Pilot Testing of SHRP 2 Reliability Data and Analytical Products: Southern California (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.

4.5 See What Factors Affect Reliability (AE1) The guide presents several examples of applications that a TTRMS could address using outputs from the TTRMS. The guide calls these examples “Use Cases.” The study team examined two related use cases: • AE1 – Described in this section, Use Case AE1 is geared toward agency administrators and planners who can potentially use the case study approach to identify the causes of unreliability on the highway system. • AE2 – Summarized in the following section, Use Case AE2 can help administrators identify the contributions of each factor (ITRE et al. 2012). Use Case AE1 examines how incidents, weather, work zones, special events, traffic control devices, and fluctuations in demand can contribute to unreliability. The study team followed the six steps outlined in the guide to see how well the guidance performed in developing the results: 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 as well as the 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, that is, each combination of nominal congestion and nonrecurring event (including none). 5. Prepare TR-CDFs for each regime identified. 6. Analyze the contributions of the various factors so that the differences in impacts can be assessed. The first two steps, selecting the system of interest and the time frame for analysis, were described in Chapter 3 of this report. The two facilities analyzed are Orange County I-5 from Jeffrey Road in Irvine to East 4th Street in Santa Ana as shown in Figure 4.6, and Los Angeles County I-210 facility from East Colorado Boulevard in Sierra Madre to Citrus Avenue in Azusa as shown in Figure 4.7. The time frame for analysis was all non-holiday weekdays in 2010. Travel rates were estimated from midnight to midnight for each of these days. The travel rate observations were developed using the database described in the previous section. 54

Figure 4.6. Orange County I-5 system of interest. Figure 4.7. Los Angeles County I-210 system of interest. Jeffrey Rd E. 4th St 6.5 miles+/- Jeffrey Rd E. Colorado Blvd 10 miles+/- 55

Labeling Observations Labeling the observations describes the process detailed in the guide to develop nonrecurrent events, that is, descriptions of the conditions prevailing during each individual trip taken on the facility. Nonrecurrent events were joined to a system-loading variable. This combination was called a regime. System loading refers to the demand conditions that might be expected during a time period, which most commonly occurs during the morning or afternoon peak commute periods. Using the Orange I-5 facility as an example, Figure 4.8 illustrates unreliable trips compared to trips taken during recurrent conditions on the facility. The x-axis represents the time of day in 5-minute increments. The y-axis represents the travel rate on the facility. Each data point is a single trip from 2010 that started in a given 5- minute interval during the year (i.e., 24 hours x 12 5-minute intervals per hour x 250 days = 72,000 trips). The plot illustrates the commonly recognized a.m. and p.m. peak periods with the absolute a.m. peak occurring typically between 8:10 a.m. and 8:30 a.m. and the absolute p.m. peak occurring at 5:30 p.m. Trips that may be considered to take place under “recurring” congested conditions are outlined in yellow. Those trips that the study team labeled as being nonrecurring are outlined in red. Figure 4.8. Recurrent and nonrecurrent congestion on I-5 facility. AM Peak PM Peak Recurrent PM Peak Recurrent AM Peak Non- Recurrent 56

The guide describes four major causes of unreliability: • Weather • Incidents and accidents • Special events, such as road closures and sporting events • Demand surges The study team found labeling the observations to be a time-consuming exercise; it required extensive manual manipulation and matching. The guide provided some technical approaches to assist with identifying nonrecurring conditions, but did not provide guidance on how to prioritize each of these conditions. For example, if an accident occurred during a weather event should the condition be classified as a weather cause of unreliability or an accident cause? It might be reasonable to create a combination category, which was done for the Sacramento/Lake Tahoe case study described in the guide, but the study team did not create a combination category for this use case. Weather Data The study team downloaded daily historical weather data from the Weather Underground website. For the Orange County I-5 facility, the study team used the Santa Ana weather location, since it is centrally located on the facility. For the Los Angeles County I-210 facility, the study team used Burbank weather, because it was available for 2010, as shown in Figure 4.9. Downloading the data was very easy and took only a minute or two to obtain data for both facilities. 57

Figure 4.9. Weather Underground daily data. Source: www.weatherunderground.com. This data was supplemented with the Caltrans Traffic Accident Surveillance and Analysis System (TASAS) database (described in the accident data section below) because the TASAS database identifies the prevailing conditions on the roadway when an accident occurred. The study team also considered supplementing these data with National Oceanic and Atmospheric Administration (NOAA) 15-minute data to obtain a better level of data granularity for the analysis. However, one aspect of Southern California weather is that it is relatively consistent during the day, compared to other regions. For each 5-minute interval in the dataset, the study team identified the record as either having a 1 for a weather impact or a 0 for no weather impact. The weather impact was determined if the interval was described as “wet” in the TASAS dataset or as having recorded precipitation in the Weather Underground dataset. Traffic Incident and Accident Data The primary source for accident data in the dataset was the TASAS data described earlier. TASAS was developed by Caltrans to compile and summarize California Highway Patrol (CHP) state highway related collision reports. TASAS data can be downloaded from PeMS and provides the location of the accident (i.e., route, direction, postmile, and lane number), the severity (i.e., property damage only, injury, and fatality), as well as the lighting and weather conditions at the time of the accident. TASAS does not provide the duration of the accident, which was estimated in all cases based partly on data from the California Highway Patrol (CHP) dispatch logs. 58

The CHP dispatch logs, as well as the dispatch communications with the CHP officers in the field, are also found in PeMS. These data attempt to classify the type of incident and can be used to estimate the duration of incidents since each incident is “opened” and “closed” when CHP arrives at and leaves the scene of the incident. The dispatch data are very difficult to work with and are not always linked to the actual duration of an incident. For example, there are cases that the study team identified when an ambulance was dispatched and the event closed only a few minutes apart. It is unlikely that the ambulance arrived at the scene, provided medical assistance, and evacuated the scene, all within a few minutes. Special Event Data The study team used three primary data sources for special event data. For the analysis, the team included planned lane closures as special events. These data came from the Caltrans PeMS Lane Closure System (LCS). The LCS is a lane closure request and tracking system used by District Traffic Managers and contractors to request, review and approve lane closures on the freeway system. The dataset provided the location and date of the closure as well as the actual closing and opening times. The secondary sources of data were from the published schedules of events near the facilities. Major special event trip generators were not identified for the Los Angeles I-210 facility, and no special event facilities lie adjacent to the I-5 in Orange County. However, on State Route 57 in Anaheim, just north of I-5, are the venues for the Anaheim Angels professional baseball team and the Anaheim Ducks professional hockey team. Both Angel Stadium and the Honda Center host an array of other events, such as concerts for which 2010 schedules could not be readily obtained. The Angels and Ducks schedules (including preseason and playoff games) were downloaded and used to label “Special Event” condition. For the third data source, the study team analyzed VDS flow data at the interchanges adjacent to Angel Stadium and the Honda Center to identify times when flows were heavier than usual and to identify when other events occurred at these venues. The flows were compared against known event times (e.g., night baseball games) to validate findings from other nights. This approach allowed the study team to identify a fairly detailed set of special event times that might impact the I-5 facility. This facilitated the flagging of special events for each 5-minute interval with a 1 for special event or a 0 for no special event. Demand Surges The study team identified a “High Demand” category using the approach outlined in the guide by looking at the VMT traveling during each 5-minute interval. If demand exceeded two standard deviations above the mean for a given 5-minute interval, it was given a “Demand” designation (a 1 or a 0 as done with the previous approaches). 59

Identifying Regimes Once all the causes were identified, a final category for expected system loading was developed using the semi-variance approach outlined in the “See What Factors Affect Unreliability (AE1)” use case provided in the guide. Figure 4.10 shows an example of the semi-variance results for the Los Angeles I-210 facility. Table 4-2 shows the system-loading results between 2:00 p.m. and 4:00 p.m. used to create Figure 4.10. Figure 4.10. Semi-variance for I-210 facility. - 2,000 4,000 6,000 8,000 10,000 12,000 12 :0 0: 00 A M 12 :3 5: 00 A M 1: 10 :0 0 AM 1: 45 :0 0 AM 2: 20 :0 0 AM 2: 55 :0 0 AM 3: 30 :0 0 AM 4: 05 :0 0 AM 4: 40 :0 0 AM 5: 15 :0 0 AM 5: 50 :0 0 AM 6: 25 :0 0 AM 7: 00 :0 0 AM 7: 35 :0 0 AM 8: 10 :0 0 AM 8: 45 :0 0 AM 9: 20 :0 0 AM 9: 55 :0 0 AM 10 :3 0: 00 A M 11 :0 5: 00 A M 11 :4 0: 00 A M 12 :1 5: 00 P M 12 :5 0: 00 P M 1: 25 :0 0 PM 2: 00 :0 0 PM 2: 35 :0 0 PM 3: 10 :0 0 PM 3: 45 :0 0 PM 4: 20 :0 0 PM 4: 55 :0 0 PM 5: 30 :0 0 PM 6: 05 :0 0 PM 6: 40 :0 0 PM 7: 15 :0 0 PM 7: 50 :0 0 PM 8: 25 :0 0 PM 9: 00 :0 0 PM 9: 35 :0 0 PM 10 :1 0: 00 P M 10 :4 5: 00 P M 11 :2 0: 00 P M Se m i‐V ar ia nc e of th e Tr av el R at e pe r O bs er va tio n ([ se c/ m i]^ 2/ n) Time of Day SV 60

Table 4.2. System Loading on the I-210 Facility Time Semi- Variance (SV) Travel Rate (TR) Average Speed Regime Calculated 2:00:00 p.m. 171 60.4 60.0 Uncong 2:05:00 p.m. 178 60.6 59.8 Uncong 2:10:00 p.m. 196 61.2 59.3 Uncong 2:15:00 p.m. 228 62.0 58.6 Uncong 2:20:00 p.m. 270 63.2 57.6 Uncong 2:25:00 p.m. 319 64.5 56.5 Uncong 2:30:00 p.m. 387 66.1 55.3 Uncong 2:35:00 p.m. 458 67.8 53.9 LowCong 2:40:00 p.m. 569 70.4 51.9 LowCong 2:45:00 p.m. 693 72.9 50.2 LowCong 2:50:00 p.m. 771 74.3 49.2 LowCong 2:55:00 p.m. 826 75.5 48.4 LowCong 3:00:00 p.m. 899 76.7 47.7 LowCong 3:05:00 p.m. 1,051 79.2 46.2 LowCong 3:10:00 p.m. 1,228 81.9 44.7 ModCong 3:15:00 p.m. 1,505 85.6 42.8 ModCong 3:20:00 p.m. 1,724 88.3 41.4 ModCong 3:25:00 p.m. 1,898 90.4 40.5 ModCong 3:30:00 p.m. 2,191 93.6 39.1 ModCong 3:35:00 p.m. 2,529 96.9 37.9 ModCong 3:40:00 p.m. 2,852 99.9 36.8 ModCong 3:45:00 p.m. 3,246 103.3 35.6 ModCong 3:50:00 p.m. 3,510 105.7 34.8 HighCong 3:55:00 p.m. 3,803 108.0 34.1 HighCong 4:00:00 p.m. 4,290 111.7 33.0 HighCong Applying the Use Case, the study team first calculated the semi-variance for each 5- minute interval over the year using the formula from Chapter 3 in the guide: , where is the semi-variance, n is the number of 5-minute intervals per year for the run start time (e.g., all the 3:55 p.m. time periods in 2010), xi is the observed travel rate (1/speed), and r is the minimum travel rate observed over the entire year. 61

The length of the segment under analysis can impact the travel rate calculation, which can impact all the analysis performed using the L02 methods. There are no clear rules of thumb on how long a segment should be. However, choosing a segment that is too long will tend to smooth out travel times and travel rates across days, which will flatten out the semi-variance. Picking too short a segment may result in high semi-variances, which may not adequately reflect the traveler’s overall trip experience on the facility. Once all the unreliability conditions and system loadings were identified, the study team labeled the regime for each 5-minute interval. An example for the I-5 facility in Orange County is shown in Table 4.3. In this table, each condition (e.g., high demand, weather, special event, and incident) is coded with a 1 or a 0 to identify the condition (noted in the “Regime1” column). The ultimate regime (i.e., Regime1 and the Congestion Level) was labeled in the final column. The study team followed a hierarchy in labeling the regimes. Weather was considered to be a prevailing cause of congestion, followed by incidents, special events, and high demand. Since a single time interval can experience multiple reliability factors, a different order of prevailing causes would change the factor analysis. The guide does not offer advice on an appropriate order or method for assigning intervals to categories. The study team tested several variations, including simplifying and expanding categories. Clearly, this is a step that is open to wide interpretation and requires extensive trial and error. Table 4.3. Regime identification on I-5 facility. Time 2 Timestamp Time Period Free Flow Low Demand High Demand Weather Special Event Incident Score Incident Unk SV Calc Congestion Level Regime1 Regime 5 1/4/10 0:00 Eve/ Early AM 1 0 0 0 0 0 0 5.8 Uncong Normal Normal-Uncong 5 1/4/10 0:05 Eve/ Early AM 1 0 0 0 0 0 0 6.6 Uncong Normal Normal-Uncong 10 1/4/10 0:10 Eve/ Early AM 1 0 0 0 0 0 0 6.2 Uncong Normal Normal-Uncong 15 1/4/10 0:15 Eve/ Early AM 1 0 0 0 0 0 0 8.4 Uncong Normal Normal-Uncong 20 1/4/10 0:20 Eve/ Early AM 1 0 0 0 0 0 0 8.5 Uncong Normal Normal-Uncong 25 1/4/10 0:25 Eve/ Early AM 1 0 0 0 0 0 0 7.2 Uncong Normal Normal-Uncong 30 1/4/10 0:30 Eve/ Early AM 1 0 0 0 0 0 0 8.9 Uncong Normal Normal-Uncong 35 1/4/10 0:35 Eve/ Early AM 1 0 0 0 0 0 0 17.4 Uncong Normal Normal-Uncong 40 1/4/10 0:40 Eve/ Early AM 1 0 0 0 0 0 0 14.7 Uncong Normal Normal-Uncong 45 1/4/10 0:45 Eve/ Early AM 1 0 0 0 0 0 0 13.9 Uncong Normal Normal-Uncong 50 1/4/10 0:50 Eve/ Early AM 1 0 0 0 0 0 0 14.8 Uncong Normal Normal-Uncong 55 1/4/10 0:55 Eve/ Early AM 1 0 0 0 0 0 0 18.4 Uncong Normal Normal-Uncong Developing Travel Rate Cumulative Distribution Functions (TR-CDFs) The fifth step in the process is to develop probability distributions and create cumulative distribution functions for the travel rates. This was done by creating a Microsoft Excel pivot table and binning the travel rates in percentiles to arrive at the TR-CDFs for each facility. Figure 4.11 shows the I-5 results and Figure 4.12 shows the I-210 results. 62

Figure 4.11. Travel rate cumulative distribution functions (TR-CDFs) for I-5. Figure 4.12. Travel rate cumulative distribution functions (TR-CDFs) for I-210. 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 255 265 275 285 295 Cu m ul at iv e Di st rib ut io n of T rip s Travel Rate (seconds per mile) Normal-Uncong Normal-Low Cong Normal-Mod Cong Normal-High Cong Demand-Uncong Demand-Low Cong Demand-Mod Cong Demand-High Cong Incident-Uncong Incident-Low Cong Incident-Mod Cong Incident-High Cong Weather-Uncong Weather-Low Cong Weather-Mod Cong Weather-High Cong Special Event-Uncong Special Event-Low Cong Special Event-Mod Cong Special Event-High Cong 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 255 265 275 285 295 Cu m ul at iv e Di st rib ut io n of T rip s Travel Rate (seconds per mile) Normal-Uncong Normal-Low Cong Normal-Mod Cong Normal-High Cong Demand-Uncong Demand-Low Cong Demand-Mod Cong Demand-High Cong Incident-Uncong Incident-Low Cong Incident-Mod Cong Incident-High Cong Weather-Uncong Weather-Low Cong Weather-Mod Cong Weather-High Cong Special Event - Uncong Special Event - Low Cong Special Event - Mod Cong Special Event - High Cong 63

As can be expected, intervals with high system loading (i.e., peak commute periods) have the highest congestion on both facilities (shown by the red lines). Weather also has a significant impact on both facilities. Days with bad weather (shown with the triangle symbol) exhibit particularly long tails to the right of the graph and show very long travel rates on some days. Of note is the impact of special events on the Orange County I-5 facility. Occasionally, lines will cross other lines. This occurs primarily when there are relatively few observations available to create smooth transitions from one data point to the next. For example, there were only 47 rainy day observations in the year 2010 from which to develop the cumulative distribution for weather. Using the I-5 facility as an example, even on days with low congestion, shown by the green lines, special events have a larger impact than days with moderate congestion and normal demand (shown by the orange lines with no symbol). Figure 4.13 shows the I-5 CDFs for normal (i.e., recurring) congested periods. Figure 4.14 shows the I-5 CDFs with the special event lines added. Figure 4.13 shows the curves under normal conditions for uncongested and low-, moderate-, and high-congested regimes for I-5 in Orange County. Under uncongested conditions (e.g., in the middle of the night), there is little difference between the slowest trip at 57 seconds per mile (or 63 mph) and the fastest trip at 47 seconds per mile (or 76 mph). The green line representing this condition is nearly vertical, with a median trip at the 50th percentile, taking just under 55 seconds per mile. In contrast, under the most severely congested conditions (the red line), travel rates slow considerably. The median travel rate has more than doubled to 128 seconds per mile (28 mph). Approximately 5 percent of the trips under extremely congested conditions can take more than 175 seconds per mile (less than 20 mph). 64

Figure 4.13. Travel rate cumulative distribution functions (TR-CDFs) for I-5 normal congestion. Figure 4.14 shows the special event curves along with the normal curves shown in Figure 4.13. Special events can dramatically increase travel rates on the facility. Even under uncongested time periods (the green curve with the circle symbol), a special event can increase the median travel rate from about 53 seconds per mile to 60 seconds (an increase of 13 percent). The 85th percentile travel rate increases even more dramatically, from 55 seconds per mile (about 65 mph) to 99 seconds (36 mph), an increase of more than 80 percent. Figure 4.14 also shows that even under moderate congestion, the travel rates for special events can exceed the rate under recurrent congestion during the most heavily traveled weekday. 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 255 265 275 285 295 Cu m ul at iv e Di st rib ut io n of T rip s Travel Rate (seconds per mile) Normal-Uncong Normal-Low Cong Normal-Mod Cong Normal-High Cong 65

Figure 4.14. Travel rate cumulative distribution functions (TR-CDFs) for I-5 special events. Figure 4.15 shows the curves under normal conditions for uncongested and low-, moderate-, and high-congested regimes for I-210 in Los Angeles County. I-210 shows a slightly wider range of variation in travel rate than does I-5. The slowest trip takes 67 seconds per mile (53 mph) and the fastest trip is at 55 seconds per mile (65 mph). This is slower than for the I-5, but unlike the I-5, the low congested regime for the I-210 is not much slower than the uncongested regime. The median trip under low congestion takes 59 seconds per mile for the uncongested regime, but only about 8 percent longer (64 seconds per mile) under the low congested regime. The most severely congested conditions on the I-210 (the red line) have travel rates slowing considerably, with the median travel rate increasing more than 144 percent to 144 seconds per mile (25 mph). Approximately 5 percent of the trips under extremely congested conditions can take more than 178 seconds per mile (20 mph), which is similar to the extreme tail for the I-5 in Orange County. 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 255 265 275 285 295 Cu m ul at iv e Di st rib ut io n of T rip s Travel Rate (seconds per mile) Normal-Uncong Normal-Low Cong Normal-Mod Cong Normal-High Cong Special Event-Uncong Special Event-Low Cong Special Event-Mod Cong Special Event-High Cong 66

Figure 4.15. Travel rate cumulative distribution functions (TR-CDFs) for I-210 normal congestion. Figure 4.16 shows the weather curves along with the normal curves that were shown in Figure 4.15. Under uncongested and low congested times (e.g., late at night) rain does not have much impact on travel rates. In peak periods (high congestion), rain appears to reduce travel rates. This may be due to the fact that there were only 46 rainy days during 2010 in Southern California and statistically the sample is too small—though likely significant—to arrive at a completely conclusive answer. Rain may cause people not to take trips that they would otherwise take, which would reduce congestion on the facility. At the extreme end of the high congestion weather curve, however, results are what would be expected. Around the 75th percentile of trips, weather begins to have a worsening effect on travel rates. Around 5 percent of peak period trips in rainy conditions can take longer than 217 seconds per mile or travel slower than 16 mph. 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% 45 55 65 75 85 95 105 115 125 135 145 155 165 175 185 195 205 215 225 235 245 255 265 275 285 295 Cu m ul at iv e Di st rib ut io n of T rip s Travel Rate (seconds per mile) Normal-Uncong Normal-Low Cong Normal-Mod Cong Normal-High Cong 67