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.
69 CHAPTER 3 Analysis of Existing Reliability Based on Data 3.1 Introduction The existing reliability of the tested facilities was assessed based on real-world data obtained from multiple sources by using the L02 project guidelines and additional measures and visualization techniques combined with other performance measures and assessment techniques. For the I-95 GPL and EL, the estimates were based on data collected from infrastructure-based traffic detectors (true-presence microwave detectors) located at one-third to one-half mile spacing. Infrastructure devices were not available for SR-7 to provide data for use in travel time estimation. Therefore, travel time data from a private-sector data provider (INRIX) were used in estimating the travel time reliability of SR-7. The travel time measurements were combined with nonrecurrent event data (e.g., weather, incidents, construction, and special events) to estimate the impacts of these factors on reliability. This chapter presents the reliability analysis only for the I- 95 NB GPL. Due to the large number of graphs and in order to improve the readability of the report, the analyses of the I-95 NB EL, I-95 SB GPL, I-95 SB EL, SR-7 NB, and SR-7 SB are presented in Appendices B, C, D, E, and F, respectively. However, a summary of the results of the analysis of all segments is presented in this chapter. 3.2 I-95 Northbound General-Purpose Lane Analysis 3.2.1 Overall Reliability Performance Initially, this project used L02 recommendations of categorizing the data by congestion level. However, it was later decided to bin the data by time of day as considering different causes of congestion at different times of the day was more appropriate because it allowed analyzing the reliability of the specific condition under consideration. Figures 3.1, 3.2, and 3.3 present an assessment of the overall system performance for the I-95 NB GPL for different times of the day. The travel time rate CDF in Figure 3.1 clearly shows the unreliability of travel during the PM peak periods. The PM peak period was subdivided into two peaks, PM1 from 3:00 to 5:00 p.m. and PM2 from 5:00 to 7:00 p.m., to reflect the differences in the observed congestion patterns and the different causes of congestion in these two periods. In Figure 3.1, it appears that the 95th percentile travel time rate in the PM peak was close to 162 second/mile (about 21 mph) compared to about 62 second/mile (58 mph) for free- flow conditions, reflecting a 95th percentile TTI of 2.6 and indicating highly unreliable travel during the PM peak. The reliability was good in the remaining periods, although the 95th percentile travel time rate appeared to be somewhat high in the after-PM (APM) period between 7:00 and 10:00 p.m., resulting in a TTI of about 1.75. MD in the figure refers to the midday period, and MN refers to the midnight and early morning period.
70 Figure 3.1. CDFs for I-95 NB GPL. The percentage of unreliability contribution (see Figure 3.2) shows that the unreliability in the PM1 and PM2 peaks as measured by the semivariance contributed to 29% and 45%, respectively, of the overall unreliability of the daily operations, with a total of 74% for the combination of these two peaks. The remaining AM, midday, and APM periods contributed to 5%, 11%, and 9%, respectively, of the total unreliability according to the semivariance measure. Some of the unreliability of the midday and APM periods occurred at the boundary of the PM peak period.
71 Figure 3.2. Percentage of unreliability contribution for I-95 NB GPL. The CDF curves in Figure 3.1 are crowded, so to provide a more straightforward visualization of reliability results, the travel time, speed, and TTI that correspond to the 50th, 80th, and 95th percentiles of CDFs for different times of day are presented in Figure 3.3. The results show that the travel time under the 50th and 80th percentile travel time was about six minutes for most of the day. However, the 50th percentile travel time increased to nine minutes and the 80th percentile increased to 13 minutes for the PM peak. The travel time for the 95th percentile condition increased from seven to eight minutes to 16 minutes. The corresponding 95th percentile speed in the PM peak was around 20 mph, and the TTI was around 2.6.
72 (a) (b) (c) Figure 3.3. I-95 NB GPL (a) travel time, (b) speed, and (c) TTI.
73 The variation of the TTI by five-minute intervals shown in Figure 3.4a confirms that, excluding the PM peak periods, the 95th percentile TTI was good for most of the day, except in the AM peak between 7:30 and 8:45 a.m., when it reached 1.4 to 1.5. This figure also shows that the 80th and 95th percentile TTIs started increasing between 2:00 and 3:00 p.m., reaching 1.9 and 2.5, respectively, by 4:00 p.m. Between 4:45 and 6:45 p.m., the 80th and 95th percentile TTIs increased to 2.3 and 3.0, respectively, then started decreasing sharply, reaching 1.0 and 1.4 by 8:00 p.m. The median travel time was 1.5 to 1.8 of the free-flow travel time in the PM peak periods between 3:30 and 6:30 p.m. Between 9:00 a.m. and 1:45 p.m. and after 9:00 p.m., the 95th percentile TTI ranged between 1.00 and 1.20. These results indicate that the period between 3:00 and 8:00 p.m. could be considered unreliable, with major influence of outliers according to L05 guidance. Figure 3.4a shows that the AM peak was moderately unreliable, while the rest of the day was reliable. The policy index was also calculated assuming a target speed of 40 mph, as shown in Figure 3.4b. This measure confirms that the period between about 2:00 and 7:00 p.m. was unreliable.
74 (a) (b) Figure 3.4. I-95 NB GPL (a) TTIs and (b) policy index variation by time of day.
75 Figure 3.5 shows that the on-time performance (both the 1.1 and 1.25 on-time performance) was close to or above 90% between 8:00 a.m. and 2:00 p.m., but dropped from 52% to 62% and 60% to 75%, respectively, between 2:00 and 7:30 p.m., with the highest drop appearing to be between 5:00 and 6:45 p.m. As stated earlier, the Florida DOT central planning office prefers the use of on-time performance and PTI to assess reliability. The misery index, semistandard deviation (shown as Semi-SD in Figure 3.6), and buffer index were worst between 3:00 and 7:00 p.m., although they were also relatively high between 7:00 and 8:30 p.m., as shown in Figure 3.6. The high values in the after-PM peak period possibly reflect the observed higher severity of crashes during this period, as discussed below. The misery index reflects the average travel time of the worst 5% of travel, indicating the high impacts of severe events. Figure 3.5. I-95 NB GPL on-time performance.
76 Figure 3.6. I-95 NB GPL other performance measures. Figures 3.7 and 3.8, respectively, show the five-minute variation of VMT and vehicle hours traveled (VHT) in 24 hours. Figure 3.7 shows that I-95 NB GPL had a relatively high VMT from 7:00 a.m. to 7:00 p.m., even though there was a slight drop in VMT during the midday. However, the curve in Figure 3.8 shows a big difference in VHT during different times of the day. In addition to the reliability measures mentioned above, the reliability rating defined in L08 as the percentage of VMT with a TTI less than 1.33 was also calculated. Figure 3.9 presents the percentage of VMT under given values of TTIs. The results in this figure indicate that for the whole day only 54% of VMT had a 95th percentile TTI less than 1.33. If the thresholds are increased to 1.5 and 2, these percentages increase to 67% and 74%, respectively.
77 Figure 3.7. I-95 NB GPL VMT variation by time of day. Figure 3.8. I-95 NB GPL VHT variation by time of day.
78 (a) (b) Figure 3.9. I-95 NB GPL reliability rating for comparison based on (a) 80th and (b) 95th percentile TTI. 3.2.2 Contributions of Influential Factors The impacts of various factors on travel time reliability were examined in this study following the procedures outlined in the L02 project. Figure 3.10 presents the CDF distributions for travel time rate under different traffic conditions. Tables 3.1, 3.2, and 3.3, respectively, summarize the percentages of occurrence, severity, and overall contribution of the no-event traffic condition (including normal traffic and high-demand conditions), incident, weather, and incident plus weather to travel time reliability. An important observation from Table 3.3 is that the no-event periods contributed significantly to the unreliability of the system in the PM peak (10% in PM1 and 21% in PM2,
79 with a total of 31% of the whole day). This level of contribution indicates serious issues with the recurrent operation. It should be mentioned that in reality the contribution of recurrent congestion was higher as even during incident intervals in the congested peak, part of the congestion or unreliability was due to the congestion due to recurrent capacity constraints. The reliability analysis also indicated that incidents were a major contributor to travel time reliability for most of the day. Table 3.3 indicates that the five-minute intervals with incidents contributed to 15% and 21% of the unreliability of the day during the PM1 and PM2 periods. However, after correcting for the no-event congestion contribution during the incidents, the contribution of incidents appeared to be 10% and 8.4% of the daily unreliability during the PM1 and PM2 periods, respectively. These percentages are the same contribution as the no-event in the PM1 period (10%) but lower than the no-event contribution during the PM2 period (8.4% versus 21%), again indicating the severity of the recurrent congestion during the PM2 period. However, Table 3.2 shows that a single average incident event caused more damage than a single average no-event. During the rest of the day, the impacts of incidents based on the semivariance were clearly lower. Weather events were relatively rare compared to incident occurrence, and the overall contribution of weather was much smaller than that due to incident and no-event high-demand conditions, as shown in Table 3.1 and Table 3.3. However, a single weather event impact measured in NSV was almost the same as the impact of a single incident in the PM1 and PM2 periods (see Table 3.2). Incidents plus weather events were even rarer than weather events; therefore, their contributions to the overall reliability were low. However, the incident plus weather combination generated the worst conditions, as indicated by the travel time distributions shown in Figure 3.10 and the NSV values shown in Table 3.2. This table shows that during PM1 and PM2, the NSV during incident plus weather events was about twice the NSV during incident conditions and also about twice the NSV during rainy conditions. It is interesting to see that the impact of a single incident plus weather event was also very high in the AM peak and relatively high in the midday period. Table 3.1. Percentage of Occurrence Time Period Nonevent Incident Weather Incident + Weather Total AM 10% 2% 1% 0% 13% MD 20% 6% 1% 0% 27% PM1 4% 2% 0% 0% 6% PM2 5% 3% 0% 0% 8% APM 9% 3% 0% 0% 12% MN 27% 4% 1% 0% 33%
80 Table 3.2. Percentage of Severity Time Period Nonevent Incident Weather Incident + Weather Total AM 0% 0% 1% 17% 19% MD 0% 1% 1% 5% 7% PM1 3% 9% 7% 17% 35% PM2 5% 8% 8% 14% 35% APM 0% 3% 0% 1% 4% MN 0% 0% 0% 0% 0% Table 3.3. Percentage of Unreliability Contribution Time Period Nonevent Incident Weather Incident + Weather Total AM 2% 1% 0% 3% 5% MD 3% 6% 1% 2% 11% PM1 10% 15% 1% 2% 29% PM2 21% 21% 2% 1% 45% APM 2% 7% 0% 0% 9% MN 1% 0% 0% 0% 1% Figure 3.10 CDFs confirm that the worst conditions in each peak occurred during incident plus bad weather events and that the no-event conditions during the PM1 and particularly the PM2 peaks were bad. Incident and weather impacts were also clear. In addition to showing the results in CDF format, Figure 3.11, Figure 3.12, and Figure 3.13 summarize results in additional ways for the purpose of helping transportation agencies better visualize the results. Figure 3.11 shows the values of travel time for 50th, 80th, and 95th percentiles, and Figure 3.12 and Figure 3.13 show the corresponding speeds and TTIs, respectively.
81 (a) (b)
82 (c) (d)
83 (e) (f) Figure 3.10. CDF by regimes for I-95 NB GPL for (a) AM peak, (b) MD, (c) PM1, (d) PM2, (e) APM peak, and (f) MN periods.
84 (a) (b)
85 (c) (d)
86 (e) (f)
87 (g) Figure 3.11. I-95 NB GPL travel times for (a) AM peak, (b) MD, (c) PM1, (d) PM2, (e) APM, (f) MN, and (g) all time periods.
88 (a) (b)
89 (c) (d)
90 (e) (f)
91 (g) Figure 3.12. I-95 NB GPL speeds for AM peak, (b) MD, (c) PM1, (d) PM2, (e) APM, (f) MN, and (g) all time periods.
92 (a) (b)
93 (c) (d)
94 (e) (f)
95 (g) Figure 3.13 I-95 NB GPL TTIs for (a) AM peak, (b) MD, (c) PM1, (d) PM2, (e) APM, (f) MN, and (g) all time periods. In addition to the above visualization and analysis techniques, other reliability performance measures were estimated for every five minutes of the day. These estimations were done based on TMC operation staff requirements for fine-grained analysis of reliability. The additional measures included the mean and 50th, 80th, and 95th percentile TTIs; semistandard deviation; buffer index; skew statistics; on-time performance (based on 1.1 and 1.25 thresholds); and the misery index, as shown in Figure 3.14 to Figure 3.23, respectively. The variation of the mean TTI by five-minute intervals during the 24 hours of the day (see Figure 3.14) shows that with no incidents and when the demand did not exceed the high- demand threshold, the mean TTI was 1.3 to 1.4 (i.e., 30% to 40% higher in travel time than the free-flow travel time) between 3:00 and 5:00 p.m., increased to about 1.45 between 5:00 and 5:35 p.m., and then dropped to 1.15 by 6:20 p.m. During higher-demand days, the mean TTI between 3:00 and 5:00 p.m. was about 1.54 but increased to 1.78 between 5:00 and 6:30 p.m. Therefore, higher-demand days not only increased the TTI but also elongated the period during which the TTI was high. A similar trend can be seen in Figure 3.17 for the 95th percentile TTI. The high-demand contribution to the 95th percentile TTI was small until 5:00 p.m., indicating that the worst 5% travel was caused mainly by other events. Between 5:00 and 7:00 p.m., however, the high demand increased the 95th percentile TTI from 2.35 for normal-demand congested conditions to 2.70. Incidents appeared to be the main contributors to the 95th percentile TTI in the early PM peak and the rest of the day. However, between 5:00 and 7:00 p.m., the impact of high demand during the no-event day was significant. Figure 3.13 also confirms that during the normal- demand period, the 95 percentile TTI still had a high value. This figure also shows that there was
96 some effect of weather on reliability, particularly during PM1 and PM2, but to a lesser degree than incidents and high demand. Another interesting finding from the figures is that the impacts of the influencing factors on the 95th percentile TTI and 80th percentile TTI were much higher than the impacts of the influencing factors on the median or mean TTI. This finding is important as it allows stronger justifications of advanced strategies to address factors such as incidents, weather, and fluctuations in demands. For example, when considering the mean TTI at 4:00 p.m., the value was 1.41 under normal demand and no-event and 1.62 under all conditions combined (a difference of about 15%). The corresponding values for the 95 percentile TTI were 1.9 and 2.5, respectively (a difference of about 31%). Another observation is that the TTIs were still high even for normal conditions, indicating the potential impacts of external factors not accounted for in the analysis such as backups from off-ramps and downstream incidents, events on the ELs or opposing traffic, diversion from other routes, seasonal variations, and unrecorded weather and special events. This observation is important because it may explain that the FREEVAL-RL tool may sometimes underestimate unreliability because it does not account for all real-world events, as shown in a later chapter. Figure 3.14. Mean TTI comparison for I-95 NB GPL.
97 Figure 3.15. 50th percentile TTI comparison for I-95 NB GPL. Figure 3.16. 80th percentile TTI comparison for I-95 NB GPL.
98 Figure 3.17. 95th percentile TTI comparison for I-95 NB GPL. Figure 3.18. Semistandard deviation comparison for I-95 NB GPL.
99 Figure 3.19. Buffer index comparison for I-95 NB GPL. Figure 3.20. Skew statistics comparison for I-95 NB GPL.
100 Figure 3.21. On-time performance comparison (based on 1.1) for I-95 NB GPL. Figure 3.22. On-time performance comparison (based on 1.25) for I-95 NB GPL.
101 Figure 3.23. Misery index comparison for I-95 NB GPL. The detailed impacts of high demand, incident, and weather were further analyzed in this study, and the corresponding results are presented in the next section. 3.2.3 Impact of Normal versus High Demands The no-event period was further classified into two categories, high-demand and normal-demand periods, to study the impacts of high demand. The categorizing was performed based on the procedure developed in the L02 project to determine the threshold that differentiates normal demands from high demands. The travel time rate CDF curves and the associated percentiles in Figure 3.24 show significant impacts of high demand, particularly during the PM2 periods. This finding indicates that implementing active traffic and demand management when a certain threshold of demand is exceeded has the potential of improving system reliability. It is interesting to note again, however, that based on the CDF curves, even with normal demand in the PM peak periods, the 95th percentile travel time rate was about twice the free-flow travel time rate, probably indicating that there were impacts of events from outside of the system or events not accounted for. The occurrence, severity, and unreliability contribution results as shown in Table 3.4, Table 3.5, and Table 3.6 indicate that the five-minute intervals with high demand based on the derived threshold contributed significantly to the unreliability of the no-event period. The contribution of high-demand intervals to whole-day unreliability was 18% and 41% in PM1 and PM2, respectively, compared to 8% and 13% corresponding values for the normal-demand periods (Table 3.6). When normalized by frequency to determine the severity of the impact of a
102 single interval, the high-demand interval NSV was 17% and 36% in the PM1 and PM2 periods, compared to corresponding 9% and 9% for normal-demand intervals, respectively (Table 3.5). (a) (b)
103 (c) (d)
104 (e) (f) Figure 3.24. CDF by regimes for I-95 NB GPL for normal and demand (a) AM, (b) MD, (c) PM1, (d) PM2, (e) APM, and (f) MN time periods.
105 Table 3.4. Percentage of Occurrence Time Period Demand Normal Total AM 1% 13% 13% MD 2% 25% 27% PM1 3% 2% 5% PM2 3% 4% 7% APM 1% 11% 12% MN 2% 35% 37% Table 3.5. Percentage of Severity Time Period Demand Normal Total AM 11% 0% 12% MD 7% 0% 7% PM1 17% 9% 26% PM2 36% 9% 45% APM 10% 1% 10% MN 0% 0% 0% Table 3.6. Percentage of Unreliability Contribution Time Period Demand Normal Total AM 3% 2% 4% MD 4% 3% 7% PM1 18% 8% 27% PM2 41% 13% 54% APM 2% 4% 6% MN 0% 1% 1% To select capacity improvements and/or active traffic management strategies, it is not sufficient to identify congestion and unreliability values and perform a general analysis of the contributing factors. Analyzing the data and visualizing the bottleneck impacts using contour (heat) maps, as shown in Figure 3.25, indicated that the main issues in the PM1 peak were two capacity problems on NW 79th Street and NW 103rd Street. The capacities on these links were found to be lower that of the capacity reported by the HCM. For the PM2 period, the main issue was a backup from the off-ramp to the Florida Turnpike.
106 Figure 3.25. I-95 NB GPL speed contour map during the PM peak period. 3.2.4 Incident Severity Impacts The incident impacts were further analyzed by the level of lane blockage for the AM, midday, PM, and APM periods. The CDF results, presented in Figure 3.26, show that the contribution of lane blockage incidents, as expected, was much higher than a single average incident. This finding is also reflected by the travel time rate CDFs, which show the highest tilt of two or more (2+) laneâblocking incidents during the PM peak, then 2+ laneâblocking incidents in the APM peaks, followed by 2+ laneâblocking incidents during the midday, one-laneâblocking incidents during the PM, and one-laneâblocking incidents in the midday and APM periods. The I-95 NB GPL incident that occurred under the good weather conditions in the AM peak did not seem to have had a significant influence. Tables 3.7 to 3.9, respectively, present the occurrence, severity, and unreliability contribution of different lane-blocking incidents. It is seen from Table 3.9 that the contribution of lane-blocking incidents was twice as much as non-lane-blocking incidents, although non-lane- blocking incident frequency was much higher than that of lane blockage frequency. On a single- event basis, the NSV indicated the high impacts of 2+ lane blockage. Additional analysis results not shown in this report indicated that crash incidents were generally more damaging than other (noncrash) incidents. During the PM period, the damage due to a single noncrash incident was about 50% of the damage caused by a crash incident. However, the overall impact on reliability was equivalent due to the higher frequency of noncrash incidents. To better understand the incident impacts, the temporal and spatial incident frequency variation needs to be determined to allow better selection of advanced incident management strategies. These variations can be visualized as shown in Figures 3.27 and Figure 3.28, which show that the crash incident frequency was clearly the highest in the PM peak period from 4:40 to 7:00 p.m. The noncrash incidents were high most of the day, but they had a relatively flat peak between 2:00 and 8:00 p.m. These figures also show that, when the investigated facility was segmented to three segments, the highest frequency of incidents occurred at Segment 3, which
107 was the most downstream segment (between NW 103 Street and the turnpike exit). Segment 2 (between NW 79th Street and NW 103rd Street) had the second-highest crash incident frequency. (a) (b)
108 (c) (d)
109 (e) (f) Figure 3.26. CDF by regimes for I-95 NB GPL for (a) AM, (b) MD, (c) PM1, (d) PM2, (e) APM, and (f) MN periods.
110 Table 3.7. Percentage of Occurrence Time Period Nonincident 0 Lanes Blocked 1 Lane Blocked 2 Lanes Blocked 3+ Lanes Blocked Total AM 10.6% 1.6% 0.3% 0.1% 0.0% 12.6% MD 21.2% 4.8% 0.9% 0.2% 0.0% 27.1% PM1 4.1% 1.5% 0.4% 0.1% 0.1% 6.2% PM2 5.2% 2.0% 0.8% 0.2% 0.1% 8.3% APM 9.3% 2.4% 0.5% 0.2% 0.1% 12.4% MN 29.0% 3.7% 0.4% 0.2% 0.1% 33.4% Table 3.8. Percentage of Severity Time Period Nonincident 0 Lanes Blocked 1 Lane Blocked 2 Lanes Blocked 3+ Lanes Blocked Total AM 0.1% 0.0% 0.4% 1.0% 8.6% 10.1% MD 0.1% 0.1% 0.8% 5.2% 8.2% 14.2% PM1 1.1% 1.2% 2.1% 20.3% 13.5% 37.2% PM2 1.7% 1.4% 3.3% 11.8% 4.5% 21.0% APM 0.1% 0.1% 0.9% 8.6% 7.8% 17.4% MN 0.0% 0.0% 0.0% 0.1% 0.0% 0.1% Table 3.9. Percentage of Unreliability Contribution Time Period Nonincident 0 Lanes Blocked 1 Lane Blocked 2 Lanes Blocked 3+ Lanes Blocked Total AM 4.1% 0.1% 0.7% 0.4% 0.2% 1.3% MD 6.7% 2.4% 3.3% 4.8% 1.4% 11.9% PM1 23.1% 8.9% 4.4% 9.7% 7.3% 30.2% PM2 44.9% 14.2% 13.2% 13.5% 1.5% 42.4% APM 4.6% 1.5% 2.1% 7.0% 3.1% 13.8% MN 1.1% 0.1% 0.1% 0.1% 0.0% 0.3%
111 Figure 3.27. Percentage of crash incidents per mile comparison.
112 Figure 3.28. Percentage of noncrash incidents per mile comparison. Figure 3.29 presents the 95th percentile TTI under three scenarios. Scenario 1 includes all traffic conditions (shown as I-95 NB ALL in the figure). Scenario 2 corresponds to the traffic conditions without lane-blocking incidents (shown as I-95 NB ALL â Lane-Blocked Incidents in the figure). Scenario 3 refers to the conditions after further removal of all remaining incidents (shown as I-95 NB ALL - Incidents in the figure). Figure 3.29 further indicates the significant impacts of lane-blocking incidents on the PTI (95th percentile TTI) between 2:30 and 8:00 p.m. During PM1 (3:00 to 5:00 p.m.), the 95th percentile TTI increased from about 2.0 to about 2.5 due to lane-blocking events. In PM2 (5:00 to 7:00 p.m.), the conditions without incidents were already unreliable mainly due to backups from the turnpike, yet incidents increased the peak 95th percentile from 2.6 to 3.0. Lane-blocking incidents also resulted in an increase in the mean TTI, but to a lesser degree (e.g., from 1.58 to 1.68 at 4:00 p.m.), as shown in Figure 3.30. However, only small impacts of lane-blocking incidents can be observed from Figures 3.31 and 3.32 on the median and 80th percentile TTIs. The higher impacts of incidents on the 95th percentile clearly indicate the need to consider reliability when assessing the impacts of incidents and incident management benefits.
113 Figure 3.29. 95th percentile TTI and lane-blocking incidents comparison for I-95 NB GPL. Figures 3.33 to 3.36 present the impacts of lane-blocking incident duration on reliability performance measures. It appears from these figures that higher-duration incidents did not contribute to unreliability more than shorter-duration incidents. All lane-blocking incidents, even those with durations less than 30 minutes, had a high impact on reliability. This effect may have been due in part to the higher frequency of these shorter-incident duration events.
114 Figure 3.30. Mean TTI and lane-blocking incidents comparison for I-95 NB GPL. Figure 3.31. 50th percentile TTI and lane-blocking incidents comparison for I-95 NB GPL.
115 Figure 3.32. 80th percentile TTI and lane-blocking incidents comparison for I-95 NB GPL. Figure 3.33. Mean TTI and incident duration comparison for I-95 NB GPL.
116 Figure 3.34. 50th percentile TTI and incident duration comparison for I-95 NB GPL. Figure 3.35. 80th percentile TTI and incident duration comparison for I-95 NB GPL.
117 Figure 3.36. 95th percentile TTI and incident duration comparison for I-95 NB GPL. 3.2.5 Weather Impacts Further analysis was conducted to examine the impacts of the severity of precipitation on travel time reliability; the results are presented in Figures 3.37 to 3.41 and Tables 3.10, 3.11, and 3.12. As shown in these tables and figures, a single moderate-to-heavy rain event caused a significant impact, with resulting NSV that was 20% to 50% higher than the no-weather event in the PM peak. Table 3.11 also seems to indicate that heavy rain in the AM peak also caused high contribution per event. However, as stated earlier, the overall contribution of weather to unreliability appeared to be relatively small due to the low number of rain events compared, for example, to incident events. Another important factor to be considered is that the weather stations from which the weather service information providers collect data are distributed around the region and were not located at the study facility. This distribution of weather data may have reduced the accuracy of the assessment of weather events on reliability.
118 (a) (b)
119 (c) (d)
120 (e) (f) Figure 3.37. CDF by regimes for I-95 NB GPL for (a) AM, (b) MD, (c) PM1, (d) PM2, (e) APM, and (f) MN periods.
121 Table 3.10. Percentage of Occurrence Time Period Nonweather Light Rain Moderate Rain Heavy Rain Total AM 12% 0% 0% 0% 13% MD 26% 1% 0% 0% 27% PM1 6% 0% 0% 0% 6% PM2 8% 0% 0% 0% 8% APM 12% 0% 0% 0% 12% MN 32% 1% 0% 0% 33% Table 3.11. Percentage of Severity Time Period Nonweather Light Rain Moderate Rain Heavy Rain Total AM 0% 0% 1% 9% 10% MD 1% 1% 1% 1% 3% PM1 8% 8% 10% 17% 43% PM2 10% 12% 15% 5% 41% APM 1% 0% 0% 0% 2% MN 0% 0% 0% 0% 0% Table 3.12. Percentage of Unreliability Contribution Time Period Nonweather Light Rain Moderate Rain Heavy Rain Total AM 2% 0% 0% 0% 3% MD 9% 0% 0% 0% 10% PM1 27% 1% 0% 0% 29% PM2 45% 1% 0% 0% 47% APM 10% 0% 0% 0% 10% MN 1% 0% 0% 0% 1%
122 Figure 3.38. Mean TTI and weather comparison for I-95 NB GPL. Figure 3.39. 50th percentile TTI and weather comparison for I-95 NB GPL.
123 Figure 3.40. 80th percentile TTI and weather comparison for I-95 NB GPL. Figure 3.41. 95th percentile TTI and weather comparison for I-95 NB GPL.
124 3.3 Summary of I-95 Northbound General-Purpose Lane Performance Based on the results presented in the previous section, it can be concluded that the I-95 NB GPL segment was extremely unreliable in the PM peak. This period of unreliability extended from about 2:00 to 8:00 p.m. Most of the unreliability in the day occurred in the PM peaks, with the unreliability between 3:00 and 5:00 p.m. and between 5:00 and 7:00 p.m. contributing 29% and 45%, respectively, of the overall daily unreliability of the NB GPL in the day (a combined contribution of 74%). This overall unreliability did not consider the additional unreliability on the shoulders between 2:00 and 3:00 p.m. and between 7:00 and 8:00 p.m. The 50th, 80th, and 95th percentile TTIs for the GPL PM peak were 1.5, 2.1, and 2.6, respectively. The analysis of detector data indicated that during earlier parts of the peak, between 3:00 and 5:00 p.m., the reliability of the system was influenced by two capacity-constrained locations at NW 79th Street and NW 103rd Street. Between 5:00 and 7:00 p.m., a more severe capacity problem occurred due to the off- ramp to the Florida Turnpike and the turnpike toll plaza downstream of that ramp. Due to the different congestion patterns and causes in the PM peak, as explained above, the PM peak was subdivided into two periods for the purposes of analysis. It was found that higher-demand days not only increased the TTI, but also elongated the period during which the TTI was high. This finding indicates that implementing active traffic and demand management when a certain threshold of demand is exceeded has the potential of improving system reliability. It is interesting to note, however, that based on the CDF curves, even with normal demand in the PM peak periods, the 95th percentile travel time rate was about twice the free-flow travel time rate, probably indicating that there were impacts of events from outside the system or events not accounted for. In addition to these capacity-constrained problems, incidents contributed significantly to unreliability during most of the day. The contribution of lane-blocking incidents, as expected, was much higher than a single average incident. The contribution of lane-blocking incidents was twice as much as non-lane-blocking incidents, although non-lane-blocking incident frequency was much higher than that of lane blockage frequency. On a single-event basis, an average 2+ laneâblocking incidents had a very high relative impact. Additional analysis results indicated that crash incidents were generally more damaging than other (noncrash) incidents. During the PM period, the damage due to a single noncrash incident was about 50% of the damage caused by a crash incident. However, the overall impact on reliability was equivalent due to the higher frequency of noncrash incidents. The proportion of severe incidents appeared to be higher in the APM peak, possibly due to dark conditions, and increased in speed differential due to the dissipations of some queues and the lingering of others. Weather events (rain) were relatively rare compared to incident occurrence, and the overall contribution of weather was much smaller than that due to incidents and the PM peak no- event high-demand conditions. However, the impact of a single weather event, particularly moderate- to heavy-intensity rain events, was almost the same as that of a single incident in the PM peak. Incidents plus weather events were even rarer than weather events. Thus, their contributions to the overall reliability were low. However, a single incident plus weather event generated on average the worst conditions and on a single-incident basis produced twice the
125 impact of an incident in dry weather conditions. It is interesting to see that the impact of a single incident plus weather event was also very high in the AM peak and relatively high in the midday period. 3.4 Summary of I-95 Northbound Express Lane Performance The results of the I-95 NB EL analysis are presented in Appendix B. The I-95 NB EL was unreliable in the PM peak, but for a shorter period than the GPL. Most of the unreliability occurred in the PM1 and PM2 peaks, when the unreliability between 3:00 and 5:00 p.m. and between 5:00 and 7:00 p.m. contributed to 32% and 47%, respectively, of the overall daily unreliability of the NB EL in the day (a combined contribution of 79%). Between 5:00 and 7:00 p.m., the 50th, 80th, and 95th percentile TTIs were about 1.1, 1.5, and 2.7. Between 3:00 and 5:00 p.m., the 50th, 80th, and 95th percentile values were 1.1, 1.2, and 2.2. These results indicate that the median travel time of the EL was good. However, the 95th percentile TTI was relatively high between 3:00 and 7:00 p.m., and the 80th percentile TTI was somewhat high between 5:00 and 7:00 p.m. Compared to the GPL values, the 80th percentile TTI had a shorter peak period on the EL (from 4:30 to 6:00 p.m. compared to 3:00 to 7:00 p.m.). The maximum 80th percentile TTI value for EL during the PM peak was also lower than that on the GPL (peak of 1.75 versus 2.2). Although the peak 95th percentile TTIs of the EL and GPL were similar (95th percentile TTI values between 2.5 and 3.0) in the PM peak, the peak 95th percentile TTI for the EL occurred between 4:15 and 6:15 p.m., while that for the GPL occurred between 3:15 and 7:30 p.m. The no-event period contribution to the unreliability of the NB EL in the PM peak period was significantly smaller than the contribution of the no-event period for the GPL. However, the no-event periods, particularly when the demand exceeded the high-demand threshold, still had a significant influence on reliability, indicating that more aggressive pricing policies between 5:00 and 7:00 p.m., when the demand exceeded the high-volume threshold, would have the potential to improve the reliability of the system. Analyzing the data and visualizing the bottleneck impacts by using contour (heat) maps indicated that the main capacity-constrained congestion issues in the PM peak on the EL were at approximately the same locations as those of the GPL (NW 79th Street, NW 103rd Street, and the Florida Turnpike exit). The reliability analysis also indicated that incident intervals were the major contributors to unreliability. Although the incident frequency was lower on the EL than on the GPL, the EL contribution to unreliability was very high in the PM peak period due to the high severity per incident when one lane of the EL was blocked, and even more when both lanes were blocked. This finding needs to be explored further to determine how these impacts can be reduced, especially considering the geometric constraints of the EL, which may increase the impacts of incidents. Rainy conditions combined with incidents also increased the impacts of a single event. The overall contribution of weather events to reliability was small. 3.5 Summary of I-95 Southbound General-Purpose Lane Performance The results of the I-95 SB GPL analysis are presented in Appendix C. The SB GPL was unreliable from 7:00 to about 10:30 a.m. Forty-six percent of the unreliability of the day
126 occurred between 7:00 and 9:00 a.m., and it appeared that additional significant contribution occurred between 9:00 and 10:30 a.m. Between 7:00 and 9:00 a.m., the 50th, 80th, and 95th TTIs were 1.4, 1.7, and 2.5, respectively. The maximum five-minute 50th, 80th, and 95th percentile values during the AM peak were 1.6, 2.0, and 2.7 to 3.0, respectively. The midday period (assumed between 9:00 a.m. and 3:00 p.m.) also had a relatively high 95th percentile TTI at about 1.9. Five-minute reliability analysis indicated that the main unreliability in the midday peak occurred between 9:00 and 11:00 a.m., at least in part due to the extension of the AM peak congestion beyond 9:00 a.m. on some days. The contribution of the no-event periods to the unreliability of the SB direction during the AM peak period was smaller than the contribution in the NB direction during the PM peak. However, the no-event periods, particularly when the demand exceeded the high-demand threshold, still had a significant influence on reliability. Analyzing the data and visualizing the bottleneck impacts by using contour (heat) maps indicated that the main capacity-constrained congestion issue in the AM peak was located at three merging areas (Miami Garden Drive, the NW 103rd Street ramp, and at the exit of the EL). During the no-event periods between 7:00 and 9:00 a.m., the contribution of a single event with high demand was three times as much as that with normal demand. The reliability analysis also indicated that incidents were a major contributor to reliability most of the day. Significant effects of these incidents were also observed in the midday and PM peak. The percentage incident contribution to unreliability of the GPL was much higher in the SB direction in the midday and AM peak periods compared to the NB PM peak period. A higher number of incidents appear to have occurred in the SB direction, particularly in the AM and PM peaks, compared to the NB direction, which increased the impact on reliability. This difference could be due to the more complex weaving and merging maneuvers in the SB direction. Safety analysis should be conducted to determine the causes. The analysis also indicated that although the frequency of lane-blocking incidents was lower than non-lane-blocking incidents, their contribution was higher than shoulder incidents in the AM peak and particularly in the midday peak. The contribution of one-laneâblocking incidents, and particularly with 2+ lane blockages, was much higher than shoulder incidents. Incident plus weather events did not significantly influence overall reliability because of the low frequency of these incidents. However, the impact of a single such incident in the AM peak, and to a lesser degree in the midday and PM peak, was high. In the AM peak, the impact of a single incident plus weather event was twice as great as a single incident and 3.75 times as great as a single no-event interval. The PM peak period between 5:00 and 7:00 p.m., considered off-peak for the SB direction, during incident plus weather conditions in the SB direction was as bad as during the AM peak. The overall contribution of weather events to reliability was small due to the low frequency of these events; however, a single moderate-to-heavy-rain event had a significant influence. 3.6 Summary of I-95 Southbound Express Lane Performance The results of the I-95 SB EL analysis are presented in Appendix D. The SB EL was unreliable in the AM peak. However, the reliability was significantly better than the SB GPL, and the
127 unreliability lasted for a shorter period of time. A large proportion of the unreliability occurred in the AM peak and midday. However, the midday unreliability appeared to occur at the shoulder of the AM peak, from 9:00 to 10:00 a.m. Between 7:30 and 9:30 a.m., the 50th, 80th, and 95th TTIs were 1.08, 1.2, and 1.7, respectively. The main contributing factor to unreliability in the EL in the SB direction was an incident in the AM peak and, to a lesser degree, the midday peak. In the AM peak, the contribution of a single incident event was very high, indicating that, as with the NB EL, the geometry and operational constraints on the EL increased the incident impacts. 3.7 Summary of SR-7 Northbound Performance The results of the SR-7 NB analysis are presented in Appendix E. SR-7 is a data-poor environment. Therefore, data from INRIX were used in analyzing the reliability of the facility based on real-world data. The analysis clearly showed that the PM peak traffic experienced the most unreliable travel time, followed by the travel times during the midday and the after-PM peak period. The semivariance of travel time rate for the PM peak when considering a single- instant contribution was close to double the midday and evening values. However, because there are more hours in the midday, the overall contribution to unreliability was higher in the midday, followed closely by the PM peak. The 50th, 80th, and 95th percentile TTIs in the PM peak were 1.50, 1.63, and 1.90, respectively. The maximum 95th percentile TTI was approximately 2.0, indicating a travel time twice the free-flow travel time. For the midday, the three TTIs were 1.40, 1.46, and 1.55, respectively; the three indexes were close to each other. Similar results were noted for the other periods. These findings may be due to the nature of the operation on SR-7 or a function of the data used in the analysis (INRIX data). On a single-event basis, normal conditions, bad weather conditions, and conditions with crashes with a longer duration than 30 minutes appeared to have had the highest influence on reliability. However, when taking occurrence into consideration, the normal conditions during the midday and PM periods had larger contributions to overall unreliability along SR-7 NB. The analysis also showed that the I-95 lane-blocking events slightly increased the semivariance of travel time rates along SR-7 NB in the PM peak period. 3.8 Summary of SR-7 Southbound Performance The results of the SR-7 SB analysis are presented in Appendix F. The analysis indicated that the reliability performance was similar for different periods of the day, with the AM peak showing only slightly higher single-event impact than the PM peak and midday peak. Because of the longer period of the midday, the overall contribution to unreliability of the midday was the highest. The 50th, 80th, and 95th TTIs in the AM peak were 1.40, 1.50, and 1.80, respectively. For the midday, the three TTIs were 1.45, 1.49, and 1.63, respectively. For the PM peak, the three indexes were 1.46, 1.49, and 1.51. Again, all indexes seemed to be close to each other, possibly due to the nature of the operation on the facility or the nature of the data used. The analysis results showed that the occurrence of crashes caused slight increases in the unreliability of SR-7 SB traffic. It appeared that weather events and crashes had comparable
128 impacts on unreliability. The percentage of semivariance was also high during the midday period, which indicates that further improvement in the existing roadway configuration and signal timing may need to be considered.
