Description
This can take the form of being protected from the event or from exposure to something that causes health or economical losses. It can include protection of people or of possessions.
ABSTRACT Title: STANDARDIZING AND SIMPLIFYING SAFETY SERVICE PATROL BENEFIT-COST RATIO ESTIMATION
Mersedeh TariVerdi, M.S., 2012
Directed By:
Prof. E. Miller-Hooks, Department of Civil and Environmental Engineering
Safety Service Patrol (SSP) programs operate nationwide with the aim of mitigating the impact of traffic incidents, especially along urban freeways. The central mission of the SPP programs is to reduce incident duration thereby reducing congestion related travel delays, fuel consumption, emission pollutants, and the likelihood of secondary incidents. The SSP-BC Tool was developed herein to fill the need for a standardized benefit-cost ratio estimation methodology for SSP programs with wide applicability and substantiated and needed updatable monetary conversion rates. The developed tool is designed to capture characteristics of incident, traffic, roadway geometry, and weather particular to the state area. VISSIM, a traffic microsimulation platform, was used to develop several multiple regression models with R-square values of 0.7 to 0.9 to assess the impact of travel delay, fuel consumption, and emission pollutants. Separate approaches were employed to estimate the savings in secondary incidents. In addition, a comprehensive method to compute fuel consumption and emissions is presented.
STANDARDIZING AND SIMPLIFYING SAFETY SERVICE PATROL BENEFITCOST RATIO ESTIMATION .
Mersedeh TariVerdi
Thesis submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Master of Science 2012
Advisory Committee: Professor Elise Miller-Hooks, Chair Professor Paul Schonfeld Assistant Professor Lei Zhang
© Copyright by Mersedeh TariVerdi 2012
Acknowledgments
First and foremost I offer my sincerest gratitude to my supervisor, Prof Elise MillerHooks, who has supported me throughout my thesis with her patience and knowledge. No doubt that without her this thesis would not have been completed or written. I am deeply indebted to my committee members Prof. Paul Schonfeld and Dr. Zhang for their time and effort in reviewing this work.
I acknowledge the I-95 Corridor Coalition for their financial support for this project and Tier 1 center (CTSM) for the 9-month fellowship.
I would like to thank my team-mates, Suvish, Xiadong, and Kara for sharing the literature and invaluable assistance; to my best friends, Nezam, Alireza and many others for their support. I am deeply and forever indebted to my parents for their love, support and encouragement throughout my entire life. I am very grateful to my brothers Sam and Amir.
ii
Contents
CHAPTER 1. Introduction...................................................................................... viii CHAPTER 2. Background and Literature Review .................................................... 6 2.1 2.2 2.3 2.4 Factors to Consider in Benefit-Cost Estimation........................................... 6 Evaluation and Estimation Methodologies .................................................. 6 Overview of Service Patrol Benefit/Cost Ratios per State ......................... 10 Summary .................................................................................................... 12
CHAPTER 3. Factors Affecting SSP Benefits and Their Calculation .................... 14 3.1 3.2 Measures of Effectiveness (MOEs)............................................................ 14 Factors Contributing to Travel Delay and Fuel Consumption ................... 14
3.2.1 Geometry of Roadway Segment .......................................................... 18 3.2.2 General Terrain .................................................................................... 19 3.2.3 Traffic Characteristics .......................................................................... 21 3.2.4 Incident Attributes ................................................................................ 24 3.2.5 Weather Conditions .............................................................................. 27 3.3 3.4 Calculating Travel Delay ........................................................................... 29 Calculating Fuel Consumption and Air Pollutants ..................................... 30
3.4.1 Emissions Calculation .......................................................................... 31 3.4.2 Fuel Consumption and Emission Calculation Methodology................ 34 3.4.3 Power Demand (Pv,t) and Instantaneous Fuel Consumption (FRt) Calculation ....................................................................................................... 35 3.4.4 CO2, CO, HC, CO & NOx Emissions Calculation............................... 37 3.4.5 SOx Emissions Calculation .................................................................. 37 3.4.6 Fuel Consumption ................................................................................ 37 3.4.7 Assumptions ......................................................................................... 37
iii
3.5 3.6
Secondary Incident Savings ....................................................................... 38 Summary .................................................................................................... 44
CHAPTER 4. Implementation of Variables in Simulation Designs ........................ 45 4.1 General Simulation Settings and Incident Modeling ................................. 47
4.1.1 Simulation Settings .............................................................................. 47 4.1.2 Control Case ......................................................................................... 48 4.1.3 Simulating Incidents and Rubbernecking Effect .................................. 48 4.1.4 VISSIM Calibration ............................................................................. 50 4.2 Single-Factor Experiments ......................................................................... 51
4.2.1 Geometry Factors: Number of Lane and Lane Blockages ................... 52 4.2.2 Geometry Factors: Ramps .................................................................... 54 4.2.3 General Terrain: Horizontal Curves ..................................................... 55 4.2.4 General Terrain: Vertical Curves ......................................................... 56 4.2.5 Traffic Characteristics: Speed of Vehicles ........................................... 57 4.2.6 Traffic Characteristics: Demand Flow Rate ......................................... 58 4.2.7 Traffic Characteristics: Truck Percentages .......................................... 60 4.2.8 Simultaneous Changes of Factors ........................................................ 60 4.3 Multiple-Regression Analysis .................................................................... 61
4.3.1 Minimum Sample Size ......................................................................... 61 4.3.2 Designing a Sample of Incidents .......................................................... 62 4.3.3 Multiple-Regression Models ................................................................ 64 4.4 Summary .................................................................................................... 75
CHAPTER 5. B/C Ratio Estimation ........................................................................ 76 5.1 Savings Computation ................................................................................. 76
5.1.1 Savings in Travel Delay ....................................................................... 77
iv
5.1.2 Saving in Fuel Consumption and Emissions ........................................ 79 5.1.3 Saving in Secondary Incidents ............................................................. 79 5.2 Total Benefit Calculation ........................................................................... 81
5.2.1 Monetary Values .................................................................................. 81 5.2.2 Computing Total Benefit ...................................................................... 82 5.3 5.4 Cost Calculation ......................................................................................... 84 The B/C ratio .............................................................................................. 85
CHAPTER 6. The Tool by Illustrative Example ..................................................... 87 6.1 The SSP-BC Tool ....................................................................................... 87
CHAPTER 7. Conclusions and Limitations ............................................................ 98 7.1 7.2 Limitations ................................................................................................. 98 Contributions .............................................................................................. 99
APPENDIX A: Fuel Consumption Computation Tables....................................... 100 APPENDIX B: Monetary Equivalents................................................................... 101 References .............................................................................................................. 112
v
List of Tables
Table 1.1 Freeway service patrol B/C ratio ............................................................. 10 Table 2.1 Previous studies on travel delay and incident duration............................ 16 Table 2.2 Maximum grade table (Adopted from roadway design manual 2010, .... 21 Table 2.3 FFS vs. Base capacity for freeways (Adopted from HCM 2010) ............ 22 Table 2.4 Average incident duration (Chou and Miller-Hooks, 2008) .................... 26 Table 2.5 Weather-Speed relation............................................................................ 28 2.6 Speed and capacity reduction based on road type (Chin et al. 2004)) ............... 28 Table 2.7 Actual speed under adverse weather conditions ...................................... 29 Table 2.8 Nomenclature for variables used in equations ......................................... 34 Table 2.9 Calculation of Road -Load coefficients [Source: USEPA, 2011c] .......... 36 Table 2.10 Secondary incident classification methods ............................................ 41 Table 3.1 Driver behavior parameters, adopted from Miller-Hooks et al. (2010) ... 51 Table 3.2 Summary of variables used in numerical experiments ............................ 52 Table 3.3 Number of lanes and lane blockage analysis ........................................... 53 Table 3.4 Ramp analysis .......................................................................................... 55 Table 3.5 Segment grade analysis ............................................................................ 56 Table 3.6 Impact of gradient on average speed ....................................................... 56 Table 3.7 Fuel economy changes due segment grade .............................................. 57 Table 3.8 Travel delay and fuel consumption changes by FFS ............................... 57 Table 3.9 Truck percentage analysis ........................................................................ 60 Table 3.10 Improved R-square comparison for new model .................................... 72 Table 3.11 LDV fuel consumption .......................................................................... 75 Table 4.1 Summary of monetary equivalents (ATRI) ............................................. 81 Table 5.1 Geometry and traffic default values......................................................... 91 Table 5.2 New York 2006 monetary values ............................................................ 95 Table 5.3 H.E.L.P result comparison ....................................................................... 96 Table A.1 Calculation of Road-load coefficients .................................................. 100 Table A.2 Emission factors .................................................................................... 100 Table A.3 Fuel properties ...................................................................................... 100 vi
Table A8.4 Transmission parameters for engine speed calculation (Source: PERE, 2005; EPA420-P-05-001) .................................................................................................. 100 Table B.1 Average gasoline prices ........................................................................ 101 Table B.2 Average labor cost by state ................................................................... 102 Table B.3 Average wage by area ........................................................................... 103 Table B.4 Truck percentage by state...................................................................... 111
vii
List of Figures
Figure 2.1 Base capacity vs. Total ramp density (adapted from HCM, 2010, P. 10-7) .............................................................................................................................................. 23 Figure 2.2 Incident timestamp flow chart (Chou and Miller-Hooks, 2008) ............ 25 Figure 3.1 Estimation model development progress................................................ 46 Figure 3.2 Incident layout on typical three-lane unidirectional freeway segment ... 50 Figure 3.3 Number of lanes and lane blockage analysis .......................................... 53 Figure 3.4 Simulating incidents close to an off-ramp .............................................. 54 Figure 3.5 Average speed Vs. Traffic flow rate....................................................... 58 Figure 3.6 Travel time vs. Traffic flow.................................................................... 59 Figure 3.7 Summery of fit diagnostic for total travel delay of LDV ....................... 68 Figure 3.8 Scatterplots of residuals against explanatory variables .......................... 69 Figure 3.9 Summery of fit diagnostic of linear model of LDT fuel consumption ... 73 Figure 3.10 Scatterplots of residuals against explanatory variables ........................ 74 Figure 4.1 Travel delay estimation procedure ......................................................... 78 Figure 4.2 Subcategory linear interpolation............................................................. 79 Figure 5.1 Main window .......................................................................................... 88 Figure 5.2 Program cost detail ................................................................................. 89 Figure 5.3 Basic data in segment level .................................................................... 89 Figure 5.4 Program information window................................................................. 91 Figure 5.5 Roadway geometry and traffic information ........................................... 93 Figure 5.6 Incident information window ................................................................. 94 Figure 5.7 Output window ....................................................................................... 95 Figure 5.8 CORSIM vs. VISSIM ............................................................................. 97
viii
CHAPTER 1. Introduction
Traffic congestion adversely effects traveler safety, cost, quality of life and the environment. Traffic congestion can be recurring or non-recurring, i.e. due to randomly arising events. Congestion caused by recurring events is a result of traffic demand exceeding the fixed capacity of a road segment during every day traffic patterns. Conversely, when the available capacity of a road segment decreases due to unpredictable events, such as vehicle incidents or adverse weather conditions, it is categorized as nonrecurrent congestion. The Federal Highway Administration (FHWA) sites non-recurrent congestion as the cause of approximately 60 percent of all road traffic in the United States. Traffic incidents (e.g. accidents, breakdowns), in particular, are a major cause of non-recurring traffic congestion. In fact, Caltrans’ Division of Research and Innovation claims that traffic incidents cause about 25 percent of this type of congestion on freeways (Caltrans, 2010). Therefore, incidents are counted as one of the most significant reasons for congestion in vehicular transportation. Increased travel time, increased risk of secondary accidents and decreased safety to other drivers and responders were identified as the most serious problems associated with an incident in a road segment. Additional problems, such as increased fuel consumption, pollutant emissions, cost of goods and services and negative impact on emergency response time, are also considered by decision-makers to be significant issues arising from traffic incidents. Moreover, the longer the incident duration and the time of impact to traffic flow, the greater the incident’s negative impacts. It is evident that traffic incidents have a strong adverse effect on urban areas. As such, reducing incident clearance time can mitigate its impacts (see, for example, Blumentritt et al., 1981). To control the impact of traffic incidents, Traffic Incident Management (TIM) programs that aim to reduce the duration and consequences of incidents and improve the safety of motorways have been introduced nationwide. A significant goal of most TIM programs is to coordinate the response by a number of public and private organizations to incidents. For example, transportation agencies are typically called to the incident scene by first responders, including law enforcement. Freeway Service Patrol (FSP) and similar Safety Service Patrol (SSP) programs are often components of a large TIM program. These 1
programs have been widely cited as very effective. In these programs, the service patrol vehicles may roam the roadways to which they are assigned (i.e. their beats), monitoring and responding to observed incidents. Alternatively, they may be dispatched to a call while roaming or from a traffic incident management center. The trained drivers of patrol vehicles may call for assistance from law enforcement and/or emergency responders, or may directly assist with motorists’ needs. SSPs receive most of their funding from state and federal taxes. Therefore, they are must have public support and may be scrutinized when budgets are limited. While states work to provide essential services they must consider their budget limitations. As such, quantifying the benefits of SSPs is important to legislators determine the effectiveness of such programs in terms of improving safety and increasing public benefits. Thus, even in times of budget crises, the benefits of these programs may be great enough that funding them is encouraged. This research effort builds on recent I-95 Corridor Coalition efforts (Chou and Miller-Hooks, 2008) in which a procedure was developed to determine the benefits (i.e. reduction in congestion, secondary incidents, fuel consumption and pollution, along with their monetary equivalents) of an existing SSP program. The methodology was employed to estimate the B/C ratio for the Highway Emergency Local Patrol (H.E.L.P.) program in New York. This prior effort revealed the need for additional study in identifying a set of best performance measures and monetary conversion rates to accurately depict the benefits and costs of such programs. Moreover, the developed approach, like other comparable methods with similar accuracy used around the country, requires significant computational effort and is, therefore, costly and time-consuming. A quicker and less data-intensive approach is desired so that it can be readily and widely utilized by all states around the US Such an approach is developed within the effort described herein. When attempting to compare an SSP program to its alternatives, it is common to compare the benefits of the program with its costs through a benefit - cost (B/C) ratio. This ratio has been estimated for many SSP programs operating around the nation. These ratios, however, range dramatically (e.g. from 2-to-1 to 36-to-1). The majority of this variability is likely due to the wide range of estimation methodologies and monetary equivalent 2
conversion factors employed within these techniques, rather than to actual differences between the program benefits. This great variability also opens these findings to questions about their accuracy. A standardized methodology that can be universally and equitably employed in such B/C ratio estimation is essential as it would aid in creating consistency and, therefore, greater confidence in the validity of the results. The main objective of this study was to develop a user-friendly tool, referred to as the SSP-BC Tool, based on consistent performance measures and monetary conversion rates that can quickly compute the B/C ratio of an existing or planned SSP program. In Chapter two, a review of existing US SSP programs and other relevant studies that evaluate their performance, as well as reported B/C ratios, are presented. There are numerous factors that might be considered in evaluating the benefits to society of a SSP program. The most common are: savings in travel time, fuel consumption, pollutant emissions and secondary incidents. These are described in detail in Chapter three. In the same chapter, a review of factors that have the greatest effect on these measures is also provided. Components of weather, roadway gradient and curvature, density of ramps in the roadway segment, and traffic composition, all of which influence the available capacity of a roadway segment and fuel consumption rates, are considered in this study. To quantify these benefits of a SSP program, VISSIM, a microscopic traffic simulation product, is employed. The Component Object Module (COM) interface of VISSIM was used for modeling freeway incidents. The COM interface permits controlled experiments with altered spatial and temporal incident characteristics. To provide a realist portrayal of an incident in a simulation environment, all factors that have been found to affect travel delay, fuel consumption and emissions must be considered in the experimental design criteria. A methodology for modeling traffic incidents is adopted that exploits this COM interface. Within this methodology factors, such as roadway length and gradient, are directly set in VISSIM, while other factors, such as volume, traffic composition and incident attributes, are defined through the interface. The effect of weather on roadway performance is captured through changes to free flow speed. This methodology is described in Chapter three.
3
Replicating real-world conditions within a simulation environment requires a primary study into the capabilities of the software and creation of specific methods needed to adequately capture desired effects. Chapter four highlights the necessity of this investigation and describes these methods. Initial runs were made to study trends in travel delay and fuel consumption estimates resulting from univariate changes in factors. Whether variables are dependent was also verified through runs in which the state of two or three factors were permitted to change simultaneously. Upon analyzing outputs from the simulation runs, and after discussion with PTV America’s support, it was determined that the built-in fuel consumption tool available within VISSIM was not suitable for this study. The tool did not provide repeatable estimates. Moreover, it seems that it overestimates fuel efficiency of vehicles available within the US and its equations were developed for emissions estimation from traffic on arterials. Specific methods for calculating fuel consumption and emissions from vehicular modal parameters gathered for each VISSIM run are presented in Chapter four. A review of factors affecting fuel consumption and emissions is given in Chapter three. Analysis of results from the initial runs provided insights into the response of the vehicles in VISSIM to changes in characteristics of the roadway segment, traffic volume and the roadway environment. As an example, these runs revealed that traffic performance was unaffected by changes in roadway curvature, an important input. Thus, efforts associated with these runs indicated that the effects of significant curvature could not be captured directly and a suitable methodology would be needed. Additional restrictions on combinations permitted in batch runs were identified, (i.e. when speed was set through code and grade was non-zero) that preclude the possibility of conducting simulation runs to capture all combinations of input. Because it was not possible to create batch runs to run all combinations of inputs, regression analysis was employed. Specifically, regression models of travel delay and fuel consumption were developed and calibrated based on simulation results from 1200 runs on a typical stretch of a three lane freeway. For each run, a randomly generated traffic incident was created. The incident scenarios involved one of three states of lane blockage (shoulder, one lane and two lane blockage) with equal likelihood. For each incident, the incident duration was set 4
according to a statistical distribution identified in previous studies. This culminated in 7 regression models capturing the response of dependent variables associated with travel delay and fuel consumption to incident duration, traffic volume, gradient and percentage of trucks. To reduce the error terms in the regression models, and improve overall fitness, an additional 73,290 runs were designed and conducted. The runs involved all possible combinations of 16 categories of incident duration, 11 categories of traffic volume and 6 speed categories, resulting in 1,056 combinations. For each combination, runs including one of 3 types of lane blockage and one of 5 possible roadway sizes in terms of number of lanes and 5 random seed for each were made. Note that no complete road closure scenario was considered. Other spatial and temporal incident characteristics were held constant. At the heart of the SSP-BC Tool is a database of five tables: tables of travel delays for light and heavy duty vehicles, average driver and police officer wages, and fuel costs. The tool pulls data from these tables to complete computations related to the benefits and costs of the studied system. Data in these tables are derived directly from the simulation run results (travel delays, fuel consumption), through regression-based estimates (travel delays, fuel consumption), computations (emissions, secondary incidents) or from publically available sources (wages, fuel costs, traffic composition, and monetary conversion rates). The regression models were used specifically to capture the effects of traffic compositions (i.e. percentage of trucks in traffic mix) and roadway grade. While Chapters three and four focus on individual incident characteristics and impacts, Chapter five describes calculations and assumptions used to obtain program benefit estimates, and ultimately the B/C ratio. Chapter six includes snapshots of the tool and a brief explanation of how the tool works through an illustrative example. Results of a case study involving the H.E.L.P program are presented Chapter seven. General findings follow in Chapter eight.
5
1
CHAPTER 2. Background and Literature Review
SSP programs exist in a large portion of the US SSP drivers can provide free assistance to motorists. Examples of service include: providing a gallon of gasoline, changing flat tires, jump starting dead batteries, pushing vehicles off the road, providing minor mechanical repairs, and helping motorists call for other emergency services. In the case of severe accidents, SSP drivers are trained to help police redirect traffic. These services are crucial for shortening the duration of incidents and, thus, diminishing their impact, and improving safety for other drivers on the roadway segment. Furthermore, SSPs can be used as probe vehicles, providing real-time updates on traffic conditions (Traffic Incident Management Handbook). In this chapter, evaluation studies on SSP programs around the nation reported in the literature are reviewed. B/C ratio computation approaches and estimates by program are reported. 1.1 Factors to Consider in Benefit-Cost Estimation The first step of estimating the benefits and costs of a SSP program is to determine the components that should be considered in the calculations. The reduction in travel delay and corresponding economic benefits for the motorists plays a significant role in the benefit estimation. Some studies also consider prevention of secondary incidents due to decreased incident duration; they assume a direct relation between number of potential secondary incidents and incident duration. In addition, environmental concerns, such as fuel consumption and pollutant emissions, are included in savings. Some studies derive an estimate for fuel consumption from delay time or use the computational tools available in some simulation software packages. There are other savings that should be counted in the benefits of a SSP program. For example, costs of towing if SSP vehicles were not at the scene, lawsuits from secondary incidents that are prevented, and additional time available for troopers for more urgent tasks that the SSP programs cannot handle. 1.2 Evaluation and Estimation Methodologies The most accurate way to evaluate the benefits of a SSP program is to conduct a “before-and-after” study that compares the incident detection, response, clearance and
6
recovery times (often marked by a return in traffic state to pre-incident flow rates) for a comparable period before and after the deployment of the SSP program. Donnell et al. (1999) evaluated the Penn-Lincoln Parkway Service Patrol in this way. The study recorded similar incidents that occurred prior to and following the implementation of SSPs, the collected data was compared to compute possible savings. Respectively, incident response time and clearance time were found to be reduced by an average of 8.7 and 8.3 minutes (17.1 minutes overall savings in incident duration), yielding a B/C ratio of 30:1. In another study, Skabardonis et al. (1995) analyzed the operation of SSP programs in San Francisco, California using field data from 24 weekdays before the SSP and 22 weekdays after the SSP program was implemented. The B/C ratio was shown to be 3.4. An assessment that was carried out by Bertini et. al. (2001) in one region of Oregon showed that the regional SSP program reduced the average cost of delay-causing incidents to travelers by 36 to 66 percent when it was upgraded from part-time to full-time. In many circumstances, however, the “before” dataset is not available as it usually requires a well-maintained and long-term managed database. Therefore, researchers adopt a “with-and-without” approach. This method compares a SSP managed incident to a similar incident that was managed by state or local police. Commissioned by the Safety Service Patrol (SSP) program in Northern Virginia (NOVA), Dougald et. al. (2008) compared the average durations of various episodes with similar incident types and roadway and traffic conditions. The main difference between each episode was whether or not the SSP program responded. The data that was compared came from two databases: 1) the incident management database (IMD) and 2) the Virginia State Police (VSP) computer-aideddispatch (CAD) system. This type of study has been applied to investigations conducted in other states, including Indiana (1999), Minnesota (2004), Florida (2005), California (1995, 2005), Maryland (2006), New York (2008) and Missouri (2010). In these circumstances where no “before” data was maintained, it is necessary to make assumptions surrounding how long the incident would last had such a SSP program not existed. In this way, the potential savings of SSP-assisted incidents can be calculated. Sensitivity analysis can be conducted to examine how the B/C ratio responds to varying incident duration savings. Generally, the range of assumed duration reduction is between
7
10 and 20 minutes. A study of the evaluation of the SSP program in Los Angeles, assumed that the SSPs would reduce incident duration by 10, 12.5, or 15 minutes, resulting in a B/C ratio that ranged from 3.75:1 to 5.5:1 (Skabardonis et al., 1998). Moreover, the average duration of crashes and in-lane incidents handled with the Hoosier Helper SSP were assumed to be lowered by 10 min while all other incident durations were reduced by 15 minutes (Latoski et al. 1999). Chou et al. (2008) lengthened the duration of without FSPassist incidents by between 5 and 25 minutes in 5-minute increments for studies on SSP program of New York State, H.E.L.P. The estimation methods of incident delay and delay savings of SSP programs draw on methods such as statistical, deterministic queuing, or simulation-based models, or surveys. Examples of studies involving each are given next. Mauch et al. (2005) examined the Big-rig SSP pilot program that provides services including heavy-duty tow trucks along the I-710 freeway in California through use of statistical modeling. Regression analysis was completed and correlations between average vehicular delay per day and big-rig and non-big-rig incidents were noted. A calibrated regression-based function was given to describe the relationship between traffic delay and incident duration. They combined the response time savings with delay estimates to forecast delay savings. It was assumed that for comparison purposes if conventional services would be required, 45 minutes of response time would be needed for dispatch of the big-rig tow truck. Expected benefits from the program were estimated to be $14,700/day with a benefit-cost ratio of 5:1, where benefit computations include travel delay only. Other studies employing statistical approaches to evaluating the benefits of SSP programs include: Mauch et al. (2005), Haghani et al.(2006). Where traffic volume profiles over the incident duration are available, a deterministic queuing model can be applied to predict travel delay. Skabardonis et al. (1995, 1998) estimated delay as the difference in travel times under normal and incident conditions using data from loop detectors and probe vehicles. Guin et al. (2007) used extrapolated capacity reduction factors during the response and clearance of incidents associated with cumulative traffic volume as inputs to the queuing model for the Georgia NaviGAtor SSP program. This approach was also employed in the evaluations of SSP 8
programs in Oregon (Bertini et al., 2001), Florida (Hadi and Zhan, 2006; Hagen et al, 2005), Virginia (Dougald wt. al, 2007) and Missouri (Sun et Al., 2010). The majority of SSP program evaluation studies rely on simulation, because it is often the case that traffic volume data and other traffic characteristics are limited and, thus, effects on traffic must be estimated. Traffic simulation models have become quite advanced, permitting control of roadway design, traffic volumes, and incident characterization, including incident duration, number of lanes blocked, and location. Latoski et al. (1999) estimated delay using the XXEXQ macroscopic traffic simulation model in studying the Hoosier Helper program in Northern Indiana. The study yielded B/C ratio of 4.71 if the program operates only in the day time and 13.28 for 24 hour operations of the program considering travel delay in benefits and annual investment, employee salaries and benefits, overhead and maintenance costs. 120 incidents were replicated in the microscopic CORSIM traffic simulation platform for the purposes of studying their effects on travel delay for Coordinated Highways Action Response Team (CHART) in Maryland (Chang et al., 2006). Benefits in reduction of travel delay, fuel consumption and emissions were estimated to be 1,006.50 million dollars in a similar study in 2009. Representative incidents with varying duration and lane blockage were simulated in the PARAMICS microscopic traffic simulation platform to analyze the Freeway Incident Response Safety Team (FIRST) program in Minnesota (MnDOT, 2004). Total incident delay was plotted against volume corresponding to different incident durations ranging from 4 to 40 minutes. Based on the plot, delay reductions can be predicted given the incident duration reduction caused by the FIRST program. The B/C ratio estimated for FIRST was 15.8, including travel delay and crash avoidance in the benefit estimation. More recently, Chou and MillerHooks (2008) replicated 693 actual incidents in CORSIM to analyze the B/C ratio of the H.E.L.P. program in New York. Incidents were simulated with H.E.L.P-assist and without H.E.L.P-assist circumstances. The rubbernecking effect set up in CORSIM was computed from capacity reduction estimates associated with number of lanes, lane blockage, and incident type. It was found that the B/C ratio range was between 2.14 and 2.68 (for different costs) using conservative monetary conversion rates for travel delay, fuel consumption, emission pollutants and avoided secondary incidents. The B/C ratio would
9
increase to between 13.2 and 16.2 if vehicle occupancy, traffic composition, and higher incident severity level were included in benefit evaluation. 1.3 Overview of Service Patrol Benefit/Cost Ratios per State The first SSP program with annual operations originated in 1960 in Chicago, Illinois. In 2006, the US DOT and Intelligent Transportation System (ITS) Joint Program Office (JPO) conducted a survey regarding service patrol programs in 106 metropolitan areas. At the time, 73 out of 99 areas that responded had a service patrol program in operation and more than 40 states had implemented at least one SSP program. From review of the literature, including journal articles, research reports and web pages from state departments of transportation, it can be concluded that service patrols reduce incident duration, improve safety, and help reduce fuel consumption and emissions. It was proved that its benefit outweighed its cost with the B/C ratio ranging from 1.48:1 to 38.25:1. A summary of B/C ratio estimates noted in the literature can be found in Table 2.1. It is evident from the table that the ratios vary widely. This is in part because there are inconsistencies in not only analysis methods and monetary conversion rates used to obtain the ratios, but also in the factors they include in benefit estimation. In fact, each state or city adopts its own set of factors that it deems relevant to calculate the B/C ratio. Underestimation may result from ignoring certain benefits while overestimation may occur from over-counting low probability events. Differences in monetary conversion rates, that provide monetary values for estimated benefits, can also greatly impact the final ratio values.
Table 1.1 Freeway service patrol B/C ratio
State Program name I710 Big-Rig Location/service area I-710 south of ocean blvd to the I-5 interchang LA Year Included benefits B/C
California
2005
Travel delay Travel delay, fuel consumption Travel delay Travel delay, fuel consumption
5 5 for 15 min duration reduction 3.4 2.3-41.5, 25.1 overall
California
FSP
1998
California Florida
FSP Road Ranger
SF district 1-7, except dis.3, Turnpike
1995 2005
10
Georgia Indiana Michigan Minnesota
NaviGAtor Hoosier Helper Freeway courtesy Patrol FIRST Freeway motorist assist program
Atlanta Northwest Indiana Southest Michigan Minneapolis, St. Paul
2007 1999 2009 2004
Travel delay, fuel consumption, emissions, avoided secondary incidents Travel delay Travel delay Air quality Travel delay, crash avoidance Travel delay, secondary crashes
4.4 4.7 daytime ,13.3 24h 15.2 15.8
Missouri
St.Louis Long Island; in New York City; the Lower Hudson Valley; Buffalo; Rochester; and the Albany Capital District Region 2 Hampton Road Northern Virginia
2010
38.25
New York
H.E.L.P
2009
Travel delay, fuel consumption, emissions, avoided secondary incidents Travel delay Fuel consumption Travel delay, fuel consumption, emissions Travel delay, fuel consumption, emissions
2.14-16.5 for 20 min reduction
Oregon Virginia Virginia
FSP SSP NOVE SSP
2001 2007 2006
32.52-3.68 overall 4.71, range of 1.48-10.17 Vs. V/C 6.2
Each mentioned previous study was conducted for a unique location given data for a specified time period and each such study typically required enormous effort to complete. These studies, however, are needed to defend and secure financial support for continued program operations. It is often the case that studies in one area re-invent methodologies created for studies in other areas. Moreover, their estimated B/C ratios cannot be directly compared, because they rarely use similar factors or monetary conversion rates. The Freeway Service Patrol Evaluation (FSPE) model, developed by the University of California at Berkeley (Skabardonis et al., 2005) for California, is a dedicated SSP evaluation tool. That can be applied more widely. The FSPE model was implemented in Excel workbook using Visual Basic for Application (VBA). This tool computes daily, annual or specified time period savings with respect to incident delay, fuel consumption and emission, as well as the B/C ratio. The tool relies on a deterministic queuing model for calculating incident delay. Benefits of SSPs are dependent on the beat’s geometric and traffic characteristics, and the frequency and type of assisted incidents. Default model
11
parameters are provided, but they can be modified by the users if empirical field data are available. This SSP model can analyze 24/7 SSP services, or can accommodate more limited weekday or weekend services, although the best final prediction model was found to be for 24 hour SSP services. They applied the 24 hour model to estimate the benefits. They count for limited operational hours and used the proportion of vehicle miles traveled (VMT) in fewer than 24 hours. Each beat is divided into segments for each travel direction and data are input accordingly. The FSPE model distributes the FSP-assists per incident type proportionally to the VMT per beat. It is assumed that the response time without the SSP is 30 minutes. The SSP response time is computed based on the beat length, average tow-truck speed, and number of trucks operating on the beat. The FSPP model later was developed based on the FSPE model to evaluate the roadway segments which does not have SSP assistance. The saving in response times are estimated based on a statewide weighted average of all incidents from the fisical year 20022003 FSP-assists database. One of the strengths of this SSP evaluation tool is the consideration of directional effects of daily traffic volume along each patrolled beat. The tool was calibrated for use in California. Significant effort and data are required to calibrate the queuing model for use in other locations (Skabardonis et al., 2005). 1.4 Summary While B/C ratio estimation models, specifically FSPP and FSPE, exist that might have general utility in B/C ratio estimation for SSP programs, the SSP-BC Tool proposed herein accounts for a wider array of traffic, environmental and program characteristics that influence benefit and cost estimates. The SSP-BC Tool accounts for factors, such as ramp density, horizontal and vertical alignments, traffic composition, and weather conditions, that have been identified as important to travel delay, fuel consumption and emissions estimation. Moreover, fuel consumption and pollutant emissions estimates used in prior related studies, when included, are made based on simplistic regionally developed rates for travel day to fuel consumption and emissions. For example, in the FSPP tool, the fuel consumption and emission calculation factors are developed for California conditions. The tool uses average vehicular speed; thus, driving modes such as acceleration or deceleration and stops of vehicles are not captured in the computations. In the proposed SSP-BC tool,
12
estimates of these factors are made from these modal parameters. The SSP-BC Tool also has the advantage of including up to two lanes blockage in the freeway segment. The FSPP and FSPE tools include at most one lane that is blocked. On the other hand, the FSPP and FSPE tools can be used to compute incident response time savings due to a SSP program. As such, they can provide needed input to the SSP-BC tool. The SSP-BC Tool also accounts for secondary incidents that statistically would arise with longer incident durations than can be expected where SSP exist, which FSPE and FSPP tools do not.
13
2
CHAPTER 3. Factors Affecting SSP Benefits and Their Calculation
Numerous methods are practiced by incident program managers for measuring and evaluating SSP program performance. Results of such studies are often used to justify the expenses of these programs, but can also provide insights that can be used to improve performance and ultimately reduce the number and impact of traffic incidents. The first step to evaluate a SSP program is to identify the contributing factors to benefit values as will be discussed in this chapter. In the first Section, MOEs employed in benefit estimation are identified. Factors effecting the estimation of these MOEs are discussed in Section 3.2. Methodologies used for their computation are described in Subsections 3.3-3.5. This is followed by a summary in Section 3.6. 2.1 Measures of Effectiveness (MOEs) A myriad of MOEs may be used in evaluating the benefits of a SSP program. Typical measures include: operational performance measures (e.g. incident response time), traffic performance measures (e.g. travel delay), environmental impacts, safety (secondary incident prevention), reliability, maintainability, and ease of use. In the context of this study, travel delay, fuel consumption, pollutant emissions, and prevention of secondary incidents have been chosen as the MOEs of interest.
2.2
Factors Contributing to Travel Delay and Fuel Consumption Numerous factors, like roadway geometry and weather, affecting MOEs have been
identified in studies on travel delay, fuel consumption and emissions. A comprehensive set of factors has been used in this study for estimating all MOEs; although, some factors are more significant for one MOE than others. For example, roadway grade will have greater impact on fuel consumption than occurrence of secondary incidents. To identify the factors of greatest importance for travel delay estimation, works in the literature were reviewed. The majority of statistical and deterministic queuing methods developed for this application area assume that the most significant factor in reducing incident-induced delays is to reduce incident duration. One of the formulae most widely used to compute travel delay was
14
developed by Wirasinghe (1978) as described in (Qi et al, 2002). For a given roadway segment, Wirasinghe’s formula is given in equation (1). This equation includes factors of incident duration (T), total capacity (S1), traffic demand (S2), and bottleneck capacity (S3). Eq.3.1 In equation 3.1, incident duration and bottleneck capacity are directly related to travel delay resulting from an incident. The number of lanes blocked and severity of the incident is a function of available capacities at bottlenecks at the time of the incident. Incident duration estimation is required within equation 3.1 for travel delay computation. Since 1987 many techniques have been developed to predict incident duration based on collected data. However, the site-specific nature of the collected data has caused disagreement as to the validity of the results. The majority of prior studies have employed different statistical models to estimate incident duration, travel delay, and similar required data to evaluate traffic conditions and the efficiency of an SSP or similar program while they can be applied in studies of roadways with similar traffic characteristics, geometry and weather. These models are neither suitable nor applicable to other regions with different traffic circumstances. A goal of the proposed SSP-BC Tool is, thus, to provide a method to uniformly estimate and compare travel delay savings associated with SSP programs. To address the need for regional or even roadway-specific estimates, microscopic traffic simulation methods are employed. In general microscopic traffic simulation can be used to estimate the consequences of an incident on travel performance. In developing the SSP-BC tool, to produce realistic estimates, its application to a nationwide model, calibrated parameters for typical US highways, comprehensive information and details of the typical incident sites were considered. When considering travel delay, it is important to analyze congestion, incident duration and the causes of both. Geometry of the roadway, traffic characteristics, demand, construction and major maintenance operations, traffic accidents or vehicular breakdowns, and weather conditions are the factors suggested in the Highway Capacity Manual that affect the actual capacity of a highway segment. Table 3.1 shows factors that have been
15
considered in some studies of incident duration and travel delay. In this study, the simulation experiments were designed to account for nearly all of these suggested factors.
Table 2.1 Previous studies on travel delay and incident duration
Authors, year Dependent variable Variables used in the model development Peak-hour Character of incident Severity level Disposal type Number of lanes blocked Personal injuries Response units(fire department, police, FSP) Temporal characteristics Weather (snow, rain) Incident location (with respect to ramps) Incident characteristics Involved vehicle Response source Physical variables: accident time of the day, the day of the week, weather Vehicle variables: number of vehicles, truck involvement, Response variables Number of lanes affected Presence of trucks Time of the day Police response time Weather
Zhao et al. 2010
Incident delay
Boyles wt al. 2006
Incident duration
Qi (2002)
Incident duration
Smith et al. (2001)
Clearance time
Garib et al. (1997)
Incident duration
A basic freeway segment is chosen as the control sample for simulation designs herein as it is in HCM for estimating the capacity of a roadway under different circumstances (HCM, 2010, 10-1). The method of the HCM 2010 assumes that under basic conditions a freeway segment can reach its full capacity. These basic conditions include clear fine weather and visibility, no congestion due to incidents, no work zone activity (short- or long-term), and acceptable pavement conditions which support normal operations. In addition, presumably all the drivers are familiar with the area. Models of typical base geometries and additional estimates to account for special roadway features, such as curvature and weather, were developed. The frequency and
16
severity level of incidents were permitted to vary over time and space. Temporal characteristics include season, day of week, and time of day of an incident. Other attributes, such as location of the incident in the roadway and lane blockage due to incident occurrence, are considered as spatial characteristics. Regardless of the time or location that an incident occurs, as long as there is at least one vehicle traveling behind the incident, the result is capacity and/or speed reduction, which affects time delay. A list of factors that are studied herein is given as follows. ? Geometry of the roadway segment o Segment length o Number of lanes and average lane width o Lateral clearance (shoulder) o Ramps o General terrain o Horizontal curves o Segment gradient ? Traffic Characteristics o FFS o Ramp FFS o Traffic flow rate o Percentages of trucks in traffic flow ? Incident attributes o Incident severity o Incident duration o Average incident duration o Rubbernecking effect ? Weather conditions These factors are discussed in the following subsections. As this study uses a simulation estimation method, The range associated with each chosen factor within the simulation estimation method is based on information from the literature as discussed in proceeding subsections. These factors affect not only travel delay, but fuel consumption
17
and emissions. Some factors affect these latter MOEs directly and, thus, are not only a consequence of added delay. 2.2.1 Geometry of Roadway Segment 2.2.1.1 Segment Length A large number of studies have used simulation to study the impact of traffic incidents on travel delay using a model of a generic fixed-length homogeneous section of roadway. The homogeneity of a Section relates to its traffic, geometry and weather. Chou and Miller-Hooks (2008), Saka et al. (2004), and Hobeika and Dhulipala (2004) examined a 10-mile segment. This length was typically chosen to ensure that the effects of a traffic incident could be entirely captured within the model. A similar length, thus, is used for simulation runs of this study. Greater homogeneity may exist for shorter roadway segments, especially where there is significant horizontal curvature and vertical changes; however, using a shorter length segment can lead to errors in estimates. Additionally, it has been noted that there can be an undesirable increase in accident location reporting errors and other types of errors for shorter roadway segments (Shankar et al., 1994). 2.2.1.2 Number of Lanes and Average Lane Width As the study focuses on freeways, 2 to 6 lane highway segments in each direction are considered. It is assumed that the standard minimum lane width of 12 feet is available based on the default value of lane width of urban and rural highways in HCM 2010. 2.2.1.3 Lateral Clearance (Shoulder) To ensure full operational capacity, basic freeway segments require a minimum 6feet right-side shoulder. As VISSIM does not offer the capability to model shoulders, initial numerical experiment were conducted to investigate VISSIM’s ability to capture the impact of shoulder characteristics. This is done by adding an additional lane that is closed to traffic. It was noted that VISSIM does not capture the effects of shoulder existence. Therefore, no interruption in flow or reduction to capacity due to either shoulder width or closure was modeled. Note that impact of capacity reduction due to incidents occurring in the shoulder has been studied in this effort.
18
2.2.1.4 Ramps Only major junctions were modeled and all on- and off-ramps were assumed to be located at the right-edge of the roadway. According to the HCM 2010, merge and diverge influence areas are 1500 feet downstream from the merge and 1500 feet upstream from diverge points. Thus, route decisions used within VISSIM to model vehicle movements towards off-ramps started from 1500 feet before the off-ramps. Qi (2002) considered the possibility of incorporating relationships between incident and ramp locations in incident duration modeling. Qi was unable to obtain the needed data to ascertain details of this relationship. Moreover, no other studies of this relationship could be found in the literature. Preliminarily experiments were designed here to study this relationship, but based on the results, (Section 4.2.2) only ramp density was used to capture the impact of ramps. Ramp density is defined in HCM to explain the impact of merging and diverging vehicles on the free flow speed and the capacity of the segment. Ramp density is the average number of on-ramps and off-ramps in a 6-mile segment based on the midpoint of the study segment. It varies from 0 (occurs in rural areas) to 6 (possible in dense urban areas) total ramps per mile. The free flow speed of a basic freeway segment is most sensitive to ramp density as discussed in Chapters 10 and 11 of the HCM. The HCM does not discuss the impact of the number of main lanes on ramp operations or the percentage of traffic separating\entering from\to the main lane. However, the HCM (2010), states that on average, an increase of 2 ramps per mile in total ramp density causes approximately 5 miles per hour reduction of speed in the basic segment (FFS of 75 miles per hour having zero ramp density, standard lane width and right-side clearance) , because of approximate linear relation of free flow speed and ramp density. Numerical experiments were made in this study to compare travel delay and fuel consumption due to different exiting flows. 2.2.2 General Terrain There are three categories of general terrain: level, rolling, and mountainous. Level terrain contains any combination of horizontal or vertical alignments that enables heavy vehicles to operate at the same speed as passenger cars. Typically, this terrain contains short grades with a maximum 2% incline/decline. In rolling terrain, there is a combination of vertical or horizontal grades which cause heavy vehicles to operate slightly poorer than
19
passenger cars, but still they have not reached crawl speed (i.e. the maximum constant speed that trucks can maintain on a specific grade over a given length on an uphill stretch). In mountainous terrain, heavy vehicles operate at crawl speed for significant distances or frequent intervals. The impact of horizontal curves and vertical grades were studied separately in this effort. If data pertaining to roadway grade are not available, default values for each terrain are made available in the tool. 2.2.2.1 Horizontal Curves Curvature is a signification factor in the incident duration estimation studies (Gomes et al, 2008). In most of the relevant statistical studies, however, existence of curvature enters the incident duration estimation models as a dummy variable regardless of its degree. The degree of a curve relates to its design speed. The sharper the curve, the slower the design speed. For safety related reasons, posted speed limits on curves are usually lower than the design speed of the curves based on the minimum radii. Super elevation change in horizontal curves of a road segment can vary from 0 percent for areas with severe weather conditions to 8 percent for drier areas and regions with favorable weather conditions. For areas of high-speed, such as freeways, a maximum super elevation of 6% in horizontal curvature is typical and is employed in this study. In the US it is customary that the design speed range for curves be set between 45 and 80 mile per hour (Roadway Design Manual 2010). Shankar et al. (1994) completed one of the few studies that considered this curvature in estimating incident duration. They categorized horizontal curves by their design speeds and determined an explanatory variable for each category. This method, however, requires a level of detail that is not practical for the users of the SSP-BC Tool developed herein. Within the tool, in a segment with free flow speed of 70 mph, thus, curvatures having design speeds of 70 to 75 mph were assumed as straight, 60 to 65 mph as mild and 50 to 55 mph as sharp. To capture this, 5 and 10 mile per hour speed reductions were applied for mild and sharp curvatures, respectively. 2.2.2.2 Segment gradient Roadway gradient is one of the highway-related factors known to significantly affect fuel consumption and emission rates (Park and Rakha, 2005), since vehicles need
20
more power on uphill climbs to maintain their speed and less power in descending downhill. The Roadway Design Manual has 4 percent maximum allowable grade for urban freeways as shown in Table 3.2. However, since SSP programs also operate on mountainous roads with higher grades, the impact of grade in the range of -10% to 10% is considered in this study.
Table 2.2 Maximum grade table (Adopted from roadway design manual 2010,
Functional Classification Type of Terrain 15 20 25 30 35 40 45 50 55 60 65 70 75 80
Urban and Suburban Local Collector Arterial Freeway All Level Rolling Level Rolling Level Rolling <15 9 12 ----<15 9 12 ----<15 9 12 ----<15 9 11 8 9 --<15 9 10 7 8 --<15 9 10 7 8 --<15 8 9 6 7 ---7 8 6 7 4 5 -7 8 5 6 4 5 -----3 4 -----3 4 -----3 4 -----3 4
2.2.3 Traffic Characteristics 2.2.3.1 Free-Flow Speed (FFS) and Roadway Capacity FFS is the most important factor defining the roadway capacity. Theoretically, when the density and flow rate in the segment is zero, vehicles travel with FFS. In practice, it is defined as the desired speed at flow rates between 0 and 1000 passenger cars per hour per lane. Using a systematic simple (e.g every tenth vehicles in each lane and a minimum of 100 vehicles) the mean speed of all passenger cars can be reported as FFS (HCM, 2010). The factors affecting FFS are: lane width, lateral clearance and ramp density, all of which were considered in simulation designs herein under geometric characteristics of the segment.
21
The HCM defines capacity as the “average flow rate across all lanes.” VISSIM does not have direct input for capacity. Consequently, by using the FFS as the initial desired speed, suggested reduced capacities under various circumstances can be modeled. For example, to include ramp density in the model, instead of running all the possible number of ramps, the equivalent reduced speed for capacity reduction percentages due to ramp density provided by the HCM is used. The maximum posted speed limit in US freeways varies from 55 to 80 mph according to Highway Safety Research and Communications (2012). Considering capacities in the HCM 2010 and their relationship with other factors, a 70 mph FFS was employed in the simulation runs herein. This is because speed reduction is generally taken from a base FFS of 70 mph.
Table 2.3 FFS vs. Base capacity for freeways (Adopted from HCM 2010)
FFS (mile/h) 75 70 65 60 55 Base Capacity (pc/h/ln) 2400 2400 2350 2300 2250
The capacity-speed relationship shown in Table 3.3 is based on national norms, but this relationship can change locally. Furthermore, upon review of data from the National Motorist Association it is observed that, in urban facilities, speed limits for passenger cars and trucks are identical. There are roadways in some states such as California, where the speed limits differ by vehicle class. In this study, it was assumed that speed limits are identical for all vehicle types. To consider impact of weather condition on speeds and existence of ramps, the study range of the vehicle speed is chosen to be 35 to 75 mph in this study. 2.2.3.2 Ramp FFS For simulating ramps in VISSIM, their FFS must be set. Typical speeds for ramps are in the range of 20 to 50 mph according to the HCM. 25 and 35 mph were considered in this study. The impact of ramp density was taken into account using base capacity reduction due to increase in number of ramps in a segment. The relationship between base
22
capacity and total ramp density is shown in Figure 3.1, assuming drivers are familiar with the area and no trucks are present (HCM, 2010).
Figure 2.1 Base capacity vs. Total ramp density (adapted from HCM, 2010, P. 10-7)
2420 2400 Base capacity (pc/mi/ln) 2380 2360 2340 2320 2300 2280 2260 2240 0 2 4 6 8 10 Total ramp density (ramps/mi)
2.2.3.3 Traffic Flow Rate (Demand) Traffic volume directly affects level of service of the freeway segment and savings from reduced incident duration. The hourly traffic (in each direction) was used as traffic volume. A range of maximum capacities between 200 and 2200 vehicles per lane per hour (vplph) on average was used for simulation runs. For traffic flows of less than 1000 vplph, vehicular speeds are nearly constant. At a 75 mph FFS, the minimum breakpoint of speed reduction due to flow growths occurs at a volume of 1000 vplph (HCM 2010). Thus, for each case, 11 different traffic volume categories from 200 to 2200 vplph were simulated. 2.2.3.4 Truck Percentages in Traffic Flow Traffic composition, including the percentage of heavy vehicles, is one of the details required to complete an operational analysis of a freeway segment. In addition, the rate of fuel consumption is highly dependent on truck percentages. A range between 0 and 20% trucks is used in developing the SSP-BS Tool. This range was based on information involving truck percentages in specific areas. Specifically, a 3 percent truck composition value was noted for the I-270 freeway in Maryland (Miller-Hooks et al., 2010).
23
2.2.4 Incident Attributes Incident-caused congestion and incident duration are greatly affected by incident severity. As a factor in statistical analyses related to incident duration estimation models, severity is most often noted to be significant. However, SSP program savings are typically derived from more frequent, low severity incidents. 2.2.4.1 Incident Severity In most studies on freeway capacity reduction and incident duration, traffic accidents or vehicular breakdown are modeled based on lane and shoulder blockage (Hadi et al. 2000; Saka et al.,2008; Chou and Miller-Hooks, 2008; Khattak et al., 2010 ; HCM 2010). In this study, shoulder, one lane, and two lanes blockage events are used to model incidents with different severity levels. The greater the severity, the more lanes blocked. 2.2.4.2 Incident Duration The main objectives of SSPs are to identify incidents or other causes of disruption in the traffic stream and minimize incident duration. Thus, a standard method to calculate incident duration is essential for SSP program evaluation. Incident duration is often defined as the time between incident occurrence and when response vehicles leave the scene (Garib et al. 1997, Nam and Mannering, 2000; smith and smith, 2001). The Traffic Incident Management Handbook describes incident duration based on the time required to detect an incident, time from incident report to on-scene response, and time required to clear the incident. A widely used approach to defining incident duration is depicted through timestamps in Figure 3.1.
24
Figure 2.2 Incident timestamp flow chart (Chou and Miller-Hooks, 2008)
As described in (Chou and Miller-Hooks, 2008), at the moment the incident occurs, the driver will call the management center and the “Call Start Timestamp” will be logged. Once confirmation of the incident is received, a SSP vehicle driver will be dispatched to the scene and the “Dispatched Timestamp” is logged. The “Arrival Timestamp” is marked when the patrol unit arrives at the scene. Finally, once the event is cleared, the “End Timestamp” is recorded. Thus, time for confirmation, response time, assistance duration, and incident duration can be determined from the difference between two timestamps as follows: Confirmation time = Dispatch time - Call Start time Response time = Arrival time - Dispatch time Assistance duration = End time - Arrival time Incident duration = End time - Call Start time When an incident is identified by a patrolling vehicle, i.e. is self-initiated, both the time for incident confirmation and response time will be zero. The majority of SSP program responses are made to incidents with durations of 90 minutes or less. SSP programs may also assist state troopers or local police in response to more severe, typically longer, incidents. Although the frequency of these long duration incidents is low (often less than 2%), these incidents cause very significant travel delay. In fact, in some cases, it is necessary for law enforcement to close the entire freeway.
25
Including these extreme cases in the evaluation, study of SSPs often results in an overestimation of the benefits of such programs, particularly since they play a supporting, rather than leading, role in these events. Therefore, incidents with durations greater than 90 minutes were not modeled within this study and benefit estimations can be considered conservative in this respect. 2.2.4.3 Average Incident Duration To quantify the benefits of a SSP program in New York State, Chou and MillerHooks (2008) studied the incident data for the study region. In the first phase of their study, the average incident durations over all incidents arising within a six-month period along four roadways, i.e. I-287, I-684, the Taconic State Parkway and the Sprain Brook Parkway, were calculated. In Table 3.4, a summary of the computed incident durations is shown. Savings from the SSP program can be computed from the difference between the average duration of incidents to which the SSP vehicles responded and those to which it did not.
Table 2.4 Average incident duration (Chou and Miller-Hooks, 2008)
Responds With SSP Without SSP Savings Taconic State Parkway 17.60 39.56 21.96 Sprain Brook Parkway 12.10 21.96 18.12 I-684 26.24 40.61 14.37 I-287 17.70 68.21 50.52
As shown in the table, for those incidents to which the SSP program did not respond, the average incident duration varied between 22 and 68 minutes. The range is smaller where the program is involved in the response. In a study on traffic recovery time, Saka et al. (2004) used VISSIM to simulate incidents that lasted between 10 and 60 minutes. Skabardonis et al. (2005) assumed that the average incident time without SSP response would be 30 minutes. Taken from a report in CHART in 2009 the average incident durations of incidents involving property damage and disabled vehicles were reported to be 33 and 20 minutes, respectively. Therefore, incident durations of 5 to 90 minutes with 5-minute increments were studied for the SSP-BC Tool herein. It is assumed that during this 5 minute time period, estimated savings change linearly.
26
2.2.4.4 Rubbernecking Effect When an incident occurs in a freeway, vehicles in unaffected lanes often reduce their speeds. The effect of these speed reductions is called the rubbernecking effect. Assuming that a warning sign is set up for upstream traffic to inform other drivers of the incident, the rubbernecking effect can be modeled by a reduced speed area in the segment from the warning sign to the incident location. It is assumed in the models developed herein that warning signs are set up 500 feet prior to the incident as is recommended in emergency traffic control guidelines. The effect of rubbernecking is an important piece of an incident to consider. It is often the case that many accidents are caused by drivers looking at other vehicle crashes and other roadside traffic incidents. A 2003 study by the Virginia Commonwealth University’s Transportation Safety Training Center (TSTC) revealed that rubbernecking was the leading cause of vehicle crashes. It accounted for 16 percent of all vehicle crashes. Other distractions arising external to the vehicle, such as the presence of deer, accounted for 35 percent (Masinick and Teng, 2004) of such crashes. 2.2.5 Weather Conditions A growing concern of roadway management agencies is the impact of adverse weather on freeway traffic operations. It is understood that severe weather conditions reduce freeway capacities, but few works have studied its precise impact. In addition, the results obtained from many studies are from outside the US or relate to rural freeway segments within the US. These statistics may not be applicable to urban freeway segments due to different roadway geometries, driver behaviors, and traffic characteristics. In pertinent studies, weather conditions are classified into one of three types: "rain", "snow" and "others" (wind, fog, etc.). Each category is divided in terms of intensity (light versus heavy). The effect of each classification of weather on FFS is summarized in Table 3.5.
27
Table 2.5 Weather-Speed relation
Researcher/year Ibrahim and Hall/ 1994 HCM 2000 Kyte/ 2001 Manish et al. /2005 Rakha et al./ 2007
N/A Not Availble
Light rain 1.2-8 mph 2-14% 15.3 mph 1-5 % N/A
Heavy Rain 3-10 mph 5-17% 15.3 mph 4-7% N/A
Light snow 0.6 mph 8-10% 23-26 mph 3-10 % 13%
Heavy Snow 26.4 mph 30-40% 26.4 mph 11-15% 40%
visibility N/A N/A N/A 6-11% 13%
Wind N/A N/A N/A 1-1.5% 10%
It has been shown that that the impacts of weather on traffic flow and its parameters are dependent on the class of road. Chin et al. (2004) used loop detector data from different regions of the US. These data were linked to different weather. The weather conditions were classified into 6 categories: light rain, heavy rain, light snow, heavy snow, fog, and ice. The impact of each adverse weather condition was then translated into loss of capacity and speed as shown in Table 3.6.
2.6 Speed and capacity reduction based on road type (Chin et al. 2004))
Weather Condition Light rain Heavy rain Light snow Heavy snow Fog Ice Urban Freeway Capacity (%) 4 8 7.5 27.5 6 27.5 Speed (%) 10 16 15 38 13 38 Rural Freeway Capacity (%) 4 10 7.5 27.5 6 27.5 Speed 10 25 15 38 13 38
The impact of weather can be modeled through its effects on speed and capacity reduction. Because VISSIM does not have explicit capacity input, suggested speed reduction along urban freeways due to adverse weather conditions was used for this study. Table 3.6 shows no significant difference between urban and rural freeways except under the condition of heavy rain. To categorize intensity of weather conditions, thresholds have been developed in the HCM as follows. Light rain: perception below 0.25 inch/hour Heavy rain: Perception greater than .25 inch/hour Light snow: perception below 0.5 inch/hour
28
Heavy snow: Perception greater than 0.5 inch/hour In this study, using this knowledge from previous studies, speed reduction on urban freeways was based on the 6 different adverse weather conditions. For this investigation, the levels are selected as: 5% speed reduction due to light rain, 10% for heavy rain, light snow, and low visibility, 15% for fog, 35% for heavy snow, and 40% for icy conditions. Table 3.7 shows suggested speeds under different weather states.
Table 2.7 Actual speed under adverse weather conditions
speed limit 75 70 65 60 55 speed reduction percentage 5 71.25 66.5 61.75 57 52.25 10 67.5 63 58.5 54 49.5 15 63.75 59.5 55.25 51 46.75 35 48.75 45.5 42.25 39 35.75 40 45 42 39 36 33
It is assumed that weather conditions are uniform along a segment. In addition, simulated weather does not change during the simulation. In other words, a specified speed reduction value was used for any simulation run. The focus of this section has been primarily on factors affecting travel delay. However, many of the factors that affect travel delay also directly impact fuel consumption and emissions of air pollutants. For example, grade is a roadway characteristic with its greatest effect on fuel consumption. Travel factors, such as speed, were also found to significantly impact fuel consumption and emissions.
2.3
Calculating Travel Delay A simulation-based evaluation method was developed to estimate travel delay and
input data needed to compute fuel consumption and emissions. The platform employed in this study is PTV America's VISSIM (version 5.3) software, a micro-simulation tool for traffic operations modeling. VISSIM is used to obtain estimates of travel characteristics
29
and other metrics for roadways with operational SSP programs and those without (for comparison) through a host of simulation runs in which solitary incidents are simulated and their effects are estimated. VISSIM computes the travel delay of each vehicle and total travel delay in the network in terms of the average total delay per vehicle (in seconds). Total delay is computed over all vehicles passing through a travel segment. For a given vehicle, its value is determined by subtracting the ideal travel time (assuming FFSs can be maintained) from the realized travel time. 2.4 Calculating Fuel Consumption and Air Pollutants While VISSIM could be applied directly in estimating travel delay, it was found that it could not reliably compute fuel consumption for freeways and no module is available for computing pollutant emissions. PTV offers external (in the form of add-ons) fuel consumption and emission calculation modules for VISSIM. The user manual describes a process in which emissions data can be obtained from node evaluation at a network level. For this study, necessary licenses needed to use the emissions and fuel consumption add-ons were available and several preliminary tests were designed and run in an effort to validate fuel consumption and emissions output from these modules. Unfortunately, the modules did not function as designed. In most runs, fuel consumption and emissions were not reported. Results of those runs for which results were obtained revealed that the fuel economy estimates from VISSIM were over-estimated for US vehicle markets. In fact, an average fuel efficiency of 35 miles per gallon was obtained from the successful runs. Through discussions with the PTV America, Inc. support staff, it was determined that these tools would not be reliable, nor necessarily available, for this study. More generally, the fuel consumption and emissions models were specifically designed for signalized intersections and not freeways. A comprehensive method to calculate fuel consumption and emissions from all vehicle movements in vehicle record output files of VISSIM was adopted in this study. After recording the actual speed, acceleration, and mass of all vehicles, the data was entered into the network at one second intervals during a simulation period. For each run, calculations for the run were made to obtain fuel consumption and emissions. 30
2.4.1 Emissions Calculation Emissions in the transportation sector are primarily due to the combustion of fossil fuels. Carbon dioxide (CO2) and hydrocarbons, such as methane (CH4), are produced from the combustion products of fossil fuels, like petroleum, diesel and biofuels, as a result of the fuel’s high carbon content [USEPA, 2010a]. Nitro gases or NOx emissions are formed when nitrogen (N), either in the air or in fuel, combines with oxygen (O2) at high temperatures. Other pollutants, such as PM and CO, are formed due to incomplete combustion of fuel; whereas, SOx emissions are formed as a result of the sulfur content in the fuel [USEPA, 2009]. CO2 emissions are proportional to the carbon content of the fuel. Logically, this would mean emissions vary by fuel type. These emissions can be calculated using a simple relationship associating the amount of fuel consumed, the carbon content of the fuel (or carbon coefficient) and the fraction oxidized (usually estimated to be 99%) [USEPA, 2006]:
On the other hand, non-CO2 emissions (CH4, NOx, PM, SOx, etc.) are not directly proportional to fuel consumption and are affected by vehicle characteristics. Therefore, to accurately determine the effects, vehicle-specific emission rates/factors (e.g. mass of pollutant/mile) are used in combination with vehicle activity. These factors are a function of vehicle type, age, fuel, and emission control technology. Vehicle activity can be defined by vehicle miles travelled (VMT) or hours of operation, and depend on the units of the emission factor [NCHRP, 2006]. Vehicle activity can be used directly to calculate emissions by either using the vehicle fuel economy (miles per gallon) or fuel-based emission factors (grams per mile).
While these general mathematical relationships are typically used to calculate emissions, the level of accuracy is dependent on the approach used to define emission
31
production processes and number of variables considered in determining emission factors [IPCC, 2006]. With a top-down approach, emission estimates are obtained using aggregate fuel consumption. These estimates can be reasonable when emissions calculation at the macro-scale level is needed (e.g. national emissions inventory). With a bottom-up approach, often used at a micro-scale level (e.g. project level emissions), more detailed inputs for fuel consumption and emission factors are required. It must be noted that while the bottom-up approach may also be used for inventory purposes, obtaining large-scale, detailed data inputs for emissions calculations often proves to be difficult. Both fuel consumption and emission factors are associated with/dependent on several variables which influence the performance of a vehicle and, therefore, the amount of fuel consumed or mass of pollutants emitted. The power-demand of a vehicle is dependent on various inter-linked parameters, such as vehicle characteristics (type, age, mass, etc.), operating mode (start/stop, running, idle and vehicular speed) and environmental parameters (road characteristics, temperature, humidity, etc.). A schematic illustrating the relationships of these factors to emissions is shown in Figure 3.1.
Figure 3.1. Schematic illustrating relationship of variables to vehicular emissions (SHA)
32
As illustrated in the figure, modal and environmental parameters, vehicle and road characteristics, fuel type, and traffic conditions, all affect a vehicle’s power demand, which in turn affects the resulting emissions produced. For example, use of alternative fuels or newer and more efficient vehicles or driving on a relatively flat road would typically produce lower emissions. Also, modal parameters significantly affect emissions. For example, emissions produced when a vehicle is turned off and restarted before the engine has cooled down (hot start) are lower as compared with those when the vehicle is initially turned on (cold start). Depending on traffic conditions and road characteristics, vehicles at higher speeds or accelerating from low speeds produce more emissions as compared to at higher speeds. Moreover, road characteristics (e.g. road grade, number of lanes, the number and type of specific traffic control devices, surface conditions, etc.) also influence the traffic flow and density. For example, large road grade and high traffic volume result in a large number of stops and starts and, therefore, emissions [Bachman, 1997]. The incorporation of these variables in determining emission factors and fuel consumption, that is, working on a micro-scale level, would produce more accurate and realistic results. Of the many models that currently offer emission estimates for on-road vehicles are traffic simulation models, like CORSIM, S-PARAMICS, INTEGRATION, and TRANSIMS. These are some of the most widely used traffic simulation models. These, along with other models, such as DYNASMART-P, represent driver behavior and vehicle kinematics for individual vehicles and trips, and are able to replicate and enable analysis of a variety of traffic-related activities. Many of these traffic simulation models account for environmental impacts of traffic related activities either by using emissions estimation modules integrated within the model or by using external microscopic emissions estimation models as plug-ins to determine emission outputs. The plug-ins provides instantaneous emission rates based on the vehicular inputs from the simulation models. However, a major disadvantage of using these models is that these models use averages for speeds, acceleration, deceleration and fuel consumption in order to generate emissions output. In most cases, the methodology employed to estimate emissions is not described. Therefore, while these models that best simulate traffic have the potential to provide a level of detail required for micro-level emissions estimation, they use a macroscopic or undefined methodology for emissions estimation. 33
Of the external microscopic emissions estimation models (sometimes used as plugins with traffic simulation models), CMEM and MOVES created by University of California-Riverside and USEPA, respectively, are the most notable and comprehensive. These models use a power demand approach to capture the physical processes of emissions production, incorporating a vehicle’s modal parameters, and hence, provide more accurate emission estimates [CMEM, 2010; USEPA, 2011]. Although both models have many benefits and are capable of producing microscopic, modal emission results, neither accounts for several important factors relevant to assessing the effects on emissions produced from traffic conditions and changes to vehicle composition on roadways. Some of the disadvantages of using these models lie in the scope of the variables they cover, level of detail captured in the outputs, and the limited flexibility they offer users. 2.4.2 Fuel Consumption and Emission Calculation Methodology Similar to MOVES and CMEM, a power-based approach was used for this research project to estimate emissions wherein vehicle characteristics and modal parameters, namely vehicle mass, velocity and acceleration, are used to calculate the instantaneous power demand (Pv,t) for a vehicle type category. When combined with the speed-based engine parameters (e.g. K, N, V), this approach provides an instantaneous fuel rate (FRt). The fuel rate is then multiplied by fuel-based emission factors (EF) to produce emission estimates for criteria air pollutants, such as HC, CO, NOx, CO2. Furthermore, the calculated instantaneous fuel rate (FR) when multiplied by the fuel-based sulfur content and other variables provides the associated instantaneous SOx emissions output for the vehicle. The equations and related data used to determine second-by-second emissions output for this research project are described next. The nomenclature used in the Equations (Eq. 3.1) is listed in Table 3.1.
Table 2.8 Nomenclature for variables used in equations LDV LDT PV,t M vt at c1 A : : : : : : : : Light-duty Vehicle (e.g. passenger cars, SUV, etc) Light-duty Trucks Instantaneous Tractive Power of vehicle V at time t (KW) Vehicle mass (metric tonne) Vehicle speed at time t (mph) Vehicle acceleration at time t (mph/s) Conversion factor for speed: 0.447ms-1/mph Rolling resistance coefficient (KW/mps)
34
B C r g FRt HV ? Kt Nt V S g/gtop K0 EMPol EFPol ?Fuel Fuel EconomyLDV/LDT T EMSOX SCFuel
: : : : : : : : : : : : : : : : : : : :
Rotational resistance coefficient (KW/mps2) Aerodynamic drag coefficient (KW/mps3) Road grade i.e. slope (%) Gravitational constant: 9.81 m/s2 Fuel consumption rate (g of fuel/s) Heating Value (KJ/g) Engine Efficiency = 0.4 Engine friction factor at time t (KJ/rev*L) Engine speed at time t (rps) Engine displacement volume (L) Engine Speed to Vehicle Speed Ratio (rpm/mph) Gear ratio 0.22 KJ/rev-litre (average based on range 0.19-0.25 KJ/rev-L) Emission for pollutant (g) Emission factor for pollutant (g/mile) Denisty of fuel (g/gal) Fuel Economy for vehicle category (gal/mile) Total time travelled by vehicle category (s) SOx Emission (g) Sulfur Content of Fuel (ppm)
2.4.3
Power Demand (Pv,t) and Instantaneous Fuel Consumption (FRt) Calculation The power-demand approach breaks down the emissions generation process of a
vehicle into the physical processes of the vehicle’s engine that correspond with vehicle operation and emissions production. As previously discussed vehicle performance during various driving conditions directly contributes to fuel consumption and resulting emissions. For example, vehicle characteristics, like age and engine size, would determine how quickly the vehicle can move in and out of periods of high power demand (e.g. overcoming high gradients or reaching desired speeds by accelerating). Therefore, estimating the physical processes that a vehicle undergoes during operation can provide higher resolution in defining a vehicle’s emissions production process. These processes are best captured by the engine’s tractive power (Pv,t), which in turn is based on the vehicle’s modal parameters (e.g. vt and at) and road characteristics (i.e. r). At a given time t, the instantaneous tractive power is defined as:
[ { ( ? )}] Eq. 3.2
35
The instantaneous modal parameters, speed (vt) and acceleration (at) at time t, were obtained directly from VISSIM outputs (or by other means); whereas, the vehicle parameters, such as mass (M) and the vehicle track road-load coefficients (A, B and C) for each vehicle category were obtained from USEPA’s MOVES model. These values were estimated by USEPA using vehicular data for LDVs and LDTs from inspection and maintenance programs and developing linear models to determine the coefficients from vehicle mass, M [Koupal et al, 2004]. The vehicle parameters used here are listed in Table 3.9.
Table 2.9 Calculation of Road -Load coefficients [Source: USEPA, 2011c]
Vehicle Category LDV(passenger cars) LDT(Trucks, SUVs,etc) LHD<=14K LHD<=19.5K Source mass(metric tons) 1.4788 1.86686 7.64159 6.25047 A(KW/mps) 0.156461 0.22112 0.561933 0.498699 B(KW/mps2) 0.00200193 0.00283757 0 0 C(MW/mps3) 0.000492646 0.000698282 0.00160302 0.00147383
Equ atio n 2.1 Equ atio n 2.2
The vehicle’s power demand directly influences the amount of fuel consumed, and therefore the mass of pollutant produced. The fuel consumption rate (FRt) or the energy used per second to operate the vehicle is a function of engine speed (Nt) and the engine friction factor (Kt), which captures the energy used to overcome engine friction per engine revolution and unit displacement. Both Nt and Kt are dependent on various speed-related vehicle parameters. The simplified equation used to calculate FR appropriate for mesoscale emissions estimation (as required here) was obtained from Barth et al. (2000) and is expressed as:
[ ], Eq. 2.3
where
Eq. 2.4 [ ] Eq. 2.5
36
The values for S (based on vehicle type category) and g/gtop (based on vehicle speed, vt) and fuel based variables (i.e. HV) are recorded in Table A.4 and Table A.3 of Appendix A respectively. 2.4.4 CO2, CO, HC, CO & NOx Emissions Calculation For the purpose of this project, fuel-based emission factors (i.e. mass of pollutant produced per unit of vehicle activity), EFPol, for the LDV and LDT vehicle categories for major fuel types (i.e. gasoline and diesel) were obtained from the USEPA (refer to Table A.2 in Appendix A). These emission factors in combination with other variables specific to the vehicle categories (e.g. fuel economy, time spent on roads, etc) and fuel (i.e. density, EFPol) were then used to calculate the emissions output for each pollutant (EMPol) using Equation 3.6.
( ) { } Eq.3.6
2.4.5 SOx Emissions Calculation The sulfur-content in a fuel affects the amount of SOx emissions produced when fuel is consumed. Therefore, the sulfur content (SCFuel as obtained from Table A.3 in Appendix A) for gasoline and diesel were used to estimate the SOx emissions for a vehicle category using the following relationship:
(
)
Eq. 2.6
2.4.6 Fuel Consumption To estimate the total fuel consumed by a vehicle category due to effects on its modal profile caused by changes within the traffic scenario, the power-demand based FR as calculated previously was used as shown in Equation 3.8.
? ( ), Eq. 2.7
2.4.7 Assumptions While there are many other variables that might have been considered, such as engine speed, air-to-fuel ratio, fuel use and catalyst pass fraction, vehicle emissions are
37
most influenced by engine power and fuel use. Also, since the scale of this project requires meso-scale emission results, and since all other variables require additional detailed vehicle-specific parameters (based on dynamometer measurements of each vehicle type by brand), these variables were not used in determining Pv,t [Barth et al., 2000].
2.5
Secondary Incident Savings An incident is called “secondary” if it is a consequence of a primary incident. The
occurrence of such a secondary incident is related to the duration of a primary incident (Khattak et al., 2008; Zhan et al., 2008). Therefore, as SSP programs aim to decrease the duration of primary incidents, they also decrease the risk of secondary incidents. In fact, it was noted in Karlaftis et al. (1998) that for every minute of additional incident duration, the risk of occurrence of a secondary incident increases by 1.7% in the winter and 3.1% in all other seasons, for an average of 2.8%. They fitted two logistic regression models to primary crashes assisted by SSP vehicles associated with the Hoosier Helper program in Indiana. Crashes within 3 miles upstream and within the clearance time plus15 minutes of a primary crash were classified as secondary. The odds ratio, which measures the strength of connotation between a primary incident characteristic and the probability of secondary incident occurrence, is presented. Odd ratios of clearance time in winter and all other seasons are estimated as 1.018 and 1.032, respectively. In other words, the SSP program could reduce the probability of secondary incident occurrence by 18.5% in the winter and 36.3% in all other seasons per incident to which they respond. The first step in quantifying the savings in secondary incidents is to estimate the number of incidents that are secondary. However, there is little agreement among researchers in terms of the validity of methods aimed at identifying and classifying secondary incidents. The primary approaches to such identification and classification are static threshold, dynamic threshold, and simulation-based filtering methods. Table 3.10 lists the classification methods in the literature. Raub (1997) employed temporal and spatial thresholds to classify incidents. Any incident that occurs within 15 minutes of the resolution of another incident and within 1
38
mile is defined as a secondary incident. Applying this method, 15-percent of all incidents reported by police were found to be secondary. Other studies used similar fixed thresholds,. For example, Moore et al. (2004) defined incidents as secondary if they occurred within 2 hours and 2 miles from incident identification. This static method is also adopted by Karlaftis et al. (1999), Hirunyanitiwattana et al. (2006) and Zhan et al. (2008). One drawback of the static threshold method is that it cannot capture field conditions (i.e. changing demand), and therefore, leads to misclassification. The dynamic threshold method was developed to compensate for the shortcomings of the static approach. Sun et al. (2007, 2010) proposed a master incident progression curve to identify secondary incidents. The progression curve is constructed from affected distance, which is measured from incident location to the end of its queue. Instead of using a static maximum or average queue length, the author marked the end of varying queue throughout the entire incident duration. Incidents that fall under the curve are considered to be secondary. It was concluded that the method reduced Type I errors by 24.38-percent and Type II errors by 3.13-percent. Similarly, Zhang and Khattak (2010) employed a dynamic queue-based method in which queue length is calculated by a deterministic queuing model (D/D/1). Zhan et al. (2009) classified secondary incidents as those that occur within the boundary of the estimated maximum queue length and dissipation time of a lane-blocked primary incident. To arrive at this conclusion they used a cumulative arrival and departure traffic delay model. A simulation-based secondary incident filtering (SBSIF) method was proposed by Chou and Miller-Hooks (2008). This model used geometric boundaries to analyze the incident impact area in a time-space contour map of traffic speeds. Regression models are established for identifying the corner points of the impact area. The authors conclude that the SBSIF method can reduce the misclassifications by up to 58-percent. They noted that 4% of 693 incidents to which H.E.L.P. responded to were secondary using this approach. Regardless of the method used in distinguishing secondary incidents from primary incidents, once the number of secondary incidents as a fraction of primary incidents is known, the next step is to estimate savings in secondary incidents, i.e. the number of incidents that did not arise secondary to a primary incident as a result of reduction in 39
incident duration. Chou and Miller-Hooks (2008) assumed the number of secondary incidents without SSP is linearly correlated with the total delay ratio between with and without SSP response cases. Similarly, Guin et al. (2007) employed an equation based on the ratio of average incident duration of SSP versus non-SSP incident responded cases. Table 3.10 provides an overview of secondary incident methodologies, assumptions and assumed rates from the literature. In general, Federal Highway Administration (FHWA) noted that approximately 20% of all incidents are secondary incidents. They include not only crashes, but engine stalls, overheating, and running out of gas as types of secondary incidents.
40
Table 2.10 Secondary incident classification methods
Data Sample Size
Authors (Year)
Definition
Method
Claimed Percentage/Main Finding
Data Source
Data Time and Location
Raub (1997)
Occur within 15 min from incident resolution and within 1 mile
Fixed temporal and spatial parameters
>15% of all incidents reported, average secondary crash occurs within 36.4 min, 600 meters after primary accident, average primary accident duration is 45 min, added of delay 69 min
1,796
Police reports
1995 Northern Chicago, IL
Karlaftis et al. (1999)
Occur within 15 min from incident resolution and within 1 mile
Fixed temporal and spatial parameters
>15% of all incidents reported
Indiana DOT
1992-1995, Borman Expressway
Moore et al. (2004)
Occur within 2 hrs and 2 miles from incident identification
Fixed temporal and spatial parameters
Secondary accidents are considerably rarer events than previous studies suggest, lower frequency of secondary crashes, 1.5%~3% per primary incident
84,684
CA Highway patrol
March, May, July 1999 and Dec.1998, LA Freeway
Hirunyanitiwattana et al. (2006)
Occur within 1 hrs and 2 miles from incident identification
Fixed temporal and spatial parameters
Secondary crashes occur more often during rush hour traffic in the morning and evening, rear-end collision is the predominant secondary collision type, accounting for 2/3 of all secondary crashes.
70,6 in 1999 and 183,988 in 2000
FHWA
1999 and 2000, CA highway system
41
Sun (2007,2010)
Occur under a master incident progression curve
Dynamic threshold method by marking the end of varying queue throughout the entire incident
dynamic method reduce Type I error by 24.38% and Type II by 3.13%; Results from using dynamic method versUSstatic method can differ by more than 30% in identifying secondary incidents
5,514
Highway patrol in St. Louis, MO
2002, I-70, MO
Zhan et al. (2008)
Occur within 2 miles and 15 min from incident resolution
Fixed temporal and spatial parameters
Average rate of 7.94% as primary incidents, 5.22% as secondary crashes, secondary crashes are usually much less severe than other crashes. Traveler sight conditions and lane blockage durations of primary incidents are significant contributing factors for determining the severity of secondary crashes
4,435
SMART database of FDOT
Jan 2005-Jan 2007 Fort Lauderdale, FL
Zhan et al. (2009)
Occur within the boundary of estimated maximum queue length and dissipation time of the potential lane-blockage primary incident
Cumulative arrival and departure traffic delay model to Estimate maximum queue length and associated queue recovery time
5% as primary incidents , 3.23% as secondary crash, accidents occur in daytime and with long lane-blockage duration increase the possibility of secondary crashes
4,435
SMART database of FDOT
Jan 2005-Jan 2007 Fort Lauderdale, FL
42
Zhang and Khattak (2010)
Occur within the a queue associated with primary incident and duration of primary incident, contained event duration: durations of all secondary incidents are contained within primary incident duration, extended event duration: duration of one or more secondary incidents partially overlaps with primary incident duration but extend beyond it
Dynamic queuebased method which queue length is calculated through a deterministic queuing model(D/D/1)
Contained and extended events show different characteristics and operational response patterns. Factors associated with durations of longer cascading events include primary incident being a crash, secondary incident being crash, multiple vehicles involved in secondary incidents, and longer time gap between primary and secondary incidents
37,379
Traffic Operations Center, VDOT
2005, Hampton Roads
Zhang and Khattak (2010)
Occur within a queue associated with primary incident and duration of primary incident
Dynamic queuebased method which queue length is calculated through a deterministic queuing model(D/D/1)
96.93% independent incidents with average of 14 min primary incident duration, 2.7% primary-secondary pairs of 40 min duration of primary incident, 0.37% primary-multiple secondary events of 68 min primary incident duration, characteristics of primary incident including crash, long duration, multiplevehicle involvement and lane blockage and road geometric variable increase secondary incident frequency
37,379
Traffic Operations Center, VDOT
2005, Hampton Roads, VA
Chou (2010)
Occur within the incident impact area on time-space traffic speed contour map
Simulation based secondary incident filtering method
24 and 27 out of 630 potential secondary incidents are identified employing visual and regression implementations for corner point identification with SBSIF method
693
New York DOT
2006, I-287, NY
43
2.6
Summary In this chapter, the essential measures required to evaluate the benefits of a SSP
program are identified. Factors affecting travel delay, fuel consumption, and pollutant emissions were reviewed. Assumptions and computational methodologies for computing the MOEs were also introduced. In the next chapter, the importance and ability to experimentally capture the effects of each factor in MOE estimation is studied.
44
3
CHAPTER 4. Implementation of Variables in Simulation Designs
The SSP-BC Tool must be comprehensive to enable the estimation of program benefits under all possible incident scenarios for which any user may require results. A simulation-based method was developed to estimate travel delay and fuel consumption of an incident scenario. Travel delay is a direct output of the simulation software, but fuel consumption is derived from modal parameters related to vehicular movement details, obtained from the simulation (e.g., velocity, acceleration, mass, etc.). Savings in emissions are computed from total fuel consumption. Savings calculation in secondary incidents partially benefits from obtained travel delay values. Any given incident falls into one combination of the identified factors as each incident is defined by its location, lanes blocked, and duration, as well as by the settings related to roadway grade, weather and other contributing factors discussed in Chapter 3. To account for all combinations of these factors related to incidents on freeways with between 2 and 6 lanes, up to 2 lanes blocked (no road closure considered), and clearance times of less than 90 minutes, i.e. all combinations within the study scope, combinations over 8,000,000 simulation runs would need to be conducted. Additionally, some of the factors cannot be controlled through code and thus, automating all of the runs through batch runs is not possible. Even using a powerful computer and even if all runs could be automated; it would take more than a year to complete. This means that simulating all incident scenarios is impractical. Instead, simulation runs are supplemented through statistical approaches described in this chapter. A set of preliminary experiments were conducted to gain insight into the impact of each factor on travel delay and fuel consumption and to choose appropriate statistical estimation (i.e. supplemental) models to employ and methodologies for their calibration. A control case was designed and simulation runs were taken in which the state of a single factor was changed univeriately. These experiments were designed to answer two questions: (1) Does the simulation technique replicate the factors appropriately in the simulation environment? (2) What form should the statistical model take and it include?
45
Multiple-regression modeling was chosen to estimate travel delay and fuel consumption of a given incident from a sample of the possible incident characteristic combinations (i.e. incident scenarios). Results of this initial set of experiments showed that it is best to develop different regression models for shoulder, one- and two-lane blockage scenarios. Moreover, travel delay for LDVs and LDTs must be computed separately, because the monetary benefits associated with savings incurred by trucks is incorporated within the truck driver’s hourly costs. Therefore, for travel delay, one multi-regression model was developed for each of the six categories of incidents and vehicle types as shown in Figure 4.1. Based on results of additional preliminary experiments to assess the impact of the various factors on fuel consumption estimation, one general model was calibrated to account for all incident scenarios.
Figure 3.1 Estimation model development progress
The goodness-of-fit and modeling assumptions were tested. Results from these tests indicated that the developed models needed improvement. Thus, a hybrid statisticalsimulation data approach was introduced for improving the travel delay regression models,
46
and a non-linear regression model was calibrated to improve the goodness-of-fit of the fuel consumption estimation model. This progression is also shown in Figure 4.1. The simulation platform, general settings and incident simulation techniques are described in Section 4.1. Results of single-factor experiments are described in Section 4.2. Regression models and investigation of their goodness-of-fit are presented in Section 4.3. This is followed by a summary in Section 4.4. 3.1 General Simulation Settings and Incident Modeling While many high quality microscopic simulation software programs, like PARMICS, VISSIM and CORSIM, areavailable that for many purposes adequately model traffic (Brockfield et al., 2004; Ranjitkar et al., 2004; Jones et al., 2004; Bloomberg et al., 2003), some studies have revealed that some platforms, like VISSIM and PARAMICS are better than others, specifically CORSIM (Choa et al., 2002). Given positive experience with VISSIM in the literature, as well as prior experience, including extensive calibration studies (Miller-Hooks et al, 2010), by the authors with VISSIM, VISSM was chosen as the simulation modeling platform. . An additional benefit of working with VISSIM is its COM interface. The interface provides great flexibility in controlling various aspects of the simulation environment. Primary calibrated VISSIM parameters, assumptions, and method for modeling incidents are described in Section 4.1. In Section 4.2, methods of implementation for each of the factors in the simulation environment and their impacts on travel delay and fuel consumption given by results of numerical experiments on individual factors are explained. Results of these experiments are combined to reproduce more realistic incident scenarios in which multiple factors change concurrently. Multiple-regression modeling is employed to estimate travel delay and fuel consumption as explained in Section 4.4. This is followed by an improved estimation model for travel delay exhibited in Section 4.5 and a summary of the Chapter in Section 4.6. 3.1.1 Simulation Settings For each incident scenario and seed, one run of VISSIM involves 7,200 seconds of simulation time. A typical incident scenario is explained in next section and shown in
47
Figure 4.1. The software user manual suggests the use of a warm-up period. This period includes the first 1,800 seconds of each run; this period is also required to achieve a steady traffic flow along the segment from the start of the analysis period. Incidents are designed to occur 300 seconds after the end of the warm-up period or 2,100 seconds into the simulation. To get more accurate results, the VISSIM user manual suggests running a minimum of three runs with different random seeds for each simulation model and reporting the average over random seeds in the final results. Average results, based on the
5,400 seconds of simulation run time, over runs with five randomly chosen seeds were collected. The software user manual suggests 1 time step per simulation second in terms of
simulation resolution for models that contain only vehicles (i.e. that are not multimodal). However, five time steps per simulation second was used in this study. It was observed in preliminary runs that in modeling congested roadway conditions, using a higher resolution reduces loss of vehicles due to difficulty by the vehicles in entering the network. Each run required approximately 2.5 minutes on a Dell Precision T7500 personal computer with a 3.20 gigahertz quad core processor and 12 gigabytes of RAM, running a 64 bit Windows 7 operating system. 3.1.2 Control Case A typical incident to which SSP vehicles responded was selected and designed based on the information in Chapters 2 and 3 to be the control case incident for the singlefactor experiment runs. The studied segment is a three lane, 10-mile, unidirectional straight freeway segment with no on-ramps, off-ramps, grade, or lane drops as shown in Figure 4.1. The incident duration is 20 minutes (described in Section 4.1.3) which would not change with experiments, the traffic volume is set to be 1200 vplph. The FFS of vehicles is set to be 75 mph. In each experiment, all characteristics are set as base except for the factor under study. 3.1.3 Simulating Incidents and Rubbernecking Effect The COM interface of VISSIM allows users to control various aspects of the simulation, which makes the software highly flexible. However, VISSIM and its COM interface do not include a specific feature for modeling traffic incidents. Simulation-based
48
traffic studies relying on the VISSIM platform have used a variety of methods to model incidents, including: defining a parking lot with one space and assigning a car to park in the space for a fixed time period (Wang et al. ,2003; Pulugurtha et al. , 2002), setting a bus stop at the incident location in which the bus stops for a fixed period of time (Hadi et al.,2000), using active traffic signal control in the affected lanes for the duration of an incident (Saka et al. 2004) and setting a passenger vehicle with speed of zero in the incident location from the start time of the incident to the end of clearance time (Chou and Miller-Hooks, 2010). In this study, two cars with zero speed are co-located at the incident location (lane, and location with respect to ramp) from the start time of the incident for a pre-set incident duration time. VISSIM provides a function that allows users to set a temporary reduced speed area on a link. Such reduced speed areas can capture the effects of rubbernecking during an incident. That is, they can be used to model the reduction in speed in unblocked lanes during the incident time period. Hadi et al. (2000) found that a speed of 20 mph for vehicles modeled in VISSIM in which an incident is active results in the suggested available capacity by the HCM. At the end of the incident period, vehicles involved in the incident are removed from the blocked lanes. Once the reduced speed areas become inactive, vehicles traveling in the affected lanes accelerate until they reach their initial desired speed (Figure 4.2).
49
Figure 3.2 Incident layout on typical three-lane unidirectional freeway segment
3.1.4 VISSIM Calibration To correctly predict system response, it is essential to calibrate the simulation software to existing traffic conditions. Miller-Hooks et al. (2010) identified five carfollowing and lane-changing parameters in VISSIM that had very significant effect on travel delay estimation. After completing an extensive effort to calibrate a model of a 41mile Maryland freeway (82 miles in both directions) against actual travel time measurements, they suggested changes to four of the five values. The suggested parameter settings are: ‘Following’ Variation (CC2), ‘Following’ Thresholds (CC4&5), Safety Distance Reduced Factor (SDRF), and Look Back Distance (LBD). Their definitions, default values, possible range in VISSIM, and the final set that are used in this study are listed in Table 4.1. CC2 to CC5 belong to the Wiedmann 99 car following model, which is
50
mainly suitable for interurban and freeways. SDRF and LBD are lane-changing parameters associated with driver behavior.
Table 3.1 Driver behavior parameters, adopted from Miller-Hooks et al. (2010)
Parameter CC2 CC4&5 SDRF LBD Definition Following variation: desired safety following distance Lower & Upper following threshold Safety distance reduced factor: effects safety distance during lane changing Look back distance: defines the distance at which vehicles will begin to attempt to change lanes Default Value 4 meters 0.35 mph 0.6 200 meters Range 1.5-20 meters (16.40-65.62 ft) 0.1-2.0 0.1-0.9 50-1000 meters Final 39.37 (ft) 0.1 0.1 3280.83
Note: the sign of Lower following threshold (CC4) is ‘-’ and the sign of Upper following threshold (CC5) is ‘+’. 3.2 Single-Factor Experiments Finding a way to model significant factors of travel delay and fuel consumption in VISSIM simulation environment and monitoring their impact on travel delay and fuel consumption are discussed in this subsection. Results of this section helped to choose the appropriate explanatory variables for estimating travel delay and fuel consumption. When a factor was identified as insignificant in the relevant travel delay or fuel consumption estimation model, it was not included in the estimation model or the approach used to capture the impact of that factor in the simulation environment was changed. Factors were designed to change univariantly in the control case in various simulation runs. Results in terms of travel delay and fuel consumption estimates were analyzed for a set of experiments associated with each factor. A summary of the studied factors and ranges on their values is presented in Table 4.2.
51
Table 3.2 Summary of variables used in numerical experiments
General Attributes Factors Segment length Number of lanes and average lane width Geometry of the roadway segment Lateral clearance (shoulder) Ramps Horizontal curves Segment gradient FFS Traffic characteristics Ramp FFS Traffic flow rate Percentages of trucks in traffic flow Incident severity Incident attributes Weather conditions Average incident duration Rubbernecking effect Rang used 10 mile 2-6 lane, 12 feet 6 feet 0 to 10 ramp/mile Straight, Mild, Sharp -10 to +10 percent 55 to 75 mph 25 and 35 mph 200 to 2200 vplph 0 to 18 percent Shoulder, 1-lane and 2-lane Blockage 0 to 90 minutes (5-minute increment) 500 feet Upstream of Incident Location
Clear, Light Rain, Heavy Rain, Snow, Fog, Icy condition, Low Visibility, Wind
3.2.1 Geometry Factors: Number of Lane and Lane Blockages To describe the number of main lanes that are blocked due to an incident, average travel delay per vehicle is calculated for possible combinations of number of lanes and lane blockage within the study range (Table 4.3). Total travel delay divided by total number of vehicles gives average travel delay per vehicle in a simulation run. Table 4.3 indicates a decrease in the average travel delay per vehicle when the number of lanes in a segment increases. The impact of one lane closure on travel delay is much higher in freeways with fewer of lanes, but the difference between travel delay of one lane and two lanes blockage scenarios was not large. One closed lane due to an incident results in a travel delay increase of approximately 13 times that in a two-lane freeway and 5 times that for a six-lane freeway. However, when two lanes are closed due to an incident, travel delay is approximately two times larger compare to one lane blockage for a comparable roadway segment. It is shown in Table 4.3 that fuel consumption is not very sensitive to the number of lanes blocked. This is because the relationship between fuel consumption and vehicular speed has a parabolic shape. Fuel consumption at lower speeds in a stop-and-start mode due to congestion is nearly the same as at higher speeds (e.g. an average speed of 70 mph) with no stops and starts. Therefore, number of lanes blocked due
52
to incident can be eliminated from the fuel consumption estimation model. An approximate trend for travel delay versus fuel consumption is shown in Figure 4.4.
Table 3.3 Number of lanes and lane blockage analysis
Travel delay (hour/vehicle) state of factor two lane three lane four lane five lane six lane
N/A: Not Applicable
Fuel consumption (gallon) 2-lane blockage N/A 240.76 199.53 129.87 109.45 shoulder blockage 307.5198 275.2585 614.0988 775.1365 925.1616 1-lane blockage 305.4774 331.2327 613.5244 777.0599 923.8499 2-lane blockage N/A 319.7124 614.5162 781.0039 933.9603
shoulder blockage 7.44 10.12 13.94 14.93 17.35
1-lane blockage 100.23 110.34 104.45 88.543 77.84
Figure 3.3 Number of lanes and lane blockage analysis
53
A linear trend can be seen in both travel delay and fuel consumption graphs given in Figure 4.3. However, travel delay is greatly affected by lane blockage. Therefore, it was necessary to develop separate estimation models for travel delay for different lane blockage incident scenarios. 3.2.2 Geometry Factors: Ramps The FFS along the ramps, exiting traffic flow and location of incident with respect to ramp locations are important aspects that have been considered in simulating incidents in close proximity to ramps. The impact of incident-ramp proximity on travel delay and fuel consumption is studied. The simulation roadway segment is set to have one off-ramp. (Figure 4.2 compare to Figure 4.1). Two ramp FFSs (25 and 35 mph) are tested. Additionally, two exiting flow percentages (25 and 50) of the main lane traffic volumes are considered. Studies have shown that ramps are significant sources of bottlenecks in freeway operations, affecting traffic as far as a quarter-mile upstream and downstream of merge and diverge points (Zhang et all, 2009). Five incident locations, each set within a half-mile of the off-ramp location, are modeled as depicted in Figure4.4. The average travel delays and fuel consumption over 5 random seeds for each case are reported in Table 4.4.
Figure 3.4 Simulating incidents close to an off-ramp
The distance needed for a vehicle to start a diverging maneuver is set as the number of lanes the vehicle must pass multiplied by the look-back distance. For example, if the look-back distance is set as 200 feet, a vehicle in lane three will start changing lanes to reach the right-most lane beginning from 400 feet upstream of the off-ramp.
54
Table 3.4 Ramp analysis
State of the factor (ramp speed-exiting volume) 25mph- 25% 25mph- 50% 35mph- 25% 35mph- 50% base case Travel delay (vehicle-hour) Passenger cars 13.106 13.106 10.484 10.484 10.295 Trucks 0.180 0.180 0.131 0.131 0.116 Fuel consumption (gal) Passenger cars 613.1 613.3 614.2 614.1 614.1 Trucks 15.4 15.7 15.6 15.7 15.8
As anticipated, the lower the ramp FFS, the higher the travel delays incurred and greater the fuel consumed. However, even for arising incidents close to the ramp with design speed of 35 mph and the main lane FFS 70 mph, the ramp does not affect travel delay significantly. However, fuel consumption decrease since portion of vehicles moves slower nearer to ramp and vehicles generally have higher fuel efficacy in the range of 30 to 50 mph range than at 70 mph. To include the incident-ramp proximity impact on travel delay and fuel consumption in SSP-BC Tool, the speed design of every ramp on the study segment should be included as input. However, the information does not make significant difference in travel delay and fuel consumption outputs. A more practical way to include the impact of ramps, therefore, was selected for the SSP-BC Tool as explain in Sections 3.2.1.4 and 3.2.3.2 in which capacity reduction due to ramp density is applied. 3.2.3 General Terrain: Horizontal Curves Given a specific number of points, VISSIM provides an option to draw a Bezier curve when creating connector links. To assess whether or not the software adequately captures the effects of curvature in freeway operations on speed, preliminary simulation runs were conducted in which the travel time of vehicles traversing two similar segments, one curvy and one straight, were compared. No significant difference was noted between travel times in these runs. Thus, it was concluded that the operational effects of roadway curvature are not captured. To capture these effects, reduced speed areas and lower approaching desired speeds were used within curved areas of the test segment. This is consistent with freeways where speed limits are reduced around curvy roadway segments.
55
1000 feet of the base segment is modeled with curvature. Curves with three design speed were considered. The effect of the curvature is captured by setting speeds of 65, 55, and 50 mph along the 1000-foot length of the 70 mph segment. Incidents were placed randomly in different positions within the curve. Results of test cases were not significantly different from the base case in lower speed categories. Therefore, for roads with mild and sharp curves, 5 mph and 10 mph speed reduction respectively were applied to posted speed limits if the exact speed limit of the segment close to a curve is not available in detail. 3.2.4 General Terrain: Vertical Curves The gradient of the segment can be set manually in VISSIM. According to HCM 2010, the maximum grade in level terrain is 2% (-2%), in rolling terrain is 5% (-5%), and in mountainous terrain is 10% (-10%) as discussed in Section 3.2.2.2. The base case has 0% gradient. The results for different grades are shown in Table 4.5. Travel delay increases significantly as the gradient of the segment increases. Negative grades (downhill) do not have a significant impact on travel delay. The impact of grade on average vehicular speed can be discerned from results given in Table 4.6. The average speed of trucks is more affected by grade. The difference between average speeds of passenger vehicles and trucks widens as the grade increases.
Table 3.5 Segment grade analysis
percentage of grade -5% 5% 10% base case Travel delay (vehicle-hour) Passenger cars Trucks 10.5 0.1 26.1 0.2 83.6 5.3 10.295 0.116 Fuel consumption (gal) Passenger cars Trucks 515.5 12.9 397.4 9.6 289.6 14.5 614.1 15.8
Table 3.6 Impact of gradient on average speed
Average Speed (mph) percentage of grade -5% 5% 10% base case Passenger cars 61 52 37 60 Trucks 52 35 23 51
56
Fuel consumption reduces with reduced speed resulting from higher grades. The VISSIM manual explains that the impact of gradient on traffic flow is found in acceleration and deceleration of vehicles, “The possible acceleration decreases by 0.1 m/s² per percent of positive gradient (road incline).” To capture the effects of gradient and not changes in speed, on fuel consumption, all other factors, including speed, remain constant in this set of experiments. Thus, for different gradient settings, a constant speed is forced. This permits analysis of travel delay fuel consumption and emissions changes due to changes in gradient. Results are provided in Table 4.7. To maintain a constant speed along a gradient, a significant increase in fuel consumption rate is required.
Table 3.7 Fuel economy changes due segment grade
Fuel Economy (mpg) percentage of grade 0 5% 10% Passenger cars 22.06 16.18 11.50 Trucks 21.34 16.27 10.09
3.2.5 Traffic Characteristics: Speed of Vehicles In the simulation runs, the speed of vehicles is defined by the desired speed. VISSIM has a speed distribution for desired speeds from which it assigns a speed to every entering vehicle. If the speed of any vehicle for any reason (e.s. lane blockage) changes, vehicles reach their assigned desired speed after passing the obstacle. Four different speed regimes from 37.5 mph (60kmh) to 75 mph (120 kmh) were modeled. It was found that fuel consumption is highly sensitive to desired speed. In fact almost 50% reduction in fuel consumption was noted when desired speed changes from 120 to 60 kmh. Travel delay, however, slightly decreased with decreasing desired speed.
Table 3.8 Travel delay and fuel consumption changes by FFS
FFS (kmh) 60.00 80.00 100.00 Base Case (120) Travel delay (vehicle-hour) Passenger cars 7.66 8.16 8.42 10.295 Trucks 1.64 0.4 0.14 0.116 Fuel consumption (gal) Passenger cars 257.82 306.46 397.98 614.1 Trucks 9.36 11.04 12.54 15.8
57
3.2.6 Traffic Characteristics: Demand Flow Rate The average speed of vehicles in incident cases with prevailing traffic volumes prior to incidents between 200 and 2200 vplph, were obtained through simulation runs, results from which are shown in Figure 4.5. The sold line depicts the HCM-suggested speed for a basic freeway segment with FFS of 70 mph. The dash-square line Indicates VISSIM average speed, obtained from this study for the base segment under different flow rates. The dash-triangle shows average speed over the segment for an incident with 90minute duration. The dot line indicates the boundary of Level of Service E and F when density passes 45 passenger car per mile per hour (pc/mi/hr).
Figure 3.5 Average speed Vs. Traffic flow rate
80
70 60 Speed (mi/hr) 50 40 30 20 10 0 0 500 1000 1500 2000 2500 Flow rate (vplph) HCM 2010 45 pc/mi/hr no incident 90mins incident
From the 90-minute incident line, the average speed reported for the study segment with incident at a density of 45 pc/mi/hr, where 90-minute incident line passes the dot line, is approximately 35 mph. The HCM suggests a 15% capacity reduction due to shoulder incidents in a 4-lane freeway. It, also, provides adjusted speed-flow curves for indicated capacity reductions (i.e. due to incidents, Exhibit 10-9, HCM 2010). From the mentioned graph in HCM, the average vehicular speed under 15% capacity reduction in a basic segment at 45 pc/mi/hr is approximately 40 mph (FFS of 70 mph, capacity of 2400 pcplph). This indicates that for the simulated segment, capacities associated with lower 58
traffic flow rates are reached in comparison to ideal conditions in a basic freeway segment suggested in the HCM. Furthermore, from the “no incident” line in Figure 4.3, the capacity of the simulated segment closes in on 2000 vplph as average speeds for higher traffic flows become almost constant from this flow rate. Also, after 2000 vplph, the travel times don’t change significantly as traffic flow increases (Figure 4.6). Having 2000 vplph capacity in mind for the study segment under normal conditions, the corresponding average speed of vehicles to the capacity given a 15% capacity reduction in case of incident must be in the range of 35 to 38 mph. 35 mph value has been found from simulated incident case.
Figure 3.6 Travel time vs. Traffic flow
400 350 Total travel time (Sec) 300 250 200 no incident 90mins incident
150
100 50 0 0 500 1000 1500 2000 2500 Flow rate (pc/h/ln)
Findings reported in Figure 4.5 also indicate lower average speeds resulting from the simulation runs under lower traffic volumes as compared with expectations given in the HCM suggested values because the gap between previously mentioned adjusted speed-flow curve and actual speed-flow curves is smaller as low flow rates (Exhibit 10-9, HCM 2010). From Section 4.2.5, it was found that travel delay slightly decreased with desired speed. Thus, if maintaining the suggested speeds of at in lower flow rates, higher travel delay as HCM suggested values might be obtained assumed the capacity to be 2400 vplph for the basic segment.
59
3.2.7 Traffic Characteristics: Truck Percentages To test the impact of trucks on the operation of vehicles in the segment seven traffic composition cases with 0 to 20 percent truck traffic were tested. One-lane blockage incidents were modeled instead of shoulder blockage incidents to capture the impact of vehicle maneuvers. Maneuvers of trucks differ from those passenger cars and, therefore, it affects travel time and fuel consumption differently. Higher Percentages of trucks in traffic flow increase the travel delay and fuel consumption as expected (Table 4.9).
Table 3.9 Truck percentage analysis
percentage of trucks 0.00 0.05 0.10 0.12 0.15 0.17 0.20 Travel delay (vehicle-hour) Passenger cars 171.0673 174.5893 166.5927 160.7813 149.6053 144.7567 140.1433 Trucks 0 13.296 25.51267 30.78933 38.094 42.16 47.52067 SUM 171.0673 187.8853 192.1053 191.5707 187.6993 186.9167 342.1347 Fuel consumption (gal) Passenger cars 137.5791 134.7476 123.1948 119.4744 112.8164 109.7824 105.1306 Trucks 0 13.50501 25.98375 31.20691 37.56797 42.52419 49.08134 SUM 137.5791 148.2526 149.1785 150.6813 150.3844 152.3066 154.212
3.2.8 Simultaneous Changes of Factors In the prior section, the effects of individual factors on travel delay and fuel consumption in the presence of incidents was studied. In reality, multiple factors will exist that will simultaneous the impact these measures under such incident conditions and their effects are often nonadditive. In brief, the factors that directly were considered in the travel delay and fuel consumption estimation models are: Incident duration, number of lanes, number of lanes blocked, prevailing traffic volume, FFS, percentage of trucks and gradient as determined in Sections 4.2.1 to 4.2.7. No prior published study could be found that described a relationship between truck composition and/or roadway gradient with travel delay. Thus, additional analysis to test the independence of these factors and their impact on travel delay was completed. Simulation runs in which the number of lanes, truck percentage and/or segment gradient change concurrently were conducted. Results of these additional simulation runs indicated a constant increase in travel delay due to increased truck composition regardless of the
60
number of lanes. The same pattern was found for segment gradient. These results infer that additional delay due to percentage of trucks and grade change on the three-lane freeway test segment can be added directly to estimate for segments with any number of lanes. However, this was not the case for fuel consumption. 3.3 Multiple-Regression Analysis Multiple-regression relates two or more independent variables ( ) to a dependent variable (Y). Seven multiple-regression models are presented for travel delay and fuel consumption of cars and trucks of different lane blockage incident scenarios based on a design sample of incidents. The general form of multiple regressions is shown in Equation 4.1 where dependent and independent variables are i dimensional vectors. The parameters were estimated using a least squares method.
Eq. 3.1
To obtain the travel delay estimation models, the regression models were developed for different lane blockage scenarios as discussed in Section 4.2.1. In each category of lane blockage, two models are presented for light-duty and heavy-duty vehicles. Explanatory variables of the travel delay regression model were chosen to be incident duration, traffic volume, percentage of trucks and gradient of the roadway. The only parameter found insignificant in the conducted preliminarily experiments for fuel consumption was number of blocked lanes due to incidents. Thus, explanatory variables of fuel consumption regression model are: number of lanes in the segment, incident duration, traffic volume, , speed, percentage of trucks and gradient. 3.3.1 Minimum Sample Size A balance between accuracy and computation time must be chosen in selecting an appropriate sample size of incident scenarios for simulation runs. The larger the sample size, the more accurate the model and the better the estimation of parameters, but the greater the computational effort. Determining the minimum sample size of incidents, thus, is necessary. For this study, the population means (µ) method is employed to determine the minimum sample size required for the multiple-regression models. With anticipated effect size (ƒ2) of 0.05, statistical power level of 0.95, four explanatory variables and probability
61
level of 0.05, the minimum sample size required is 376 (Cohen et al. 2003). Thus, a sample size of 400 was used for each lane-blockage state. For the fuel consumption model 300 observations were used. 3.3.2 Designing a Sample of Incidents To create a random sample of incidents for estimating the parameters of the multiple-regression models, the explanatory variables were used as the design criteria of each incident. Since the correlation between the explanatory variables is unknown, it was assumed that the explanatory variables are independent and uncorrelated with one another. Where this assumption invalid some incidents might have a low likelihood. While some incident cases with very low probability may be generated in the sample used within the simulation and later to calibrate regression models, if appropriate modeling techniques are used, these samples will have little effect on the development of a regression model passing goodness-of-fit tests. In addition, if the domain of a variable is dependent to one another, the estimation model should be developed for that domain, since a general model of all points might have different local behavior. This issue is addressed here with using the real world ranges for generating the random variables and the appropriate probabilistic distributions best describing each criterion. The random values for each incident in the sample are generated as follows. If incident duration is a random variable, it will have a probability density function (PDF). Statistical methods have been employed by researchers to explain and estimate incident duration when treating it as a random variable. These methods treat the random variables with probabilistic distributions, conditional probabilities, linear and non-linear regression models, time sequential and others as discussed in Chapter 3. Golob et al. (1987), GIuliano (1989), Garib et al. (1997), Suvilllivan (1997) and Ozbay et al. (1999) found that the log-normal distribution very closely fit their freeway incident data. Ozbay et al. (1999) claimed that incidents with the same severity level have normal distributions, supporting the theory that incident duration is a random variable (Smith and Smith, 2002). Nam and Mannering (2000) found that the Weibull distribution is also capable to estimate incident duration of an incident sample. Smith and Smith (2002),
62
however, found that required goodness-of-fit tests for log-normal and Weibull distributions failed. To create a sample of incidents representing real data, therefore, incident duration cannot simply be generated randomly from a uniform distribution. The average and standard deviation of the incident durations used in the design sample for this study need to be close to the incidents to which SSP vehicles responded. Chou and Miller-Hooks (2008) found that the average incident duration of 80% of incidents that SSP vehicles responded to is 17.6 minutes in New York State with standard deviation of 18.07 minutes. First, the Weibull distribution was used to generate incident duration times for the sample, but calibration of its parameters to provide desirable average and standard deviation were not successful. MATLAB was employed for generating random variables from the inverse of the Weibull distribution. Boyles and Waller (2207) used a log-normal distribution with µ (mean)=3 and ?(standard deviation)=1.6 to describe the incident duration of the incidents. Herein, by searching within the vicinity of those parameters, a log-normal distribution with mean 2.8 and standard deviation of 1.4 was found that best fit the distribution parameters that were sought. Using the inverse of the defined distribution (Equation 4.2), a set of 400 random incident durations having a mean of 17.8 minutes, standard deviation of 16.9 minutes, maximum of 70 minutes, and minimum of 5 minutes was generated. The lognormal inverse function is defined in terms of its CDF as in Equation 4.2.
{ }, Eq. 3.2
where
?
?
Hourly traffic volume is an important input used in the simulation runs. However, in travel delay studies, traffic volume is not often addressed directly and a factor that can be related to it is time-of-day variable: a.m. and p.m. peak hours on weekdays and off peak hours during weekdays and weekends. However, due to lack of information on the connection between time-of-day and volume, a uniform distribution is used to generate random traffic volumes in incident cases. Percentage of trucks in traffic composition, and
63
gradient were assumed to be independent of one another and a uniform distribution was used to generate each of them. For fuel consumption sample, number of lanes also assigned to incident scenarios from a uniform distribution. 3.3.3 Multiple-Regression Models As mentioned in Section 4.3.1, multiple-regression was selected to model travel delay and fuel consumption of both light- and heavy-duty vehicles in each lane blockage scenario. First, the linear regression models of travel delay and fuel consumption based on explanatory variables were obtained. Next, composition of variables is introduced to the model and a set of non-linear regression models are presented for validation by various goodness-of-fit tests. A stepwise technique is employed to find the best subset of explanatory variables for models. The stepwise technique starts the regression with the best regressor. It then finds the next best variable to add to the model, and finally it checks all variables in each equation to see if the previously entered variables remain significant. Other techniques that might be used in place of the stepwise technique include MAXR which chooses the variables to add to the model so as to achieve the highest possible R-square value. The stepwise method terminates based on the Mallow’s statistics. Mallow’s is a goodness-of-fit test for
regression that used ordinary least squares for estimating the parameters. When the expectation of becomes close to the P value, the stepwise procedure terminates and the
final set of explanatory variables are introduced. SAS statistical software package was employed for the statistical analysis conducted herein. SAS is a combination of programs that were designed for statistical analysis of data. The package offers six variable selection methods. These methods present results in a set of candidate regression models from which the best is chosen. To choose the best estimation model for travel delay and fuel consumption from the set of candidates models, six approaches are exhibited as goodness-of-fit tests: coefficient of determination (R-square), adjusted R-square, Mallow’s , Akaike Information Criterion under the name of “an information criterion” (AIC), Bayesian Information Criterion (BIC) and Schwarz’s Bayesian information criterion (SBC) as exhibited in Equations 5.1a to 5.1d.
64
The first method, the coefficient of determination method, is not always reliable. The goodness-of-fit increases with the number of regressors added to the model and, thus, the more explanatory variables, the better the model appears to be. The adjusted R-square method can be used to compare models with different numbers of explanatory variables, because regressors are added to the model only if their entry leads to statistically significant improvements in the model. Like the adjusted R-square technique, the AIC method penalizes any additional unnecessary estimators and discourages overfitting. Assuming the error term within the model is normally distributed, the maximum log likelihood was derived for each candidate models. The derived likelihood of each model is then to compute AIC, BIC and SBC (Equation 5.1b, 5.1c, 5.1d). From this set of candidate models, the one having minimum value of AIC, BIC and SBC would be selected as the final model. For example, to find the travel delay regression model having 8 explanatory variables, first using the stepwise and MAXR are used two sets of variables. Each set results regression model. The one best fit the data is identified throughout each of goodness-of-fits that may lead to a first choice model. If we decide to use R-square as goodness-of-fit test, the model which has higher R-square would be the best model,but if we want to use AIC as the goodness-of-fit test, the model that gives the lowest value of AIC would be presented as the final model. 5.1a 5.1b 5.1c ( where N = number of observations, SSE= Sum of squared errors, P= number of explanatory variables, K= number of free variables <= P+1, (k= # independent variables + intercept), ) , 5.1d
65
MML= Maximum log likelihood of the model. 3.3.3.1 Travel Delay Regression Models Regression models were developed for each lane blockage scenario and vehicle type. Note that each scenario is assumed to arise with the same likelihood. Let TTD stand for total travel delay of an incident case, ID be incident duration, vol be traffic volume at the time of the incident, PT be percentage of trucks in the traffic composition and G represent the gradient of the road. First, the linear model is developed for light- and heavyduty vehicles and the shoulder lane blockage scenario with four explanatory variables as shown in Equations 5.1a and 5.1b. All variables were found to be significant at the 0.15 level. The models have R-square values of 0.6772 and 0.5682, respectively. LDV:
Eq. 3.3
LDT:
Eq. 3.4
The above regression models are based on four assumptions related to the dependent variables: independence, normality, homoscedasticity (constant variance of response variable) and linearity. The regression assumptions can be re-expressed in terms of modeling errors to validate the assumptions on which model is built. Random errors are independent, normally distributed, have constant variance and zero mean (Equation 4.1). Having these . In
conditions the random errors can be considered as a random sample from
addition, the best representation of errors is through standard residuals. Standard residuals are the difference between actual and predicted response variables for each observation with constant variance over different dependent variables. SAS calculates residuals with a variance of 1. In general, any systematic pattern in residuals indicates a violation in assumptions and systematic error. Fit diagnostics for the models, including residual graphs for each parameter were obtained and analyzed. A summary of goodness-of-fit test results for travel
66
delay of LDV is presented in Figure 4.7 as a sample of a full analysis. The behavior of other models and the analysis were very similar to this case. In this model, it appears that the linearity assumption is violated, because the residuals are not scattered randomly around zero and form a clear pattern. Also the variance of residuals seems to have two values and they value is not constant. It shows that the model does not have the same accuracy for all data points. A Quantile-Quantile plot indicates that theoretical and actual data distributions do not agree, as the plotted points are not approximately on the y=x line. The slope of the curve of the plotted points increases from left to right, which indicates that a theoretical distribution that is skewed to the right, such as a log-normal distribution, might better fit the data. In addition, the mild curve indicates a small shape parameter for the chosen distribution (i.e. ? for log-normal). Cook’s D shows no outlier points, as all data points are within a distance of 2 of the zero line.
67
Figure 3.7 Summery of fit diagnostic for total travel delay of LDV
As part of additional analysis, the residuals are plotted separately for each explanatory variable (Figure 4.8). Since the variables are uncorrelated by design, each graph shows the direct relationship of dependent variable and explanatory variable. Incident duration has a random scatter plot matching its log-normal distribution. Residuals associated with the truck composition are also randomly scattered around zero, so the linear assumption seems reasonable. Residual graphs of volume and grade suggests a parabolic curve, then it may make sense to regress travel delay on the squared form of these two variables. Notice that the range of changes in truck composition is low inferring that the linear relationship with travel delay may be correct.
68
Figure 3.8 Scatterplots of residuals against explanatory variables
As a result, to improve the model, new variables are introduced. These variables are either the original variables squared (i.e. vol_sq indicates volume squared) or a multiple of two of the variables. Non-linear regression models were fitted to the data accordingly. The R-square of the models were improved slightly but the systematic errors were not eliminated. Search for an appropriate multiple-regression model was repeated for one- and twolane blockage incident cases using a similar procedure as described previously for the shoulder blockage case. The linear models are presented in Equations 4.5 to 4.8. One-lane blockage travel delay linear regression model: LDV:
Eq. 3.5
LDT:
Eq. 3.6
69
Two-lane blockage travel delay linear regression model: LDV:
Eq. 3.7
LDT:
Eq.3.8
Similar to the shoulder lane blockage category, non-linear regression models were calibrated for the one- and two-lane blockage incident categories. However, the travel delay estimation models of one- and two- lane blockage scenarios did not improve statistically compared with linear counterparts. The R-square of these non-linear models are presented in Table 4.16. A hybrid approach mentioned previously was established for improving travel delay regression models as described in the following section. 3.3.3.2 Hybrid Approach In linear regression, the coefficient of a single variable will not change by removing or adding a new independent and uncorrelated variable to the model. In Section 4.2 truck percentage and segment grade were found to be uncorrelated with other explanatory variables. A hybrid approach in which travel delay obtained from simulated incidents is integrated with estimates obtained from developed regression models is created to reduce the error of the estimation models and capture the relationship between travel delay, number of lanes in the segment, incident duration, traffic volume and the speed of vehicles more accurately. In the hybrid approach, the primary linear regression model is broken into two parts: (a) a travel delay function on number of lanes, incident duration, traffic volume and speed and (b) a travel delay function on percentage of trucks in traffic composition and roadway gradient. Assume an incident in which all the factors (the explanatory variables) are nonzero. The first part (a) is identical to the same incident case in which tucks percentage and gradient are zero. Travel delay associated with this incident ( ) was then directly
computed from the simulation runs. The additional travel delay due to percentage of trucks and different gradients then was included in the model using linear regression estimation
70
equations. For example, travel delay regression model for light-duty vehicles and shoulder blockage incidents (Equation 4.3) would be reformed as in Equation 4.9. a
[ ]
b
Eq. 3.9
To validate this hybrid approach, simulation runs for 300 incidents in a three-lane highway were completed. Incident durations and traffic volumes were set following the design described in Section 4.3.2. Truck percentage and gradient were set to random values from uniform distributions. Then, the equation 4.9 was applied to obtained travel delay data of the incidents with zero percentage of truck and grade, . Refer to these values
as “predicted values” for the travel delay of designed incidents. Then, the coefficient of determination of the hybrid model can be computed as follows:
where ?
? = Residual sum of squares, = total sum of squares, = Observed values, =Mean of observed values, = Predicted values by model. The R-square of the linear regression for travel delay for shoulder incidents was 0.672 while the R-square of the hybrid model is 0.939. Thus, we can conclude that this estimation approach better captures travel delay. The same approach was applied to linear models of each of the six categories (3 lane blockage categories for each vehicle class) and the R-square was calculated. A comparison of R-square values between the linear
71
regression models, non-linear regression models, and the hybrid approach is presented in Table 4.16. It can be noted that the R-square of regression models of all categories has improved significantly with the hybrid modeling approach given in Equation 4.9.
Table 3.10 Improved R-square comparison for new model
Lane blockage Shoulder lane blockage 1-lane blockage 2-lane blockage Vehicle class LDV LDT LDV LDT LDV LDT Linear model 0.677 0.568 0.195 0.153 0.142 0.129 Nonlinear model 0.878 0.698 0.244 0.236 0.141 0.075 Hybrid Model 0.939 0.875 0.768 0.719 0.784 0.725
To use this hybrid approach in estimating travel delays in the SSP-BC Tool, simulation runs for all possible cases of number of lanes in a segment, number of lanes blocked due to the incident, incident duration, traffic volume, and speed of vehicles must be made. The travel delay obtained from the runs is then integrated with the regressed travel delay due to truck percentage and segment grade. 3.3.3.3 Fuel Consumption Regression Model The same approach described in the previous section is used to obtain a fuel consumption prediction model for the light-duty vehicles. Later in Chapter 5 it is explained why the fuel consumption of LDT is not required for benefit computation of a SSP program. The linear model is presented in Equation 4.10 in addition to travel delay variables. Here, spd stands for speed of vehicles and lane is the number of lanes in the segment.
Eq. 3.10
The coefficient of determination of these linear models is 0.8210. The R-square for the light-duty vehicle model is very high, indicating excellent model fit to the data. However, it appears from Figure 4.9 that the linearity assumption is violated. This is indicated by a curve pattern in the residual scatterplot. Additionally, the variance of residuals increase indicates heteroscedasiticity assumption does not hold. From Figure 72
4.10, it can be seen that the residuals associated with volume are randomly scattered around zero, but the variance is not constant. This shows that the model is less accurate for some data points. Back in Figure 4.9, since the residual distribution is close to normally distributed and plotted points in the Q-Q chart are almost on the y=x line, it is reasonable to assume that the residuals are normally distributed
Figure 3.9 Summery of fit diagnostic of linear model of LDT fuel consumption
73
Figure 3.10 Scatterplots of residuals against explanatory variables
To address the linearity and variance problem, a set of non-linear models were derived and tested. From Figure 4.10, we can guess that a transform on gradient and speed might improve the model. Fuel consumption changes linearly with number of lanes in the segment (“lane”) as found preciously in Section 4.3.2. Using the stepwise method, some parameters were chosen to enter the model. The final chosen nonlinear model for fuel consumption (Table 4.11) obtained after consideration of a variety of models.
74
Table 3.11 LDV fuel consumption Root MSE 67.16175 R-Square0.8293 Dependent Mean200.47409Adj R-Sq0.8211 Coeff Var 33.50146 Parameter Estimates Parameter Variable DF Estimate Intercept 1 -255.22686 vol 1 0.05440 g 1 18.83305 lane 1 15.77193 spd 1 4.50066 Ptruck 1 -3.56345 ID 1 1.86832 Spd^(2) 1 -0.02294 ID^(1/6) 1 -0.18012 gsq 1 0.08399
Pr > |t| 0.0368 <.0001 <.0001 <.0001 0.0008 <.0001 0.0149 0.0233 0.0928 0.0050
The R-square of this model is not significantly improved by relaxing the linearity assumption. However, other goodness-of-fit tests show significant improvements and variables in the model agree with the similar studies in the same area, which make the model be adopted for the purpose of this study.
3.4
Summary VISSIM was employed to estimate travel delay and fuel consumption of individual
incidents in a segment. Impact of previously identified factors on travel delay and fuel consumption studied and multiple-regression models for estimating travel delay and fuel consumption were presented. A hybrid approach was introduced for improving the obtained travel delay models. Note that once fuel consumption is obtained, emissions can be computed (Section 3.4).
75
4
CHAPTER 5. B/C Ratio Estimation
This chapter discusses the computation of the B/C ratio, which is designed to provide insight into the return on investment received from operating a SSP program. Evaluation of the benefits in the B/C ratio requires a method for the amalgamation of chosen MOEs. In the developed tool, these are the savings in travel delay, fuel consumption, emissions and secondary incidents. These MOEs are given in a variety of units of measurement. Savings in travel delay is in vehicle-hours, savings in fuel consumption is in gallons of fuel, savings in emissions is in metric tons, and secondary incident savings is in number of prevented incidents. Thus, conversion to a common unit of measurement is required to develop a single numeric value for the numerator of the B/C ratio. Moreover, the unit of measurement must be commensurate with the units used in the B/C ratio’s denominator, namely cost. Consequently, the most common approach is to convert the individual benefit measures to their monetary equivalents using monetary conversion factors. Methodologies for computing the total program savings associated with each of the chosen MOEs are provided in Section 5.1. Computation of the total benefit, i.e. the numerator of the B/C ratio, is discussed in Section 5.2. Section 5.3 describes total cost calculation in the deamination. This is followed by B/C ratio calculation in Section 5.4. 4.1 Savings Computation To compute the benefit of a SSP program during a time period, benefits derived from each individual incident due to response by an SSP vehicle must first be determined. This is because the duration of the incident decreases as a result of the SSP vehicle response as explained in Section 3.2.4.2. To assess the value of the reduction in incident duration, travel delay, fuel consumption, and emissions can be estimated for the incident with reduced duration as a result of the SSP response “with-and-without” approach was employed (Section 2.2). Typically, it is the case that no such pre-program measurements were made. Thus, estimation of the “without” case must be made by assuming an increase in the duration of each realized incident. The amount of increase should be commensurate with the program’s incident response time.
76
4.1.1 Savings in Travel Delay To estimate travel delay of actual incidents in an area, using the travel delay hybrid model proposed in Section 4.3, simulation results from all studied incident scenarios are required. Simulation runs were made and travel delays associated with 14,784 incident scenarios were collected. The runs involved all possible combinations of 16 categories of incident duration, 11 categories of traffic volume and 6 speed categories, resulting in 1,056 combinations. For each combination, runs are including one of 3 types of lane blockage and one of 5 possible roadway sizes in terms of number of lanes. Results were saved in a table contained within the data base that supports the tool for further computations. To estimate the travel delay associated with an incident with known characteristics using the proposed simulation-based procedure, the incident characteristics and associated traffic volume and speed must be known. As discussed in Chapter 3, the impact of the ramp density (Section 3.2.1.4), horizontal curvature (Section 3.2.2.1), and weather conditions (Section 3.2.5) on the capacity of the segment is captured through a reduction in FFS, affecting the speed category of the incidents. The maximum speed reduction due to existence of ramps, curvature, and adverse weather conditions determines the speed category of the incident. For example, consider an incident case for which speed in clear weather is 65 mph. To include the impact of heavy rain (a reduction in speed by 10percent) and a full cloverleaf interchange in a one-mile segment (two on-ramps and two off-ramps in each direction, 5 mph reduction for each 2 ramps/mile),the employed speed for the incident cases in that segment would be 55 mph;
{ }
When the final speed (after all reductions are taken) to be associated with an incident is determined, the estimation of travel delay for each incident can be completed with the use of the proposed regression models (Equations 4.3 to 4.8). The savings are computed from the difference between travel delays for the “without” and “with” incident cases. An overview of this procedure is provided in Figure 5.1.
77
Figure 4.1 Travel delay estimation procedure
For incidents for which particular incident duration, traffic volume and speed is not one of the categories in above data set, travel delay is obtained by assuming linear changes in between upper and lower bound categories. If category of incident duration i, denotes travel delay of incident duration i and and stand for travel delay of lower
and upper bound of the incident duration i category, respectively, Equation 5.1(a) can be used to obtain travel delay of desired incident duration.
Eq. 4.1(a)
Traffic volume is assumed to be rounded to nearest volume category. A similar linear estimation approach to incident duration was used to interpolate when given speeds outside the tested categories. Likewise, a similar equation to Equation 5.1(a) but for speed can be used as follows.
78
Eq. 4.2(b)
Where in incident i,
denotes speed of vehicles prior to incident i and
and
stand for the speed of lower and upper bound of the incident i, respectively, if neither incident duration nor speed associated with an incident were in the provided data set, by linear assumptions on speed and using Equation 5.1(b) for SPu and SPl, the travel delay associated with upper and lower bound on of incident duration, IDu and IDl, are computed and then Equation 5.1 is applied to compute the final travel delay. This process is shown in Figure 5.2.
Figure 4.2 Subcategory linear interpolation Incident Duration, Speed SPu IDu SPl SPu IDl SPl
4.1.2 Saving in Fuel Consumption and Emissions A similar with/without incident approach is taken to estimate fuel consumption savings associated with an incident scenario. Fuel consumption corresponding to each incident scenario is obtained directly from equations described in Tables 4.15 to 4.22. Emissions are calculated directly from equations 3.6-3.8 based on the fuel consumption estimates. 4.1.3 Saving in Secondary Incidents The probability of occurrence of a secondary incident grows with an increase in the primary incident duration (Section 3.5). To estimate the number of prevented secondary incidents, the number of secondary incidents when SSP is not operating is assumed to be linearly correlated with the travel delay ratio of without and with incidents to which SSP responded in a period of time. This approach to estimating secondary incident savings is discussed in (Chou and Miller-Hooks, 2008). It supposes that total travel delay is a
79
reasonable surrogate for impact area size of primary incidents in which incidents classifies as secondary. This relationship is shown in Equation 5.2. , where : Number of secondary incidents for extended incident duration case (without case), : Number of secondary incidents in base case (with case), : Travel delay for the extended case, : Travel delay for the base case. As discussed in Chapter 3, for this analysis, the number of secondary incidents ( as a fraction of primary incidents must be known regardless of the chosen secondary and are estimated as explained in Section 5.1.1.
Eq. 4.2
incident classification method.
Another method to calculate the benefits of SSP program in terms of prevented secondary incident is to consider the incident duration reduction contribution to likelihood of secondary incident occurrence as explained in Section 3.5. As mentioned in Section 3.5 Karlaftis et al. (1998) estimated the clearance time coefficient for winter and all other seasons as 0.017 and 0.031, respectively. Assuming that SSP vehicles reduce the incident duration by 20 minutes, the increase in the likelihood of a secondary incident occurrence would be 14.05% in winters and 18.59% in all other seasons.
The average increase in likelihood of occurrence of a secondary incident is %17.46. Using this method, the potential secondary incidents reduced due to SSP program can be computed as shown in Equation 5.3. ( where ) ,
Eq.5.3
80
N: Number of reduced potential secondary incidents, ID: Incident duration reduction due to SSP program operation in minutes, Np: Total number of incidents to which SSP vehicles responded.
4.2
Total Benefit Calculation
4.2.1 Monetary Values To isolate a single unit for evaluation of a SSP program, congestion related travel delay (vehicle-hours), fuel consumption (gallons), and number of secondary incidents prevented are converted into their monetary equivalents. Monetary equivalents in the SSPBC Tool proposed herein were provided by the American Transportation Research Institute (ATRI). Table 5.1 contains a list of the monetary equivalent variables, the variable’s corresponding output, a description for each variable and data source. Four individual tables containing this information support the B/C ratio computation within the tool. They are also designed to be updatable.
Table 4.1 Summary of monetary equivalents (ATRI)
Variable Corresponding Output Description Used to monetize the wasted fuel that would result from increased congestion if SSP did not exist NOTE: fuel is already factored into the Hourly Truck Cost, and the monetization of wasted fuel should only be performed on the passenger vehicle share Used to monetize lost productivity of passenger vehicles resulting from increased congestion if SSP did not exist Source
Average gasoline prices (Table B.1)
Gallons of fuel saved
U.S. Energy Information Administration, Gasoline and Diesel Fuel Update; updated 5:00 p.m. every Monday; http://www.eia.doe.gov/oog/info/gdu/gasdi esel.asp U.S. Department of Labor, Bureau of Labor Statistics; State Occupational Employment and Wage Estimates; http://www.bls.gov/oes/current/oessrcst.ht m An Analysis of the Operational Costs of Trucking: A 2011 Update; ATRI; http://www.atrionline.org/research/results/Op_Costs_2011 _Update_one_page_summary.pdf Based on actual operational cost data collected from motor carriers across the country, representing a cross-Section of industry sectors. The Economic Impact of Motor Vehicle Crashes: 2000. NHTSA.
Average labor costs (Table B.2, B.3)
Hours of delay prevented
Commercial vehicle costs per hour
Hours of delay prevented; Gallons of fuel saved
Used to monetize lost productivity of commercial vehicles resulting from increased congestion if SSP did not exist.
Secondary incident cost
Number of secondary incidents averted
Represents only the cost of property damage. Used to monetize the cost of additional secondary incidents that would result from increased congestion if SSP did not exist.
81
While some previously developed B/C ratio estimates made for SSP programs have included monetized emissions equivalents in the savings computation, a review of the literature indicates that the available monetary equivalents are based largely on soft, intangible costs, as opposed to other more tangible costs (e.g. price of a gallon of fuel). Thus, tons of emissions saved are reported separately and are not included in the B/C ratio computed in the SSP-BC Tool. Average hourly wages are used herein to convert savings in travel delay to a monetary equivalent. Wage values are available at metropolitan-levels and as a state average (Table B.3). Additionally, data containing the share of commercial VMT compared to total VMT by state were used for truck composition estimate for each state (Table B.4). This data is necessary to distinguish between the benefits derived from savings in travel delay due to passenger vehicles and commercial vehicles. Average operational cost of trucking for 2011 is obtained to be $59.61. The B/C ratio is highly sensitive to the cost of secondary incidents. In this study, cost represents “property only damage” incidents and for 2011 it is assumed to be $4,736. Other costs associated with higher severity incidents and congestion due to secondary incidents were not considered. 4.2.2 Computing Total Benefit To compute the total savings in travel delay, fuel consumption, emissions, and secondary incidents resulting from a SSP program in a segment over a period of time, information pertaining to the incidents arising along the studied roadway segment during the study period is needed. Specifically, the distribution of incidents with respect to lane blockage must be known (or approximated). Assuming any two incidents are independent, TSj , the total savings of type j, where j={total travel delay, fuel consumption, emission} for every incident i arising during a period of time over a road segment as described in Equation 5.3 can be computed. When using this method, it is necessary to assume that the impact of an individual incident has no influence on other incidents on the road. Furthermore, as described in Section 5.1.1, the speed of the incident scenario and so the savings in travel delay and fuel consumption is related to geometry characteristics of
82
the study segment and weather condition in the time of the incident. The geometry characteristics are similar for all incidents in a study segment. However, the weather condition might vary incident by incident in a period of time. One incident under each weather conditions (Section 3.2.5) would have different final speed. Therefore, having the probability of each weather type, Pk, saving of one incident can be estimated as exhibited in equation 5.4. ? ? where TSj = Total saving j, j = =Type of saving {Total travel delay, fuel consumption, emission pollutants}, i = Individual incidents, k= Weather conditions {Clear, light rain, heavy rain, low visibility, snow, fog, icy = Saving type j in incident i of weather condition k. Given monetary conversion rates for travel delay, fuel consumption, and secondary incidents, total program benefit can be computed. With these concepts, and assuming that benefits are uniformly distributed over length, the total benefit of the SSP program over the study period and roadway segment can be computed as in Equation 5.4.
?
Eq. 5.5
,
Eq. 5.6
where B= Total benefit of a SSP program, j = Total travel delay (1), Fuel consumption (2), Secondary incidents (3), TSj = Total savings of type j, MEj= Monetary equivalent of saving j, L= the length of the study segment.
83
4.3
Cost Calculation The total cost of a SSP program, TC, is a function of the number of roving SSP
trucks along the study segment, hourly operating cost per truck, number of working hours, number of workdays in a year, fuel cost of each vehicle, cost of giveaway fuel to the vehicles that ran out of gas, and other costs such as vehicle maintenance cost as expressed by Equation 5.4. Moreover, for some SSP programs, most often the total annual cost of the whole SSP program is available and not the cost associated with the study segment. The total annual cost can be computed from Equation 5.4.
Eq. 4.3
where TC : Total annual cost for operating the SSP program in dollars, c : Cost per truck-hour {hourly wage of driver, fuel cost of the vehicle}, n : Number of roving trucks, hr : Number of working hours in each day, day : Number of workdays in a year, fuel: annual giveaway fuel cost. The cost of many SSP programs can often be easily calculated, as many SSP programs are outsourced and the charges are provided contractually. The cost of the program by roadway segment may be less clear. Two general methodologies were considered herein for the computation of segment-based costs. First, given total program costs, costs associated with a given segment can be computed based on the proportion: number of the total incidents to which the SSP vehicles responded to those to which they responded arising only within the study segment. This computation is captured in Equation 5.8. ? where = Cost of operating the SSP program along study segment n, = Total annual cost of the SSP program,
, Eq. 4.4
84
= Number of incidents along the study segment n to which the SSP program responded, = Total number of incidents to which SSP program responded. The second methodology is to compute cost associated with a given segment by the proportion of length of it to the total length of covered roads SSP vehicles covers. In this method, it is assumed that cost is uniformly distributed over length of the roads of the SSP service area. The first method is used in SSP-BC Tool. Some SSP programs may operate a heterogeneous fleet of vehicles. Thus, those vehicles that are similar in response capability or with identical operational hours can be classified as falling within the same group. Total annual costs can be computed from costs computed for each category of vehicles. 4.4 The B/C ratio The obtained benefit from Equation 5.5 and cost from Equation 5.7 are used to assess the segment-based B/C ratio for a given SSP program over the study period. The SSP-BC Tool provides multi-segment analysis. The B/C ratio of n segments is computed from the ratio of the sum of benefits to sum of costs for all segments as in Equation 5.6. ? where ? = B/C ratio of multiple segments, = Obtained benefit of segment n, = Obtained cost of segment n. Recall that within the SSP-BC Tool, savings in emission pollutants are not translated to dollars and, thus, cannot be included in the B/C ratio. Emissions are given separately in the form of metric tons.
? ?
,
Eq. 4.5
85
4.4.1 Additional Benefits Additional savings that has not been quantified in this study are: improved safety not only in preventing secondary incidents, but in the improved feeling of security on the transportation system, congestion cost associated with the secondary incidents, improved freight transit system, environmental benefits, and benefits to other agencies like additional time available for troopers for more urgent tasks that the SSP programs cannot handle. A list of additional costs associated with incidents is: Administrative costs: the cost (monetary and temporal) associated with investigating and documenting the primary, and any secondary, incidents. In the case of fatal incidents, costs increase exponentially. In addition, there are generally administrative costs associated with insurance claims. Legal costs: Includes attorney fees and court costs associated with litigation resulting from primary and secondary incidents. Rehabilitation costs: The cost of career retraining required as a result of disability caused by roadway incident. An additional cost in this category is, replacement employee costs. That is, employers often hire temporary help or compensate other staff by paying overtime to cover the position of an injured employee. Disability/Retirement income: Should the employee suffer career-ending injury, the employer will have to make payments to fund the employee’s disability pension. Productivity reduction: this is the cost associated with lost wages and benefits over the victim’s remaining lifespan. Numerous additional sources of benefits in cost reduction have not been included in the computation of program benefits within the proposed SSP-BC tool. The exclusion of the many additional benefits from the benefit estimate used in the B/C ratio results in conservative B/C estimates.
86
5
CHAPTER 6. The Tool by Illustrative Example
The SSP-BC Tool interface was coded in Microsoft Visual Basic 2003. Data developed based on Chapter 4 is in microsoft Access (2010) format. Tables of monetary values for travel delay and fuel consumtion, and share of trucks in traffic volume were designed to be updated by the user. The SSP-BC Tool is explained in Section 6.1. 693 incidents to which the SSP program in New York (H.E.L.P) responded over a 6-month period in 2006 is used as a case study for the tool and its outputs. A comparison between previously obtained B/C ratio by Chou and Miller-Hooks (2008) and use of the propesed generic SSP-BC Tool is made in the Section 6.2. 5.1 The SSP-BC Tool The I-287 segment studied herein is approximately 10 miles in length, beginning at the junction with I-95 and continuing west to the Tappan Zee Bridge in New York. This segment is referred to as Beat 8-2 of the H.E.L.P. program. Incidents arising on this roadwaysegment will be handled by a H.E.L.P. vehicle driver, a trooper from Unit T or both. During the study period, 1,303 incidents arose along the study segment of I-287. 693 of these 1,303 incidents received service from the H.E.L.P. program during the H.E.L.P. hours of operation. Figure 6.1 shows the main window of the SSP-BC tool including the information on SSP program level and number of segments. Total annual program cost is required here. In addition, a detailed cost list is provided in the “Cost” window (Figure 6.2) in which annual SSP program cost will be calculated automatically. The figure shows the cost information of the H.E.L.P program. Chou and Miller-Hooks (2008) used costs of $40 and $50 per truck-hour, two roving trucks operated within the study roadway segment with an eight-hour workday, 126 workdays within the 6 month study period (21 days/month). The annual cost using $40/truck-hour was $161,280. Up to 5 different categories of cost groups as explained in Section 5.3 are availble for different type of service vehicles or operational hours that might exist within a SSP program.
87
The next input is for finding the associated cost to the segment, total number of incidents arose in the segment in one year. If the annual cost was the estimated cost of total SSP program, not the program cost associated with the study segment like this example, the number of incidents to which the SSP program responded in a year has to be used as input here (Section 5.3). Segments in a study area must be homogenouse in terms of geometry, weather and traffic volumes as explained in Chapter 3. The SSP-BC Tool is capable of analysing up to 30 segments at a time. This example has only 1 segment.
Figure 5.1 Main window
88
Figure 5.2 Program cost detail
For each segment information pertaining to hours of operation weather, traffic conditions, SSP program average response time, roadway geometry, and incident distribution and duration must be entered. It is assumed that SSP programs operats within a single state. Regional data at the metropolitan level are applied in setting the monetary conversion rate. Average monetary values for the state are used when the region is set to “others”.
Figure 5.3 Basic data in segment level
89
The program information window (Figure 6.4) contains details on operational hours and perfomance of the SSP program. Hour of operation are divided to four time categories. The user must to be consistant with her/his definition of each time category for analysis of each segment. For example, if she/he selects 7 to 10 AM as her/his AM peak hour, traffic and incident information for these hours must be used for the AM peak in following steps. A key impact to the SSP-BC Tool is the average incident duration reduction offered by the program. Different approches for estimating this time reduction are discussed in detail in Chapter 2. In addition, the FSPE tool (Describe in Section 2.3) computes the arrival time of the SSP vehicles and ,thus, incidnet response time. The savings in incident duration greatly depend on the severity of the incident or, as employed herein, on the number of lanes blocked as a result of incident occurence. The SSP-BC Tool assigns incident duration savings in each lane blockage category to the duration of the incidents in that category. The average value, as in the example, can also be used where sufficent data is not availble. Chou and Miller-Hooks (2008) founded an average savings of approximately 20 minutes in incident duration for incidents involving a collision and 19 minutes for incidents involving a disabled vehicle for the study area as a result of the presence of the H.E.L.P. program. The average incident duration savings of 20 minutes is used for the example solved here.
90
Figure 5.4 Program information window
The SSP-BC Tool provides default values for incident duration and related savings to the due to SSP program based on previous studies in the area. The default values associated with roadway geometry and traffic information (Figure 6.5) are based on the numerical experiments used as described in Chapters 3 and 4 and summarized in Table 6.1.
Table 5.1 Geometry and traffic default values
Input Segment length Number of traffic lanes by direction General terrain Horizontal curvature Number of ramps in segment Posted main-lane speed limit Percentages of trucks Weather Default value 10 3 Level straight 0 70 3 clear
The H.E.L.P study segment length is 10 miles and number of lanes is 4. Default values of general terrain and road curvature were used. Total ramp density for the H.E.L.P study segment is 1.4 ramps/mile with14 on/off ramps within the segment.
91
Traffic volume data for the study roadway segment was employed for the same period, but in the following year. Average weekday and weekend traffic volumes by month were matched to the incidents by their date and time information. For each of the operational hour classes of the SSP program, one traffic volume was assumed in this example. Incidents for which prevailing traffic volume was between 0 and 600 vplph were categorized for the weekend, 600 to 1000 vplph for weekday off peak, 1000 to 1400 vplph as weekday PM peak, and 1400 to 2200 as weekday AM peak. The average for the prevailing incident traffic volumes in each category was set within a time class. For example for weekday PM peak, the average of 1200 vplph was used. The percentage of trucks set to 7.8% (Table B.4) for all operational hour classes. Weather was assumed clear for all incidents. Note that this classification of volume was done to fit the available data to the SSP-BC Tool. The operational hour of the H.E.L.P. program is weekday peak periods.
92
Figure 5.5 Roadway geometry and traffic information
Incident information scenario is entered next. Average incident durations and number of incidents are required by lane blockage as shown in Figure 6.6. They are assumed to be identical for all operational hour classes for this example. Savings in prevented secondary incidents in the tool is calculated using the first method discussed in Section 5.1.3. The input is the percentage of secondary incidents out of primary incidents for the study segment.
93
Figure 5.6 Incident information window
Since the example has only one segment, the B/C ratio can be ontained. The output window is shown in Figure 6.7. Users can choose one or more segments for the B/C ratio analysis. The outputs of the tool as described in MOEs, Section 3.1, are savings in travel delay in vehicle-hours, fuel of passenger cars and light-duty in gallons, number of prevented secondary incidents and emission pollutants in metric tones. The users specifies which MOEs (of travel delay, fuel consumption and secondary incidnes) to include in the
94
B/C ratio.The total benefit of the chosen segment is then calculated as described in Section 5.2. For this example, 2006 monetary values are employed to compute the benefits for consistancy with the availble cost information of the program (Table 6.2).
Table 5.2 New York 2006 monetary values
State New York Travel delay $/hr 15 Fuel (gas) $/gallon) 2 Carbon Monoxide (CO) $/ton 6,360 Hydrocarbons (HC) $/ton 6,700 Nitrogen Oxide(NO) $/ton 12,875
Figure 5.7 Output window
95
The B/C ratio of the H.E.L.P program was estimated to be 2.83 plus additional benefits derived from emisions savings. Adding emissions to the benefits increased the B/C ratio to 3.08. Chou and Miller-Hooks (2008) estimated the B/C ratio of the H.E.L.P program 2.68. In both B/C ratio estimates that included emissions, only benefits from savings in CO, HC, and NO were included. Monetary conversion rates used are given in Table 6.2. They used technique in which they replicated incidents in CORSIM and computed the travel delay using the same with/without approach (Section 5.1). They used conversion factors for fuel consumption and emissions from travel delay. In simulating incidents they did not includethe geometry characteristics of the segment, such as ramps. Table 6.3 contains obtained savings Chou and Miller-Hooks found for 20 minutes incident duration savings due to the operation of H.E.L.P. program.
Table 5.3 H.E.L.P result comparison
Type of saving Travel delay (veh-hr) Fuel consumption (gal) Secondary incidents (#) CO (ton) HC (ton) NO (ton) Chou et al. (2008) 12,182 1,451 9 1.79 0.16 0.08 SSP-BC 10,097 12,856 17 8.34 1.1 0.55
Travel delay estimated by the SSP-BC Tool is slightly less than what Chou and Miller-Hooks obtained. On the other hand, Chou and Miller-Hooks estimated fewere saved secondary. A comparison between VISSIM and CORSIM was completed to better understand difference in travel delay estimates. The study segment was simulated in both software products. Travel delay was gathered for different traffic volumes as plotted in Figure 6.8. It was found that CORSIM estimates higher travel delays compare to VISSIM when the simulated segment reaches its capacity. It seems that in CORSIM, the capacity of the roadway segment with defult parameters applied in Chou and Miller-Hooks, worked is lower compare to VISSIM using described calibrated parameters. Relate discussion can be
96
found in Section 4.1.2. Where earlier studies note that VISSIM provides a better model of traffic than CORSIM.
Figure 5.8 CORSIM vs. VISSIM
250 200 150 100 50 0 0 500 1000 1500 2000 2500 Volume (vphpl)
VISSIM CORSIM
. The B/C ratio of the H.E.L.P program with 2011 monetary values (Appendix B),as current SSP-BC Tool monetary equivalent data, and asuming the truck-hour cost of $60 compare to $40 in 2006, total cost of the program would be $120,960, is 2.23.
Travel delay (veh-hr)
97
6
CHAPTER 7. Conclusions and Limitations
The SSP-BC Tool was developed to fill the need for a standardized B/C ratio estimation methodology with wide applicability and substantiated and needed updatable monetary conversion rates. The tool was designed to support B/C ratio estimation for roadways with existing programs, but can also be used to test numerous what-if scenarios, including the introduction of a new program or the impact of improvements in service response times. A quicker and less data-intensive approach was developed so that it can be readily and widely utilized by all states around the US. The SSP-BC Tool accounts for a wide array of traffic, environmental and program characteristics that influence benefit and cost estimates. The Factors, such as incident duration, traffic volume and composition, ramp density, horizontal and vertical alignments, and weather conditions that have been identified as important to travel delay fuel consumption and emissions estimation were included in this tool. Moreover, the per-second vehicle velocity and acceleration values were employed in the computation of fuel consumption and emissions. Numerous experimental runs were completed and seven multiple-regression models for estimating travel delay and fuel consumption were developed. For experimental runs, the techniques to simulate different geometry, traffic, and weather characteristics were suggested and tested. Additional delay caused by two regression parameters, truck percentage and segment gradient, can be applied to any travel delay estimate in which traffic composition or gradient is not included. 6.1 Limitations The SSP-BC Tool extensions are limited by a maximum incident duration of ninety minutes, that only up to two lanes can be blocked due to an incident and no consideration for roadway closure, and that any given segment has a maximum of six lanes. The SSP-BC Tool can be extended to include of longer duration or greater severity, and roadway with more than six lanes. To complete a nationwide study, it would be ideal to have all input data associated with the entire national roadway system. As this is highly impractical, a statistical approach
98
might be used in which a random sample of the required traffic and environmental data is taken. With this sample, a general estimation model can be created and calibrated based on statistical approaches. Unfortunately, this method is not an easy and cost effective method. In addition, obtaining one general estimation model for all states would not be reliable considering the fact that statistical models developed for the entirety of the nation will likely poorly fit the data of specific regions. 6.2 Contributions In addition to the development of a nationwide tool for SSP program B/C ratio estimation, the contributions of this work include the identification of the significant factors affecting SSP program benefits, techniques for simulating these factors and development of regression models to estimate travel delay, fuel consumption and emissions given traffic, roadway geometry, program characteristics and weather conditions. For travel delay estimation, the developed enhanced regression methodology takes a hybrid approach to multiple-regression model construction. This approach combines parameters obtained through regression analysis for truck percentage and roadway grade, which were found to be independent of all other factors, with results from simulation runs. The simulation results and estimates from the regression models are developed into tables employed within the tool’s database. For fuel consumption estimation, a multiple regression model is developed that is used directly within the tool. These developed regression models can be used directly where applicable.
99
7
APPENDIX A: Fuel Consumption Computation Tables
Table A.1 Calculation of Road-load coefficients
Vehicle Category LDV (passenger cars) LDT (trucks, SUVs, etc) LHD<=14K LHD<=19.5K Source Mass (metric tons) 1.4788 1.86686 7.64159 6.25047 A (KW/mps) 0.156461 0.22112 0.561933 0.498699 B (KW/mps2) 0.00200193 0.00283757 0 0 C (MW/mps3) 0.000492646 0.000698282 0.00160302 0.00147383
Table A.2 Emission factors
Fuel Economy (mile/gal)1 22.1 17.6 Engine Displacement Volume, V (L)3 1.3-3.1 2.5-5.3 Emission Factor (EF in g/mi) for gasoline2 HC 2.8 3.51 CO 20.9 27.7 NOx 1.39 1.81 CO2 451 637
Vehicle LDV LDT
Table A.3 Fuel properties
HV (KJ/g)a ?b
(g/gal)
Fuel
Base Fuel
SCFuel (ppm) 80 500
Gasoline Diesel
(Base fuel) (Base fuel)
43.448 42.791
2834.95 3210.98
Table A7.4 Transmission parameters for engine speed calculation (Source: PERE, 2005; EPA420-P05-001)
LDV & LDT S = 35.6 Speed (mph) 0-18 18-25 25-40 40-50 50+ Gear 1 2 3 4 5 g/gtop 4.04 2.22 1.44 1 0.9
100
8
APPENDIX B: Monetary Equivalents
Table B.1 Average gasoline prices
Area name Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri abb AL AK AZ AR CA CO CT DE DC FL GA HI ID IL IN IA KS KY LA ME MD MA MI MN MS MO Fuel price($) 3.163538462 3.514961538 3.514961538 3.163538462 3.586 3.163538462 3.379346154 3.339615385 3.339615385 3.267 3.242576923 3.514961538 3.198192308 3.298903846 3.298903846 3.298903846 3.298903846 3.298903846 3.163538462 3.379346154 3.339615385 3.307923077 3.298903846 3.303057692 3.163538462 3.298903846 Area name Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming abb MT NE NV NH NJ NM NY NC ND OH OK OR PA RI SC SD TN TX UT VT VA WA WV WI WY Fuel price ($) 3.198192 3.298904 3.514962 3.379346 3.339615 3.163538 3.500981 3.242577 3.298904 3.282615 3.298904 3.514962 3.339615 3.379346 3.242577 3.298904 3.298904 3.167385 3.198192 3.379346 3.242577 3.492635 3.242577 3.298904 3.198192
101
Table B.2 Average labor cost by state
Area name Alabama Alaska Arizona Arkansas Average wage ($/hr) 21.5 24.21 20.38 17.05 Area name Montana Nebraska Nevada New Hampshire Average wage ($/hr) 17.34 18.42 19.82 21.37
California Colorado Connecticut Delaware District of Columbia Florida Georgia Guam Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts
24.39 22.48 24.96 22.53 35.31 19.36 20.32 15.02 21.03 18.56 22.33 18.76 18.14 18.89 18.25 18.26 18.98 24.46 25.82
New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Puerto Rico Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virgin Islands Virginia
24.39 19.26 24.86 19.47 17.81 19.66 17.76 20.94 20.7 12.92 22.08 18.23 16.53 18.43 20.3 19.29 20.21 17.85 23
102
Michigan Minnesota Mississippi Missouri
20.81 21.86 16.31 19.13
Washington West Virginia Wisconsin Wyoming
23.53 17.01 19.7 19.96
Table B.3 Average wage by area
Area Anniston-Oxford AL Auburn-Opelika AL Birmingham-Hoover AL Columbus GA-AL Decatur AL Dothan AL Florence-Muscle Shoals AL Gadsden AL Huntsville AL Mobile AL Montgomery AL Tuscaloosa AL Northwest Alabama nonmetropolitan area Northeast Alabama nonmetropolitan area Southwest Alabama nonmetropolitan area Southeast Alabama nonmetropolitan area Anchorage AK Fairbanks AK Southeast Alaska nonmetropolitan area Railbelt / Southwest Alaska nonmetropolitan area Flagstaff AZ Lake Havasu City - Kingman AZ Phoenix-Mesa-Scottsdale AZ Prescott AZ Tucson AZ Yuma AZ Wage ($/hr) 16.92 17.92 19.82 17.83 17.72 16.59 16.45 15.93 23.12 18.39 18.43 18.26 15.62 15.62 16.21 16.73 24.75 24.21 22.33 23.69 18.89 16.97 20.89 18.04 20.27 16.4 Area Duluth MN-WI Fargo ND-MN Grand Forks ND-MN La Crosse WI-MN Mankato-North Mankato MN Minneapolis-St. Paul-Bloomington MNWI Rochester MN St. Cloud MN Northwest Minnesota nonmetropolitan area Northeast Minnesota nonmetropolitan area Southwest Minnesota nonmetropolitan area Southeast Minnesota nonmetropolitan area Gulfport-Biloxi MS Hattiesburg MS Jackson MS Pascagoula MS Northeast Mississippi nonmetropolitan area Northwest Mississippi nonmetropolitan area Southeast Mississippi nonmetropolitan area Southwest Mississippi nonmetropolitan area Columbia MO Jefferson City MO Joplin MO Springfield MO Central Missouri nonmetropolitan area North Missouri nonmetropolitan area Wage ($/hr) 18.81 18.14 17.81 19.09 18.06 23.63 23.43 18.62 17.28 17.2 16.84 17.88 17.23 15.87 17.7 18.58 15.78 14.8 15.09 15.97 17.62 17.91 16.23 17.02 15.51 15.07
103
North Arizona nonmetropolitan area Southeast Arizona nonmetropolitan area Fayetteville-Springdale-Rogers AR-MO Fort Smith AR-OK Hot Springs AR Jonesboro AR Little Rock-North Little Rock-Conway AR Memphis TN-MS-AR Pine Bluff AR Texarkana-Texarkana TX-AR Central Arkansas nonmetropolitan area East Arkansas nonmetropolitan area South Arkansas nonmetropolitan area West Arkansas nonmetropolitan area Bakersfield CA Chico CA El Centro CA Fresno CA Hanford-Corcoran CA Los Angeles-Long Beach-Glendale CA Metropolitan Division Los Angeles-Long Beach-Santa Ana CA Madera CA Merced CA Modesto CA Napa CA Oakland-Fremont-Hayward CA Metropolitan Division Oxnard-Thousand Oaks-Ventura CA Redding CA Riverside-San Bernardino-Ontario CA Sacramento--Arden-Arcade--Roseville CA Salinas CA San Diego-Carlsbad-San Marcos CA San Francisco-Oakland-Fremont CA San Francisco-San Mateo-Redwood City CA Metropolitan Division San Jose-Sunnyvale-Santa Clara CA San Luis Obispo-Paso Robles CA Santa Ana-Anaheim-Irvine CA Metropolitan Division
16.99 19.48 18.7 16.09 16.55 16 18.75 19.32 16.76 17.5 15.21 14.92 15.17 14.33 21.4 19.54 19.01 19.76 20.71 24.16 24.1 20.78 18.79 19.96 23.86 27.09 23.03 19.88 20.64 24.08 20.61 24.14 28.76 30.43 32.62 21.29 23.93
Southeast Missouri nonmetropolitan area Southwest Missouri nonmetropolitan area Billings MT Great Falls MT Missoula MT Eastern Montana nonmetropolitan area Central Montana nonmetropolitan area Southwestern Montana nonmetropolitan area Western Montana nonmetropolitan area Lincoln NE Western Nebraska nonmetropolitan area Central Nebraska nonmetropolitan area Northeastern Nebraska nonmetropolitan area Southeastern Nebraska nonmetropolitan area Carson City NV Las Vegas-Paradise NV Reno-Sparks NV Western Central Nevada nonmetropolitan area Other Nevada nonmetropolitan area Manchester NH Northern New Hampshire nonmetropolitan area Other New Hampshire nonmetropolitan area Western New Hampshire nonmetropolitan area Southwestern New Hampshire nonmetropolitan area Allentown-Bethlehem-Easton PA-NJ Atlantic City-Hammonton NJ Camden NJ Metropolitan Division Edison-New Brunswick NJ Metropolitan Division Newark-Union NJ-PA Metropolitan Division New York-White Plains-Wayne NY-NJ Metropolitan Division Ocean City NJ Trenton-Ewing NJ Vineland-Millville-Bridgeton NJ Albuquerque NM Farmington NM Las Cruces NM Santa Fe NM
14.67 14.67 18.06 16.62 17.5 16.43 16.36 17.79 16.66 18.83 15.31 15.96 15.86 16.01 21.85 19.59 20.52 18.41 20.7 22.49 16.73 19.85 21.76 20.11 20.38 20.02 22.26 24.57 25.74 27.49 18.64 26.93 20.36 19.96 18.58 18.45 20.26
104
Santa Barbara-Santa Maria-Goleta CA Santa Cruz-Watsonville CA Santa Rosa-Petaluma CA Stockton CA Vallejo-Fairfield CA Visalia-Porterville CA Yuba City CA Mother Lode Region of California nonmetropolitan area Eastern Sierra Region of California nonmetropolitan area North Coast Region of California nonmetropolitan area North Valley Region of California nonmetropolitan area Northern Mountains Region of California nonmetropolitan area Boulder CO Colorado Springs CO Denver-Aurora CO Fort Collins-Loveland CO Grand Junction CO Greeley CO Pueblo CO East and South Colorado nonmetropolitan area West Colorado nonmetropolitan area Northcentral Colorado nonmetropolitan area Central Colorado nonmetropolitan area Bridgeport-Stamford-Norwalk CT Danbury CT Hartford-West Hartford-East Hartford CT New Haven CT Norwich-New London CT-RI Springfield MA-CT Waterbury CT Worcester MA-CT Northwestern Connecticut nonmetropolitan area Eastern Connecticut nonmetropolitan area Dover DE Wilmington DE-MD-NJ Metropolitan Division Sussex County Delaware nonmetropolitan area Washington-Arlington-Alexandria DC-VA-MDWV Metropolitan Division
22.71 22.49 23.3 20.6 22.42 18.45 19.8 21.2 19.07 19.21 18.27 20.93 25.65 21.46 23.77 21.2 19.02 19.68 18.02 16.33 19.51 20.49 17.87 28.03 23.27 25.13 24.37 21.31 21.4 22.07 23.11 21.4 19.91 18.81 24.12 17.34 29.95
North and West Central New Mexico nonmetropolitan area Eastern New Mexico nonmetropolitan area Southwestern New Mexico nonmetropolitan area Los Alamos County New Mexico nonmetropolitan area Albany-Schenectady-Troy NY Binghamton NY Buffalo-Niagara Falls NY Elmira NY Glens Falls NY Ithaca NY Kingston NY Nassau-Suffolk NY Metropolitan Division New York-Northern New Jersey-Long Island NY-NJ-PA Poughkeepsie-Newburgh-Middletown NY Rochester NY Syracuse NY Utica-Rome NY Capital/Northern New York nonmetropolitan area East Central New York nonmetropolitan area Central New York nonmetropolitan area Southwest New York nonmetropolitan area Asheville NC Burlington NC Charlotte-Gastonia-Concord NC-SC Durham NC Fayetteville NC Goldsboro NC Greensboro-High Point NC Greenville NC Hickory-Lenoir-Morganton NC Jacksonville NC Raleigh-Cary NC Rocky Mount NC Virginia Beach-Norfolk-Newport News VA-NC Wilmington NC Winston-Salem NC Northeastern North Carolina nonmetropolitan area
15.91 17.01 16.6 36.42 22.24 19.84 20.2 19.37 18.66 21.89 19.75 24.45 26.48 22.25 20.77 20.87 18.58 18.3 18.94 18.46 18 17.71 17.07 21.46 25.59 17.56 16.35 19.05 18.17 16.86 16.42 21.54 16.5 19.92 18.43 19.62 16.49
105
Washington-Arlington-Alexandria DC-VA-MDWV Cape Coral-Fort Myers FL Crestview-Fort Walton Beach-Destin FL Deltona-Daytona Beach-Ormond Beach FL Fort Lauderdale-Pompano Beach-Deerfield Beach FL Metropolitan Division Gainesville FL Jacksonville FL Lakeland-Winter Haven FL Miami-Fort Lauderdale-Miami Beach FL Miami-Miami Beach-Kendall FL Metropolitan Division Naples-Marco Island FL North Port-Bradenton-Sarasota FL Ocala FL Orlando-Kissimmee FL Palm Bay-Melbourne-Titusville FL Palm Coast FL Panama City-Lynn Haven FL Pensacola-Ferry Pass-Brent FL Port St. Lucie FL Punta Gorda FL Sebastian-Vero Beach FL Tallahassee FL Tampa-St. Petersburg-Clearwater FL West Palm Beach-Boca Raton-Boynton Beach FL Metropolitan Division Northwest Florida nonmetropolitan area Northeast Florida nonmetropolitan area South Florida nonmetropolitan area Albany GA Athens-Clarke County GA Atlanta-Sandy Springs-Marietta GA Augusta-Richmond County GA-SC Brunswick GA Chattanooga TN-GA Dalton GA Gainesville GA Hinesville-Fort Stewart GA Macon GA
29.58 18.68 19.08 16.96 20.2 20.55 19.73 17.5 20.3 20.21 19.15 18.62 17.14 18.71 20.67 16.61 17.14 17.72 18.09 17.17 18.09 19.73 20.07 20.63 15.96 16.57 16.37 17.15 19.15 22.33 19.35 17.91 18.39 16.73 18.73 18.02 17.6
Other North Carolina nonmetropolitan area Western Central North Carolina nonmetropolitan area Western North Carolina nonmetropolitan area Bismarck ND Far Western North Dakota nonmetropolitan area West Central North Dakota nonmetropolitan area East Central North Dakota nonmetropolitan area Far Eastern North Dakota nonmetropolitan area Akron OH Canton-Massillon OH Cleveland-Elyria-Mentor OH Columbus OH Dayton OH Lima OH Mansfield OH Parkersburg-Marietta-Vienna WV-OH Sandusky OH Springfield OH Steubenville-Weirton OH-WV Toledo OH Wheeling WV-OH Youngstown-Warren-Boardman OH-PA West Northwestern Ohio nonmetropolitan area Other Ohio nonmetropolitan area Eastern Ohio nonmetropolitan area Southern Ohio nonmetropolitan area Lawton OK Oklahoma City OK Tulsa OK Northeastern Oklahoma nonmetropolitan area Northwestern Oklahoma nonmetropolitan area Southwestern Oklahoma nonmetropolitan area Southeastern Oklahoma nonmetropolitan area Bend OR Corvallis OR Eugene-Springfield OR Medford OR
16.39 17.09 15.98 18.23 18.63 17.43 16.04 16.57 19.74 17.54 20.59 21.03 20.39 18.29 17.62 17.23 16.64 17.84 16.39 18.98 16 17.42 17.35 16.78 16.45 17.01 16.75 18.83 18.65 15.85 15.99 16.17 15.7 19.05 22.65 19.66 18.96
106
Rome GA Savannah GA Valdosta GA Warner Robins GA North Georgia nonmetropolitan area Middle Georgia nonmetropolitan area East Georgia nonmetropolitan area South Georgia nonmetropolitan area Honolulu HI Hawaii / Maui / Kauai nonmetropolitan area Boise City-Nampa ID Coeur d'Alene ID Idaho Falls ID Lewiston ID-WA Logan UT-ID Pocatello ID North Idaho nonmetropolitan area Southwest Idaho nonmetropolitan area Southcentral Idaho nonmetropolitan area East Idaho nonmetropolitan area Bloomington-Normal IL Cape Girardeau-Jackson MO-IL Champaign-Urbana IL Chicago-Naperville-Joliet IL Metropolitan Division Chicago-Naperville-Joliet IL-IN-WI Danville IL Davenport-Moline-Rock Island IA-IL Decatur IL Kankakee-Bradley IL Lake County-Kenosha County IL-WI Metropolitan Division Peoria IL Rockford IL St. Louis MO-IL Springfield IL Northwest Illinois nonmetropolitan area West Central Illinois nonmetropolitan area East Central Illinois nonmetropolitan area South Illinois nonmetropolitan area Anderson IN
18.17 19.07 15.46 20.6 15.71 16.11 15.63 15.68 21.68 19.21 19.52 17.22 19.19 17.85 16.71 17.47 17.16 16.29 16.81 20.05 22.11 17.06 21.64 23.62 23.32 17.7 18.87 19.15 18.04 23.33 19.56 19.73 20.9 21.17 17.81 17.09 16.87 17.9 16.69
Portland-Vancouver-Beaverton OR-WA Salem OR Coastal Oregon nonmetropolitan area Southern Oregon nonmetropolitan area Eastern Oregon nonmetropolitan area Linn County Oregon nonmetropolitan area Altoona PA Erie PA Harrisburg-Carlisle PA Johnstown PA Lancaster PA Lebanon PA Philadelphia PA Metropolitan Division Philadelphia-Camden-Wilmington PANJ-DE-MD Pittsburgh PA Reading PA Scranton--Wilkes-Barre PA State College PA Williamsport PA York-Hanover PA Far Western Pennsylvania nonmetropolitan area West Central Pennsylvania nonmetropolitan area Northeastern Pennsylvania nonmetropolitan area East Central Pennsylvania nonmetropolitan area Aguadilla-Isabela-San Sebastian PR Fajardo PR Guayama PR Mayaguez PR Ponce PR San German-Cabo Rojo PR San Juan-Caguas-Guaynabo PR Yauco PR Puerto Rico nonmetropolitan area 1 Puerto Rico nonmetropolitan area 2 New Shoreham Town Rhode Island nonmetropolitan area Anderson SC Charleston-North CharlestonSummerville SC Columbia SC Florence SC
22.58 19.5 17.38 17.59 17.49 18.72 16.51 17.69 20.72 17.05 18.81 18.41 23.69 23.47 20.44 19.57 17.73 20.4 17.28 19.05 17.43 16.54 17.01 18.09 11.04 11.71 13.69 11.73 11.7 10.88 13.25 10.87 11.72 11.13 17.03 17.27 19.21 19.39 17.49
107
Bloomington IN Cincinnati-Middletown OH-KY-IN Columbus IN Elkhart-Goshen IN Evansville IN-KY Fort Wayne IN Gary IN Metropolitan Division Indianapolis-Carmel IN Kokomo IN Lafayette IN Louisville-Jefferson County KY-IN Michigan City-La Porte IN Muncie IN South Bend-Mishawaka IN-MI Terre Haute IN Northeast Indiana nonmetropolitan area Northwest Indiana nonmetropolitan area Southwest / Southeast Indiana nonmetropolitan area Ames IA Cedar Rapids IA Des Moines-West Des Moines IA Dubuque IA Iowa City IA Omaha-Council Bluffs NE-IA Sioux City IA-NE-SD Waterloo-Cedar Falls IA Northeast Iowa nonmetropolitan area Northwest Iowa nonmetropolitan area Southwest Iowa nonmetropolitan area Southeast Iowa nonmetropolitan area Kansas City MO-KS Lawrence KS Manhattan KS St. Joseph MO-KS Topeka KS Wichita KS Kansas nonmetropolitan area
17.14 20.5 19.95 17.75 18.78 18.74 19.08 20.54 20.29 18.94 19.39 16.79 18.06 18.89 16.92 16.56 16.48 17.08 20.08 19.77 20.72 17.34 20.11 19.76 16.14 17.87 16.22 16.16 15.62 16.59 21.18 17.55 17.09 16.83 18.62 19.02 15.93
Greenville-Mauldin-Easley SC Myrtle Beach-Conway-North Myrtle Beach SC Spartanburg SC Sumter SC Low Country South Carolina nonmetropolitan area Upper Savannah South Carolina nonmetropolitan area Pee Dee South Carolina nonmetropolitan area Lower Savannah South Carolina nonmetropolitan area Rapid City SD Sioux Falls SD Central South Dakota nonmetropolitan area Eastern South Dakota nonmetropolitan area Western South Dakota nonmetropolitan area Cleveland TN Jackson TN Johnson City TN Kingsport-Bristol-Bristol TN-VA Knoxville TN Morristown TN Nashville-Davidson--Murfreesboro-Franklin TN Western Tennessee nonmetropolitan area South Central Tennessee nonmetropolitan area North Central Tennessee nonmetropolitan area Eastern Tennessee nonmetropolitan area Abilene TX Amarillo TX Austin-Round Rock TX Beaumont-Port Arthur TX Brownsville-Harlingen TX College Station-Bryan TX Corpus Christi TX Dallas-Fort Worth-Arlington TX Dallas-Plano-Irving TX Metropolitan Division El Paso TX Fort Worth-Arlington TX Metropolitan Division Houston-Sugar Land-Baytown TX Killeen-Temple-Fort Hood TX
18.8 14.84 18.55 16.11 17.17 17.26 15.44 16.44 16.64 17.77 15.14 15.69 15.27 16.21 16.8 17.08 17.73 18.8 15.75 19.57 15.76 16.68 15.18 15.54 16.72 18.14 22.18 19.1 15.25 18.89 17.35 21.89 22.53 16.88 20.36 22.26 17.74
108
Bowling Green KY Clarksville TN-KY Elizabethtown KY Huntington-Ashland WV-KY-OH Lexington-Fayette KY Owensboro KY West Kentucky nonmetropolitan area South Central Kentucky nonmetropolitan area West Central Kentucky nonmetropolitan area East Kentucky nonmetropolitan area Alexandria LA Baton Rouge LA Houma-Bayou Cane-Thibodaux LA Lafayette LA Lake Charles LA Monroe LA New Orleans-Metairie-Kenner LA Shreveport-Bossier City LA Hammond nonmetropolitan area Natchitoches nonmetropolitan area Winnsboro nonmetropolitan area New Iberia nonmetropolitan area Bangor ME Lewiston-Auburn ME Portland-South Portland-Biddeford ME Portsmouth NH-ME Rochester-Dover NH-ME Northeast Maine nonmetropolitan area Southwest Maine nonmetropolitan area Baltimore-Towson MD Bethesda-Frederick-Gaithersburg MD Metropolitan Division Cumberland MD-WV Hagerstown-Martinsburg MD-WV Salisbury MD Upper Eastern Shore nonmetropolitan area Garrett County Maryland nonmetropolitan area St. Mary's County Maryland nonmetropolitan area Barnstable Town MA Boston-Cambridge-Quincy MA-NH Boston-Cambridge-Quincy MA NECTA Division Brockton-Bridgewater-Easton MA NECTA Division
17.38 16.91 17.9 17.06 18.92 16.91 17.09 15.69 17 17.25 16.89 18.9 18.24 17.87 17.43 16.7 19.72 17.62 16.48 16.14 16.1 17.61 18.68 17.84 20.52 23.15 20.6 16.73 18.39 24 28.06 17.99 18.64 19.61 17.85 16.31 29.12 21.31 27.19 28.56 22.22
Laredo TX Longview TX Lubbock TX McAllen-Edinburg-Mission TX Midland TX Odessa TX San Angelo TX San Antonio TX Sherman-Denison TX Tyler TX Victoria TX Waco TX Wichita Falls TX Northwestern Texas nonmetropolitan area North Central Texas nonmetropolitan area Eastern Texas nonmetropolitan area Central Texas nonmetropolitan area Southern Texas nonmetropolitan area Gulf Coast Texas nonmetropolitan area Ogden-Clearfield UT Provo-Orem UT St. George UT Salt Lake City UT Northern Utah nonmetropolitan area West Central Utah nonmetropolitan area South Western Utah nonmetropolitan area Eastern Utah nonmetropolitan area Burlington-South Burlington VT Southern Vermont nonmetropolitan area Northern Vermont nonmetropolitan area Blacksburg-Christiansburg-Radford VA Charlottesville VA Danville VA Harrisonburg VA Lynchburg VA Richmond VA Roanoke VA Winchester VA-WV Southwestern Virginia nonmetropolitan area Southside Virginia nonmetropolitan area Northeastern Virginia nonmetropolitan area
16.14 17.83 17.38 15.61 20.76 18.91 17.05 18.95 17.48 17.72 17.5 17.64 16.66 16.58 16.62 16.09 16.19 15.72 16.52 18.4 18.67 16.43 20.47 18.96 16.28 16.11 18.02 21.98 19.76 18.79 18.57 21.8 16.64 17.71 17.48 21.41 18.26 19.28 16.24 15.83 20.48
109
Framingham MA NECTA Division Haverhill-North Andover-Amesbury MA-NH NECTA Division Lawrence-Methuen-Salem MA-NH NECTA Division Leominster-Fitchburg-Gardner MA Lowell-Billerica-Chelmsford MA-NH NECTA Division Nashua NH-MA NECTA Division New Bedford MA Peabody MA NECTA Division Pittsfield MA Providence-Fall River-Warwick RI-MA Taunton-Norton-Raynham MA NECTA Division Nantucket Island and Martha's Vineyard nonmetropolitan area Southwest Massachusetts nonmetropolitan area Northwest Massachusetts nonmetropolitan area North Central Massachusetts nonmetropolitan area Ann Arbor MI Battle Creek MI Bay City MI Detroit-Livonia-Dearborn MI Metropolitan Division Detroit-Warren-Livonia MI Flint MI Grand Rapids-Wyoming MI Holland-Grand Haven MI Jackson MI Kalamazoo-Portage MI Lansing-East Lansing MI Monroe MI Muskegon-Norton Shores MI Niles-Benton Harbor MI Saginaw-Saginaw Township North MI Warren-Troy-Farmington Hills MI Metropolitan Division Upper Peninsula of Michigan nonmetropolitan area Northeast Lower Peninsula of Michigan nonmetropolitan area Northwest Lower Peninsula of Michigan nonmetropolitan area Balance of Lower Peninsula of Michigan nonmetropolitan area
27.86 21.7 21.58 19.67 26.67 23.36 19.84 22.73 20.28 21.62 21.67 21.93 18.66 20.16 23.64 23.44 19.78 18.1 22.85 22.64 19.37 19.72 18.67 19.29 18.92 21 19.13 18.03 18.89 18.99 22.49 17.49 15.99 17.51 17.97
Northwestern Virginia nonmetropolitan area Bellingham WA Bremerton-Silverdale WA Kennewick-Pasco-Richland WA Longview WA Mount Vernon-Anacortes WA Olympia WA Seattle-Bellevue-Everett WA Metropolitan Division Seattle-Tacoma-Bellevue WA Spokane WA Tacoma WA Metropolitan Division Wenatchee WA Yakima WA Northwestern Washington nonmetropolitan area Southwestern Washington nonmetropolitan area Central Washington nonmetropolitan area Eastern Washington nonmetropolitan area Charleston WV Morgantown WV Southern West Virginia nonmetropolitan area North Central West Virginia nonmetropolitan area Appleton WI Eau Claire WI Fond du Lac WI Green Bay WI Janesville WI Madison WI Milwaukee-Waukesha-West Allis WI Oshkosh-Neenah WI Racine WI Sheboygan WI Wausau WI Eastern Wisconsin nonmetropolitan area West Central Wisconsin nonmetropolitan area South Central Wisconsin nonmetropolitan area Southwestern Wisconsin nonmetropolitan area
17.02 19.92 22.44 22.91 20.4 20.13 22.19 26.25 25.57 20.24 21.94 18.56 18.38 19.4 18.7 18.66 20.5 18.25 17.78 16.21 16.08 18.95 17.84 18.49 19.42 18.43 21.8 21.64 19.44 18.34 18.92 18.48 17.69 17.92 16.54 16.96
110
Northern Wisconsin nonmetropolitan area Casper WY Cheyenne WY Northwestern Wyoming nonmetropolitan area Southwestern Wyoming nonmetropolitan area Northeastern Wyoming nonmetropolitan area Southeastern Wyoming nonmetropolitan area
16.61 20.37 19.57 18.14 21.03 20.32 18.73
Table B.4 Truck percentage by state
State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Dist. of Col. Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Percentage of trucks 11.8% 9.5% 12.0% 16.8% 8.6% 6.7% 6.5% 8.9% 3.5% 8.7% 9.8% 3.9% 14.6% 12.0% 14.5% 15.2% 12.6% 13.8% 15.7% 9.0% 8.5% 5.0% 7.5% 6.7% 13.7% 14.5% State Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming USA Average Percentage of trucks 13.0% 12.9% 8.4% 7.4% 7.6% 19.7% 7.8% 10.9% 17.8% 11.1% 15.0% 12.1% 10.5% 4.7% 10.1% 13.3% 11.6% 12.2% 19.2% 8.8% 7.6% 10.4% 12.7% 13.2% 19.9% 10.6%
111
9
References
Bachman, W. H. (1997). Toward a GIS-Based Modal Model of Automobile Exhaust Emissions. EPA Cooperative Agreement #: CR823020. Georgia Institute of Technology. Atlanta, Georgia. Available at: <http://www.epa.gov/nrmrl/pubs/600r98097/600R98097.pdf. Last Accessed: December, 2011. Barth et al, (2000). The Development of a Comprehensive Modal Emissions Model. NCHRP Project 25-11 Final report. Available at: http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_w122.pdf. Last Accessed: September, 2011. Bertini, R.L., Tantiyanugulchai, S., Anderson E., Lindgren, R., & Leal, M. (2001). Evaluation of Region 2 Incident Response Program Using Archived Data. Portland State University, Research Report PSU-CE-TRG-01-01. Blumentritt, C.W., Pinnell, C., McCasland, W.R., Ross, D.W., and Glazer, J. for Selection of Ramp Control Systems. NCHRP Report No. 232. Boyles, S., Fajardo, D., & Fajardo., D. (2006), Naïve Bayesian Classifier for Incident Duration Prediction Boyles, S., S. T. Waller. (2007). A Stochastic Delay Prediction Model for Real-Time Incident Management, ITE Journal 77, 18-24. Chang, G.L. & Rochon, S. (2006). Performance Evaluation of CHART – the Real Time Incident Management System in Year 2006. Maryland Department of Transportation. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates. Chou, C., & Miller-Hooks, E. (2008). Benefit-cost Analysis of Freeway Service Patrol Programs: Methodology and Case Study. Presentation at the 88th Annual Meeting of the Transportation Research Board, Maryland, 20. Chou, C. and E. Miller-Hooks, E. (2010). Violations Modeling in Studies of Concurrent-Flow Lanes. Transportation Research Record 2196, 75-82. Chou, C., & Miller-Hooks, E. (2010). Simulation-Based Secondary Incident Filtering Method. Journal of Transportation Engineering, Vol. 136, No.8, 746-754. CMEM, (2010). Comprehensive Modal Emissions Model (CMEM). Center for Environmental Research and Technology. Riverside, California. Available at: <http://www.cert.ucr.edu/cmem/>. Last Accessed: December, 2011. Cuciti, P., & Janson, B. (1995). Incident Management via Courtesy Patrol: Evaluation of a Pilot Program in Colorado. Transportation Research Record, 1494. Transportation Research Board, National Research Council, Washington DC. Donnell, E., Patten, M., & Mason, J. (1999). Evaluating a Roadside Assistance Program: PennLincoln Parkway Service Patrol. Transportation Research Record: Journal of the Transportation Research Board, 1683(-1), 143-149. Dougald, L.E., Demetsky, M.J., (2008). Assessing Return on Investment of Freeway Safety Service Patrol Programs. Transportation Research Record: Journal of the Transportation Research Board, 2047, 19-27. Garib, A., Radwan, A.E., & Al-Deek, H. (1997). Estimating magnitude and duration of incident delays. Journal of Transportation Engineering 123 (6), 459–466. GREET (2008). GREET Transportation Fuel Cycle Analysis Model, GREET 1.8b. Argonne National Laboratory. USDOE. 2008. Available at: <cta.ornl.gov/bedb/.../LowerHigher_Heating_Values_for_Various_Fuels.xls>. Last Accessed: January, 2012.
112
Guin, A., Porter, C., Smith, B., & Holmes, C. (2007). Benefits Analysis for Incident Management Program Integrated with Intelligent Transportation Systems Operations: Case Study. Transportation Research Record: Journal of the Transportation Research Board, 2000, 78-87. Hadi, M., Sinha, P., & Wang, A. (2007). Modeling Reductions in Freeway Capacity due to Incidents in Microscopic Simulation Models. Transportation Research Record: Journal of the Transportation Research Board, 1999(-1), 62-68. Choa, F. , Milam , R. T. , & Stanek D. (2002). Hagen, L., Zhou, H., & Singh, H. (2005). Road Ranger Benefit Cost Analysis. Florida Department of Transportation. Haghani, A., Iliescu, D., Hamedi, M. & Yang, S. (2006). Methodology for Quantifying the Cost Effectiveness of Freeway Service Patrol Programs: A Case Study. University of Maryland.Guidelines Han, L. D., Chin, S., & Hwang, H. (2003). Estimating Adverse Weather Impacts on Major U.S. Highway Network, Transportation Research Board, Washington, D.C. Hani S. Mahmassani, Jing Dong, Jiwon Kim, Roger B. Chen and Byungkyu (Brian) Park (2009). Incorporating Weather Impacts in Traffic Estimation and Prediction Systems, Final Report. Hawkins, P. (1993). Evaluation of the Southwest Freeway Motorist Assistance Program in Houston. Texas Transportation Institute, Texas Department of Transportation, Report No. 1922-1F. Hirunyanitiwattana, W., & Mattingly, S. (2006). Identifying Secondary Crash Characteristics for California Highway System. Presented at 85th Annual Meeting of the Transportation Research Board, Washington, D.C. Hobeika, A.G. & Dhulipala, S. (2004). Estimation of Travel Times on Urban Freeways under Incident Conditions, Transportation Research Record: Journal of the Transportation Research Board 1867, TRB, National Research Council, Washington D.C., 97-106 International Panel on Climate Change (IPCC), (2006). 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Available at: <http://www.ipccnggip.iges.or.jp/public/2006gl/index.html>. Last Accessed: November, 2011. Karlaftis, M. G., Richards, N. J., Latoski, S. P., & Sinha, K. C. (1998). Empirical analysis of secondary crash causes. Proc., 77th Annu. Meeting, Transportation Research Board, Washington, D.C. Karlaftis, M., S. Latoski, N. Richards, & K. Sinha. (1999). ITS Impacts on Safety and Traffic Management: An Investigation of Secondary Crash Causes. ITS Journal, Vol. 5, No. 1, 39–52. Khattak, A., J. Schofer, and M. Wang. (1995) A simple time-sequential procedure for predicting freeway incident duration. IVHS Journal, Vol. 2, , 113-138. Khattak, A., Wang, X., & Zhang, H. (2008). Are Incident Durations and Secondary Incidents Interdependent Presented at the 2009 Transportation Research Board Annual Meeting, Washington, D.C. Kim, W., Chang, G., & Rochon, S. M.(2008)Analysis of Freeway Incident Duration for ATIS Application, Presented the 15th world congress on intelligent transportation system, New York Koupal et al., (2005). Koupal, J, Landamn,l, Nam,E, Scarbro,C, Glover,E and Ginannelli, R. MOVES Energy and Emissions Inputs. Office of Transportation and Air Quality. Available at: http://www.epa.gov/otaq/models/ngm/420p05003.pdf. Last Accessed: December, 2011. Latoski, S. P., Pal, R., & Sinha K.C. (1999). Cost-Effectiveness Evaluation of Hoosier Helper Freeway Service Patrol. Journal of Transportation Engineering, 125(5), 429-437.
113
Lindley, J.A. (1987). Urban Freeway Congestion: Quantification of the Problem and Effectiveness of Potential Solution. ITE Journal 57 (1), 27-32. Martin Fellendorf , Peter Vortisch, (2001) Validation of the Microscopic Traffic Flow Model VISSIM in Different Real-Workd Situations. Annual Meeting TRB. Masinick, J. P., & Tang, H. (2004).An analysis on the Impact of Rubbernecking onUrban Freeway Traffic, Research Report No. UVACTS-15-0-62. Mauch, M., Ahn, S., Chung, K., & Skabardonis, A. (2005). Baseline Evaluation of the Freeway Service Patrol (FSP) I-710 Big-Rig Demonstration Program. Institute of Transportation Studies, University of California-Bekerley, Working Paper UCB-ITS-WP-2005-3. Minnesota Department of Transportation, (2004). FIRST Program Evaluation. ITS annual report. Moore, J., G. Giuliano, & Cho, S. (2004). Secondary Accident Rates on Los Angeles Freeways. Journal of Transportation Engineering, Vol. 130, No. 3, 280–285. Murty, A. S. R., & Verma, A. K. (1986). Inverse normal and lognormal distributions for reliability design. Reliability Engineering, 15(1), 55-60. Nam, D., & Mannering, F. (2000). An exploratory hazard-based analysis of highway incident duration. Transportation Research Part A: Policy and Practice 34 (2), 85–102. Office of Transportation and Air Quality (OTAQ), 2012. Regulations and Standards. Available at: <http://www.epa.gov/otaq/climate/regulations.htm>. Last Accessed: January, 2012. Pulugurtha, S., Dangeti, M., Nambisan, S., & Kaseko, M.(2002). Simulating and Analyzing 27 Incidents Using CORSIM and VISSIM Traffic Simulation Software. Proceedings of the 28 Seventh International Conference, Applications of Advanced Technologies in Transportation. Rakha, H., Farzaneh, M., Arafeh, M., Hranac, R., Sterzin, E. & Krechmer, D. (2007). Empirical Studies on Traffic Flow in Inclement Weather, Final Report - Phase I. Raub, R. A., (1997). Occurrence of Secondary Crashes on Urban Arterial Roadways. Transportation Research Record 1581, TRB, National Research Council, Washington, D.C., 1997. pp. 53–58. Roadway Design Manual, 2010. Saka, A. A., Jeihani, M., & James, P. A. (2008) Estimation of Traffic Recovery Time for Different Flow Regimes on Freeways. Final Report No. SP708B4L. July 2008 Shankar, V., Mannering, F., & Barfield, W. (1995). Effect of roadway geometrics and environmental factors on rural freeway accident frequencies. Accident Analysis & Prevention, 27(3), 371-389. Skabardonis, A., & Mauch, M. (2005). SSP Beat Evaluation and Predictor Models: User’s Manual. Institute of Transportation Studies, University of California-Berkeley, Research Report: UCB-ITSRR-2005-XX. Skabardonis, A., Noeimi, H., Petty, K., Rydzewski, D., Varaiya, P., & Haitham, A. (1995). Freeway Service Patrol Evaluation. Institute of Transportation Studies, University of CaliforniaBerkeley, California PATH Research Report No. UCB-ITS-PRR-95-5. Skabardonis, A., Petty, K., Varaiya, P., & Bertini, R. (1998). Evaluation of the Freeway Service Patrol (FSP) in Los Angeles. Institute of Transpotation Studies, University of California-Berkeley, California PATH Research Report No. UCB-ITS-PRR-98-31.
114
Smith, K., & Smith, B.L., (2001). Forecasting the Clearance Time of Freeway Accidents. No. STL2001-01, Center for Transportation Studies (University of Virginia), Charlottesville, VA. Sun, C. (2007). Secondary Accident Data Fusion for Assessing Long-Term Performance of Transportation Systems. Midwest Transportation Consortium, Center for Transportation Research and Education, Cedar Falls, Iowa. Sun, C., Chilukuri, V., Ryan, T., & Trueblood, M. (2010). Evaluation of Freeway Motorist Assist Program. University of Missouri-Columbia, Final Report RD 09-004. Sun,C., Chilukuri, V., (2010). Dynamic Incident Progression Curve for Classifying Secondary Traffic Crashes. Journal of Transportation Engineering, Vol. 136, No. 12, 1153-1158. Traffic Incident Management Handbook, 2000 USDOE, (2011). Alternative Fuels & Advanced Vehicles Data Center. Energy Efficiency & Renewable Energy. USDOE. Available at: <http://www.afdc.energy.gov/afdc/>. Last Accessed: December, 2011. USDOE, (2012). Fuel Economy. Energy Efficiency & Renewable Energy. USDOE. Available at: < http://www.fueleconomy.gov/feg/findacar.shtml>. Last Accessed: January, 2012. USEPA, (2004). Update of Methane and NitroUSOxide Emission Factors on On-Highway Vehicles. USEPA Report Number: EPA420-P-04-016. Available at: < http://www.epa.gov/otaq/models/ngm/420p04016.pdf >. Last Accessed: January, 2012. USEPA, (2006). High Global Warming Potential (GWP) Gases. Available at: http://www.epa.gov/highgwp/scientific.html. Last Accessed: January, 2011. USEPA, (2007). Emission Facts: Average Annual Emissions and Fuel Consumption for Passenger Cars and Light Trucks. Consumer Information. USEPA. Available at: < http://www.epa.gov/oms/consumer/f00013.htm> . Last Accessed: January, 2012. USEPA, (2009). Transportation and Air Quality. Available at: http://www.epa.gov/otaq/. Last Accessed: August, 2011. USEPA, (2011). Motor Vehicle Emissions Model (MOVES). Modeling and Inventories. USEPA. Available at: <http://www.epa.gov/otaq/models/moves/index.htm>. Last Accessed: January, 2012. Wang, Y.H., Hallenbeck, M., & Cheevarunothai, P. (2008).Quantifying Incident-Induced 21 Travel Delays on Freeways Using Traffic Sensor Data Washington State Department of 22 Transportation, Report No. TNW2008-02, 23 Wirasinghe, S. C. (1978) Determination of traffic delays from shock-wave analysis. Transportation Research 12, 343-348. Zhan, C., A. Gan, and M. Hadi., (2009). Identifying Secondary Crashes and Their Contributing Factors. In Transportation Research Record: Journal of the Transportation Research Board, No. 2102, Transportation Research Board of the National Academies, Washington, D.C., 68-75. Zhan, C., Shen, L., Hadi, M., and Gan, A. (2008) . Understanding the characteristic of secondary crashes on freeways. CD-ROM, Transportation Research Board, Washington, D.C. Zhang, H., & Khattak, A. (2010). Analysis of Cascading Incident Event Durations on Urban Freeways. Transportation Research Record: Journal of the Transportation Research Board, No. 2178, Transportation Research Board of the National Academies, Washington, D.C., pp. 30–39.
115
doc_587649159.pdf
This can take the form of being protected from the event or from exposure to something that causes health or economical losses. It can include protection of people or of possessions.
ABSTRACT Title: STANDARDIZING AND SIMPLIFYING SAFETY SERVICE PATROL BENEFIT-COST RATIO ESTIMATION
Mersedeh TariVerdi, M.S., 2012
Directed By:
Prof. E. Miller-Hooks, Department of Civil and Environmental Engineering
Safety Service Patrol (SSP) programs operate nationwide with the aim of mitigating the impact of traffic incidents, especially along urban freeways. The central mission of the SPP programs is to reduce incident duration thereby reducing congestion related travel delays, fuel consumption, emission pollutants, and the likelihood of secondary incidents. The SSP-BC Tool was developed herein to fill the need for a standardized benefit-cost ratio estimation methodology for SSP programs with wide applicability and substantiated and needed updatable monetary conversion rates. The developed tool is designed to capture characteristics of incident, traffic, roadway geometry, and weather particular to the state area. VISSIM, a traffic microsimulation platform, was used to develop several multiple regression models with R-square values of 0.7 to 0.9 to assess the impact of travel delay, fuel consumption, and emission pollutants. Separate approaches were employed to estimate the savings in secondary incidents. In addition, a comprehensive method to compute fuel consumption and emissions is presented.
STANDARDIZING AND SIMPLIFYING SAFETY SERVICE PATROL BENEFITCOST RATIO ESTIMATION .
Mersedeh TariVerdi
Thesis submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Master of Science 2012
Advisory Committee: Professor Elise Miller-Hooks, Chair Professor Paul Schonfeld Assistant Professor Lei Zhang
© Copyright by Mersedeh TariVerdi 2012
Acknowledgments
First and foremost I offer my sincerest gratitude to my supervisor, Prof Elise MillerHooks, who has supported me throughout my thesis with her patience and knowledge. No doubt that without her this thesis would not have been completed or written. I am deeply indebted to my committee members Prof. Paul Schonfeld and Dr. Zhang for their time and effort in reviewing this work.
I acknowledge the I-95 Corridor Coalition for their financial support for this project and Tier 1 center (CTSM) for the 9-month fellowship.
I would like to thank my team-mates, Suvish, Xiadong, and Kara for sharing the literature and invaluable assistance; to my best friends, Nezam, Alireza and many others for their support. I am deeply and forever indebted to my parents for their love, support and encouragement throughout my entire life. I am very grateful to my brothers Sam and Amir.
ii
Contents
CHAPTER 1. Introduction...................................................................................... viii CHAPTER 2. Background and Literature Review .................................................... 6 2.1 2.2 2.3 2.4 Factors to Consider in Benefit-Cost Estimation........................................... 6 Evaluation and Estimation Methodologies .................................................. 6 Overview of Service Patrol Benefit/Cost Ratios per State ......................... 10 Summary .................................................................................................... 12
CHAPTER 3. Factors Affecting SSP Benefits and Their Calculation .................... 14 3.1 3.2 Measures of Effectiveness (MOEs)............................................................ 14 Factors Contributing to Travel Delay and Fuel Consumption ................... 14
3.2.1 Geometry of Roadway Segment .......................................................... 18 3.2.2 General Terrain .................................................................................... 19 3.2.3 Traffic Characteristics .......................................................................... 21 3.2.4 Incident Attributes ................................................................................ 24 3.2.5 Weather Conditions .............................................................................. 27 3.3 3.4 Calculating Travel Delay ........................................................................... 29 Calculating Fuel Consumption and Air Pollutants ..................................... 30
3.4.1 Emissions Calculation .......................................................................... 31 3.4.2 Fuel Consumption and Emission Calculation Methodology................ 34 3.4.3 Power Demand (Pv,t) and Instantaneous Fuel Consumption (FRt) Calculation ....................................................................................................... 35 3.4.4 CO2, CO, HC, CO & NOx Emissions Calculation............................... 37 3.4.5 SOx Emissions Calculation .................................................................. 37 3.4.6 Fuel Consumption ................................................................................ 37 3.4.7 Assumptions ......................................................................................... 37
iii
3.5 3.6
Secondary Incident Savings ....................................................................... 38 Summary .................................................................................................... 44
CHAPTER 4. Implementation of Variables in Simulation Designs ........................ 45 4.1 General Simulation Settings and Incident Modeling ................................. 47
4.1.1 Simulation Settings .............................................................................. 47 4.1.2 Control Case ......................................................................................... 48 4.1.3 Simulating Incidents and Rubbernecking Effect .................................. 48 4.1.4 VISSIM Calibration ............................................................................. 50 4.2 Single-Factor Experiments ......................................................................... 51
4.2.1 Geometry Factors: Number of Lane and Lane Blockages ................... 52 4.2.2 Geometry Factors: Ramps .................................................................... 54 4.2.3 General Terrain: Horizontal Curves ..................................................... 55 4.2.4 General Terrain: Vertical Curves ......................................................... 56 4.2.5 Traffic Characteristics: Speed of Vehicles ........................................... 57 4.2.6 Traffic Characteristics: Demand Flow Rate ......................................... 58 4.2.7 Traffic Characteristics: Truck Percentages .......................................... 60 4.2.8 Simultaneous Changes of Factors ........................................................ 60 4.3 Multiple-Regression Analysis .................................................................... 61
4.3.1 Minimum Sample Size ......................................................................... 61 4.3.2 Designing a Sample of Incidents .......................................................... 62 4.3.3 Multiple-Regression Models ................................................................ 64 4.4 Summary .................................................................................................... 75
CHAPTER 5. B/C Ratio Estimation ........................................................................ 76 5.1 Savings Computation ................................................................................. 76
5.1.1 Savings in Travel Delay ....................................................................... 77
iv
5.1.2 Saving in Fuel Consumption and Emissions ........................................ 79 5.1.3 Saving in Secondary Incidents ............................................................. 79 5.2 Total Benefit Calculation ........................................................................... 81
5.2.1 Monetary Values .................................................................................. 81 5.2.2 Computing Total Benefit ...................................................................... 82 5.3 5.4 Cost Calculation ......................................................................................... 84 The B/C ratio .............................................................................................. 85
CHAPTER 6. The Tool by Illustrative Example ..................................................... 87 6.1 The SSP-BC Tool ....................................................................................... 87
CHAPTER 7. Conclusions and Limitations ............................................................ 98 7.1 7.2 Limitations ................................................................................................. 98 Contributions .............................................................................................. 99
APPENDIX A: Fuel Consumption Computation Tables....................................... 100 APPENDIX B: Monetary Equivalents................................................................... 101 References .............................................................................................................. 112
v
List of Tables
Table 1.1 Freeway service patrol B/C ratio ............................................................. 10 Table 2.1 Previous studies on travel delay and incident duration............................ 16 Table 2.2 Maximum grade table (Adopted from roadway design manual 2010, .... 21 Table 2.3 FFS vs. Base capacity for freeways (Adopted from HCM 2010) ............ 22 Table 2.4 Average incident duration (Chou and Miller-Hooks, 2008) .................... 26 Table 2.5 Weather-Speed relation............................................................................ 28 2.6 Speed and capacity reduction based on road type (Chin et al. 2004)) ............... 28 Table 2.7 Actual speed under adverse weather conditions ...................................... 29 Table 2.8 Nomenclature for variables used in equations ......................................... 34 Table 2.9 Calculation of Road -Load coefficients [Source: USEPA, 2011c] .......... 36 Table 2.10 Secondary incident classification methods ............................................ 41 Table 3.1 Driver behavior parameters, adopted from Miller-Hooks et al. (2010) ... 51 Table 3.2 Summary of variables used in numerical experiments ............................ 52 Table 3.3 Number of lanes and lane blockage analysis ........................................... 53 Table 3.4 Ramp analysis .......................................................................................... 55 Table 3.5 Segment grade analysis ............................................................................ 56 Table 3.6 Impact of gradient on average speed ....................................................... 56 Table 3.7 Fuel economy changes due segment grade .............................................. 57 Table 3.8 Travel delay and fuel consumption changes by FFS ............................... 57 Table 3.9 Truck percentage analysis ........................................................................ 60 Table 3.10 Improved R-square comparison for new model .................................... 72 Table 3.11 LDV fuel consumption .......................................................................... 75 Table 4.1 Summary of monetary equivalents (ATRI) ............................................. 81 Table 5.1 Geometry and traffic default values......................................................... 91 Table 5.2 New York 2006 monetary values ............................................................ 95 Table 5.3 H.E.L.P result comparison ....................................................................... 96 Table A.1 Calculation of Road-load coefficients .................................................. 100 Table A.2 Emission factors .................................................................................... 100 Table A.3 Fuel properties ...................................................................................... 100 vi
Table A8.4 Transmission parameters for engine speed calculation (Source: PERE, 2005; EPA420-P-05-001) .................................................................................................. 100 Table B.1 Average gasoline prices ........................................................................ 101 Table B.2 Average labor cost by state ................................................................... 102 Table B.3 Average wage by area ........................................................................... 103 Table B.4 Truck percentage by state...................................................................... 111
vii
List of Figures
Figure 2.1 Base capacity vs. Total ramp density (adapted from HCM, 2010, P. 10-7) .............................................................................................................................................. 23 Figure 2.2 Incident timestamp flow chart (Chou and Miller-Hooks, 2008) ............ 25 Figure 3.1 Estimation model development progress................................................ 46 Figure 3.2 Incident layout on typical three-lane unidirectional freeway segment ... 50 Figure 3.3 Number of lanes and lane blockage analysis .......................................... 53 Figure 3.4 Simulating incidents close to an off-ramp .............................................. 54 Figure 3.5 Average speed Vs. Traffic flow rate....................................................... 58 Figure 3.6 Travel time vs. Traffic flow.................................................................... 59 Figure 3.7 Summery of fit diagnostic for total travel delay of LDV ....................... 68 Figure 3.8 Scatterplots of residuals against explanatory variables .......................... 69 Figure 3.9 Summery of fit diagnostic of linear model of LDT fuel consumption ... 73 Figure 3.10 Scatterplots of residuals against explanatory variables ........................ 74 Figure 4.1 Travel delay estimation procedure ......................................................... 78 Figure 4.2 Subcategory linear interpolation............................................................. 79 Figure 5.1 Main window .......................................................................................... 88 Figure 5.2 Program cost detail ................................................................................. 89 Figure 5.3 Basic data in segment level .................................................................... 89 Figure 5.4 Program information window................................................................. 91 Figure 5.5 Roadway geometry and traffic information ........................................... 93 Figure 5.6 Incident information window ................................................................. 94 Figure 5.7 Output window ....................................................................................... 95 Figure 5.8 CORSIM vs. VISSIM ............................................................................. 97
viii
CHAPTER 1. Introduction
Traffic congestion adversely effects traveler safety, cost, quality of life and the environment. Traffic congestion can be recurring or non-recurring, i.e. due to randomly arising events. Congestion caused by recurring events is a result of traffic demand exceeding the fixed capacity of a road segment during every day traffic patterns. Conversely, when the available capacity of a road segment decreases due to unpredictable events, such as vehicle incidents or adverse weather conditions, it is categorized as nonrecurrent congestion. The Federal Highway Administration (FHWA) sites non-recurrent congestion as the cause of approximately 60 percent of all road traffic in the United States. Traffic incidents (e.g. accidents, breakdowns), in particular, are a major cause of non-recurring traffic congestion. In fact, Caltrans’ Division of Research and Innovation claims that traffic incidents cause about 25 percent of this type of congestion on freeways (Caltrans, 2010). Therefore, incidents are counted as one of the most significant reasons for congestion in vehicular transportation. Increased travel time, increased risk of secondary accidents and decreased safety to other drivers and responders were identified as the most serious problems associated with an incident in a road segment. Additional problems, such as increased fuel consumption, pollutant emissions, cost of goods and services and negative impact on emergency response time, are also considered by decision-makers to be significant issues arising from traffic incidents. Moreover, the longer the incident duration and the time of impact to traffic flow, the greater the incident’s negative impacts. It is evident that traffic incidents have a strong adverse effect on urban areas. As such, reducing incident clearance time can mitigate its impacts (see, for example, Blumentritt et al., 1981). To control the impact of traffic incidents, Traffic Incident Management (TIM) programs that aim to reduce the duration and consequences of incidents and improve the safety of motorways have been introduced nationwide. A significant goal of most TIM programs is to coordinate the response by a number of public and private organizations to incidents. For example, transportation agencies are typically called to the incident scene by first responders, including law enforcement. Freeway Service Patrol (FSP) and similar Safety Service Patrol (SSP) programs are often components of a large TIM program. These 1
programs have been widely cited as very effective. In these programs, the service patrol vehicles may roam the roadways to which they are assigned (i.e. their beats), monitoring and responding to observed incidents. Alternatively, they may be dispatched to a call while roaming or from a traffic incident management center. The trained drivers of patrol vehicles may call for assistance from law enforcement and/or emergency responders, or may directly assist with motorists’ needs. SSPs receive most of their funding from state and federal taxes. Therefore, they are must have public support and may be scrutinized when budgets are limited. While states work to provide essential services they must consider their budget limitations. As such, quantifying the benefits of SSPs is important to legislators determine the effectiveness of such programs in terms of improving safety and increasing public benefits. Thus, even in times of budget crises, the benefits of these programs may be great enough that funding them is encouraged. This research effort builds on recent I-95 Corridor Coalition efforts (Chou and Miller-Hooks, 2008) in which a procedure was developed to determine the benefits (i.e. reduction in congestion, secondary incidents, fuel consumption and pollution, along with their monetary equivalents) of an existing SSP program. The methodology was employed to estimate the B/C ratio for the Highway Emergency Local Patrol (H.E.L.P.) program in New York. This prior effort revealed the need for additional study in identifying a set of best performance measures and monetary conversion rates to accurately depict the benefits and costs of such programs. Moreover, the developed approach, like other comparable methods with similar accuracy used around the country, requires significant computational effort and is, therefore, costly and time-consuming. A quicker and less data-intensive approach is desired so that it can be readily and widely utilized by all states around the US Such an approach is developed within the effort described herein. When attempting to compare an SSP program to its alternatives, it is common to compare the benefits of the program with its costs through a benefit - cost (B/C) ratio. This ratio has been estimated for many SSP programs operating around the nation. These ratios, however, range dramatically (e.g. from 2-to-1 to 36-to-1). The majority of this variability is likely due to the wide range of estimation methodologies and monetary equivalent 2
conversion factors employed within these techniques, rather than to actual differences between the program benefits. This great variability also opens these findings to questions about their accuracy. A standardized methodology that can be universally and equitably employed in such B/C ratio estimation is essential as it would aid in creating consistency and, therefore, greater confidence in the validity of the results. The main objective of this study was to develop a user-friendly tool, referred to as the SSP-BC Tool, based on consistent performance measures and monetary conversion rates that can quickly compute the B/C ratio of an existing or planned SSP program. In Chapter two, a review of existing US SSP programs and other relevant studies that evaluate their performance, as well as reported B/C ratios, are presented. There are numerous factors that might be considered in evaluating the benefits to society of a SSP program. The most common are: savings in travel time, fuel consumption, pollutant emissions and secondary incidents. These are described in detail in Chapter three. In the same chapter, a review of factors that have the greatest effect on these measures is also provided. Components of weather, roadway gradient and curvature, density of ramps in the roadway segment, and traffic composition, all of which influence the available capacity of a roadway segment and fuel consumption rates, are considered in this study. To quantify these benefits of a SSP program, VISSIM, a microscopic traffic simulation product, is employed. The Component Object Module (COM) interface of VISSIM was used for modeling freeway incidents. The COM interface permits controlled experiments with altered spatial and temporal incident characteristics. To provide a realist portrayal of an incident in a simulation environment, all factors that have been found to affect travel delay, fuel consumption and emissions must be considered in the experimental design criteria. A methodology for modeling traffic incidents is adopted that exploits this COM interface. Within this methodology factors, such as roadway length and gradient, are directly set in VISSIM, while other factors, such as volume, traffic composition and incident attributes, are defined through the interface. The effect of weather on roadway performance is captured through changes to free flow speed. This methodology is described in Chapter three.
3
Replicating real-world conditions within a simulation environment requires a primary study into the capabilities of the software and creation of specific methods needed to adequately capture desired effects. Chapter four highlights the necessity of this investigation and describes these methods. Initial runs were made to study trends in travel delay and fuel consumption estimates resulting from univariate changes in factors. Whether variables are dependent was also verified through runs in which the state of two or three factors were permitted to change simultaneously. Upon analyzing outputs from the simulation runs, and after discussion with PTV America’s support, it was determined that the built-in fuel consumption tool available within VISSIM was not suitable for this study. The tool did not provide repeatable estimates. Moreover, it seems that it overestimates fuel efficiency of vehicles available within the US and its equations were developed for emissions estimation from traffic on arterials. Specific methods for calculating fuel consumption and emissions from vehicular modal parameters gathered for each VISSIM run are presented in Chapter four. A review of factors affecting fuel consumption and emissions is given in Chapter three. Analysis of results from the initial runs provided insights into the response of the vehicles in VISSIM to changes in characteristics of the roadway segment, traffic volume and the roadway environment. As an example, these runs revealed that traffic performance was unaffected by changes in roadway curvature, an important input. Thus, efforts associated with these runs indicated that the effects of significant curvature could not be captured directly and a suitable methodology would be needed. Additional restrictions on combinations permitted in batch runs were identified, (i.e. when speed was set through code and grade was non-zero) that preclude the possibility of conducting simulation runs to capture all combinations of input. Because it was not possible to create batch runs to run all combinations of inputs, regression analysis was employed. Specifically, regression models of travel delay and fuel consumption were developed and calibrated based on simulation results from 1200 runs on a typical stretch of a three lane freeway. For each run, a randomly generated traffic incident was created. The incident scenarios involved one of three states of lane blockage (shoulder, one lane and two lane blockage) with equal likelihood. For each incident, the incident duration was set 4
according to a statistical distribution identified in previous studies. This culminated in 7 regression models capturing the response of dependent variables associated with travel delay and fuel consumption to incident duration, traffic volume, gradient and percentage of trucks. To reduce the error terms in the regression models, and improve overall fitness, an additional 73,290 runs were designed and conducted. The runs involved all possible combinations of 16 categories of incident duration, 11 categories of traffic volume and 6 speed categories, resulting in 1,056 combinations. For each combination, runs including one of 3 types of lane blockage and one of 5 possible roadway sizes in terms of number of lanes and 5 random seed for each were made. Note that no complete road closure scenario was considered. Other spatial and temporal incident characteristics were held constant. At the heart of the SSP-BC Tool is a database of five tables: tables of travel delays for light and heavy duty vehicles, average driver and police officer wages, and fuel costs. The tool pulls data from these tables to complete computations related to the benefits and costs of the studied system. Data in these tables are derived directly from the simulation run results (travel delays, fuel consumption), through regression-based estimates (travel delays, fuel consumption), computations (emissions, secondary incidents) or from publically available sources (wages, fuel costs, traffic composition, and monetary conversion rates). The regression models were used specifically to capture the effects of traffic compositions (i.e. percentage of trucks in traffic mix) and roadway grade. While Chapters three and four focus on individual incident characteristics and impacts, Chapter five describes calculations and assumptions used to obtain program benefit estimates, and ultimately the B/C ratio. Chapter six includes snapshots of the tool and a brief explanation of how the tool works through an illustrative example. Results of a case study involving the H.E.L.P program are presented Chapter seven. General findings follow in Chapter eight.
5
1
CHAPTER 2. Background and Literature Review
SSP programs exist in a large portion of the US SSP drivers can provide free assistance to motorists. Examples of service include: providing a gallon of gasoline, changing flat tires, jump starting dead batteries, pushing vehicles off the road, providing minor mechanical repairs, and helping motorists call for other emergency services. In the case of severe accidents, SSP drivers are trained to help police redirect traffic. These services are crucial for shortening the duration of incidents and, thus, diminishing their impact, and improving safety for other drivers on the roadway segment. Furthermore, SSPs can be used as probe vehicles, providing real-time updates on traffic conditions (Traffic Incident Management Handbook). In this chapter, evaluation studies on SSP programs around the nation reported in the literature are reviewed. B/C ratio computation approaches and estimates by program are reported. 1.1 Factors to Consider in Benefit-Cost Estimation The first step of estimating the benefits and costs of a SSP program is to determine the components that should be considered in the calculations. The reduction in travel delay and corresponding economic benefits for the motorists plays a significant role in the benefit estimation. Some studies also consider prevention of secondary incidents due to decreased incident duration; they assume a direct relation between number of potential secondary incidents and incident duration. In addition, environmental concerns, such as fuel consumption and pollutant emissions, are included in savings. Some studies derive an estimate for fuel consumption from delay time or use the computational tools available in some simulation software packages. There are other savings that should be counted in the benefits of a SSP program. For example, costs of towing if SSP vehicles were not at the scene, lawsuits from secondary incidents that are prevented, and additional time available for troopers for more urgent tasks that the SSP programs cannot handle. 1.2 Evaluation and Estimation Methodologies The most accurate way to evaluate the benefits of a SSP program is to conduct a “before-and-after” study that compares the incident detection, response, clearance and
6
recovery times (often marked by a return in traffic state to pre-incident flow rates) for a comparable period before and after the deployment of the SSP program. Donnell et al. (1999) evaluated the Penn-Lincoln Parkway Service Patrol in this way. The study recorded similar incidents that occurred prior to and following the implementation of SSPs, the collected data was compared to compute possible savings. Respectively, incident response time and clearance time were found to be reduced by an average of 8.7 and 8.3 minutes (17.1 minutes overall savings in incident duration), yielding a B/C ratio of 30:1. In another study, Skabardonis et al. (1995) analyzed the operation of SSP programs in San Francisco, California using field data from 24 weekdays before the SSP and 22 weekdays after the SSP program was implemented. The B/C ratio was shown to be 3.4. An assessment that was carried out by Bertini et. al. (2001) in one region of Oregon showed that the regional SSP program reduced the average cost of delay-causing incidents to travelers by 36 to 66 percent when it was upgraded from part-time to full-time. In many circumstances, however, the “before” dataset is not available as it usually requires a well-maintained and long-term managed database. Therefore, researchers adopt a “with-and-without” approach. This method compares a SSP managed incident to a similar incident that was managed by state or local police. Commissioned by the Safety Service Patrol (SSP) program in Northern Virginia (NOVA), Dougald et. al. (2008) compared the average durations of various episodes with similar incident types and roadway and traffic conditions. The main difference between each episode was whether or not the SSP program responded. The data that was compared came from two databases: 1) the incident management database (IMD) and 2) the Virginia State Police (VSP) computer-aideddispatch (CAD) system. This type of study has been applied to investigations conducted in other states, including Indiana (1999), Minnesota (2004), Florida (2005), California (1995, 2005), Maryland (2006), New York (2008) and Missouri (2010). In these circumstances where no “before” data was maintained, it is necessary to make assumptions surrounding how long the incident would last had such a SSP program not existed. In this way, the potential savings of SSP-assisted incidents can be calculated. Sensitivity analysis can be conducted to examine how the B/C ratio responds to varying incident duration savings. Generally, the range of assumed duration reduction is between
7
10 and 20 minutes. A study of the evaluation of the SSP program in Los Angeles, assumed that the SSPs would reduce incident duration by 10, 12.5, or 15 minutes, resulting in a B/C ratio that ranged from 3.75:1 to 5.5:1 (Skabardonis et al., 1998). Moreover, the average duration of crashes and in-lane incidents handled with the Hoosier Helper SSP were assumed to be lowered by 10 min while all other incident durations were reduced by 15 minutes (Latoski et al. 1999). Chou et al. (2008) lengthened the duration of without FSPassist incidents by between 5 and 25 minutes in 5-minute increments for studies on SSP program of New York State, H.E.L.P. The estimation methods of incident delay and delay savings of SSP programs draw on methods such as statistical, deterministic queuing, or simulation-based models, or surveys. Examples of studies involving each are given next. Mauch et al. (2005) examined the Big-rig SSP pilot program that provides services including heavy-duty tow trucks along the I-710 freeway in California through use of statistical modeling. Regression analysis was completed and correlations between average vehicular delay per day and big-rig and non-big-rig incidents were noted. A calibrated regression-based function was given to describe the relationship between traffic delay and incident duration. They combined the response time savings with delay estimates to forecast delay savings. It was assumed that for comparison purposes if conventional services would be required, 45 minutes of response time would be needed for dispatch of the big-rig tow truck. Expected benefits from the program were estimated to be $14,700/day with a benefit-cost ratio of 5:1, where benefit computations include travel delay only. Other studies employing statistical approaches to evaluating the benefits of SSP programs include: Mauch et al. (2005), Haghani et al.(2006). Where traffic volume profiles over the incident duration are available, a deterministic queuing model can be applied to predict travel delay. Skabardonis et al. (1995, 1998) estimated delay as the difference in travel times under normal and incident conditions using data from loop detectors and probe vehicles. Guin et al. (2007) used extrapolated capacity reduction factors during the response and clearance of incidents associated with cumulative traffic volume as inputs to the queuing model for the Georgia NaviGAtor SSP program. This approach was also employed in the evaluations of SSP 8
programs in Oregon (Bertini et al., 2001), Florida (Hadi and Zhan, 2006; Hagen et al, 2005), Virginia (Dougald wt. al, 2007) and Missouri (Sun et Al., 2010). The majority of SSP program evaluation studies rely on simulation, because it is often the case that traffic volume data and other traffic characteristics are limited and, thus, effects on traffic must be estimated. Traffic simulation models have become quite advanced, permitting control of roadway design, traffic volumes, and incident characterization, including incident duration, number of lanes blocked, and location. Latoski et al. (1999) estimated delay using the XXEXQ macroscopic traffic simulation model in studying the Hoosier Helper program in Northern Indiana. The study yielded B/C ratio of 4.71 if the program operates only in the day time and 13.28 for 24 hour operations of the program considering travel delay in benefits and annual investment, employee salaries and benefits, overhead and maintenance costs. 120 incidents were replicated in the microscopic CORSIM traffic simulation platform for the purposes of studying their effects on travel delay for Coordinated Highways Action Response Team (CHART) in Maryland (Chang et al., 2006). Benefits in reduction of travel delay, fuel consumption and emissions were estimated to be 1,006.50 million dollars in a similar study in 2009. Representative incidents with varying duration and lane blockage were simulated in the PARAMICS microscopic traffic simulation platform to analyze the Freeway Incident Response Safety Team (FIRST) program in Minnesota (MnDOT, 2004). Total incident delay was plotted against volume corresponding to different incident durations ranging from 4 to 40 minutes. Based on the plot, delay reductions can be predicted given the incident duration reduction caused by the FIRST program. The B/C ratio estimated for FIRST was 15.8, including travel delay and crash avoidance in the benefit estimation. More recently, Chou and MillerHooks (2008) replicated 693 actual incidents in CORSIM to analyze the B/C ratio of the H.E.L.P. program in New York. Incidents were simulated with H.E.L.P-assist and without H.E.L.P-assist circumstances. The rubbernecking effect set up in CORSIM was computed from capacity reduction estimates associated with number of lanes, lane blockage, and incident type. It was found that the B/C ratio range was between 2.14 and 2.68 (for different costs) using conservative monetary conversion rates for travel delay, fuel consumption, emission pollutants and avoided secondary incidents. The B/C ratio would
9
increase to between 13.2 and 16.2 if vehicle occupancy, traffic composition, and higher incident severity level were included in benefit evaluation. 1.3 Overview of Service Patrol Benefit/Cost Ratios per State The first SSP program with annual operations originated in 1960 in Chicago, Illinois. In 2006, the US DOT and Intelligent Transportation System (ITS) Joint Program Office (JPO) conducted a survey regarding service patrol programs in 106 metropolitan areas. At the time, 73 out of 99 areas that responded had a service patrol program in operation and more than 40 states had implemented at least one SSP program. From review of the literature, including journal articles, research reports and web pages from state departments of transportation, it can be concluded that service patrols reduce incident duration, improve safety, and help reduce fuel consumption and emissions. It was proved that its benefit outweighed its cost with the B/C ratio ranging from 1.48:1 to 38.25:1. A summary of B/C ratio estimates noted in the literature can be found in Table 2.1. It is evident from the table that the ratios vary widely. This is in part because there are inconsistencies in not only analysis methods and monetary conversion rates used to obtain the ratios, but also in the factors they include in benefit estimation. In fact, each state or city adopts its own set of factors that it deems relevant to calculate the B/C ratio. Underestimation may result from ignoring certain benefits while overestimation may occur from over-counting low probability events. Differences in monetary conversion rates, that provide monetary values for estimated benefits, can also greatly impact the final ratio values.
Table 1.1 Freeway service patrol B/C ratio
State Program name I710 Big-Rig Location/service area I-710 south of ocean blvd to the I-5 interchang LA Year Included benefits B/C
California
2005
Travel delay Travel delay, fuel consumption Travel delay Travel delay, fuel consumption
5 5 for 15 min duration reduction 3.4 2.3-41.5, 25.1 overall
California
FSP
1998
California Florida
FSP Road Ranger
SF district 1-7, except dis.3, Turnpike
1995 2005
10
Georgia Indiana Michigan Minnesota
NaviGAtor Hoosier Helper Freeway courtesy Patrol FIRST Freeway motorist assist program
Atlanta Northwest Indiana Southest Michigan Minneapolis, St. Paul
2007 1999 2009 2004
Travel delay, fuel consumption, emissions, avoided secondary incidents Travel delay Travel delay Air quality Travel delay, crash avoidance Travel delay, secondary crashes
4.4 4.7 daytime ,13.3 24h 15.2 15.8
Missouri
St.Louis Long Island; in New York City; the Lower Hudson Valley; Buffalo; Rochester; and the Albany Capital District Region 2 Hampton Road Northern Virginia
2010
38.25
New York
H.E.L.P
2009
Travel delay, fuel consumption, emissions, avoided secondary incidents Travel delay Fuel consumption Travel delay, fuel consumption, emissions Travel delay, fuel consumption, emissions
2.14-16.5 for 20 min reduction
Oregon Virginia Virginia
FSP SSP NOVE SSP
2001 2007 2006
32.52-3.68 overall 4.71, range of 1.48-10.17 Vs. V/C 6.2
Each mentioned previous study was conducted for a unique location given data for a specified time period and each such study typically required enormous effort to complete. These studies, however, are needed to defend and secure financial support for continued program operations. It is often the case that studies in one area re-invent methodologies created for studies in other areas. Moreover, their estimated B/C ratios cannot be directly compared, because they rarely use similar factors or monetary conversion rates. The Freeway Service Patrol Evaluation (FSPE) model, developed by the University of California at Berkeley (Skabardonis et al., 2005) for California, is a dedicated SSP evaluation tool. That can be applied more widely. The FSPE model was implemented in Excel workbook using Visual Basic for Application (VBA). This tool computes daily, annual or specified time period savings with respect to incident delay, fuel consumption and emission, as well as the B/C ratio. The tool relies on a deterministic queuing model for calculating incident delay. Benefits of SSPs are dependent on the beat’s geometric and traffic characteristics, and the frequency and type of assisted incidents. Default model
11
parameters are provided, but they can be modified by the users if empirical field data are available. This SSP model can analyze 24/7 SSP services, or can accommodate more limited weekday or weekend services, although the best final prediction model was found to be for 24 hour SSP services. They applied the 24 hour model to estimate the benefits. They count for limited operational hours and used the proportion of vehicle miles traveled (VMT) in fewer than 24 hours. Each beat is divided into segments for each travel direction and data are input accordingly. The FSPE model distributes the FSP-assists per incident type proportionally to the VMT per beat. It is assumed that the response time without the SSP is 30 minutes. The SSP response time is computed based on the beat length, average tow-truck speed, and number of trucks operating on the beat. The FSPP model later was developed based on the FSPE model to evaluate the roadway segments which does not have SSP assistance. The saving in response times are estimated based on a statewide weighted average of all incidents from the fisical year 20022003 FSP-assists database. One of the strengths of this SSP evaluation tool is the consideration of directional effects of daily traffic volume along each patrolled beat. The tool was calibrated for use in California. Significant effort and data are required to calibrate the queuing model for use in other locations (Skabardonis et al., 2005). 1.4 Summary While B/C ratio estimation models, specifically FSPP and FSPE, exist that might have general utility in B/C ratio estimation for SSP programs, the SSP-BC Tool proposed herein accounts for a wider array of traffic, environmental and program characteristics that influence benefit and cost estimates. The SSP-BC Tool accounts for factors, such as ramp density, horizontal and vertical alignments, traffic composition, and weather conditions, that have been identified as important to travel delay, fuel consumption and emissions estimation. Moreover, fuel consumption and pollutant emissions estimates used in prior related studies, when included, are made based on simplistic regionally developed rates for travel day to fuel consumption and emissions. For example, in the FSPP tool, the fuel consumption and emission calculation factors are developed for California conditions. The tool uses average vehicular speed; thus, driving modes such as acceleration or deceleration and stops of vehicles are not captured in the computations. In the proposed SSP-BC tool,
12
estimates of these factors are made from these modal parameters. The SSP-BC Tool also has the advantage of including up to two lanes blockage in the freeway segment. The FSPP and FSPE tools include at most one lane that is blocked. On the other hand, the FSPP and FSPE tools can be used to compute incident response time savings due to a SSP program. As such, they can provide needed input to the SSP-BC tool. The SSP-BC Tool also accounts for secondary incidents that statistically would arise with longer incident durations than can be expected where SSP exist, which FSPE and FSPP tools do not.
13
2
CHAPTER 3. Factors Affecting SSP Benefits and Their Calculation
Numerous methods are practiced by incident program managers for measuring and evaluating SSP program performance. Results of such studies are often used to justify the expenses of these programs, but can also provide insights that can be used to improve performance and ultimately reduce the number and impact of traffic incidents. The first step to evaluate a SSP program is to identify the contributing factors to benefit values as will be discussed in this chapter. In the first Section, MOEs employed in benefit estimation are identified. Factors effecting the estimation of these MOEs are discussed in Section 3.2. Methodologies used for their computation are described in Subsections 3.3-3.5. This is followed by a summary in Section 3.6. 2.1 Measures of Effectiveness (MOEs) A myriad of MOEs may be used in evaluating the benefits of a SSP program. Typical measures include: operational performance measures (e.g. incident response time), traffic performance measures (e.g. travel delay), environmental impacts, safety (secondary incident prevention), reliability, maintainability, and ease of use. In the context of this study, travel delay, fuel consumption, pollutant emissions, and prevention of secondary incidents have been chosen as the MOEs of interest.
2.2
Factors Contributing to Travel Delay and Fuel Consumption Numerous factors, like roadway geometry and weather, affecting MOEs have been
identified in studies on travel delay, fuel consumption and emissions. A comprehensive set of factors has been used in this study for estimating all MOEs; although, some factors are more significant for one MOE than others. For example, roadway grade will have greater impact on fuel consumption than occurrence of secondary incidents. To identify the factors of greatest importance for travel delay estimation, works in the literature were reviewed. The majority of statistical and deterministic queuing methods developed for this application area assume that the most significant factor in reducing incident-induced delays is to reduce incident duration. One of the formulae most widely used to compute travel delay was
14
developed by Wirasinghe (1978) as described in (Qi et al, 2002). For a given roadway segment, Wirasinghe’s formula is given in equation (1). This equation includes factors of incident duration (T), total capacity (S1), traffic demand (S2), and bottleneck capacity (S3). Eq.3.1 In equation 3.1, incident duration and bottleneck capacity are directly related to travel delay resulting from an incident. The number of lanes blocked and severity of the incident is a function of available capacities at bottlenecks at the time of the incident. Incident duration estimation is required within equation 3.1 for travel delay computation. Since 1987 many techniques have been developed to predict incident duration based on collected data. However, the site-specific nature of the collected data has caused disagreement as to the validity of the results. The majority of prior studies have employed different statistical models to estimate incident duration, travel delay, and similar required data to evaluate traffic conditions and the efficiency of an SSP or similar program while they can be applied in studies of roadways with similar traffic characteristics, geometry and weather. These models are neither suitable nor applicable to other regions with different traffic circumstances. A goal of the proposed SSP-BC Tool is, thus, to provide a method to uniformly estimate and compare travel delay savings associated with SSP programs. To address the need for regional or even roadway-specific estimates, microscopic traffic simulation methods are employed. In general microscopic traffic simulation can be used to estimate the consequences of an incident on travel performance. In developing the SSP-BC tool, to produce realistic estimates, its application to a nationwide model, calibrated parameters for typical US highways, comprehensive information and details of the typical incident sites were considered. When considering travel delay, it is important to analyze congestion, incident duration and the causes of both. Geometry of the roadway, traffic characteristics, demand, construction and major maintenance operations, traffic accidents or vehicular breakdowns, and weather conditions are the factors suggested in the Highway Capacity Manual that affect the actual capacity of a highway segment. Table 3.1 shows factors that have been
15
considered in some studies of incident duration and travel delay. In this study, the simulation experiments were designed to account for nearly all of these suggested factors.
Table 2.1 Previous studies on travel delay and incident duration
Authors, year Dependent variable Variables used in the model development Peak-hour Character of incident Severity level Disposal type Number of lanes blocked Personal injuries Response units(fire department, police, FSP) Temporal characteristics Weather (snow, rain) Incident location (with respect to ramps) Incident characteristics Involved vehicle Response source Physical variables: accident time of the day, the day of the week, weather Vehicle variables: number of vehicles, truck involvement, Response variables Number of lanes affected Presence of trucks Time of the day Police response time Weather
Zhao et al. 2010
Incident delay
Boyles wt al. 2006
Incident duration
Qi (2002)
Incident duration
Smith et al. (2001)
Clearance time
Garib et al. (1997)
Incident duration
A basic freeway segment is chosen as the control sample for simulation designs herein as it is in HCM for estimating the capacity of a roadway under different circumstances (HCM, 2010, 10-1). The method of the HCM 2010 assumes that under basic conditions a freeway segment can reach its full capacity. These basic conditions include clear fine weather and visibility, no congestion due to incidents, no work zone activity (short- or long-term), and acceptable pavement conditions which support normal operations. In addition, presumably all the drivers are familiar with the area. Models of typical base geometries and additional estimates to account for special roadway features, such as curvature and weather, were developed. The frequency and
16
severity level of incidents were permitted to vary over time and space. Temporal characteristics include season, day of week, and time of day of an incident. Other attributes, such as location of the incident in the roadway and lane blockage due to incident occurrence, are considered as spatial characteristics. Regardless of the time or location that an incident occurs, as long as there is at least one vehicle traveling behind the incident, the result is capacity and/or speed reduction, which affects time delay. A list of factors that are studied herein is given as follows. ? Geometry of the roadway segment o Segment length o Number of lanes and average lane width o Lateral clearance (shoulder) o Ramps o General terrain o Horizontal curves o Segment gradient ? Traffic Characteristics o FFS o Ramp FFS o Traffic flow rate o Percentages of trucks in traffic flow ? Incident attributes o Incident severity o Incident duration o Average incident duration o Rubbernecking effect ? Weather conditions These factors are discussed in the following subsections. As this study uses a simulation estimation method, The range associated with each chosen factor within the simulation estimation method is based on information from the literature as discussed in proceeding subsections. These factors affect not only travel delay, but fuel consumption
17
and emissions. Some factors affect these latter MOEs directly and, thus, are not only a consequence of added delay. 2.2.1 Geometry of Roadway Segment 2.2.1.1 Segment Length A large number of studies have used simulation to study the impact of traffic incidents on travel delay using a model of a generic fixed-length homogeneous section of roadway. The homogeneity of a Section relates to its traffic, geometry and weather. Chou and Miller-Hooks (2008), Saka et al. (2004), and Hobeika and Dhulipala (2004) examined a 10-mile segment. This length was typically chosen to ensure that the effects of a traffic incident could be entirely captured within the model. A similar length, thus, is used for simulation runs of this study. Greater homogeneity may exist for shorter roadway segments, especially where there is significant horizontal curvature and vertical changes; however, using a shorter length segment can lead to errors in estimates. Additionally, it has been noted that there can be an undesirable increase in accident location reporting errors and other types of errors for shorter roadway segments (Shankar et al., 1994). 2.2.1.2 Number of Lanes and Average Lane Width As the study focuses on freeways, 2 to 6 lane highway segments in each direction are considered. It is assumed that the standard minimum lane width of 12 feet is available based on the default value of lane width of urban and rural highways in HCM 2010. 2.2.1.3 Lateral Clearance (Shoulder) To ensure full operational capacity, basic freeway segments require a minimum 6feet right-side shoulder. As VISSIM does not offer the capability to model shoulders, initial numerical experiment were conducted to investigate VISSIM’s ability to capture the impact of shoulder characteristics. This is done by adding an additional lane that is closed to traffic. It was noted that VISSIM does not capture the effects of shoulder existence. Therefore, no interruption in flow or reduction to capacity due to either shoulder width or closure was modeled. Note that impact of capacity reduction due to incidents occurring in the shoulder has been studied in this effort.
18
2.2.1.4 Ramps Only major junctions were modeled and all on- and off-ramps were assumed to be located at the right-edge of the roadway. According to the HCM 2010, merge and diverge influence areas are 1500 feet downstream from the merge and 1500 feet upstream from diverge points. Thus, route decisions used within VISSIM to model vehicle movements towards off-ramps started from 1500 feet before the off-ramps. Qi (2002) considered the possibility of incorporating relationships between incident and ramp locations in incident duration modeling. Qi was unable to obtain the needed data to ascertain details of this relationship. Moreover, no other studies of this relationship could be found in the literature. Preliminarily experiments were designed here to study this relationship, but based on the results, (Section 4.2.2) only ramp density was used to capture the impact of ramps. Ramp density is defined in HCM to explain the impact of merging and diverging vehicles on the free flow speed and the capacity of the segment. Ramp density is the average number of on-ramps and off-ramps in a 6-mile segment based on the midpoint of the study segment. It varies from 0 (occurs in rural areas) to 6 (possible in dense urban areas) total ramps per mile. The free flow speed of a basic freeway segment is most sensitive to ramp density as discussed in Chapters 10 and 11 of the HCM. The HCM does not discuss the impact of the number of main lanes on ramp operations or the percentage of traffic separating\entering from\to the main lane. However, the HCM (2010), states that on average, an increase of 2 ramps per mile in total ramp density causes approximately 5 miles per hour reduction of speed in the basic segment (FFS of 75 miles per hour having zero ramp density, standard lane width and right-side clearance) , because of approximate linear relation of free flow speed and ramp density. Numerical experiments were made in this study to compare travel delay and fuel consumption due to different exiting flows. 2.2.2 General Terrain There are three categories of general terrain: level, rolling, and mountainous. Level terrain contains any combination of horizontal or vertical alignments that enables heavy vehicles to operate at the same speed as passenger cars. Typically, this terrain contains short grades with a maximum 2% incline/decline. In rolling terrain, there is a combination of vertical or horizontal grades which cause heavy vehicles to operate slightly poorer than
19
passenger cars, but still they have not reached crawl speed (i.e. the maximum constant speed that trucks can maintain on a specific grade over a given length on an uphill stretch). In mountainous terrain, heavy vehicles operate at crawl speed for significant distances or frequent intervals. The impact of horizontal curves and vertical grades were studied separately in this effort. If data pertaining to roadway grade are not available, default values for each terrain are made available in the tool. 2.2.2.1 Horizontal Curves Curvature is a signification factor in the incident duration estimation studies (Gomes et al, 2008). In most of the relevant statistical studies, however, existence of curvature enters the incident duration estimation models as a dummy variable regardless of its degree. The degree of a curve relates to its design speed. The sharper the curve, the slower the design speed. For safety related reasons, posted speed limits on curves are usually lower than the design speed of the curves based on the minimum radii. Super elevation change in horizontal curves of a road segment can vary from 0 percent for areas with severe weather conditions to 8 percent for drier areas and regions with favorable weather conditions. For areas of high-speed, such as freeways, a maximum super elevation of 6% in horizontal curvature is typical and is employed in this study. In the US it is customary that the design speed range for curves be set between 45 and 80 mile per hour (Roadway Design Manual 2010). Shankar et al. (1994) completed one of the few studies that considered this curvature in estimating incident duration. They categorized horizontal curves by their design speeds and determined an explanatory variable for each category. This method, however, requires a level of detail that is not practical for the users of the SSP-BC Tool developed herein. Within the tool, in a segment with free flow speed of 70 mph, thus, curvatures having design speeds of 70 to 75 mph were assumed as straight, 60 to 65 mph as mild and 50 to 55 mph as sharp. To capture this, 5 and 10 mile per hour speed reductions were applied for mild and sharp curvatures, respectively. 2.2.2.2 Segment gradient Roadway gradient is one of the highway-related factors known to significantly affect fuel consumption and emission rates (Park and Rakha, 2005), since vehicles need
20
more power on uphill climbs to maintain their speed and less power in descending downhill. The Roadway Design Manual has 4 percent maximum allowable grade for urban freeways as shown in Table 3.2. However, since SSP programs also operate on mountainous roads with higher grades, the impact of grade in the range of -10% to 10% is considered in this study.
Table 2.2 Maximum grade table (Adopted from roadway design manual 2010,
Functional Classification Type of Terrain 15 20 25 30 35 40 45 50 55 60 65 70 75 80
Urban and Suburban Local Collector Arterial Freeway All Level Rolling Level Rolling Level Rolling <15 9 12 ----<15 9 12 ----<15 9 12 ----<15 9 11 8 9 --<15 9 10 7 8 --<15 9 10 7 8 --<15 8 9 6 7 ---7 8 6 7 4 5 -7 8 5 6 4 5 -----3 4 -----3 4 -----3 4 -----3 4
2.2.3 Traffic Characteristics 2.2.3.1 Free-Flow Speed (FFS) and Roadway Capacity FFS is the most important factor defining the roadway capacity. Theoretically, when the density and flow rate in the segment is zero, vehicles travel with FFS. In practice, it is defined as the desired speed at flow rates between 0 and 1000 passenger cars per hour per lane. Using a systematic simple (e.g every tenth vehicles in each lane and a minimum of 100 vehicles) the mean speed of all passenger cars can be reported as FFS (HCM, 2010). The factors affecting FFS are: lane width, lateral clearance and ramp density, all of which were considered in simulation designs herein under geometric characteristics of the segment.
21
The HCM defines capacity as the “average flow rate across all lanes.” VISSIM does not have direct input for capacity. Consequently, by using the FFS as the initial desired speed, suggested reduced capacities under various circumstances can be modeled. For example, to include ramp density in the model, instead of running all the possible number of ramps, the equivalent reduced speed for capacity reduction percentages due to ramp density provided by the HCM is used. The maximum posted speed limit in US freeways varies from 55 to 80 mph according to Highway Safety Research and Communications (2012). Considering capacities in the HCM 2010 and their relationship with other factors, a 70 mph FFS was employed in the simulation runs herein. This is because speed reduction is generally taken from a base FFS of 70 mph.
Table 2.3 FFS vs. Base capacity for freeways (Adopted from HCM 2010)
FFS (mile/h) 75 70 65 60 55 Base Capacity (pc/h/ln) 2400 2400 2350 2300 2250
The capacity-speed relationship shown in Table 3.3 is based on national norms, but this relationship can change locally. Furthermore, upon review of data from the National Motorist Association it is observed that, in urban facilities, speed limits for passenger cars and trucks are identical. There are roadways in some states such as California, where the speed limits differ by vehicle class. In this study, it was assumed that speed limits are identical for all vehicle types. To consider impact of weather condition on speeds and existence of ramps, the study range of the vehicle speed is chosen to be 35 to 75 mph in this study. 2.2.3.2 Ramp FFS For simulating ramps in VISSIM, their FFS must be set. Typical speeds for ramps are in the range of 20 to 50 mph according to the HCM. 25 and 35 mph were considered in this study. The impact of ramp density was taken into account using base capacity reduction due to increase in number of ramps in a segment. The relationship between base
22
capacity and total ramp density is shown in Figure 3.1, assuming drivers are familiar with the area and no trucks are present (HCM, 2010).
Figure 2.1 Base capacity vs. Total ramp density (adapted from HCM, 2010, P. 10-7)
2420 2400 Base capacity (pc/mi/ln) 2380 2360 2340 2320 2300 2280 2260 2240 0 2 4 6 8 10 Total ramp density (ramps/mi)
2.2.3.3 Traffic Flow Rate (Demand) Traffic volume directly affects level of service of the freeway segment and savings from reduced incident duration. The hourly traffic (in each direction) was used as traffic volume. A range of maximum capacities between 200 and 2200 vehicles per lane per hour (vplph) on average was used for simulation runs. For traffic flows of less than 1000 vplph, vehicular speeds are nearly constant. At a 75 mph FFS, the minimum breakpoint of speed reduction due to flow growths occurs at a volume of 1000 vplph (HCM 2010). Thus, for each case, 11 different traffic volume categories from 200 to 2200 vplph were simulated. 2.2.3.4 Truck Percentages in Traffic Flow Traffic composition, including the percentage of heavy vehicles, is one of the details required to complete an operational analysis of a freeway segment. In addition, the rate of fuel consumption is highly dependent on truck percentages. A range between 0 and 20% trucks is used in developing the SSP-BS Tool. This range was based on information involving truck percentages in specific areas. Specifically, a 3 percent truck composition value was noted for the I-270 freeway in Maryland (Miller-Hooks et al., 2010).
23
2.2.4 Incident Attributes Incident-caused congestion and incident duration are greatly affected by incident severity. As a factor in statistical analyses related to incident duration estimation models, severity is most often noted to be significant. However, SSP program savings are typically derived from more frequent, low severity incidents. 2.2.4.1 Incident Severity In most studies on freeway capacity reduction and incident duration, traffic accidents or vehicular breakdown are modeled based on lane and shoulder blockage (Hadi et al. 2000; Saka et al.,2008; Chou and Miller-Hooks, 2008; Khattak et al., 2010 ; HCM 2010). In this study, shoulder, one lane, and two lanes blockage events are used to model incidents with different severity levels. The greater the severity, the more lanes blocked. 2.2.4.2 Incident Duration The main objectives of SSPs are to identify incidents or other causes of disruption in the traffic stream and minimize incident duration. Thus, a standard method to calculate incident duration is essential for SSP program evaluation. Incident duration is often defined as the time between incident occurrence and when response vehicles leave the scene (Garib et al. 1997, Nam and Mannering, 2000; smith and smith, 2001). The Traffic Incident Management Handbook describes incident duration based on the time required to detect an incident, time from incident report to on-scene response, and time required to clear the incident. A widely used approach to defining incident duration is depicted through timestamps in Figure 3.1.
24
Figure 2.2 Incident timestamp flow chart (Chou and Miller-Hooks, 2008)
As described in (Chou and Miller-Hooks, 2008), at the moment the incident occurs, the driver will call the management center and the “Call Start Timestamp” will be logged. Once confirmation of the incident is received, a SSP vehicle driver will be dispatched to the scene and the “Dispatched Timestamp” is logged. The “Arrival Timestamp” is marked when the patrol unit arrives at the scene. Finally, once the event is cleared, the “End Timestamp” is recorded. Thus, time for confirmation, response time, assistance duration, and incident duration can be determined from the difference between two timestamps as follows: Confirmation time = Dispatch time - Call Start time Response time = Arrival time - Dispatch time Assistance duration = End time - Arrival time Incident duration = End time - Call Start time When an incident is identified by a patrolling vehicle, i.e. is self-initiated, both the time for incident confirmation and response time will be zero. The majority of SSP program responses are made to incidents with durations of 90 minutes or less. SSP programs may also assist state troopers or local police in response to more severe, typically longer, incidents. Although the frequency of these long duration incidents is low (often less than 2%), these incidents cause very significant travel delay. In fact, in some cases, it is necessary for law enforcement to close the entire freeway.
25
Including these extreme cases in the evaluation, study of SSPs often results in an overestimation of the benefits of such programs, particularly since they play a supporting, rather than leading, role in these events. Therefore, incidents with durations greater than 90 minutes were not modeled within this study and benefit estimations can be considered conservative in this respect. 2.2.4.3 Average Incident Duration To quantify the benefits of a SSP program in New York State, Chou and MillerHooks (2008) studied the incident data for the study region. In the first phase of their study, the average incident durations over all incidents arising within a six-month period along four roadways, i.e. I-287, I-684, the Taconic State Parkway and the Sprain Brook Parkway, were calculated. In Table 3.4, a summary of the computed incident durations is shown. Savings from the SSP program can be computed from the difference between the average duration of incidents to which the SSP vehicles responded and those to which it did not.
Table 2.4 Average incident duration (Chou and Miller-Hooks, 2008)
Responds With SSP Without SSP Savings Taconic State Parkway 17.60 39.56 21.96 Sprain Brook Parkway 12.10 21.96 18.12 I-684 26.24 40.61 14.37 I-287 17.70 68.21 50.52
As shown in the table, for those incidents to which the SSP program did not respond, the average incident duration varied between 22 and 68 minutes. The range is smaller where the program is involved in the response. In a study on traffic recovery time, Saka et al. (2004) used VISSIM to simulate incidents that lasted between 10 and 60 minutes. Skabardonis et al. (2005) assumed that the average incident time without SSP response would be 30 minutes. Taken from a report in CHART in 2009 the average incident durations of incidents involving property damage and disabled vehicles were reported to be 33 and 20 minutes, respectively. Therefore, incident durations of 5 to 90 minutes with 5-minute increments were studied for the SSP-BC Tool herein. It is assumed that during this 5 minute time period, estimated savings change linearly.
26
2.2.4.4 Rubbernecking Effect When an incident occurs in a freeway, vehicles in unaffected lanes often reduce their speeds. The effect of these speed reductions is called the rubbernecking effect. Assuming that a warning sign is set up for upstream traffic to inform other drivers of the incident, the rubbernecking effect can be modeled by a reduced speed area in the segment from the warning sign to the incident location. It is assumed in the models developed herein that warning signs are set up 500 feet prior to the incident as is recommended in emergency traffic control guidelines. The effect of rubbernecking is an important piece of an incident to consider. It is often the case that many accidents are caused by drivers looking at other vehicle crashes and other roadside traffic incidents. A 2003 study by the Virginia Commonwealth University’s Transportation Safety Training Center (TSTC) revealed that rubbernecking was the leading cause of vehicle crashes. It accounted for 16 percent of all vehicle crashes. Other distractions arising external to the vehicle, such as the presence of deer, accounted for 35 percent (Masinick and Teng, 2004) of such crashes. 2.2.5 Weather Conditions A growing concern of roadway management agencies is the impact of adverse weather on freeway traffic operations. It is understood that severe weather conditions reduce freeway capacities, but few works have studied its precise impact. In addition, the results obtained from many studies are from outside the US or relate to rural freeway segments within the US. These statistics may not be applicable to urban freeway segments due to different roadway geometries, driver behaviors, and traffic characteristics. In pertinent studies, weather conditions are classified into one of three types: "rain", "snow" and "others" (wind, fog, etc.). Each category is divided in terms of intensity (light versus heavy). The effect of each classification of weather on FFS is summarized in Table 3.5.
27
Table 2.5 Weather-Speed relation
Researcher/year Ibrahim and Hall/ 1994 HCM 2000 Kyte/ 2001 Manish et al. /2005 Rakha et al./ 2007
N/A Not Availble
Light rain 1.2-8 mph 2-14% 15.3 mph 1-5 % N/A
Heavy Rain 3-10 mph 5-17% 15.3 mph 4-7% N/A
Light snow 0.6 mph 8-10% 23-26 mph 3-10 % 13%
Heavy Snow 26.4 mph 30-40% 26.4 mph 11-15% 40%
visibility N/A N/A N/A 6-11% 13%
Wind N/A N/A N/A 1-1.5% 10%
It has been shown that that the impacts of weather on traffic flow and its parameters are dependent on the class of road. Chin et al. (2004) used loop detector data from different regions of the US. These data were linked to different weather. The weather conditions were classified into 6 categories: light rain, heavy rain, light snow, heavy snow, fog, and ice. The impact of each adverse weather condition was then translated into loss of capacity and speed as shown in Table 3.6.
2.6 Speed and capacity reduction based on road type (Chin et al. 2004))
Weather Condition Light rain Heavy rain Light snow Heavy snow Fog Ice Urban Freeway Capacity (%) 4 8 7.5 27.5 6 27.5 Speed (%) 10 16 15 38 13 38 Rural Freeway Capacity (%) 4 10 7.5 27.5 6 27.5 Speed 10 25 15 38 13 38
The impact of weather can be modeled through its effects on speed and capacity reduction. Because VISSIM does not have explicit capacity input, suggested speed reduction along urban freeways due to adverse weather conditions was used for this study. Table 3.6 shows no significant difference between urban and rural freeways except under the condition of heavy rain. To categorize intensity of weather conditions, thresholds have been developed in the HCM as follows. Light rain: perception below 0.25 inch/hour Heavy rain: Perception greater than .25 inch/hour Light snow: perception below 0.5 inch/hour
28
Heavy snow: Perception greater than 0.5 inch/hour In this study, using this knowledge from previous studies, speed reduction on urban freeways was based on the 6 different adverse weather conditions. For this investigation, the levels are selected as: 5% speed reduction due to light rain, 10% for heavy rain, light snow, and low visibility, 15% for fog, 35% for heavy snow, and 40% for icy conditions. Table 3.7 shows suggested speeds under different weather states.
Table 2.7 Actual speed under adverse weather conditions
speed limit 75 70 65 60 55 speed reduction percentage 5 71.25 66.5 61.75 57 52.25 10 67.5 63 58.5 54 49.5 15 63.75 59.5 55.25 51 46.75 35 48.75 45.5 42.25 39 35.75 40 45 42 39 36 33
It is assumed that weather conditions are uniform along a segment. In addition, simulated weather does not change during the simulation. In other words, a specified speed reduction value was used for any simulation run. The focus of this section has been primarily on factors affecting travel delay. However, many of the factors that affect travel delay also directly impact fuel consumption and emissions of air pollutants. For example, grade is a roadway characteristic with its greatest effect on fuel consumption. Travel factors, such as speed, were also found to significantly impact fuel consumption and emissions.
2.3
Calculating Travel Delay A simulation-based evaluation method was developed to estimate travel delay and
input data needed to compute fuel consumption and emissions. The platform employed in this study is PTV America's VISSIM (version 5.3) software, a micro-simulation tool for traffic operations modeling. VISSIM is used to obtain estimates of travel characteristics
29
and other metrics for roadways with operational SSP programs and those without (for comparison) through a host of simulation runs in which solitary incidents are simulated and their effects are estimated. VISSIM computes the travel delay of each vehicle and total travel delay in the network in terms of the average total delay per vehicle (in seconds). Total delay is computed over all vehicles passing through a travel segment. For a given vehicle, its value is determined by subtracting the ideal travel time (assuming FFSs can be maintained) from the realized travel time. 2.4 Calculating Fuel Consumption and Air Pollutants While VISSIM could be applied directly in estimating travel delay, it was found that it could not reliably compute fuel consumption for freeways and no module is available for computing pollutant emissions. PTV offers external (in the form of add-ons) fuel consumption and emission calculation modules for VISSIM. The user manual describes a process in which emissions data can be obtained from node evaluation at a network level. For this study, necessary licenses needed to use the emissions and fuel consumption add-ons were available and several preliminary tests were designed and run in an effort to validate fuel consumption and emissions output from these modules. Unfortunately, the modules did not function as designed. In most runs, fuel consumption and emissions were not reported. Results of those runs for which results were obtained revealed that the fuel economy estimates from VISSIM were over-estimated for US vehicle markets. In fact, an average fuel efficiency of 35 miles per gallon was obtained from the successful runs. Through discussions with the PTV America, Inc. support staff, it was determined that these tools would not be reliable, nor necessarily available, for this study. More generally, the fuel consumption and emissions models were specifically designed for signalized intersections and not freeways. A comprehensive method to calculate fuel consumption and emissions from all vehicle movements in vehicle record output files of VISSIM was adopted in this study. After recording the actual speed, acceleration, and mass of all vehicles, the data was entered into the network at one second intervals during a simulation period. For each run, calculations for the run were made to obtain fuel consumption and emissions. 30
2.4.1 Emissions Calculation Emissions in the transportation sector are primarily due to the combustion of fossil fuels. Carbon dioxide (CO2) and hydrocarbons, such as methane (CH4), are produced from the combustion products of fossil fuels, like petroleum, diesel and biofuels, as a result of the fuel’s high carbon content [USEPA, 2010a]. Nitro gases or NOx emissions are formed when nitrogen (N), either in the air or in fuel, combines with oxygen (O2) at high temperatures. Other pollutants, such as PM and CO, are formed due to incomplete combustion of fuel; whereas, SOx emissions are formed as a result of the sulfur content in the fuel [USEPA, 2009]. CO2 emissions are proportional to the carbon content of the fuel. Logically, this would mean emissions vary by fuel type. These emissions can be calculated using a simple relationship associating the amount of fuel consumed, the carbon content of the fuel (or carbon coefficient) and the fraction oxidized (usually estimated to be 99%) [USEPA, 2006]:
On the other hand, non-CO2 emissions (CH4, NOx, PM, SOx, etc.) are not directly proportional to fuel consumption and are affected by vehicle characteristics. Therefore, to accurately determine the effects, vehicle-specific emission rates/factors (e.g. mass of pollutant/mile) are used in combination with vehicle activity. These factors are a function of vehicle type, age, fuel, and emission control technology. Vehicle activity can be defined by vehicle miles travelled (VMT) or hours of operation, and depend on the units of the emission factor [NCHRP, 2006]. Vehicle activity can be used directly to calculate emissions by either using the vehicle fuel economy (miles per gallon) or fuel-based emission factors (grams per mile).
While these general mathematical relationships are typically used to calculate emissions, the level of accuracy is dependent on the approach used to define emission
31
production processes and number of variables considered in determining emission factors [IPCC, 2006]. With a top-down approach, emission estimates are obtained using aggregate fuel consumption. These estimates can be reasonable when emissions calculation at the macro-scale level is needed (e.g. national emissions inventory). With a bottom-up approach, often used at a micro-scale level (e.g. project level emissions), more detailed inputs for fuel consumption and emission factors are required. It must be noted that while the bottom-up approach may also be used for inventory purposes, obtaining large-scale, detailed data inputs for emissions calculations often proves to be difficult. Both fuel consumption and emission factors are associated with/dependent on several variables which influence the performance of a vehicle and, therefore, the amount of fuel consumed or mass of pollutants emitted. The power-demand of a vehicle is dependent on various inter-linked parameters, such as vehicle characteristics (type, age, mass, etc.), operating mode (start/stop, running, idle and vehicular speed) and environmental parameters (road characteristics, temperature, humidity, etc.). A schematic illustrating the relationships of these factors to emissions is shown in Figure 3.1.
Figure 3.1. Schematic illustrating relationship of variables to vehicular emissions (SHA)
32
As illustrated in the figure, modal and environmental parameters, vehicle and road characteristics, fuel type, and traffic conditions, all affect a vehicle’s power demand, which in turn affects the resulting emissions produced. For example, use of alternative fuels or newer and more efficient vehicles or driving on a relatively flat road would typically produce lower emissions. Also, modal parameters significantly affect emissions. For example, emissions produced when a vehicle is turned off and restarted before the engine has cooled down (hot start) are lower as compared with those when the vehicle is initially turned on (cold start). Depending on traffic conditions and road characteristics, vehicles at higher speeds or accelerating from low speeds produce more emissions as compared to at higher speeds. Moreover, road characteristics (e.g. road grade, number of lanes, the number and type of specific traffic control devices, surface conditions, etc.) also influence the traffic flow and density. For example, large road grade and high traffic volume result in a large number of stops and starts and, therefore, emissions [Bachman, 1997]. The incorporation of these variables in determining emission factors and fuel consumption, that is, working on a micro-scale level, would produce more accurate and realistic results. Of the many models that currently offer emission estimates for on-road vehicles are traffic simulation models, like CORSIM, S-PARAMICS, INTEGRATION, and TRANSIMS. These are some of the most widely used traffic simulation models. These, along with other models, such as DYNASMART-P, represent driver behavior and vehicle kinematics for individual vehicles and trips, and are able to replicate and enable analysis of a variety of traffic-related activities. Many of these traffic simulation models account for environmental impacts of traffic related activities either by using emissions estimation modules integrated within the model or by using external microscopic emissions estimation models as plug-ins to determine emission outputs. The plug-ins provides instantaneous emission rates based on the vehicular inputs from the simulation models. However, a major disadvantage of using these models is that these models use averages for speeds, acceleration, deceleration and fuel consumption in order to generate emissions output. In most cases, the methodology employed to estimate emissions is not described. Therefore, while these models that best simulate traffic have the potential to provide a level of detail required for micro-level emissions estimation, they use a macroscopic or undefined methodology for emissions estimation. 33
Of the external microscopic emissions estimation models (sometimes used as plugins with traffic simulation models), CMEM and MOVES created by University of California-Riverside and USEPA, respectively, are the most notable and comprehensive. These models use a power demand approach to capture the physical processes of emissions production, incorporating a vehicle’s modal parameters, and hence, provide more accurate emission estimates [CMEM, 2010; USEPA, 2011]. Although both models have many benefits and are capable of producing microscopic, modal emission results, neither accounts for several important factors relevant to assessing the effects on emissions produced from traffic conditions and changes to vehicle composition on roadways. Some of the disadvantages of using these models lie in the scope of the variables they cover, level of detail captured in the outputs, and the limited flexibility they offer users. 2.4.2 Fuel Consumption and Emission Calculation Methodology Similar to MOVES and CMEM, a power-based approach was used for this research project to estimate emissions wherein vehicle characteristics and modal parameters, namely vehicle mass, velocity and acceleration, are used to calculate the instantaneous power demand (Pv,t) for a vehicle type category. When combined with the speed-based engine parameters (e.g. K, N, V), this approach provides an instantaneous fuel rate (FRt). The fuel rate is then multiplied by fuel-based emission factors (EF) to produce emission estimates for criteria air pollutants, such as HC, CO, NOx, CO2. Furthermore, the calculated instantaneous fuel rate (FR) when multiplied by the fuel-based sulfur content and other variables provides the associated instantaneous SOx emissions output for the vehicle. The equations and related data used to determine second-by-second emissions output for this research project are described next. The nomenclature used in the Equations (Eq. 3.1) is listed in Table 3.1.
Table 2.8 Nomenclature for variables used in equations LDV LDT PV,t M vt at c1 A : : : : : : : : Light-duty Vehicle (e.g. passenger cars, SUV, etc) Light-duty Trucks Instantaneous Tractive Power of vehicle V at time t (KW) Vehicle mass (metric tonne) Vehicle speed at time t (mph) Vehicle acceleration at time t (mph/s) Conversion factor for speed: 0.447ms-1/mph Rolling resistance coefficient (KW/mps)
34
B C r g FRt HV ? Kt Nt V S g/gtop K0 EMPol EFPol ?Fuel Fuel EconomyLDV/LDT T EMSOX SCFuel
: : : : : : : : : : : : : : : : : : : :
Rotational resistance coefficient (KW/mps2) Aerodynamic drag coefficient (KW/mps3) Road grade i.e. slope (%) Gravitational constant: 9.81 m/s2 Fuel consumption rate (g of fuel/s) Heating Value (KJ/g) Engine Efficiency = 0.4 Engine friction factor at time t (KJ/rev*L) Engine speed at time t (rps) Engine displacement volume (L) Engine Speed to Vehicle Speed Ratio (rpm/mph) Gear ratio 0.22 KJ/rev-litre (average based on range 0.19-0.25 KJ/rev-L) Emission for pollutant (g) Emission factor for pollutant (g/mile) Denisty of fuel (g/gal) Fuel Economy for vehicle category (gal/mile) Total time travelled by vehicle category (s) SOx Emission (g) Sulfur Content of Fuel (ppm)
2.4.3
Power Demand (Pv,t) and Instantaneous Fuel Consumption (FRt) Calculation The power-demand approach breaks down the emissions generation process of a
vehicle into the physical processes of the vehicle’s engine that correspond with vehicle operation and emissions production. As previously discussed vehicle performance during various driving conditions directly contributes to fuel consumption and resulting emissions. For example, vehicle characteristics, like age and engine size, would determine how quickly the vehicle can move in and out of periods of high power demand (e.g. overcoming high gradients or reaching desired speeds by accelerating). Therefore, estimating the physical processes that a vehicle undergoes during operation can provide higher resolution in defining a vehicle’s emissions production process. These processes are best captured by the engine’s tractive power (Pv,t), which in turn is based on the vehicle’s modal parameters (e.g. vt and at) and road characteristics (i.e. r). At a given time t, the instantaneous tractive power is defined as:
[ { ( ? )}] Eq. 3.2
35
The instantaneous modal parameters, speed (vt) and acceleration (at) at time t, were obtained directly from VISSIM outputs (or by other means); whereas, the vehicle parameters, such as mass (M) and the vehicle track road-load coefficients (A, B and C) for each vehicle category were obtained from USEPA’s MOVES model. These values were estimated by USEPA using vehicular data for LDVs and LDTs from inspection and maintenance programs and developing linear models to determine the coefficients from vehicle mass, M [Koupal et al, 2004]. The vehicle parameters used here are listed in Table 3.9.
Table 2.9 Calculation of Road -Load coefficients [Source: USEPA, 2011c]
Vehicle Category LDV(passenger cars) LDT(Trucks, SUVs,etc) LHD<=14K LHD<=19.5K Source mass(metric tons) 1.4788 1.86686 7.64159 6.25047 A(KW/mps) 0.156461 0.22112 0.561933 0.498699 B(KW/mps2) 0.00200193 0.00283757 0 0 C(MW/mps3) 0.000492646 0.000698282 0.00160302 0.00147383
Equ atio n 2.1 Equ atio n 2.2
The vehicle’s power demand directly influences the amount of fuel consumed, and therefore the mass of pollutant produced. The fuel consumption rate (FRt) or the energy used per second to operate the vehicle is a function of engine speed (Nt) and the engine friction factor (Kt), which captures the energy used to overcome engine friction per engine revolution and unit displacement. Both Nt and Kt are dependent on various speed-related vehicle parameters. The simplified equation used to calculate FR appropriate for mesoscale emissions estimation (as required here) was obtained from Barth et al. (2000) and is expressed as:
[ ], Eq. 2.3
where
Eq. 2.4 [ ] Eq. 2.5
36
The values for S (based on vehicle type category) and g/gtop (based on vehicle speed, vt) and fuel based variables (i.e. HV) are recorded in Table A.4 and Table A.3 of Appendix A respectively. 2.4.4 CO2, CO, HC, CO & NOx Emissions Calculation For the purpose of this project, fuel-based emission factors (i.e. mass of pollutant produced per unit of vehicle activity), EFPol, for the LDV and LDT vehicle categories for major fuel types (i.e. gasoline and diesel) were obtained from the USEPA (refer to Table A.2 in Appendix A). These emission factors in combination with other variables specific to the vehicle categories (e.g. fuel economy, time spent on roads, etc) and fuel (i.e. density, EFPol) were then used to calculate the emissions output for each pollutant (EMPol) using Equation 3.6.
( ) { } Eq.3.6
2.4.5 SOx Emissions Calculation The sulfur-content in a fuel affects the amount of SOx emissions produced when fuel is consumed. Therefore, the sulfur content (SCFuel as obtained from Table A.3 in Appendix A) for gasoline and diesel were used to estimate the SOx emissions for a vehicle category using the following relationship:
(
)
Eq. 2.6
2.4.6 Fuel Consumption To estimate the total fuel consumed by a vehicle category due to effects on its modal profile caused by changes within the traffic scenario, the power-demand based FR as calculated previously was used as shown in Equation 3.8.
? ( ), Eq. 2.7
2.4.7 Assumptions While there are many other variables that might have been considered, such as engine speed, air-to-fuel ratio, fuel use and catalyst pass fraction, vehicle emissions are
37
most influenced by engine power and fuel use. Also, since the scale of this project requires meso-scale emission results, and since all other variables require additional detailed vehicle-specific parameters (based on dynamometer measurements of each vehicle type by brand), these variables were not used in determining Pv,t [Barth et al., 2000].
2.5
Secondary Incident Savings An incident is called “secondary” if it is a consequence of a primary incident. The
occurrence of such a secondary incident is related to the duration of a primary incident (Khattak et al., 2008; Zhan et al., 2008). Therefore, as SSP programs aim to decrease the duration of primary incidents, they also decrease the risk of secondary incidents. In fact, it was noted in Karlaftis et al. (1998) that for every minute of additional incident duration, the risk of occurrence of a secondary incident increases by 1.7% in the winter and 3.1% in all other seasons, for an average of 2.8%. They fitted two logistic regression models to primary crashes assisted by SSP vehicles associated with the Hoosier Helper program in Indiana. Crashes within 3 miles upstream and within the clearance time plus15 minutes of a primary crash were classified as secondary. The odds ratio, which measures the strength of connotation between a primary incident characteristic and the probability of secondary incident occurrence, is presented. Odd ratios of clearance time in winter and all other seasons are estimated as 1.018 and 1.032, respectively. In other words, the SSP program could reduce the probability of secondary incident occurrence by 18.5% in the winter and 36.3% in all other seasons per incident to which they respond. The first step in quantifying the savings in secondary incidents is to estimate the number of incidents that are secondary. However, there is little agreement among researchers in terms of the validity of methods aimed at identifying and classifying secondary incidents. The primary approaches to such identification and classification are static threshold, dynamic threshold, and simulation-based filtering methods. Table 3.10 lists the classification methods in the literature. Raub (1997) employed temporal and spatial thresholds to classify incidents. Any incident that occurs within 15 minutes of the resolution of another incident and within 1
38
mile is defined as a secondary incident. Applying this method, 15-percent of all incidents reported by police were found to be secondary. Other studies used similar fixed thresholds,. For example, Moore et al. (2004) defined incidents as secondary if they occurred within 2 hours and 2 miles from incident identification. This static method is also adopted by Karlaftis et al. (1999), Hirunyanitiwattana et al. (2006) and Zhan et al. (2008). One drawback of the static threshold method is that it cannot capture field conditions (i.e. changing demand), and therefore, leads to misclassification. The dynamic threshold method was developed to compensate for the shortcomings of the static approach. Sun et al. (2007, 2010) proposed a master incident progression curve to identify secondary incidents. The progression curve is constructed from affected distance, which is measured from incident location to the end of its queue. Instead of using a static maximum or average queue length, the author marked the end of varying queue throughout the entire incident duration. Incidents that fall under the curve are considered to be secondary. It was concluded that the method reduced Type I errors by 24.38-percent and Type II errors by 3.13-percent. Similarly, Zhang and Khattak (2010) employed a dynamic queue-based method in which queue length is calculated by a deterministic queuing model (D/D/1). Zhan et al. (2009) classified secondary incidents as those that occur within the boundary of the estimated maximum queue length and dissipation time of a lane-blocked primary incident. To arrive at this conclusion they used a cumulative arrival and departure traffic delay model. A simulation-based secondary incident filtering (SBSIF) method was proposed by Chou and Miller-Hooks (2008). This model used geometric boundaries to analyze the incident impact area in a time-space contour map of traffic speeds. Regression models are established for identifying the corner points of the impact area. The authors conclude that the SBSIF method can reduce the misclassifications by up to 58-percent. They noted that 4% of 693 incidents to which H.E.L.P. responded to were secondary using this approach. Regardless of the method used in distinguishing secondary incidents from primary incidents, once the number of secondary incidents as a fraction of primary incidents is known, the next step is to estimate savings in secondary incidents, i.e. the number of incidents that did not arise secondary to a primary incident as a result of reduction in 39
incident duration. Chou and Miller-Hooks (2008) assumed the number of secondary incidents without SSP is linearly correlated with the total delay ratio between with and without SSP response cases. Similarly, Guin et al. (2007) employed an equation based on the ratio of average incident duration of SSP versus non-SSP incident responded cases. Table 3.10 provides an overview of secondary incident methodologies, assumptions and assumed rates from the literature. In general, Federal Highway Administration (FHWA) noted that approximately 20% of all incidents are secondary incidents. They include not only crashes, but engine stalls, overheating, and running out of gas as types of secondary incidents.
40
Table 2.10 Secondary incident classification methods
Data Sample Size
Authors (Year)
Definition
Method
Claimed Percentage/Main Finding
Data Source
Data Time and Location
Raub (1997)
Occur within 15 min from incident resolution and within 1 mile
Fixed temporal and spatial parameters
>15% of all incidents reported, average secondary crash occurs within 36.4 min, 600 meters after primary accident, average primary accident duration is 45 min, added of delay 69 min
1,796
Police reports
1995 Northern Chicago, IL
Karlaftis et al. (1999)
Occur within 15 min from incident resolution and within 1 mile
Fixed temporal and spatial parameters
>15% of all incidents reported
Indiana DOT
1992-1995, Borman Expressway
Moore et al. (2004)
Occur within 2 hrs and 2 miles from incident identification
Fixed temporal and spatial parameters
Secondary accidents are considerably rarer events than previous studies suggest, lower frequency of secondary crashes, 1.5%~3% per primary incident
84,684
CA Highway patrol
March, May, July 1999 and Dec.1998, LA Freeway
Hirunyanitiwattana et al. (2006)
Occur within 1 hrs and 2 miles from incident identification
Fixed temporal and spatial parameters
Secondary crashes occur more often during rush hour traffic in the morning and evening, rear-end collision is the predominant secondary collision type, accounting for 2/3 of all secondary crashes.
70,6 in 1999 and 183,988 in 2000
FHWA
1999 and 2000, CA highway system
41
Sun (2007,2010)
Occur under a master incident progression curve
Dynamic threshold method by marking the end of varying queue throughout the entire incident
dynamic method reduce Type I error by 24.38% and Type II by 3.13%; Results from using dynamic method versUSstatic method can differ by more than 30% in identifying secondary incidents
5,514
Highway patrol in St. Louis, MO
2002, I-70, MO
Zhan et al. (2008)
Occur within 2 miles and 15 min from incident resolution
Fixed temporal and spatial parameters
Average rate of 7.94% as primary incidents, 5.22% as secondary crashes, secondary crashes are usually much less severe than other crashes. Traveler sight conditions and lane blockage durations of primary incidents are significant contributing factors for determining the severity of secondary crashes
4,435
SMART database of FDOT
Jan 2005-Jan 2007 Fort Lauderdale, FL
Zhan et al. (2009)
Occur within the boundary of estimated maximum queue length and dissipation time of the potential lane-blockage primary incident
Cumulative arrival and departure traffic delay model to Estimate maximum queue length and associated queue recovery time
5% as primary incidents , 3.23% as secondary crash, accidents occur in daytime and with long lane-blockage duration increase the possibility of secondary crashes
4,435
SMART database of FDOT
Jan 2005-Jan 2007 Fort Lauderdale, FL
42
Zhang and Khattak (2010)
Occur within the a queue associated with primary incident and duration of primary incident, contained event duration: durations of all secondary incidents are contained within primary incident duration, extended event duration: duration of one or more secondary incidents partially overlaps with primary incident duration but extend beyond it
Dynamic queuebased method which queue length is calculated through a deterministic queuing model(D/D/1)
Contained and extended events show different characteristics and operational response patterns. Factors associated with durations of longer cascading events include primary incident being a crash, secondary incident being crash, multiple vehicles involved in secondary incidents, and longer time gap between primary and secondary incidents
37,379
Traffic Operations Center, VDOT
2005, Hampton Roads
Zhang and Khattak (2010)
Occur within a queue associated with primary incident and duration of primary incident
Dynamic queuebased method which queue length is calculated through a deterministic queuing model(D/D/1)
96.93% independent incidents with average of 14 min primary incident duration, 2.7% primary-secondary pairs of 40 min duration of primary incident, 0.37% primary-multiple secondary events of 68 min primary incident duration, characteristics of primary incident including crash, long duration, multiplevehicle involvement and lane blockage and road geometric variable increase secondary incident frequency
37,379
Traffic Operations Center, VDOT
2005, Hampton Roads, VA
Chou (2010)
Occur within the incident impact area on time-space traffic speed contour map
Simulation based secondary incident filtering method
24 and 27 out of 630 potential secondary incidents are identified employing visual and regression implementations for corner point identification with SBSIF method
693
New York DOT
2006, I-287, NY
43
2.6
Summary In this chapter, the essential measures required to evaluate the benefits of a SSP
program are identified. Factors affecting travel delay, fuel consumption, and pollutant emissions were reviewed. Assumptions and computational methodologies for computing the MOEs were also introduced. In the next chapter, the importance and ability to experimentally capture the effects of each factor in MOE estimation is studied.
44
3
CHAPTER 4. Implementation of Variables in Simulation Designs
The SSP-BC Tool must be comprehensive to enable the estimation of program benefits under all possible incident scenarios for which any user may require results. A simulation-based method was developed to estimate travel delay and fuel consumption of an incident scenario. Travel delay is a direct output of the simulation software, but fuel consumption is derived from modal parameters related to vehicular movement details, obtained from the simulation (e.g., velocity, acceleration, mass, etc.). Savings in emissions are computed from total fuel consumption. Savings calculation in secondary incidents partially benefits from obtained travel delay values. Any given incident falls into one combination of the identified factors as each incident is defined by its location, lanes blocked, and duration, as well as by the settings related to roadway grade, weather and other contributing factors discussed in Chapter 3. To account for all combinations of these factors related to incidents on freeways with between 2 and 6 lanes, up to 2 lanes blocked (no road closure considered), and clearance times of less than 90 minutes, i.e. all combinations within the study scope, combinations over 8,000,000 simulation runs would need to be conducted. Additionally, some of the factors cannot be controlled through code and thus, automating all of the runs through batch runs is not possible. Even using a powerful computer and even if all runs could be automated; it would take more than a year to complete. This means that simulating all incident scenarios is impractical. Instead, simulation runs are supplemented through statistical approaches described in this chapter. A set of preliminary experiments were conducted to gain insight into the impact of each factor on travel delay and fuel consumption and to choose appropriate statistical estimation (i.e. supplemental) models to employ and methodologies for their calibration. A control case was designed and simulation runs were taken in which the state of a single factor was changed univeriately. These experiments were designed to answer two questions: (1) Does the simulation technique replicate the factors appropriately in the simulation environment? (2) What form should the statistical model take and it include?
45
Multiple-regression modeling was chosen to estimate travel delay and fuel consumption of a given incident from a sample of the possible incident characteristic combinations (i.e. incident scenarios). Results of this initial set of experiments showed that it is best to develop different regression models for shoulder, one- and two-lane blockage scenarios. Moreover, travel delay for LDVs and LDTs must be computed separately, because the monetary benefits associated with savings incurred by trucks is incorporated within the truck driver’s hourly costs. Therefore, for travel delay, one multi-regression model was developed for each of the six categories of incidents and vehicle types as shown in Figure 4.1. Based on results of additional preliminary experiments to assess the impact of the various factors on fuel consumption estimation, one general model was calibrated to account for all incident scenarios.
Figure 3.1 Estimation model development progress
The goodness-of-fit and modeling assumptions were tested. Results from these tests indicated that the developed models needed improvement. Thus, a hybrid statisticalsimulation data approach was introduced for improving the travel delay regression models,
46
and a non-linear regression model was calibrated to improve the goodness-of-fit of the fuel consumption estimation model. This progression is also shown in Figure 4.1. The simulation platform, general settings and incident simulation techniques are described in Section 4.1. Results of single-factor experiments are described in Section 4.2. Regression models and investigation of their goodness-of-fit are presented in Section 4.3. This is followed by a summary in Section 4.4. 3.1 General Simulation Settings and Incident Modeling While many high quality microscopic simulation software programs, like PARMICS, VISSIM and CORSIM, areavailable that for many purposes adequately model traffic (Brockfield et al., 2004; Ranjitkar et al., 2004; Jones et al., 2004; Bloomberg et al., 2003), some studies have revealed that some platforms, like VISSIM and PARAMICS are better than others, specifically CORSIM (Choa et al., 2002). Given positive experience with VISSIM in the literature, as well as prior experience, including extensive calibration studies (Miller-Hooks et al, 2010), by the authors with VISSIM, VISSM was chosen as the simulation modeling platform. . An additional benefit of working with VISSIM is its COM interface. The interface provides great flexibility in controlling various aspects of the simulation environment. Primary calibrated VISSIM parameters, assumptions, and method for modeling incidents are described in Section 4.1. In Section 4.2, methods of implementation for each of the factors in the simulation environment and their impacts on travel delay and fuel consumption given by results of numerical experiments on individual factors are explained. Results of these experiments are combined to reproduce more realistic incident scenarios in which multiple factors change concurrently. Multiple-regression modeling is employed to estimate travel delay and fuel consumption as explained in Section 4.4. This is followed by an improved estimation model for travel delay exhibited in Section 4.5 and a summary of the Chapter in Section 4.6. 3.1.1 Simulation Settings For each incident scenario and seed, one run of VISSIM involves 7,200 seconds of simulation time. A typical incident scenario is explained in next section and shown in
47
Figure 4.1. The software user manual suggests the use of a warm-up period. This period includes the first 1,800 seconds of each run; this period is also required to achieve a steady traffic flow along the segment from the start of the analysis period. Incidents are designed to occur 300 seconds after the end of the warm-up period or 2,100 seconds into the simulation. To get more accurate results, the VISSIM user manual suggests running a minimum of three runs with different random seeds for each simulation model and reporting the average over random seeds in the final results. Average results, based on the
5,400 seconds of simulation run time, over runs with five randomly chosen seeds were collected. The software user manual suggests 1 time step per simulation second in terms of
simulation resolution for models that contain only vehicles (i.e. that are not multimodal). However, five time steps per simulation second was used in this study. It was observed in preliminary runs that in modeling congested roadway conditions, using a higher resolution reduces loss of vehicles due to difficulty by the vehicles in entering the network. Each run required approximately 2.5 minutes on a Dell Precision T7500 personal computer with a 3.20 gigahertz quad core processor and 12 gigabytes of RAM, running a 64 bit Windows 7 operating system. 3.1.2 Control Case A typical incident to which SSP vehicles responded was selected and designed based on the information in Chapters 2 and 3 to be the control case incident for the singlefactor experiment runs. The studied segment is a three lane, 10-mile, unidirectional straight freeway segment with no on-ramps, off-ramps, grade, or lane drops as shown in Figure 4.1. The incident duration is 20 minutes (described in Section 4.1.3) which would not change with experiments, the traffic volume is set to be 1200 vplph. The FFS of vehicles is set to be 75 mph. In each experiment, all characteristics are set as base except for the factor under study. 3.1.3 Simulating Incidents and Rubbernecking Effect The COM interface of VISSIM allows users to control various aspects of the simulation, which makes the software highly flexible. However, VISSIM and its COM interface do not include a specific feature for modeling traffic incidents. Simulation-based
48
traffic studies relying on the VISSIM platform have used a variety of methods to model incidents, including: defining a parking lot with one space and assigning a car to park in the space for a fixed time period (Wang et al. ,2003; Pulugurtha et al. , 2002), setting a bus stop at the incident location in which the bus stops for a fixed period of time (Hadi et al.,2000), using active traffic signal control in the affected lanes for the duration of an incident (Saka et al. 2004) and setting a passenger vehicle with speed of zero in the incident location from the start time of the incident to the end of clearance time (Chou and Miller-Hooks, 2010). In this study, two cars with zero speed are co-located at the incident location (lane, and location with respect to ramp) from the start time of the incident for a pre-set incident duration time. VISSIM provides a function that allows users to set a temporary reduced speed area on a link. Such reduced speed areas can capture the effects of rubbernecking during an incident. That is, they can be used to model the reduction in speed in unblocked lanes during the incident time period. Hadi et al. (2000) found that a speed of 20 mph for vehicles modeled in VISSIM in which an incident is active results in the suggested available capacity by the HCM. At the end of the incident period, vehicles involved in the incident are removed from the blocked lanes. Once the reduced speed areas become inactive, vehicles traveling in the affected lanes accelerate until they reach their initial desired speed (Figure 4.2).
49
Figure 3.2 Incident layout on typical three-lane unidirectional freeway segment
3.1.4 VISSIM Calibration To correctly predict system response, it is essential to calibrate the simulation software to existing traffic conditions. Miller-Hooks et al. (2010) identified five carfollowing and lane-changing parameters in VISSIM that had very significant effect on travel delay estimation. After completing an extensive effort to calibrate a model of a 41mile Maryland freeway (82 miles in both directions) against actual travel time measurements, they suggested changes to four of the five values. The suggested parameter settings are: ‘Following’ Variation (CC2), ‘Following’ Thresholds (CC4&5), Safety Distance Reduced Factor (SDRF), and Look Back Distance (LBD). Their definitions, default values, possible range in VISSIM, and the final set that are used in this study are listed in Table 4.1. CC2 to CC5 belong to the Wiedmann 99 car following model, which is
50
mainly suitable for interurban and freeways. SDRF and LBD are lane-changing parameters associated with driver behavior.
Table 3.1 Driver behavior parameters, adopted from Miller-Hooks et al. (2010)
Parameter CC2 CC4&5 SDRF LBD Definition Following variation: desired safety following distance Lower & Upper following threshold Safety distance reduced factor: effects safety distance during lane changing Look back distance: defines the distance at which vehicles will begin to attempt to change lanes Default Value 4 meters 0.35 mph 0.6 200 meters Range 1.5-20 meters (16.40-65.62 ft) 0.1-2.0 0.1-0.9 50-1000 meters Final 39.37 (ft) 0.1 0.1 3280.83
Note: the sign of Lower following threshold (CC4) is ‘-’ and the sign of Upper following threshold (CC5) is ‘+’. 3.2 Single-Factor Experiments Finding a way to model significant factors of travel delay and fuel consumption in VISSIM simulation environment and monitoring their impact on travel delay and fuel consumption are discussed in this subsection. Results of this section helped to choose the appropriate explanatory variables for estimating travel delay and fuel consumption. When a factor was identified as insignificant in the relevant travel delay or fuel consumption estimation model, it was not included in the estimation model or the approach used to capture the impact of that factor in the simulation environment was changed. Factors were designed to change univariantly in the control case in various simulation runs. Results in terms of travel delay and fuel consumption estimates were analyzed for a set of experiments associated with each factor. A summary of the studied factors and ranges on their values is presented in Table 4.2.
51
Table 3.2 Summary of variables used in numerical experiments
General Attributes Factors Segment length Number of lanes and average lane width Geometry of the roadway segment Lateral clearance (shoulder) Ramps Horizontal curves Segment gradient FFS Traffic characteristics Ramp FFS Traffic flow rate Percentages of trucks in traffic flow Incident severity Incident attributes Weather conditions Average incident duration Rubbernecking effect Rang used 10 mile 2-6 lane, 12 feet 6 feet 0 to 10 ramp/mile Straight, Mild, Sharp -10 to +10 percent 55 to 75 mph 25 and 35 mph 200 to 2200 vplph 0 to 18 percent Shoulder, 1-lane and 2-lane Blockage 0 to 90 minutes (5-minute increment) 500 feet Upstream of Incident Location
Clear, Light Rain, Heavy Rain, Snow, Fog, Icy condition, Low Visibility, Wind
3.2.1 Geometry Factors: Number of Lane and Lane Blockages To describe the number of main lanes that are blocked due to an incident, average travel delay per vehicle is calculated for possible combinations of number of lanes and lane blockage within the study range (Table 4.3). Total travel delay divided by total number of vehicles gives average travel delay per vehicle in a simulation run. Table 4.3 indicates a decrease in the average travel delay per vehicle when the number of lanes in a segment increases. The impact of one lane closure on travel delay is much higher in freeways with fewer of lanes, but the difference between travel delay of one lane and two lanes blockage scenarios was not large. One closed lane due to an incident results in a travel delay increase of approximately 13 times that in a two-lane freeway and 5 times that for a six-lane freeway. However, when two lanes are closed due to an incident, travel delay is approximately two times larger compare to one lane blockage for a comparable roadway segment. It is shown in Table 4.3 that fuel consumption is not very sensitive to the number of lanes blocked. This is because the relationship between fuel consumption and vehicular speed has a parabolic shape. Fuel consumption at lower speeds in a stop-and-start mode due to congestion is nearly the same as at higher speeds (e.g. an average speed of 70 mph) with no stops and starts. Therefore, number of lanes blocked due
52
to incident can be eliminated from the fuel consumption estimation model. An approximate trend for travel delay versus fuel consumption is shown in Figure 4.4.
Table 3.3 Number of lanes and lane blockage analysis
Travel delay (hour/vehicle) state of factor two lane three lane four lane five lane six lane
N/A: Not Applicable
Fuel consumption (gallon) 2-lane blockage N/A 240.76 199.53 129.87 109.45 shoulder blockage 307.5198 275.2585 614.0988 775.1365 925.1616 1-lane blockage 305.4774 331.2327 613.5244 777.0599 923.8499 2-lane blockage N/A 319.7124 614.5162 781.0039 933.9603
shoulder blockage 7.44 10.12 13.94 14.93 17.35
1-lane blockage 100.23 110.34 104.45 88.543 77.84
Figure 3.3 Number of lanes and lane blockage analysis
53
A linear trend can be seen in both travel delay and fuel consumption graphs given in Figure 4.3. However, travel delay is greatly affected by lane blockage. Therefore, it was necessary to develop separate estimation models for travel delay for different lane blockage incident scenarios. 3.2.2 Geometry Factors: Ramps The FFS along the ramps, exiting traffic flow and location of incident with respect to ramp locations are important aspects that have been considered in simulating incidents in close proximity to ramps. The impact of incident-ramp proximity on travel delay and fuel consumption is studied. The simulation roadway segment is set to have one off-ramp. (Figure 4.2 compare to Figure 4.1). Two ramp FFSs (25 and 35 mph) are tested. Additionally, two exiting flow percentages (25 and 50) of the main lane traffic volumes are considered. Studies have shown that ramps are significant sources of bottlenecks in freeway operations, affecting traffic as far as a quarter-mile upstream and downstream of merge and diverge points (Zhang et all, 2009). Five incident locations, each set within a half-mile of the off-ramp location, are modeled as depicted in Figure4.4. The average travel delays and fuel consumption over 5 random seeds for each case are reported in Table 4.4.
Figure 3.4 Simulating incidents close to an off-ramp
The distance needed for a vehicle to start a diverging maneuver is set as the number of lanes the vehicle must pass multiplied by the look-back distance. For example, if the look-back distance is set as 200 feet, a vehicle in lane three will start changing lanes to reach the right-most lane beginning from 400 feet upstream of the off-ramp.
54
Table 3.4 Ramp analysis
State of the factor (ramp speed-exiting volume) 25mph- 25% 25mph- 50% 35mph- 25% 35mph- 50% base case Travel delay (vehicle-hour) Passenger cars 13.106 13.106 10.484 10.484 10.295 Trucks 0.180 0.180 0.131 0.131 0.116 Fuel consumption (gal) Passenger cars 613.1 613.3 614.2 614.1 614.1 Trucks 15.4 15.7 15.6 15.7 15.8
As anticipated, the lower the ramp FFS, the higher the travel delays incurred and greater the fuel consumed. However, even for arising incidents close to the ramp with design speed of 35 mph and the main lane FFS 70 mph, the ramp does not affect travel delay significantly. However, fuel consumption decrease since portion of vehicles moves slower nearer to ramp and vehicles generally have higher fuel efficacy in the range of 30 to 50 mph range than at 70 mph. To include the incident-ramp proximity impact on travel delay and fuel consumption in SSP-BC Tool, the speed design of every ramp on the study segment should be included as input. However, the information does not make significant difference in travel delay and fuel consumption outputs. A more practical way to include the impact of ramps, therefore, was selected for the SSP-BC Tool as explain in Sections 3.2.1.4 and 3.2.3.2 in which capacity reduction due to ramp density is applied. 3.2.3 General Terrain: Horizontal Curves Given a specific number of points, VISSIM provides an option to draw a Bezier curve when creating connector links. To assess whether or not the software adequately captures the effects of curvature in freeway operations on speed, preliminary simulation runs were conducted in which the travel time of vehicles traversing two similar segments, one curvy and one straight, were compared. No significant difference was noted between travel times in these runs. Thus, it was concluded that the operational effects of roadway curvature are not captured. To capture these effects, reduced speed areas and lower approaching desired speeds were used within curved areas of the test segment. This is consistent with freeways where speed limits are reduced around curvy roadway segments.
55
1000 feet of the base segment is modeled with curvature. Curves with three design speed were considered. The effect of the curvature is captured by setting speeds of 65, 55, and 50 mph along the 1000-foot length of the 70 mph segment. Incidents were placed randomly in different positions within the curve. Results of test cases were not significantly different from the base case in lower speed categories. Therefore, for roads with mild and sharp curves, 5 mph and 10 mph speed reduction respectively were applied to posted speed limits if the exact speed limit of the segment close to a curve is not available in detail. 3.2.4 General Terrain: Vertical Curves The gradient of the segment can be set manually in VISSIM. According to HCM 2010, the maximum grade in level terrain is 2% (-2%), in rolling terrain is 5% (-5%), and in mountainous terrain is 10% (-10%) as discussed in Section 3.2.2.2. The base case has 0% gradient. The results for different grades are shown in Table 4.5. Travel delay increases significantly as the gradient of the segment increases. Negative grades (downhill) do not have a significant impact on travel delay. The impact of grade on average vehicular speed can be discerned from results given in Table 4.6. The average speed of trucks is more affected by grade. The difference between average speeds of passenger vehicles and trucks widens as the grade increases.
Table 3.5 Segment grade analysis
percentage of grade -5% 5% 10% base case Travel delay (vehicle-hour) Passenger cars Trucks 10.5 0.1 26.1 0.2 83.6 5.3 10.295 0.116 Fuel consumption (gal) Passenger cars Trucks 515.5 12.9 397.4 9.6 289.6 14.5 614.1 15.8
Table 3.6 Impact of gradient on average speed
Average Speed (mph) percentage of grade -5% 5% 10% base case Passenger cars 61 52 37 60 Trucks 52 35 23 51
56
Fuel consumption reduces with reduced speed resulting from higher grades. The VISSIM manual explains that the impact of gradient on traffic flow is found in acceleration and deceleration of vehicles, “The possible acceleration decreases by 0.1 m/s² per percent of positive gradient (road incline).” To capture the effects of gradient and not changes in speed, on fuel consumption, all other factors, including speed, remain constant in this set of experiments. Thus, for different gradient settings, a constant speed is forced. This permits analysis of travel delay fuel consumption and emissions changes due to changes in gradient. Results are provided in Table 4.7. To maintain a constant speed along a gradient, a significant increase in fuel consumption rate is required.
Table 3.7 Fuel economy changes due segment grade
Fuel Economy (mpg) percentage of grade 0 5% 10% Passenger cars 22.06 16.18 11.50 Trucks 21.34 16.27 10.09
3.2.5 Traffic Characteristics: Speed of Vehicles In the simulation runs, the speed of vehicles is defined by the desired speed. VISSIM has a speed distribution for desired speeds from which it assigns a speed to every entering vehicle. If the speed of any vehicle for any reason (e.s. lane blockage) changes, vehicles reach their assigned desired speed after passing the obstacle. Four different speed regimes from 37.5 mph (60kmh) to 75 mph (120 kmh) were modeled. It was found that fuel consumption is highly sensitive to desired speed. In fact almost 50% reduction in fuel consumption was noted when desired speed changes from 120 to 60 kmh. Travel delay, however, slightly decreased with decreasing desired speed.
Table 3.8 Travel delay and fuel consumption changes by FFS
FFS (kmh) 60.00 80.00 100.00 Base Case (120) Travel delay (vehicle-hour) Passenger cars 7.66 8.16 8.42 10.295 Trucks 1.64 0.4 0.14 0.116 Fuel consumption (gal) Passenger cars 257.82 306.46 397.98 614.1 Trucks 9.36 11.04 12.54 15.8
57
3.2.6 Traffic Characteristics: Demand Flow Rate The average speed of vehicles in incident cases with prevailing traffic volumes prior to incidents between 200 and 2200 vplph, were obtained through simulation runs, results from which are shown in Figure 4.5. The sold line depicts the HCM-suggested speed for a basic freeway segment with FFS of 70 mph. The dash-square line Indicates VISSIM average speed, obtained from this study for the base segment under different flow rates. The dash-triangle shows average speed over the segment for an incident with 90minute duration. The dot line indicates the boundary of Level of Service E and F when density passes 45 passenger car per mile per hour (pc/mi/hr).
Figure 3.5 Average speed Vs. Traffic flow rate
80
70 60 Speed (mi/hr) 50 40 30 20 10 0 0 500 1000 1500 2000 2500 Flow rate (vplph) HCM 2010 45 pc/mi/hr no incident 90mins incident
From the 90-minute incident line, the average speed reported for the study segment with incident at a density of 45 pc/mi/hr, where 90-minute incident line passes the dot line, is approximately 35 mph. The HCM suggests a 15% capacity reduction due to shoulder incidents in a 4-lane freeway. It, also, provides adjusted speed-flow curves for indicated capacity reductions (i.e. due to incidents, Exhibit 10-9, HCM 2010). From the mentioned graph in HCM, the average vehicular speed under 15% capacity reduction in a basic segment at 45 pc/mi/hr is approximately 40 mph (FFS of 70 mph, capacity of 2400 pcplph). This indicates that for the simulated segment, capacities associated with lower 58
traffic flow rates are reached in comparison to ideal conditions in a basic freeway segment suggested in the HCM. Furthermore, from the “no incident” line in Figure 4.3, the capacity of the simulated segment closes in on 2000 vplph as average speeds for higher traffic flows become almost constant from this flow rate. Also, after 2000 vplph, the travel times don’t change significantly as traffic flow increases (Figure 4.6). Having 2000 vplph capacity in mind for the study segment under normal conditions, the corresponding average speed of vehicles to the capacity given a 15% capacity reduction in case of incident must be in the range of 35 to 38 mph. 35 mph value has been found from simulated incident case.
Figure 3.6 Travel time vs. Traffic flow
400 350 Total travel time (Sec) 300 250 200 no incident 90mins incident
150
100 50 0 0 500 1000 1500 2000 2500 Flow rate (pc/h/ln)
Findings reported in Figure 4.5 also indicate lower average speeds resulting from the simulation runs under lower traffic volumes as compared with expectations given in the HCM suggested values because the gap between previously mentioned adjusted speed-flow curve and actual speed-flow curves is smaller as low flow rates (Exhibit 10-9, HCM 2010). From Section 4.2.5, it was found that travel delay slightly decreased with desired speed. Thus, if maintaining the suggested speeds of at in lower flow rates, higher travel delay as HCM suggested values might be obtained assumed the capacity to be 2400 vplph for the basic segment.
59
3.2.7 Traffic Characteristics: Truck Percentages To test the impact of trucks on the operation of vehicles in the segment seven traffic composition cases with 0 to 20 percent truck traffic were tested. One-lane blockage incidents were modeled instead of shoulder blockage incidents to capture the impact of vehicle maneuvers. Maneuvers of trucks differ from those passenger cars and, therefore, it affects travel time and fuel consumption differently. Higher Percentages of trucks in traffic flow increase the travel delay and fuel consumption as expected (Table 4.9).
Table 3.9 Truck percentage analysis
percentage of trucks 0.00 0.05 0.10 0.12 0.15 0.17 0.20 Travel delay (vehicle-hour) Passenger cars 171.0673 174.5893 166.5927 160.7813 149.6053 144.7567 140.1433 Trucks 0 13.296 25.51267 30.78933 38.094 42.16 47.52067 SUM 171.0673 187.8853 192.1053 191.5707 187.6993 186.9167 342.1347 Fuel consumption (gal) Passenger cars 137.5791 134.7476 123.1948 119.4744 112.8164 109.7824 105.1306 Trucks 0 13.50501 25.98375 31.20691 37.56797 42.52419 49.08134 SUM 137.5791 148.2526 149.1785 150.6813 150.3844 152.3066 154.212
3.2.8 Simultaneous Changes of Factors In the prior section, the effects of individual factors on travel delay and fuel consumption in the presence of incidents was studied. In reality, multiple factors will exist that will simultaneous the impact these measures under such incident conditions and their effects are often nonadditive. In brief, the factors that directly were considered in the travel delay and fuel consumption estimation models are: Incident duration, number of lanes, number of lanes blocked, prevailing traffic volume, FFS, percentage of trucks and gradient as determined in Sections 4.2.1 to 4.2.7. No prior published study could be found that described a relationship between truck composition and/or roadway gradient with travel delay. Thus, additional analysis to test the independence of these factors and their impact on travel delay was completed. Simulation runs in which the number of lanes, truck percentage and/or segment gradient change concurrently were conducted. Results of these additional simulation runs indicated a constant increase in travel delay due to increased truck composition regardless of the
60
number of lanes. The same pattern was found for segment gradient. These results infer that additional delay due to percentage of trucks and grade change on the three-lane freeway test segment can be added directly to estimate for segments with any number of lanes. However, this was not the case for fuel consumption. 3.3 Multiple-Regression Analysis Multiple-regression relates two or more independent variables ( ) to a dependent variable (Y). Seven multiple-regression models are presented for travel delay and fuel consumption of cars and trucks of different lane blockage incident scenarios based on a design sample of incidents. The general form of multiple regressions is shown in Equation 4.1 where dependent and independent variables are i dimensional vectors. The parameters were estimated using a least squares method.
Eq. 3.1
To obtain the travel delay estimation models, the regression models were developed for different lane blockage scenarios as discussed in Section 4.2.1. In each category of lane blockage, two models are presented for light-duty and heavy-duty vehicles. Explanatory variables of the travel delay regression model were chosen to be incident duration, traffic volume, percentage of trucks and gradient of the roadway. The only parameter found insignificant in the conducted preliminarily experiments for fuel consumption was number of blocked lanes due to incidents. Thus, explanatory variables of fuel consumption regression model are: number of lanes in the segment, incident duration, traffic volume, , speed, percentage of trucks and gradient. 3.3.1 Minimum Sample Size A balance between accuracy and computation time must be chosen in selecting an appropriate sample size of incident scenarios for simulation runs. The larger the sample size, the more accurate the model and the better the estimation of parameters, but the greater the computational effort. Determining the minimum sample size of incidents, thus, is necessary. For this study, the population means (µ) method is employed to determine the minimum sample size required for the multiple-regression models. With anticipated effect size (ƒ2) of 0.05, statistical power level of 0.95, four explanatory variables and probability
61
level of 0.05, the minimum sample size required is 376 (Cohen et al. 2003). Thus, a sample size of 400 was used for each lane-blockage state. For the fuel consumption model 300 observations were used. 3.3.2 Designing a Sample of Incidents To create a random sample of incidents for estimating the parameters of the multiple-regression models, the explanatory variables were used as the design criteria of each incident. Since the correlation between the explanatory variables is unknown, it was assumed that the explanatory variables are independent and uncorrelated with one another. Where this assumption invalid some incidents might have a low likelihood. While some incident cases with very low probability may be generated in the sample used within the simulation and later to calibrate regression models, if appropriate modeling techniques are used, these samples will have little effect on the development of a regression model passing goodness-of-fit tests. In addition, if the domain of a variable is dependent to one another, the estimation model should be developed for that domain, since a general model of all points might have different local behavior. This issue is addressed here with using the real world ranges for generating the random variables and the appropriate probabilistic distributions best describing each criterion. The random values for each incident in the sample are generated as follows. If incident duration is a random variable, it will have a probability density function (PDF). Statistical methods have been employed by researchers to explain and estimate incident duration when treating it as a random variable. These methods treat the random variables with probabilistic distributions, conditional probabilities, linear and non-linear regression models, time sequential and others as discussed in Chapter 3. Golob et al. (1987), GIuliano (1989), Garib et al. (1997), Suvilllivan (1997) and Ozbay et al. (1999) found that the log-normal distribution very closely fit their freeway incident data. Ozbay et al. (1999) claimed that incidents with the same severity level have normal distributions, supporting the theory that incident duration is a random variable (Smith and Smith, 2002). Nam and Mannering (2000) found that the Weibull distribution is also capable to estimate incident duration of an incident sample. Smith and Smith (2002),
62
however, found that required goodness-of-fit tests for log-normal and Weibull distributions failed. To create a sample of incidents representing real data, therefore, incident duration cannot simply be generated randomly from a uniform distribution. The average and standard deviation of the incident durations used in the design sample for this study need to be close to the incidents to which SSP vehicles responded. Chou and Miller-Hooks (2008) found that the average incident duration of 80% of incidents that SSP vehicles responded to is 17.6 minutes in New York State with standard deviation of 18.07 minutes. First, the Weibull distribution was used to generate incident duration times for the sample, but calibration of its parameters to provide desirable average and standard deviation were not successful. MATLAB was employed for generating random variables from the inverse of the Weibull distribution. Boyles and Waller (2207) used a log-normal distribution with µ (mean)=3 and ?(standard deviation)=1.6 to describe the incident duration of the incidents. Herein, by searching within the vicinity of those parameters, a log-normal distribution with mean 2.8 and standard deviation of 1.4 was found that best fit the distribution parameters that were sought. Using the inverse of the defined distribution (Equation 4.2), a set of 400 random incident durations having a mean of 17.8 minutes, standard deviation of 16.9 minutes, maximum of 70 minutes, and minimum of 5 minutes was generated. The lognormal inverse function is defined in terms of its CDF as in Equation 4.2.
{ }, Eq. 3.2
where
?
?
Hourly traffic volume is an important input used in the simulation runs. However, in travel delay studies, traffic volume is not often addressed directly and a factor that can be related to it is time-of-day variable: a.m. and p.m. peak hours on weekdays and off peak hours during weekdays and weekends. However, due to lack of information on the connection between time-of-day and volume, a uniform distribution is used to generate random traffic volumes in incident cases. Percentage of trucks in traffic composition, and
63
gradient were assumed to be independent of one another and a uniform distribution was used to generate each of them. For fuel consumption sample, number of lanes also assigned to incident scenarios from a uniform distribution. 3.3.3 Multiple-Regression Models As mentioned in Section 4.3.1, multiple-regression was selected to model travel delay and fuel consumption of both light- and heavy-duty vehicles in each lane blockage scenario. First, the linear regression models of travel delay and fuel consumption based on explanatory variables were obtained. Next, composition of variables is introduced to the model and a set of non-linear regression models are presented for validation by various goodness-of-fit tests. A stepwise technique is employed to find the best subset of explanatory variables for models. The stepwise technique starts the regression with the best regressor. It then finds the next best variable to add to the model, and finally it checks all variables in each equation to see if the previously entered variables remain significant. Other techniques that might be used in place of the stepwise technique include MAXR which chooses the variables to add to the model so as to achieve the highest possible R-square value. The stepwise method terminates based on the Mallow’s statistics. Mallow’s is a goodness-of-fit test for
regression that used ordinary least squares for estimating the parameters. When the expectation of becomes close to the P value, the stepwise procedure terminates and the
final set of explanatory variables are introduced. SAS statistical software package was employed for the statistical analysis conducted herein. SAS is a combination of programs that were designed for statistical analysis of data. The package offers six variable selection methods. These methods present results in a set of candidate regression models from which the best is chosen. To choose the best estimation model for travel delay and fuel consumption from the set of candidates models, six approaches are exhibited as goodness-of-fit tests: coefficient of determination (R-square), adjusted R-square, Mallow’s , Akaike Information Criterion under the name of “an information criterion” (AIC), Bayesian Information Criterion (BIC) and Schwarz’s Bayesian information criterion (SBC) as exhibited in Equations 5.1a to 5.1d.
64
The first method, the coefficient of determination method, is not always reliable. The goodness-of-fit increases with the number of regressors added to the model and, thus, the more explanatory variables, the better the model appears to be. The adjusted R-square method can be used to compare models with different numbers of explanatory variables, because regressors are added to the model only if their entry leads to statistically significant improvements in the model. Like the adjusted R-square technique, the AIC method penalizes any additional unnecessary estimators and discourages overfitting. Assuming the error term within the model is normally distributed, the maximum log likelihood was derived for each candidate models. The derived likelihood of each model is then to compute AIC, BIC and SBC (Equation 5.1b, 5.1c, 5.1d). From this set of candidate models, the one having minimum value of AIC, BIC and SBC would be selected as the final model. For example, to find the travel delay regression model having 8 explanatory variables, first using the stepwise and MAXR are used two sets of variables. Each set results regression model. The one best fit the data is identified throughout each of goodness-of-fits that may lead to a first choice model. If we decide to use R-square as goodness-of-fit test, the model which has higher R-square would be the best model,but if we want to use AIC as the goodness-of-fit test, the model that gives the lowest value of AIC would be presented as the final model. 5.1a 5.1b 5.1c ( where N = number of observations, SSE= Sum of squared errors, P= number of explanatory variables, K= number of free variables <= P+1, (k= # independent variables + intercept), ) , 5.1d
65
MML= Maximum log likelihood of the model. 3.3.3.1 Travel Delay Regression Models Regression models were developed for each lane blockage scenario and vehicle type. Note that each scenario is assumed to arise with the same likelihood. Let TTD stand for total travel delay of an incident case, ID be incident duration, vol be traffic volume at the time of the incident, PT be percentage of trucks in the traffic composition and G represent the gradient of the road. First, the linear model is developed for light- and heavyduty vehicles and the shoulder lane blockage scenario with four explanatory variables as shown in Equations 5.1a and 5.1b. All variables were found to be significant at the 0.15 level. The models have R-square values of 0.6772 and 0.5682, respectively. LDV:
Eq. 3.3
LDT:
Eq. 3.4
The above regression models are based on four assumptions related to the dependent variables: independence, normality, homoscedasticity (constant variance of response variable) and linearity. The regression assumptions can be re-expressed in terms of modeling errors to validate the assumptions on which model is built. Random errors are independent, normally distributed, have constant variance and zero mean (Equation 4.1). Having these . In
conditions the random errors can be considered as a random sample from
addition, the best representation of errors is through standard residuals. Standard residuals are the difference between actual and predicted response variables for each observation with constant variance over different dependent variables. SAS calculates residuals with a variance of 1. In general, any systematic pattern in residuals indicates a violation in assumptions and systematic error. Fit diagnostics for the models, including residual graphs for each parameter were obtained and analyzed. A summary of goodness-of-fit test results for travel
66
delay of LDV is presented in Figure 4.7 as a sample of a full analysis. The behavior of other models and the analysis were very similar to this case. In this model, it appears that the linearity assumption is violated, because the residuals are not scattered randomly around zero and form a clear pattern. Also the variance of residuals seems to have two values and they value is not constant. It shows that the model does not have the same accuracy for all data points. A Quantile-Quantile plot indicates that theoretical and actual data distributions do not agree, as the plotted points are not approximately on the y=x line. The slope of the curve of the plotted points increases from left to right, which indicates that a theoretical distribution that is skewed to the right, such as a log-normal distribution, might better fit the data. In addition, the mild curve indicates a small shape parameter for the chosen distribution (i.e. ? for log-normal). Cook’s D shows no outlier points, as all data points are within a distance of 2 of the zero line.
67
Figure 3.7 Summery of fit diagnostic for total travel delay of LDV
As part of additional analysis, the residuals are plotted separately for each explanatory variable (Figure 4.8). Since the variables are uncorrelated by design, each graph shows the direct relationship of dependent variable and explanatory variable. Incident duration has a random scatter plot matching its log-normal distribution. Residuals associated with the truck composition are also randomly scattered around zero, so the linear assumption seems reasonable. Residual graphs of volume and grade suggests a parabolic curve, then it may make sense to regress travel delay on the squared form of these two variables. Notice that the range of changes in truck composition is low inferring that the linear relationship with travel delay may be correct.
68
Figure 3.8 Scatterplots of residuals against explanatory variables
As a result, to improve the model, new variables are introduced. These variables are either the original variables squared (i.e. vol_sq indicates volume squared) or a multiple of two of the variables. Non-linear regression models were fitted to the data accordingly. The R-square of the models were improved slightly but the systematic errors were not eliminated. Search for an appropriate multiple-regression model was repeated for one- and twolane blockage incident cases using a similar procedure as described previously for the shoulder blockage case. The linear models are presented in Equations 4.5 to 4.8. One-lane blockage travel delay linear regression model: LDV:
Eq. 3.5
LDT:
Eq. 3.6
69
Two-lane blockage travel delay linear regression model: LDV:
Eq. 3.7
LDT:
Eq.3.8
Similar to the shoulder lane blockage category, non-linear regression models were calibrated for the one- and two-lane blockage incident categories. However, the travel delay estimation models of one- and two- lane blockage scenarios did not improve statistically compared with linear counterparts. The R-square of these non-linear models are presented in Table 4.16. A hybrid approach mentioned previously was established for improving travel delay regression models as described in the following section. 3.3.3.2 Hybrid Approach In linear regression, the coefficient of a single variable will not change by removing or adding a new independent and uncorrelated variable to the model. In Section 4.2 truck percentage and segment grade were found to be uncorrelated with other explanatory variables. A hybrid approach in which travel delay obtained from simulated incidents is integrated with estimates obtained from developed regression models is created to reduce the error of the estimation models and capture the relationship between travel delay, number of lanes in the segment, incident duration, traffic volume and the speed of vehicles more accurately. In the hybrid approach, the primary linear regression model is broken into two parts: (a) a travel delay function on number of lanes, incident duration, traffic volume and speed and (b) a travel delay function on percentage of trucks in traffic composition and roadway gradient. Assume an incident in which all the factors (the explanatory variables) are nonzero. The first part (a) is identical to the same incident case in which tucks percentage and gradient are zero. Travel delay associated with this incident ( ) was then directly
computed from the simulation runs. The additional travel delay due to percentage of trucks and different gradients then was included in the model using linear regression estimation
70
equations. For example, travel delay regression model for light-duty vehicles and shoulder blockage incidents (Equation 4.3) would be reformed as in Equation 4.9. a
[ ]
b
Eq. 3.9
To validate this hybrid approach, simulation runs for 300 incidents in a three-lane highway were completed. Incident durations and traffic volumes were set following the design described in Section 4.3.2. Truck percentage and gradient were set to random values from uniform distributions. Then, the equation 4.9 was applied to obtained travel delay data of the incidents with zero percentage of truck and grade, . Refer to these values
as “predicted values” for the travel delay of designed incidents. Then, the coefficient of determination of the hybrid model can be computed as follows:
where ?
? = Residual sum of squares, = total sum of squares, = Observed values, =Mean of observed values, = Predicted values by model. The R-square of the linear regression for travel delay for shoulder incidents was 0.672 while the R-square of the hybrid model is 0.939. Thus, we can conclude that this estimation approach better captures travel delay. The same approach was applied to linear models of each of the six categories (3 lane blockage categories for each vehicle class) and the R-square was calculated. A comparison of R-square values between the linear
71
regression models, non-linear regression models, and the hybrid approach is presented in Table 4.16. It can be noted that the R-square of regression models of all categories has improved significantly with the hybrid modeling approach given in Equation 4.9.
Table 3.10 Improved R-square comparison for new model
Lane blockage Shoulder lane blockage 1-lane blockage 2-lane blockage Vehicle class LDV LDT LDV LDT LDV LDT Linear model 0.677 0.568 0.195 0.153 0.142 0.129 Nonlinear model 0.878 0.698 0.244 0.236 0.141 0.075 Hybrid Model 0.939 0.875 0.768 0.719 0.784 0.725
To use this hybrid approach in estimating travel delays in the SSP-BC Tool, simulation runs for all possible cases of number of lanes in a segment, number of lanes blocked due to the incident, incident duration, traffic volume, and speed of vehicles must be made. The travel delay obtained from the runs is then integrated with the regressed travel delay due to truck percentage and segment grade. 3.3.3.3 Fuel Consumption Regression Model The same approach described in the previous section is used to obtain a fuel consumption prediction model for the light-duty vehicles. Later in Chapter 5 it is explained why the fuel consumption of LDT is not required for benefit computation of a SSP program. The linear model is presented in Equation 4.10 in addition to travel delay variables. Here, spd stands for speed of vehicles and lane is the number of lanes in the segment.
Eq. 3.10
The coefficient of determination of these linear models is 0.8210. The R-square for the light-duty vehicle model is very high, indicating excellent model fit to the data. However, it appears from Figure 4.9 that the linearity assumption is violated. This is indicated by a curve pattern in the residual scatterplot. Additionally, the variance of residuals increase indicates heteroscedasiticity assumption does not hold. From Figure 72
4.10, it can be seen that the residuals associated with volume are randomly scattered around zero, but the variance is not constant. This shows that the model is less accurate for some data points. Back in Figure 4.9, since the residual distribution is close to normally distributed and plotted points in the Q-Q chart are almost on the y=x line, it is reasonable to assume that the residuals are normally distributed
Figure 3.9 Summery of fit diagnostic of linear model of LDT fuel consumption
73
Figure 3.10 Scatterplots of residuals against explanatory variables
To address the linearity and variance problem, a set of non-linear models were derived and tested. From Figure 4.10, we can guess that a transform on gradient and speed might improve the model. Fuel consumption changes linearly with number of lanes in the segment (“lane”) as found preciously in Section 4.3.2. Using the stepwise method, some parameters were chosen to enter the model. The final chosen nonlinear model for fuel consumption (Table 4.11) obtained after consideration of a variety of models.
74
Table 3.11 LDV fuel consumption Root MSE 67.16175 R-Square0.8293 Dependent Mean200.47409Adj R-Sq0.8211 Coeff Var 33.50146 Parameter Estimates Parameter Variable DF Estimate Intercept 1 -255.22686 vol 1 0.05440 g 1 18.83305 lane 1 15.77193 spd 1 4.50066 Ptruck 1 -3.56345 ID 1 1.86832 Spd^(2) 1 -0.02294 ID^(1/6) 1 -0.18012 gsq 1 0.08399
Pr > |t| 0.0368 <.0001 <.0001 <.0001 0.0008 <.0001 0.0149 0.0233 0.0928 0.0050
The R-square of this model is not significantly improved by relaxing the linearity assumption. However, other goodness-of-fit tests show significant improvements and variables in the model agree with the similar studies in the same area, which make the model be adopted for the purpose of this study.
3.4
Summary VISSIM was employed to estimate travel delay and fuel consumption of individual
incidents in a segment. Impact of previously identified factors on travel delay and fuel consumption studied and multiple-regression models for estimating travel delay and fuel consumption were presented. A hybrid approach was introduced for improving the obtained travel delay models. Note that once fuel consumption is obtained, emissions can be computed (Section 3.4).
75
4
CHAPTER 5. B/C Ratio Estimation
This chapter discusses the computation of the B/C ratio, which is designed to provide insight into the return on investment received from operating a SSP program. Evaluation of the benefits in the B/C ratio requires a method for the amalgamation of chosen MOEs. In the developed tool, these are the savings in travel delay, fuel consumption, emissions and secondary incidents. These MOEs are given in a variety of units of measurement. Savings in travel delay is in vehicle-hours, savings in fuel consumption is in gallons of fuel, savings in emissions is in metric tons, and secondary incident savings is in number of prevented incidents. Thus, conversion to a common unit of measurement is required to develop a single numeric value for the numerator of the B/C ratio. Moreover, the unit of measurement must be commensurate with the units used in the B/C ratio’s denominator, namely cost. Consequently, the most common approach is to convert the individual benefit measures to their monetary equivalents using monetary conversion factors. Methodologies for computing the total program savings associated with each of the chosen MOEs are provided in Section 5.1. Computation of the total benefit, i.e. the numerator of the B/C ratio, is discussed in Section 5.2. Section 5.3 describes total cost calculation in the deamination. This is followed by B/C ratio calculation in Section 5.4. 4.1 Savings Computation To compute the benefit of a SSP program during a time period, benefits derived from each individual incident due to response by an SSP vehicle must first be determined. This is because the duration of the incident decreases as a result of the SSP vehicle response as explained in Section 3.2.4.2. To assess the value of the reduction in incident duration, travel delay, fuel consumption, and emissions can be estimated for the incident with reduced duration as a result of the SSP response “with-and-without” approach was employed (Section 2.2). Typically, it is the case that no such pre-program measurements were made. Thus, estimation of the “without” case must be made by assuming an increase in the duration of each realized incident. The amount of increase should be commensurate with the program’s incident response time.
76
4.1.1 Savings in Travel Delay To estimate travel delay of actual incidents in an area, using the travel delay hybrid model proposed in Section 4.3, simulation results from all studied incident scenarios are required. Simulation runs were made and travel delays associated with 14,784 incident scenarios were collected. The runs involved all possible combinations of 16 categories of incident duration, 11 categories of traffic volume and 6 speed categories, resulting in 1,056 combinations. For each combination, runs are including one of 3 types of lane blockage and one of 5 possible roadway sizes in terms of number of lanes. Results were saved in a table contained within the data base that supports the tool for further computations. To estimate the travel delay associated with an incident with known characteristics using the proposed simulation-based procedure, the incident characteristics and associated traffic volume and speed must be known. As discussed in Chapter 3, the impact of the ramp density (Section 3.2.1.4), horizontal curvature (Section 3.2.2.1), and weather conditions (Section 3.2.5) on the capacity of the segment is captured through a reduction in FFS, affecting the speed category of the incidents. The maximum speed reduction due to existence of ramps, curvature, and adverse weather conditions determines the speed category of the incident. For example, consider an incident case for which speed in clear weather is 65 mph. To include the impact of heavy rain (a reduction in speed by 10percent) and a full cloverleaf interchange in a one-mile segment (two on-ramps and two off-ramps in each direction, 5 mph reduction for each 2 ramps/mile),the employed speed for the incident cases in that segment would be 55 mph;
{ }
When the final speed (after all reductions are taken) to be associated with an incident is determined, the estimation of travel delay for each incident can be completed with the use of the proposed regression models (Equations 4.3 to 4.8). The savings are computed from the difference between travel delays for the “without” and “with” incident cases. An overview of this procedure is provided in Figure 5.1.
77
Figure 4.1 Travel delay estimation procedure
For incidents for which particular incident duration, traffic volume and speed is not one of the categories in above data set, travel delay is obtained by assuming linear changes in between upper and lower bound categories. If category of incident duration i, denotes travel delay of incident duration i and and stand for travel delay of lower
and upper bound of the incident duration i category, respectively, Equation 5.1(a) can be used to obtain travel delay of desired incident duration.
Eq. 4.1(a)
Traffic volume is assumed to be rounded to nearest volume category. A similar linear estimation approach to incident duration was used to interpolate when given speeds outside the tested categories. Likewise, a similar equation to Equation 5.1(a) but for speed can be used as follows.
78
Eq. 4.2(b)
Where in incident i,
denotes speed of vehicles prior to incident i and
and
stand for the speed of lower and upper bound of the incident i, respectively, if neither incident duration nor speed associated with an incident were in the provided data set, by linear assumptions on speed and using Equation 5.1(b) for SPu and SPl, the travel delay associated with upper and lower bound on of incident duration, IDu and IDl, are computed and then Equation 5.1 is applied to compute the final travel delay. This process is shown in Figure 5.2.
Figure 4.2 Subcategory linear interpolation Incident Duration, Speed SPu IDu SPl SPu IDl SPl
4.1.2 Saving in Fuel Consumption and Emissions A similar with/without incident approach is taken to estimate fuel consumption savings associated with an incident scenario. Fuel consumption corresponding to each incident scenario is obtained directly from equations described in Tables 4.15 to 4.22. Emissions are calculated directly from equations 3.6-3.8 based on the fuel consumption estimates. 4.1.3 Saving in Secondary Incidents The probability of occurrence of a secondary incident grows with an increase in the primary incident duration (Section 3.5). To estimate the number of prevented secondary incidents, the number of secondary incidents when SSP is not operating is assumed to be linearly correlated with the travel delay ratio of without and with incidents to which SSP responded in a period of time. This approach to estimating secondary incident savings is discussed in (Chou and Miller-Hooks, 2008). It supposes that total travel delay is a
79
reasonable surrogate for impact area size of primary incidents in which incidents classifies as secondary. This relationship is shown in Equation 5.2. , where : Number of secondary incidents for extended incident duration case (without case), : Number of secondary incidents in base case (with case), : Travel delay for the extended case, : Travel delay for the base case. As discussed in Chapter 3, for this analysis, the number of secondary incidents ( as a fraction of primary incidents must be known regardless of the chosen secondary and are estimated as explained in Section 5.1.1.
Eq. 4.2
incident classification method.
Another method to calculate the benefits of SSP program in terms of prevented secondary incident is to consider the incident duration reduction contribution to likelihood of secondary incident occurrence as explained in Section 3.5. As mentioned in Section 3.5 Karlaftis et al. (1998) estimated the clearance time coefficient for winter and all other seasons as 0.017 and 0.031, respectively. Assuming that SSP vehicles reduce the incident duration by 20 minutes, the increase in the likelihood of a secondary incident occurrence would be 14.05% in winters and 18.59% in all other seasons.
The average increase in likelihood of occurrence of a secondary incident is %17.46. Using this method, the potential secondary incidents reduced due to SSP program can be computed as shown in Equation 5.3. ( where ) ,
Eq.5.3
80
N: Number of reduced potential secondary incidents, ID: Incident duration reduction due to SSP program operation in minutes, Np: Total number of incidents to which SSP vehicles responded.
4.2
Total Benefit Calculation
4.2.1 Monetary Values To isolate a single unit for evaluation of a SSP program, congestion related travel delay (vehicle-hours), fuel consumption (gallons), and number of secondary incidents prevented are converted into their monetary equivalents. Monetary equivalents in the SSPBC Tool proposed herein were provided by the American Transportation Research Institute (ATRI). Table 5.1 contains a list of the monetary equivalent variables, the variable’s corresponding output, a description for each variable and data source. Four individual tables containing this information support the B/C ratio computation within the tool. They are also designed to be updatable.
Table 4.1 Summary of monetary equivalents (ATRI)
Variable Corresponding Output Description Used to monetize the wasted fuel that would result from increased congestion if SSP did not exist NOTE: fuel is already factored into the Hourly Truck Cost, and the monetization of wasted fuel should only be performed on the passenger vehicle share Used to monetize lost productivity of passenger vehicles resulting from increased congestion if SSP did not exist Source
Average gasoline prices (Table B.1)
Gallons of fuel saved
U.S. Energy Information Administration, Gasoline and Diesel Fuel Update; updated 5:00 p.m. every Monday; http://www.eia.doe.gov/oog/info/gdu/gasdi esel.asp U.S. Department of Labor, Bureau of Labor Statistics; State Occupational Employment and Wage Estimates; http://www.bls.gov/oes/current/oessrcst.ht m An Analysis of the Operational Costs of Trucking: A 2011 Update; ATRI; http://www.atrionline.org/research/results/Op_Costs_2011 _Update_one_page_summary.pdf Based on actual operational cost data collected from motor carriers across the country, representing a cross-Section of industry sectors. The Economic Impact of Motor Vehicle Crashes: 2000. NHTSA.
Average labor costs (Table B.2, B.3)
Hours of delay prevented
Commercial vehicle costs per hour
Hours of delay prevented; Gallons of fuel saved
Used to monetize lost productivity of commercial vehicles resulting from increased congestion if SSP did not exist.
Secondary incident cost
Number of secondary incidents averted
Represents only the cost of property damage. Used to monetize the cost of additional secondary incidents that would result from increased congestion if SSP did not exist.
81
While some previously developed B/C ratio estimates made for SSP programs have included monetized emissions equivalents in the savings computation, a review of the literature indicates that the available monetary equivalents are based largely on soft, intangible costs, as opposed to other more tangible costs (e.g. price of a gallon of fuel). Thus, tons of emissions saved are reported separately and are not included in the B/C ratio computed in the SSP-BC Tool. Average hourly wages are used herein to convert savings in travel delay to a monetary equivalent. Wage values are available at metropolitan-levels and as a state average (Table B.3). Additionally, data containing the share of commercial VMT compared to total VMT by state were used for truck composition estimate for each state (Table B.4). This data is necessary to distinguish between the benefits derived from savings in travel delay due to passenger vehicles and commercial vehicles. Average operational cost of trucking for 2011 is obtained to be $59.61. The B/C ratio is highly sensitive to the cost of secondary incidents. In this study, cost represents “property only damage” incidents and for 2011 it is assumed to be $4,736. Other costs associated with higher severity incidents and congestion due to secondary incidents were not considered. 4.2.2 Computing Total Benefit To compute the total savings in travel delay, fuel consumption, emissions, and secondary incidents resulting from a SSP program in a segment over a period of time, information pertaining to the incidents arising along the studied roadway segment during the study period is needed. Specifically, the distribution of incidents with respect to lane blockage must be known (or approximated). Assuming any two incidents are independent, TSj , the total savings of type j, where j={total travel delay, fuel consumption, emission} for every incident i arising during a period of time over a road segment as described in Equation 5.3 can be computed. When using this method, it is necessary to assume that the impact of an individual incident has no influence on other incidents on the road. Furthermore, as described in Section 5.1.1, the speed of the incident scenario and so the savings in travel delay and fuel consumption is related to geometry characteristics of
82
the study segment and weather condition in the time of the incident. The geometry characteristics are similar for all incidents in a study segment. However, the weather condition might vary incident by incident in a period of time. One incident under each weather conditions (Section 3.2.5) would have different final speed. Therefore, having the probability of each weather type, Pk, saving of one incident can be estimated as exhibited in equation 5.4. ? ? where TSj = Total saving j, j = =Type of saving {Total travel delay, fuel consumption, emission pollutants}, i = Individual incidents, k= Weather conditions {Clear, light rain, heavy rain, low visibility, snow, fog, icy = Saving type j in incident i of weather condition k. Given monetary conversion rates for travel delay, fuel consumption, and secondary incidents, total program benefit can be computed. With these concepts, and assuming that benefits are uniformly distributed over length, the total benefit of the SSP program over the study period and roadway segment can be computed as in Equation 5.4.
?
Eq. 5.5
,
Eq. 5.6
where B= Total benefit of a SSP program, j = Total travel delay (1), Fuel consumption (2), Secondary incidents (3), TSj = Total savings of type j, MEj= Monetary equivalent of saving j, L= the length of the study segment.
83
4.3
Cost Calculation The total cost of a SSP program, TC, is a function of the number of roving SSP
trucks along the study segment, hourly operating cost per truck, number of working hours, number of workdays in a year, fuel cost of each vehicle, cost of giveaway fuel to the vehicles that ran out of gas, and other costs such as vehicle maintenance cost as expressed by Equation 5.4. Moreover, for some SSP programs, most often the total annual cost of the whole SSP program is available and not the cost associated with the study segment. The total annual cost can be computed from Equation 5.4.
Eq. 4.3
where TC : Total annual cost for operating the SSP program in dollars, c : Cost per truck-hour {hourly wage of driver, fuel cost of the vehicle}, n : Number of roving trucks, hr : Number of working hours in each day, day : Number of workdays in a year, fuel: annual giveaway fuel cost. The cost of many SSP programs can often be easily calculated, as many SSP programs are outsourced and the charges are provided contractually. The cost of the program by roadway segment may be less clear. Two general methodologies were considered herein for the computation of segment-based costs. First, given total program costs, costs associated with a given segment can be computed based on the proportion: number of the total incidents to which the SSP vehicles responded to those to which they responded arising only within the study segment. This computation is captured in Equation 5.8. ? where = Cost of operating the SSP program along study segment n, = Total annual cost of the SSP program,
, Eq. 4.4
84
= Number of incidents along the study segment n to which the SSP program responded, = Total number of incidents to which SSP program responded. The second methodology is to compute cost associated with a given segment by the proportion of length of it to the total length of covered roads SSP vehicles covers. In this method, it is assumed that cost is uniformly distributed over length of the roads of the SSP service area. The first method is used in SSP-BC Tool. Some SSP programs may operate a heterogeneous fleet of vehicles. Thus, those vehicles that are similar in response capability or with identical operational hours can be classified as falling within the same group. Total annual costs can be computed from costs computed for each category of vehicles. 4.4 The B/C ratio The obtained benefit from Equation 5.5 and cost from Equation 5.7 are used to assess the segment-based B/C ratio for a given SSP program over the study period. The SSP-BC Tool provides multi-segment analysis. The B/C ratio of n segments is computed from the ratio of the sum of benefits to sum of costs for all segments as in Equation 5.6. ? where ? = B/C ratio of multiple segments, = Obtained benefit of segment n, = Obtained cost of segment n. Recall that within the SSP-BC Tool, savings in emission pollutants are not translated to dollars and, thus, cannot be included in the B/C ratio. Emissions are given separately in the form of metric tons.
? ?
,
Eq. 4.5
85
4.4.1 Additional Benefits Additional savings that has not been quantified in this study are: improved safety not only in preventing secondary incidents, but in the improved feeling of security on the transportation system, congestion cost associated with the secondary incidents, improved freight transit system, environmental benefits, and benefits to other agencies like additional time available for troopers for more urgent tasks that the SSP programs cannot handle. A list of additional costs associated with incidents is: Administrative costs: the cost (monetary and temporal) associated with investigating and documenting the primary, and any secondary, incidents. In the case of fatal incidents, costs increase exponentially. In addition, there are generally administrative costs associated with insurance claims. Legal costs: Includes attorney fees and court costs associated with litigation resulting from primary and secondary incidents. Rehabilitation costs: The cost of career retraining required as a result of disability caused by roadway incident. An additional cost in this category is, replacement employee costs. That is, employers often hire temporary help or compensate other staff by paying overtime to cover the position of an injured employee. Disability/Retirement income: Should the employee suffer career-ending injury, the employer will have to make payments to fund the employee’s disability pension. Productivity reduction: this is the cost associated with lost wages and benefits over the victim’s remaining lifespan. Numerous additional sources of benefits in cost reduction have not been included in the computation of program benefits within the proposed SSP-BC tool. The exclusion of the many additional benefits from the benefit estimate used in the B/C ratio results in conservative B/C estimates.
86
5
CHAPTER 6. The Tool by Illustrative Example
The SSP-BC Tool interface was coded in Microsoft Visual Basic 2003. Data developed based on Chapter 4 is in microsoft Access (2010) format. Tables of monetary values for travel delay and fuel consumtion, and share of trucks in traffic volume were designed to be updated by the user. The SSP-BC Tool is explained in Section 6.1. 693 incidents to which the SSP program in New York (H.E.L.P) responded over a 6-month period in 2006 is used as a case study for the tool and its outputs. A comparison between previously obtained B/C ratio by Chou and Miller-Hooks (2008) and use of the propesed generic SSP-BC Tool is made in the Section 6.2. 5.1 The SSP-BC Tool The I-287 segment studied herein is approximately 10 miles in length, beginning at the junction with I-95 and continuing west to the Tappan Zee Bridge in New York. This segment is referred to as Beat 8-2 of the H.E.L.P. program. Incidents arising on this roadwaysegment will be handled by a H.E.L.P. vehicle driver, a trooper from Unit T or both. During the study period, 1,303 incidents arose along the study segment of I-287. 693 of these 1,303 incidents received service from the H.E.L.P. program during the H.E.L.P. hours of operation. Figure 6.1 shows the main window of the SSP-BC tool including the information on SSP program level and number of segments. Total annual program cost is required here. In addition, a detailed cost list is provided in the “Cost” window (Figure 6.2) in which annual SSP program cost will be calculated automatically. The figure shows the cost information of the H.E.L.P program. Chou and Miller-Hooks (2008) used costs of $40 and $50 per truck-hour, two roving trucks operated within the study roadway segment with an eight-hour workday, 126 workdays within the 6 month study period (21 days/month). The annual cost using $40/truck-hour was $161,280. Up to 5 different categories of cost groups as explained in Section 5.3 are availble for different type of service vehicles or operational hours that might exist within a SSP program.
87
The next input is for finding the associated cost to the segment, total number of incidents arose in the segment in one year. If the annual cost was the estimated cost of total SSP program, not the program cost associated with the study segment like this example, the number of incidents to which the SSP program responded in a year has to be used as input here (Section 5.3). Segments in a study area must be homogenouse in terms of geometry, weather and traffic volumes as explained in Chapter 3. The SSP-BC Tool is capable of analysing up to 30 segments at a time. This example has only 1 segment.
Figure 5.1 Main window
88
Figure 5.2 Program cost detail
For each segment information pertaining to hours of operation weather, traffic conditions, SSP program average response time, roadway geometry, and incident distribution and duration must be entered. It is assumed that SSP programs operats within a single state. Regional data at the metropolitan level are applied in setting the monetary conversion rate. Average monetary values for the state are used when the region is set to “others”.
Figure 5.3 Basic data in segment level
89
The program information window (Figure 6.4) contains details on operational hours and perfomance of the SSP program. Hour of operation are divided to four time categories. The user must to be consistant with her/his definition of each time category for analysis of each segment. For example, if she/he selects 7 to 10 AM as her/his AM peak hour, traffic and incident information for these hours must be used for the AM peak in following steps. A key impact to the SSP-BC Tool is the average incident duration reduction offered by the program. Different approches for estimating this time reduction are discussed in detail in Chapter 2. In addition, the FSPE tool (Describe in Section 2.3) computes the arrival time of the SSP vehicles and ,thus, incidnet response time. The savings in incident duration greatly depend on the severity of the incident or, as employed herein, on the number of lanes blocked as a result of incident occurence. The SSP-BC Tool assigns incident duration savings in each lane blockage category to the duration of the incidents in that category. The average value, as in the example, can also be used where sufficent data is not availble. Chou and Miller-Hooks (2008) founded an average savings of approximately 20 minutes in incident duration for incidents involving a collision and 19 minutes for incidents involving a disabled vehicle for the study area as a result of the presence of the H.E.L.P. program. The average incident duration savings of 20 minutes is used for the example solved here.
90
Figure 5.4 Program information window
The SSP-BC Tool provides default values for incident duration and related savings to the due to SSP program based on previous studies in the area. The default values associated with roadway geometry and traffic information (Figure 6.5) are based on the numerical experiments used as described in Chapters 3 and 4 and summarized in Table 6.1.
Table 5.1 Geometry and traffic default values
Input Segment length Number of traffic lanes by direction General terrain Horizontal curvature Number of ramps in segment Posted main-lane speed limit Percentages of trucks Weather Default value 10 3 Level straight 0 70 3 clear
The H.E.L.P study segment length is 10 miles and number of lanes is 4. Default values of general terrain and road curvature were used. Total ramp density for the H.E.L.P study segment is 1.4 ramps/mile with14 on/off ramps within the segment.
91
Traffic volume data for the study roadway segment was employed for the same period, but in the following year. Average weekday and weekend traffic volumes by month were matched to the incidents by their date and time information. For each of the operational hour classes of the SSP program, one traffic volume was assumed in this example. Incidents for which prevailing traffic volume was between 0 and 600 vplph were categorized for the weekend, 600 to 1000 vplph for weekday off peak, 1000 to 1400 vplph as weekday PM peak, and 1400 to 2200 as weekday AM peak. The average for the prevailing incident traffic volumes in each category was set within a time class. For example for weekday PM peak, the average of 1200 vplph was used. The percentage of trucks set to 7.8% (Table B.4) for all operational hour classes. Weather was assumed clear for all incidents. Note that this classification of volume was done to fit the available data to the SSP-BC Tool. The operational hour of the H.E.L.P. program is weekday peak periods.
92
Figure 5.5 Roadway geometry and traffic information
Incident information scenario is entered next. Average incident durations and number of incidents are required by lane blockage as shown in Figure 6.6. They are assumed to be identical for all operational hour classes for this example. Savings in prevented secondary incidents in the tool is calculated using the first method discussed in Section 5.1.3. The input is the percentage of secondary incidents out of primary incidents for the study segment.
93
Figure 5.6 Incident information window
Since the example has only one segment, the B/C ratio can be ontained. The output window is shown in Figure 6.7. Users can choose one or more segments for the B/C ratio analysis. The outputs of the tool as described in MOEs, Section 3.1, are savings in travel delay in vehicle-hours, fuel of passenger cars and light-duty in gallons, number of prevented secondary incidents and emission pollutants in metric tones. The users specifies which MOEs (of travel delay, fuel consumption and secondary incidnes) to include in the
94
B/C ratio.The total benefit of the chosen segment is then calculated as described in Section 5.2. For this example, 2006 monetary values are employed to compute the benefits for consistancy with the availble cost information of the program (Table 6.2).
Table 5.2 New York 2006 monetary values
State New York Travel delay $/hr 15 Fuel (gas) $/gallon) 2 Carbon Monoxide (CO) $/ton 6,360 Hydrocarbons (HC) $/ton 6,700 Nitrogen Oxide(NO) $/ton 12,875
Figure 5.7 Output window
95
The B/C ratio of the H.E.L.P program was estimated to be 2.83 plus additional benefits derived from emisions savings. Adding emissions to the benefits increased the B/C ratio to 3.08. Chou and Miller-Hooks (2008) estimated the B/C ratio of the H.E.L.P program 2.68. In both B/C ratio estimates that included emissions, only benefits from savings in CO, HC, and NO were included. Monetary conversion rates used are given in Table 6.2. They used technique in which they replicated incidents in CORSIM and computed the travel delay using the same with/without approach (Section 5.1). They used conversion factors for fuel consumption and emissions from travel delay. In simulating incidents they did not includethe geometry characteristics of the segment, such as ramps. Table 6.3 contains obtained savings Chou and Miller-Hooks found for 20 minutes incident duration savings due to the operation of H.E.L.P. program.
Table 5.3 H.E.L.P result comparison
Type of saving Travel delay (veh-hr) Fuel consumption (gal) Secondary incidents (#) CO (ton) HC (ton) NO (ton) Chou et al. (2008) 12,182 1,451 9 1.79 0.16 0.08 SSP-BC 10,097 12,856 17 8.34 1.1 0.55
Travel delay estimated by the SSP-BC Tool is slightly less than what Chou and Miller-Hooks obtained. On the other hand, Chou and Miller-Hooks estimated fewere saved secondary. A comparison between VISSIM and CORSIM was completed to better understand difference in travel delay estimates. The study segment was simulated in both software products. Travel delay was gathered for different traffic volumes as plotted in Figure 6.8. It was found that CORSIM estimates higher travel delays compare to VISSIM when the simulated segment reaches its capacity. It seems that in CORSIM, the capacity of the roadway segment with defult parameters applied in Chou and Miller-Hooks, worked is lower compare to VISSIM using described calibrated parameters. Relate discussion can be
96
found in Section 4.1.2. Where earlier studies note that VISSIM provides a better model of traffic than CORSIM.
Figure 5.8 CORSIM vs. VISSIM
250 200 150 100 50 0 0 500 1000 1500 2000 2500 Volume (vphpl)
VISSIM CORSIM
. The B/C ratio of the H.E.L.P program with 2011 monetary values (Appendix B),as current SSP-BC Tool monetary equivalent data, and asuming the truck-hour cost of $60 compare to $40 in 2006, total cost of the program would be $120,960, is 2.23.
Travel delay (veh-hr)
97
6
CHAPTER 7. Conclusions and Limitations
The SSP-BC Tool was developed to fill the need for a standardized B/C ratio estimation methodology with wide applicability and substantiated and needed updatable monetary conversion rates. The tool was designed to support B/C ratio estimation for roadways with existing programs, but can also be used to test numerous what-if scenarios, including the introduction of a new program or the impact of improvements in service response times. A quicker and less data-intensive approach was developed so that it can be readily and widely utilized by all states around the US. The SSP-BC Tool accounts for a wide array of traffic, environmental and program characteristics that influence benefit and cost estimates. The Factors, such as incident duration, traffic volume and composition, ramp density, horizontal and vertical alignments, and weather conditions that have been identified as important to travel delay fuel consumption and emissions estimation were included in this tool. Moreover, the per-second vehicle velocity and acceleration values were employed in the computation of fuel consumption and emissions. Numerous experimental runs were completed and seven multiple-regression models for estimating travel delay and fuel consumption were developed. For experimental runs, the techniques to simulate different geometry, traffic, and weather characteristics were suggested and tested. Additional delay caused by two regression parameters, truck percentage and segment gradient, can be applied to any travel delay estimate in which traffic composition or gradient is not included. 6.1 Limitations The SSP-BC Tool extensions are limited by a maximum incident duration of ninety minutes, that only up to two lanes can be blocked due to an incident and no consideration for roadway closure, and that any given segment has a maximum of six lanes. The SSP-BC Tool can be extended to include of longer duration or greater severity, and roadway with more than six lanes. To complete a nationwide study, it would be ideal to have all input data associated with the entire national roadway system. As this is highly impractical, a statistical approach
98
might be used in which a random sample of the required traffic and environmental data is taken. With this sample, a general estimation model can be created and calibrated based on statistical approaches. Unfortunately, this method is not an easy and cost effective method. In addition, obtaining one general estimation model for all states would not be reliable considering the fact that statistical models developed for the entirety of the nation will likely poorly fit the data of specific regions. 6.2 Contributions In addition to the development of a nationwide tool for SSP program B/C ratio estimation, the contributions of this work include the identification of the significant factors affecting SSP program benefits, techniques for simulating these factors and development of regression models to estimate travel delay, fuel consumption and emissions given traffic, roadway geometry, program characteristics and weather conditions. For travel delay estimation, the developed enhanced regression methodology takes a hybrid approach to multiple-regression model construction. This approach combines parameters obtained through regression analysis for truck percentage and roadway grade, which were found to be independent of all other factors, with results from simulation runs. The simulation results and estimates from the regression models are developed into tables employed within the tool’s database. For fuel consumption estimation, a multiple regression model is developed that is used directly within the tool. These developed regression models can be used directly where applicable.
99
7
APPENDIX A: Fuel Consumption Computation Tables
Table A.1 Calculation of Road-load coefficients
Vehicle Category LDV (passenger cars) LDT (trucks, SUVs, etc) LHD<=14K LHD<=19.5K Source Mass (metric tons) 1.4788 1.86686 7.64159 6.25047 A (KW/mps) 0.156461 0.22112 0.561933 0.498699 B (KW/mps2) 0.00200193 0.00283757 0 0 C (MW/mps3) 0.000492646 0.000698282 0.00160302 0.00147383
Table A.2 Emission factors
Fuel Economy (mile/gal)1 22.1 17.6 Engine Displacement Volume, V (L)3 1.3-3.1 2.5-5.3 Emission Factor (EF in g/mi) for gasoline2 HC 2.8 3.51 CO 20.9 27.7 NOx 1.39 1.81 CO2 451 637
Vehicle LDV LDT
Table A.3 Fuel properties
HV (KJ/g)a ?b
(g/gal)
Fuel
Base Fuel
SCFuel (ppm) 80 500
Gasoline Diesel
(Base fuel) (Base fuel)
43.448 42.791
2834.95 3210.98
Table A7.4 Transmission parameters for engine speed calculation (Source: PERE, 2005; EPA420-P05-001)
LDV & LDT S = 35.6 Speed (mph) 0-18 18-25 25-40 40-50 50+ Gear 1 2 3 4 5 g/gtop 4.04 2.22 1.44 1 0.9
100
8
APPENDIX B: Monetary Equivalents
Table B.1 Average gasoline prices
Area name Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri abb AL AK AZ AR CA CO CT DE DC FL GA HI ID IL IN IA KS KY LA ME MD MA MI MN MS MO Fuel price($) 3.163538462 3.514961538 3.514961538 3.163538462 3.586 3.163538462 3.379346154 3.339615385 3.339615385 3.267 3.242576923 3.514961538 3.198192308 3.298903846 3.298903846 3.298903846 3.298903846 3.298903846 3.163538462 3.379346154 3.339615385 3.307923077 3.298903846 3.303057692 3.163538462 3.298903846 Area name Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming abb MT NE NV NH NJ NM NY NC ND OH OK OR PA RI SC SD TN TX UT VT VA WA WV WI WY Fuel price ($) 3.198192 3.298904 3.514962 3.379346 3.339615 3.163538 3.500981 3.242577 3.298904 3.282615 3.298904 3.514962 3.339615 3.379346 3.242577 3.298904 3.298904 3.167385 3.198192 3.379346 3.242577 3.492635 3.242577 3.298904 3.198192
101
Table B.2 Average labor cost by state
Area name Alabama Alaska Arizona Arkansas Average wage ($/hr) 21.5 24.21 20.38 17.05 Area name Montana Nebraska Nevada New Hampshire Average wage ($/hr) 17.34 18.42 19.82 21.37
California Colorado Connecticut Delaware District of Columbia Florida Georgia Guam Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts
24.39 22.48 24.96 22.53 35.31 19.36 20.32 15.02 21.03 18.56 22.33 18.76 18.14 18.89 18.25 18.26 18.98 24.46 25.82
New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Puerto Rico Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virgin Islands Virginia
24.39 19.26 24.86 19.47 17.81 19.66 17.76 20.94 20.7 12.92 22.08 18.23 16.53 18.43 20.3 19.29 20.21 17.85 23
102
Michigan Minnesota Mississippi Missouri
20.81 21.86 16.31 19.13
Washington West Virginia Wisconsin Wyoming
23.53 17.01 19.7 19.96
Table B.3 Average wage by area
Area Anniston-Oxford AL Auburn-Opelika AL Birmingham-Hoover AL Columbus GA-AL Decatur AL Dothan AL Florence-Muscle Shoals AL Gadsden AL Huntsville AL Mobile AL Montgomery AL Tuscaloosa AL Northwest Alabama nonmetropolitan area Northeast Alabama nonmetropolitan area Southwest Alabama nonmetropolitan area Southeast Alabama nonmetropolitan area Anchorage AK Fairbanks AK Southeast Alaska nonmetropolitan area Railbelt / Southwest Alaska nonmetropolitan area Flagstaff AZ Lake Havasu City - Kingman AZ Phoenix-Mesa-Scottsdale AZ Prescott AZ Tucson AZ Yuma AZ Wage ($/hr) 16.92 17.92 19.82 17.83 17.72 16.59 16.45 15.93 23.12 18.39 18.43 18.26 15.62 15.62 16.21 16.73 24.75 24.21 22.33 23.69 18.89 16.97 20.89 18.04 20.27 16.4 Area Duluth MN-WI Fargo ND-MN Grand Forks ND-MN La Crosse WI-MN Mankato-North Mankato MN Minneapolis-St. Paul-Bloomington MNWI Rochester MN St. Cloud MN Northwest Minnesota nonmetropolitan area Northeast Minnesota nonmetropolitan area Southwest Minnesota nonmetropolitan area Southeast Minnesota nonmetropolitan area Gulfport-Biloxi MS Hattiesburg MS Jackson MS Pascagoula MS Northeast Mississippi nonmetropolitan area Northwest Mississippi nonmetropolitan area Southeast Mississippi nonmetropolitan area Southwest Mississippi nonmetropolitan area Columbia MO Jefferson City MO Joplin MO Springfield MO Central Missouri nonmetropolitan area North Missouri nonmetropolitan area Wage ($/hr) 18.81 18.14 17.81 19.09 18.06 23.63 23.43 18.62 17.28 17.2 16.84 17.88 17.23 15.87 17.7 18.58 15.78 14.8 15.09 15.97 17.62 17.91 16.23 17.02 15.51 15.07
103
North Arizona nonmetropolitan area Southeast Arizona nonmetropolitan area Fayetteville-Springdale-Rogers AR-MO Fort Smith AR-OK Hot Springs AR Jonesboro AR Little Rock-North Little Rock-Conway AR Memphis TN-MS-AR Pine Bluff AR Texarkana-Texarkana TX-AR Central Arkansas nonmetropolitan area East Arkansas nonmetropolitan area South Arkansas nonmetropolitan area West Arkansas nonmetropolitan area Bakersfield CA Chico CA El Centro CA Fresno CA Hanford-Corcoran CA Los Angeles-Long Beach-Glendale CA Metropolitan Division Los Angeles-Long Beach-Santa Ana CA Madera CA Merced CA Modesto CA Napa CA Oakland-Fremont-Hayward CA Metropolitan Division Oxnard-Thousand Oaks-Ventura CA Redding CA Riverside-San Bernardino-Ontario CA Sacramento--Arden-Arcade--Roseville CA Salinas CA San Diego-Carlsbad-San Marcos CA San Francisco-Oakland-Fremont CA San Francisco-San Mateo-Redwood City CA Metropolitan Division San Jose-Sunnyvale-Santa Clara CA San Luis Obispo-Paso Robles CA Santa Ana-Anaheim-Irvine CA Metropolitan Division
16.99 19.48 18.7 16.09 16.55 16 18.75 19.32 16.76 17.5 15.21 14.92 15.17 14.33 21.4 19.54 19.01 19.76 20.71 24.16 24.1 20.78 18.79 19.96 23.86 27.09 23.03 19.88 20.64 24.08 20.61 24.14 28.76 30.43 32.62 21.29 23.93
Southeast Missouri nonmetropolitan area Southwest Missouri nonmetropolitan area Billings MT Great Falls MT Missoula MT Eastern Montana nonmetropolitan area Central Montana nonmetropolitan area Southwestern Montana nonmetropolitan area Western Montana nonmetropolitan area Lincoln NE Western Nebraska nonmetropolitan area Central Nebraska nonmetropolitan area Northeastern Nebraska nonmetropolitan area Southeastern Nebraska nonmetropolitan area Carson City NV Las Vegas-Paradise NV Reno-Sparks NV Western Central Nevada nonmetropolitan area Other Nevada nonmetropolitan area Manchester NH Northern New Hampshire nonmetropolitan area Other New Hampshire nonmetropolitan area Western New Hampshire nonmetropolitan area Southwestern New Hampshire nonmetropolitan area Allentown-Bethlehem-Easton PA-NJ Atlantic City-Hammonton NJ Camden NJ Metropolitan Division Edison-New Brunswick NJ Metropolitan Division Newark-Union NJ-PA Metropolitan Division New York-White Plains-Wayne NY-NJ Metropolitan Division Ocean City NJ Trenton-Ewing NJ Vineland-Millville-Bridgeton NJ Albuquerque NM Farmington NM Las Cruces NM Santa Fe NM
14.67 14.67 18.06 16.62 17.5 16.43 16.36 17.79 16.66 18.83 15.31 15.96 15.86 16.01 21.85 19.59 20.52 18.41 20.7 22.49 16.73 19.85 21.76 20.11 20.38 20.02 22.26 24.57 25.74 27.49 18.64 26.93 20.36 19.96 18.58 18.45 20.26
104
Santa Barbara-Santa Maria-Goleta CA Santa Cruz-Watsonville CA Santa Rosa-Petaluma CA Stockton CA Vallejo-Fairfield CA Visalia-Porterville CA Yuba City CA Mother Lode Region of California nonmetropolitan area Eastern Sierra Region of California nonmetropolitan area North Coast Region of California nonmetropolitan area North Valley Region of California nonmetropolitan area Northern Mountains Region of California nonmetropolitan area Boulder CO Colorado Springs CO Denver-Aurora CO Fort Collins-Loveland CO Grand Junction CO Greeley CO Pueblo CO East and South Colorado nonmetropolitan area West Colorado nonmetropolitan area Northcentral Colorado nonmetropolitan area Central Colorado nonmetropolitan area Bridgeport-Stamford-Norwalk CT Danbury CT Hartford-West Hartford-East Hartford CT New Haven CT Norwich-New London CT-RI Springfield MA-CT Waterbury CT Worcester MA-CT Northwestern Connecticut nonmetropolitan area Eastern Connecticut nonmetropolitan area Dover DE Wilmington DE-MD-NJ Metropolitan Division Sussex County Delaware nonmetropolitan area Washington-Arlington-Alexandria DC-VA-MDWV Metropolitan Division
22.71 22.49 23.3 20.6 22.42 18.45 19.8 21.2 19.07 19.21 18.27 20.93 25.65 21.46 23.77 21.2 19.02 19.68 18.02 16.33 19.51 20.49 17.87 28.03 23.27 25.13 24.37 21.31 21.4 22.07 23.11 21.4 19.91 18.81 24.12 17.34 29.95
North and West Central New Mexico nonmetropolitan area Eastern New Mexico nonmetropolitan area Southwestern New Mexico nonmetropolitan area Los Alamos County New Mexico nonmetropolitan area Albany-Schenectady-Troy NY Binghamton NY Buffalo-Niagara Falls NY Elmira NY Glens Falls NY Ithaca NY Kingston NY Nassau-Suffolk NY Metropolitan Division New York-Northern New Jersey-Long Island NY-NJ-PA Poughkeepsie-Newburgh-Middletown NY Rochester NY Syracuse NY Utica-Rome NY Capital/Northern New York nonmetropolitan area East Central New York nonmetropolitan area Central New York nonmetropolitan area Southwest New York nonmetropolitan area Asheville NC Burlington NC Charlotte-Gastonia-Concord NC-SC Durham NC Fayetteville NC Goldsboro NC Greensboro-High Point NC Greenville NC Hickory-Lenoir-Morganton NC Jacksonville NC Raleigh-Cary NC Rocky Mount NC Virginia Beach-Norfolk-Newport News VA-NC Wilmington NC Winston-Salem NC Northeastern North Carolina nonmetropolitan area
15.91 17.01 16.6 36.42 22.24 19.84 20.2 19.37 18.66 21.89 19.75 24.45 26.48 22.25 20.77 20.87 18.58 18.3 18.94 18.46 18 17.71 17.07 21.46 25.59 17.56 16.35 19.05 18.17 16.86 16.42 21.54 16.5 19.92 18.43 19.62 16.49
105
Washington-Arlington-Alexandria DC-VA-MDWV Cape Coral-Fort Myers FL Crestview-Fort Walton Beach-Destin FL Deltona-Daytona Beach-Ormond Beach FL Fort Lauderdale-Pompano Beach-Deerfield Beach FL Metropolitan Division Gainesville FL Jacksonville FL Lakeland-Winter Haven FL Miami-Fort Lauderdale-Miami Beach FL Miami-Miami Beach-Kendall FL Metropolitan Division Naples-Marco Island FL North Port-Bradenton-Sarasota FL Ocala FL Orlando-Kissimmee FL Palm Bay-Melbourne-Titusville FL Palm Coast FL Panama City-Lynn Haven FL Pensacola-Ferry Pass-Brent FL Port St. Lucie FL Punta Gorda FL Sebastian-Vero Beach FL Tallahassee FL Tampa-St. Petersburg-Clearwater FL West Palm Beach-Boca Raton-Boynton Beach FL Metropolitan Division Northwest Florida nonmetropolitan area Northeast Florida nonmetropolitan area South Florida nonmetropolitan area Albany GA Athens-Clarke County GA Atlanta-Sandy Springs-Marietta GA Augusta-Richmond County GA-SC Brunswick GA Chattanooga TN-GA Dalton GA Gainesville GA Hinesville-Fort Stewart GA Macon GA
29.58 18.68 19.08 16.96 20.2 20.55 19.73 17.5 20.3 20.21 19.15 18.62 17.14 18.71 20.67 16.61 17.14 17.72 18.09 17.17 18.09 19.73 20.07 20.63 15.96 16.57 16.37 17.15 19.15 22.33 19.35 17.91 18.39 16.73 18.73 18.02 17.6
Other North Carolina nonmetropolitan area Western Central North Carolina nonmetropolitan area Western North Carolina nonmetropolitan area Bismarck ND Far Western North Dakota nonmetropolitan area West Central North Dakota nonmetropolitan area East Central North Dakota nonmetropolitan area Far Eastern North Dakota nonmetropolitan area Akron OH Canton-Massillon OH Cleveland-Elyria-Mentor OH Columbus OH Dayton OH Lima OH Mansfield OH Parkersburg-Marietta-Vienna WV-OH Sandusky OH Springfield OH Steubenville-Weirton OH-WV Toledo OH Wheeling WV-OH Youngstown-Warren-Boardman OH-PA West Northwestern Ohio nonmetropolitan area Other Ohio nonmetropolitan area Eastern Ohio nonmetropolitan area Southern Ohio nonmetropolitan area Lawton OK Oklahoma City OK Tulsa OK Northeastern Oklahoma nonmetropolitan area Northwestern Oklahoma nonmetropolitan area Southwestern Oklahoma nonmetropolitan area Southeastern Oklahoma nonmetropolitan area Bend OR Corvallis OR Eugene-Springfield OR Medford OR
16.39 17.09 15.98 18.23 18.63 17.43 16.04 16.57 19.74 17.54 20.59 21.03 20.39 18.29 17.62 17.23 16.64 17.84 16.39 18.98 16 17.42 17.35 16.78 16.45 17.01 16.75 18.83 18.65 15.85 15.99 16.17 15.7 19.05 22.65 19.66 18.96
106
Rome GA Savannah GA Valdosta GA Warner Robins GA North Georgia nonmetropolitan area Middle Georgia nonmetropolitan area East Georgia nonmetropolitan area South Georgia nonmetropolitan area Honolulu HI Hawaii / Maui / Kauai nonmetropolitan area Boise City-Nampa ID Coeur d'Alene ID Idaho Falls ID Lewiston ID-WA Logan UT-ID Pocatello ID North Idaho nonmetropolitan area Southwest Idaho nonmetropolitan area Southcentral Idaho nonmetropolitan area East Idaho nonmetropolitan area Bloomington-Normal IL Cape Girardeau-Jackson MO-IL Champaign-Urbana IL Chicago-Naperville-Joliet IL Metropolitan Division Chicago-Naperville-Joliet IL-IN-WI Danville IL Davenport-Moline-Rock Island IA-IL Decatur IL Kankakee-Bradley IL Lake County-Kenosha County IL-WI Metropolitan Division Peoria IL Rockford IL St. Louis MO-IL Springfield IL Northwest Illinois nonmetropolitan area West Central Illinois nonmetropolitan area East Central Illinois nonmetropolitan area South Illinois nonmetropolitan area Anderson IN
18.17 19.07 15.46 20.6 15.71 16.11 15.63 15.68 21.68 19.21 19.52 17.22 19.19 17.85 16.71 17.47 17.16 16.29 16.81 20.05 22.11 17.06 21.64 23.62 23.32 17.7 18.87 19.15 18.04 23.33 19.56 19.73 20.9 21.17 17.81 17.09 16.87 17.9 16.69
Portland-Vancouver-Beaverton OR-WA Salem OR Coastal Oregon nonmetropolitan area Southern Oregon nonmetropolitan area Eastern Oregon nonmetropolitan area Linn County Oregon nonmetropolitan area Altoona PA Erie PA Harrisburg-Carlisle PA Johnstown PA Lancaster PA Lebanon PA Philadelphia PA Metropolitan Division Philadelphia-Camden-Wilmington PANJ-DE-MD Pittsburgh PA Reading PA Scranton--Wilkes-Barre PA State College PA Williamsport PA York-Hanover PA Far Western Pennsylvania nonmetropolitan area West Central Pennsylvania nonmetropolitan area Northeastern Pennsylvania nonmetropolitan area East Central Pennsylvania nonmetropolitan area Aguadilla-Isabela-San Sebastian PR Fajardo PR Guayama PR Mayaguez PR Ponce PR San German-Cabo Rojo PR San Juan-Caguas-Guaynabo PR Yauco PR Puerto Rico nonmetropolitan area 1 Puerto Rico nonmetropolitan area 2 New Shoreham Town Rhode Island nonmetropolitan area Anderson SC Charleston-North CharlestonSummerville SC Columbia SC Florence SC
22.58 19.5 17.38 17.59 17.49 18.72 16.51 17.69 20.72 17.05 18.81 18.41 23.69 23.47 20.44 19.57 17.73 20.4 17.28 19.05 17.43 16.54 17.01 18.09 11.04 11.71 13.69 11.73 11.7 10.88 13.25 10.87 11.72 11.13 17.03 17.27 19.21 19.39 17.49
107
Bloomington IN Cincinnati-Middletown OH-KY-IN Columbus IN Elkhart-Goshen IN Evansville IN-KY Fort Wayne IN Gary IN Metropolitan Division Indianapolis-Carmel IN Kokomo IN Lafayette IN Louisville-Jefferson County KY-IN Michigan City-La Porte IN Muncie IN South Bend-Mishawaka IN-MI Terre Haute IN Northeast Indiana nonmetropolitan area Northwest Indiana nonmetropolitan area Southwest / Southeast Indiana nonmetropolitan area Ames IA Cedar Rapids IA Des Moines-West Des Moines IA Dubuque IA Iowa City IA Omaha-Council Bluffs NE-IA Sioux City IA-NE-SD Waterloo-Cedar Falls IA Northeast Iowa nonmetropolitan area Northwest Iowa nonmetropolitan area Southwest Iowa nonmetropolitan area Southeast Iowa nonmetropolitan area Kansas City MO-KS Lawrence KS Manhattan KS St. Joseph MO-KS Topeka KS Wichita KS Kansas nonmetropolitan area
17.14 20.5 19.95 17.75 18.78 18.74 19.08 20.54 20.29 18.94 19.39 16.79 18.06 18.89 16.92 16.56 16.48 17.08 20.08 19.77 20.72 17.34 20.11 19.76 16.14 17.87 16.22 16.16 15.62 16.59 21.18 17.55 17.09 16.83 18.62 19.02 15.93
Greenville-Mauldin-Easley SC Myrtle Beach-Conway-North Myrtle Beach SC Spartanburg SC Sumter SC Low Country South Carolina nonmetropolitan area Upper Savannah South Carolina nonmetropolitan area Pee Dee South Carolina nonmetropolitan area Lower Savannah South Carolina nonmetropolitan area Rapid City SD Sioux Falls SD Central South Dakota nonmetropolitan area Eastern South Dakota nonmetropolitan area Western South Dakota nonmetropolitan area Cleveland TN Jackson TN Johnson City TN Kingsport-Bristol-Bristol TN-VA Knoxville TN Morristown TN Nashville-Davidson--Murfreesboro-Franklin TN Western Tennessee nonmetropolitan area South Central Tennessee nonmetropolitan area North Central Tennessee nonmetropolitan area Eastern Tennessee nonmetropolitan area Abilene TX Amarillo TX Austin-Round Rock TX Beaumont-Port Arthur TX Brownsville-Harlingen TX College Station-Bryan TX Corpus Christi TX Dallas-Fort Worth-Arlington TX Dallas-Plano-Irving TX Metropolitan Division El Paso TX Fort Worth-Arlington TX Metropolitan Division Houston-Sugar Land-Baytown TX Killeen-Temple-Fort Hood TX
18.8 14.84 18.55 16.11 17.17 17.26 15.44 16.44 16.64 17.77 15.14 15.69 15.27 16.21 16.8 17.08 17.73 18.8 15.75 19.57 15.76 16.68 15.18 15.54 16.72 18.14 22.18 19.1 15.25 18.89 17.35 21.89 22.53 16.88 20.36 22.26 17.74
108
Bowling Green KY Clarksville TN-KY Elizabethtown KY Huntington-Ashland WV-KY-OH Lexington-Fayette KY Owensboro KY West Kentucky nonmetropolitan area South Central Kentucky nonmetropolitan area West Central Kentucky nonmetropolitan area East Kentucky nonmetropolitan area Alexandria LA Baton Rouge LA Houma-Bayou Cane-Thibodaux LA Lafayette LA Lake Charles LA Monroe LA New Orleans-Metairie-Kenner LA Shreveport-Bossier City LA Hammond nonmetropolitan area Natchitoches nonmetropolitan area Winnsboro nonmetropolitan area New Iberia nonmetropolitan area Bangor ME Lewiston-Auburn ME Portland-South Portland-Biddeford ME Portsmouth NH-ME Rochester-Dover NH-ME Northeast Maine nonmetropolitan area Southwest Maine nonmetropolitan area Baltimore-Towson MD Bethesda-Frederick-Gaithersburg MD Metropolitan Division Cumberland MD-WV Hagerstown-Martinsburg MD-WV Salisbury MD Upper Eastern Shore nonmetropolitan area Garrett County Maryland nonmetropolitan area St. Mary's County Maryland nonmetropolitan area Barnstable Town MA Boston-Cambridge-Quincy MA-NH Boston-Cambridge-Quincy MA NECTA Division Brockton-Bridgewater-Easton MA NECTA Division
17.38 16.91 17.9 17.06 18.92 16.91 17.09 15.69 17 17.25 16.89 18.9 18.24 17.87 17.43 16.7 19.72 17.62 16.48 16.14 16.1 17.61 18.68 17.84 20.52 23.15 20.6 16.73 18.39 24 28.06 17.99 18.64 19.61 17.85 16.31 29.12 21.31 27.19 28.56 22.22
Laredo TX Longview TX Lubbock TX McAllen-Edinburg-Mission TX Midland TX Odessa TX San Angelo TX San Antonio TX Sherman-Denison TX Tyler TX Victoria TX Waco TX Wichita Falls TX Northwestern Texas nonmetropolitan area North Central Texas nonmetropolitan area Eastern Texas nonmetropolitan area Central Texas nonmetropolitan area Southern Texas nonmetropolitan area Gulf Coast Texas nonmetropolitan area Ogden-Clearfield UT Provo-Orem UT St. George UT Salt Lake City UT Northern Utah nonmetropolitan area West Central Utah nonmetropolitan area South Western Utah nonmetropolitan area Eastern Utah nonmetropolitan area Burlington-South Burlington VT Southern Vermont nonmetropolitan area Northern Vermont nonmetropolitan area Blacksburg-Christiansburg-Radford VA Charlottesville VA Danville VA Harrisonburg VA Lynchburg VA Richmond VA Roanoke VA Winchester VA-WV Southwestern Virginia nonmetropolitan area Southside Virginia nonmetropolitan area Northeastern Virginia nonmetropolitan area
16.14 17.83 17.38 15.61 20.76 18.91 17.05 18.95 17.48 17.72 17.5 17.64 16.66 16.58 16.62 16.09 16.19 15.72 16.52 18.4 18.67 16.43 20.47 18.96 16.28 16.11 18.02 21.98 19.76 18.79 18.57 21.8 16.64 17.71 17.48 21.41 18.26 19.28 16.24 15.83 20.48
109
Framingham MA NECTA Division Haverhill-North Andover-Amesbury MA-NH NECTA Division Lawrence-Methuen-Salem MA-NH NECTA Division Leominster-Fitchburg-Gardner MA Lowell-Billerica-Chelmsford MA-NH NECTA Division Nashua NH-MA NECTA Division New Bedford MA Peabody MA NECTA Division Pittsfield MA Providence-Fall River-Warwick RI-MA Taunton-Norton-Raynham MA NECTA Division Nantucket Island and Martha's Vineyard nonmetropolitan area Southwest Massachusetts nonmetropolitan area Northwest Massachusetts nonmetropolitan area North Central Massachusetts nonmetropolitan area Ann Arbor MI Battle Creek MI Bay City MI Detroit-Livonia-Dearborn MI Metropolitan Division Detroit-Warren-Livonia MI Flint MI Grand Rapids-Wyoming MI Holland-Grand Haven MI Jackson MI Kalamazoo-Portage MI Lansing-East Lansing MI Monroe MI Muskegon-Norton Shores MI Niles-Benton Harbor MI Saginaw-Saginaw Township North MI Warren-Troy-Farmington Hills MI Metropolitan Division Upper Peninsula of Michigan nonmetropolitan area Northeast Lower Peninsula of Michigan nonmetropolitan area Northwest Lower Peninsula of Michigan nonmetropolitan area Balance of Lower Peninsula of Michigan nonmetropolitan area
27.86 21.7 21.58 19.67 26.67 23.36 19.84 22.73 20.28 21.62 21.67 21.93 18.66 20.16 23.64 23.44 19.78 18.1 22.85 22.64 19.37 19.72 18.67 19.29 18.92 21 19.13 18.03 18.89 18.99 22.49 17.49 15.99 17.51 17.97
Northwestern Virginia nonmetropolitan area Bellingham WA Bremerton-Silverdale WA Kennewick-Pasco-Richland WA Longview WA Mount Vernon-Anacortes WA Olympia WA Seattle-Bellevue-Everett WA Metropolitan Division Seattle-Tacoma-Bellevue WA Spokane WA Tacoma WA Metropolitan Division Wenatchee WA Yakima WA Northwestern Washington nonmetropolitan area Southwestern Washington nonmetropolitan area Central Washington nonmetropolitan area Eastern Washington nonmetropolitan area Charleston WV Morgantown WV Southern West Virginia nonmetropolitan area North Central West Virginia nonmetropolitan area Appleton WI Eau Claire WI Fond du Lac WI Green Bay WI Janesville WI Madison WI Milwaukee-Waukesha-West Allis WI Oshkosh-Neenah WI Racine WI Sheboygan WI Wausau WI Eastern Wisconsin nonmetropolitan area West Central Wisconsin nonmetropolitan area South Central Wisconsin nonmetropolitan area Southwestern Wisconsin nonmetropolitan area
17.02 19.92 22.44 22.91 20.4 20.13 22.19 26.25 25.57 20.24 21.94 18.56 18.38 19.4 18.7 18.66 20.5 18.25 17.78 16.21 16.08 18.95 17.84 18.49 19.42 18.43 21.8 21.64 19.44 18.34 18.92 18.48 17.69 17.92 16.54 16.96
110
Northern Wisconsin nonmetropolitan area Casper WY Cheyenne WY Northwestern Wyoming nonmetropolitan area Southwestern Wyoming nonmetropolitan area Northeastern Wyoming nonmetropolitan area Southeastern Wyoming nonmetropolitan area
16.61 20.37 19.57 18.14 21.03 20.32 18.73
Table B.4 Truck percentage by state
State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Dist. of Col. Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Percentage of trucks 11.8% 9.5% 12.0% 16.8% 8.6% 6.7% 6.5% 8.9% 3.5% 8.7% 9.8% 3.9% 14.6% 12.0% 14.5% 15.2% 12.6% 13.8% 15.7% 9.0% 8.5% 5.0% 7.5% 6.7% 13.7% 14.5% State Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming USA Average Percentage of trucks 13.0% 12.9% 8.4% 7.4% 7.6% 19.7% 7.8% 10.9% 17.8% 11.1% 15.0% 12.1% 10.5% 4.7% 10.1% 13.3% 11.6% 12.2% 19.2% 8.8% 7.6% 10.4% 12.7% 13.2% 19.9% 10.6%
111
9
References
Bachman, W. H. (1997). Toward a GIS-Based Modal Model of Automobile Exhaust Emissions. EPA Cooperative Agreement #: CR823020. Georgia Institute of Technology. Atlanta, Georgia. Available at: <http://www.epa.gov/nrmrl/pubs/600r98097/600R98097.pdf. Last Accessed: December, 2011. Barth et al, (2000). The Development of a Comprehensive Modal Emissions Model. NCHRP Project 25-11 Final report. Available at: http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_w122.pdf. Last Accessed: September, 2011. Bertini, R.L., Tantiyanugulchai, S., Anderson E., Lindgren, R., & Leal, M. (2001). Evaluation of Region 2 Incident Response Program Using Archived Data. Portland State University, Research Report PSU-CE-TRG-01-01. Blumentritt, C.W., Pinnell, C., McCasland, W.R., Ross, D.W., and Glazer, J. for Selection of Ramp Control Systems. NCHRP Report No. 232. Boyles, S., Fajardo, D., & Fajardo., D. (2006), Naïve Bayesian Classifier for Incident Duration Prediction Boyles, S., S. T. Waller. (2007). A Stochastic Delay Prediction Model for Real-Time Incident Management, ITE Journal 77, 18-24. Chang, G.L. & Rochon, S. (2006). Performance Evaluation of CHART – the Real Time Incident Management System in Year 2006. Maryland Department of Transportation. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates. Chou, C., & Miller-Hooks, E. (2008). Benefit-cost Analysis of Freeway Service Patrol Programs: Methodology and Case Study. Presentation at the 88th Annual Meeting of the Transportation Research Board, Maryland, 20. Chou, C. and E. Miller-Hooks, E. (2010). Violations Modeling in Studies of Concurrent-Flow Lanes. Transportation Research Record 2196, 75-82. Chou, C., & Miller-Hooks, E. (2010). Simulation-Based Secondary Incident Filtering Method. Journal of Transportation Engineering, Vol. 136, No.8, 746-754. CMEM, (2010). Comprehensive Modal Emissions Model (CMEM). Center for Environmental Research and Technology. Riverside, California. Available at: <http://www.cert.ucr.edu/cmem/>. Last Accessed: December, 2011. Cuciti, P., & Janson, B. (1995). Incident Management via Courtesy Patrol: Evaluation of a Pilot Program in Colorado. Transportation Research Record, 1494. Transportation Research Board, National Research Council, Washington DC. Donnell, E., Patten, M., & Mason, J. (1999). Evaluating a Roadside Assistance Program: PennLincoln Parkway Service Patrol. Transportation Research Record: Journal of the Transportation Research Board, 1683(-1), 143-149. Dougald, L.E., Demetsky, M.J., (2008). Assessing Return on Investment of Freeway Safety Service Patrol Programs. Transportation Research Record: Journal of the Transportation Research Board, 2047, 19-27. Garib, A., Radwan, A.E., & Al-Deek, H. (1997). Estimating magnitude and duration of incident delays. Journal of Transportation Engineering 123 (6), 459–466. GREET (2008). GREET Transportation Fuel Cycle Analysis Model, GREET 1.8b. Argonne National Laboratory. USDOE. 2008. Available at: <cta.ornl.gov/bedb/.../LowerHigher_Heating_Values_for_Various_Fuels.xls>. Last Accessed: January, 2012.
112
Guin, A., Porter, C., Smith, B., & Holmes, C. (2007). Benefits Analysis for Incident Management Program Integrated with Intelligent Transportation Systems Operations: Case Study. Transportation Research Record: Journal of the Transportation Research Board, 2000, 78-87. Hadi, M., Sinha, P., & Wang, A. (2007). Modeling Reductions in Freeway Capacity due to Incidents in Microscopic Simulation Models. Transportation Research Record: Journal of the Transportation Research Board, 1999(-1), 62-68. Choa, F. , Milam , R. T. , & Stanek D. (2002). Hagen, L., Zhou, H., & Singh, H. (2005). Road Ranger Benefit Cost Analysis. Florida Department of Transportation. Haghani, A., Iliescu, D., Hamedi, M. & Yang, S. (2006). Methodology for Quantifying the Cost Effectiveness of Freeway Service Patrol Programs: A Case Study. University of Maryland.Guidelines Han, L. D., Chin, S., & Hwang, H. (2003). Estimating Adverse Weather Impacts on Major U.S. Highway Network, Transportation Research Board, Washington, D.C. Hani S. Mahmassani, Jing Dong, Jiwon Kim, Roger B. Chen and Byungkyu (Brian) Park (2009). Incorporating Weather Impacts in Traffic Estimation and Prediction Systems, Final Report. Hawkins, P. (1993). Evaluation of the Southwest Freeway Motorist Assistance Program in Houston. Texas Transportation Institute, Texas Department of Transportation, Report No. 1922-1F. Hirunyanitiwattana, W., & Mattingly, S. (2006). Identifying Secondary Crash Characteristics for California Highway System. Presented at 85th Annual Meeting of the Transportation Research Board, Washington, D.C. Hobeika, A.G. & Dhulipala, S. (2004). Estimation of Travel Times on Urban Freeways under Incident Conditions, Transportation Research Record: Journal of the Transportation Research Board 1867, TRB, National Research Council, Washington D.C., 97-106 International Panel on Climate Change (IPCC), (2006). 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Available at: <http://www.ipccnggip.iges.or.jp/public/2006gl/index.html>. Last Accessed: November, 2011. Karlaftis, M. G., Richards, N. J., Latoski, S. P., & Sinha, K. C. (1998). Empirical analysis of secondary crash causes. Proc., 77th Annu. Meeting, Transportation Research Board, Washington, D.C. Karlaftis, M., S. Latoski, N. Richards, & K. Sinha. (1999). ITS Impacts on Safety and Traffic Management: An Investigation of Secondary Crash Causes. ITS Journal, Vol. 5, No. 1, 39–52. Khattak, A., J. Schofer, and M. Wang. (1995) A simple time-sequential procedure for predicting freeway incident duration. IVHS Journal, Vol. 2, , 113-138. Khattak, A., Wang, X., & Zhang, H. (2008). Are Incident Durations and Secondary Incidents Interdependent Presented at the 2009 Transportation Research Board Annual Meeting, Washington, D.C. Kim, W., Chang, G., & Rochon, S. M.(2008)Analysis of Freeway Incident Duration for ATIS Application, Presented the 15th world congress on intelligent transportation system, New York Koupal et al., (2005). Koupal, J, Landamn,l, Nam,E, Scarbro,C, Glover,E and Ginannelli, R. MOVES Energy and Emissions Inputs. Office of Transportation and Air Quality. Available at: http://www.epa.gov/otaq/models/ngm/420p05003.pdf. Last Accessed: December, 2011. Latoski, S. P., Pal, R., & Sinha K.C. (1999). Cost-Effectiveness Evaluation of Hoosier Helper Freeway Service Patrol. Journal of Transportation Engineering, 125(5), 429-437.
113
Lindley, J.A. (1987). Urban Freeway Congestion: Quantification of the Problem and Effectiveness of Potential Solution. ITE Journal 57 (1), 27-32. Martin Fellendorf , Peter Vortisch, (2001) Validation of the Microscopic Traffic Flow Model VISSIM in Different Real-Workd Situations. Annual Meeting TRB. Masinick, J. P., & Tang, H. (2004).An analysis on the Impact of Rubbernecking onUrban Freeway Traffic, Research Report No. UVACTS-15-0-62. Mauch, M., Ahn, S., Chung, K., & Skabardonis, A. (2005). Baseline Evaluation of the Freeway Service Patrol (FSP) I-710 Big-Rig Demonstration Program. Institute of Transportation Studies, University of California-Bekerley, Working Paper UCB-ITS-WP-2005-3. Minnesota Department of Transportation, (2004). FIRST Program Evaluation. ITS annual report. Moore, J., G. Giuliano, & Cho, S. (2004). Secondary Accident Rates on Los Angeles Freeways. Journal of Transportation Engineering, Vol. 130, No. 3, 280–285. Murty, A. S. R., & Verma, A. K. (1986). Inverse normal and lognormal distributions for reliability design. Reliability Engineering, 15(1), 55-60. Nam, D., & Mannering, F. (2000). An exploratory hazard-based analysis of highway incident duration. Transportation Research Part A: Policy and Practice 34 (2), 85–102. Office of Transportation and Air Quality (OTAQ), 2012. Regulations and Standards. Available at: <http://www.epa.gov/otaq/climate/regulations.htm>. Last Accessed: January, 2012. Pulugurtha, S., Dangeti, M., Nambisan, S., & Kaseko, M.(2002). Simulating and Analyzing 27 Incidents Using CORSIM and VISSIM Traffic Simulation Software. Proceedings of the 28 Seventh International Conference, Applications of Advanced Technologies in Transportation. Rakha, H., Farzaneh, M., Arafeh, M., Hranac, R., Sterzin, E. & Krechmer, D. (2007). Empirical Studies on Traffic Flow in Inclement Weather, Final Report - Phase I. Raub, R. A., (1997). Occurrence of Secondary Crashes on Urban Arterial Roadways. Transportation Research Record 1581, TRB, National Research Council, Washington, D.C., 1997. pp. 53–58. Roadway Design Manual, 2010. Saka, A. A., Jeihani, M., & James, P. A. (2008) Estimation of Traffic Recovery Time for Different Flow Regimes on Freeways. Final Report No. SP708B4L. July 2008 Shankar, V., Mannering, F., & Barfield, W. (1995). Effect of roadway geometrics and environmental factors on rural freeway accident frequencies. Accident Analysis & Prevention, 27(3), 371-389. Skabardonis, A., & Mauch, M. (2005). SSP Beat Evaluation and Predictor Models: User’s Manual. Institute of Transportation Studies, University of California-Berkeley, Research Report: UCB-ITSRR-2005-XX. Skabardonis, A., Noeimi, H., Petty, K., Rydzewski, D., Varaiya, P., & Haitham, A. (1995). Freeway Service Patrol Evaluation. Institute of Transportation Studies, University of CaliforniaBerkeley, California PATH Research Report No. UCB-ITS-PRR-95-5. Skabardonis, A., Petty, K., Varaiya, P., & Bertini, R. (1998). Evaluation of the Freeway Service Patrol (FSP) in Los Angeles. Institute of Transpotation Studies, University of California-Berkeley, California PATH Research Report No. UCB-ITS-PRR-98-31.
114
Smith, K., & Smith, B.L., (2001). Forecasting the Clearance Time of Freeway Accidents. No. STL2001-01, Center for Transportation Studies (University of Virginia), Charlottesville, VA. Sun, C. (2007). Secondary Accident Data Fusion for Assessing Long-Term Performance of Transportation Systems. Midwest Transportation Consortium, Center for Transportation Research and Education, Cedar Falls, Iowa. Sun, C., Chilukuri, V., Ryan, T., & Trueblood, M. (2010). Evaluation of Freeway Motorist Assist Program. University of Missouri-Columbia, Final Report RD 09-004. Sun,C., Chilukuri, V., (2010). Dynamic Incident Progression Curve for Classifying Secondary Traffic Crashes. Journal of Transportation Engineering, Vol. 136, No. 12, 1153-1158. Traffic Incident Management Handbook, 2000 USDOE, (2011). Alternative Fuels & Advanced Vehicles Data Center. Energy Efficiency & Renewable Energy. USDOE. Available at: <http://www.afdc.energy.gov/afdc/>. Last Accessed: December, 2011. USDOE, (2012). Fuel Economy. Energy Efficiency & Renewable Energy. USDOE. Available at: < http://www.fueleconomy.gov/feg/findacar.shtml>. Last Accessed: January, 2012. USEPA, (2004). Update of Methane and NitroUSOxide Emission Factors on On-Highway Vehicles. USEPA Report Number: EPA420-P-04-016. Available at: < http://www.epa.gov/otaq/models/ngm/420p04016.pdf >. Last Accessed: January, 2012. USEPA, (2006). High Global Warming Potential (GWP) Gases. Available at: http://www.epa.gov/highgwp/scientific.html. Last Accessed: January, 2011. USEPA, (2007). Emission Facts: Average Annual Emissions and Fuel Consumption for Passenger Cars and Light Trucks. Consumer Information. USEPA. Available at: < http://www.epa.gov/oms/consumer/f00013.htm> . Last Accessed: January, 2012. USEPA, (2009). Transportation and Air Quality. Available at: http://www.epa.gov/otaq/. Last Accessed: August, 2011. USEPA, (2011). Motor Vehicle Emissions Model (MOVES). Modeling and Inventories. USEPA. Available at: <http://www.epa.gov/otaq/models/moves/index.htm>. Last Accessed: January, 2012. Wang, Y.H., Hallenbeck, M., & Cheevarunothai, P. (2008).Quantifying Incident-Induced 21 Travel Delays on Freeways Using Traffic Sensor Data Washington State Department of 22 Transportation, Report No. TNW2008-02, 23 Wirasinghe, S. C. (1978) Determination of traffic delays from shock-wave analysis. Transportation Research 12, 343-348. Zhan, C., A. Gan, and M. Hadi., (2009). Identifying Secondary Crashes and Their Contributing Factors. In Transportation Research Record: Journal of the Transportation Research Board, No. 2102, Transportation Research Board of the National Academies, Washington, D.C., 68-75. Zhan, C., Shen, L., Hadi, M., and Gan, A. (2008) . Understanding the characteristic of secondary crashes on freeways. CD-ROM, Transportation Research Board, Washington, D.C. Zhang, H., & Khattak, A. (2010). Analysis of Cascading Incident Event Durations on Urban Freeways. Transportation Research Record: Journal of the Transportation Research Board, No. 2178, Transportation Research Board of the National Academies, Washington, D.C., pp. 30–39.
115
doc_587649159.pdf