 westbound, San Diego. Figure D.12. CDFs by regime for the average travel rate, I-8 westbound, San Diego. 61 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.12. C Fs by regime for the aver g travel rate, I-8 westbound, S iego. The next three (right to left) are low congestion, weather; moderate congestion, weather; and high congestion, normal. The CDFs for the remaining regimes are all nearly identical. It is important to note the number of instances in which weather is a significant, nonrecurring factor. Table D.21 shows the average SV values and the number of 5-minute time periods by regime, and Table D.22 shows the number of hours for those regimes in which the SVs exceed 100. [COMPOSITION: Please align decimal places in table.] Table D.21. Average Semivariances and Number of 5-Minute Time Periods by Regime Route Condition Normal Demand Weather Special Events Incidents SV n SV n SV n SV n SV n

532 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Several trends are clear from Tables D.21 and D.22: • Although many 5-minute time periods are spent in the normal condition, the SV is never particularly large, reaching only 45 in the high-congestion, normal regime; • The regime with the highest SV is high congestion, weather, followed by high conges- tion, incidents; moderate congestion, special events; moderate congestion, incidents; and low congestion, weather. The SVs for all other regimes are relatively small; • There are 99 (of a total of 1,872) hours of operation for regimes when the SV exceeds 100 (5% of the total hours); and • The regime with the most hours is low congestion, weather, with 51 hours. It seems clear that focusing on enhancing facility reliability during inclement weather would be a logical action to take. Determine When a Route Is Unreliable (AP5) In this use case, the agency planner wants to see when a route’s travel time reliability is unacceptable. The planner can use this analysis to determine if travel time variability is an all-day problem, or if it is confined to specific time periods. She or he can use this insight to decide where and when to implement corrective measures like ramp meter- ing to help mitigate congestion-induced variability. The analysis can also help plan- ners determine where and when high-occupancy vehicle (HOV) or high-occupancy toll (HOT) lanes could be an alternative to provide more consistent travel times for car- pools or paying drivers. Planners might also use this analysis to see what can be done in rural areas to mitigate the impacts of beach traffic in the summer or recreational skiing traffic in the winter. Table D.23 presents the key question, steps, inputs, and the result for this use case. TABLE D.21. AVERAGE SEMIVARIANCES AND NUMBER OF 5-MINUTE TIME PERIODS BY REGIME Route Condition Normal Demand Weather Special Events Incidents SV n SV n SV n SV n SV n I-8 WB Uncongested 5 1,973 16 29 27 89 0 0 17 15 Low 9 12,840 21 276 101 610 20 16 24 220 Moderate 11 2,633 35 110 80 147 473 37 337 115 High 45 2,916 50 17 1,180 176 0 0 805 245 Note: WB = westbound. TABLE D.22. AVERAGE SEMIVARIANCES AND HOURS OF UNRELIABLE OPERATION BY REGIME Route Condition Normal Demand Weather Special Events Incidents SV Hours SV Hours SV Hours SV Hours SV Hours I-8 WB Uncongested 5 0 16 0 27 0 0 0 17 0 Low 9 0 21 0 101 51 20 0 24 0 Moderate 11 0 35 0 80 0 473 3 337 10 High 45 0 50 0 1,180 15 0 0 805 20

533 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY This use case can be addressed using any one of the routes examined before. The metric can again be the SV, and the time frames can be those for which data were avail- able: all of 2011 in the case of I-5, SR-15, and SR-163; and November 3, 2008, until February 27, 2009, in the case of I-8. The SV data suggest that • I-5 is unreliable only during the p.m. peak. • SR-15 is somewhat unreliable during midday and significantly unreliable during the p.m. peak. • SR-163 is more unreliable during the midday and equally unreliable during the p.m. peak. • I-8 is unreliable during the a.m. peak and to a lesser extent during the p.m. peak and into the early evening. The reasons for the unreliability are predominantly as follows: • High congestion during the p.m. peak in the case of I-5, SR-15, and SR-163; and the a.m. peak in the case of I-8; • Weather, which is a significant source of unreliability for all four routes, especially during regimes involving high congestion, but for other regimes, as well; • Incidents, although they predominantly have an impact during regimes involving high congestion; and • Special events, especially during the early evening on I-8. Tables D.24 and D.25 provide hour and percentage breakdowns, respectively, of the times (regimes) when each facility’s average SV exceeds 100. TABLE D.23. DETERMINE WHEN A ROUTE IS UNRELIABLE (AP5) User Agency planner Question When does a route have unreliable travel times? Steps 1. Select the route for which the reliability assessment is desired. 2. Select a metric to assess reliability and a value to be used to distinguish between reliable and unreliable operation. 3. Determine the time frame for the analysis (e.g., a year, a quarter, a season, only weekdays; peak hours, weekdays, all times). 4. Assemble TR-PDFs for the route for the time period and system operating conditions of interest. 5. Determine the times when the route has unreliable travel times. 6. Search for reasons why the route might have had unreliable travel times under those conditions (e.g., weather, incidents, work zones). 7. Create a list of those reasons and figures that show the percentage of time during which those conditions exist. Inputs TR-PDFs for the route and across time for the date ranges of interest and other specifications desired. Result A list of those reasons and figures that show the percentage of time during which those conditions exist.

534 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Notice that the trends are quite different for I-8 versus the other three facilities. I-8 has all of its hours of unreliable operation during nonrecurring events (weather, spe- cial events, and incidents), and none during normal operation. In contrast, the other three routes have more than 76% of their unreliable operation during high-congestion, normal operation. Moreover, I-8 has no percentages attributable to demand condi- tions, but the other three do. TABLE D.24. HOURS OF UNRELIABLE OPERATION BY REGIME AND FACILITY Route Condition Normal Demand Weather Special Events Incidents Total HoursSV Hours SV Hours SV Hours SV Hours SV Hours I-5 Uncongested 7 0 60 0 46 0 111 11 172 24 35 High 205 1,065 1,415 39 2,563 15 1,399 9 1,769 39 1,167 SR-15 Uncongested 15 0 47 0 68 0 29 0 139 5 5 Low 27 0 118 9 106 16 0 0 97 0 25 Moderate 46 0 127 1 151 23 0 0 93 0 24 High 241 1,160 2,415 55 3,751 14 3,113 14 3,032 49 1,292 SR-163 Uncongested 11 0 13 0 61 0 21 0 54 0 0 Moderate 56 0 169 43 399 28 601 29 684 30 129 High 261 1,064 1,789 86 1,924 21 1,424 20 1,385 80 1,271 I-8 Westbound Uncongested 5 0 16 0 27 5 0 0 17 0 0 Low 9 0 21 0 101 51 20 0 24 0 51 Moderate 11 0 35 0 80 0 473 3 337 10 13 High 45 0 50 0 1,180 15 0 0 805 20 35 TABLE D.25. PERCENTAGE OF UNRELIABLE OPERATION BY REGIME AND FACILITY Route Condition Normal Demand Weather Special Events Incidents Total (%)SV % SV % SV % SV % SV % I-5 Uncongested 7 0 60 0 46 0 111 1 172 2 3 High 205 89 1,415 3 2,563 1 1,399 1 1,769 3 97 SR-15 Uncongested 15 0 47 0 68 0 29 0 139 0 0 Low 27 0 118 1 106 1 0 0 97 0 2 Moderate 46 0 127 0 151 2 0 0 93 0 2 High 241 86 2,415 4 3,751 1 3,113 1 3,032 4 96 SR-163 Uncongested 11 0 13 0 61 0 21 0 54 0 0 Moderate 56 0 169 3 399 2 601 2 684 2 9 High 261 76 1,789 6 1,924 2 1,424 1 1,385 6 91 I-8 Westbound Uncongested 5 0 16 0 27 0 0 0 17 0 0 Low 9 0 21 0 101 52 20 0 24 0 52 Moderate 11 0 35 0 80 0 473 3 337 10 13 High 45 0 50 0 1,180 15 0 0 805 21 36

535 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Assist Rural Freight Operations Decisions (AP6) In this use case, an agency planner wants to help out with rural freight operations. He or she wants to understand where and when rural freight delivery times are unreliable. The planner would need access to truck-related travel times, which are different from cars and other vehicles. Assuming that such data are available, this type of analysis would help the planner make decisions about how to help rural freight operations, such as special climbing lanes, snow removal, and highway geometry, especially in areas with little or no real-time traffic data. Table D.26 presents the key question, steps, inputs, and the result for this use case. The Bluetooth data from US-50 between Placerville and South Lake Tahoe, Cali- fornia, provide an excellent basis for examining this use case. Travel times along US-50 were monitored by CalTrans using a set of Bluetooth detectors. The area is rural; no system detectors exist. Figure D.13 shows approximately where four detectors were placed. For the pur- poses of this analysis, the Placerville Bluetooth sensor is the pushpin furthest to the left (west); the South Lake Tahoe sensor is the one furthest to the right (east). The distance between the sensors is about 50 miles; the highway is four lanes wide (two lanes in each direction) and undivided. The travel time under good weather conditions is about 50 minutes. From January 28, 2011, until April 21, 2011, 29,533 observations of trip times were observed on 82 days (about 360 observations per day). All of the observations are displayed in Figure D.14. One can see that the observations are a combination of trip times and travel times, and some of the observations range up to 400 minutes (the time difference limit used for matching the Bluetooth pings). TABLE D.26. ASSIST RURAL FREIGHT OPERATIONS DECISIONS (AP6) User Agency planner Question Where and when are rural freight travel times unreliable? Steps 1. Select the routes and areas of interest. 2. Select the time periods and network operating conditions of interest (could be all). 3. Assemble TR-PDFs for freight trips for the routes and areas of interest and the time periods and network operating conditions of interest. 4. Search for reasons why the routes and areas might have had variations in the travel rates under those conditions (e.g., weather, incidents, work zones). 5. Create a list of the reasons why the travel rates might be more variable and a pie chart showing the percentage of time during which those conditions exist. Inputs TR-PDFs for the routes and areas of interest and across time and for the system operating conditions of interest. Result A list of the reasons why the travel rates might be more variable and a pie chart showing the percentage of time during which those conditions exist.

536 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.13. Placerville to South Lake Tahoe. Map data © 2012 Google. Figure D.14. Observations of trip times, Placerville to South Lake Tahoe. 71 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.13 shows approximately where four detectors were placed. For the purposes of this analysis, the Placerville Bluetooth sensor is the pushpin furthest to the left (west); the South Lake Tahoe sensor is the one furthest to the right (east). The distance between the sensors is about 50 miles; the highway is four lanes wide (two lanes in each direction) and undivided. The travel time under good weather conditions is about 50 minutes. [Insert Figure D.13] [caption] Figure D.13. Placerville to South Lake Tahoe. [source] Source: © 2012 Google. From January 28, 2011, until April 21, 2011, 29,533 observations of trip times were observed on 82 days (about 360 observations per day). All of the observations are displayed in Figure D.14. One can see that the observations are a combination of trip times and travel times, 72 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx and some of the observations range up to 400 minutes (the time difference limit used for matching the Bluetooth pings). [Insert Figure D.14] [caption] Figure D.14. Observations of trip times, Placerville to South Lake Tahoe. The observations, filtered and plotted against date and time, are shown in Figure D.15. Clearly, there were some days when the travel times were very large, ranging up to 250 minutes during a heavy snow storm. On days when conditions were good, the times ranged from 50 to 75 minutes. [Insert Figure D.15] The observations, filtered and plotted against date and time, are shown in Figure D.15. Clearly, there were some days when the travel times were very large, ranging up to 250 minutes during a heavy snowstorm. On days when conditions were good, the times ranged from 50 to 75 minutes.

537 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY To address the use case, Step 1 is to select the route and area of interest. In this instance, it is US-50 from Placerville to South Lake Tahoe. Step 2 is to select the time period and network operating conditions of interest. In this instance, they are Janu- ary 28, 2011, until April 21, 2011, or the duration of time during which the detectors were installed, and all conditions. Step 3 is to assemble TR-PDFs for freight trips. Since no vehicle classification data were associated with the Bluetooth observations, it will be assumed that all the observations are for freight vehicles (clearly, this assumption is not true). Step 4 is to search for reasons why there were variations in the travel rates. For this purpose, two data sources were employed: the incident data contained within PeMS and weather data. The most useful weather information related to precipitation, fog, and wind gusts. It was perceived that these three would have the most effect on travel conditions. All incidents within the monitored section were deemed to be note- worthy; none were omitted. Care was taken to associate the incidents, given their loca- tions, with those trips most likely to be affected. Trips were presumed to be affected if they would have been at about the location of the incident at the time when it occurred given the time at which they passed by the Placerville sensor (going east) or the South Lake Tahoe sensor (going west). A speed of 60 mph was assumed to estimate the travel time to the incident location. A disappointing aspect of the incident data was that some of the durations appeared to be spurious. They were often zero minutes. Consideration of the nonzero values suggested that 30 minutes would be a sensible default. That value was used in lieu of the recorded information if the recorded value was smaller. In summary, weather- and incident-related nonrecurring event information was added to each record. Figure D.15. Filtered observations of trip times, Placerville to South Lake Tahoe. 73 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx [caption] Figure D.15. Filtered observations of trip times, Placerville to South Lake Tahoe. To address the use case, Step 1 is to select the route and area of interest. In this instance, it is US-50 from Placerville to South Lake Tahoe. Step 2 is to select the time period and network operating conditions of interest. In this instance, they are January 28, 2011, until April 21, 2011, or the duration of time during which the detectors were installed, and all conditions. Step 3 is to assemble TR-PDFs for freight trips. Since no vehicle classification data were associated with the Bluetooth observations, it will be assumed that all the observations are for freight vehicles (clearly, this assumption is not true). Step 4 is to search for reasons why there were variations in the travel rates. For this purpose, two data sources were employed: the incident data contained within PeMS and weather data. The most useful weather information related to precipitation, fog, and wind gusts. It was perceived that these hree would have t most effect on travel conditions. All incidents within the monitored section w re deemed to be noteworthy; none were omitted.

538 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Table D.27 presents a summary of the travel times for each condition. One can see that the three most adverse conditions in the absence of incidents are heavy snow, snow and fog, and freezing fog. The worst is heavy snow, for which the average travel time is 125 minutes instead of 61 minutes under normal conditions. Incidents raise the average to 133 minutes. Highlighted in gray are the conditions with the largest values for the four trip time metrics. Figure D.16 presents the CDFs for the various conditions listed in Table D.27. As the labeling shows, the distributions farthest to the right are incidents, snow; freezing fog; incidents, freezing fog; incidents, heavy snow; incidents, snow; incidents, snow and fog; heavy snow; and snow and fog. Farthest to the left are a set of density func- tions lying to the left of the normal label: incidents, gusts; rain; gusts; light rain; inci- dents, light rain; normal; and incidents. Based on the data in Table D.27 and Figure D.16, the guidance one might give to freight operators would be that (a) under normal conditions, the travel time should be about 61 minutes; (b) gusty wind and rain do not seem to have a significant effect, but light snow adds about 10 to 20 minutes, as does limited visibility; and (c) heavy snow, snow and fog, and fog with freezing conditions add another 40 to 50 minutes and can TABLE D.27. TRIP TIME IMPACTS FROM ADVERSE CONDITIONS Condition No. of Observations Trip Time (min) Minimum Average Maximum Standard Deviation Normal 16,683 47.5 61.3 256.7 9.9 Gusts 5,286 47.7 60.2 176.7 9.7 Rain 908 48.1 64.3 93.8 7.9 Light snow 2,066 50.9 83.7 203.4 24.8 Limited visibility 25 58.6 85.7 128.4 23.1 Snow 136 56.3 97.3 143.2 16.5 Freezing fog 27 54.1 112.4 153.2 26.6 Snow and fog 276 63.5 125.7 237.2 39.9 Heavy snow 298 59.9 125.5 200.0 27.7 Incidents alone 191 52.0 81.6 136.8 25.7 Incidents with rain 18 53.3 56.1 65.6 3.0 Incidents with gusts 107 48.5 62.3 136.2 16.3 Incidents with light snow 78 58.8 80.9 125.1 16.5 Incidents with snow 105 64.2 106.0 153.8 18.5 Incidents in freezing fog 11 93.4 113.5 141.2 16.0 Incidents with snow and fog 49 72.4 123.6 179.1 27.7 Incidents with heavy snow 89 106.3 133.3 173.6 14.6 Note: Shading indicates conditions with the largest values for the four trip time metrics.

539 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.16. CDFs for trip times during adverse conditions. make the trip as long as 2 hours. Incidents have an impact, but often not a substantial one. The exception is incidents that occur under normal conditions, which can add 20 or more minutes to the travel time. HIGHWAY SYSTEM OPERATORS AND USERS This section focuses on questions that might be asked by people who both manage (supply side) and use (demand side) the highway network. Each use case is defined fol- lowed by a description of the context in which it is applied and the results obtained. The order is consistent with the use case template presented in Table D.1. Roadway System Managers Roadway system managers need to make decisions about how to operate the system. They are directly responsible for maintaining and improving reliability. These people often work for transportation management centers, metropolitan agencies, and state DOTs. Reliability information helps them better manage the network’s operation. For example, a roadway system manager might use reliability information to determine what actions to take when conditions are awry and set goals for reliability improve- ments. Such people might also use reliability information to determine how to respond to incidents and other events. View Historical Reliability Impacts of Adverse Conditions (MM1) In this use case, a system manager wants to know how adverse conditions affect travel time reliability. Knowing the impacts of conditions such as special events, bad weather, incidents, and lane closures can help him or her manage the system better. For ex- ample, such knowledge helps provide better real-time information to the public by 76 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx freezing fog; incidents, heavy snow; incidents, snow; incidents, snow and fog; heavy snow; and snow and fog. Farthest to the left are a set of density functions lying to the left of the normal label: incidents, gusts; rain; gusts; light rain; incidents, light rain; normal; and incidents. [Insert Figure D.16] [caption] Figure D.16. CDFs for trip times during adverse conditions. Based on the data in Table D.27 and Figure D.16, the guidance one might give to freight operators would be that (a) under normal conditions, the travel time should be about 61 minutes; (b) gusty wind and rain do not seem to have a significant effect, but light snow adds about 10 to 20 minutes, as does limited visibility; and (c) heavy snow, snow and fog, and fog with freezing conditions add another 40 to 50 minutes and can make the trip as long as 2 hours. Incidents have an impact, but often not a subst tial one. The exception is incidents that occur under normal conditions, which can add 20 or more inutes to the travel time.

540 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY timely updating variable message signs with expected delay information. Knowing the impacts also helps the operator provide better pretrip information, such as anticipated speeds and delay times, and encourage people to plan ahead or take alternate routes or modes. Table D.28 presents the key question, steps, inputs, and the result for this use case. Answering this question is complicated. It gets at the heart of the difference between travel time reliability viewed from the perspective of an operator versus that of a user. To see the difference, one needs to examine both system detector data and individual vehicle data. Fortunately, both were collected simultaneously in the Lake Tahoe case study for I-5 in Sacramento, and those data are used here. Step 1 involves selecting the network area of interest. In this case it will be I-5 in Sacramento, just south of the junction with US-50. Figure D.17 shows the location of the study area. The four markers indicate the locations where Bluetooth readers were set up to record vehicles moving northbound and southbound. The use case focuses on travel times between the first and last reader. Step 2 involves selecting the adverse condition of interest. Rain is a good choice. It has a significant impact on the travel time, as shown below. Step 3 focuses on assembling the data for the location and condition of interest. In this instance, the time frame is January 24, 2011, until March 16, 2011, the dates during which the Bluetooth data were collected. Four data sets are used: Bluetooth data, PeMS detector data, PeMS incident data, and weather data. These four data sets are combined to create a unified database in which each Bluetooth observation can be cross referenced to the weather conditions and incidents. The same is true for the PeMS 5-minute travel times. Step 4 involves determining the impacts of the adverse condition; in this case, rain. It makes sense to begin exploring the impacts with a broad-brush perspective. Figure D.18 shows the daily trends in the 5-minute system detector–based travel times for the southbound direction. Clearly, weather has an adverse impact, as do inci- dents during adverse weather conditions. In addition, there is an afternoon peak, but its impact on the travel times is not significant; it boosts the times from 4 to 5 minutes up to about 7 minutes. Figure D.19 shows the CDFs for the 5-minute travel times TABLE D.28. VIEW HISTORICAL RELIABILITY IMPACTS OF ADVERSE CONDITIONS (MM1) User Roadway system manager Question What are the historical reliability impacts of different adverse conditions? Steps 1. Select the network area of interest. 2. Select the adverse conditions of interest. 3. Assemble a historical database of normal and adverse conditions for segments in the study area that have been adversely affected. 4. Determine the relative impacts of the adverse conditions. Inputs Historical TT-PDFs for the segments in the study area under various normal and adverse conditions. Result A table that shows the range of reliability impacts that has arisen due to the various adverse conditions.

541 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.17. Interstate 5 in Sacramento. Map data © 2012 Google. 79 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Fortunately, both were collected simultaneously in the Lake Tahoe case study for I-5 in Sacramento, and those data are used here. Step 1 involves selecting the network area of interest. In this case it will be I-5 in Sacramento, just south of the junction with US-50. Figure D.17 shows the location of the study area. The four markers indicate the locations where Bluetooth readers were set up to record vehicles moving northbound and southbound. The use case focuses on travel times between the first and last reader. [Insert Figure D.17] [caption] Figure D.17. Interstate 5 in Sacramento. 81 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx It is apparent that weather significantly affects the travel times when the congestion is high, but not otherwise. Hence, one answer to the question is that I-5 becomes unreliable when weather is the adverse condition. [Insert Fi re D.18] [caption] Figure D.18. Daily trends in 5-minute average travel times, I-5 southbound. [Insert Figure D.19] [caption] Figure D.18. Daily trends in 5-minute average travel times, I-5 southbound.

542 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY in the southbound direction under four regimes: high congestion, normal; high con- gestion, weather; uncongested, normal; and uncongested, weather. Not enough data points exist for weather combined with incidents to produce meaningful CDFs. It is apparent that weather significantly affects the travel times when the conges- tion is high, but not otherwise. Hence, one answer to the question is that I-5 becomes unreliable when weather is the adverse condition. Even though the question has been answered, the individual vehicle travel times lend a slightly different perspective. In this case, the question is: By how much do the individual vehicle travel times vary when the conditions are adverse? To see what the impacts are, it is useful to begin with a high-level examination of the data. Figure D.20 shows daily weekday trends in the travel times experienced by individual vehicles on I-5 southbound. Plotted are data points for 119,528 weekday trips (of 156,855 observations) across the 3-month time period. These travel time trends can be directly compared with Figure D.18, which displayed the 5-minute aver- age travel times based on the system detectors. It is clear that when the freeway is uncongested, the travel times range from just under 5 minutes (about 4.2 minutes) to about 7 minutes, with some times reaching up to 10 minutes (these might be outliers). However, during the peak (high congestion), even though there are days when vehicles still experience travel times of 5 minutes, there are days when the travel times reach up to 10 minutes. There are a few days when nonrecurring events cause travel times up to 35 minutes. But are these data scattered on a given day, or are there trends? Figure D.21 (top graph) shows the 10,000 travel times observed between February 16, 2011, at 1:14 a.m. and February 21, 2011, at 4:33 p.m. A short, abrupt transient can be seen fol- lowed by a much longer, less dramatic one. Several other smaller transients are also Figure D.19. CDFs of 5-minute average travel rates for four regimes, I-5 southbound. 82 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.19. CDFs of 5-minute average travel rates for four regimes, I-5 southbound. Even though t e q estion has been answered, the individual vehicle travel times lend a slightly different perspective. In this case, the question is: By how much do the individual vehicle travel times vary when the conditions are adverse? To see what the impacts are, it is useful to begin with a high-level examination of the data. Figure D.20 shows daily weekday trends in the travel times experienced by individual vehicles on I-5 southbound. Plotted are data points for 119,528 weekday trips (of 156,855 observations) across the 3-month time period. These travel time trends can be directly compared with Figure D.18, which displayed the 5-minute average travel times based on the system detectors. [Insert Figure D.20] [caption]

543 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.20. Daily trends in individual vehicle travel times, I-5 southbound. 83 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.20. Daily trends in individual vehicle travel times, I-5 southbound. It is clear that when the freeway is uncongested, the travel times range from just under 5 minutes (about 4.2 minutes) to about 7 minutes, with some times reaching up to 10 minutes (these might be outliers). However, during the peak (high congestion), even though there are days when vehicles still experience travel times of 5 minutes, there are days when the travel times reach up to 10 minutes. There are a few days when nonrecurring events cause travel times up to 35 minutes. But are these data scattered on a given day, or are there trends? Figure D.21 (top graph) shows the 10,000 travel times observed between February 16, 2011, at 1:14 a.m. and February 21, 2011, at 4:33 p.m. A short, abrupt transient can be seen followed by a much longer, less dramatic one. Several other smaller transients are also evident. The major transient was due to rain on February 18, 2011, and the short blip was due to an incident that same day. Figure D.21 (bottom graph) zooms in on the roughly 1,500 observations from February 18, 2011, during two major events. Although there is still some overplotting, the dots from evident. The major transient was due to rain on February 18, 2011, and the short blip was due to an incident that same day. Figure D.21 (bottom graph) zooms in on the roughly 1,500 observations from Feb- ruary 18, 2011, during two major events. Although there is still some overplotting, the dots from individual trip times can be seen. Midmorning it began to rain (at about Observation 46,000). Around 11:00 a.m. there was an incident (at about Observa- tion 46,110). The rain had no effect on the travel times. At about 1:30 p.m. there was a second incident (at about Observation 46,500), which also had no impact on the travel times. Then, at about 2:40 p.m., there was a third incident (at about Observa- tion 46,700). Although at first it had a small impact, at about 3:00 p.m. it had a major impact. The travel times rose dramatically to about 35 minutes, but by 4:00 p.m. they had dropped to about 9 minutes. The travel times began climbing again (at about Obser- vation 47,000) and peaked at about 15 minutes. While this was under way, at about 5:40 p.m. there was another incident, but it did not have a discernible impact. Hence, the long transient seems to have been due to the rain. By 6:45 p.m., with the peak over, the travel times were back down to 5 to 6 minutes (about Observation 47,700). Returning to the high-level perspective, one can ask whether this narrow banding in travel times is always present. Figure D.22 provides some insight. Plotted in the top half is the difference between the 5th and 75th percentile travel times versus the 5th percentile travel time for half-overlapping sequences of 50 Bluetooth observations across the three months. Although the overplotting makes it difficult to see how many data points there are for each combination, one can see that the spread between the 75th and the 5th percentile travel times is much larger when the 5th percentile travel time is low; in other words, when the freeway is congested, the spread in travel times is m ch smaller.

544 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 84 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx individual trip times can be seen. Midmorning it began to rain (at about Observation 46,000). Around 11:00 a.m. there was an incident (at about Observation 46,110). The rain had no effect on the travel times. At about 1:30 p.m. there was a second incident (at about Observation 46,500), which also had no impact on the travel times. Then, at about 2:40 p.m., there was a third incident (at about Observation 46,700). Although at first it had a small impact, at about 3:00 p.m. it had a major impact. The travel times rose dramatically to about 35 minutes, but by 4:00 p.m. they had dropped to about 9 minutes. The travel times began climbing again (at about Observation 47,000) and peaked at about 15 minutes. While this was underway, at about 5:40 p.m. there was another incident, but it did not have a discernible impact. Hence, the long transient seems to have been due to the rain. By 6:45 p.m., with the peak over, the travel times were back down to 5 to 6 minutes (about Observation 47,700). [Insert Figure D.21] [caption] 0 5 10 15 20 25 30 35 40 40000 41000 42000 43000 44000 45000 46000 47000 48000 49000 50000 Tra ve l T im e ( mi n) Chronological Observation Number Individual Vehicle Travel Times / I-5 Southbound 85 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.21. Daily trends in individual vehicle travel times, I-5 southbound. Returning to the high-level perspective, one can ask whether this narrow banding in travel times is always present. Figure D.22 provides some insight. Plotted in the top half is the difference between the 5th and 75th percentile travel times versus the 5th percentile travel time for half-overlapping sequences of 50 Bluetooth observations across the three months. Although the overplotting makes it difficult to see how many data points there are for each combination, one can see that the spread between the 75th and the 5th percentile travel times is much larger when the 5th percentile travel time is low; in other words, when the freeway is congested, the spread in travel times is much smaller. [Insert Figure D.22] [caption] 0 5 10 15 20 25 30 35 40 46500 46700 46900 47100 47300 47500 47700 47900 Tra ve l T im e ( mi n) Chronological Observation Number Individual Vehicle Travel Times / I-5 Southbound Figure D.21. Daily trends in individual vehicle travel times, I-5 southbound.

545 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.22. Spreads in individual vehicle travel times. 86 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.22. Spreads in individual vehicle travel times. Figure D.22 provides an example of how the 5th and 75th percentile values vary. There is some evidence that the spread between the 5th and 75th percentile travel times decreases as the 5th percentile increases. 0 5 10 15 20 25 0 5 10 15 20 25 30 35 75 th -5 th Pe rce nti le Tra ve l T im e ( mi n) Fifth Percentile Travel Time (min) (75th - 5th) Percentile Travel Time versus the 5th Percentile 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM Tra ve l T im e ( mi n) Time of Day Trends in the 5th and 75th Percentile Travel Times t(5) t(75)

546 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.22 provides an example of how the 5th and 75th percentile values vary. There is some evidence that the spread between the 5th and 75th percentile travel times decreases as the 5th percentile increases. The implication of Figure D.22 is that as the facility becomes more congested and the travel times increase, the variation in the travel times decreases. That is, the congestion makes it difficult for the vehicles to travel at widely varying speeds. Hence, when the travel time is larger, the consistency is greater. This observation means that the issues in travel time reliability are more complex than just the system detector data might suggest. Although the system is clearly unreliable during high congestion because it cannot provide the same travel time day to day for the same flow (conges- tion) condition, the travel time it does provide in each situation is very similar for all the vehicles involved (i.e., it is difficult for vehicles to have widely varying travel times), and the consistency in the travel times is high. Hence, what people may actually be complaining about when they focus on reli- ability is their inability to achieve desired travel times when the system is congested (i.e., from one day to another, or one peak load instance to another, or one situation to another). They cannot achieve the travel time they want, and the travel times they experience under those conditions are widely variable. This situation is not the incon- sistency in travel times among drivers at a given point in time. Rather, it is the inability of the system to provide the same travel time under the same operating conditions, or the same travel time when traffic is congested every time it is congested. Conversely, users regard the system as being reliable when it is consistent in providing the same travel time from one trip to the next, as it can when it is uncongested. This occurs in spite of the fact that under that condition (uncongested) the variance in individual vehicle travel times is high, because people can achieve the travel times they want. Thus, the reliability of the system is perceived to be high when the variation in the individual vehicle travel times is also high, and reliability is perceived to be low when the variance in individual vehicle travel times is low. Figure D.22 helps to show this. The message is this: people perceive reliability is high when the system lets them travel at their desired travel speeds (and achieve their desired travel times); but this is the con- dition when the variance in individual vehicle travel times is high. When the variance is high, such as when the system is lightly loaded, the system is able to deliver consistent performance from one condition occurrence to another and from one day to another, even though the variance is high, and when nonrecurring events arise. That the travel time the system can deliver varies from one condition realization to another can be seen in Figure D.23. In spite of the fact that Figure D.22 shows that individual vehicle travel times are consistent within a given situation, the travel times are different across realizations of that situation. Hence, in spite of the consistency for a given situation, the CDFs of the individual vehicle travel times, combined across situ- ations, provide the same trend in travel times as those reported in Figure D.19. In summary, from an individual vehicle travel time perspective, the answer to the original use case question is that the impact of adverse conditions (in this case, rain) is as follows: (a) under such conditions the system cannot deliver a reliable travel time;

547 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.23. CDFs for individual vehicle travel times in several regimes. 88 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx this: people perceive reliability is high when the system lets them travel at their desired travel speeds (and achieve their desired travel times); but this is the condition when the variance in individual vehicle travel times is high. When the variance is high, such as when the system is lightly loaded, the system is able to deliver consistent performance from one condition occurrence to another and from one day to another, even though the variance is high, and when nonrecurring events arise. That the travel time the system can deliver varies from one condition realization to another can be seen in Figure D.23. In spite of the fact that Figure D.22 shows that individual vehicle travel times are consistent within a given situation, the travel times are different across realizations of that situation. Hence, in spite of the consistency for a given situation, the CDFs of the individual vehicle travel times, combined across situations, provide the same trend in travel times as those reported in Figure D.19. [Insert Figure D.23 [caption] Figure D.23. CDFs for individual vehicle travel ti es in several regimes. (b) the time it can deliver is different from the time the traveler wants; and (c) the travel time it delivers is different from one occurrence of the condition to another. Be Alerted When the System Is Struggling with Reliability (MM2) The user wants to know when the travel times on a facility have become unreliable or are about to do so. Consistent with the discussions in MM1, this alert tells the user the system is entering a condition in which travelers cannot achieve the travel times they want because the congestion is too high (travel is constrained), or the system’s ability to provide consistent travel times has become low, or both. A roadway system manager might use information as the basis for sending out alerts to variable message signs or route guidance devices. Table D.29 presents the key question, steps, inputs, and the result for this use case. TABLE D.29. BE ALERTED WHEN THE SYSTEM IS STRUGGLING WITH RELIABILITY (MM2) User Roadway system manager Question Has a route or system become unreliable? Steps 1. Select the segments or routes being monitored. 2. Select the conditions for which notification is desired. 3. Design the test that will be used to identify the condition selected. 4. Monitor the TT-PDFs (or TR-PDFs) to see if an unreliable condition has arisen. Inputs Real-time information about the status of the segment or route plus historical TT-PDFs for the segments and routes under surveillance. In addition, real-time information about the network conditions as explanatory variables. Result An alert message that displays the facility and location where the travel time reliability is adverse and TT- CDFs that compare the current segment or route travel time against the ones that would be expected. Note: TT-CDF = travel time cumulative density function.

548 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY This use case is similar to incident detection or identification of times when the system’s behavior has become unstable. As in Use Case MM1 and elsewhere, under heavy congestion or under adverse nonrecurring conditions (i.e., when the system is under stress), individual vehicle travel times become less variable (i.e., more consistent) because congestion keeps people from being able to travel at the speeds they want. Moreover, the system struggles to provide the same travel times each time these condi- tions occur. It seems the system cannot control the manner in which vehicles interact or the effects on system capacity from the nonrecurring event. The steps in the use case are as follows. Step 1 is to select the segments or routes to be monitored. I-5 southbound in Sacramento will be employed. Step 2 involves select- ing the conditions for which notification is desired. The choice will be any condition in which the facility has difficulty letting users travel at their desired speeds. Step 3 involves designing the test that will be used. Step 4 is to monitor the TT-PDFs (or TR-PDFs) to see if reliability is suffering. The data collected will be used ex post facto to see when reliability was affected. Based on data for individual vehicle travel times, it appears that the lower- percentile travel times (the higher speeds) are affected earliest and most (i.e., the higher- percentile speeds decrease the most). Without question, the other percentile travel times also change, but not as dramatically, at least not initially. That is, the lower-percentile travel times provide the earliest sign that the system is entering an unreliable condition. As the system becomes more heavily loaded from either congestion or a nonrecurring event, it loses the ability to permit drivers to achieve the lowest travel times. It cannot let those vehicles thread their way through the traffic stream. Either the traffic density is too high, or the nonrecurring event has interfered with that ability (e.g., as a result of queuing). The clearest evidence of this can be seen in time traces. Figure D.24 shows the percentiles of individual vehicle travel times for I-5 southbound in Sacramento on March 1, 2011, the same data set used previously. One can see that as incidents occur or the system becomes more heavily loaded, the 5th percentile travel time increases. The first event occurs at about 11:00 a.m., when the travel times abruptly rise and then return to normal. This was an incident. The second event starts at about 3:00 p.m., when the travel times begin to increase in advance of the p.m. peak. In the first 11:00 a.m. instance, notice that all the percentile travel times increase, without any advance warning that they were going to; however, in the 3:00 p.m. instance, the 5th percentile travel time begins to increase (and the standard deviation begins to decrease) well in advance of the changes in the other percentile travel times. The message seems clear. In the case of recurring congestion, the low-percentile travel times (e.g., 5th percentile) and the standard deviation both provide leading indi- cations that a period of congested operation is approaching. The travel times (from day to day) are about to become unreliable in the sense that people will not be able to achieve their desired travel times, and the travel time they experience from one instance to the next will not be the same. However, when an unexpected nonrecurring event such as an incident affects the system’s operation, it does so abruptly, without leading indications (from the 5th percentile travel time or the standard deviation) that condi- tions are about to change. In addition, in the latter case, when the event (especially an

549 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.24. Trends in travel time percentiles and their standard deviations. incident) creates a bottleneck that affects all drivers, then the travel times not only all increase, but they become very consistent. When the event affects only some lanes, and thus only some drivers, the travel times increase, but there remains a significant varia- tion in the travel times achieved. People in the less-affected lanes are able to achieve significantly shorter travel times than those in the more-affected lanes. Finally, one thing remains true in all conditions: the lower-percentile (e.g., 5th percentile) travel times are always affected, either in advance of the full-fledged condition (e.g., when congested operation occurs) or immediately upon its onset (e.g., with a nonrecurring event during otherwise uncongested operation). The implication of these observations is that a test that identifies these periods of unreliable operation can be predicated on the lower-percentile travel times. One test will probably not fit all conditions—most likely such tests need to be tuned to the facilities being observed—but a test based on the lower-percentile travel times is likely to always work. Tests based on reductions in the standard deviation will also work for the recurring event conditions; such tests appear to provide an even earlier warning that conditions are in the process of changing. The results of applying one test are shown in the bottom of Figure D.24. The test involves checking three conditions: the average of the 5th percentile travel time in four successive observations rises above 4.7 minutes; the 5th percentile rises above 5 minutes in the current observation; or the 75th percentile travel time rises above 6 minutes. The latter test seems to catch instances in which some, but not all, of the lanes are affected by an incident. 0 2 4 6 8 10 12 10:00 12:00 14:00 16:00 18:00 Tr av el T im e Pe rc en ti le o r S ta nd ar d De vi ati on (m in ) Time on March 1, 2011 tAvg t(5) t(25) t(50) t(75) t(95) 10*StdDev R-Flag

550 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Compare a Recent Adverse Condition with Prior Ones (MM3) In this use case, the roadway manager wants to see if the agency did a better job of handling a recent adverse condition than it has in the past with similar adverse condi- tions. The objective is to compare the reliability of the system in the recent event with similar events in the past. Perhaps the agency’s response was different. This helps the agency decide whether the actions taken were sufficient, or if more could be done to reduce the traffic impacts. In rural areas, this information can help operators provide more accurate pretrip information in inclement conditions. Table D.30 presents the key question, steps, inputs, and the result for this use case. Although data were not available for before-and-after situations in which agency actions were involved, it is still possible to address this use case by examining some of the Bluetooth data collected. Those data allow a comparison of system performance under different occurrences of the same or similar adverse conditions. For the purposes of MM3, the data set for trips from South Lake Tahoe to Placerville, the same data set that was used in Use Case AP6, is useful. As Figure D.15 indicates, the travel time increased on this route several times in response to bad weather conditions. These conditions can be compared and contrasted to see similari- ties and differences. As Figure D.16 indicates, heavy snow has a dramatic effect on the distribution of travel times. The snows that occurred during the observation period increased the mini- mum travel time from 50 to 70 minutes, the median travel time from 55 to 140 minutes, and the 90th percentile travel time from 70 to 210 minutes. There were six heavy snowstorms during the period of observation: one on Janu- ary 29; a second on February 15, a third on February 24, and three more on March 18, 19, and 24. The impacts of these storms were felt across the following times: TABLE D.30. COMPARE A RECENT ADVERSE CONDITION WITH PRIOR ONES (MM3) User Roadway system manager Question How do the reliability impacts from a recent adverse condition compare with prior, similar adverse conditions? Steps 1. Select the network area of interest. 2. Identify the recent adverse condition of interest. 3. Assemble a database of historical normal and adverse conditions for segments in the study area that have been adversely affected. 4. Compare the impacts of the recent event with the impacts of prior occurrences. Inputs Historical TT-PDFs for the arterial segments in the study area for the recent condition and similar prior adverse conditions, as well as normal ones. Result One or more figures comparing the TT-CDFs for the recent adverse condition with other similar events in the past and the normal conditions.

551 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY • January 29 at 9:30 p.m. to January 30 at 9:00 p.m.; • February 15 at 6:00 p.m. to February 18 at 6:00 p.m.; • February 24 at noon to February 27 at midnight; • March 18 at noon to March 19 at 3:00 p.m.; • March 19 at 3:00 p.m. to March 22 at 6:00 p.m.; and • March 24 at midnight to March 25 at 3:00 p.m. Figure D.25 presents the TT-CDFs for the six storms. As can be seen, the March 19 event produced the most significant effect on the travel times: the 90th percentile reached 160 minutes. By comparison, for the other heavy snows, the 90th percen- tile reached 140 minutes on March 18 and between 110 and 120 minutes for the other four storms. These values are all in contrast with the 90th percentile value of 90 min- utes for normal conditions. These findings could be summarized by saying that heavy snow adds up to 50 minutes for 90% of the trips, and only 10% of the trips manage to complete the trip with an additional 20 minutes or less (50 versus 70 minutes). Figure D.25. CDFs for individual vehicle travel times in several regimes. 97 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx [Insert Figure D.25] [caption] Figure D.25. CDFs for individual vehicle travel times in several regimes. These findings could be summarized by saying that heavy snow adds up to 50 minutes for 90% of the trips, and only 10% of the trips manage to complete the trip with an additional 20 minutes or less (50 versus 70 minutes). <H3>Gauge the Impacts of New Arterial Management Strategies (MM4) The user wants to gauge the effectiveness of new arterial management strategies in terms of travel times and travel time variability. The analysis compares before-and-after conditions. This supports the analysis of changes like signal timing updates, new access management policies,

552 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Gauge the Impacts of New Arterial Management Strategies (MM4) The user wants to gauge the effectiveness of new arterial management strategies in terms of travel times and travel time variability. The analysis compares before-and- after conditions. This supports the analysis of changes like signal timing updates, new access management policies, and geometric modifications. Table D.31 presents the key question, steps, inputs, and the result for this use case. No data were collected during the case studies that could be used to address this use case. However, the analysis procedure is effectively identical to that used in Use Cases AE4 and AE5. The impacts of such changes could also be assessed using a simu- lation model. Gauge the Impacts of New Freeway Management Strategies (MM5) The user wants to gauge the effectiveness of new freeway management strategies in terms of travel times and travel time variability. The analysis compares before-and- after conditions. This supports the analysis of the impacts of strategies such as ramp- metering rate changes, geometric modifications, speed limit updates, or HOV lanes. Table D.32 presents the key question, steps, inputs, and the result for this use case. TABLE D.31. GAUGE THE IMPACTS OF NEW ARTERIAL MANAGEMENT STRATEGIES (MM4) User Roadway system manager Question Has a new arterial management strategy improved travel time reliability? Steps 1. Select the arterial segments of interest. 2. Identify the before-and-after conditions. 3. Assemble a database of segment and route TT-PDFs for the before-and-after conditions. 4. Analyze the differences in the TT-PDFs. Inputs Historical TT-PDFs for arterial segments and routes in the study area for the before-and-after conditions, minimizing external factors like adverse conditions that might create differences due to other reasons. Result A figure comparing the TT-CDFs for the before-and-after conditions for the segments and routes of interest, with an overall assessment of the impact. TABLE D.32. GAUGE THE IMPACTS OF NEW FREEWAY MANAGEMENT STRATEGIES (MM5) User Roadway system manager Question Has a new freeway management strategy improved travel time reliability? Steps 1. Select the freeway segments of interest. 2. Identify the before-and-after conditions. 3. Assemble a database of segment and route TT-PDFs for the before-and-after conditions. 4. Analyze the differences in the TT-PDFs. Inputs Historical TT-PDFs for freeway segments and routes in the study area for the before-and-after conditions, minimizing external factors like adverse conditions that might create differences due to other reasons. Result A figure comparing the TT-CDFs for the before-and-after conditions for the segments and routes of interest, with an overall assessment of the impact.

553 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY No data were collected during the case studies that could be used to address this use case. However, the analysis procedure is effectively identical to that used in Use Cases AE4 and AE5. The impacts of such changes could also be assessed using a simu- lation model. Determine Pricing Levels Using Reliability Data (MM6) In this use case, the user wants to see what the pricing levels should be on HOT lanes. The hypothesis is that better and more reliable travel times have value, and people will pay for better service. The question is: How much will they pay? Table D.33 presents the key question, steps, inputs, and the result for this use case. No data were obtained during the project that could be used to provide an exam- ple of this use case. However, the procedure would be similar to that used in Use Cases AE4 and AE5, in which the focus was on changes in system performance over time. In this case, however, the emphasis would be on performance without and with the HOT lane instead of without and with the improvements made to the facility. Drivers with Constrained Trips The driver-related use cases fall into two categories: constrained and unconstrained. This section deals with the constrained trips. A trip is constrained if it has a specific time of arrival. Hence, early and late have definite meanings. Examples of constrained trips include trips for doctor’s appointments, scheduled deliveries, and scheduled bus stops. Unconstrained trips have no particular arrival time. Some constrained trips are made frequently; others, infrequently. In the case of the frequent ones, the driver has a sense of how long the trip should take and how much the trip time should vary due to congestion and other factors. In the case of the infrequent ones, the driver may not have a sense of the travel time, but needs to be on time nevertheless. TABLE D.33. DETERMINE PRICING LEVELS USING RELIABILITY DATA (MM6) User Roadway system manager Question What toll can be charged on HOT lanes given the comparative reliability of those lanes versus the mixed-use lanes? Steps 1. Identify the route and conditions under study. 2. Assemble the TT-PDFs for the HOT lanes and the mixed-use lanes. 3. Compare the TT-PDFs and determine the fee that can be charged given the differential TT-PDFs that exist for the various conditions. 4. Implement the HOT lane fee structure. Inputs Historical TT-PDFs for the routes and conditions under study both in the HOT lanes and the mixed-use lanes. Econometric information about the prices people are willing to pay for lower travel times and better travel time reliability (two separate concepts). Result A pricing guide for different conditions on the routes studied.

554 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Understand Departure Times and Routes for a Trip (MC1) A driver wants to understand in advance the travel times and routing options for a trip that will be made frequently. This is something the driver might do if she or he has a new job or has selected a new day care center. The trip is not to be made right away, but it will be soon, and the driver wants to know how much time to allow and what route (or routes) to choose. The answer is obtained by analyzing historical data, in- cluding all conditions under which the trip might have been made, including inclement weather, incidents, and so forth, because the user wants to know what to do depend- ing on the network conditions that exist. Table D.34 presents the key question, steps, inputs, and the result for this use case. Step 1 involves selecting the origin and destination. In this use case, the same origin and destination as in Figure D.1 are used, which draws on the three routes in San Diego. Step 2 involves selecting the desired arrival time. In this specific case, most of the time, the trip is going to be made between 3:00 and 7:00 p.m. It is worth looking at some details on how individual travel times can be synthesized from 5-minute mean travel times; the subsequent paragraphs provide that description. Most traffic management centers collect mean travel times every 5 minutes from system detectors. These inputs provide insight into the average operating conditions of the facility, but they do not provide sufficient information for understanding varia- tions in individual driver travel times. Automated vehicle identification and automated vehicle location (AVL) are useful in answering questions pertaining to individual driver travel times, but these data sources are not very common. However, it is possible to TABLE D.34. UNDERSTAND DEPARTURE TIMES AND ROUTES FOR A TRIP (MC1) User Driver Question What departure times are needed and what routes should be selected in order to arrive on time given past TT-PDFs for this trip? Steps 1. Select the origin and destination. 2. Select the desired arrival time. 3. Decide what being on time (probability of being late) means. 4. Analyze the TT-PDFs to identify options for departure times and routes. 5. Create a table that shows how much time to allow and what route to select depending on the network conditions that might exist. Inputs Historical TT-PDFs for the routes and departure times that are logical based on the conditions that might exist. Result Table of routes and departure times to select depending on the conditions that exist. The table is created by analyzing the TT-PDFs for the various routes and network conditions.

555 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY synthesize individual vehicle travel times from 5-minute mean travel times obtained from loop detector data by using automatic vehicle identification data. In this case, Bluetooth data collected in Sacramento were used. The data collection involved a 5-mile section on I-5 in Sacramento. The location is depicted in Figure D.17. Data were collected in the northbound and southbound direc- tions for approximately three months (January to April) in 2010. Bluetooth readers were stationed at four locations as shown in Figure D.17. Every tag read in the data set has an encrypted media access control address, the tag reader ID at both origin and destination, and time stamps associated with them. There were around 150,000 tag reads in each direction. About 110,000 tag reads remained after removing out- liers. Since nonrecurring events like weather and incidents have a significant impact on travel times experienced by individual drivers, data were further classified based on nonrecurring event type (normal, weather, and incidents). The O–D pair data were first sorted on the basis of chronological order (i.e., begin- ning with origin time). In the first pass, which was based on the 50 most recent trip rate observations, trip rates associated with each percentile and the mean trip rate associated with that set of observations were computed, and the ratio of each percen- tile travel rate to that of mean travel rate was computed. The data were stored in bins based on mean travel rates. Last, ratios to the mean were computed for each percentile travel rate. These ratios were used in conjunction with the average travel rates from the sys- tem detectors (in this case for the three routes in San Diego) to synthesize estimates of the travel rates (and then the travel times) for vehicles that would have represented each percentile (101 values) in the distribution for each 5-minute observation. Step 3 involves deciding what being on time (probability of being late) means. In this specific use case the assumption is that the driver wants to know how wide the arrival time window is (in time) to encompass 90% of the arrivals. In other words, his or her on-time arrival window is 90%. Step 4 involves assembling the data and creating TT-CDFs for each operating condition for each route. The analysis presented here is a two-step process. First, syn- thesized individual vehicle travel time data between 15:00 and 19:00 were used to gen- erate TT-CDFs. These CDFs are for all percentile drivers. Figure D.26 presents CDF plots for each route for each regime. Under normal operating conditions, SR-163 and I-5 have almost identical 90th percentile travel times (22 minutes); SR-15 performs marginally better with a travel time of about 19 minutes. However, under incident, weather, and special events, SR-163 performs consistently better than the other two routes. It seems that travel times grow significantly higher in the cases of weather events. Table D.35 summarizes these results.

556 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.26. Travel time CDFs for three routes in San Diego. 107 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.26. Travel time CDFs fo r three routes in San Diego. Table D.35. 90th Percentile Travel Times for Each Route Under Various Conditions TT CDFs between 15:00–19:00 on CA-163 TT CDFs between 15:00–19:00 on I-5 TT CDFs between 15:00–19:00 on I-15

557 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.35. 90TH PERCENTILE TRAVEL TIMES FOR EACH ROUTE UNDER VARIOUS CONDITIONS Route 90th Percentile Travel Time (min) Normal Demand Incidents Weather Special Events SR-163 21.48 29.65 31.42 34.82 32.07 I-5 21.55 28.48 32.68 39.98 32.28 SR-15 19.38 29.62 35.73 41.35 39.42 The analysis presented above does not take into account the risk receptiveness of individual drivers; that is, some drivers are more or less aggressive than others. For example, one can and should focus on specific classes of drivers (classified here by aggressiveness) to provide meaningful guidance. In this case, three categories were employed: more aggressive drivers, whose travel times are centered on 10th percentile travel times; median drivers, whose travel times are centered on 50th percentile travel times; and conservative drivers, whose travel times are centered on 90th percentile travel times. This leads to a second analysis for guidance based on driver type. Without loss of generality, the discussion here is based on the aggressive drivers (those with travel times centered on 10th percentile travel times). It is assumed that the distribution of travel times for any given driver type is normally distributed. For example, times for the aggressive driver class are normally distributed between the 5th and 15th percen- tiles, with a mean equivalent to the 10th percentile travel time. Based on this frame- work, travel times for 100 drivers were synthesized for each 5-minute travel time observation from the system detectors. Figure D.27 presents the same corridor conditions as shown in Figure D.26 but for the 10th percentile (aggressive) drivers. Although the choice of best routes has not changed for various nonrecurring conditions, as expected, the 90th percentile travel times are lower for these drivers. Under normal operating conditions, aggressive drivers have travel times about 2 minutes shorter on all three routes. Under demand and incident conditions, they are about 4 to 5 minutes shorter. Although there was not a significant drop in travel times on SR-163 and I-5 under weather and special event conditions, aggressive drivers had a travel time on SR-15 about 7 minutes shorter com- pared with the overall driver population. Table D.36 summarizes these results. Step 5 involves creating a table that shows what time to allow and what route to select depending on the network conditions that might exist.

558 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.27. Travel time CDFs for aggressive drivers. 110 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.27. Travel time CDFs for aggressive drivers. TT CDFs between 15:00–19:00 on CA-163 (Agg drivers) TT CDFs between 15:00–19:00 on I-5 (Agg drivers) TT CDFs between 15:00–19:00 on I-15 (Agg drivers)

559 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.36. 90TH PERCENTILE TRAVEL TIMES FOR AGGRESSIVE DRIVERS FOR EACH ROUTE UNDER VARIOUS CONDITIONS Route 90th Percentile Travel Time (min) Normal Demand Incidents Weather Special Events SR-163 19.05 25.57 26.83 31.00 28.52 I-5 19.23 25.03 27.43 34.90 29.05 SR-15 17.25 25.40 30.85 34.05 32.57 Determine a Departure Time and Route Just Before a Trip (MC2) In this use case, the driver wants to know when to leave and what route to take shortly before making a trip in order to arrive on time. The driver needs to have predictions of what the travel times are likely to be on the routes he or she is most likely to select. Table D.37 presents the key question, steps, inputs, and the result for this use case. Step 1 involves selecting the origin and destination. In this use case, the ones shown in Figure D.1 are again selected. Step 2 involves selecting the desired arrival time. In this specific case, the driver wants to make the trip between 3:00 and 7:00 p.m., wants to know when to leave, and wants to know the likelihood of arriving at the destination on time with 90% certainty (in other words, with a 90% probability of on-time arrival). Step 3 involves assembling the data and creating TT-CDFs for the normal operat- ing condition for each route. The driver wants to know the best route and best time to make the trip. The analysis presented here attempts to provide this information in the following way: the 4-hour time window of interest is divided into eight half-hour windows (3:00 to 3:30 p.m., 3:30 to 4:00 p.m., and so forth). For the purpose of this illustration, the expected arrival time of the driver is the end of each half-hour period (3:30 p.m., 4:00 p.m., and so forth). In preparation for Step 4, the question can be TABLE D.37. DETERMINE A DEPARTURE TIME AND ROUTE JUST BEFORE A TRIP (MC2) User Driver Question Just before making a trip, when should the driver leave and what route should he or she take to arrive at a destination on time? Steps 1. Select the origin and destination. 2. Decide what being on time (probability of being late) means. 3. Obtain predictions of the TT-PDFs for routes that might be selected. 4. Determine the options for departure times and routes. 5. Select the departure time and route that minimizes the travel time but assures an on-time arrival. Inputs Forecasts of TT-PDFs by departure time for the routes that might be chosen given the current and anticipated network conditions. Result Overlapping plots of the TT-CDFs for the various routes so that the distributions of their travel times can be compared, and the one with the shortest travel time at a particular probability can be selected.

560 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY reformulated as follows: Which among the eight expected arrival times of interest will minimize the driver’s travel time, and on which route? Two analyses are presented for answering this question. The first uses synthesized individual vehicle travel time data between 3:00 and 7:00 p.m. and generates TT-CDFs for eight half-hours of interest. These CDFs provide guidance about when the driver should leave in order to arrive within a specific half-hour of interest. These CDFs are for all percentile drivers. Figure D.28 presents CDF plots for each route for the normal operating condition. The TT-CDFs for the first and eighth half-hour time windows for SR-163 are at the left, suggesting that those time windows have reliable travel times. Travel times from the second to the sixth half-hour windows get increasingly worse; operating conditions seem to improve in the seventh half-hour period, and the facility seems to resume normal operating conditions during the eighth half-hour period. The situation is similar for both I-5 and SR-15, although the I-5 TT-CDF for the seventh half-hour period is significantly different. Step 5 involves creating a table that shows how much time to allow and what route to select depending on the network conditions that might exist. Table D.38 shows the 90th percentile travel times for various operating conditions for these three routes. The 90th percentile travel time is minimum (18.43 minutes) for the first half-hour window of interest (3:00 to 3:30 p.m.). Travel time for the second to sixth half-hours of inter- est (3:30 to 4:00 p.m. through 5:30 to 6:00 p.m.) has a positive slope, meaning it was constantly increasing from one time window to the next. The 90th percentile travel time was 19.62 minutes for the second half-hour window and 23.82 minutes for the sixth half-hour window. Operating conditions seem to improve, as travel time for the seventh half-hour window (6:00 to 6:30 p.m.) was less than that of the sixth. The story is similar for both I-5 and SR-15. According to Table D.38, the best time to leave is 3:00 to 3:30 p.m., and the best route is SR-15.

561 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.28. Half-hour TT-CDFs for three routes in San Diego under the normal condition. 115 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.28. Half-hour TT-CDFs for three routes in San Diego under the normal condition.

562 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY The analysis above considers all drivers. A second analysis was conducted to see what the distribution of travel times would be for aggressive (10th percentile) drivers. Figure D.29 presents CDF plots for each route for aggressive drivers for the normal operating condition. Although the trends in travel times in these plots at first seem sim- ilar to those presented in Figure D.28, they are different. Shorter travel times are rea- sonably expected of the aggressive driver population. The SR-15 TT-CDFs for most of the half-hour time periods are at the left, suggesting that most of the aggressive driver population has a tight distribution of travel times; there is more variation in travel times on I-5, and SR-163 shows the most variation. The 90th percentile travel time in any half-hour time period is about 2 to 3 minutes shorter than that for the overall driver population. Table D.39 summarizes values of 90th percentile travel times for these three routes for aggressive drivers. TABLE D.38. 90TH PERCENTILE DEPARTURE TIMES AND TRAVEL TIMES CORRESPONDING TO DIFFERENT ARRIVAL TIMES FOR THREE ROUTES Route Travel and Departure Times Expected Arrival Time 3:30 p.m. 4:00 p.m. 4:30 p.m. 5:00 p.m. 5:30 p.m. 6:00 p.m. 6:30 p.m. 7:00 p.m. SR-163 Travel time (min) 18.43 19.62 20.85 21.50 21.85 23.82 23.37 19.35 Departure time 3:11 p.m. 3:40 p.m. 4:09 p.m. 4:38 p.m. 5:08 p.m. 5:36 p.m. 6:06 p.m. 6:40 p.m. I-5 Travel time (min) 17.58 18.68 20.25 21.90 22.78 24.43 21.87 18.28 Departure time 3:12 p.m. 3:41 p.m. 4:09 p.m. 4:38 p.m. 5:07 p.m. 5:35 p.m. 6:08 p.m. 6:41 p.m. SR-15 Travel time (min) 16.45 17.18 18.70 19.98 19.88 22.10 20.75 17.10 Departure time 3:13 p.m. 3:42 p.m. 4:11 p.m. 4:40 p.m. 5:10 p.m. 5:37 p.m. 6:09 p.m. 6:42 p.m.

563 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.29. Half-hour TT-CDFs for aggressive drivers for three routes in San Diego under the normal condition. 118 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.29. Half-hour TT-CDFs for aggressive drivers for three routes in San Diego under the normal condition.

564 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Understand the Extra Time Needed for a Trip (MC3) This is a variant on the prior two use cases. The question is this: How much extra time is needed to make an on-time arrival? Implicitly, this use case assumes that the driver has a sense of how long the trip should take. In the example presented here, it is as- sumed that the driver has queried a static path-choice application that does not pay at- tention to time-of-day or real-time conditions, and he or she wants to know how much extra time is needed for the time of departure given the actual, current conditions. Table D.40 presents the key question, steps, inputs, and the result for this use case. TABLE D.39. 90TH PERCENTILE DEPARTURE TIMES AND TRAVEL TIMES CORRESPONDING TO DIFFERENT ARRIVAL TIMES FOR AGGRESSIVE DRIVERS Route Travel and Departure Times Expected Arrival Time 3:30 p.m. 4:00 p.m. 4:30 p.m. 5:00 p.m. 5:30 p.m. 6:00 p.m. 6:30 p.m. 7:00 p.m. SR-163 Travel time (min) 16.30 17.52 18.55 18.97 19.18 21.18 20.38 16.77 Departure time 3:13 p.m. 3:42 p.m. 4:11 p.m. 4:41 p.m. 5:10 p.m. 5:38 p.m. 6:09 p.m. 6:43 p.m. I-5 Travel time (min) 14.80 16.53 18.18 19.43 20.23 21.72 18.95 16.03 Departure time 3:15 p.m. 3:43 p.m. 4:11 p.m. 4:40 p.m. 5:09 p.m. 5:38 p.m. 6:11 p.m. 6:43 p.m. SR-15 Travel time (min) 14.25 15.05 16.55 17.92 17.82 19.67 18.02 14.92 Departure time 3:15 p.m. 3:44 p.m. 4:13 p.m. 4:42 p.m. 5:12 p.m. 5:40 p.m. 6:11 p.m. 6:45 p.m. TABLE D.40. UNDERSTAND THE EXTRA TIME NEEDED FOR A TRIP (MC3) User Driver Question How much extra time needs to be allowed for a trip in order to arrive on time? Steps 1. Select the origin and destination. 2. Identify when the trip will be made and the arrival time of interest. 3. Decide what being on time (probability of being late) means. 4. Determine the options for departure times and routes. 5. Develop the distribution of extra time needed for the trip. Inputs Historical, individual vehicle TT-PDFs for the routes and departure times that might be chosen based on network conditions. Result Overlapping plots of the TT-CDFs for the various routes and departure times (including the distribution corresponding to the assumed travel time) so that the distributions of their travel times can be compared, and the extra travel time needed for the various options can be understood.

565 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin and destination. In this use case, the ones shown in Figure D.1 are selected. Step 2 involves selecting the desired arrival time. In this example, the driver wants to make the trip between 5:00 to 5:30 p.m., and he or she wants to know when to leave and the likelihood of arriving at the desired destination no later than 5:30 p.m. Step 3 involves deciding what being on time (probability of being late) means. In this specific use case the driver wants to know the likelihood of arriving at the destina- tion on time with 90% certainty, in other words, with a 90% probability of on-time arrival. Step 4 involves assembling the data, and Step 5 involves creating a distribution of extra time for the normal operating conditions for each route. The driver has a general idea of the trip’s duration under normal conditions, but wants to have a sense of the extra time that might be required under other conditions. For the purposes of representing this scenario, two half-hour time periods (3:00 to 3:30 p.m. and 5:00 to 5:30 p.m.) for normal operating conditions were chosen. Travel time distribution in the first half-hour period (3:00 to 3:30 p.m.) represents the driver’s perception of how long the trip would take under normal conditions; travel time distribution in the second half-hour period (5:00 to 5:30 p.m.) represents actual travel times on that route under current conditions. The TT-CDFs for the normal and current conditions are obviously different, but they can be used to compute the distribution of the extra time needed. One way to compute this distribution is to sample TT-CDFs, calculate the algebraic difference in travel times, and look at the distribution of differences in travel times. In this use case, TT-CDFs were sampled 10,000 times, and a CDF was created based on these differences in travel times. Figure D.30 presents the distribution of extra time needed for the three routes. If the driver is making a trip on SR-163, he or she needs to add a cushion time of about 6.5 minutes to ensure on-time arrival 90% of times; that value is about 6 minutes for both SR-15 and I-5. Figure D.31 presents these plots again for the aggressive driver population.

566 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Decide How to Compensate for an Adverse Condition (MC4) A driver wants to know how much extra time to allow so that he or she arrives at an event on time given that the event will create congestion, or wants to be aware of the changes to the travel time because of an incident, bad weather, special event, or lane closure for a trip she or he plans to take. This information helps the driver plan for a nonrecurrent scenario and determine how much extra time to allow when such an event occurs. Table D.41 presents the key question, steps, inputs, and the result for this use case. Figure D.31. Distribution of extra time needed for aggressive drivers. 122 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx on-time arrival 90% of times; that value is about 6 minutes for both SR-15 and I-5. Figure D.31 presents these plots again for the aggressive driver population. [Insert Figure D.30] [caption] Figure D.30. Distribution of extra time needed. [Insert Figure D.31] [caption] Figure D.30. Distribution of extra time needed. 122 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx on-time arrival 90% of times; that value is about 6 minutes for both SR-15 and I-5. Figure D.31 presents these plots again for the aggressive driver population. [Insert Figure D.30] [caption] Figure D.30. Distribution of extra time needed. [Insert Figure D.31] [caption]

567 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin and destination. In this use case, the ones shown in Figure D.1 are selected. Step 2 involves selecting the desired arrival time. In this specific case, the driver wants to make the trip between 5:00 and 5:30 p.m., and he or she wants to know when to leave and the likelihood of arriving at the destination by no later than 5:30 p.m. Step 3 involves deciding what being on time (probability of being late) means. In this specific use case the driver wants to know the likelihood of arriving at the destina- tion on time with 90% certainty, in other words, with a 90% probability of on-time arrival. Steps 4 through 8 involve assembling the data and creating a distribution of extra time for each operating condition for each route. The procedure used for generating extra time distributions is the same as that used in Use Case MC3. Figure D.32 rep- resents CDF plots for extra time needed for these three routes for each regime. As the figure shows, for SR-163 and I-5 weather has the worst effect on the distribution of extra time needed. Incidents, special events, and demand (in that order) have the next most significant impact in extra time distribution on those two routes. For SR-15, special events have the worst effect on extra time distribution; after special events, weather and incidents (in that order) have significant impacts on the extra time distri- bution. Table D.42 summarizes these results for the three routes, and Figure D.33 and Table D.43 present these results for aggressive drivers. TABLE D.41. DECIDE HOW TO COMPENSATE FOR AN ADVERSE CONDITION (MC4) User Driver Question For the route the driver plans to use, how much time should be allowed to compensate for the congestion that will be caused by an adverse condition? Steps 1. Select the origin, destination, and route. 2. Decide what being on time (probability of being late) means. 3. Determine what travel time is expected. 4. Select the condition that will exist just before an adverse condition occurs (e.g., peak congestion). 5. Select the adverse condition (e.g., incident, bad weather, special event). 6. Examine historical TT-PDFs for the adverse condition. 7. Compare these TT-PDFs with the one for the condition on which the expected travel time is based. 8. Identify the distribution of extra travel time needed to compensate for the adverse condition. Inputs Historical TT-PDFs for the route and adverse condition and TT-PDFs for the condition on which the expected travel time is predicated. Result A PDF for the extra time that should be allowed to ensure that the driver can arrive on time given the adverse condition.

568 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 126 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.32. CDF of distribution of extra time needed for three routes. Table D.42. 90th Percentile Values of Extra Time Needed for Three Routes Figure D.32. CDF representing distribution of extra time needed for three routes.

569 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 128 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.33. CDF of distribution of extra time needed for aggressive drivers for three routes. [COMPOSITION: Please align decimal places in table.] Figure D.33. CDF representing distribution of extra time needed for aggressive drivers for three routes.

570 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.42. 90TH PERCENTILE VALUES OF EXTRA TIME NEEDED FOR THREE ROUTES Route 90th Percentile Extra Travel Time Needed (min) Normal Demand Incidents Weather Special Events SR-163 6.42 10.32 10.92 17.49 15.72 I-5 7.72 10.80 16.15 24.29 12.50 SR-15 5.90 NA 17.70 20.54 23.32 Note: NA = not available. TABLE D.43. 90TH PERCENTILE VALUES OF EXTRA TIME NEEDED FOR AGGRESSIVE DRIV- ERS FOR THREE ROUTES Route 90th Percentile Extra Travel Time Needed (min) Normal Demand Incidents Weather Special Events SR-163 5.10 8.40 9.03 15.44 13.46 I-5 6.43 8.71 13.00 19.12 9.30 SR-15 4.83 NA 13.67 16.40 17.23 Note: NA = not available. Decide En Route Whether to Change Routes (MC5) A driver en route wants to determine whether an alternate route would increase the likelihood of an on-time arrival. At major splits in the roadway, travelers want in- formation that will help them decide whether to stay on their planned route or de- tour to an alternate route. Travelers can receive this information from traditional data dissemination technologies (e.g., variable message signs) or from emergent in-vehicle technologies (e.g., route guidance systems). Although this information is currently dis- tributed in the form of average travel times, it could be improved if augmented with reliability information. Table D.44 presents the key question, steps, inputs, and the result for this use case. TABLE D.44. DECIDE EN ROUTE WHETHER TO CHANGE ROUTES (MC5) User Driver Question Should an alternate route be chosen while en route to increase the likelihood of being on time? Steps 1. Reaffirm the desired arrival time and definition of being on time. 2. Examine TT-PDFs for routes that could be chosen based on the driver’s current location. 3. See if the PDF for one of those routes would provide a better on-time arrival than the route currently being followed. 4. If so, change routes. If not, continue using the current route. Inputs Forecasts of TT-PDFs for alternate routes from the current location to the destination for the current time and the driver’s current location. Result TT-CDFs for the current route and the alternate routes and a choice of the route that is best.

571 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY For the purposes of illustrating this use case, picture the following scenario. In Step 1 assume the driver is traveling on the freeway and at around 5:00 p.m., the driver reaches the junction where the three routes (SR-163, I-5, and SR-15) in San Diego meet. The traveler has to make a route choice decision as the operational regime has changed. To further expand this thought, consider the following two scenarios. In Scenario 1, the driver started the trip on a day with normal operating conditions; however, at around 5:00 p.m. there has been an incident, thus changing the operating conditions. Now, he or she consults a route guidance system for choosing the best alternate route. In Scenario 2, the traveler started the trip under inclement weather conditions. So, obvi- ously he or she would expect travel times to be longer than what they would have been under normal operating conditions. However, at around 5:00 p.m. (the time around which he or she reaches the junction) there is an incident, and therefore the expected travel time is different from when he or she originally started the trip. So, guidance is sought from the route guidance system for choosing the best alternate route. Step 2 involves assembling the data and creating TT-CDFs for the two scenarios presented above. In Scenario 1, the two possible nonrecurring events types are normal and incident; in Scenario 2, they are weather and incident. If the driver started the trip on a normal-condition day, then the probability of a weather event is zero; if the driver started the trip on a day with inclement weather conditions, then the prob- ability of a normal nonrecurring event is zero. Furthermore, possible nonrecurring event types in both scenarios are assumed to be independent: that is, the occurrence or non occurrence of one does not affect the occurrence or nonoccurrence of the other. Since the driver wants to make a path-choice decision at around 5:00 p.m., historical data between 3:30 and 6:30 p.m. were used to generate these CDFs. Using these data, the probability of occurrence of these two events was computed, and a mean TR-CDF for each nonrecurring type was generated. Monte Carlo simulation was used for sam- pling the CDFs. The number of Monte Carlo runs performed was such that 500 data points were ensured for the least-frequent event. Two random variables between zero and one were generated from a uniform distribution. The first variable was used to determine the event type, and the second was used to sample mean travel time for the corresponding nonrecurring event. One hundred individual travel times were synthe- sized for each driver class (aggressive, median, and conservative). Using these data, TT-CDFs were created for the three routes. Figure D.34 represents TT-CDFs for three driver classes for the three routes when operating conditions change from normal to normal with incidents. It can be inferred that SR-15 is the best alternate route to choose for all driver classes. The 90th percen- tile travel time for the aggressive driver class on SR-15 is about 18 minutes, but on both SR-163 and I-5 it is about 20 minutes. For the median driver class, it is about 20 minutes on SR-15 and about 22 minutes on both SR-163 and I-5. Finally, for the conservative driver class, the 90th percentile travel time is about 22 minutes on SR-15 and about 24 to 25 minutes on SR-163 and I-5. These results are summarized in Table D.45.

572 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 133 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.34. CDF representing distribution of travel times when operating conditions change from normal to normal with incident for various driver classes for three routes. TT CDFs for aggressive drivers on three routes (Normal+Incidents) TT CDFs for median drivers on three routes (Normal+Incidents) TT CDFs for conservative drivers on three routes (Normal+Incidents) Figure D.34. CDF representing distribution of travel times when operating conditions change from normal to normal with incident for various driver classes for three routes.

573 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.45. 90TH PERCENTILE TRAVEL TIME VALUES FOR NORMAL WITH INCIDENT OPERATING CONDITIONS FOR VARIOUS DRIVER CLASSES FOR THREE ROUTES Driver Type 90th Percentile Travel Time Normal with Incident (min) SR-163 I-5 SR-15 Aggressive 19.50 19.87 18.05 Median 21.45 21.83 19.80 Conservative 24.52 24.78 22.57 Figure D.35 represents TT-CDFs for various driver classes for three routes when operating conditions change from weather to weather and incidents. It can be inferred that SR-163 is the best alternate route for all driver classes. The 90th percentile travel time for the aggressive driver class on SR-163 is about 31 minutes, but on both SR-15 and I-5 it is about 36 to 37 minutes; for the median driver class it is about 34 minutes on SR-163 and about 40 minutes on both SR-15 and I-5; finally, for the conservative driver class it is about 39 minutes on SR-163 and about 44 to 45 minutes on SR-15 and I-5. These results are summarized in Table D.46. It can be inferred from these TT-CDFs that the impact of an incident on travel times is significantly higher on days with inclement weather than on normal days.

574 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 135 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.35. CDF representing distribution of travel times when operating conditions change from weather to weather with incident for various driver classes for three routes. TT CDFs for aggressive drivers on three routes (Weather+Incidents) TT CDFs for median drivers on three routes (Weather+Incidents) TT CDFs for conservative drivers on three routes (Weather+Incidents) Figure D.35. CDF representing distribution of travel times when operating conditions change from weather to weather with incident for various driver classes for three routes.

575 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.46. 90TH PERCENTILE TRAVEL TIMES WHEN OPERATING CONDITIONS CHANGE FROM WEATHER TO WEATHER WITH INCIDENT FOR VARIOUS DRIVER CLASSES FOR THREE ROUTES Driver Type 90th Percentile Travel Time Combined Weather and Incident (min) SR-163 I-5 SR-15 Aggressive 31.18 36.18 36.68 Median 34.08 39.83 40.27 Conservative 38.88 44.98 45.45 Drivers with Unconstrained Trips Unconstrained trips have no particular arrival time against which a measure of sched- ule delay can be calculated. Consistency becomes the focus. Examples of unconstrained trips include shopping trips and visits to zoos and museums. The main question posed is: When should the trip departure time occur so that the trip has the most reliable travel time? Determine the Best Time of Day to Make Trip (MU1) In this use case, the driver has a range of times when the trip could be made and wants to know what time is best: that is, what departure time has the most reliable travel time? Table D.47 presents the key question, steps, inputs, and the result for this use case. TABLE D.47. DETERMINE THE BEST TIME OF DAY TO MAKE TRIP (MU1) User Driver Question When should a trip be made so that it has the most reliable travel time? Steps 1. Select the origin and destination. 2. Decide what is meant by a reliable trip time (e.g., probability of being late). 3. Assemble TT-PDFs for the times and conditions when the trip might be made (e.g., during the midday on days without adverse conditions). 4. Find the departure time and route that provides the most reliable travel time. Inputs Historical TT-PDFs for the departure times that might be selected and plausible routes given those departure times. Result A set of TT-CDFs for the various departure times and routes that could be chosen and a diagram that identifies the best of these (i.e., the best time to make the trip and the best route to use).

576 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin and destination. In this use case, the ones shown in Figure D.1 are selected. Step 2 involves selecting the desired arrival time. In this specific case, the driver wants to make the trip between 3:00 and 7:00 p.m., and he or she wants to know when to leave and the likelihood of arriving at the destination on time with 90% cer- tainty (i.e., a 90% probability of an on-time arrival). Step 3 involves assembling the data and creating TT-CDFs for a given operating condition for each route. In this use case two nonrecurring event conditions were con- sidered: normal and weather. It was further assumed that the driver is in the aggres- sive driver class. The driver wants to make the trip between 3:00 and 7:00 p.m. and wants to know the best route and best time to make the trip. The methodology used in developing TT-CDFs here is similar to the one used in Use Case MC2, and details of the methodology are omitted. Figure D.36 presents TT-CDF plots for each route for the normal operating condi- tion. The SR-15 TT-CDFs for most of the half-hour time periods are at the left, sug- gesting a tight distribution of travel times; variation in travel times is more obvious for I-5, and most obvious for SR-163. TT-CDFs for the first and eighth half-hour time windows are at the left, suggesting that those time widows have reliable travel times. Travel times from the second to the sixth half-hour windows get increasingly worse; operating conditions seem to improve in the seventh half-hour period, and the facil- ity seems to resume normal operating conditions during the eighth half-hour period. Under normal operating conditions the best time for making the trip is between 3:00 and 3:30 p.m., and the best route to choose is SR-15. Table D.48 summarizes values of the 90th percentile travel times for the three routes.

577 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.36. Half-hour TT-CDFs for three routes in San Diego under the normal condition, for aggressive drivers. 139 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.36. Half-hour TT-CDFs for three routes in San Die der the normal condition.

578 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.37 presents TT-CDF plots for each route under inclement weather con- ditions. It can be inferred that travel times on SR-163 are highly unreliable under bad weather, and that it is best to avoid this route under these conditions. Although 90th percentile travel times under normal operating conditions got increasingly worse between the second and sixth half-hour periods, the story is different under weather conditions. The 90th percentile travel times were increasingly longer between the sec- ond and fifth half-hour periods; the value was reduced for the sixth half-hour period, and it was worst for the seventh half-hour period (about 34 minutes). It can be inferred that TT-CDF trends for both I-5 and SR-15 are also highly unreliable, with the excep- tion of the first half-hour period (3:00 to 3:30 p.m.) on SR-15. Under inclement weather conditions the best time for making the trip is between 3:00 and 3:30 p.m., and the best route to choose is SR-15. Table D.49 summarizes values of 90th percentile travel times for these three routes. TABLE D.48. 90TH PERCENTILE DEPARTURE TIMES AND TRAVEL TIMES CORRESPONDING TO DIFFERENT ARRIVAL TIMES UNDER NORMAL CONDITIONS FOR THREE ROUTES Route Travel and Departure Times Expected Arrival Time 3:30 p.m. 4:00 p.m. 4:30 p.m. 5:00 p.m. 5:30 p.m. 6:00 p.m. 6:30 p.m. 7:00 p.m. SR-163 Travel time (min) 16.30 17.52 18.55 18.97 19.18 21.18 20.38 16.77 Departure time 3:13 p.m. 3:42 p.m. 4:11 p.m. 4:41 p.m. 5:10 p.m. 5:38 p.m. 6:09 p.m. 6:43 p.m. I-5 Travel time (min) 14.80 16.53 18.18 19.43 20.23 21.72 18.95 16.03 Departure time 3:15 p.m. 3:43 p.m. 4:11 p.m. 4:40 p.m. 5:09 p.m. 5:38 p.m. 6:11 p.m. 6:43 p.m. SR-15 Travel time (min) 14.25 15.05 16.55 17.92 17.82 19.67 18.02 14.92 Departure time 3:15 p.m. 3:44 p.m. 4:13 p.m. 4:42 p.m. 5:12 p.m. 5:40 p.m. 6:11 p.m. 6:45 p.m.

579 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.37. Half-hour TT-CDFs for three routes in San Diego under the weather condition, for aggressive drivers. 142 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.37. Half-hour TT-CDFs for three routes in San Diego r the weather condition.

580 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Determine How Much Extra Time Is Needed (MU2) A user wants to know how much extra time to allow so that he or she can arrive on time for a special event (e.g., a baseball game). This use case assumes that the user has some idea how long the trip ought to take. Table D.50 presents the key question, steps, inputs, and the result for this use case. TABLE D.49. 90TH PERCENTILE DEPARTURE TIMES AND TRAVEL TIMES CORRESPONDING TO DIFFERENT ARRIVAL TIMES UNDER WEATHER CONDITIONS FOR THREE ROUTES Route Travel and Departure Times Expected Arrival Time 3:30 p.m. 4:00 p.m. 4:30 p.m. 5:00 p.m. 5:30 p.m. 6:00 p.m. 6:30 p.m. 7:00 p.m. SR-163 Travel time (min) 22.50 26.90 27.42 33.32 31.10 32.22 33.90 29.30 Departure time 3:07 p.m. 3:33 p.m. 4:02 p.m. 4:26 p.m. 4:58 p.m. 5:27 p.m. 5:56 p.m. 6:30 p.m. I-5 Travel time (min) 17.33 21.98 25.12 26.32 28.57 29.90 26.42 18.17 Departure time 3:12 p.m. 3:38 p.m. 4:04 p.m. 4:33 p.m. 5:01 p.m. 5:30 p.m. 6:03 p.m. 6:41 p.m. SR-15 Travel time (min) 16.55 25.87 27.12 26.90 30.45 33.80 31.27 26.23 Departure time 3:13 p.m. 3:34 p.m. 4:02 p.m. 4:33 p.m. 4:59 p.m. 5:26 p.m. 5:58 p.m. 6:33 p.m. TABLE D.50. DETERMINE HOW MUCH EXTRA TIME IS NEEDED (MU2) User Driver Question How much extra time is needed for a common trip to generally arrive on time? Steps 1. Select the origin and destination. 2. Select the condition(s) under which the trip is to be made. 3. Decide what is meant by arriving on time (e.g., the probability of being late). 4. Assemble TT-PDFs for the time at which the trip is to be made, the routes that might be chosen, and the network conditions that would exist. 5. Develop a PDF of the extra time that might be needed given the TT-PDFs from Step 4 and the probabilities that the trip might be made under the conditions indicated. Inputs Historical TT-PDFs for the departure times that might be selected and the routes that might be used given the network conditions that might exist for those times. Result A PDF of the extra time that should be allowed to arrive on time.

581 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin and destination. In this use case, the ones shown in Figure D.1 are again selected. Step 2 involves selecting the conditions under which the trip is being made. In this specific case, the driver wishes to attend a special event that starts by 7:00 p.m., and hence wants to make the trip between 6:30 and 7:00 p.m. Step 3 involves deciding what being on time (probability of being late) means. In this specific use case the driver wants to know the likelihood of arriving at the destina- tion on time with 90% certainty (i.e., 90% probability of on-time arrival). Step 4 involves assembling the data and creating TT-CDFs for the given operating condition for each route. Because this is a special event, the likelihood of an incident taking place is high. To accommodate for any delay that might be caused by an inci- dent, two nonrecurring event conditions were considered: special events and incidents. It was further assumed that the driver falls into the aggressive driver class. The meth- odology used in developing TT-CDFs here is similar to the one used in Use Case MC5, and details of the methodology are omitted. TRANSIT USE CASES Transit use cases focus on operators and users of transit services. On the supply (ser- vice provider) side, three main emphases exist: service planning (developing bus routes, identifying required bus headways, and considering ways to improve ridership), sched- uling (assigning buses and drivers as efficiently as possible to meet planning, opera- tional, and customer needs), and operations (using real-time and archived information to better manage day-to-day service issues). For these use case examples, San Diego bus system vehicles equipped with an AVL-based tracking system are used. Buses on 10 of the routes were AVL equipped: • Route 6: Fashion Valley–North Park; • Route 7: Downtown–Balboa Park/Zoo–La Mesa; • Route 10: Old Town–University and College; • Route 11: Paradise Valley and Meadowbrook Drive to San Diego State University Transit Center • Route 15: Downtown–San Diego State University; • Route 20: Downtown–Del Lago Transit Station; • Route 41: Fashion Valley–USCD/VA Medical Center; • Route 50: 9th Avenue and C Street to the UTC Transit Center • Route 88: Old Town–Fashion Valley; and • Route 150: Downtown–UTC/VA Express. On some routes, many of the buses had AVL equipment installed, so there was plenty of data; other routes had less. The routes chosen for analysis here had the most data and served O–D pairs that were instructive to analyze.

582 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY The AVL-equipped buses generate the following information for every stop: • Route ID; • Trip ID (based on factors such as the time of departure); • The latitude and longitude of the bus when it stops; • Stop ID of the location where the bus stops (based on the latitude and longitude of the location where it stops); • The distance of the bus from the predefined location of the Stop ID; • The time when the doors open and when they close; • The time when the bus was supposed to stop at the Stop ID given the trip it was on; and • Many other data items not used in this analysis, including the number of people who get on and off the bus and how many are on-board the bus when it leaves the stop. This is a rich data set that can be used extensively and effectively to study the qual- ity of the transit services provided. Transit Planners The following use cases demonstrate system functionalities that are helpful for transit system planners who are responsible for determining where major improvements are needed, where the buses should go, and how much service should be provided. Determine Routes with the Least Travel Time Variability (TP1) A service planner wants to determine which routes have the poorest reliability, that is, which routes have the most variability in on-time performance at the stops. This information is helpful as a supplement to passenger demand analysis when designing express bus routes, as it allows the planner to analyze which routings minimize the average travel time or the travel time variability, or both. Table D.51 presents the key question, steps, inputs, and the result for this use case. TABLE D.51. DETERMINE ROUTES WITH THE LEAST TRAVEL TIME VARIABILITY (TP1) User Transit planner Question Which routes have the least travel time variability? Steps 1. Select the routes and conditions of interest. 2. Assemble PDFs for the deviations from scheduled stop times. 3. Rank the routes based on greatest deviations from the scheduled stop times. Inputs A database of deviations from scheduled stop times by route for the conditions of interest (could be all conditions). Result A rank ordering of the routes based on their deviations from scheduled stop times, from poorest to best performance.

583 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the routes and conditions of interest. In this case, the routes are the nine with AVL-equipped buses, and the time frame is August 2011, the month for which data were collected. Step 2 involves assembling PDFs for the deviations from the scheduled stop times. Figure D.38 presents these PDFs for the 10 routes that had AVL-equipped buses. As the figure shows, in the worst cases the departures can be as much as (or in one case more than) 6 minutes early and as much as 16 minutes late. Route 88 has the best performance, which can be observed in Figure D.38 by com- paring the CDFs and also seen numerically in Table D.52. The standard deviation for Route 88 is only 2.0; all the other routes have higher values. Route 11 has the largest standard deviation. It appears to be buried in the cluster of CDFs in the figure, but in fact it has a significant tail in terms of early departures (negative values), reaching down to 5 minutes early. Figure D.38. Deviations from scheduled stop times by route. 149 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx in the worst cases the departures can be as much as (or in one case more than) 6 minutes early and as much as 16 minutes l te. [Insert Figure D.38] [caption] Figure D.38. Deviations from scheduled stop times by route. Route 88 has the best performance, which can be observed in Figure D.38 by comparing the CDFs and also seen numerically in Table D.52. The standard deviation for Route 88 is only 2.0; all the other routes have higher values. Route 11 has the largest standard deviation. It appears to be buried in the cluster of CDFs in the figure, but in fact it has a significant tail in terms of early departures (negative values), reaching down to 5 minutes early. Table D.52. Standard Deviations for Differences from Scheduled Stop Times by Route

584 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.52. STANDARD DEVIATIONS FOR DIFFERENCES FROM SCHEDULED STOP TIMES BY ROUTE Route Standard Deviation (min) 6 4.49 7 4.02 10 4.63 11 9.37 15 3.24 20 4.07 41 3.49 50 3.01 88 2.00 150 3.62 Step 3 involves ranking the routes based on the reliability of their services. For the nine routes shown in Figure D.39, this means creating a list going from the most- to the least reliable route, effectively from the CDF with the narrowest range of values to the one with the largest. This ranking is reflected in the standard deviations shown in Table D.52. A reasonable ranking of reliable routes based on the values displayed, from the most reliable to the least reliable, is 88, 50, 15, 41, 150, 7, 20, 6, 10, and finally 11. Interestingly, and compared with the standard deviation rankings, the routes that seem to stand out in Figure D.39 are different: Route 88, the one with the narrowest range of values and the smallest deviations at the higher percentiles; Route 15, the one that seems to be latest the most often; Route 11, which apparently has the worst reliability according to Table D.52; Route 7, which has the greatest deviations for the percentile values above 80%; and Route 6, which seems to have a significant percent- age of early departures. These other metrics provide additional ways to extract infor- mation from the CDFs about the relative performance of the routes. Compare Exclusive Bus Lanes with Mixed-Traffic Operations (TP2) A planner wants to compare the reliability of buses traveling in mixed traffic with those operating on exclusive bus lanes. This type of assessment is useful when studying the possibility of moving an existing service to a dedicated lane, such as might happen during the development of a bus rapid transit route. Table D.53 presents the key ques- tion, steps, inputs, and the result for this use case. The bus system in San Diego does not technically have a route that operates on an exclusive bus lane, so this use case cannot be addressed directly. However, a hypotheti- cal analysis can be performed based on, say, the contrast between the performance of Route 88 and Route 7.

585 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY TABLE D.53. COMPARE EXCLUSIVE BUS LANES WITH MIXED-TRAFFIC OPERATIONS (TP2) User Transit planner Question How much would an exclusive bus lane help with reliability? Steps 1. Select the routes and conditions of interest. 2. Assemble PDFs for the deviations from scheduled stop times for routes that do and do not have exclusive bus lanes. If there are no routes with exclusive bus lanes, assemble data for those routes that operate in the least-congested conditions, or create a simulation model. 3. Assess the reduction in deviations from scheduled stop times that can be achieved by having the buses operate in their own lane. Inputs A database of deviations from scheduled stop times for routes with and without exclusive bus lanes. If none exists, use data for routes that operate in the least-congested conditions. Result An assessment of the extent to which on-time performance is improved by exclusive bus lanes. Figure D.39. Distributions of the deviation from scheduled stop times for Route 88 versus Route 7. 154 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.39. Distributions of the deviation from scheduled stop times for Route 88 versus Route 7. <H2>Transit Scheduler The following use cases illustrate how the monit ring system can help transit s hedulers, who need to know what schedules are feasible, what reliability they can have, and whether existing schedules need to be adjusted so that advertised stop times can be achieved. <H3>Acquire Reliability Data for Building Schedules (TS1) A scheduler wants to determine the amount of recovery time to include in an existing bus route’s schedule to ensure that most late-arriving buses will be able to depart for their next trip on time. The steps in the analysis process are as follows. Step 1 is to select the routes and con- ditions of interest. In this instance, August data for the nine routes with AVL-equipped buses will be used. Step 2 is to assemble PDFs for the deviations from scheduled stop times for routes that do and do not have exclusive bus lanes. If there are no routes with exclusive bus lanes, assemble data for those routes that operate in the least-congested conditions, or create a simulation model. In this instance, the best- performing bus route will be used as the basis for comparison. Step 3 involves assessing the reduction in deviations from scheduled stop times that can be achieved by having the buses oper- ate in their own lane.

586 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY As Figure D.39 shows, Route 88 performs much better than Route 7. At the 90th percentile, the buses on Route 7 are as much as 8 minutes late leaving their stops, but the buses on Route 88 are only 3 minutes late. In addition, the buses on Route 7 are on rare occasions more than 3 minutes early leaving their stops, but the buses on Route 88 are never more than slightly over 2 minutes early. If the Route 88 buses were operating on a busway and the Route 7 buses were in mixed traffic, there would be evidence that the busway helped to improve on-time performance. Transit Schedulers The following use cases illustrate how the monitoring system can help transit sched- ulers, who need to know what schedules are feasible, what reliability they can have, and whether existing schedules need to be adjusted so that advertised stop times can be achieved. Acquire Reliability Data for Building Schedules (TS1) A scheduler wants to determine the amount of recovery time to include in an existing bus route’s schedule to ensure that most late-arriving buses will be able to depart for their next trip on time. This is a basic scheduling activity. The total round-trip time (or cycle time) on a bus route typically is based on four elements (1): • The average round-trip running time for the route; • The minimum break time (layover time) for drivers required by policy or contract; • Any additional recovery time (ideally zero) required (the greater of zero or the round-trip buffer time minus the layover time); and • Any additional time (ideally zero) required to achieve a desired headway. This use case helps schedulers to build timetables based on the variability associ- ated with roadway and environmental conditions, as well as with passenger demand or any other variability specifically associated with the provision of transit service. Table D.54 presents the key question, steps, inputs, and the result for this use case.

587 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY This use case addresses an issue that lies at the heart of building good route sched- ules. As with the highway- (auto-) focused use cases, in transit scheduling travel time matters less than the reliability of the travel time. The objective is to build schedules that can actually be followed. Otherwise, the bus drivers get frustrated: no matter how hard they try, they cannot make the stops at the scheduled times. Riders too are frus- trated because they cannot get to their destinations at the times listed in the schedule, and they miss their transfers. An unreliable schedule has no value. A little background about Route 7 is helpful before addressing the use case directly. Route 7 runs between downtown San Diego and La Mesa along Broadway, Park Bou- levard, and University Avenue as shown in the top portion of Figure D.40. The length of the route is about 17.6 miles, and it requires a little more than an hour to traverse end to end. The timetable shows varying stop patterns, as shown in the bottom portion of Figure D.40. There are some even shorter stop patterns, not shown, that seem to be used only when the public schools are in session. The database contains eight stopping patterns, ranging from 32 stops up to 66 stops. The steps in the process for this use case are as follows. Step 1 is to select the route of interest. Route 7 will be employed (as shown above). Step 2 is to assemble PDFs for the deviations from advertised stop times by trip for the typical and adverse condi- tions of interest. Step 3 is to define what is meant by being on time (the limits of being either early or late and the probability of being within that window). Step 4 involves assessing the extent to which adjustments in the schedule would improve the on-time performance. TABLE D.54. ACQUIRE RELIABILITY DATA FOR BUILDING SCHEDULES (TS1) User Transit scheduler Question What schedule can be achieved given the variability in the travel times along the route? Steps 1. Select the route of interest. 2. Assemble PDFs for the deviations from advertised stop times by trip for the typical and adverse conditions of interest. 3. Define what is meant by being on time (the limits of being either early or late and the probability of being within that window). 4. Assess the extent to which adjustments in the schedule would improve the on-time performance. Inputs A database of deviations from scheduled stop times by trip for the route and conditions of interest. Result An assessment of the extent to which adjustments in the schedule would improve the on-time performance of the route.

588 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY That there are challenges with schedule reliability is evident from Figure D.41. Shown are all the differences from scheduled stop times for all equipped buses and all routes for the data received. The nth stop does not always refer to the same location in this particular plot because the stop sequences are different, but even with this caveat, it is clear that the bus drivers find it challenging to keep the buses on time. They some- times start the run early, by as much as 20 minutes, and leave intermediate stops early so they can be on time for the stops whose times are printed in the timetable (the stop times for all stops are not included in the published timetable). Unfortunately, even with their efforts, some of the buses are as much as 10 to 20 minutes late by the time they reach the end of the route; that is, they slip to about the times published for the bus that left ahead of them. 157 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx A little background about Route 7 is helpful before addressing the use case directly. Route 7 runs between downtown San Diego and La Mesa along Broadway, Park Boulevard, and University Avenue as shown in the top portion of Figure D.40. The length of the route is about 17.6 miles, and it requires a little more than an hour to traverse end to end. The timetable shows varying stop patterns, as shown in the bottom portion of Figure D.40. There are some even shorter stop patterns, not shown, that seem to be used only when the public schools are in session. The database contains eight stopping patterns, ranging from 32 stops up to 66 stops. [Insert Figure D.40] [caption] Figure D.40. Route 7 in San Diego.

589 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 159 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.41. Schedule deviations for Route 7 in San Diego. Figure D.42 is more precise in that it shows the growth in average lateness for the stop sequence that involves 66 stops (the maximum number of stops). If the average lateness remained close to zero, then the buses would, on average, be on time. But it is clear they are not. The average lateness continues to increase along the route. This means the schedule is too aggressive; the buses cannot keep up, and more slack is needed. The one exception is the stopping pattern involving 53 stops. The buses following that stop schedule seem to stay on time. [Insert Figure D.42] [caption] Figure D.41. Schedule deviations for Route 7 in San Diego. Figure D.42 is more precise in that it shows the growth in average lateness for the stop sequ nce that inv lves 66 tops (the maximum number of stops). If the average lateness remained close to zero, then the buses would, on average, be on time. But it is clear they are not. The average lateness continues to increase along the route. This means the schedule is too aggressive; the buses cannot keep up, and more slack is needed. The ne exception is the stopping pattern involving 53 stops. The buses fol- lowing that stop schedule seem to stay on time. The standard deviation of the lateness shows similar trends. The standard devia- tion could easily be constant across the route, meaning buses would be likely to be both early and late, and the extent to which they deviated from the scheduled stop times would be constant across the route. This is in fact true for the stopping pattern involving 53 stops. But for the others, the standard deviation grows, which says that the deviations from the average lateness are growing as the bus progresses along the route. This reemphasizes the fact that more schedule slack is needed.

590 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY For qualitative affirmation that the route has schedule challenges, the on-time per- formance of several buses can be examined. Plotted in Figure D.43 are the schedule deviations by stop for five buses. It is clear that Bus 2 encountered problems. Between about the fourth and sixth stop it got delayed, and it never recovered. It was late for the rest of the stops. Its pattern closely resembles that of Bus 5, which also struggled to stay on time as it progressed along the route. In fact, all of the buses except Bus 3 get farther behind as they progress along their routes. The message seems clear. The schedule needs to be lengthened. Perhaps the exist- ing schedule dates from an earlier year when the streets were not as congested as they were when the data were collected. Perhaps uncongested travel times were used instead of the conditions that pertain when the buses are operating. 160 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.42. Schedule deviations for Route 7 in San Diego for 66 stops. The standard deviation of the lateness shows similar trends. The standard deviation could easily be constant across the route, meaning buses would be likely to be both early and late, and the extent to which they deviated from the scheduled stop times would be constant across the route. This is in fact true for the stopping pattern involving 53 stops. But for the others, the standard d viation grows, which says that the d vi tions from the average laten ss ar growing Figure D.42. Schedule deviations for Route 7 in San Diego.

591 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY A good sense of how much time to add to the route can be obtained by examin- ing Figure D.44. Shown is the CDF for the lateness of buses at Stop 60, almost at the end of the route. Two stop IDs pertain to this stop; the CDFs for both are shown. The actual end of the route is not used because it is the depot, and lateness at the depot is not of concern from a schedule standpoint. As can be seen, 90% of the buses are late by 12 minutes or less. There is not a right answer for how much time to add, but add- ing 12 minutes so that 90% of the buses can complete the route on time seems reason- able. Only one in 10 will be late. The tail begins to extend significantly beyond that point, probably reflecting delays over which the bus drivers have little or no control. Figures D.41, D.42, and D.43 all suggest that the place to add this time is after about the 25th stop, so that is the strategy suggested here. 161 2014.04.23 15 L02 Guide Appendix D_Final for comp sition.docx as the bus progresses along the route. This reemphasizes the fact that more schedule slack is needed. For qualitative affirmation that the route has schedule challenges, the on-time performance of several buses can be examined. Plotted in Figure D.43 are the schedule deviations by stop for five buses. It is clear that Bus 2 encountered problems. Between about the fourth and sixth stop it got delayed, and it never recovered. It was late for the rest of the stops. Its pattern closely resembles that of Bus 5, which also struggled to stay on time as it progressed along the route. In fact, all of the buses except Bus 3 get farther behind as they progress along their routes. The message seems clear. The schedule needs to be lengthened. Perhaps the existing schedule dates from an earlier year when the streets were not as congested as they were when the data were collected. Perhaps uncongested travel times were used instead of the conditions that pertain when the buses are operating. [Insert Figure D.43] [caption] Figure D.43. Schedule deviations for five buses. Figure D.44. Schedule deviations at Stop 60 (out of 66 stops). 162 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.43. Schedule deviations for five buses. A good sense of how much time to add to the route can be obtained by examining Figure D.44. Shown is the CDF for the lateness of buses at Stop 60, almost at the end of the route. Two stop IDs pertain to this stop; the CDFs for both are shown. The actual end of the route is not used because it is the depot, and lateness at the depot is not of concern from a schedule standpoint. As can be seen, 90% of the buses are late by 12 minutes or less. There is not a right answer for how much time to add, but adding 12 minutes so that 90% of the buses can complete the route on time seems reasonable. Only one in 10 will be late. The tail begins to extend significantly beyond that point, probably reflecting delays over which the bus drivers have little or no control. Figures D.41, D.42, and D.43 all suggest that the place to add this time is after about the 25th stop, so that is the strategy suggested here. [Insert Figure D.44] [caption] Figure D.44. Schedule deviations at Stop 60 (out of 66 stops).

592 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Choose Departure Times to Minimize Arrival Uncertainty (TS2) A scheduler wants to choose departure times for a route that will minimize the likeli- hood that the stops are not on time. This analysis is useful when planning the schedule for a specific route. The scheduled departure and arrival times may be flexible, but the transit vehicle needs to arrive when scheduled. This analysis helps a user schedule a route to minimize travel time variability. Table D.55 presents the key question, steps, inputs, and the result for this use case. This use case is very much like TS1 except the focus is on the departure time for the route rather than the intermediate stop times. It is as though an earlier stop time helps ensure the stops are made on time. To some degree, this is a reasonable thought, especially if the first stop is some distance from the depot. The analysis of Route 7 conducted in TS1 can be used to illustrate the application of this use case. Assume the objective was to reach Stop 60 on time based on the cur- rent timetable. Two things would need to be done. First, as suggested by Figure D.45, the buses would need to start their routes about 12 minutes earlier than they presently do; second, the intermediate stop times would need to be adjusted to earlier times so that the timetable reflected a schedule that got the bus to Stop 60 on time. TABLE D.55. CHOOSE DEPARTURE TIMES TO MINIMIZE ARRIVAL UNCERTAINTY (TS2) User Transit scheduler Question When should a specific trip start to minimize the likelihood that the stops will not be on time? Steps 1. Select the route and trip of interest. 2. Assemble PDFs for the deviations from advertised stop times for the typical and adverse conditions of interest. 3. Define what is meant by being on time (probabilities of being late and early). 4. Assess the extent to which an adjustment in the departure time will improve the on-time performance at the stops. Inputs A database of stop times by trip for the route, trip, and conditions of interest. Result An assessment of the extent to which an adjustment in the departure time will improve the on-time performance of the route.

593 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 176 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx The TT-PDF between 99343 and 10314 is shown in Figure D.45. The CDF is based on 115 observations of trip times between these two stops during August. It seems the trip can take anywhere from 35 minutes to nearly 80 minutes. Even the 90th percentile travel time is 58 minutes, nearly double the shortest time observed. [Insert Figure D.45] [caption] Figure D.45. Travel Times on Route 7 from Broadway and 11th Street (99343) to University Avenue and Maple Street (10314). Table D.61 shows the distribution of trips based on whether the departure from the origin was early, on time, or late; and whether the arrival at the destination was early, on time, or late. On time was defined as being a half-minute early to 2 minutes late. [COMPOSITION: Please do not align decimal places in table.] Figure D.45. Travel Times on Route 7 from Broadway and 1th Street ( t University Avenue and Maple Street (10314). Transit Operators Transit operators care about travel time reliability because it is strongly linked with • Costs. Less reliable services have higher costs. Extra buses and drivers are needed to fill in for buses that are late. Bus bunching occurs. The bus loadings are uneven. The amount of time required for a bus to make a round trip ultimately determines how many buses and drivers are required to provide a given service frequency. Simply put, the need for more buses equates to higher operations and capital costs; • Ridership. More people use the service when it is more reliable. When the ser- vice is unreliable, bus loadings become highly variable, and bus bunching occurs. Passengers have to wait longer than expected for a bus and experience over- crowded conditions. More reliable service makes transit more competitive with other travel options; and • Revenue. Reliable service is more attractive to passengers, which generates more ridership and, in turn, more revenue. As suggested by TCRP Report 100: Transit Capacity and Quality of Service Manual (2), many factors influence transit service reliability: traffic conditions and road construction, vehicle and maintenance quality, vehicle and driver availability, existence of transit preferential treatments, schedule achievability, evenness of pas- senger demand, variations in bus operator experience, wheelchair lift and ramp usage, route length and number of stops, and operations control strategies.

594 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY A reliability monitoring system needs to give transit system service providers two types of reliability information. First, reliability metrics for general traffic along bus routes that operate in mixed traffic (more than 99% of all route miles in the United States) are needed; this requires access to the same data described in the previous use cases. Second, transit-specific travel time information measured relative to a schedule or designated headway is needed; this requires collecting data from transit vehicles equipped with tracking devices, typically a GPS-based AVL system. To support transit-specific reliability monitoring, these use cases introduce spe- cialized metrics such as schedule adherence and headway regularity. Note that transit providers may use a travel time slightly faster than the average travel time (e.g., a 40th percentile travel time) for scheduling purposes (to reduce the possibility that buses will need to sit and wait at a time point), and the buffer time used in scheduling may be based on something other than a 95th percentile travel time (e.g., a 90th percentile travel time).The following use cases demonstrate system functionalities that are help- ful for a transit provider’s bus operations department. Identify Routes with the Poorest Reliability (TO1) A transit operator wants to determine which routes have the poorest reliability. This information helps prioritize routes for further analysis to find the sources of the unreliability. Table D.56 presents the key question, steps, inputs, and the result for this use case. This use case is effectively the same as Use Case TP1. The objective is to identify routes that have the poorest reliability. Table D.52 helps answer that question in the context of the routes with AVL-equipped buses. At a standard deviation of 9.37 min- utes, Route 11 has the poorest reliability. The second worst reliable route is Route 10, with a standard deviation of 4.63 minutes. Note that the ranking of the routes, based on Table D.52, is slightly different from the impression conveyed by Figure D.38. In viewing the CDFs it appears that Routes 7 and 15 are the ones whose performance is the poorest, because Route 7 has TABLE D.56. IDENTIFY ROUTES WITH THE POOREST RELIABILITY (TO1) User Transit operator Question What routes have the poorest reliability performance? Steps 1. Select the routes of interest (could be all routes). 2. Select the operating conditions of interest (could be all conditions). 3. Assemble a database of variations from scheduled stop times for all the stops on the routes, classified by operating condition. 4. Identify the routes with the worst PDFs for the variations from scheduled stop times. Inputs Historical database of stop times by stop for the routes and conditions of interest plus the scheduled stopping times. Result A rank-ordered list of the routes based on the extent to which the actual stopping times for the stops on the routes differ from the intended stopping times.

595 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY the greatest deviations from the scheduled stop times from the 80th percentile onward, and Route 15 has the greatest deviations from the 20th to the 80th percentiles. These differences help illustrate the fact that there is not a single right answer to the question of which route is the poorest. The answer depends on the perspective one has on how poorest should be measured. Review Reliability for a Route (TO2) A user wants to review the reliability performance of a single route. This helps the transit operator identify the possible cause(s) of variation in on-time performance and, ultimately, a potential solution. Possible reasons for the unreliability include insuf- ficient running time in the schedule (trips arriving consistently late at the end of the route), insufficient recovery time (trips consistently starting later than scheduled), spot issues along a route (reliability issues consistently starting within a particular segment of a route), driver performance (trips operated by a particular driver being consistently late), or bus bunching. Table D.57 presents the key question, steps, inputs, and the result for this use case. This use case is similar to Use Case TS1. The question is: How does the on-time performance vary? The answer in the context of Route 7 is that it varies a lot: as Figure D.44 shows, buses on a 1-hour run may be as late as 20 minutes. Figure D.45 shows that the buses tend to be late, on average, and that the deviation in that lateness increases along the route. Figures D.46 and D.47 suggest that the problems with the route are more related to the later stops than the earlier stops. The biggest delays arise after Stop 20. It appears that adding 12 minutes to the timetable would resolve the on-time problems. Examine the Potential Impacts of Bus Priority on a Route (TO3) This use case explores the merits of introducing bus priority on a route. Bus priority is one way to enhance reliability. The buses can control the green times and extend the green so they are not delayed, or delay the green to avoid having to merge with TABLE D.57. REVIEW RELIABILITY FOR A ROUTE (TO2) User Transit operator Question How does the on-time performance vary for a specific route? Steps 1. Select the route of interest. 2. Select the operating conditions of interest (could be all conditions). 3. Assemble a database of variations from scheduled stop times for all the trips on the route, classified by operating condition. 4. Investigate why the worst deviations arise (could be due to a variety of reasons). Inputs Historical database of stop times for the route of interest and the conditions of interest relative to the scheduled stop times. Result PDFs for the deviations of specific stop times from their intended stop times, a ranking of the stops based on this deviation, and ideas about why the deviations arise.

596 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 178 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx occurred 11 times, and for which the average travel time was 43.05 minutes with a standard deviation of 4.93 minutes. About 48% of the time (3 + 25 + 27 observations) the bus arrives late at the destination; about 30% of the time (2 + 2 + 27), it is late leaving. Figure D.46 shows the CDF for deviation from the scheduled stop time at the destination. Although about 35% of the time the bus will arrive at or before the scheduled stop time, 65% of the time it is late; and at the 90th percentile, it is about 12 minutes late. [Insert Figure D.46] [caption] Figure D.46. CDF for lateness (deviation from scheduled stop time) at the destination. [Same caption for Figure D.47. CDF for lateness (deviation from scheduled stop time) at the destination??] 180 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Step 3 involves determining the options for departure times. Step 4 is to select the departure time that ensures an acceptable arrival time. At the heart of this analysis is a plot like the one in Figure D.47, which shows the travel times from the origin stop to the destination by time of day (by bus departure time, which means the same run). It is clear that the time required to make the trip varies widely and is heavily dependent on the time of day. [Inse t Figure D.47] [caption] Figure D.47. Travel times from the origin stop to the destination by time of day. Not enough data were available for any given departure time to create a complete CDF, but the spread in the data for August can certainly be observed. Figure D.46. CDF for lateness (deviation from scheduled stop time) at the destination. Figure D.47. Travel ti es fr m the origin stop to the desti ation by time of day.

597 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY conflicting traffic. Data on the amount of signal delay experienced by buses would help support the case for transit signal priority and justify its potential impacts on other traffic. Table D.58 presents the key question, steps, inputs, and the result for this use case. This use case is much like Use Case TP2. In TP2 the question was whether an exclusive bus lane would add value. Here the question is whether bus priority would help. In both cases, the question is whether some special treatment for buses would help improve reliability, either by removing them from the mixed-traffic stream or giving them the ability to be served by priority traffic signal operations. The method of analysis is the same in both cases. As illustrated in Use Case TP2, CDFs should be assembled for the on-time performance based on the deviations from planned stop times. If bus priority improves this performance, which was the case in comparing Route 7 with Route 88, then the treatment has merit. Assess a Mitigating Action for an Adverse Condition (TO4) In this use case, a transit operator wants to plan operational changes for future adverse conditions (e.g., incidents, severe weather, construction, or special events) and prepare riders for what the service will be like during these conditions. Table D.59 presents the key question, steps, inputs, and the result for this use case. TABLE D.58. EXAMINE THE POTENTIAL IMPACTS OF BUS PRIORITY ON A ROUTE (TO3) User Transit operator Question By how much would bus priority (e.g., at signals) improve the reliability of a transit route? Steps 1. Select the route of interest. 2. Select the operating conditions of interest (could be all conditions). 3. Identify other routes on which bus priority has been introduced or the impacts of signal delay are less significant. If none exist, build a simulation model that can be used to do the assessment. 4. Compare the TR-PDFs (rates) for the route of interest with TR-PDFs for the routes on which bus priority has been introduced or the impacts are less significant. Do this using simulation if necessary. 5. See how much impact bus priority has. Inputs A database of trip times for the route and conditions of interest and for routes on which bus priority has already been introduced. Result Differential TT-PDFs that show how the route’s reliability could be improved if bus priority were introduced.

598 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY This use case focuses on seeing if a mitigating action improves the reliability of a bus route under adverse conditions. Unfortunately, during August 2011, the time period during which the data were collected, no significant adverse conditions arose, nor were actions taken to mitigate those adverse conditions. Hence, this use case can- not be addressed directly by the data available. However, the use case analysis is again comparable to Use Cases TP2 and TO3. The objective is to see if the on-time performance is enhanced during the adverse con- dition by the action taken. For example, if heavy rains were an issue, then the question would be whether a mitigating action like adding extra buses would help mitigate the negative impacts. The analysis technique would be the same as that used in those two use cases. Transit Passengers This section focuses on transit riders, who benefit from reliability information when planning specific trips, either pretrip or en route. Like motorists, transit passengers are primarily concerned with trips: when should they start their trip, what route(s) they should take, and how much extra time they should allow for on-time arrivals. Unlike motorists, however, their options are constrained by the transit service network and schedule. Determine the On-Time Performance of a Trip (TC1) A user wants to know the on-time performance of a specific trip on a bus route. The trip has an origin, destination, and departure time. The user wants to know, in general, how long the trip will take. Table D.60 presents the key question, steps, inputs, and the result for this use case. TABLE D.59. ASSESS A MITIGATING ACTION FOR AN ADVERSE CONDITION (TO4) User Transit operator Question What benefit would be obtained by taking an action intended to mitigate the impacts of an adverse condition? Steps 1. Select the route of interest. 2. Select the adverse condition of interest. 3. Select the mitigating action to be tested (one or more could be assessed). 4. Assemble a database of TR-PDFs (rates) for routes under typical conditions and under conditions for which the mitigating action has been implemented. If none exist, build a simulation model that can be used to conduct that assessment. 5. Compare the TR-PDFs (rates) for typical and adverse conditions with and without the mitigating action. 6. See if the mitigating action has a significant impact. Inputs A database of trip times for the route and conditions of interest and for routes on which bus priority has already been introduced. Use simulation if necessary. Result Differential TR-PDFs that show how the route’s reliability would be improved if the mitigating actions were used.

599 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY This use case is the fundamental building block of the analysis procedures related to transit riders. It focuses on determining the on-time performance for a specific trip, from an origin stop to a destination stop, on the same bus route. Step 1 is to select the bus route, departure stop, and arrival stop. In this instance Route 7 will be used with boarding Stop 99343 (Broadway and 11th Street) and alight- ing stop 10314 (University Avenue and Maple Street). Step 2 is to assemble a database of TT-PDFs for the route and stop pair of interest, and Step 3 is to develop a PDF for the deviation from scheduled arrival time at the stop of interest. The TT-PDF between 99343 and 10314 is shown in Figure D.45. The CDF is based on 115 observations of trip times between these two stops during August. It seems the trip can take anywhere from 35 minutes to nearly 80 minutes. Even the 90th percentile travel time is 58 minutes, nearly double the shortest time observed. Table D.61 shows the distribution of trips based on whether the departure from the origin was early, on time, or late; and whether the arrival at the destination was early, on time, or late. On time was defined as being a half-minute early to 2 minutes late. TABLE D.61. ON-TIME ASSESSMENT FOR ROUTE 7 FROM BROADWAY AND 11TH STREET (99343) TO UNIVERSITY AVENUE AND MAPLE STREET (10314) Origin Destination Early On Time Late Early No. of observations 11 9 3 Average travel time (min) 43.05 48.33 53.01 Standard deviation 4.93 5.41 4.88 On time No. of observations 24 11 25 Average travel time (min) 44.73 50.89 55.14 Standard deviation 5.59 6.40 7.72 Late No. of observations 2 2 27 Average travel time (min) 44.55 47.88 55.15 Standard deviation 12.49 3.72 6.57 TABLE D.60. DETERMINE THE ON-TIME PERFORMANCE OF A TRIP (TC1) User Transit passenger Question How often does a specific bus trip arrive at a specific stop on time? Steps 1. Select the bus route, departure stop, and arrival stop. 2. Assemble a database of TT-PDFs for this bus route and stop pair of interest (if available) for the arrival time and condition(s) of interest. 3. Develop a PDF for the deviation from scheduled arrival time at the stop of interest. Inputs Historical database of TT-PDFs for the bus route and trip of interest (if available) for the conditions of interest. Result A PDF for the deviation from scheduled arrival time at the stop of interest.

600 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Of the 115 observations, the buses are late departing from the origin stop and late arriving at the destination 27 times. The average travel time in this condition is 55.15 minutes, and the standard deviation is 6.57 minutes. This is strikingly different from the best performance, which occurred 11 times, and for which the average travel time was 43.05 minutes with a standard deviation of 4.93 minutes. About 48% of the time (3 + 25 + 27 observations) the bus arrives late at the destination; about 30% of the time (2 + 2 + 27), it is late leaving. Figure D.46 shows the CDF for deviation from the scheduled stop time at the des- tination. Although about 35% of the time the bus will arrive at or before the scheduled stop time, 65% of the time it is late; and at the 90th percentile, it is about 12 minutes late. Determine an Arrival Time Just Before a Trip (TC2) A rider wants to determine, just before a trip, when he or she might arrive. This is similar to TC1, but it uses data about current conditions. In addition, since the inquiry is occurring just before a trip, the departure time is relatively fixed and the arrival time becomes the focus. Table D.62 presents the key question, steps, inputs, and the result for this use case. Step 1 involves selecting an origin and destination. The same two stops employed in TC1 will be used. Step 2 involves determining what on time (the probability of being late) means. Step 3 involves determining the options for departure times. Step 4 is to select the departure time that ensures an acceptable arrival time. At the heart of this analysis is a plot like the one in Figure D.47, which shows the travel times from the origin stop to the destination by time of day (by bus departure time, which means the same run). It is clear that the time required to make the trip varies widely and is heavily dependent on the time of day. Not enough data were available for any given departure time to create a complete CDF, but the spread in the data for August can certainly be observed. A logical answer to the question is something like this. In the morning, say before 10:00 a.m., an hour needs to be allowed. The bus might actually arrive in 50 minutes, but allowing for 60 minutes is safe. Across the middle of the day, 70 minutes should be allowed, although (unfortunately) there are times when only 50 minutes will be TABLE D.62. DETERMINE AN ARRIVAL TIME JUST BEFORE A TRIP (TC2) User Transit rider Question Just before making a trip, when does the rider have to leave to arrive at a destination on time? Steps 1. Select the origin and destination. 2. Decide what being on time (probability of being late) means. 3. Determine the options for departure times. 4. Select the departure time that ensures an acceptable arrival time. Inputs Forecasts of passenger TT-PDFs by departure time for the routes that might be chosen given the current and anticipated network conditions. Result Plots of the TT-CDFs for the various departure times so the most appropriate departure time can be selected.

601 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY needed. Toward evening, the time needed drops back to 60 minutes and then to less than 50 minutes, and by about 11:00 p.m., only 40 minutes needs to be allowed. Moreover, the travel times are very consistent. The value in having real information about the trip times—in this case by time of day—is strikingly apparent. Determine a Friend’s Arrival Time (TC3) In this use case, the TTRMS user wants to know when his or her friend will arrive. The use case is similar to TC2 because it focuses on the arrival time and depends on infor- mation about current conditions. However, in this case, the route is already known, so the question is: When will the friend arrive? Table D.63 presents the key question, steps, inputs, and the result for this use case. In state-of-the-art bus systems, the progress of individual buses can be tracked via a web page or cell phone app. Such systems make it easy to answer this question. But most systems today are not state of the art, so historical information needs to be used instead. Step 1 is to select the origin, destination, route, and bus being ridden. Without loss of generality, Route 7 will again be selected, origin 99343, destination 10314, and—to make it interesting—departure time 1:00 p.m. As can be seen in Figure D.50 later in the appendix, the trip for this departure time can take anywhere from 55 to 72 min- utes, which is a spread of 17 minutes. The guidance might be to plan on being ready for the friend as early as 55 minutes after he or she leaves, do not plan on doing something until at least 72 minutes after he or she departs, and have his or her cell phone number. There is enough variation that making a call to find out where the bus is would be valuable. In contrast, late at night, say with an approximately 11:00 p.m. departure, there is less need for contingency planning. The shortest travel time is about 35 minutes, and the longest is about 40 minutes. Hence, it would be fine to show up at the bus stop about 35 minutes after the friend leaves the origin and expect to wait no more than about 5 minutes for the bus to arrive. TABLE D.63. DETERMINE A FRIEND’S ARRIVAL TIME (TC3) User Friend of a transit rider Question When will the friend arrive? Steps 1. Select the origin, destination, route, and bus being ridden. 2. Assemble current information for this bus run (if available) and historical TT-PDFs for this bus route and this specific run (if available). 3. Develop an arrival time PDF for this bus trip at the stop of interest. Inputs Information about this bus run (if available) and historical TT-PDFs for this bus route and this specific run (if available). Adjust for the current conditions. Result Arrival time PDF for this bus trip at the stop of interest.

602 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Understand a Trip with a Transfer (TC4) A rider wants to find out, in advance, what time to leave and what route(s) to take to reach a destination on time when a transfer is involved. Missing the bus at the origin and missing the transfer are both concerns, as well as being late or early at the destina- tion. The use case is based on historical data, rather than current conditions, and as such is suitable for travelers who want to plan a trip in advance rather than immedi- ately before they leave. Table D.64 presents the key question, steps, inputs, and the result for this use case. This use case focuses on understanding departure times and routes for a trip that involves a transfer. Without loss of generality, a bus trip from the Gaslight district to the San Diego Zoo can be used to illustrate. In August 2010 when the bus data were collected, the San Diego Zoo was open from 9:00 a.m. to 9:00 p.m. on weekdays. Hence, arrival at the zoo will be targeted within this window. A transfer is an important feature of this trip. The San Diego transit trip planner suggests several possible paths, one of which uses bus Routes 11 and 7. For a 1:00 p.m. trip on a weekday, the trip planner suggests the following path: 1. Walk 0.4 mile north from the Gaslamp Quarter Trolley Station to Market Street at 6th Avenue. 2. At 12:50 p.m. take the MTS BUS Route 11 SDSU via Downtown/Adams Ave. 3. Get off the stop on Park Boulevard and University Avenue at approximately 01:19 p.m. 4. Walk 0.1 mile south from Park Boulevard and University Avenue to Park Boulevard at University Avenue. 5. At 01:25 p.m. take the MTS BUS Route 7 Downtown via University Ave. 6. Get off the bus at the stop on Park Boulevard at Zoo Place at approximately 01:29 p.m. 7. Walk 0.2 mile west to the Zoo. TABLE D.64. UNDERSTAND A TRIP WITH A TRANSFER (TC4) User Transit rider Question What departure times are needed when a transfer is involved? Steps 1. Select the origin, destination, desired arrival time, and route. 2. Decide what being on time (probability of being late) means. 3. Analyze the TT-PDFs to identify options for departure times. 4. Create a CDF of the travel times so that an acceptable departure bus can be selected. Inputs Historical passenger TT-PDFs for the departure times that are logical based on conditions. Result Table of departure times depending on the conditions. The table is created by analyzing the TT-PDFs for the routes involved and network conditions.

603 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY The team hoped that all the buses on Routes 11 and 7 would be equipped with monitoring devices, but such was not the case. Only about half of them were. If they all had been equipped, it would have been possible to construct trip travel times by selecting a boarding a bus, riding it to the transfer point, transferring from it to the next transfer bus, and then riding to the destination. This proved infeasible because all the buses were not instrumented. A different strategy was needed. The analysis involved two steps: preprocessing the bus trip data to develop infor- mation needed to conduct the analysis and generating a synthesized set of hypotheti- cal representative trips through Monte Carlo simulation. These steps are described in detail in Appendix B: Methodological Details, in the section on developing transit rider PDFs for trips. To conduct these analyses with the San Diego data, a somewhat lengthy but straightforward nine-step process is involved: 1. For each route separately, combine the data for the individual weekdays into a combined data set for all the weekdays in August. 2. Extract the records for which the latitude and longitude fields are not blank (this happens to be the case for about 50% of the records). 3. Compute average latitude and longitude values for every Stop ID based on the nonblank records. 4. Backfill the latitude and longitude fields with average values computed above for all the records that are blank so that every record has these fields filled. 5. If any records remained with no latitude and longitude data, use the latitude and longitude values for the stops on either side to determine what the latitude and longi- tude most likely were for the stops with no latitude and longitude data (this situation arose for one Route 7 stop). 6. Create a set of unique stop names (cross-street descriptors) for every Stop ID based on the latitude and longitude for the stop. 7. Add the stop names to every record in the database. 8. Extract from the data set the stops of interest for the trip time analysis (Sixth and Market and University Avenue and Park Boulevard for Route 11; University Avenue and Park Boulevard and Park Boulevard and Zoo Place for Route 7). 9. Extract the records for these stops into two new data sets, one for each route. To provide a sense of the result, the Route 7 database for the two stops of inter- est in the direction of interest includes 484 records. The Route 11 database contains 1,415 records. Some analysis results taken directly from these data sets indicate how these two routes operate for the stops involved in the trip being studied. Figure D.48 shows the distribution of relative departure times for the two stops on Route 11 and the two stops on Route 7. Sometimes the buses leave early, but most of the time they leave late, by as much as 5 minutes or more. The figure also shows the distribution of travel times between the two stops of interest. For Route 11, the travel times are much longer, and

604 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Figure D.48. Distributions of relative departure times and travel times between stops. 188 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.48. Distributions of relative departure times and travel times between stops. Figure D.49 displays the correlations among these various times. The top two plots show the correlations between the differential stop time at the first and second stops. The correlation is

605 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY they are skewed toward longer values. For Route 7, the distribution is almost uniform between 120 and 300 seconds (2 to 5 minutes). Figure D.49 displays the correlations among these various times. The top two plots show the correlations between the differential stop time at the first and second stops. The correlation is very pronounced in the case of Route 7, probably because the stops are close to one another. The bottom two plots show the absence of correlation between the differential stop time at the first stop and the travel time to the transfer (or destination) stop. The correlations between the differential stop times motivated a change in the strategy for conducting the Monte Carlo simulations. Instead of fitting PDFs to the data, which is what the team expected to do, it was decided to directly sample the observed data sets. That is, during simulation, to obtain values for Δt1, Δt2, and Δt3 for the first Route 11 bus (or Δt5, Δt6, and Δt7 for the second one), the team directly sampled values from the 484 Route 11 values developed by the data analysis. In the case of Δt10, Δt11, and Δt12 for the first Route 7 bus (or Δt14, Δt15, and Δt16 for the second one), the same action was performed: values were sampled from the 1,415 records Figure D.49. Correlations among the differential stop and travel times. 189 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx ve y pronounced in the case of Route 7, probably because the stops are close to one another. The bottom two plots show the absence of correl ti betw en the differential stop time at the first stop and the travel time to the transfer (or destination) stop. [Insert Figure D.49] [caption] Figure D.49. Correlations among the differential stop and travel times. The correlations between the differential stop times motivated a change in the strategy for conducting the Monte Carlo simulations. Instead of fitting PDFs to the data, which is what the team expected to do, it was decided to directly sample the observed data sets. That is, during

606 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY developed by the data analysis. A large-scale simulation would have been needed to sample all of the records in the two data sets; but by sampling the data sets directly, the team could ensure that the combinations of values used in the simulation were combinations observed in the field, not ones created through a synthesizing process. Finally, for the passenger arrival times at the initial stop, the distribution observed by Bowman and Turnquist (3) was used. Consistent with the guidance they provide, the original observations for buses with a 20-minute headway service were scaled to reflect values for a 15-minute headway service. The resulting distribution is shown in Figure D.50. Again, for purposes of simulation, the distribution shown was directly sampled to obtain arrival times for the passengers relative to the scheduled arrival time of the bus. A simulation of 1,000 trips produced the cumulative density function shown in Figure D.51. The shortest travel time is 30 minutes, and the longest is 72 minutes. The 50th percentile is reached at 44 minutes, and the 95th percentile is reached at 58 min- utes. The average is 44 minutes. Thus, the longest travel time is 32% longer than the mean and twice as long as the shortest time. Guidance to potential passengers might be that they should expect the trip to take 44 minutes, but one of every 20 trips takes longer than 58 minutes. Figure D.50. Arrival times for transit passengers relative to the scheduled arrival time of the bus. 191 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.50. Arrival times for transit passengers relative to the scheduled arrival time of the bus. A simulation of 1,000 trips produced the cumulative density function shown in Figure D.51. The shortest travel time is 30 minutes, and the longest is 72 minutes. The 50th percentile is reached at 44 minutes, and the 95th percentile is reached at 58 minutes. The average is 44 minutes. Thus, the longest travel time is 32% longer than the mean and twice as long as the shortest time. Guidance to potential passengers might be that they should expect the trip to take 44 minutes, but one of every 20 trips takes longer than 58 minutes. [Insert Figure D.51] [caption]

607 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY FREIGHT USE CASES This section presents use cases for freight service providers and customers. The vehicles involved are assumed to be trucks and vans, but the methodologies pertain to other freight modes, such as rail, water, and air freight. The use cases are clustered in two groups: those that pertain to the customers (e.g., shippers and receivers) and those that pertain to the service providers (e.g., trucking companies). Service providers are further subdivided into three groups relative to the motiva- tions for the use cases: • Truckload carriers. Entire truckloads of freight are picked up at one location and delivered to another. Chemical Leaman is an example. • Less-than-truckload carriers. Shipments (e.g., pallets, not packages) are trans- ported from shippers to receivers. Yellow Freight is an example. Local trucks make pickups and bring the shipments to a terminal. The shipments are then transported by one or more line haul trucks to a terminal that services the area where the re- ceiver is located. A local delivery truck then carries the shipment to the receiver (along with other shipments going to other receivers). • Parcel carriers. Shipments (predominantly packages) are carried from shippers to receivers. United Parcel Service is an example. The logistics are much the same as the less-than-truckload carriers in that the trucks make a series of pickups and deliveries of relatively small packages along a route that may vary from day to day. Figure D.51. Distribution of travel times for 1,000 simulated trips. 192 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.51. Distribution of travel times for 1,000 simulated trips. <H1>Freight Use Cases This section presents use cases for freight service providers and customers. The vehicles involved are assumed to be trucks and vans, but the methodologies pertain to other freight modes, s ch s rail, water, and air freight. Th use cases are clus ered in tw groups: those that pertain to the customers (e.g., shippers and receivers) and those that pertain to the service providers (e.g., trucking companies). Service providers are further subdivided into three groups relative to the motivations for the use cases:

608 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Travel time reliability matters to all these carriers, especially the ones that have promised delivery times, like FedEx. Shippers and receivers often specify pickup and delivery windows because they are ship or receive multiple items per day. The shippers and receivers want the shipments to be picked up and delivered within those windows, and the carriers strive to ensure that they do the pickups and deliveries during those windows. In fact, as illustrated by FedEx or United Parcel Service, people are willing to pay more for the service either to have the shipment delivered faster or to ensure that it is delivered on time. Most freight drivers accumulate information about travel times and reliability through experience and know how to avoid spots in the network where the travel times are unreliable. Carriers also use route guidance devices to direct their trucks around problem spots to make sure that deliveries are made on time. Freight Service Providers The next set of use cases shows how the monitoring system can be helpful to trucking companies trying to ensure that their deliveries will occur on time. Identify the Most Reliable Delivery Time (FP1) A trucking company wants to find the best time of day and route for a specific delivery. Both the probability of being on time and the travel time are important. This analysis is useful for carriers with flexible departure and arrival times, but who need to arrive as scheduled. Table D.65 presents the key question, steps, inputs, and the result for this use case. In this use case, Step 1 involves selecting the origin and destination. The example presented here uses the same origin and destination used in Figure D.1 involving the three routes in San Diego. Step 2 is to assemble the TT-PDFs by route and time of day and perhaps network condition. In this use case, only the normal condition is examined, reflecting decision making under normal conditions. The same Monte Carlo simulation technique used in the traveler use cases was employed to generate individual vehicle travel times for the I-5, SR-15, and SR-163 routes. For each normal 5-minute time period during the TABLE D.65. IDENTIFY THE MOST RELIABLE DELIVERY TIME (FP1) User Trucking company Question When should a delivery be made if the most reliable delivery time is desired? Steps 1. Select the origin and destination. 2. Assemble TT-PDFs by route and time of day and perhaps network condition. 3. Define on-time delivery (e.g., percentage within an on-time window). 4. Identify the time of day and route that minimize the duration of the on-time window. Inputs A database of TT-PDFs for truck travel times by route and time of day from the origin to the destination under the network operating conditions that pertain. Result A rank ordering of the departure times and routes based on the duration of the on-time window (narrowest to widest).

609 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY weekdays in 2009, 100 truck trips were simulated, one for each percentile based on the average travel time observed. Step 3 is to define what is meant by the on-time delivery window. In this use case, it was decided that attention would be focused on the narrowest window in which 80% of the arrivals occurred. That meant that the analysis would seek the ith and jth percentile arrival times such that (a) the total arrivals between the two percentiles was 80% (e.g., between the 5th and 85th percentiles), and (b) the difference in the arrival times between these two percentiles was the smallest (e.g., the difference in arrival times between the 7th and the 87th percentile arrival travel times was the smallest of all possible percentile travel times encompassing 80% of the arrivals). Step 4 involves identifying the time of day and route that minimize the duration of the on-time window. In this case, each percentile delivery time was calculated for every 5-minute time period to obtain the minimum on-time window that would satisfy the 80% requirement. All of the minimums for every 5-minute period for each route were compared. Finally, a table was created showing the routes and on-time window for each 5-minute period between 8:00 a.m. and 5:00 p.m. Figure D.52 shows the result of routes corresponding to the time window. The x-axis stands for departure times, and the y-axis stands for the corresponding on-time windows. The two vertical lines stand for the boundary from 8:00 a.m. to 5:00 p.m. Minimum_Window stands for the smallest on-time window of each 5-minute depar- ture time among the three routes. The figure shows that SR-15 performs better during peak hours, and I-5 performs better during off-peak hours. The minimum on-time win- dow is between 2 and 3 minutes during off-peak hours and varies from 3 to 6 minutes Figure D.52. Route selection. (The x-axis stands for departure times, and the y-axis stands for the corresponding on-time windows. Minimum_Window = the smallest on-time window of each 5-minute departure time among the three routes.) 197 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx 5-minute time period, it can be seen that the smallest on-time window (2.02 minutes) occurs when the departure time is 8:00 a.m. on route SR-15. [Insert Figure D.52] [caption] Figure D.52. Route selection. (The x-axis stands for departure times, and the y-axis stands for the corresponding on-time windows. Minimum_Window = the smallest on-time window of each 5- minute departure time among the three routes.) <H3>Estimate a Delivery Window (FP2) A user wants to see what the delivery window is for a departure time and route. This helps the trucking company find a window that minimizes the impact of traffic. It also helps with

610 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY during the p.m. peak. After comparing the three routes and each 5-minute time period, it can be seen that the smallest on-time window (2.02 minutes) occurs when the depar- ture time is 8:00 a.m. on route SR-15. Estimate a Delivery Window (FP2) A user wants to see what the delivery window is for a departure time and route. This helps the trucking company find a window that minimizes the impact of traffic. It also helps with deliveries in rural areas where a limited number of routes exist for an origin and destination. Table D.66 presents the key question, steps, inputs, and the result for this use case. Step 1 involves selecting the origin and destination. In this instance, the same ori- gin and destination are used as those in Figure D.1, which shows the three routes in San Diego. For departure times, 8:00 a.m. to 5:00 p.m. are used without loss of gen- erality inasmuch as these are typical working hours. For the route, SR-163 is selected. Step 2 involves deciding what the on-time percentage is to be. In this use case, the 80% on-time window was selected, consistent with the previous use case. Step 3 involves assembling the TT-PDFs by time of day and route, and possibly by the network operating conditions. Step 4 involves determining when the delivery window begins and ends given the on-time percentage. Figure D.53 shows the travel times to the beginning and end of the on-time window for departures between 8:00 a.m. and 5:00 p.m., marked with vertical lines. The figure indicates, for example, that if the truck were to leave at 8:00 a.m., the smallest on-time window begins about 13 minutes later (at 8:13 a.m.) and ends about 16 minutes later (shortly after 8:16 a.m.). The x-axis is the departure time, and the y-axis shows the travel time. As Figure D.53 shows, the arrival times do not vary by much, and the travel time grows to above 20 minutes late in the afternoon. The departure times with the small- est on-time windows (the ones with the smallest difference between the starting and ending times) occur between 10:00 and 10:30 a.m. It is helpful to plot the length of the on-time window as a function of the departure time during the day, or the difference between the red and blue lines in Figure D.53. This trend is presented in Figure D.54. TABLE D.66. ESTIMATE A DELIVERY WINDOW (FP2) User Trucking company Question What delivery window should be promised? Steps 1. Select the origin, destination, departure time, and route. 2. Decide what the on-time percentage is to be (i.e., probability of being neither late nor early). 3. Assemble TT-PDFs by time of day (and perhaps network operating condition) for the route. 4. Determine when the delivery window begins and ends given the on-time percentage. Inputs A database of TT-PDFs for truck travel times by time of day for the selected route under the network operating conditions of interest. Result The delivery window (beginning to end) for the on-time percentage.

611 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY By reviewing the plot in Figure D.54, it is possible to see that the smallest on-time window occurs at 10:15 a.m. and lasts for approximately 2.5 minutes. Identify How to Maximize the Probability of an On-Time Delivery (FP3) In this use case, the trucking company wants to know when a truck needs to leave and what route it should follow to maximize the probability of an on-time delivery. This is important for time-sensitive shipments. Table D.67 presents the key question, steps, inputs, and the result for this use case. 199 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Step 2 involves deciding what the on-time percentage is to be. In this use case, the 80% on-time window was selected, consistent with the previous use case. Step 3 involves assembling the TT-PDFs by time of day and route, and possibly by the network operating conditions. Step 4 involves determining when the delivery window begins and ends given the on- time percentage. Figure D.53 shows the travel times to the beginning and end of the on-time window for departures between 8:00 a.m. and 5:00 p.m., marked with vertical lines. The figure indicates, for example, that if the truck were to leave at 8:00 a.m., the smallest on-time window begins about 13 minutes later (at 8:13 a.m.) and ends about 16 minutes later (shortly after 8:16 a.m.). The x-axis is the departure time, and the y-axis shows the travel time. [Insert Figure D.53] [caption] Figure D.53. Travel times to the beginning and end of the on-time window. Figure D.53. Travel times to the beginning and end of the on-time window. Figure D.54. Length of the on-time window as a function of departure time. 200 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx As Figure D.53 shows, the arrival times do not vary by much, and the travel time grows to above 20 minutes late in the afternoon. The departure times with the smallest on-time windows (the ones with the smallest difference between the starting and ending times) occur between 10:00 and 10:30 a.m. It is helpful to plot the length of the on-time window as a function of the departure time during the day, or the difference between the red and blue lines in Figure D.53. This trend is presented in Figure D.54. [Insert i re . 4] [caption] Figure D.54. Length of the on-time window as a function of departure time. By reviewing the plot in Figure D.54, it is p ssible to see that the smallest on-time window occurs at 10:15 a.m. and lasts for approximately 2.5 minutes.

612 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin and destination. In this use case, the O–D pair shown in Figure D.1 was used again. Step 2 is to define the on-time window, or the time frame during which shipments are considered to be delivered on-time. The half hour from 4:00 to 4:30 p.m. was selected. Step 3 involves assembling the TT-PDFs by time of day and route. The TT-PDFs mentioned in FP1 was used, for SR-163 under normal conditions. Step 4 involves finding the route and departure time that maximize the probability of arriving within the on-time window. The same Monte Carlo simulation technique used in the traveler use cases was employed to generate individual vehicle travel times for each route. Figure D.55 shows the results. The graph shows that for all three routes the departure times between 3:50 and 4:05 p.m. maximize the on-time arrival percentage. TABLE D.67. IDENTIFY HOW TO MAXIMIZE THE PROBABILITY OF AN ON-TIME DELIVERY (FP3) User Trucking company Question When should a truck leave and what route should it use to maximize the likelihood of an on-time delivery? Steps 1. Select the origin and destination. 2. Define the on-time window (earliest to latest delivery times). 3. Assemble TT-PDFs by time of day and route (and perhaps network operating condition). 4. Find the route and departure time that maximize the probability of arriving within the on-time window. Inputs A database of TT-PDFs for truck travel times by time of day and route under the network operating conditions of interest. Result The departure time and route to use to maximize the probability of making an on-time delivery. Figure D.55. Effects of departure time on the on-time arrival percentage. 203 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx traveler use cases was employed to generate individual vehicle travel times for each route. Figure D.55 shows the results. The graph shows that for all three routes the departure times between 3:50 and 4:05 p.m. maximize the on-time arrival percentage. [Insert Figure D.55] [caption] Figure D.55. Effects of departure time on the on-time arrival percentage. <H3>Assess the On-Time Probability for a Scheduled Shipment (FP4) A user wants to obtain the probability of an on-time arrival based on a scheduled departure time, a desired delivery time, and route. This information helps the trucking company assess whether a proposed delivery schedule incorporates too much risk of not being on-time. Table D.68 presents the key question, steps, inputs, and the result for this use case. Table D.68. Assess the On-Time Probability for a Scheduled Shipment (FP4)

613 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Assess the On-Time Probability for a Scheduled Shipment (FP4) A user wants to obtain the probability of an on-time arrival based on a scheduled de- parture time, a desired delivery time, and route. This information helps the trucking company assess whether a proposed delivery schedule incorporates too much risk of not being on-time. Table D.68 presents the key question, steps, inputs, and the result for this use case. Step 1 involves selecting the origin, destination, and scheduled departure and delivery times. In this use case, the O–D pair shown in Figure D.1 was used again. Step 2 involves defining the on-time delivery window. Here, the on-time delivery window was defined as being from 4:00 to 4:30 p.m. Step 3 involves assembling TT-PDFs by time of day and route (see Figure D.58 later in the appendix). Step 4 involves identifying the probability that the truck will arrive during the on-time delivery window. As shown in Figure D.58, the probability of being on-time can be seen for given departure times. When the departure times are between 3:00 and 4:20 p.m., the probability of being on time grows from zero to almost one between 3:50 and 4:05 p.m. and then drops back to zero. Assess the Impacts of Adverse Highway Conditions (FP5) A trucking company wants to obtain information concerning the likelihood that a route will involve large delays (e.g., due to inclement weather). Knowing this will re- duce the likelihood of severe delays. This information would help the company assess the likelihood of a given (rural) route being closed prior to the arrival of a truck or during the passage of the truck along the route and, if necessary, identify alternative routes. The assessment is based on a combination of current weather forecasts, current road conditions, and historical experience. Table D.69 presents the key question, steps, inputs, and the result for this use case. TABLE D.68. ASSESS THE ON-TIME PROBABILITY FOR A SCHEDULED SHIPMENT (FP4) User Trucking company Question What on-time probability can be expected for a scheduled shipment? Steps 1. Select the origin, destination, and scheduled departure and delivery times. 2. Define the on-time delivery window. 3. Assemble TT-PDFs by time of day and route (and perhaps network operating condition). 4. Identify the probability that the truck will arrive during the on-time delivery window. Inputs A database of TT-PDFs for truck travel times by time of day and route under the network operating conditions of interest. Result The probability of being on-time given the departure time and delivery time window.

614 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin, destination, and route of interest. In this use case, the O–D pair shown in Figure D.1 was used again. The route of interest is SR-163. Step 2 involves assembling TT-PDFs for the route and time frames during which the trips would take place. The TT-PDFs for the SR-163 route employed in FP1 can again be used. Step 3 involves analyzing the TT-CDFs to identify the likelihood that the trips will take more than various amounts of time. In this case, a comparison is made of the effects on travel time of events such as special events, incidents, weather events, and high demand. Step 4 involves determining how long the trips might take and the probabilities that those times might materialize. Figure D.56 shows the travel times for SR-163 under different conditions. For this route, the minimum travel time is 15.43 minutes, and the corresponding departure time is 10:15 a.m. From the figure it can be seen that the biggest impact at 10:15 a.m. is a special event that causes the travel time to increase to 21.68 minutes. TABLE D.69. ASSESS THE IMPACTS OF ADVERSE HIGHWAY CONDITIONS (FP5) User Trucking company Question What is the likelihood that a route will involve large delays? Steps 1. Select the origin, destination, and route of interest. 2. Assemble TT-PDFs for the route and time frames during which the trips will take place. 3. Analyze the TT-CDFs to identify the likelihood that the trips will take more than various amounts of time. 4. Determine how long the trips might take and the probabilities of those times. Inputs A database of TT-PDFs for truck trips on the route of interest during the seasons (time frames) of interest. Result The probability that the trip might take as long as or longer than specific amounts of time.

615 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Determine the Start Time for a Delivery Route (FP6) This use case focuses on a truck making multiple deliveries. The truck follows a tour from the depot to the delivery locations and back to the depot. The objective is to determine when the tour should start and what the delivery times should be, or how much schedule slack (extra time) should be provided, to maximize the probability that all the deliveries will be made on time. In the literature, this is referred to as the stochastic vehicle routing problem. If the travel times and unloading times are all fixed, then no slack is needed; but if either or both are variable, as they most often are, then extra time is needed. For this use case, the important piece of information is the schedule slack. The length of the tour and the return time to the depot are ignored. (Sometimes it is not possible to do this.) Table D.70 presents the key question, steps, inputs, and the result for this use case. Figure D.56. Travel times for SR-163. 207 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Step 4 involves determining how long the trips might take and the probabilities that those times might materialize. Figure D.56 shows the travel times for SR-163 under different conditions. For this route, the minimum travel time is 15.43 minutes, and the corresponding departure time is 10:15 a.m. From the figure it can be seen that the biggest impact at 10:15 a.m. is a special event that causes the travel time to increase to 21.68 minutes. [Insert Figure D.56] [caption] Figure D.56. Travel times for SR-163. <H3>Determine the Start Time for a Delivery Route (FP6) This use case focuses on a truck making multiple deliveries. The truck follows a tour from the depot to the delivery locations and back to the depot. The objective is to determine when the tour

616 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Find the Departure Time and Routing for a Set of Deliveries (FP7) A trucking company wants to plan the schedule for a delivery route (tour) that has both flexible and inflexible deliveries. The objective is to devise a plan that minimizes cost yet ensures on-time arrival for those deliveries requiring it. This type of plan can help less-than-truckload and package delivery carriers set schedules and routings for a series of deliveries with varying degrees of flexibility. As with FP6, some of the deliver- ies have specific on-time windows; the others do not. Sequencing is an issue for the deliveries that do not. In the classic literature this is referred to as the stochastic vehicle routing problem with some on-time windows. Table D.71 presents the key question, steps, inputs, and the result for this use case. TABLE D.70. DETERMINE THE START TIME FOR A DELIVERY ROUTE (FP6) User Trucking company Question When should the truck leave (how much schedule slack should be provided) to maximize the likelihood that all of the deliveries in a tour will be on time? Steps 1. Select the tour (delivery route) of interest. 2. Establish the on-time windows for the destinations. 3. For every i to j pair of nodes in the tour, assemble TT-PDFs for tij for every reasonable departure time from i (i.e., either the depot or the previous delivery location). 4. For each possible departure time from the depot use the tij TT-PDFs to develop PDFs for the deliveries at each of the destinations. (This assumes the time spent making delivery is fixed.) 5. Find the departure time that maximizes the smallest of these on-time delivery probabilities or some weighted combination of them. Inputs A database of TT-PDFs for truck trips on the links between the delivery nodes for every reasonable departure time from the departure node. Result An estimate of the extra time at the beginning of the trip needed to maximize the probability that the deliveries are made on time. TABLE D.71. FIND THE DEPARTURE TIME AND ROUTING FOR A SET OF DELIVERIES (FP7) User Trucking company Question When does a truck need to leave and in what sequence should it make its deliveries to maximize its on- time delivery performance? Steps 1. Select the locations to which deliveries need to be made. 2. Specify the on-time windows for deliveries that have them. 3. Assemble TT-PDFs for all the nonstop paths between the places to be visited (because the sequence is unknown). If the TT-PDFs are time dependent, then assemble TT-PDFs by departure time. 4. Solve the stochastic vehicle routing problem with partial on-time windows. Inputs A database of truck TT-PDFs for nonstop paths between the delivery locations, differentiated by departure time if necessary. Result A tour that that maximizes the probability that the time-constrained deliveries occur within their on- time windows. Also, for each delivery node, the range of delivery times to be expected.

617 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Solve the Multiple Vehicle Routing Problem Under Uncertainty (FP8) A user wants to create a set of routes (tours) for pickups and deliveries that minimizes the number of drivers required, while ensuring that the drivers return to the depot no later than a set time. The earliest time the trucks can leave and the latest they can return are both constrained. This use case extends Use Case FP7 to multiple trucks and imposes a return time. Parcel delivery companies solve this problem every day. They determine how many trucks are needed and what packages go on what trucks so that the drivers return within the allowed time. This analysis helps parcel carriers create routes that use their drivers and vehicles as efficiently as possible, while avoiding potential costs (e.g., overtime). Table D.72 presents the key question, steps, inputs, and the result for this use case. Alter Delivery Schedules in Real Time (FP9) In this use case, a trucking company needs to alter the delivery schedules for one or more trucks in real time due to some type of unexpected event. For example, an inci- dent on a freeway might have created severe delays on one or more of the paths the trucks were planning to use. The trucking company would note the estimated severity of the incident, identify which delivery vehicles are affected (based on their present positions and deliveries yet to be made), and generate new tours for each to minimize the impact of the incident on the on-time delivery likelihoods. This analysis is particu- larly helpful to parcel carriers that would have to revise their tours due to unexpected sources of delay that occur on the roadway system while deliveries are in progress. Table D.73 presents the key question, steps, inputs, and the result for this use case. TABLE D.72. SOLVE THE MULTIPLE VEHICLE ROUTING PROBLEM UNDER UNCERTAINTY (FP8) User Trucking company Question How many trucks and drivers are needed to make a set of deliveries, what departure times should be used, and how should the trucks be routed to maximize on-time deliveries and pickups? Steps 1. Select the locations for pickups and deliveries. 2. Identify the on-time windows for pickups and deliveries that have them. 3. Assemble TT-PDFs for the paths between the places to be visited. If the TT-PDFs for the paths vary by time of day, then assemble TT-PDFs for each departure time for each path. 4. Solve the stochastic multiple vehicle routing problem. Inputs A database of TT-PDFs for truck trips on paths between the locations to be visited (including the depot), differentiated by departure time when necessary. Result A set of tours (pickup and delivery sequences) that minimizes the number of trucks and drivers required while maximizing the on-time delivery percentages and complying with the maximum allowed return. Also, for each destination, the range of pickup or delivery times to be expected.

618 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Freight Customers The following use cases demonstrate monitoring system functionalities that are helpful for freight customers. Minimize Shipping Costs Due to Unreliability (FC1) A user wants to ship packages from an origin to a destination with minimum costs due to unreliability. That means there is a cost for being early and late, similar to a utility function. This analysis helps customers identify the shipping methods that will meet their specifications at the lowest cost. Table D.74 presents the key question, steps, in- puts, and the result for this use case. TABLE D.73. ALTER DELIVERY SCHEDULES IN REAL TIME (FP9) User Trucking company Question How should the existing multivehicle delivery schedule be altered when there is a disruption in the network? Steps 1. Identify the locations to which deliveries need to be made for each vehicle. 2. Identify those vehicles that are affected by the disruption. 3. Reaffirm the on-time windows for each remaining delivery. 4. Update the TT-PDFs for the paths between the places to be visited based on the network disruption. If the paths and TT-PDFs are time dependent, then differentiate by departure time. 5. Solve the stochastic vehicle routing problem for the remaining delivery locations based on the updated TT-PDFs. Inputs Updated TT-PDFs for truck trips on paths from one location to another, differentiated by departure time if necessary. Result A new delivery schedule that maximizes the on-time probabilities. Also, for each destination, the range of delivery times to be expected. TABLE D.74. MINIMIZE SHIPPING COSTS DUE TO UNRELIABILITY (FC1) User Freight customer Question How can the shipping costs due to unreliability be minimized? Steps 1. Select the origin and destination. 2. Assemble costs and the PDFs for travel times for various shipping options. 3. Identify the costs of being early and late by certain amounts. 4. Compute the expected cost of each option based on its TT-PDF. Inputs A database of TT-PDFs for various shipping options and the costs of being early and late by certain amounts. Result The shipping option with the lowest cost due to unreliability.

619 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Step 1 involves selecting the origin and destination. In this use case, an origin and destination between which there are four routes for delivery options are assumed. Step 2 involves assembling costs and the PDFs for travel times for various shipping options. In this use case, four routes are assumed, each of which has a specific travel time distribution for delivery options. The assumed on-time window is 100 to 110 minutes. Figure D.57 shows the CDFs for travel times for the four routes. Figure D.57 shows that the four options have different minimum travel times, but they do not vary by much. In this step, the cost for traveling must be calculated, as well as the cost for being early or late. The travel time for the whole trip, including arrivals and returns, must be obtained. The same method for getting the arrival travel times is used to get the total travel times for the four routes, as shown in Figure D.58. To get the cost for total travel time, it is assumed that the rate is $0.01/minute. Figure D.59 shows that the cost for travel time is consistent with total travel time. Options with less travel time have less travel time cost. Step 3 involves identifying the costs of being early and late by certain amounts. For arrival travel time, it is assumed the on-time window is from 100 minutes to 110 minutes. The penalty rate for being early is $1/minute, but the penalty rate for being late is $2/minute. Figure D.60 shows the cost for penalties of the four options. Figure D.60 shows that for almost 93% probability, Option 3 is the best option among the four options, but Option 4 is the best option when probability is larger than 93%. Figure D.57. CDFs of minimum travel times for four routes. (The assumed on-time window is 100 to 110 minutes.) 216 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Step 2 involves assembling costs and the PDFs for travel times for various shipping options. In this use case, four routes are assumed, each of which has a specific travel time distribution for delivery options. The assumed on-time window is 100 to 110 minutes. Figure D.57 shows the CDFs for travel times for the four routes. [Insert Figure D.57] [caption] Figure D.57. CDFs of travel times for four routes. (The assumed on-time window is 100 to 110 minutes.) Figure D.57 shows that the four options have different minimum travel times, but they do not vary by much. In this step, the cost for traveling must be calculated, as well as the cost for being early or late. The travel time for the whole trip, including arrivals and returns, must be obtained. The same method for getting the arrival travel times is used to get the total travel times for the four routes, as shown in Figure D.58.

620 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY 217 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx [Insert Figure D.58] [caption] Figure D.58. Total travel times for four routes. To get the cost for total travel time, it is assumed that the rate is $0.01/minute. Figure D.59 shows that the cost for travel time is consistent with total travel time. Options with less travel time have less travel time cost. [Insert Figure D.59] [caption] Figure D.58. Total travel times for four routes. 218 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.59. Cost for total travel time. Step 3 involves identifying the costs of being early and late by certain amounts. For arrival travel time, it is assumed the on-time window is from 100 minutes to 110 minutes. The penalty rate for being early is $1/minute, but the penalty rate for being late is $2/minute. Figure D.60 shows the cost for penalties of the four options. [Insert Figure D.60] [caption] Figure D.59. Cost for total travel time. Step 4 involves computing the expected cost of each option based on its TT-PDF. The cost is calculated for the whole trip, including the penalty rate for being late or early, as well as arrival travel time and return travel time. Figure D.61 shows the cumu- lative distribution function.

621 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY From Figure D.61, it can be seen that when probability is below 40%, the best option is Option 4. When probability is greater than 40% and smaller than 70%, the best option is Option 2, but Option 3 is best when probability is 75% to 95%. When probability is near 100%, either Option 4 or Option 1 is preferable. Figure D.60. Cost for on-time penalties. 219 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.60. Cost for on-time penalties. Figure D.60 shows that for almost 93% probability, Option 3 is the best option among the four options, but Option 4 is the best option when probability is larger than 93%. Step 4 involves computing the expected cost of each option based on its TT-PDF. The cost is calculated for the whole trip, including the penalty rate for being late or early, as well as arrival travel time and return travel time. Figure D.61 shows the cumulative distribution function. [Insert Figure D.61] [caption] Figure D.61. Cost for the whole trip (travel time and penalty). 220 2014.04.23 15 L02 Guide Appendix D_Final for composition.docx Figure D.61. Cost for the whole trip (travel time and penalty). From Figure D.61, it can be seen that hen probability is below 40%, the best option is Option 4.When probability is greater than 40% and smaller than 70%, the best option is Option 2, but Option 3 is best when probability is 75% to 95%. When probability is near 100%, either Option 4 or Option 1 is preferable. <H3>Determine Storage Space for Just-in-Time Deliveries (FC2) A user wishes to set up a just-in-time delivery system and wants to know how much storage space is needed at the receiving location. Because shipments may arrive late, some stock needs to be carried in inventory; and since shipments may be early, some storage space for those arrivals needs to be provided. This information helps size the facility effectively. Table D.75 presents the key question, steps, inputs, and the result for this use case.

622 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Determine Storage Space for Just-in-Time Deliveries (FC2) A user wishes to set up a just-in-time delivery system and wants to know how much storage space is needed at the receiving location. Because shipments may arrive late, some stock needs to be carried in inventory; and since shipments may be early, some storage space for those arrivals needs to be provided. This information helps size the facility effectively. Table D.75 presents the key question, steps, inputs, and the result for this use case. Find the Lowest-Cost Reliable Origin (FC3) A user needs items shipped to a location by a specified time while minimizing ship- ping costs. This use case is similar to Use Case FC1, but more than one origin is pos- sible. This information helps a customer assess the trade-off between cost and delivery reliability. Table D.76 presents the key question, steps, inputs, and the result for this use case. TABLE D.75. DETERMINE STORAGE SPACE FOR JUST-IN-TIME DELIVERIES (FC2) User Freight customer Question How much storage space is needed given the reliability of deliveries for a just-in-time manufacturing facility? Steps 1. Select the manufacturing facility of interest. 2. Assemble PDFs for the deviations from intended delivery times. Separate these by shipment size and origin if there are different reliabilities. 3. Determine a tolerable stock-out probability. 4. Build a Monte Carlo model of the storage facility given the PDFs for delivery time deviations and determine how large the on-site storage needs to be to ensure that the stock-out probability is met. Inputs A database of PDFs for deviations from intended delivery times separated by shipment sizes and origin. Result A three-dimensional trade-off surface between travel time, on-time performance, and shipping cost. Given weights among these three, the optimal shipping option can be selected. TABLE D.76. FIND THE LOWEST-COST RELIABLE ORIGIN (FC3) User Freight customer Question What origin provides the most reliable deliveries at an acceptable cost? Steps 1. Select the destination of interest and the possible origins. 2. Define reliable delivery (e.g., percentage within the on-time window). 3. Define two cost functions: one related to travel time and the other related to whether the shipment is early or late. 4. Assemble TT-PDFs and PDFs for the deviations from intended delivery times by origin. 5. Select the origin that has the best CDF for cost. Inputs A database of PDFs for travel times and deviations from intended delivery times by origin. Result The origin with the best CDF in terms of cost.

623 GUIDE TO ESTABLISHING MONITORING PROGRAMS FOR TRAVEL TIME RELIABILITY Find the Warehouse Site with the Best Distribution Reliability (FC4) A user wants to site a distribution center so that it maximizes delivery reliability to the destinations it serves. Several sites are possible. The number of truck trips to each location served is important because the best choice needs to account for the trip fre- quency. The times of day are also important because the travel time reliability varies depending on when the deliveries take place. Table D.77 presents the key question, steps, inputs, and the result for this use case. REFERENCES 1. Boyle, D., J. Pappas, P. Boyle, B. Nelson, D. Sharfarz, and H. Benn. TCRP Report 135: Controlling System Costs: Basic and Advanced Scheduling Manuals and Contemporary Issues in Transit Scheduling. Transportation Research Board of the National Academies, Washington, D.C., 2009. 2. Kittelson and Associates, Inc., KFH Group, Inc., Parsons Brinckerhoff Quade and Douglass, Inc., and K. Hunter-Zaworski. TCRP Report 100: Transit Capacity and Quality of Service Manual, 2nd ed. Transportation Research Board of the National Academies, Washington, D.C., 2003. 3. Bowman, L. A., and M. A. Turnquist. Service Frequency, Schedule Reliability and Passenger Wait Times at Transit Stops. Transportation Research Part A, Vol. 15A, No. 6, 1981, pp. 465–471. TABLE D.77. FIND THE WAREHOUSE SITE WITH THE BEST DISTRIBUTION RELIABILITY (FC4) User Freight customer Question What warehouse location has the best delivery reliability? Steps 1. Select the possible warehouse sites, the destinations to be served, the number of truck trips per week to those destinations, and the times when the deliveries would take place. 2. Assemble TT-PDFs for trips from the warehouse sites to the destinations for the times when the deliveries would take place. 3. Identify the warehouse site that maximizes the likelihood that trucks will be on time in reaching the destinations. Inputs A database of TT-PDFs for truck trips from the warehouse sites to the destinations for the times when the deliveries would take place. Result A rank ordering of the warehouse sites based on the truck trip–weighted reliability of reaching the destinations on time.