ORANGE COUNTY EMPIRICAL ANALYSIS: USING LOOP DETECTOR DATA TO IDENTIFY BOTTLENECK LOCATIONS AND CHARACTERISTICS

ORANGE COUNTY EMPIRICAL ANALYSIS: USING LOOP DETECTOR DATA TO IDENTIFY BOTTLENECK LOCATIONS AND CHARACTERISTICS by LEAH S. TOMLINSON A research project report submitted in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE in CIVIL AND ENVIRONMENTAL ENGINEERING Portland State University 2009

PROJECT APPROVAL The research project report of Leah S. Tomlinson for the Master of Science in Civil and Environmental Engineering submitted on April 28, 2009, is accepted by the faculty advisor and the department. ADVISOR APPROVAL Robert L. Bertini, Advisor DEPARTMENT APPROVAL Scott A. Wells, Chair Department of Civil and Environmental Engineering

iii ACKNOWLEDGEMENTS I would like to thank Dr. Robert Bertini for supervising this project and for his support, guidance and motivation on this project and during the course of my graduate career. I would like to acknowledge the California DOT and the University of California at Berkeley for allowing access to the PeMS database and Lianyu Chu for providing clarification and troubleshooting on issues that arose during the course of the project. Finally, I would like to the Department of Civil Engineering and Environmental Engineering at Portland State University that funded a portion of this project.

iv Abstract An abstract of the research project report of Leah S. Tomlinson for the Master of Science in Civil and Environmental Engineering submitted on April 28, 2009. Title: Orange County Empirical Analysis: Using Loop Detector Data to Identify Bottleneck Locations and Characteristics. In order to evaluate and understand congestion issues of a freeway, it is important to understand how traffic on that freeway flows. Using archived freeway data and manipulating it in ways that allow researchers to better understand that flow is a very useful tool. This study, inspired by previous studies of the same freeway segment, determines bottleneck locations and queue characteristics using archived freeway data. The freeway segment, a 6-mile stretch of Northbound I-405 in Orange County, California was studied. Data were taken from three dates. This data included occupancy, speed and count measures from on- and off-ramps and lane by lane from the mainline of the freeway. Speed diagrams were created for the mainline lanes to determine bottleneck locations, queue locations when they occurred and how long those queues lasted. Then, oblique plots of count, occupancy and speed were created to determine exactly how traffic was moving in the mainline lanes. Then the on-ramp data was examined and it was determined that surges in on-ramp traffic were creating the bottlenecks. There were some limitations that stemmed from inaccurate data that future research would be able to correct.

v TABLE OF CONTENTS 1. Introduction...1 1.1. Background...1 2. Methodology...2 3. Data...4 4. Analysis...6 4.1. May 22, 2001...6 4.1.1. Speed Diagrams...6 4.1.2. Oblique Plots...6 4.2. May 23, 2001...12 5. Results/Conclusions...18 6. Further Research...18 References...19

vi LIST OF FIGURES Figure Title Page 1 Bottleneck example...1 2 Speed Diagram and Site Map (May 22, 2001)...3 3 University and I-405...5 4 Oblique Count Plot (May 22, 2001)...7 5 Oblique Occupancy Plot (May 22, 2001)...8 6 Oblique Speed Plot (May 22, 2001)...9 7 On-ramp Oblique Count Plot (5/22/03)...11 8 Speed Diagram (May 23, 2001)...13 9 Oblique Count Plot (May 23, 2001)...14 10 Oblique Occupancy Plot (May 23, 2001)...15 11 Oblique Occupancy Plot (May 23, 2001)...16 12 On-ramp Oblique Count Plot (May 23, 2001)...17

1 1. Introduction The objective of this study is to identify bottleneck locations and characteristics given inductive loop detector data (vehicle counts, lane occupancies and speed data). The data used is from a sixmile stretch of Interstate 405 (northbound lanes) in Orange County, California. This study was inspired by and is related to a previous study titled, Use of Performance Measurement System Data to Diagnose Freeway Bottleneck Locations Empirically in Orange County, California (1). That study was conducted using data from a summer day in 1998 and it was concluded in that study that there were limitations in bottleneck locations, no on-ramp detector data and that there may be a need for a larger scale analysis. The current study will take another look at this same freeway segment to see if detector data at more frequent locations (including on-ramps) can give more detailed bottleneck information and characteristics. The data came from the California Performance Measurement System (PeMS) database. This resource was developed by the California Department of Transportation and The University of California at Berkeley. PeMS collects historical and real-time freeway data from California freeways in order to compute freeway performance measures (1). This study focused on the AM peak period from 7:00 to 9:00. Because of data restrictions, only 3 different queues were analyzed. The tools used to examine the bottlenecks are several oblique plots of different performance measures (speed, occupancy and count) along with speed diagrams. 1.1. Background A bottleneck is a point at which upstream of the point, traffic is flowing freely and downstream, there is a queue (Figure 1). Daganzo (2) defines a bottleneck such that it would be serving vehicles at a maximum rate, and discharge outflows measured downstream of a queue would not be affected by traffic conditions further downstream. A bottleneck is considered active when it meets the foregoing conditions and deactivated when there is a decrease in demand or when there is a spillover from a downstream bottleneck (1). Figure 1 Bottleneck Example (3)

2 Because bottlenecks can propagate for many miles and last for hours, it is important to understand the way they form their characteristics while active and how they dissipate. There have been many studies and much research done on the subject of bottlenecks. There are several implications of this research. They include the demonstration of the benefits of traffic management sensor investments, the revelation of the benefits of archived raw data, the better understanding of traffic behavior at a merge and much more. (3) In this study, data are taken from loop detectors and analyzed. This allowed for the identification of the bottleneck locations, the times at which the bottlenecks were activated and deactivated and perhaps, what caused the bottlenecks. 2. Methodology As noted previously, this study was conducted using loop detector data. The archived, raw speed data were used to create a speed contour plot (Figure 2).

Figure 2 - Speed Diagram and Site Map (May 22, 2001) 3

4 The red areas represent a slower speed while the green areas represent faster speeds closer to freely flowing conditions. The three aforementioned queues are labeled as Queue 1, 2 and 3. Because these queues form at Milepost 4.03 (University Drive), subsequent focus will take place at Milepost 3.86. Since we have data from the on-ramp at Milepost 3.86 and from all lanes at Milepost 4.03, our evaluation at Milepost 3.86 should provide a good idea of what is happening on that segment of the freeway. The site map is to the right of the speed diagram. The diamonds represent loop detector locations. Loops shown in red are those that are located in high occupancy vehicle (HOV) lanes and the loops that are grayed out show the locations where the data is either missing or corrupt. As can be seen from the diagram, the second on-ramp detector at Sand Canyon Avenue is grayed out because the data for that location is missing. This makes the evaluation of the queues at Milepost 3.31 impossible since we don t know the volume of traffic entering the freeway at that location. For this reason, only the queues and bottlenecks at Milepost 3.86 will be considered. After identifying where the queues were and subsequently, where the bottlenecks formed and dissipated, the next step in the process was to take the same speed data along with the occupancy and count data and examine them. This gave a better picture of what is causing the bottlenecks and allowed us to determine other characteristics of the traffic flow at this location during this time period. The next step in evaluating the bottlenecks and queues was to create and examine oblique plots of speed, occupancy and vehicle count. From these plots, it can be seen when speed, count and occupancy increase and decrease. When a bottleneck activates, speed decreases, count also decreases and occupancy increases. This makes sense because as traffic moves more slowly, fewer vehicles pass over the loop detectors in a given time period and stay over the loops for a larger percentage of that time period. By knowing the exact time of speed, count and occupancy changes and having the same information for the upstream on-ramp traffic, it is possible to determine what affect the on-ramp traffic has on the mainline traffic. Since it was determined where the queues formed from the speed diagrams, oblique plots were made at the same location and everything was looked at together. 3. Data As can be seen from Figure 1 and Figure 3, the study site is located on northbound I-405 between Mileposts 0.93 and 6.21 in Orange County, California. Along the study site, there are four northbound mainline travel lanes. On the left-hand side of these lanes, there is a high occupancy vehicle (HOV) lane that is separated by a wide, striped buffer. There are several locations where an auxiliary lane is present as well. These can be seen in Figure 1 at Mileposts 1.11, 1.73, 2.35, 5.74 and 6.21. There are 14 mainline loop detector stations that are labeled

5 according to their corresponding milepost. In addition, all on- and off-ramps have detectors on them. The following figure shows the site where a majority of the analysis takes place. The locations of the on-ramp loop detector and the upstream and downstream freeway loop detectors are shown. Data were pulled from the PeMS database for the days of May 22, 2001, May 23, 2001 and June 1, 2001. Loop detectors collected count, occupancy and speed data. The data was aggregated in 30 second intervals for each lane. As was previously mentioned, there are some locations where the data were missing or the data were collected but somehow corrupted and inaccurate. These loops are indicated as being faulty by being grayed over where the site map is displayed. Figure 3 - University and I 405 (Courtesy of Google Maps)

6 4. Analysis 4.1 May 22, 2001 May 22, 2001 was the first day s data to be examined. It was manipulated and analyzed by processes previously explained. The plots and figures in this section explain what was found after processing that data. 4.1.1 Speed Diagrams As explained earlier, the first thing that was done to the data was the creation of a speed diagram. The data from the 3 different days showed similar trends so only the graphs from the May 22, 2001 will be displayed. The speed diagram for this day is seen in Figure 2. Queue 1 was activated at 7:34AM and deactivated at 7:44AM. Queue 2 was activated at 7:56AM and deactivated at 8:11AM. The last queue, Queue 3 was activated at 8:23AM and deactivated at 8:49AM. After determining these queue times, the next step was to take a look at the oblique plots to get a picture of the flow of traffic at this location at the times that the queues began and ended. 4.1.2 Oblique Plots Figure 4 shows the oblique count plots for Milepost 3.86. The red lines indicate the beginning and ending of the queue and are labeled as Queue 1, 2 and 3. As can be seen from this plot, at around 7:34AM, the time of the activation of Queue 1, there is a sharp decline in count in lanes 1, 3 and 4 (the data for lane 2 in this instance was corrupt). The trend continues until approximately 7:44AM, at which time, the count begins to increase again. This too is consistent with the end of Queue 1. The same trends occur at the beginning and end of Queue 2 and Queue 3. The numbers on the plot are values for the flow of vehicles in vehicles per hour. They show bottleneck pre-queue flows and discharge flows. Because all lanes follow the same trends, only lane 1 flows were labeled. Figures 5 and 6 show oblique occupancy and oblique speed counts respectively. Again, the queue beginnings and endings are outlined in red. Similar trends can be observed in these plots. At approximately the same times as the beginnings and ends of the queues that correspond with the increases and decreases in count (as shown on the oblique count plot), decreases in speeds and increases in occupancy can be observed. Note that during the when between the queues are active, there are steep increases (occupancy) or decreases (speed) and in between the queues, speed and occupancy level off. Again, this is logical and expected since a queue would slow traffic speeds and therefore increase the amount of time vehicles spend over the loop detectors (occupancy).

7 Queue 1 Figure 4 - Oblique Count Plot

8 Queue 3 Figure 5 - Oblique Occupancy Plot

9 Queue 1 Queue 2 Queue 3 Figure 6 - Oblique Speed Plot

10 At this point in the study, it had become apparent that at this particular location, the data were good (with the exception of lane 2). It was making sense and the different types of data were lining up together in a way that reinforced what we know about traffic flow. After it was understood how the traffic was flowing at that location, where the queue was forming and how traffic is responding to that queue, attempts were made to solve the question of why the queue is forming in first place. In order to do that, the next step was to look at the on-ramp flow at that location. It was hypothesized that where the formation of queues and the slowing of traffic occurred, there would be a surge in traffic from the on-ramp onto the mainline freeway. Upon examining the on-ramp count, it was determined that the hypothesis was correct.

11 Queue 1 Queue 2 Queue 3 Figure 7 On-ramp Oblique Count Plot

12 The arrows and number in Figure 7 show surges in on-ramp flow. As can be seen, the queues were activated almost immediately after the surges. The arrow labeled 1, points to an onramp surge that most likely caused Queue 1. There was a sharp surge in on-ramp flow, this increased traffic volume on the mainline, slowing speeds, decreasing count and increasing occupancy. The same series of events occur with Queue 2 and Queue 3. Again, surges in onramp flow occur right before these queues form as indicated by the arrows labeled 2 and 3. Arrow 4 shows another very sharp surge in on-ramp traffic. This surge can be seen in Figure 1 as well. Over the duration of Queue 3, there is a slight increase in traffic speed followed by another period of very slow speeds that correspond with the on-ramp flow increase shown as arrow 4. 4.2. May 23, 2001 The same trends were seen for the other 2 days that were evaluated. Figures 8-12 show the same plots for May 23, 2001 which were previously shown for May 22, 2001. Although the same trends can be seen for June 1, 2001, they are not as defined and sharp as the other 2 days. For this reason, plots for that date have been omitted from this paper. The information from the two dates given should provide enough information and illustrate the objectives and results of this study well enough to not include the third day. Again, the queues beginning and ending times are shown by red lines and are labeled. In the oblique occupancy plot, during queues, the occupancy decreases. The same is true for the speeds on the oblique speed plot. Just the opposite occurs on the oblique occupancy plot for reasons previously explained. Also, as can been seen in the on-ramp flow from May 22, 2001, there are surges just prior to the queue beginning times as can be seen on the oblique count plots for May 23, 2001 (Figure 12).

13 Milepost 6.21 5.74 5.55 Culver Drive 5.01 Travel Direction Queue 1 Queue 2 Queue 3 7:20 7:20 7:29 7:44 8:01 7:30 7:45 8:01 8:03 8:03 8:35 8:34 4.03 3.86 University Drive 7:14 7:21 8:02 8:04 8:31 8:59 3.31 7:15 8:59 3.04 Sand Canyon Ave. 40 7:16 8:58 2.99 7:17 8:58 2.35 7:18 8:57 1.93 Hwy. 133 7:19 8:57 1.73 7:31 8:48 1.11 0.93 Irvine Center Drive Figure 8 - Speed Diagram (May 23, 2001)

14 Queue 1 Queue 3 Figure 9 - Oblique Count Plot

15 Queue 1 Queue 2 Queue 3 Figure 10 - Oblique Occupancy Plot

16 Queue 1 Queue 2 Queue 3 Figure 11 - Oblique Speed Plot

17 342 vph Figure 12 - On-ramp Oblique Count Plot

18 5. Results/Conclusions As previously noted, it was able to be determined that the data were accurate and that different data sets matched up with one another. By looking at the data in oblique plots, we can see exactly how traffic is moving and to what degree it is slowing or getting faster. The main thing that this study determined was that surges in on-ramp flow from University Drive onto Northbound Interstate 405 in Orange County, California cause bottlenecks and queues at this location. The data proves this in the three cases presented in this study and it is certain that it would do so over and over again with data from other dates. The fact that so much can be determined about the dynamics of freeway travel from the data that is collected speaks volumes. By recording speed, occupancy and count data and manipulating it in the ways shown in this study (and in other ways), it is possible to see exactly how traffic flows without having to directly observe it. One of the real successes of this study was being able to use this performance measurement system to identify the bottleneck locations. It shows that not only is the use of loop detector warranted but also extremely helpful and the data they pick up very useful. Solving the problem of the congestion in this location is not as simple as identifying the problem itself. This area of California has some of the worst congestion in the country and no one change will repair it. A ramp signal at this location could improve the flow of traffic by providing a more even flow of vehicles to the freeway but ramp signals alone will not completely erase the congestion in this area. Another issue that is present is the lack of data and the quality of data at certain loops. The more accurate the data, the closer researchers can come to identifying issues and the more precise they can be in their proposed solutions to these problems. 6. Further Research The same technique used on the correct data from the on-ramp at Sand Canyon Avenue would be beneficial as the queue from the bottlenecks identified at University Drive could be backing up into that area. There also seems to be bottlenecks forming at that location so it would be useful to know if on-ramp surges were causing them as well. We know that this process can show us where bottlenecks occur and where queues form. Further research could include a system that automatically creates the plots shown in this paper so that more detailed trends could be observed. This would also help see if problem areas were getting better or worse over time or after an improvement like the installation of a ramp signal.

Having not only the archived freeway data, but also archived plots showing how traffic moved on any given date could be extremely useful in evaluating the long term congestion issues of a freeway. 19

20 References 1. Bertini, Robert L. and Myton, Aaron M. Use of Performance Measurement System Data to Diagnose Freeway Bottleneck Locations Empirically in Orange County, California. Transportation Research Record 1925. 2004. 2. Daganzo, C.F. Fundamentals of Transportation Engineering and Traffic Operations. Elsevier Science. Oxford, 1997. 3. Bertini, Robert L. Capacity and the Breakdown Phenomenon at a Freeway Merge Bottleneck: Unlocking the Potential of Loop Sensor Data. Portland State University. February 2002.