Microscopic Assessment Of Transportation Emissions On Limited Access Highways

Transcription

1 University of Central Florida Electronic Theses and Dissertations Doctoral Dissertation (Open Access) Microscopic Assessment Of Transportation Emissions On Limited Access Highways 2012 Hatem Abou-Senna University of Central Florida Find similar works at: University of Central Florida Libraries Part of the Civil Engineering Commons STARS Citation Abou-Senna, Hatem, "Microscopic Assessment Of Transportation Emissions On Limited Access Highways" (2012). Electronic Theses and Dissertations This Doctoral Dissertation (Open Access) is brought to you for free and open access by STARS. It has been accepted for inclusion in Electronic Theses and Dissertations by an authorized administrator of STARS. For more information, please contact

2 MICROSCOPIC ASSESSMENT OF TRANSPORTATION EMISSIONS ON LIMITED ACCESS HIGHWAYS by HATEM AHMED ABOU-SENNA B.S. Cairo University, 1993 M.E. Cairo University, 2000 M.S. University of Central Florida, 2003 A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Civil, Environmental and Construction Engineering in the College of Engineering and Computer Science at the University of Central Florida Orlando, Florida Fall Term 2012 Major Professor: Ahmed E. Radwan

3 2012 Hatem A. Abou-Senna ii

4 ABSTRACT On-road vehicles are a major source of transportation carbon dioxide (CO 2 ) greenhouse gas emissions in all the developed countries, and in many of the developing countries in the world. Similarly, several criteria air pollutants are associated with transportation, e.g., carbon monoxide (CO), nitrogen oxides (NOx), and particulate matter (PM). The need to accurately quantify transportation-related emissions from vehicles is essential. Transportation agencies and researchers in the past have estimated emissions using one average speed and volume on a long stretch of roadway. With MOVES, there is an opportunity for higher precision and accuracy. Integrating a microscopic traffic simulation model (such as VISSIM) with MOVES allows one to obtain precise and accurate emissions estimates. The new United States Environmental Protection Agency (USEPA) mobile source emissions model, MOVES2010a (MOVES) can estimate vehicle emissions on a second-by-second basis creating the opportunity to develop new software VIMIS 1.0 (VISSIM/MOVES Integration Software) to facilitate the integration process. This research presents a microscopic examination of five key transportation parameters (traffic volume, speed, truck percentage, road grade and temperature) on a 10-mile stretch of Interstate 4 (I- 4) test bed prototype; an urban limited access highway corridor in Orlando, Florida. iii

5 The analysis was conducted utilizing VIMIS 1.0 and using an advanced custom design technique; D-Optimality and I-Optimality criteria, to identify active factors and to ensure precision in estimating the regression coefficients as well as the response variable. The analysis of the experiment identified the optimal settings of the key factors and resulted in the development of Micro-TEM (Microscopic Transportation Emissions Meta- Model). The main purpose of Micro-TEM is to serve as a substitute model for predicting transportation emissions on limited access highways in lieu of running simulations using a traffic model and integrating the results in an emissions model to an acceptable degree of accuracy. Furthermore, significant emission rate reductions were observed from the experiment on the modeled corridor especially for speeds between 55 and 60 mph while maintaining up to 80% and 90% of the freeway s capacity. However, vehicle activity characterization in terms of speed was shown to have a significant impact on the emission estimation approach. Four different approaches were further examined to capture the environmental impacts of vehicular operations on the modeled test bed prototype. First, (at the most basic level), emissions were estimated for the entire 10-mile section by hand using one average traffic volume and average speed. Then, three advanced levels of detail were studied using VISSIM/MOVES to analyze smaller links: average speeds and volumes (AVG), second-bysecond link driving schedules (LDS), and second-by-second operating mode distributions (OPMODE). This research analyzed how the various approaches affect predicted emissions of CO, NOx, PM and CO 2. iv

6 The results demonstrated that obtaining accurate and comprehensive operating mode distributions on a second-by-second basis improves emission estimates. Specifically, emission rates were found to be highly sensitive to stop-and-go traffic and the associated driving cycles of acceleration, deceleration, frequent braking/coasting and idling. Using the AVG or LDS approach may overestimate or underestimate emissions, respectively, compared to an operating mode distribution approach. Additionally, model applications and mitigation scenarios were examined on the modeled corridor to evaluate the environmental impacts in terms of vehicular emissions and at the same time validate the developed model Micro-TEM. Mitigation scenarios included the future implementation of managed lanes (ML) along with the general use lanes (GUL) on the I-4 corridor, the currently implemented variable speed limits (VSL) scenario as well as a hypothetical restricted truck lane (RTL) scenario. Results of the mitigation scenarios showed an overall speed improvement on the corridor which resulted in overall reduction in emissions and emission rates when compared to the existing condition (EX) scenario and specifically on link by link basis for the RTL scenario. The proposed emission rate estimation process also can be extended to gridded emissions for ozone modeling, or to localized air quality dispersion modeling, where temporal and spatial resolution of emissions is essential to predict the concentration of pollutants near roadways. v

7 IN THE NAME OF GOD, THE MOST COMPASSIONATE, THE MOST MERCIFUL To my dearest parents Abla and Ahmed Whose constant love, kindness and support made me the person I am To my precious grandmother Nonna Who had supported me with her endless love and care To my gorgeous wife Ghada Whose eternal love and patience gave me hope and happiness To my marvelous children Nada and Adham Whose smiles and hugs were nurturing To my beloved sister Lamiaa You are all that a brother could ask for To all my family and friends Who gave me joy and precious memories through tough moments vi

8 ACKNOWLEDGMENTS First, I praise God for giving me the strength and patience to accomplish this monumental task. I would like to express my deepest gratitude for the constant support and guidance from my committee chair, advisor, and friend Dr. Essam Radwan throughout my PhD study and throughout the years at UCF especially during the toughest times of my life. His continuous support, insights, and suggestions contributed in large measure to the success of this research. I would like also to thank my committee members Dr. Haitham Al-Deek, Dr. Mohammed Abdel-Aty, Dr. David Cooper and Dr. Mark Johnson for their help and support in conducting this research and for serving in my committee. I sincerely thank Dr. Cooper for his valuable insights with the environmental portion as well as Dr. Johnson for his help with the experimental design portion which ignited the whole idea about this research. Dr. Aty s help with my candidacy procedure was very supportive and Dr. Al-Deek s understanding during my GTA was thoughtful. I can t thank enough Dr. Hesham Eldeeb who committed his time and energy to develop the key software for this research. Words can t express my gratitude towards his contribution and assistance throughout my study. I would like also to thank the staff members of the Civil Engineering Department for providing me an opportunity to pursue the graduate program and for their help. I would like to express my heartfelt appreciation to my parents, sister and all my family for their endless love, precious advice and support throughout the years. My wife s eternal love helped me to overcome tough moments. My grateful thanks are due to Orange County Transportation Planning staff for their help, support and understanding during my internship at the County. I m also grateful to Dr. Adel El-Safty for his support during tough times and backing. Also, my stay at UCF has given me many moments and experiences that I will cherish for all times to come. My friends and colleagues from the ITS Lab have confirmed my belief that there is always something to learn from everybody. vii

9 TABLE OF CONTENTS LIST OF FIGURES... x LIST OF TABLES... xii 1. INTRODUCTION Background Research Objective Research Value to Practitioners Research Response to Current or Future Needs Statewide LITERATURE REVIEW Greenhouse Gas Components and Climate Change Criteria Pollutants and Health Effects Transportation Greenhouse Gas Emissions & CO 2 Share Calculating CO 2 Emissions Feasibility of Field Capturing CO US CAFE Program Overview of Joint EPA/NHTSA National Program Summary of the Joint Final Rule Previous Studies to Model and Estimate Traffic Emissions EPA Emissions Models and Analysis Tools Softwares Examples of Non-EPA Emissions Models RESEARCH APPROACH Design of Experiments (DOE) Development of Calibrated Base Scenario Using VISSIM Model The VISSIM Model Calibration & Validation of VISSIM Model Estimation of Scenario-Based Emissions using MOVES Model The MOVES Model Validation of MOVES Model Statistical Analysis Using JMP Software Development of Emission Prediction Model (Micro-TEM) Application of Mitigation Strategies Findings of Research Results and Conclusions VISSIM/MOVES INTEGRATION SOFTWARE (VIMIS) Overview Modules Description viii

10 5. EVALUATION OF THE I-4 CORRIDOR Overview of the I-4 Downtown Corridor Model Calibration Statistical Analysis Model Validation DEVELOPING MICROSCOPIC EMISSION PREDICTION MODEL Overview Design of Experiments (DOE) Test Bed Modeling Moves Project Level Data Operating Modes & Link Driving Schedules Design Settings versus Actual Settings Analysis of Results Discussion Meta Model for Transportation Emissions Micro-TEM Introduction to Meta Models Micro-TEM EMISSION ESTIMATION APPROACHES Overview VISSIM Input/Output Data Moves Project Level Data Vehicle Activity Characterization Average Speeds, Link Drive Schedules & Operating Modes Vehicle Specific Power (VSP) Emissions Results and Analysis Discussion MODEL APPLICATIONS Overview Managed Lanes (ML) Restricted Truck Lanes (RTL) Evaluation of Scenarios CONCLUSIONS AND RECOMMENDATIONS APPENDIX A CUSTOM DESIGN APPENDIX B ANALYSIS OF CUSTOM DESIGN OUTPUT BY LINK APPENDIX C I-4 PROTOTYPE OPERATING MODE DISTRIBUTIONS REFERENCES ix

11 LIST OF FIGURES Figure 1-1: U.S. Emissions of CO 2 by Energy Consuming Sector and Fuel Type (2006)... 8 Figure 1-2: U.S. GHG Emissions by Gas Type, 2008 (MMT of CO 2 equivalent)... 8 Figure 2-1: U.S. Transportation Greenhouse Gas Emissions by Gas, CO2e (2006) Figure 2-2: U.S. Greenhouse Gas Emissions by End Use Economic Sector, Figure 2-3: U.S. Greenhouse Gas Emissions from Transportation by Mode, Figure 2-4: Vehicle Miles Traveled by Light Duty Vehicles Figure 2-5: GHG Emissions from US Freight Sources Figure 3-1: I-4 Downtown Corridor (Orlando, FL) Figure 3-2: Application of Managed Lanes in VISSIM Figure 3-3: Research Approach Scheme Figure 4-1: VIMIS 1.0 Software Figure 4-2: Design File for Input into VIMIS Figure 4-3: VISSIM Module in VIMIS Figure 4-4: OPMODE Module in VIMIS Figure 4-5: MOVES Module in VIMIS Figure 5-1: I-4 Downtown Corridor and Master Link Count Locations Figure 5-2: Small Portion of Network Overlaid on Aerial Map Figure 5-3: Peak Hour Variable Speed Limits on I Figure 5-4: DMS Travel Time Information on I Figure 5-5: Field Congestion on I-4 at SR 408 Off Ramp Figure 5-6: Simulated Congestion on I-4 at SR 408 Off Ramp Figure 6-1: Test-bed Prototype of the I-4 Corridor Figure 6-2: Summary of Stepwise Regression for Initial Model (D-Design) Figure 6-3: Summary of Stepwise Regression for Final Model (D-Design) Figure 6-4: Validation of the Regression Model by Link x

12 Figure 6-5: Summary of Stepwise Regression for Final Model (I-Design) Figure 6-6: Prediction Variance for the Design Factors at Center Points Figure 6-7: Prediction Variance for the Design Factors at Center Points (Log Space) Figure 6-8: Interaction Profiles for the Design Factors at Optimal Settings (Log Space) Figure 6-9: Prediction Variance for the Design Factors at Optimal Settings (Log Space). 108 Figure 6-10: Speed - CO 2 Emission Rates Relationship Figure 6-11: Speed-CO 2 Emission Rates at Different Temp, Truck & Grade Levels Figure 6-12: Volume-CO 2 Emission Rates at Different Temp, Truck & Grade Levels Figure 6-13: Traffic Volume CO 2 Emission Rates Relationship (0%Trucks- 0%Grade) 120 Figure 6-14: Speed Spectrum on Volume-CO 2 Emission Rate Curves Figure 6-15: Stochastic Speed Density Relationship Figure 6-16: CO 2 Surface Profiler for the Predicted model Figure 7-1: Test-bed Prototype of the I-4 Corridor Figure 7-2: Total Emissions by Vehicle Type & Estimation Approach Figure 7-3: Emissions Variation on Corridor Links for PC by Estimation Approach Figure 7-4: Link Operating Mode Distribution by Vehicle Type on Selected Links Figure 8-1: Pollutant Emissions Comparisons by Scenario Figure 8-2: Emissions Scenario Comparison by Vehicle Type Figure 8-3: Link Emission Rate Comparison RTL vs. EX Scenarios xi

13 LIST OF TABLES Table 2-1: National Emissions Estimates Table 2-2: US Transportation Sector Green House Gas Emissions, Table 3-1: Partial Layout of a Generic Experimental Design Table 3-2: Summary of Project Level Parameters Table 5-1: Master Link Counts Comparison Based on Best Seed Number Table 5-2: Master Link Counts Comparison Based on Other Seed Numbers Table 5-3: Paired T-test of Actual vs. Simulated Data Table 5-4: Network Evaluation for I-4 during Peak Hour Table 6-1: Factors and Levels Table 6-2 Partial Layout of D-Optimal Design for Five Seven-Level Continuous Factors.. 89 Table 6-3: Summary of Project Level Parameters Table 6-4: MOVES Speed Bins Table 6-5: Design Setting Versus Actual Setting Table 6-6: Volume-Speed-CO 2 Emission Rates at Zero Truck and Zero Grade Levels Table 7-1: Excerpt from VISSIM Vehicle Trajectory Data Table 7-2: Summary of project level parameters Table 7-3: Emissions by pollutant, source type, link & vehicle activity characterization Table 7-4: Link emissions per vehicle-mile by source type Table 8-1: Pollutant Emissions Comparison by Scenario Table 8-2: MOE Scenario Comparisons xii

14 1. INTRODUCTION 1.1 Background Emissions of greenhouse gases (GHGs), primarily carbon dioxide (CO 2 ), are contributing to global climate change, which is believed by many to be one of the most critical environmental issues facing the world this century. CO 2 from transportation is expected to remain the major source of total U.S. greenhouse gas emissions (IPCC 2008). To help Florida reduce GHGs, the state adopted the California motor vehicle emission standards for GHG in July 2007 (Executive Order ). Transportation as a whole represents about 40 percent of Florida's total GHG emissions, second only to the electric utility sector. Moreover, ambient air quality standards have been established for several pollutants associated with transportation, including carbon monoxide (CO) and particulate matter (PM-10 and PM-2.5). In addition to these criteria pollutant emissions, motor vehicles emit volatile organic compounds (VOCs) and nitrogen oxides (NOx), both of which are ozone precursors. Nationally, on-road transportation sources are responsible for 21 percent of VOCs emissions, 32 percent of NOx emissions, and 50 percent of CO emissions (NEI Trends, 2008). Transportation agencies and researchers have a long history of implementing techniques to calculate transportation-related emissions. Traditional methods for creating 1

15 emission inventories utilized annual average estimates. One comparison of annual estimates with monthly estimates of vehicular emission provided similar results, implying that detailed calculations were not necessary for annual emissions inventories (Cooper and Arbrandt, 2004). Travel demand models have been utilized to provide an intermediate level of detail (daily values). However, static planning models were found to ignore individual vehicle activity, which leads to underestimation of pollutant emissions, as they do not account for link capacity and other dynamic variables. As a result, estimates of emissions based on static planning models suffer from significant biases in different traffic conditions (You, et al. 2010). Currently, more accuracy has been established using microscopic analyses through the reduction of time and distance scales while splitting the network links into sub-links and utilizing second-by-second operations to calculate emissions. As stated in the USEPA report (2006): The emission factors (EFs) used in the U.S. GHG Inventory for highway vehicles are based on laboratory testing of vehicles. Although the measured testing environment simulates actual driving conditions, results merely approximate real world vehicle activity and interactions due to the stochastic nature of the transportation system. The USEPA reported that for some vehicle and control technology types, the testing did not yield statistically significant results within the 95 percent confidence interval, requiring reliance on expert judgment when developing the EFs. In those cases, the EFs were developed based on comparisons of fuel 2

16 consumption between similar vehicle and control technology categories (USEPA, 2006). Since 95% of transportation GHG emissions are in the form of CO 2 (USEPA, 2009), uncertainty in the CO 2 estimates has a much greater effect on the transportation sector estimates than uncertainty associated with nitrous oxide (N 2 O), methane (CH 4 ), or hydrofluorocarbon (HFC) emissions. Other vehicular pollutants are important in spite of their small contribution to the total. CO is a criterion pollutant with two national ambient air quality standards (NAAQS), and is used in project level analyses. NOx is a criterion pollutant that is crucial due to its role (along with volatile organic compounds VOCs) in ozone formation. Despite all the past studies conducted on GHG and criteria pollutants emissions, there is still a crucial need to identify and examine microscopically, key transportation-related parameters that contribute to vehicle emissions. Several analyses have been conducted in the literature to identify the main factors that contribute to the increase in vehicular emissions. The majority of these factors are found to be traffic related. However, the transportation system encompasses various disciplines with different traffic, planning, design, and environmental factors. Therefore, based on the literature findings, key parameters that are traffic-related (traffic volume, truck percentage, speed limits); geometry-related (road grade) and environment-related (temperature) are selected for detailed evaluation in this research. 3

17 1.2 Research Objective The primary objective of this research is to present a more sophisticated approach based on stochastic microscopic simulation of traffic and air quality impacts to identify major key factors contributing to vehicle emissions through experimental design methodologies. Several studies have shown that microscopic simulation provides better estimates of vehicular emissions as it models explicitly second by second vehicles accelerations/decelerations, lane changing and merging/diverging, which are typical in congested conditions. This research developed a framework for utilizing assessment tools to estimate CO 2 GHG emissions as well as CO, NOx and PM pollutant emissions and then used this process to apply certain mitigation strategies to a severely congested downtown corridor in Orlando, Florida, along Interstate 4 (I-4). Mitigation strategies included developing simulation scenarios that are focused on testing proposed future improvements to the downtown corridor in central Florida. For example, the introduction of Managed Lanes (ML) on the I-4 corridor was modeled using the VISSIM model and emissions were calculated from the simulated traffic during peak periods using MOVES, the latest EPA emissions model. The research developed techniques to enhance the current methodology in calculating emissions factors on limited access highways to ensure that adequate 4

18 mitigation is provided. This research is important because it contributed to improving the interface between traffic simulation models and the next generation of modal emissions models MOVES2010; measuring real-world second-by-second vehicle dynamics and comparison to corresponding simulated conditions. The main objectives of the dissertation can be summarized, as follows: 1- To understand the contribution of key transportation parameters to vehicular emissions and air pollution in a stochastic-microscopic environment via experimental design methodologies. 2- To develop and assess a new technique through a model that identifies and calculates vehicular emissions on limited access highways. 3- To improve the interface between traffic simulation models and the next generation modal emissions model MOVES To evaluate operational improvements and assess mitigation strategies that advance the congestion management process on limited access highways. 1.3 Research Value to Practitioners The research methodology provided detailed information on traffic parameters as well as air quality issues to a reasonable level of detail, taking into account modeling capabilities and the cost of acquiring data. By examining the input and output parameters 5

19 of traffic and emissions models, the research identified the most efficient forms of connection between them and the possibility of developing a hierarchy of models. The developed model has a scenario modeling capability which could be used to inform practitioners of the potential effectiveness of proposed mitigation strategies and measures. Furthermore, the proposed method provided traffic practitioners with improved emission estimates on limited access highways based on a microscopic-stochastic approach. The methodology can also be expanded to include other types of street characteristics and traffic conditions such as arterials and emission processes. 1.4 Research Response to Current or Future Needs Statewide According to the Florida Department of Environmental Protection (FDEP, 2011) website, The Florida Clean Car Emission Rule, , F.A.C., came into effect on February 15, This rule, however, will only apply to future makes and models of passenger cars, light-duty trucks, and sport utility vehicles. The U.S. Environmental Protection Agency (USEPA) is considering whether to take action regarding GHG, and the National Highway Traffic Safety Administration (NHTSA) has proposed Corporate Average Fuel Economy (CAFE) standards that would achieve GHG reductions, indirectly, through raising the federal standard for minimum miles per gallon (mpg). Once enacted, the CAFE standards will apply in Florida (FDEP, 2011). Therefore, understanding the contribution of these types of vehicles emissions to GHG along with 6

20 CAFE standards will result in greater GHG reductions compared to just relying on the federal CAFE rules. Figure 1-1 and Figure 1-2 show two graphics that have been prepared as part of (Cooper and Alley, 2010) 4th edition of the air pollution control textbook. These graphs demonstrate how much CO 2 is emitted in the U.S. and how much from each sector. The environmental impacts of transportation systems are significant, responsible for 20% to 25% of the world s energy consumption and carbon dioxide emissions every year. The social and economic impacts of transportation systems have formed a strong need for a sustainable transportation system which can meet the increasing traffic mobility needs by building new infrastructure such as Managed Lanes while minimizing the negative effects from GHG emissions to the society, economy, and environment. 7

21 Figure 1-1: U.S. Emissions of CO 2 by Energy Consuming Sector and Fuel Type (2006) Figure 1-2: U.S. GHG Emissions by Gas Type, 2008 (MMT of CO 2 equivalent) 8

22 2. LITERATURE REVIEW This section is divided into several parts. First, a brief background on green house gases (GHG) and criteria pollutants is given and their effect on climate change and human health is explained. Second, detailed information on transportation GHG emissions and in particular the impacts of carbon dioxide (CO 2 ) emissions with respect to transportation factors are provided. Third, information on carbon content and how CO 2 emissions are calculated is discussed. Synopsis of US corporate average fuel economy (CAFE) program and its final rule are presented in the next section. Examples of past research about traffic emissions models, micro-simulation and transportation factors are discussed and finally, the development of EPA and non-epa emissions models and software tools throughout the years is presented. It should be noted that a majority of the information provided in this section is obtained from the US EPA website, Intergovernmental Panel on Climate Change (IPCC 2008) and previous reports. 2.1 Greenhouse Gas Components and Climate Change Common Greenhouse gases (GHG) include carbon dioxide (CO 2 ), methane (CH 4 ), nitrous oxide (N 2 O), ozone, water vapor, and chlorofluorocarbons (CFC). Many of these gases are naturally occurring and are necessary to maintain an atmospheric 9

23 temperature that supports human life (IPCC, 1996). However, GHGs trap heat in the earth s atmosphere. GHGs are produced by both natural and human activities and can be removed through natural processes as well. However, human-produced GHGs have significantly exceeded natural absorption rates since the industrial revolution due to the increased combustion of fossil fuel. Unlike other pollutants, CO 2 as well as other GHGs take several years to leave the atmosphere. Atmospheric lifetimes are estimated to be years for CO 2, 9-15 years for CH 4, and 120 years for N 2 O (IPCC, 1996). The combinations of the previously mentioned conditions (fossil fuel combustion, deforestation and atmospheric decay) have contributed to the increased concentration of these gases. Since the beginning of the industrial revolution, atmospheric concentrations of CO 2 have increased by 36 percent, CH 4 concentrations have more than doubled, and N 2 O concentrations have risen by approximately 18 percent (IPCC, 2007). Human activities over the past 70 years have also produced synthetic chemicals that are powerful greenhouse gases with atmospheric lifetimes ranging from years to millennia (IPCC, 1996). These substances include hydroflurocarbons (HFCs), chlorofluorocarbons (CFCs) and sulfur hexafluoride (SF 6 ). GHG emissions are projected to continue to rise. The Intergovernmental Panel on Climate Change (IPCC) estimates that in the absence of additional climate policies to reduce GHG emissions, baseline global GHG emissions will increase anywhere from 25 to 90 percent between the years 2000 and 2030, with CO 2 10

24 emissions from energy use growing between 40 and 110 percent over the same period (IPCC, 2007). According to the IPCC, Warming of the climate system is unequivocal, as is now evident from observations of increases in global average air and ocean temperatures, widespread melting of snow and ice, and rising global average sea level (IPCC, 2007). The IPCC s report also describes the anticipated consequences of climate change with potential temperature increases above 2 C (3.6 F). According to the IPCC (2007), global GHGs must be reduced to 50-to-85 percent below year 2000 levels by 2050 to keep warming to 2.0-to-2.4 C (3.6-to-4.3 F). 2.2 Criteria Pollutants and Health Effects The Clean Air Act requires EPA to set National Ambient Air Quality Standards (NAAQS) for six common air pollutants which are ozone, particulate matter (PM), carbon monoxide (CO), nitrogen oxides (NOx), sulphur dioxide (SO 2 ) and lead (Pb). These are commonly known as "criteria pollutants". Significant portions of mobile source emissions are composed mainly of three of these criteria pollutants primarily CO, NOx, PM and one other class of pollutants volatile organic compounds (VOCs) (Air Emission Sources, 2011). 11

25 Vehicle emissions also contribute to the formation of photochemical smog (smoke and fog). During the hot season, pollutants such as NO x and VOCs react to form ground level ozone (O 3 ) and other pollutants. During the cold season, vehicle emissions can be trapped near the ground during winter time, a phenomenon known as temperature inversion where colder air is trapped beneath a layer of warmer air. This phenomenon leads to high concentrations of primary pollutants such as nitrogen dioxide (NO 2 ), CO and PM 2.5. Generally, studies relate smog to several respiratory and cardiovascular illnesses. Also, numerous studies have concluded that pollutants are found in greater concentration near major roadways and intersections than local roads. The following outlines these common pollutants in more detail along with some of the health effects associated with each pollutant: Carbon Monoxide (CO): CO results from the vehicle s incomplete combustion of fuels especially at low temperatures. Gasoline engines emit higher amounts of CO than diesel engines, due to their lower combustion temperature compared to diesel. Carbon monoxide has the impact of decreasing the amount of oxygen in the blood. At extremely high levels, CO can cause death but these are not found outdoors. The highest levels of CO usually occur during the colder months as mentioned earlier especially during night time due to temperature inversion where vehicle emissions are high and inversion conditions are more frequent. Motor vehicle exhaust contributes about 60 percent of all CO emissions nationwide (National Air Quality, 2002). 12

26 Nitrogen oxides (NO x ): NOx is a generic term used to describe the grouping of NO, NO 2 and other oxides of nitrogen. NOx is a group of gases that play a major role in the formation of ozone. Most NOx is colorless and odorless except for NO 2 feature is brown in color. They are mainly created during fuel combustion especially at high temperatures where engines burn a small amount of the nitrogen in the air along with nitrogen compounds from the vehicle fuels. Diesel engines generally produce greater amounts of NO x than gasoline engines due to their higher combustion temperatures. NOx can irritate airways leading to lung illnesses. NOx also are precursors of smog components such as Ozone (O 3 ). Particulate Matter (PM): PM can be a primary or secondary pollutant. "Primary" particles, such as dust or black carbon come from several sources such as passenger cars, trucks, buses, factories and construction sites. "Secondary" particles are formed from chemical reactions with other emissions. They are formed when gases from fuel combustion such as motor vehicles or power plants react with sunlight and water vapor, indirectly. PM 2.5 describes the "fine" particles that are less than or equal to 2.5 µm in diameter. PM 10 refers to all particles less than or equal to 10 µm in diameter (about one-seventh the diameter of a human hair). Diesel engines emit significantly more PM than gasoline engines. Fine particulate matter can be inhaled deeply in the lungs which can aggravate symptoms in individuals suffering from respiratory or cardiovascular diseases. Diesel PM is recognized by many agencies such as the World Health 13

27 Organization (WHO), United States Environmental Protection Agency (USEPA), and California Air Resources Board (CARB) to be very toxic compared to gasoline PM and a potential human carcinogen. Volatile Organic Compounds (VOCs): VOCs represent hundreds of different compounds. They come from incomplete fuel combustion. Other VOC emissions come from evaporation of fuel especially during refueling. Gasoline engines emit higher amounts of VOCs than diesel engines due to the greater volatility of fuel. Different VOCs vary broadly in toxicity, and many are precursors of ozone. Sulphur Dioxide (SO 2 ): SO 2 is emitted from the sulphur combustion found in the fuel. Most of SO 2 come from diesel engines since they contain more Sulphur than gasoline engines. Lead, Air Toxics, Coolants and Other emissions: Vehicles also emit toxic air pollutants such as benzene, butadiene, soot, acrolein, and formaldehyde. Some components are VOCs, while others are in the form of particles. Freon or R12 used in older air conditioning systems are known as ozone depleting substances which are emitted through leaks or during repairs. Newer vehicles refrigerant (R134a) are still considered as GHG pollutants although they are non-ozone depleting coolants. The storage and distribution of vehicle fuels also cause air pollution emissions such as the emission of hydrocarbon (HC) vapors during refueling of vehicles. 14

28 EPA (Air Emission Sources, 2011) estimates nationwide emissions of ambient air pollutants and their precursors. These estimates are based on actual monitored readings or engineering calculations of the amounts and types of pollutants emitted by vehicles, factories, and other sources. Emission estimates are based on many factors, including levels of industrial activity, technological developments, fuel consumption, and vehicle miles traveled. Table 2-1 below shows that emissions of the common air pollutants and their precursors have been reduced substantially since Table 2-1: National Emissions Estimates Millions of Tons Per Year Carbon Monoxide (CO) Lead Nitrogen Oxides (NO x ) Volatile Organic Compounds (VOC) Particulate Matter (PM) PM 10 PM 2.5 Sulfur Dioxide (SO 2 ) 6 NA 4 NA Totals

29 2.3 Transportation Greenhouse Gas Emissions & CO 2 Share GHGs are produced from multiple sectors of the economy, including industrial sources, electric power plants, residences, and agriculture, as well as different transportation modes. Unlike air pollutants, GHGs are global in nature. They do not create toxic hot spots, but rather are well-mixed in the atmosphere. Thus, the impacts of one ton of carbon dioxide emissions are the same no matter where it is emitted, or by what sector of the economy. In that sense, the relative effect of transportation emissions on the global climate can be approximated by their relative magnitude compared to all other global emissions. The primary GHGs produced by the transportation sector are carbon dioxide, methane, nitrous oxide, and hydrofluorocarbons (HFC) (CCSP, 2008). Carbon dioxide, a product of fossil fuel combustion, accounts for 95 percent of transportation GHG emissions in the United States, as illustrated in Figure 2-1. Hydrofluorocarbons, which are used in automobile, truck, and rail air conditioning and refrigeration systems, account for another 3 percent of U.S. transportation emissions. Nitrous oxide and methane, which are both emitted as byproducts of combustion, account for the remainder of the U.S. transportation GHG emissions inventory (EPA, 2009). 16

30 Figure 2-1: U.S. Transportation Greenhouse Gas Emissions by Gas, CO2e (2006) Source: U.S. EPA (2008). Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990 to

31 Transportation emissions account for 29 percent of U.S. GHG emissions, and over 5 percent of global GHG emissions (EPA, 2008). Most of the domestically produced emissions are included in the industry sector shown in Figure 2-2. Figure 2-2: U.S. Greenhouse Gas Emissions by End Use Economic Sector,2006 (million metric tons CO 2 equivalent) Source: U.S. EPA (2008). Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990 to

32 As shown in Figure 2-3 and Table 2-2, direct emissions from light-duty vehicles, which include passenger cars and light duty trucks accounted for about 59 percent of U.S. transportation GHG emissions in Emissions from freight accounted for about 19 percent of emissions. Commercial aircraft accounted for about 12 percent. All other modes accounted for about 10 percent of total emissions. Overall, on-road vehicles accounted for 79 percent of emissions (EPA, 2008). Figure 2-3: U.S. Greenhouse Gas Emissions from Transportation by Mode, 2006 Source: U.S. EPA (2008). Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990 to

33 GHG emissions from the U.S. transportation sector increased about 34 percent from 1990 to The growth in U.S. transportation GHG emissions accounted for almost one-half (47 percent) of the increase in total U.S. GHG emissions for the period. Emission trends vary by transportation mode. Medium and heavy-duty truck GHG emissions increased 77 percent from 1990 to 2006, while light duty vehicles increased 24 percent. On-road vehicles accounted for 96 percent of the increase in transportation emissions during that period; 55 percent from light-duty vehicles, 40 percent from medium and heavy-duty trucks, and one percent from other modes (EPA, 2006). Table 2-2: US Transportation Sector Green House Gas Emissions,

34 From 1990 to 2006, an increase in vehicle-miles traveled (VMT) and a stagnation of fuel economy across the U.S. vehicle fleet, caused light-duty vehicle GHG emissions to grow by 24 percent. VMT increased 39.4 percent between 1990 and 2006, as shown in Figure 2-4. Trends in transportation GHGs can generally be seen as a race between fuel economy and VMT. If VMT growth outpaces improvements in fuel economy, emissions will grow. If fuel economy improvements outpace VMT growth, emissions will decline. Recent trends indicate that light duty vehicle emissions are leveling off as VMT growth slows and fuel economy improves. Growth in passenger vehicle VMT slowed from an annual rate of 2.6 percent from 1990 to 2004 to an average annual rate of 0.6 percent from 2004 to 2007 (EPA 2009). In 2008, VMT on all streets and roads in the United States decreased for the first time since 1980, likely due to a combination of high fuel prices and a weakened economy. In addition, average new vehicle fuel economy improved from 2005 to 2007 as the market share of passenger cars increased compared to light-duty trucks; also a response to higher fuel prices and a weakening economy (EPA 2009). 21

35 Figure 2-4: Vehicle Miles Traveled by Light Duty Vehicles Source: Bureau of Transportation Statistics. National Transportation Statistics 22

36 GHG emissions from freight trucks have increased at a greater rate than all other freight sources, as shown in Figure 2-5. Figure 2-5: GHG Emissions from US Freight Sources Source: U.S. EPA (2008). Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990 to

37 2.4 Calculating CO 2 Emissions As described in the (EPA, 2006) report for calculating average CO 2 Emissions resulting from gasoline and diesel fuel, One of the primary determinants of carbon dioxide (CO 2 ) emissions from mobile sources is the amount of carbon in the fuel. Carbon content varies, but typically we use average carbon content values to estimate CO 2 emissions. The Code of Federal Regulations (40 CFR ) provides values for carbon content per gallon of gasoline and diesel fuel which EPA uses in calculating the fuel economy of vehicles: Gasoline carbon content per gallon: 2,421 grams Diesel carbon content per gallon: 2,778 grams IPCC guidelines for calculating emissions inventories require that an oxidation factor be applied to the carbon content to account for a small portion of the fuel that is not oxidized into CO 2. For all oil and oil products, the oxidation factor used is 0.99 (99 percent of the carbon in the fuel is eventually oxidized, while 1 percent remains unoxidized (EPA, 2006). 24

38 Finally, to calculate the CO 2 emissions from a gallon of fuel, the carbon emissions are multiplied by the ratio of the molecular weight of CO 2 (m.w. 44) to the molecular weight of carbon (m.w.12): 44/12. - CO 2 emissions from a gallon of gasoline = 2,421 grams x 0.99 x (44/12) = 8,788 grams = 8.8 kg/gallon = 19.4 pounds/gallon - CO 2 emissions from a gallon of diesel = 2,778 grams x 0.99 x (44/12) =10,084 grams = 10.1 kg/gallon = 22.2 pounds/gallon 2.5 Feasibility of Field Capturing CO 2 Capturing CO 2 from air is possible either naturally through plants or chemically through many ways such as bubbling air through a calcium hydroxide (CAOH) solution to remove CO 2. However, economic considerations must be given to the dilution ratio in air, which is roughly one part in three thousand. It would be possible to move the air mechanically but only at speeds that are easily achieved by natural flows such as passing over some recyclable sorbent (Klaus et al., 1999). Once this process of extracting CO 2 out of the air is done, the downstream process deals with volumes and masses and is therefore not subject to the amplification factor resulting from the dilution process. Scrubbing out CO 2 is not the only cost considered in extracting CO 2 from air. Moreover, the sorbent used has to be recovered to release CO 2 in a concentrated disposal stream and 25

39 then the CO 2 has to be disposed. These process steps are far more expensive than the capture process. Another difficult process to manage is capturing CO 2 from the transportation sector (Klaus et al., 1999). Although transitioning towards the electric or hydrogen fueled vehicles is in process, it would take a long time to accomplish. Even though it has been proposed (Seifritz et al., 1993), it is not economically viable to collect the carbon dioxide of a vehicle directly at the source since the mass flows would be prohibitively large. A unit mass of fuel results in roughly three mass units of gaseous CO 2 that would need to be stored and then shipped to another disposal site. Therefore, capturing CO 2 is simply not practical because of the mass, storage and shipping costs involved. 2.6 US CAFE Program Overview of Joint EPA/NHTSA National Program In December 2007, Congress enacted the Energy Independence and Securities Act (EISA), amending the Energy Policy Conservation Act (EPCA) to require substantial, continuing increases in fuel economy standards. The Corporate Average Fuel economy (CAFE) standards address most, but not all, of the real world CO 2 emissions since vehicle air conditioner are turned off during fuel economy testing. Fuel economy is determined by measuring the amount of CO 2 emitted from the tailpipe. The carbon content of the test 26

40 fuel is then used to calculate the amount of fuel that had to be consumed per mile in order to produce that amount of CO 2. Finally, that fuel consumption figure is converted into a miles-per-gallon figure. CAFE standards also do not address the 5 8 percent of GHG emissions that come from nitrous oxide (N 2 O), and methane (CH 4 ) as well as emissions of CO 2 and hydrofluorocarbons (HFCs) related to operation of the air conditioning system (EPA, 2009). In 2004, the California Air Resources Board (CARB) approved standards for new light duty vehicles, which regulate the emission of not only CO 2, but also other GHGs. Thirteen states and the District of Columbia, comprising about 40 percent of the light duty vehicle market, have adopted California s standards. These standards apply to model years (MY) 2009 through 2016 and require CO 2 emissions for passenger cars and the smallest light trucks of 323 g/mi in 2009 and 205 g/mi in 2016 and for the remaining light trucks of 439 g/ mi in 2009 and 332 g/mi in To promote the National Program, in May 2009, California announced its commitment to take several actions in support of the National Program, including revising its program for MYs and MYs This will allow the single national fleet produced by automakers to meet the two Federal requirements and to meet California requirements as well. Several automakers and their trade associations also announced their commitment to take several actions in support of the National Program, including not contesting the final GHG and CAFE standards for MYs (EPA, 2009). 27

41 2.6.2 Summary of the Joint Final Rule In this joint rulemaking, EPA is establishing GHG emissions standards under the Clean Air Act (CAA), and NHTSA is establishing CAFÉ standards under the EPCA of 1975, as amended by the EISA of This joint final rule covers passenger cars, lightduty trucks, and medium duty passenger vehicles built in MYs 2012 through 2016 (NHTSA, 2009). These vehicle categories are responsible for almost 60 percent of all U.S. transportation-related GHG emissions. The National Program is estimated to result in approximately 960 million metric tons of total carbon dioxide equivalent emissions reductions and approximately 1.8 billion barrels of oil savings over the lifetime of vehicles sold in MYs 2012 through In total, the combined EPA and NHTSA standards will reduce GHG emissions from the U.S. light-duty fleet by approximately 21 percent by 2030 over the level that would occur in the absence of the National Program (NHTSA, 2009). 2.7 Previous Studies to Model and Estimate Traffic Emissions Many studies have reported on the variation of emission factors (EFs) with average vehicle speed. The largest EFs for CO and other pollutants tend to occur at speeds of less than 20 mph because of inefficient engine operation and travel at low speeds. CO 2 emissions are linked directly to fuel consumption, and so CO 2 emissions per mile go up at very low or very high speeds. Knowledge of traffic-flow patterns is also 28

42 relevant because local pollutant concentrations (more important for CO and PM not so much for CO 2 ) are directly proportional to vehicle numbers and their characteristics (Bogo et al., 2001). Marsden et al. (2001) studied CO emissions in microscopic traffic modeling based on vehicle speed and classification using vehicle acceleration, deceleration, cruising and idle inputs, enriched acceleration, state of repair of the vehicle emission control system and type of engine. They showed that vehicle-exhaust emissions depend strongly on the fuel-to-air ratio. Sturm et al. (1998) described three approaches for compiling emission inventories based on actual driving behavior, specific streets and vehicle-miles traveled (VMT). The parameters considered were travel demand, traffic condition, vehicle-operating mode (cruising, idling, accelerating or decelerating), and vehicle-operating condition (cold or hot start, average speed, load, trip length, frequency of trips). Furthermore, vehicle parameters (model and year, state of maintenance, engine type and size, emission reduction devices, accrued mileage, fuel-delivery system), and the fuel characteristics (type, volatility, chemical composition) were also included. Besides, driver behavior, local climate conditions, and topography were considered. A study by Hallmark et al.(2002) found that driving patterns (e.g., speeds) at different intersections are significantly influenced by queue position, downstream and upstream lane volume, incidents, percent of heavy vehicles, and posted link speed. 29

43 Emissions also vary with respect to drivers attitude, experience, gender, physical condition, and age. Aggressive driving increases emissions compared to normal driving (De Vlieger et al., 2000). Sierra Research found that most drivers spend about 2% of total driving time in aggressive mode, which contributes about 40% of total emissions (Samuel et al., 2002). Nesamani et al. (2007) proposed an intermediate model component that can provide better estimates of link speeds by considering a set of Emission Specific Characteristics (ESC) for each link. The intermediate model was developed using multiple linear regression and evaluated using a microscopic traffic simulation model. The evaluation results showed that the proposed emission estimation method performed better than current practice and was capable of estimating time-dependent emissions if traffic sensor data are available as model input. Chu and Meyer (2009) described an analysis that utilized EPA's MOBILE6.2 vehicle emissions modeling software to identify freeway locations with large pollutant emissions and estimated the changes in emission associated with Truck-only toll (TOT) lanes. Emissions including hydrocarbon (HC), carbon monoxide (CO), nitrogen oxide (NO x ), and CO 2 were estimated by emission factors associated with various vehicle types and average speeds. The CO 2 calculation was limited due to lack of sensitivity in the model to speed variation, which was one of the benefits of the implementation of TOT 30

44 lanes. The change in vehicle speeds was applied to estimate the change in fuel consumption and CO 2 emissions. The results showed that voluntary and mandatory use of TOT lanes would reduce total CO 2 emissions on all freeway lanes by 62%. In an effort by Int Panis et al. (2011) to determine PM, NOx and CO 2 emission reductions from speed management policies in Europe, they examined the impact on urban versus highway traffic with different modeling approaches--microscopic (VeTESStool) versus macroscopic (COPERT). Results indicated that emissions of most classic pollutants do not rise or fall dramatically. The effects of specific speed reduction schemes on PM emissions from trucks were ambiguous but lower maximum speed (e.g., mph) for trucks consistently result in lower fuel consumption and in lower emissions of CO 2. In an earlier attempt by Int Panis et al. (2006) to model instantaneous traffic emissions and the influence of traffic speed limits, they concluded that the speed management impact on vehicle emissions is complex. The frequent acceleration and deceleration movements in the network may significantly reduce the benefits of changing the overall average speed. The conclusion from that study was that active speed management had no significant impact on total pollutant emissions. Boriboonsomsin and Barth (2009) evaluated the effect of road grade on vehicle fuel consumption (and thus carbon dioxide emissions). The real-world experimental 31

45 results showed that road grade does have significant effects on the fuel economy of lightduty vehicles both at the roadway link level and at the route level. Bachman et al. (2000) investigated a GIS-based modeling approach called the Mobile Emission Assessment System for Urban and Regional Evaluation (MEASURE). MEASURE provides researchers and planners with a means of assessing motor vehicle emission reduction strategies. Estimates of spatially resolved fleet composition and activity are combined with activity-specific emission rates to predict engine start and running exhaust emissions. Engine start emissions are estimated using aggregate zonal information. Running exhaust emissions are predicted using road segment specific information and aggregate zonal information. Liping and Yaping (2005) calculated emission factors for three emission modes; hot emissions, emissions from vehicles after they have warmed up to their normal operating temperature, cold-start emissions, the emissions from vehicles while they are warming up and the water temperature is below 70 C, and the evaporative emissions. Husch (1998) applied SYNCHRO, a macroscopic traffic-flow model with a builtin simplified emission model, for estimating vehicle emissions by first predicting the fuel consumption as a function of vehicle-miles, delay in vehicles hour-per-hour, and stops in stops per hour. Fuel consumption was then multiplied by an adjustment factor (differs depending on the type of emissions) to estimate the vehicle emissions. 32

46 These studies facilitated the advancement of emission models that account for start emissions, vehicle activity and roadway types. For example, the latest version of MOBILE6 included emission rates and off-cycle emissions that reflect real-world traffic conditions more accurately and can account separately for start emissions and running emissions. It is capable of estimating emissions by roadway type, time of day, and other characteristics (US EPA, 2002). Also EMFAC included low emission vehicle standards and EPA Tier II standards (CARB, 2000) and assumes modest emission reductions for appropriate inspection and maintenance programs. It produces separate emission factors for cold starts, hot starts, and hot stabilized conditions. Modal emission models based on various vehicle-operating modes have also been emerged as alternatives (Barth et al., 1996a) and (Guensler et al., 1998). The accuracy of these models, however, depends on estimates of traffic-network activity obtained from travel forecasting models, which are still based on steady state analyses. Barth et al. (1996b) developed a methodology to utilize both traffic sensor and microscopic data to estimate emissions, but it does not consider road geometry and cannot be used for links without loop detectors. Models based on it incorporate standard conditions established in laboratory dynamometer driving tests and predicted CO, HC, and nitric Oxide (NO) emissions (US EPA, 1997). A number of microscopic traffic models estimate vehicle emissions as a function of vehicle type, speed, and acceleration on a second-to-second basis. The US Federal Highway Administration (1997) developed CORSIM, a microscopic model, and used 33

47 vehicle emission rates from the dynamometer testing as the basis of its emissions model. It determined the emissions on each link by applying speed and acceleration related emission rates to each vehicle for each second the vehicle traveled on the given link. Van Aerde (1995) also computed the fuel consumption for each vehicle on a second-bysecond basis as a function of speed and acceleration using the INTEGRATION traffic emission model. Further, he included an estimation of vehicle emissions on a second-bysecond basis as a function of fuel consumption, ambient air temperature, and the extent to which a particular vehicle s catalytic-converter already warmed up during an earlier portion of a trip. Yu (1998) developed the ONROAD model for estimating vehicular CO and HC emissions based on on-road data and establishes relationships between the on-road vehicle-exhaust emission rates and vehicle instantaneous speed profile, which is a function of different traffic demand and control scenarios. The model estimates the implications of alternative traffic control and management strategies on emissions. A traffic simulation model easily incorporates ONROAD in situations where a vehicle s instantaneous speed profile can be tracked consistently. Further, it indicates that MOBILE and EMFAC underestimate on-road vehicle emissions for all vehicle types. Yu s work also includes comparisons of instantaneous emission rates among emission models, showing that emissions from the TRANSYT model deviate from those from 34

48 ONROAD, while MOBILE and EMFAC exhibit consistency in their emission rate estimations. Studies as old as Alexopoulos et al. (1993) developed a model for spatial and temporal evaluation of traffic emissions in metropolitan areas. Gertler and Pierson (1994) showed that improving the inputs to mobile source emission models, rather than developing new models, can reduce the differences between their predictions and observed concentrations of CO and HC. Electronic fuel injection systems optimize the fuel flow to ensure balanced combustion (Heywood, 1988). In this case, CO 2 and NOx are the main products of the combustion. In contrast, diesel engines operate with a lower fuel to air ratio than petrol engines using lean burning fuel and air mixtures (Al-Omishy and Al-Samarrai, 1988). Al-Deek et al. (1997) liaised with the UCF air quality research team and conducted ozone modeling using UAM. They developed FLINT (the FLorida INTersection) air quality model, a Gaussian-based model based on macroscopic theory for calculating idling emissions and predicting CO concentrations near intersections. Density, flow, vehicle composition, v/c ratio, the number of traffic lights per mile, signal coordination, and the number of stops per mile are traffic-related variables. Congested traffic conditions increase emissions and reduces speed compared to free flow 35

49 conditions (Andre and Hammarstrom, 2000 and Vlieger et al., 2000). Rakha et al. (2000) concluded that proper signal coordination could reduce emissions up to 50%. Roadway environmental characteristics along the road can have a significant influence on link speed. A study by Galin (1981) found that the land use adjacent to roads strongly influences speed. The type of land use (e.g., residential or commercial) is especially influential. Shu et al. (2010) developed a multiple linear regression model to disaggregate traffic-related CO 2 emission estimates from the parish-level scale to a 1 1 km grid scale. Considering the allocation factors (population density, urban area, income, road density) together, they used a correlation and regression analysis to determine the relationship between these factors and traffic-related CO 2 emissions, and developed the best-fit model. The result reveals that high CO 2 emissions are concentrated in dense road network of urban areas with high population density, and low CO 2 emissions are distributed in rural areas with low population density and sparse road networks. The proposed method of Shu can be used to identify the emission hot spots at fine scale and is considered more accurate and less time-consuming than the previous methods. A microscopic simulation platform for estimating vehicle emissions that can capture the vehicles' instantaneous modal activities was developed by Chen et al. (2007). They integrated the microscopic traffic-emission simulation platform by using the 36

50 microscopic traffic simulation model VISSIM and the modal emission model CMEM on a sub-network selected from the Haidian district of Beijing to evaluate the network s traffic and emission conditions. 2.8 EPA Emissions Models and Analysis Tools Softwares There are a variety of tools available to transportation practitioners for analyzing, measuring, and projecting vehicular emissions. These also include tools and methods to estimate GHG emissions and develop inventories for quantifying the effects of transportation projects, technologies, and strategies. Results from such programs are used to guide policy and planning decisions at reducing emissions. The following information is obtained from the US EPA website ( MOBILE6 was used to produce motor vehicle emission factors for use in transportation analysis and can be used at any geographic level within the U.S. NONROAD Model links to information on the NONROAD emission inventory model, which is a software tool for predicting emissions of hydrocarbons, carbon monoxide, oxides of nitrogen, particulate matter, and sulfur dioxides from small and large nonroad vehicles, equipment, and engines. This model produces estimates of criteria pollutant emissions and CO 2 from all non-road sources, with the exception of commercial marine vessels, locomotives, and aircraft. The model calculates past, present, 37

51 and future emission inventories for 80 basic and 260 specific non-transportation equipment categories. MOVES (MOtor Vehicle Emission Simulator) is EPA s current official model for estimating air pollution emissions from cars, trucks and motorcycles. This modeling system estimates emissions for on-road and non-road sources for a broad range of pollutants and allow multiple scale analysis. This system replaced MOBILE6 and will eventually replace NONROAD. NMIM or National Mobile Inventory Model, is a free, desktop computer application developed by EPA to help develop estimates of current and future emission inventories for on-road motor vehicles and nonroad equipment. NMIM uses current versions of MOBILE6 and NONROAD to calculate emission inventories, based on multiple input scenarios that can be entered into the system to calculate national, individual state, or county inventories. Fuels Models links to information on EPA's heavy-duty diesel fuel analysis program which seeks to quantify the air pollution emission effects of diesel fuel parameters on various nonroad and highway heavy-duty diesel engines. It also links to the Complex Model and the Simple Model used for the Reformulated Gasoline Program. 38

52 OMEGA the Optimization Model for Reducing Emissions of Greenhouse Gases from Automobiles, which estimates the technology cost for automobile manufacturers to achieve variable fleet-wide levels of vehicle greenhouse gas emissions. Climate Leadership in Parks (CLIP) currently is not publicly available. This tool calculates emissions based on fuel consumption and/or vehicle miles traveled and thus allows for greenhouse gas (GHG) and criteria pollutant emissions estimation at a more local level. COMMUTER Model This model analyzes the impacts of transportation control measures (TCMs) on vehicle miles traveled (VMT), criteria pollutant emissions, and CO 2. State Inventory Tool (SIT) This tool will help develop a comprehensive GHG inventory at the state level by allowing users to enter their own state-specific activity data to estimate emissions. State Inventory Project Tool This tool is based on the State Inventory Tool (SIT) and forecasts emissions through 2020 to allow users to compare trends back to Examples of Non-EPA Emissions Models Comprehensive Modal Emission Model (CMEM), which is one of the newest power demand-based emission models, was developed at the University of California, 39

53 Riverside (Barth et al., 2000). The model estimates LDV and LDT emissions as a function of the vehicle s operating mode. The term comprehensive is utilized to reflect the ability of the model to predict emissions for a wide variety of LDVs and LDTs in various operating states (e.g., properly functioning, deteriorated, malfunctioning). Vehicles were categorized in the CMEM model based on a vehicle s total emission contribution. Twenty-eight vehicle categories were constructed based on a number of vehicle variables. These vehicle variables included the vehicle s fuel and emission control technology (e.g., catalyst and fuel injection), accumulated mileage, power-toweight ratio, emission certification level (tier0 and tier1), and emitter level category (high and normal emitter). In total 24 normal vehicles and 4 high emitter categories were considered (Barth et al., 2000). The Virginia Tech microscopic energy and emission model (VT-Micro model) was developed from experimentation with numerous polynomial combinations of speed and acceleration levels. Specifically, linear, quadratic, cubic, and quartic terms of speed and acceleration were tested using chassis dynamometer data collected at the Oak Ridge National Laboratory (ORNL). The final regression model included a combination of linear, quadratic, and cubic speed and acceleration terms because it provided the least number of terms with a relatively good fit to the original data (R 2 in excess of 0.92 for all measures of effectiveness (MOE)). 40

54 EMFAC Model, California Air Resources Board (CARB) This model produces emission rates and inventories for criteria air pollutants and CO 2. It is the approved emissions model used in the State of California for SIP development, conformity analysis, and other analyses that are typically conducted using MOBILE6 in other states. The Greenhouse Gases, Regulated Emissions, and Energy Use in Transportation (GREET) model Argonne National Laboratory, This full life-cycle model was designed to evaluate energy and emission impacts of advanced vehicle technologies and new transportation fuel combinations on a full fuel-cycle/vehicle-cycle basis. Intelligent Transportation Systems Deployment Analysis System (IDAS) Federal Highway Administration, This sketch planning analysis tool is used to estimate the impacts, benefits, and costs resulting from the deployment of Intelligent Transportation System (ITS) components of over 60 types of ITS investments. Lifecycle Emissions Model (LEM) (not publicly available), University of California, Davis, This model estimates energy use, criteria pollutant emissions, and CO 2 -equivalent GHG emissions from transportation and energy sources. Long Range Energy Alternatives Planning (LEAP) System Commend: Community for Energy, Environment and Development. LEAP is a software tool for energy policy analysis and climate change mitigation assessment that uses integrated modeling to track energy consumption, production, and resource extraction in all sectors of an economy. 41

55 The MARKAL-MACRO Model, Department of Energy. This model is an integration of two models, MARKAL and MACRO to link the use of energy and environmental resources to the economy. It can forecast emissions sources and levels for CO 2, SOx, and NOx, and any user-specified pollutants and wastes. MiniCAM Model, Pacific Northwest National Laboratory (PNNL). This model forecasts CO 2 and other GHG emissions and estimates the impacts on GHG atmospheric concentrations, climate, and the environment. As of 2005, this model was superseded by ObjECTS-MiniCAM, a C++ version of the model that incorporates object-oriented programming designs for increased flexibility, maintenance, and modeling detail. National Energy Modeling System (NEMS), Energy Information Administration (EIA), USDOE. This modeling system represents the behavior of energy markets and their interactions with the U.S. economy. It contains a transportation demand module (TRAN) that has several sub-modules and that uses NEMS inputs. System for the Analysis of Global Energy Markets (SAGE), (Not publicly available), USDOE, SAGE is an integrated set of regional models that provides a technology-rich basis for estimating regional energy supply and demand. For each region, reference case estimates of end-use energy service demands (e.g., car, commercial truck, and heavy truck road travel; residential lighting; steam heat requirements in the paper industry) are developed on the basis of economic and demographic projections. 42

56 Transitional Alternative Fuels and Vehicle Model (TAFV). University of Maine. The TAFV model represents economic decisions among auto manufacturers, vehicle purchasers, and fuel suppliers and can predict the choice of alternative fuel technologies for light-duty motor vehicles. VISION Model, Argonne National Laboratory. This model forecasts energy use until 2050 and provides estimates for advanced light- and heavy-duty highway vehicle technologies and alternatives, including potential energy use, oil use, and carbon emissions impacts. The model was designed as a simplified and fast way to assess the potential impact of new fuel technologies on energy use and carbon emissions. World Energy Protection System (WEPS) Transportation Energy Model (TEM), USDOE. This structural accounting model for transportation energy use generates midterm forecasts of the transportation sector's energy use in order to evaluate the effect of changes in fuel economy on carbon emissions. Based on the above literature review, it is concluded that limited research has been done on quantifying the impacts of transportation emissions through experimental design methodologies in a stochastic-microscopic environment. The models listed above show that although there are a variety of traffic emissions models to calculate general vehicular emissions, very few focused on developing and calculating carbon dioxide and other pollutants emissions on limited access highways and particularly considering most 43

57 of the transportation factors (volume, length, speed, temperature and grade) in one model. Also, even though micro-simulation is gaining strength in many transportation aspects, it is still a new subject in emissions applications. The simulations that have been applied so far focused mostly on one of the transportation factors, which is speed and were limited to model existing conditions. Furthermore, research directed at investigating decision processes underlying emissions mitigation strategies is still in its infancy. This research attempts to develop experimental design methodologies to quantify and mitigate transportation emissions in response to current environmental challenges. In addition, this approach is extended to integrate a powerful microscopic emissions package MOVES2010a along with a powerful microscopic traffic simulation package VISSIM. VISSIM is stochastic in nature and models explicitly second by second accelerations/decelerations, lane changing and merging/diverging, which are typical in stop-and-go traffic. VISSIM output is integrated with MOVES to quantify emission inventories and rates. 44

58 3. RESEARCH APPROACH In order to achieve the stated objectives, the following research methodology was implemented: 1. Design of Experiments (DOE) 2. Development of Calibrated Base Scenario using VISSIM Model 3. Estimation of Scenario-Based Emissions using MOVES Model 4. Statistical Analysis Using JMP Software 5. Development of Emission Prediction Model (Micro-TEM) 6. Application of Mitigation Strategies 7. Findings of Research Results and Conclusions 3.1 Design of Experiments (DOE) In many scientific investigations, the concern is to optimize the system. Experimentation is one of the popular activities used to understand and/or improve a system. This can be achieved by studying the effects of two or more factors on the response at two or more values known as levels or settings simultaneously. This type of standard experiment is known as factorial design. Cost and practical constraints must be considered in choosing factors and levels. Therefore, two-level factorial designs are common for factor screening in industrial applications (Jones and Montgomery, 2010). However, if a non-standard model is required to adequately explain the response or the model contains a mix of factors with different levels, types or results in an enormous number of runs, the requirements of a standard experimental design will not fit the research requirements (Johnson et. al, 2011). As stated in Johnson s Expository Paper on 45

59 Optimal Designs: Designing experiments for these types of problems requires a different approach. We can t look in the textbook or course notes and try to match the designs we find there to the problem, (Johnson et. al, 2011). Under such conditions, optimal custom designs are the recommended design approach. Choosing an optimality criterion to select the design points to be run is another requirement. Since the factors of interest in the experiment consist of five (5) quantitative factors, each with six (6) levels and one quantitative response (CO 2 ), the standard number of full factorial design needed to cover all cases would amount to 7,776 runs (6 5 ). Even choosing to run a fractional factorial of l (k-p) where l is the number of levels of each factor investigated, k is the number of factors investigated, and p describes the size of the fraction of the full factorial used, the methodology needed to generate designs for more than two levels is too cumbersome. Accordingly, the D-optimality and I-optimality criteria were the two custom designs employed for this experiment which are explained in greater detail in Chapter 6. The factors and levels ranges included: 1) Volume (2,000 vph - 7,000 vph) 2) Speed (20 mph - 70 mph) 3) Trucks (0% - 15%) 4) Grade (0% - 5%) 5) Temperature (50 F F) The optimal custom design of this experiment resulted in 140 runs (70 runs for each design) obtained by considering all main effects, quadratic and cubic effects along with two-way and three-way factor interactions of the studied factors with six possible 46

60 levels in addition to center points. Table 3-1 provides a partial basic layout of the planning matrix in a standard order which describes the experimental plan in terms of the actual values or settings of the factors. Each row of the table represents one set of experimental conditions that when run produces a value of the response variable y. The response variable was the amount of Carbon Dioxide (CO 2 ) emissions in (kg) produced in each scenario. The five factors were designated A though E with the levels (-1) as the low setting and (+1) as the high setting. The experiment was conducted in a microscopic and stochastic environment. Table 3-1: Partial Layout of a Generic Experimental Design Run# A B C D E Y Volume Speed Truck % Grade% Temperature CO Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y15 47

61 3.2 Development of Calibrated Base Scenario Using VISSIM Model Because the collection of a large representative second-by-second vehicle operation dataset for every traffic circumstance is not realistic, the use of microscopic traffic simulation models to replicate the real world second-by-second driver behavior for hundreds of vehicles and traveling patterns is essential The VISSIM Model VISSIM 5.3 is a microscopic, stochastic and behavior based simulation model developed by PTV to model urban traffic and public transit operations. The program can analyze traffic and transit operations under constraints such as lane configuration, traffic composition, traffic signals, transit stops, etc., thus making it a useful tool for the evaluation of various alternatives based on ITS-based transportation engineering and planning measures of effectiveness. VISSIM can be applied as a powerful tool in a variety of transportation problem settings ( Calibration & Validation of VISSIM Model As mentioned previously, the study corridor was the Interstate 4 (I-4) located in downtown Orlando, Florida as shown in Figure 3-1. Peak hour information was gathered for the eastbound direction of the evening peak period including traffic counts, travel 48

62 time and delays for the distance traveled, truck percentages and temperature of the day. The existing network was coded in the micro-simulation model VISSIM, calibrated and validated to replicate existing conditions. The calibrated base scenario was used for comparison purposes against the mitigation scenarios as discussed in Chapter 8. The following methodology was utilized to calibrate and validate the VISSIM simulation model. The Florida Department of Transportation (FDOT) monitors the I-4 corridor and collects traffic counts at several field stations on a yearly basis. The FDOT traffic information database formed the basis of the calibration and validation step. Furthermore, travel time and delay information posted on the Dynamic Message Signs (DMS) on the I-4 downtown Corridor were compared against the simulated data. Figure 3-1: I-4 Downtown Corridor (Orlando, FL) 49

63 The network geometry including horizontal curves, grades and ramp locations is a critical step in the calibration process. Initial evaluation was conducted to compare the distribution of simulation outputs (traffic volumes, travel times, delays, and queues) with the field data as well as initial identification of the calibration parameters. The calibration involved accurately modeling drivers driving behavior and ensuring reasonable throughput values on the network. Travel times, delay and queuing were also used for validation. Multiple runs were conducted with each combination of parameters and the distribution of traffic volumes and queues were compared to the field data. If the field data lies within the simulated distribution of the simulated data, thus the parameters and their ranges were considered as appropriate. Evaluation of parameter sets was completed by comparing the simulations results of default parameters against adjusted parameters as well as corridor visualization as explained in detail in Chapter Estimation of Scenario-Based Emissions using MOVES Model Estimating vehicle emissions based on second-by-second vehicle operation encourages the integration of microscopic traffic simulation models with more accurate vehicle activity-based regional mobile emissions models. Therefore, the simulated vehicle driving cycle data was integrated into the latest EPA s mobile emissions model MOVES2010 to calculate CO 2 emissions based on project-level constraints. A prototype of the calibrated I-4 corridor was utilized to test for the 140 scenarios mentioned earlier. 50

64 Network output from the microscopic traffic simulation model VISSIM were input into MOVES to calculate emissions The MOVES Model MOVES2010 model is the state-of-the-art upgrade to EPA s previous modeling tools for estimating emissions from vehicles. MOVES2010 replaces the previous model MOBILE6.2, for estimating on-road mobile source emissions. MOVES modeling process requires the input of vehicle types, time periods, geographical areas, pollutants, vehicle operating modes and road types. The model then accurately reflect vehicle operating condition, such as extended idle or running emissions and provide estimates of total emissions or emission rates. The characteristics of a particular scenario are developed by creating a Run Specification database (RunSpec). The MOVES model is totally different from previous EPA mobile source emissions models in that it is purposely designed to be flexible with databases. And so, if new data becomes available, it would be easily incorporated into the model. Furthermore, MOVES enables the import of specific data needs. The MOVES model includes a default emissions database for the entire United States. The data was collected from several sources such as EPA research studies, Census Bureau vehicle surveys, Federal Highway Administration, and other federal, state, local, industry and academic sources. MOVES is 51

65 continuously updated, especially for analyses concerning State Implementation Plans (SIPs) and conformity purposes. MOVES uses Vehicle Specific Power (VSP) and instantaneous speeds to calculate emission rates on a second-by-second basis. This approach added flexibility to emissions covering all achievable combinations of instantaneous speeds and accelerations which were essential in developing emissions for any driving pattern. VISSIM generates trajectories on a second by second basis, vehicle number and types, vehicle location, speed and acceleration for each vehicle within each link on a second-by-second basis. The same temporal resolution and level of detail modeled in VISSIM supports the integration of MOVES. The trajectory data from VISSIM were converted to match MOVES input. MOVES calculates the emissions for each vehicle over time using the trajectory data based on a specified time step in each calculation Validation of MOVES Model The Clean Air Act (CAA) requires EPA to regularly update its mobile source emission models and therefore EPA continuously collects data and measures vehicle emissions to ensure they have the most accurate information on mobile source emissions. The model is based on results of millions of emission tests and substantial improvements related to the understanding of vehicle emissions. 52

66 MOVES2010a estimates air pollution emissions from cars, trucks, motorcycles, and buses. It is approved by EPA for assessing official state implementation plan (SIP) submissions as well as transportation conformity analyses outside of California. It can be used also to evaluate the benefits of several mobile source control strategies either local or regional and for policy evaluation. MOVES2010a is considered the best available tool for quantifying criteria pollutant and GHG emissions for the transportation sector. This detailed approach to modeling allowed the incorporation of large amounts of validated data from several sources, such as vehicle inspection data and maintenance (I/M) programs, remote sensing device (RSD) testing, and portable emission measurement systems (PEMS). This approach also helped in scenario comparisons and identified differences resulting from changes to vehicle speeds and acceleration patterns. As mentioned earlier, the output from the VISSIM model was used as input into the MOVES model. The first input step was to create the Project Level database where the imported data is stored in a MySQL database. The data was specified in separate text files for each input parameter such as volumes, speed, link length, grade, etc. This process was repeated for each of the following main input files: 1. Meteorology Data Importer 2. Source Type Population Importer 3. Age Distribution Importer 4. Vehicle Type VMT and VMT Fractions 5. Link Source Types Importer 6. Links Importer 53

67 7. Operating Mode Distribution Importer 8. Link Drive Schedules Importer 9. Fuel Supply Importer 10. Fuel Formulation Importer As stated in the MOVES technical guidance report (EPA, 2010): Meteorology Data Importer The Meteorology Data Importer includes temperature and humidity data for months, zones, counties, and hours that are included in the RunSpec. While the MOVES model contains 30-year average temperature and humidity data for each county, month, and hour, specific data for the modeled location and time should be used Source Type Population Importer The Source Type Population Importer includes the number of vehicles in the geographic area which is to be modeled for each vehicle or "source type" selected in the RunSpec. Data must be supplied for each source type (e.g., passenger car, passenger trucks, light commercial trucks, etc.) selected in the RunSpec Age Distribution Importer The Age Distribution Importer includes data that provides the distribution of vehicle counts by age for each calendar year (yearid) and vehicle type (sourcetypeid). The distribution of ageids (the variable for age) must sum to one for each vehicle type and year. 54

68 Vehicle Type VMT and VMT Fractions The Vehicle Type VMT importer includes yearly vehicle miles traveled (VMT) and the monthly, type of day, and hourly VMT fractions. These values will represent county-specific values for the County Data Manager Link Source Types Importer The Link Source Types Importer is used to enter the fraction of the link traffic volume which is driven by each source type. For each linkid, the sourcetypehourfraction must sum to one across all source types Links Importer The Links Importer is used to define individual roadway links of the study network. The MOVES links need not correspond to traffic modeling "links" but each link should be uniform in its activity. Each link requires a linkid that is used to reference the link in the network). Other required inputs for each link are countyid, zoneid, and, the length of the roadway link in units of miles, the traffic volume on the roadway link in units of vehicles per hour, the average speed of all of the vehicles on the roadway link in the given hour, and the average road grade of a particular link. In addition to roadway links, a project may include a single off-network (parking lot or other non-road zone) link. 55

69 Operating Mode Distribution Importer The Operating Mode Distribution Importer is used to import operating mode fraction data for source types, hour / day combinations, roadway links and pollutant / process combinations that are included in the RunSpec and Project domain. These data are entered as a distribution across operating modes. Operating modes are "modes" of vehicle activity that each have a distinct emission rate. For example, "running" activity has modes that are distinguished by their Vehicle Specific Power and instantaneous speed. "Start" activity has modes that are distinguished by the time the vehicle has been parked prior to the start ("soak time") Link Drive Schedules Importer The Link Drive Schedules Importer is used to define the precise speed and grade as a function of time (seconds) on a particular roadway link. The time domain is entered in units of seconds, the speed variable in miles per hour and the grade variable in percent grade (i.e., vertical distance / lateral distance; 100% grade equals a 45 degree slope) Fuel Supply Importer The Fuel Supply importer is used to assign existing fuels to counties, months, and years, and to assign the associated market share for each fuel. The market share for a 56

70 given fuel type (gasoline, diesel, etc.) must sum to one for each county, fuelyear (i.e., calendar year), and month Fuel Formulation Importer The Fuel Formulation importer and the Fuel Supply importer should be used together to input appropriate fuel data. The Fuel Formulation importer is used to select an existing fuel in the MOVES database and change its properties, or create a new fuel formulation with different fuel properties. Table 3-2 also shows a summary of the data parameters used explicitly for this research to conduct the experimental design analysis. Table 3-2: Summary of Project Level Parameters Location County Orange County, Florida Calendar Year 2010 Time One hour Weekday/Weekend Weekday Humidity 70.0 % Road type Urban Restricted Access represents urban freeway road (I-4 downtown links) with 3 lanes in each direction Types of Vehicles Passenger cars, SUVs, Vans & Trucks Type of Fuel Gasoline for cars and Diesel for Trucks Roadway Length 10 miles Link Traffic Volume 2,000-7,000 vehicles per hour Link Speed Limit miles per hour Truck Traffic percent Average Road Grade percent Temperature 50 F F 57

71 3.4 Statistical Analysis Using JMP Software JMP 9.0 (pronounced as "jump"), a module from SAS, is a state-of-the-art statistical analysis software program used to perform complex statistical analyses. It dynamically links statistics with graphics to interactively understand and visualize data. JMP provides a comprehensive set of statistical tools for the design of experiments and can work with a variety of data formats such as Excel, text and SAS files. JMP was used to facilitate the generation and understanding of the results of the experiment. Of immediate concern was the determination of the impact of the factors individually on the response variable which included the main effects, any two-way, three-way interactions as well as any quadratic or cubic effect. 3.5 Development of Emission Prediction Model (Micro-TEM) The purpose of the experiment was to explore all possible settings of the studied factors and develop a Microscopic Transportation Emissions Meta-Model Micro-TEM, a predictive model with the ability to calculate emission rates on freeways or limited access highways based on a stochastic-microscopic manner. Contour plots were generated for the response surface model. With the predictive model, tentative considerations of optimal settings were identified. The predictive model consisted of a function of the estimated main effects and significant interactions between the factors. 58

72 The experiment was analyzed using forward stepwise regression with all main effects and two-way factor interactions or more as candidate effects. Alternate mathematical models were envisaged based on examination of the residuals which were the collection of predictive values minus the observed values. It should be noted that the experiment was conducted for CO 2. However, the methodology can be expanded to include all other criteria pollutants of CO, NOx and PM. 3.6 Application of Mitigation Strategies As mentioned earlier, the I-4 downtown central corridor was the network under study. Several mitigation strategies were applied and tested in search for a plausible solution to the congestion problem on the I-4 corridor and at the same time mitigate the impacts of transportation emissions. Congestion pricing is one of the major applications tested and proved its effectiveness in the largest impact on reducing GHG emissions. However, congestion pricing along with any fee-based systems are not considered feasible without the implementation of ITS technologies. Consequently, real-time information on traffic conditions is essential for dynamically changing the price with demand fluctuation as is done with Managed Lane systems. The following summarizes the proposed mitigation strategies that were applied on the I-4 study corridor to manage future congestion and evaluate future transportation emissions: 59

73 o Managed Lanes (ML) (see Figure 3-2) o Restricted Truck Lanes (RTL) o Variable Speed Limits (VSL) Each of the above mentioned strategies were tested and compared against the base scenario in terms of CO 2 emissions in order to validate the developed model (Micro- TEM). Other pollutants such as CO, NOx and PM were analyzed as well. Figure 3-2: Application of Managed Lanes in VISSIM 60

74 3.7 Findings of Research Results and Conclusions The results were analyzed and documented in a useful and practical format; including a summary of the analysis, graphical illustrations and the necessary output files to support each scenario evaluation as well as the validation of the developed emission prediction model (Micro-TEM). Final recommendations were also provided regarding adaptation strategies that would improve the impact of transportation on the environment. Figure 3-3 provides a summary of the previously described research methodology which is presented in the following flow chart: 61

75 Figure 3-3: Research Approach Scheme 62

76 4. VISSIM/MOVES INTEGRATION SOFTWARE (VIMIS) 4.1 Overview This section presents the programming efforts carried out in the development of VIMIS 1.0 software; custom software developed to assist with automating the experimental design and analysis. VIMIS integrates between VISSIM and MOVES in order to facilitate the design of experiment portion on the modeled network as well as the conversion process of the VISSIM files into MOVES files. The program was developed in collaboration between me and my colleague (Dr. Hesham Eldeeb). I provided all the necessary technical information and the algorithms needed for the calculations and Dr. Eldeeb wrote the code using C Sharp (C#) programming language which can handle most sophisticated applications. The Graphical User Interface (GUI) developed for the software is shown in Figure Modules Description The program consists of four (4) main modules which were compiled using Microsoft Visual Studio The function of the first module Design Cases is to generate all the design cases (developed from JMP ) needed for the experiment in 63

77 VISSIM file format. As mentioned earlier, the custom design employed the D-optimality and I-optimality criteria; each resulted in 70 cases (a total of 140 cases). Figure 4-1: VIMIS 1.0 Software 64

78 The Design Cases module requires two main inputs: 1. The Design file is the text file (from JMP) containing a list of the design cases where each row represents a design of the 70 cases as shown in Figure The Base filename is the base name for VISSIM files. The output is 70 VISSIM input files. The main purpose is to prepare all VISSIM input files with the corresponding information in each design case. Figure 4-2: Design File for Input into VIMIS 65

79 The second module VISSIM automates the simulation runs for each case with different seed number in order to account for the randomness and variability of the simulation output. It also requires two main inputs as shown in Figure 4-3: 1. Case folder is the folder containing the input files from previous step. 2. Layout filename is the VISSIM initialization file containing all the necessary settings of all types of output needed in each run. For example, trajectory file aggregated for the whole hour of the simulation run or minute-by-minute or on a second-by-second basis. Figure 4-3: VISSIM Module in VIMIS 66

80 The third module is the operating mode OPMODE. This is a crucial step in the conversion process. It converts the trajectory output file from VISSIM into an operating mode distribution for input into MOVES. The output from VISSIM can be a very large file that cannot be opened by a conventional program such as excel or word because these programs attempt to load the whole file in memory and it is too big to fit in memory. However, VIMIS doesn t load the whole file all at once, it extracts one time step at a time (sec by sec or minute by minute), and so VIMIS is not affected by the file size. VIMIS processes each time step as it reads it. In most cases, if the output was set on a second-by-second basis, the file size can reach 10 gigabytes and cannot be opened. The magnificence of this module is that it converts this 10 gigabyte file into 300 kilobyte file (max), without the need to load the whole file as mentioned previously, and in a MOVES input format containing all the necessary links to be analyzed, types of pollutants and emission processes (running, extended idle, etc., ) to be executed by MOVES as shown in Figure 4-4. It is worth noting that MOVES can create an operating mode distribution from two default driving schedules based on average speeds. However, it may not be representative of the actual driving schedules of the modeled corridor or the specific vehicle trajectories generated from each run which highlights the importance of this module. 67

81 Figure 4-4: OPMODE Module in VIMIS The fourth and last module is the MOVES module. The main purpose was to prepare all the necessary folders and files needed as input into MOVES to run the design cases. It requires three main inputs as shown in Figure 4-5: 1. The Design file as in Module-1 which contains the list of the design cases. 2. The VISSIM output folder which contains all the output files from VISSIM runs that will be used in MOVES in addition to the operating mode file. 3. The MOVES template folder which contains all the specification files needed to be created and converted into an SQL database format to run MOVES as explained in section

82 Figure 4-5: MOVES Module in VIMIS Afterwards, the MOVES SQL browser was utilized to export the output of all the runs in excel format to be analyzed. VIMIS was used in all different stages of the research including the design of experiment, base scenario as well as the application scenarios. 69

83 5. EVALUATION OF THE I-4 CORRIDOR 5.1 Overview of the I-4 Downtown Corridor I-4 is a primary east-west transportation corridor between Tampa and Daytona cities, serving commuters, commercial and recreational traffic. I-4 is known to have severe recurring congestion during peak hours. The congestion spans about 11 miles in the evening peak period in the central corridor area as it is considered the only non-tolled limited access facility connecting the Orlando Central Business District (CBD) and the tourist attractions area (Walt Disney World). The traffic on the I-4 freeway section is collected from double inductance loops embedded in the pavement every 0.5 miles, which extends from the Walt Disney World area on the west side of the corridor to Lake Mary boulevard on the east side, for a total length of 39 miles. The interstate carries an average annual daily traffic of 200,000 vehicles on segments in Orlando. It is imperative to evaluate the environmental impacts of this corridor in terms of vehicular emissions. The modeled section was composed of approximately a 10 mile stretch of the urban limited access highway with three lanes in each direction, 12 on ramps and 13 off ramps as shown in Figure 5-1. Traffic composition included 60% passenger cars, 37% SUV s and 3% heavy-duty diesel trucks. Traffic counts were obtained from the latest FDOT online traffic information. 70

84 Figure 5-1: I-4 Downtown Corridor and Master Link Count Locations 71

85 5.2 Model Calibration The calibration was conducted for the eastbound direction during the evening peak period from 4:45 6:00 pm with the first 15 minutes as a warm up period to reflect field operations with regards to the mainline volume, on and off ramp volumes, speeds and observed queues during the peak hour. The definition of parameter calibration refers to minimizing the misfit between observed data from the real network and simulation results by fine-tuning parameter values. When running VISSIM, the user can assess the results from a visual or from a numerical point of view. The visual inspection can be observed to see the vehicle movements on the screen visualization, in order to check for network geometry which reflects that the traffic is moving in a realistic manner. Realistic manner means that the vehicles on the highway do not make U-turns at the nodes or sharp turns at the beginning or at the end of links, causing a drop in the vehicle speed. Also, sudden stops can cause shockwaves leading to disruption in the traffic flow. This emphasizes the importance of geometry coding. A small portion of the network geometry at SR 408 off-ramp overlaid on an aerial map is shown in Figure 5-2. The quantitative analysis was also carried out in parallel; when the comparison between the simulation and observed (field) data is not within recommended guidelines, it is necessary to make some changes with selected model input parameter values. 72

86 Figure 5-2: Small Portion of Network Overlaid on Aerial Map 73

87 Traffic simulation models contain numerous parameters and variables to define. With regards to VISSIM, there are two main models; car following models and lane change models. Car following models are concerned with the vehicle following behavior that affect the flow rates depending on the selected car following model. The two carfollowing models are: - Wiedemann 74: Model mainly suitable for urban traffic (arterials) and - Wiedemann 99: Model mainly suitable for interurban traffic (freeways) The lane changing models affect the driving behavior based on an extensive range of parameters. However, these are very sensitive parameters and should be adjusted with care. The driving behavior can be defined for each link type as well as for each vehicle class even within the same link. In some cases, these variables affect the entire network while others are specific to individual links. To accomplish the calibration process, a field data set, obtained from FDOT online traffic information ( was selected for the evening peak period. The data set consisted of 11 locations on the network known as master links. Traffic counts on these master links were compared with the simulated traffic volumes. As a first step in the calibration process, an initial set of runs was conducted using the VISSIM default parameters and with different seed numbers. The seed value is a 74

88 starting value for the random number generator which are called by the program and used in processes that calculate many different parameters within the simulation. After each model run, the output produced on the master link volumes were compared to the actual data. In order to gain results of high reliability, relative error between the simulated data and actual data was calculated. If any volume has relative error in excess of a specific threshold within 10%, the traffic volume on the links related to this route with the higher error were increased or decreased according to the error. Relative error was found using the formula: Actual traffic volumes - Simulated traffic volumes Relative error = * 100 Actual traffic volumes The parameters that use random numbers include car following, lane changing, driver s behavior, and release of demand. Using different seed values on the same network produce different simulation results. After several parameter adjustments and numerous runs, the best-fit seed values were selected that matched the field data within the 10% thresholds. Table 5-1 and Table 5-2 show the results of three seed values that were tested in the calibration process and matched closely the field data being within the 10% relative error threshold. Statistical analysis was conducted using Paired t-tests as explained in the next section. 75

89 Master Links Table 5-1: Master Link Counts Comparison Based on Best Seed Number Count Locations FDOT Counts Seed# VISSIM_1 Relative Error_1 JYP - OBT W. OBT % OBT - Kaley Ave S. Michigan Ave % Kaley Ave - Gore St S. SR 408 Off-Ramp % Gore St - Church St S. SR 408 On-Ramp % Church St - SR 50 N. Robinson Rd % SR 50 - Ivanhoe Blvd N. SR % Ivanhoe Blvd - Princeton St S. Princeton St % Princeton St - Par Ave N. Princeton St % Par Ave - Fairbanks Ave N. Par Ave % Fairbanks Ave - Lee Rd S. Lee Rd % Lee Rd - Maitland Ave N. Lee Rd % Overall Average % Table 5-2: Master Link Counts Comparison Based on Other Seed Numbers Master Links Count Locations FDOT Counts Seed#53 VISSIM2 Seed#1029 VISSIM3 Relative Error_2 Relative Error_3 JYP - OBT W. OBT % 6.36% OBT - Kaley Ave S. Michigan Ave % 8.14% Kaley - Gore St S. SR 408 Off-Rmp % 5.23% Gore St - Church St S. SR 408 On-Rmp % 4.84% Church St - SR 50 N. Robinson Rd % -9.95% SR 50 - Ivanhoe Bl N. SR % -9.22% Ivanhoe - Princeton S. Princeton St % 5.65% Princeton - Par Ave N. Princeton St % -7.72% Par - Fairbanks Ave N. Par Ave % -7.96% Fairbanks - Lee Rd S. Lee Rd % -7.81% Lee - Maitland Ave N. Lee Rd % 2.34% Overall Average % -1.20% 76

90 5.3 Statistical Analysis Comparing the simulated and the actual volumes on the master links for all the datasets included visual inspection and Confidence Interval method (t-test). The Confidence Interval (C.I.) is a reliable approach for comparing a simulation model with the realworld system. C.I. is performed for m collected sets of data from the field and n sets of data from the model (m and n are 11 observations each). Generalization of Gossett's t- distribution helps in testing whether or not two-sample mean come from equal or nonequal populations. However, Paired t-test is appropriate for testing the mean difference between paired observations when the paired differences follow a normal distribution. The Paired t is used to compute a confidence interval and perform a hypothesis test of the mean difference between paired observations in the population. A paired t-test matches responses that are dependent or related in a pairwise manner. The matching helps to account for variability between the pairs, usually resulting in a smaller error term, thus increasing the sensitivity of the hypothesis test or confidence interval. The null hypothesis H 0 that is tested was: H 0 : d = 0 versus H 1 : d 0 Where d is the population mean of the differences and 0 is the hypothesized mean of the differences. 77

91 If the null hypothesis is rejected, this infers that the two-sample means come from different populations and are different. To compute Paired t-test, two main computations were needed before computing the t-test. First, the pooled standard deviation of the two samples needs to be estimated. The pooled standard deviation gives a weighted average of the standard deviations of the two samples. The pooled standard deviation is going to be between the two standard deviations, with greater weight given to the standard deviation from a larger sample. The equation for the pooled standard deviation is: Where: n 1 is the sample size for the first sample, n 2 is the sample size for the second sample, S 1 is the standard deviation of the first sample, S 2 is the standard deviation of the second sample, and SP is the pooled standard deviation of the two samples. In all work with t-test, the degrees of freedom or df is The formula used for the Paired t-test is: 78

92 Where, the top of the formula is the sum of the differences (i.e. the sum of d). The bottom of the formula reads as: The square root of the following: n times the sum of the differences squared minus the sum of the squared differences, all over n-1. The sum of the squared differences: d 2 means take each difference in turn, square it, and add up all those squared numbers. The sum of the differences squared: ( d) 2 means add up all the differences and square the result. Table 5-3 shows the Paired t-test results using a confidence interval of 95% extracted from Minitab statistics software. Upper and lower limits of the resulting confidence intervals for the master links included zero. Therefore, the confidence interval method suggested that there was no significant difference between the simulated and the actual volumes on the master links of the datasets. 79

93 Table 5-3: Paired T-test of Actual vs. Simulated Data Paired T-Test and CI: FDOT, VISSIM_1 Paired T for FDOT - VISSIM_1 N Mean StDev SE Mean FDOT VISSIM_ Difference % CI for mean difference: ( , ) T-Test of mean difference = 0 (vs not = 0): T-Value = P-Value = Paired T-Test and CI: FDOT, VISSIM_2 Paired T for FDOT - VISSIM_2 N Mean StDev SE Mean FDOT VISSIM_ Difference % CI for mean difference: ( , ) T-Test of mean difference = 0 (vs not = 0): T-Value = P-Value = Paired T-Test and CI: FDOT, VISSIM_3 Paired T for FDOT - VISSIM_3 N Mean StDev SE Mean FDOT VISSIM_ Difference % CI for mean difference: ( , ) T-Test of mean difference = 0 (vs not = 0): T-Value = P-Value =

94 5.4 Model Validation As a second step, the validation process was basically validation of measures of effectiveness (MOEs), which were part of the simulation output data. MOEs were representatives of the system performance. One of the qualitative MOEs was the queue length on the I-4. Quantitative MOEs included the travel time, delay and average speeds observed on the I-4 corridor during the peak hour. As mentioned earlier, field observations and photos took place on the corridor during the evening peak hour and were compared with the simulated queues. Furthermore, Dynamic Message Signs (DMS) along the I-4 corridor display data regarding the travel time and expected delays on the 10-mile stretch during the peak hour. It should be noted that Variable Speed Limits (VSL) along the corridor were in effect as part of a safety program to improve safety. Speed limits were observed as 30 and 40 mph as shown in Figure 5-3. Speed limits, queue comparisons as well as the travel time and expected delays resulting from the simulated corridor were compared with the observed field data. 81

95 Figure 5-3: Peak Hour Variable Speed Limits on I-4 As shown in Figure 5-4 from the DMS on I-4, travel time along the 10-mile section ranged from 24 to 40 minutes during the peak hour. Simulated travel times resulted in the same range of 30 to 40 minutes reflecting an overall average speeds ranging from 15 to 20 miles per hour. Table 5-4 is an excerpt from VISSIM output for network evaluation for seed# shows the overall average network speed of mph during the peak hour. Simulated queues at Orange Blossom Trail (OBT), Kaley Avenue and SR 408 Off-Ramp were compared with field photos and were found to match the field congestion at the same time step as shown in Figure 5-5 and Figure

96 Figure 5-4: DMS Travel Time Information on I-4 Table 5-4: Network Evaluation for I-4 during Peak Hour Simulation time from to Seed Value = Parameter Value Number of vehicles in the network, All Vehicle Types 3670 Number of vehicles that have left the network, All Vehicle Types Total Distance Traveled [mi], All Vehicle Types Total travel time [h], All Vehicle Types Total delay time [h], All Vehicle Types Total stopped delay [h], All Vehicle Types Average delay time per vehicle [s], All Vehicle Types Average number of stops per vehicles, All Vehicle Types Average speed [mph], All Vehicle Types Average stopped delay per vehicle [s], All Vehicle Types

97 Figure 5-5: Field Congestion on I-4 at SR 408 Off Ramp 84

98 Figure 5-6: Simulated Congestion on I-4 at SR 408 Off Ramp 85

99 6. DEVELOPING MICROSCOPIC EMISSION PREDICTION MODEL 6.1 Overview The main objective of this research was to study the effect of major transportation related parameters on vehicular emissions on limited access highways and specifically in a microscopic and stochastic manner, hence developing an emission prediction model that is stochastic and microscopic in nature. This section explains in greater detail the procedures of the experiment taken to arrive at the resulting model. It should be noted that CO 2 was selected as the GHG pollutant, as an example, to be studied in this experiment. However, the same methodology can be expanded to include all other criteria pollutants such as CO, NOx and PM. 6.2 Design of Experiments (DOE) As mentioned in the research approach section, standard experimental designs either using full factorial or fractional factorial did not fit this research requirements and therefore, optimal custom designs were selected as the recommended design approach. Also, choosing an optimality criterion to select the design points to be run was another requirement. The custom design approach in JMP (statistical software created by SAS) generates designs using a mathematical optimality criterion. Optimal designs are 86

100 computer-generated designs that aim at solving specific research problem to optimize the respective criterion. The optimal designs fall under two main categories: 1. Designs that are optimized with respect to the regression coefficients (D-Optimality Criteria) and 2. Designs that are optimized with respect to the prediction variance of the response (I- Optimality Criteria). D-Optimal designs are most appropriate for screening experiments because the optimality criterion focuses on estimating of the coefficients precisely. The D-optimal design criterion minimized the volume of the simultaneous confidence region of the regression coefficients when selecting the design points (Johnson et al., 2011). This was achieved by maximizing the determinant of X X over all possible designs with specific number of runs. Since the volume of the confidence region is related to the accuracy of the regression coefficients, a smaller confidence region means more precise estimates even for the same level of confidence (Johnson et al., 2011). The experiment included five (5) main factors and one quantitative response (carbon dioxide emissions). The factors levels were chosen to cover all possible scenarios on limited access highways covering all level of service (LOS) ranging from LOS A to LOS F as follows: 1) Volume (from 2,000 vph to 7,000 vph) 2) Speed (from 20 mph to 70 mph) 3) Trucks (from 0% to 15%) 4) Grade (from 0% to 5%) 5) Temperature (from 50 F to 100 F) 87

101 JMP statistical software was used to generate the custom design for this experiment. In order to increase the number of levels in the custom design, additional search points were added to the coordinate exchange algorithm in JMP resulting in 6 levels for each factor plus the center points; a total of seven levels. The developed D-design and I- Design from JMP are included in Appendix A. The D-optimal design of this experiment resulted in a minimum of 36 runs obtained by considering all 36 combinations of the (5) main effects, (10) two-way and (10) three-way factor interactions as well as the (5) quadratic and (5) cubic size effects in addition to the intercept. However, 64 runs were chosen to increase the sample size. Additional six (6) degrees of freedom were introduced for lack of fit. These extra runs estimated the error variance in the model and increased the denominator s degrees of freedom for significance tests of the model coefficients. The total number of runs amounted to 70 runs. The factors levels were designated from (-1) as the low setting and (+1) as the high setting. The resulting factor settings and levels are summarized in Table 6-1. Table 6-2 provides a partial basic layout of the planning matrix in a standard order, which describes the experimental plan in terms of the actual values of the factors. Each row of the table represented one set of experimental conditions that produced a value of the response variable (Y) which was the amount of Carbon Dioxide (CO 2 ) emissions produced in kilograms (kg). 88

102 Table 6-1: Factors and Levels Setting Volume Speed Truck % Grade % Temperature Table 6-2 Partial Layout of D-Optimal Design for Five Seven-Level Continuous Factors Run# Volume Speed Truck % Grade % Temperature CO Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y16 89

103 6.3 Test Bed Modeling The test bed corridor under study was a prototype of the calibrated I-4 downtown corridor. The modeled section was composed of approximately 10 mile stretch of the urban limited access highway with three lanes in each direction. However, for experimentation purposes, the freeway section links were aggregated into 11 links, 5 main on-ramps and 5 main off-ramps as shown in Figure 6-1. Nine (9) of the segments were approximately 1-mile in length including horizontal curves and the first and last links were 0.5 mile each. Traffic composition included passenger cars, SUV s and heavyduty diesel trucks which correspond to MOVES vehicle types 21, 31 and 62, respectively. The analysis was conducted for the eastbound direction and the experimental design encompassed all peak and off-peak conditions throughout the day reflecting all different factors levels. Figure 6-1: Test-bed Prototype of the I-4 Corridor 90

104 VISSIM licenses currently do not provide an integrated emissions model for North America. Higher-level emissions Carbon Monoxide (CO), Nitrogen Oxides (NOx), Volatile Organic Compounds (VOC) statistics are available only via node evaluation. There were no results for CO 2 emissions from link evaluation which also created the motivation to link VISSIM quantitative output with MOVES. VIMIS 1.0 was used to prepare and run the 70 cases of the D-experimental design in VISSIM using the Component Object Module (COM) interface. Trajectory files generated from VISSIM were configured to output vehicle speed, acceleration, location, link length and number of vehicles and were mapped with MOVES input files through VIMIS software as well. 6.4 Moves Project Level Data As mentioned earlier, the output from the VISSIM model was used as input into the MOVES model. The Project Level database was specified in separate text files for each input parameter such as link volumes, link average speed, link length, and link grade. This process was repeated for each run of the input files. The main project parameters are described in Table

105 Table 6-3: Summary of Project Level Parameters Location County Orange County, Florida Calendar Year 2010 Month November Time One hour Weekday/Weekend Weekday Temperature 50 F F Humidity 70.0 % Roadway type Urban Restricted Access represents freeway urban road with 3 lanes per direction Types of Vehicles Type of Fuel Roadway Length (11 links - Total) Link Traffic Volume Link Truck traffic Average road grade Link Average Speed Operating Mode Output Passenger cars, Passenger trucks & Long haul combination diesel trucks Gasoline for cars and diesel for trucks Approx. 1 mile/link Total of 10 miles 2,000-7,000 vehicles per hour 0 15% Trucks 0 5 % upgrade miles per hour Running Exhaust Emissions Atmospheric CO 2, Total Energy Consumption & CO 2 Equivalent 92

106 6.5 Operating Modes & Link Driving Schedules The simulated vehicle driving cycle data was integrated into the MOVES model based on the above mentioned project-level traffic conditions. Processing the VISSIM output files for input into MOVES required the calculation of the operating mode for each vehicle in the network. The operating mode is a measure of the state of the vehicle s engine based on its speed and acceleration expressed in vehicle specific power (VSP). In MOVES, operating mode distribution input will take calculation precedence over an imported drive schedule, which will take calculation precedence over an average link speed input when more than one is entered for a given link (EPA 2010). There are 16 "speed bins" in MOVES which describe the average driving speed on a road type or link. Based on the experiment, speed levels corresponded to MOVES speed bins ID# 5 to 15 (20 to 70 mph). However, in some cases, actual operating speeds recorded from VISSIM fell below or higher than the specified speeds. In this case, other speed bins were included (less than 20 mph or greater than 70 mph). On the other hand, maximum speeds were constrained by the type of road and type of vehicles used. MOVES calculations only consider average speed ranges of ( mph) for cars and ( mph) for heavy-duty trucks (EPA 2010). All VISSIM simulated link speeds were checked to comply with MOVES speed ranges. 93

107 Table 6-4: MOVES Speed Bins Speed Bin ID Average Bin Speed Speed Bin Range speed < 2.5mph mph <= speed < 7.5mph mph <= speed < 12.5mph mph <= speed < 17.5mph mph <= speed <22.5mph mph <= speed < 27.5mph mph <= speed < 32.5mph mph <= speed < 37.5mph mph <= speed < 42.5mph mph <= speed < 47.5mph mph <= speed < 52.5mph mph <= speed < 57.5mph mph <= speed < 62.5mph mph <= speed < 67.5mph mph <= speed < 72.5mph mph <= speed Use of VISSIM s link speed and link road grade allowed the MOVES model to create an operating mode distribution from two built-in driving schedules based on its predefined 16 speed bins and an interpolation algorithm to produce a default operating mode distribution whose speeds bracket the given speed. In this research, we were concerned only with the running modes. It should be noted that "running" activity has modes that are distinguished by their Vehicle Specific Power (VSP) and instantaneous speed. The emission rate (mass of emissions per unit of time) varies with the vehicle s 94

108 operating mode which is a function of the speed and the vehicle specific power (VSP) or the related concept, Scaled Tractive Power (STP). Both VSP and STP are calculated based on a vehicle s speed and acceleration. They differ in how they are scaled. The VSP equation is used for light duty vehicles (source types 11-32) and the STP equation is used for heavy-duty vehicles (source types 41-62) (EPA 2010). VSP and operating modes are explained in detail in the next Chapter. 6.6 Design Settings versus Actual Settings VIMIS was used to run the factorial experiment with the different factor levels and settings as mentioned earlier. Each run of the VISSIM simulation was for one hour and using different random number generator to account for reasonable randomness and variability. The randomness in the experiment was also based on the random arrival patterns of the vehicles in each run. The combined effects or interactions between specific factors demand careful thought prior to conducting the experiment. Because we were performing the experiment through simulation, there were a lot of uncontrolled variables that need to be taken into account. For example, a lot of attention was given specifically to Volume and Speed. Random arrival of vehicles, dynamic network loading, stochasticity of the traffic system, and unexpected traffic demand variation due to the capacity of the highway clarifies that the design setting of the volume and/or speed were not always the same. In other words, 95

109 setting the volume dial at 7,000 vph input resulted in an actual volume output less than or greater than the designed 7,000 vph instead. Likewise were the speed settings. Furthermore, increasing the volume level more than a specific threshold affects the speed. Compounding this complexity of the main effects identified that we should be analyzing the actual output values instead of the design values as shown in Table 6-5 (sample of 20 runs). The rest of the runs are included in Appendix B. Table 6-5: Design Setting Versus Actual Setting Run# Input Volume Posted Speed Limit Trucks Grade Temp CO 2 (kg) Output Volume Output Speed

110 6.7 Analysis of Results The results were analyzed using JMP s forward stepwise regression approach with all 36 main effects (quadratic, cubic, two-way and 3-way interactions) as candidate effects according to the effect hierarchy principle. The carbon dioxide (CO 2 ) emissions output values were transformed to log space for better correlation as well as improved presentation. Preliminary analysis of the 70 runs using the stepwise regression showed an initial model including all main effects of Volume, Speed, Trucks, Grade, and Temperature. Other two-way factor interactions included Speed*Grade and Trucks*Grade in addition to two quadratic effects for the Volume and Speed factors as shown in Figure 6-2. There was no confounding between any main effects and two-way or three-way factor interactions that are aliased with each other. Furthermore, the model showed an adjusted correlation value of 99.47%. However, the model failed to pass the lack of fit test. The lack of fit was significant. As mentioned earlier, six (6) center points were added to the D-Optimal design. The center points were used to provide an estimate of the pure error as well as testing the significance of active factors. The lack of fit table shows a special diagnostic test and appears only when the data and the model provide the opportunity. The idea is to estimate the error variance independently of whether this is the right form of the model. This occurs when multiple observations occur all at the same (x) variable, known as center 97

111 points. The error that is measured for these exact replicates is called pure error. This is the portion of the sample error that cannot be explained or predicted no matter what form of model is used. The difference between the residual error from the model and the pure error is called lack of fit error. Lack of fit error can be significantly greater than pure error, which means that the model represents the wrong functional form of the regressor. In that case, a different kind of model fit should be tried. This step-by-step iterative construction of the regression model that involved automatic selection of independent variables can be achieved either by trying out one independent variable at a time and including it in the regression model if it is statistically significant, or by including all potential independent variables in the model and eliminating those that are not statistically significant, or by a combination of both methods. An initial assumption was made to remove the main effect of the Temperature then the Speed*Grade interaction. The stepwise regression was done for both cases. However, the analysis showed no significant difference between the two cases. With further examination of the speed factor and its interaction terms, the Temperature main effect only had to be eliminated in order to represent a right form of the model with no indication of lack of fit. This improved form of the model including the Volume, Speed, 98

112 Trucks and Grade effects with their interaction terms showed an adjusted R 2 of 97.9 as shown in Figure 6-3. The Temperature factor in the model represented the effect of air condition (AC) being turned on in the vehicles during high or low ambient temperature which was an indirect measure of the emission rates, thus statistically insignificant. It should be noted that the high value of the model s R 2 is attributed to the fact that emission rates calculated from MOVES are based on its built-in default database. MOVES database is calibrated and validated from field datasets as explained in Chapter 3. The model is based on results of millions of emission tests and large amounts of validated data from several sources, such as vehicle inspection data and maintenance (I/M) programs, remote sensing device (RSD) testing, and portable emission measurement systems (PEMS). The main variability in the experiment is actually attributed to the stochastic nature of the traffic simulation environment. The resulting prediction expression is included in Appendix A. Since the traffic network consisted of 11 links, further analysis was conducted on link by link basis. That is each run of the experimental design resulted in 11 outputs (one output for each link). Therefore, total sample size of 770 points was analyzed using the stepwise regression and the same model was valid across all the links with adjusted correlation value ranging from 94% to 98% as shown in Figure 6-4 (a) for link 1 and (b) for link 9. 99

113 To recall, the primary focus of the D-optimal design was to find the main effects that significantly influenced the response where the goal was to estimate the regression coefficients. Another goal to the experiment was to achieve precision in terms of the response variable and the overall prediction model which can be achieved through the I- optimality criterion especially when the resulting model included a second-degree order. Therefore, another design was created from JMP 9.0 using the I-optimality criterion resulting in another 64 runs with six levels and 6 more center points for the determination of lack of fit totaling another 70 runs. This second set of I-optimal design actually served as a confirmation experiment to the D-optimal design as the results were identical. We came to the conclusion of the final form of the I-Design model terms showing an adjusted R 2 value of 97.3 with no indication of lack of fit as shown in Figure

114 Analysis of Variance Source Model Error C. Total Lack Of Fit Source Lack Of Fit Pure Error Total Error DF DF Sum of Squares Sum of Squares Sorted Parameter Estimates Term Trucks(0,0.15) Grade(0,0.05) Volume(2000,7000) Volume*Volume Temp(50,100) Speed(20,70) Trucks*Grade Speed*Grade Speed*Speed Estimate Mean Square Summary of Fit RSquare RSquare Adj Root Mean Square Error Mean of Response Observations (or Sum Wgts) F Ratio Prob > F <.0001* F Ratio Mean Square Std Error t Ratio Prob > F * Max RSq Prob> t <.0001* <.0001* <.0001* <.0001* <.0001* <.0001* <.0001* <.0001* <.0001* Figure 6-2: Summary of Stepwise Regression for Initial Model (D-Design) 101

115 Summary of Fit RSquare RSquare Adj Root Mean Square Error Mean of Response Observations (or Sum Wgts) Lack Of Fit Source Lack Of Fit Pure Error Total Error DF Sum of Squares F Ratio Mean Square Prob > F Max RSq Figure 6-3: Summary of Stepwise Regression for Final Model (D-Design) 102

116 (a) LINK# 1 103

117 (b) LINK# 9 Figure 6-4: Validation of the Regression Model by Link 104

118 Actual by Predicted Plot Ln CO2 Predicted P<.0001 RSq=0.98 RMSE= Analysis of Variance Source Model Error C. Total Lack Of Fit Source Lack Of Fit Pure Error Total Error DF DF Sum of Squares Sum of Squares Sorted Parameter Estimates Term Trucks(0,0.15) Grade(0,0.05) Volume(2000,7000) Volume*Volume Trucks*Grade Speed(20,70) Speed*Grade Speed*Speed Estimate Mean Square Summary of Fit RSquare RSquare Adj Root Mean Square Error Mean of Response Observations (or Sum Wgts) F Ratio Prob > F <.0001* F Ratio Mean Square Std Error t Ratio Prob > F Max RSq Prob> t <.0001* <.0001* <.0001* <.0001* <.0001* * * * Figure 6-5: Summary of Stepwise Regression for Final Model (I-Design) 105

119 6.8 Discussion Prediction profiles for the significant factors selected by the model are shown in Figure 6-6 showing the actual values of the response variable as CO 2 while Figure 6-7 shows the response variable as the log space for CO 2 (Log). The prediction profiler displays prediction traces for each factor. The steepness of a prediction trace reveals a factor s significance. This shows the significant effect of the trucks and grade on the increase of carbon dioxide. Furthermore, Interaction profiles are shown on Figure 6-8. Evidence of interaction shows as non-parallel lines. The substantial slope shift proves the significant interaction involving speed and trucks with the grade. When analyzing the results of the custom design, a separate prediction equation for each dependent variable (containing different coefficients but the same terms) is fitted to the observed responses on the respective dependent variable. Once these equations are constructed, predicted values for the dependent variables can be computed at any combination of levels of the predictor variables. If appropriate current values for the independent variables were selected, inspecting the prediction profile can show which levels of the predictor variables produce the most desirable predicted response on the dependent variable. Through the desirability function, JMP adjusts the graph to display the optimal settings at which CO 2 would be minimized as shown on Figure

120 Figure 6-6: Prediction Variance for the Design Factors at Center Points. Figure 6-7: Prediction Variance for the Design Factors at Center Points (Log Space). 107

121 Figure 6-8: Interaction Profiles for the Design Factors at Optimal Settings (Log Space). Figure 6-9: Prediction Variance for the Design Factors at Optimal Settings (Log Space). 108

122 On the other hand, the results showed that there s a strong correlation between the speed and the emission rates instead of total emissions. The potential for emissions rate reductions through travel speed adjustments was significant for speeds between about 55 and 60 mph while maintaining volume levels up to 80% and 90% of the roadway capacity on limited access highways provided that the grade and truck percentages are at their minimum. It was observed from the runs that at certain volume levels and speeds ranging from 55 to 60 mph; the CO 2 emission rate was minimized resulting in efficient operation with higher miles per gallon on link by link basis as highlighted in Table 6-6 and plotted on Figure This also showed the effect of other factors such as truck percentages and roadway grades. Emissions at a given speed appeared to be influenced by the vehicle s engine loading getting to that speed from previous speed, which is the acceleration rate. The difference in CO 2 emissions between speeds lower than 55 mph and higher than 60 mph were represented by a rapid succession of high speed and/or high power events represented by the Vehicle Specific Power (VSP), which resulted in more aggressive driving cycles likely resulting in higher emissions. The Speed was almost flat especially in the range between 55 mph and 60 mph, while increases outside this range. Conversely, the relationship developed between the volume and CO 2 was found to be quadratic as was the case with the speed. Figure 6-6 and Figure 6-7 also show a good 109

123 fit between the modeled data and the fitted curve especially when compared with the freeway capacity. The increase in CO 2 emissions was limited by the capacity of the roadway and the amount of traffic that it can handle in our case 7,200 vph - capacity on a 3 lane freeway. The fitted data was also compared with the modeled data at the same settings to examine the model validity. For example, run# 25 (Appendix A, Table A-1) showed an output of 8232 kg while the fitted data resulted in approximately 8390 kg of CO 2 emissions, a difference of 1.9%. Further model validation is conducted in Chapter 8. The analysis of the experiment determined the lowest settings for volume, trucks and grade while the speed setting between 55 and 60 mph; optimum speed for engines that requires the least amount of work to complete fuel combustion. These settings yielded approximately 7900 kg of CO 2 emissions. This also proved that the optimal design approach was an accurate and powerful way to approach these types of maximization/minimization problems on a response variable. 110

124 Table 6-6: Volume-Speed-CO 2 Emission Rates at Zero Truck and Zero Grade Levels VMT Length (ft) Volume (vph) Speed (mph) Truck% Grade% CO 2 (kg) Miles per Gallon CO 2 Per mile/veh

125 VMT Length Volume Speed Truck% CO 2 (ft) (vph) (mph) Grade% (kg) Miles/Gallon CO 2 /mile/veh

126 Figure 6-10: Speed - CO 2 Emission Rates Relationship 113

127 6.9 Meta Model for Transportation Emissions Micro-TEM Introduction to Meta Models In most engineering problems, experimentation and/or simulations are needed to optimize a system or evaluate a design objective as a function of several design variables. Alternatively, many real world problems require extensive simulation runs to be explained. As a result, design optimization, design space exploration, or sensitivity analysis (what-if scenarios) becomes time consuming since they involve hundreds or even thousands of simulation evaluations. Therefore to resolve this problem, approximation models, known as Meta models or response surface models are constructed that replicate the performance of a simulation model as closely as possible while being computationally accurate and cheaper to evaluate. A Meta model is an engineering technique used when an outcome of interest or response is not easily calculated or directly measured (such as transportation emissions); therefore a substitute model of this response is developed instead. Meta models are constructed using datadriven bottom-up approach as was constructed in this research using optimal design along with micro-simulation. Since the particular internal functioning of simulation models (i.e. VISSIM, MOVES) is not always explicitly defined and depends on several underlying models, thus the input-output activities are essential. 114

128 The developed Meta model was constructed based on modeling the CO 2 emissions response from the microscopic simulation to a sufficient number of intelligently designed data points using (D) and (I) optimal designs. This approach is also known as black-box modeling or behavioral modeling. The main purpose of the developed Micro-TEM (Microscopic Transportation Emissions Meta Model) is to serve as an alternative model for predicting transportation emissions on limited access highways in lieu of running simulations using a traffic model and integrating the results in an emissions model. The model accuracy was validated against different applications and compared with several simulation outputs on the link level as well as the route level covering all possible combination scenarios as explained in the following section Micro-TEM As mentioned earlier, the main objective of this research was to study the effect of each of the significant transportation parameter on transportation emissions represented in CO 2 emissions as a case study. The optimal design approach facilitated this process. Combining the 140 runs from the D-design and I-design on the modeled network with 11 links resulted in 1,540 data points creating the opportunity to model the effect of each of the studied parameter on CO 2 emissions at different settings for the remaining parameters represented by the Meta model displayed in Figure 6-11 and Figure The modeled 115

129 curves represent the results of the response surface model developed from the custom design. The curves on Figure 6-11 show the effect of speed on CO 2 emission rates at different temperature, truck% and grade% levels. As can be seen from the curves, there are three main conclusions. First, emission rates are the lowest at speeds between 55 and 60 mph as concluded previously. Second, the effect of temperature increases gradually with the increase in truck and grade %. Third, there is an enormous increase in emission rates (from 0.5 to 1.7 kg/veh-mile)) due to the effect of grade (0-5)% and truck (0-15)% amounting to approximately 340%. The practicality of the developed emission rate versus speed curves lie in the determination of an average speed on a link level or route level. If an average link speed is determined, emission rates can be predicted, thus total emissions at specific volume, grade%, truck% and ambient temperature level. 116

130 Figure 6-11: Speed-CO 2 Emission Rates at Different Temp, Truck & Grade Levels 117

131 Similarly, the polynomial curves on Figure 6-12 show the effect of the traffic volume on CO 2 emission rates at different temperature, truck% and grade% levels. Although it appears that the speed was not taken into account; when examining the data points, a V-like shape was observed as shown on Figure The polynomial fitted curves were based on the best fit of the data points by minimizing the sum of squares of their deviations through the least squares method. However, the effect of the speed was observed in the spectrum created by the V-shaped curve, ranging from speeds of 20 mph to 60 mph with emission rates higher at the 20 mph curve. Furthermore, there was another observed deviation in the data points closer to the 60 mph curve. These data points reflect a higher emission rates for speeds higher than 60 mph which matches the speed curves shown on Figure To demonstrate the observed Speed Spectrum on the volume curves, more detailed data points were plotted at different speeds as shown on Figure It was found that the speed spectrum curves follow a power-law function. The developed traffic volume curves can be used to predict emissions per mile, thus total emissions based on a link or group of links with a specific volume or flow rate at different parameter settings. 118

132 Figure 6-12: Volume-CO 2 Emission Rates at Different Temp, Truck & Grade Levels 119

133 3000 CO2 Emission Rate (kg/mile) mph 60 mph Volume (vph) Figure 6-13: Traffic Volume CO 2 Emission Rates Relationship (0%Trucks- 0%Grade) Another major finding from this experiment is the speed-density relationship. Based on the traffic flow theory, the speed-density relationship is known to be linear. However, the curve fitting shown on Figure 6-15 showed otherwise; a third-order model with a 94% coefficient of correlation (R 2 ). As stated by Wang et al. (2009): Speeddensity (or concentration) models in a deterministic sense, whether single or multiregimes, have a pairwise relationship; that is, given a density there exists a corresponding speed from a deterministic formula.., There is a distribution of traffic speeds at a certain density level due to the stochastic nature of traffic flow, this is in 120

134 contrast to the pairwise pattern from deterministic models. This statement and the graph shown proved that the developed Micro-TEM model is stochastic and microscopic in nature. The distribution of speeds around the same density level was attributed to many factors that correspond to the stochastic feature of freeways and their associated capacity such as the dynamic traffic flow, random arrival and interaction between vehicles, in addition to the randomness in the driving behavior. The speeddensity curve is a useful tool to predict density, hence freeway level of service (LOS) CO2 Emissions (kg/m) R² = R² = T100, T0, G0, S17 20 T100, T0, G0, S20 25 T50, T0, G0, S17 20 T100, T0, G0, S25 30 T100, T0, G0, S30 40 T100, T0, G0, S T50, T0, G0, S Flow Rate (vph) Figure 6-14: Speed Spectrum on Volume-CO 2 Emission Rate Curves 121

135 Figure 6-15: Stochastic Speed Density Relationship The developed model (Micro-TEM) can be used to predict emission rates as well as total emissions on limited access highways at different parameter settings without the need to run any micro-simulation whether traffic or emission models. Micro-TEM was validated through different scenarios and applications as will be explained in Chapter 8. The resulting emissions were compared with MOVES output and the relative error did not exceed 10% in the worst case scenario. This approach added flexibility to emissions testing covering all the achievable combinations of instantaneous speeds and accelerations, which was used to develop emissions for all desired driving patterns. 122

136 The analysis of the experiment developed a predictive model (Micro-TEM) for CO 2 emissions and identified optimal settings of the key factors. Contour plots were generated for the surface model shown in Figure The predictive model consisted of a function of all the significant main effects and the speed interaction with the truck and grade factors as well as the quadratic effect of the Volume and the Speed. The predictive model expression is included in Appendix A. A simple form for the model equation can be represented as: Emission Rates = a + b (volume 2 ) + c (speed 2 ) + d (truck%) + e (grade%) Finally, specific model limitations included the following: 1) Freeways (up to 3 lanes in each direction plus auxillary lanes) 2) Parameter settings are within the studied limits (speed, volume, etc.) 3) Response variable for CO 2 emissions (other pollutants can be included) 4) Running exhaust emissions mode only Figure 6-16: CO 2 Surface Profiler for the Predicted model. 123

137 7. EMISSION ESTIMATION APPROACHES 7.1 Overview There are several ways to estimate emissions and transportation agencies and researchers have a long history of implementing techniques in estimating emissions. However, precision and accuracy depends mainly on the methodology used. For example, old traditional methods for creating emission inventories utilized annual average estimates; others have estimated emissions using one average speed and volume on a long stretch of roadway. Currently, more accuracy has been established using microscopic analyses through the reduction of time and distance scales while splitting the network links into sub-links and utilizing second-by-second operations to calculate emissions as demonstrated in this research. However, even at this level of detail, there exist different emission estimation approaches that can be investigated through the integration of VISSIM and MOVES models. This section explains in greater detail the main differences between these estimation approaches and provides an emission sensitivity analysis in calculating CO 2, CO, NOx and PM emissions in each approach. 7.2 VISSIM Input/Output Data The VISSIM model generates a significant amount of output data detailing each vehicle s performance within the network which are critical for calculating air pollutant 124

138 emissions. These details include second-by-second speed-acceleration profiles, network characteristics, and other vehicle parameters. For this study, three types of output data were generated from VISSIM runs to correspond with vehicle characterization inputs within MOVES. The first output included link-average speeds during the entire peak hour; the second output was a set of linkinstantaneous speeds but on a second-by-second basis, and the third output included vehicle trajectory data: length, speed, acceleration, weight, location and grade, on a second-by-second basis as shown in Table 7-1. All of the inputs required for MOVES emissions model were generated from VISSIM. Table 7-1: Excerpt from VISSIM Vehicle Trajectory Data t Link# Veh Length VehNr Weight a V DistX WorldX WorldY Grade

139 The subject test-bed is the previously described 10-mile prototype of I-4 as shown in Figure 7-1. Traffic composition was set at 60% passenger cars (LDGV), 37% passenger trucks (LDGT) and 3% heavy-duty diesel trucks (HDDV) as obtained from Florida Department of Transportation (FDOT) traffic information. The study period encompassed the eastbound evening peak hour from 5:00 to 6:00 pm which carries more than 6,000 vehicles per hour. The speed limits on the study corridor over the 10-mile section during the peak hour range from mph as part of a Variable Speed Limit (VSL) safety program. Therefore, VISSIM input volumes were assigned a speed distribution based around the posted speed limits. Roadway links were coded with 0% grade as nominal grade changes exist on the study corridor. Figure 7-1: Test-bed Prototype of the I-4 Corridor 126

140 7.3 Moves Project Level Data As mentioned earlier, MOVES project level database files include meteorology data, traffic composition & percentage of trucks, length, volume, average speeds & grade, distribution of vehicles age, operating mode distribution for running emissions, link drive schedules and fuel information (gasoline, diesel). A summary of MOVES project level parameters used can be seen in Table 7-2. Table 7-2: Summary of project level parameters Location County Orange County, Florida Calendar Year 2010 Month November Time 5:00 PM to 6:00 PM (one hour) Weekday/Weekend Weekday Temperature 75 F Humidity 70.0 % Roadway type Urban Restricted Access represents freeway urban road with 3 lanes in each direction 60% Passenger cars LDGV(21), 37% (%)Types of Vehicles(Source type) Passenger trucks LDGT(31) & 3% Long haul combination diesel trucks HDDV(62) Gasoline for passenger cars (LDGV) and Type of Fuel trucks (LDGT); diesel for heavy duty diesel trucks (HDDV) Roadway Length (11 links - Total) Approx. 1 mile/link Total of 10 miles Link Traffic Volume ,500 vehicles per hour Link Truck traffic 3% Heavy Duty Diesel Trucks (HDDV) Average road grade 0 % Link Average Speed miles per hour Pollutant Process Running Exhaust Emissions Output CO, NOx and Atmospheric CO 2 127

141 7.4 Vehicle Activity Characterization Average Speeds, Link Drive Schedules & Operating Modes Selection of vehicle speeds and volumes on network links is a complex process due to the fundamental relationship between speed and volume. The recommended approach for estimating average speeds and volumes is to post-process the output from a traffic model. Therefore, the simulated vehicle driving cycle output data from VISSIM was input into the MOVES model based on the above mentioned project-level traffic conditions to calculate CO, NOx and CO 2 emissions. Four approaches were used to estimate vehicle emissions for the hour. The first was a simple hand calculation that estimated emissions from total VMT at one average speed for the whole 10-mile stretch just to illustrate the old method of creating a mobile source emission inventory. Three other estimation approaches were used, all of which used 1-mile sub-sections: average speeds (AVG), link drive schedules (LDS), and operating mode distributions (OPMODE). The MOVES operating mode distribution allows one to define the amount of travel time spent in various operating modes including: braking, idling, coasting, and cruising/accelerating within various speed ranges and at various ranges of vehicle specific power (VSP). In all of the model runs, only the running exhaust emissions were modeled. 128

142 Use of the AVG approach forces MOVES to use built-in driving schedules based on predefined speed bins and an interpolation algorithm to produce a default operating mode distribution. On the other hand, in the LDS approach, each vehicle, or group of similarly performing vehicles, is modeled on a second-by-second basis using instantaneous speeds. However, even with this great amount of activity detail, MOVES will convert it to an operating mode distribution based on its internal algorithms. In the third approach (OPMODE), all vehicle activity data from VISSIM are pre-processed to develop the simulated operating mode distribution on a second-by-second basis, and this is input directly into MOVES. Thus the main differences among these three approaches lie in the distinctions among the representations of each operating mode distribution Vehicle Specific Power (VSP) MOVES calculates emissions by calculating a weighted average of emissions by operating mode. For running exhaust emissions, the operating modes are defined by Vehicle Specific Power (VSP) from cars or the related concept from trucks, Scaled Tractive Power (STP). Both VSP and STP are calculated based on a vehicle s speed and acceleration, but they differ in how they are scaled. The VSP, as shown in equation 1, is used for light duty vehicles (source types 11-32), while the STP, as shown in equation 2, is used for heavy-duty vehicles (source types 41-62) (USEPA, November 2010). 129

143 (1) (2) Where: VSP=Vehicle Specific Power (kw/ton) STP= Scaled Tractive Power (kw/ton) M = vehicle Mass (metric tons) A = rolling Term A (kw-s/m) B = rotating Term B (kw-s 2 /m 2 ) C = aerodynamic drag Term C (kw-s 3 /m 3 ) v = Instantaneous vehicle velocity (m/s) a = Instantaneous vehicle acceleration (m/s 2 ) f = fixed mass factor, g = gravitational acceleration (m/s 2 ) = Road grade (fraction) Since "running" activity has modes that are distinguished by their VSP and instantaneous speed, the Operating Mode Distribution Generator (OMDG) classifies vehicle operating modes into different bins associated with vehicle specific power and speed, and develops mode distributions based on pre-defined driving schedules. The MOVES emission rates are a direct function of VSP, a measure that has been shown to have a better correlation with emissions than average vehicle speeds (USEPA, 2002), and users can input locally-specific VSP distributions based on the exclusive characteristics of the modeled system. VSP represents the power demand placed on a vehicle when the vehicle operates in various modes and at various speeds. In other words, the operating mode is a measure of the state of the vehicle s engine at that particular moment. This function produces operating mode fractions for each bin, which are used as one of several 130

144 inputs for computing base emission rates. The output of the simulation run is a vehicle trajectory file that, for every second of the simulation, indicates the instantaneous speed and acceleration of every vehicle in the network, an excerpt is shown in Table 7-1. Since both speed and acceleration are available in the micro-simulation output for every vehicle for every second of simulation, MOVES operating mode distributions based on VSP were computed only in the OPMODE vehicle activity characterization approach. This is thought to be a more accurate way of capturing driving cycle patterns when literally thousands of vehicles have their trajectories traced, as in simulation. Since the grade was set at 0 percent in the simulation, the term for it falls out of the equation and is not used. 7.5 Emissions Results and Analysis Table 7-3 provides a comparison of the results for CO, NOx and CO 2 emissions (kg) when the MOVES analysis was conducted using the three simulation approaches. For the same 10-mile stretch of I-4, those three approaches resulted in CO 2 emission estimates ranging from more than 19,000 kg to almost 26,000 kg. By comparison, the hand calculation method gave 29,100 kg. In general, the AVG approach estimated higher total emissions than the OPMODE approach while the LDS approach estimated lower total emissions as shown on Figure 7-2. If indeed the OPMODE approach is the most accurate, the AVG approach resulted in overestimation of emissions while the LDS approach resulted in underestimation of emissions. 131

145 Heavy Duty Diesel Trucks (HDDT) Passenger Trucks - Gasoline (LDGT) Passenger Cars-Gasoline (LDGV) Table 7-3: Emissions by pollutant, source type, link & vehicle activity characterization Source Type LNK CO Emissions (kg/hr) NOx Emissions (kg/hr) CO 2 Emissions (kg/hr) AVG LDS OPMODE AVG LDS OPMODE AVG LDS OPMODE Total (LDGV) Total (LDGT) Total (HDDV) Total Emissions

146 (a) CO (b) NOx (c) CO 2 Figure 7-2: Total Emissions by Vehicle Type & Estimation Approach 133

147 Figure 7-3 (a-c) shows in greater detail the differences between the three approaches, link by link, and show the greater variability in emissions from the OPMODE approach when compared with the AVG and the LDS approaches. This is attributed to the fact that average speeds generally omit detailed vehicle activity such as acceleration and deceleration. Furthermore, this variability increases at certain locations (links 1-3 and 6-8) and decreases at other locations (links 4-5 and 9-10). When examining the network, it is found that links 1 and 2 are considered as loading points on the network from the mainline as well as the on-ramp, while link 3 is a discharging location (off-ramp) which creates a weaving area for vehicles trying to enter the network and others leaving. Weaving areas cause excessive acceleration and deceleration resulting in an increase in braking, deceleration, idling, and acceleration. Large fractions of vehicles spend a substantial amount of time operating in these modes of stop and go operation which are characterized by relatively high vehicle specific power and low speeds. The same pattern was seen for links (6, 7, and 8). However, emissions are lower on links 6-8 due to the relatively longer weaving distance between the on-ramp and the off-ramp resulting in a relatively smoother operation in addition to lower volumes compared to the volume and weaving distance on links 1-3. Figure 7-4 (a-c) maps the operating mode distribution in details for links 1, 2 and 10 for comparison purposes. The remaining links are included in Appendix C. 134

148 (a) (b) 135

149 (c) Figure 7-3: Emissions Variation on Corridor Links for PC by Estimation Approach 136

150 (a) (b) 137

151 (c) Figure 7-4: Link Operating Mode Distribution by Vehicle Type on Selected Links Furthermore, the results displayed in Table 7-3 enable us to evaluate the behavior of the studied pollutants with respect to each other. According to Figure 7-2b, there is an apparent increase in NOx emissions in all estimation approaches when compared to the fleet composition; contrary to CO and CO 2. The 37% passenger gasoline trucks generated higher emissions than the 60% passenger gasoline cars while only 3% heavy duty diesel trucks generated higher emissions than the passenger cars (60%) and the passenger trucks (37%). This is attributed to a combination of increased engine loading in heavier vehicles 138

152 (and thus higher combustion temperature) and the fact that diesel engines produce much more NOx than gasoline engines. Table 7-4 addresses the effect of VMT along with the operating mode distribution (speeds and accelerations) on the CO, NOx, and CO 2 emissions on selected corridor links. These links were selected for comparison purposes. As shown in Table 7-4, emission rates (emissions per vehicle-mile) are the highest on link 1 when compared to the rest of the network links although Figure 7-3(a), (b), (c) seem to show otherwise. The difference lies in the distance traveled (link 1 was only one-half mile long). All parameters should have the same scale for a fair comparison between them. By normalizing the emissions to vehicle-miles, it was found that link 1 has the highest emission rate (e.g. 583 grams/veh-mile CO 2 ). Furthermore, Link 1 has a greater fraction of the passenger car activity (about 30%) in the braking, idling and low speed coasting operating modes (0, 1 and 11), as well as 20% in cruise/acceleration modes (12-16) at lower speeds (1-25 mph) as shown in Figure 7-4(a). Link 2 shows nearly similar operating mode distribution patterns but with lower percentages than Link 1, especially in operating mode 11 (coasting), and had a lower emission rate (486 grams/veh-mile CO 2 ). It should be noted that link 2 has 152 vehicles more than link 1. However, emission rates are lower which is attributed to improved traffic operations compared to link 1. Link 10 has the smallest emission rates on the corridor links (362 grams/veh-mile CO 2 ). A greater fraction of the vehicle activity is in operating modes (moderate speed coasting and 139

153 cruise/acceleration); here there are relatively higher speeds (25-50 mph) with almost 0% idling or braking and 11% coasting. It is concluded that increased braking, idling and coasting at lower speeds, along with the consequent re-accelerating, described as acceleration events, have a significant impact on pollutant emission rates. Table 7-4: Link emissions per vehicle-mile by source type Link# Emissions (kg) Link Emissions (Kg/Veh-Mile) Link Source Link Avg. Dist. Type Volume Speed CO NOx CO (miles) 2 CO NOx CO 2 (mph) LDGV LDGT HDDV Total LDGV LDGT HDDV Total LDGV LDGT HDDV Total

154 7.6 Discussion This section presented a detailed examination of three different vehicle activity characterization approaches to capture the environmental impacts of vehicular travel on a limited access urban highway corridor. The VISSIM/MOVES integration using VIMIS 1.0 was used to estimate emissions derived from three approaches characterizing vehicle activity, namely average speeds (AVG), link drive schedules (LDS), and operating mode distribution (OPMODE). The OPMODE approach covered all the simulated combinations of instantaneous speeds and accelerations, and was used to develop detailed emissions for all desired driving patterns. The results demonstrated that obtaining second-by-second vehicle operations from a traffic simulation model are essential to achieve the most accurate operating mode distributions and presumably the most accurate emissions estimates. Specifically, emission rates are found to be highly sensitive to the frequent acceleration events that occur at lower speeds, that is, frequent braking/coasting, idling (operating mode bins 0, 11 and 1, respectively) and re-accelerating. In the lower speed range (< 25 mph), the emission rates for VSP bins up to 12 kw/ton are actually higher than the emission rates from the same VSP bins in the higher speed range (> 25 mph). In addition, results from VISSIM show that there were more frequent speed changes in the lower speed range, perhaps due to increased weaving and more aggressive driving. These two facts likely 141

155 accounted for the higher emissions on links 1-3 compared with emissions on links 4, 5, and 10 regardless of the amount of vehicle miles traveled (VMT). Moreover, the use of average speeds often conceals the effects of acceleration/deceleration on emissions. Using AVG and LDS approaches resulted in overestimation or underestimation of emissions, respectively, when compared to the OPMODE approach. In addition, the results of this section addressed previous conclusions (Int. Panis et al., 2006) regarding evaluating speed management policies in Europe through modeling instantaneous traffic emissions and the influence of using an average speed approach. They concluded that active speed management has no significant impact on pollutant emissions. They also concluded that the analysis of the environmental impacts of any traffic management and control policies is a complex issue and requires detailed analysis of not only their impact on average speeds but also on other aspects of vehicle operation such as acceleration and deceleration (Int. Panis et al., (2006). This study limited the pollutants to only CO 2, CO, and NOx; however, methods have been demonstrated that can be used for other mobile-source pollutants. Furthermore, CO and PM dispersion modeling analyses, which are often required for roadway projects, can use the resulting spatially-determinate EFs in roadway dispersion models such as CAL3QHC or AERMOD to predict concentrations of various pollutants near roadways, or in gridded ozone modeling. 142

156 8. MODEL APPLICATIONS 8.1 Overview The main objective of the model applications in this chapter is to study congestion mitigation strategies on the I-4 corridor and evaluate the environmental impacts in each scenario in terms of vehicular emissions and at the same time validate the developed model Micro-TEM. Since the VISSIM model was properly calibrated and validated, it was ready to perform and evaluate a range of operational strategies through various scenarios and analysis of the model outputs. Three main applications were proposed in the research methodology; Variable Speed Limits (VSL), Managed Lanes (ML) and Restricted Truck Lanes (RTL). However, the VSL application was conducted and evaluated as part of the experimental design approach since it was already implemented on the I-4 corridor during the peak hour. Therefore, the remaining strategies would include the ML and RTL as explained in the following sections. 8.2 Managed Lanes (ML) Roadway agencies face several challenges to expand freeway capacity due to the increase in construction costs, restricted right-of-way as well as environmental 143

157 regulations. Thus, transportation professionals are seeking solutions for managing the demand on existing limited access facilities efficiently and providing options for travelers. The concept of MLs is an increasingly accepted countermeasure that aims at efficiently utilizing the existing limited access facilities by restricting access to one or more lanes to certain vehicle types on a facility that is parallel to existing general use lanes (GUL). ML is also considered one of the congestion pricing applications, also called Value Pricing, Variable Pricing, or Peak Hour Pricing, which is the practice of charging motorists more to use a roadway, bridge, tunnel or parking spot during periods of heaviest use. The term, congestion pricing, comes into play at places where a charge, fee or toll is applied with the intent of reducing car trips or encouraging shorter parking stays. Johnston et al. (1996) used travel demand simulations to demonstrate that new HOV lanes may increase travel (vehicle-miles) and increase emissions when compared to transit alternatives. They recommended better travel demand modeling methods for such evaluation. Sinprasertkool (2010) concluded that higher toll rates tend to generate higher toll revenues, reduce overall CO and NOx emissions, and shift demand to general purpose lanes. However, HOV privileged treatments (HOV2 paying 50% and free service for HOV3+) at any given toll level tend to reduce toll revenue, have no impact or reduce system performance on managed lanes, and increase CO and NOx emissions, when compared to SOV. Kall et al. (2009) studied air quality impacts on I-85 managed lanes in 144

158 Atlanta. They concluded that emissions results were mixed, with small estimated increases for CO, NOx, PM10 and small decreases for HCs. Higher concentrations were found in most of the study area for the modeled pollutants (CO, NOx, and benzene), with the largest increase near the corridor. Overall, changes in emissions were small indicating little impacts of the managed lane project on air quality. Quantitative research on the air quality benefits of managed lane facilities, also known as High Occupancy Toll (HOT) lanes has been limited and inconclusive. With growing interest in mitigating climate change, primarily greenhouse gas emissions in addition to other pollutants from transportation sources and in strategies to reduce congestion, research is needed to examine the emissions benefits of ML treatments. In this section, the air quality impacts of ML and GUL were examined and compared with the existing conditions (EX) on the calibrated I-4 corridor using VIMIS 1.0. The future scenario for the I-4 managed lanes was obtained from the FDOT. The FDOT plans to implement two express toll lanes in each direction (Managed Lanes) with variable tolls concept during the congested travel periods. Detailed information regarding the I-4 ML project as well as the dynamic algorithm for calculating tolls on the ML are not released to the public yet and therefore, only the released information on the I-4 website can be disclosed at this moment ( The new concept as well as the modeled network including the dynamic assignment model on the basis which 145

159 they calculated the variable toll system was provided by the FDOT project manager to ensure the same results replicated. It should be noted that the future scenario also includes major improvements along the general use lanes over the total project length. Ultimate improvements along the corridor totaled 20 miles in length starting from west of Kirkman Road to east of SR 434. It includes the reconstruction of 15 interchanges, 60 new bridges; improving overall safety with the goal to increase the design speed to 60 mph. Access to and from the tolled ML will be limited since the intention is to increase speed over longer distances and reduce disruption to the traffic flow. The GUL anticipated opening year 2020 traffic demand is 8,000 vph and expected to reach 9,700 by 2030 (3,250 vphpl), while the opening year traffic demand for the ML is expected to be 2,500 vph and anticipated to reach 3,000 vph by the year 2030 (1,500 vphpl). 8.3 Restricted Truck Lanes (RTL) Restricted Truck Lanes (RTLs) are lanes designated only for the use of trucks. The main purpose of RTL is to separate the heavy truck traffic from other mixed-flow traffic in order to improve operations and safety. Although the concept of RTL is not new, very few truck-only lanes exist in the US. The majority of the states that have implemented RTL restrict trucks to certain lanes. However, all other vehicles are still allowed to use all lanes including the truck lanes. Most studies in the literature revealed that exclusive truck lanes are the most plausible solution for congested highways based on specific factors 146

160 such as truck volumes exceeding certain percentages of the vehicle mix during off peak and peak periods. Heavy truck traffic often results in significant congestion, safety issues, emissions, and noise impacts. These impacts not only affect lifestyle, but also economy and the quality of the environment especially on key commercial corridors such as the I-4. Segregation of truck traffic from passenger traffic has the potential to mitigate traffic operations by enhancing speeds, thus improving air quality as well as improving safety. Rakha et al. (2005) evaluated alternative truck management strategies along I-81, the results demonstrated that separation of heavy-duty trucks from the regular traffic physically offers the maximum benefits. Also, restricting trucks from the leftmost lane offers the second-highest benefits in terms of efficiency, energy, and environmental impacts. Samba et al. (2011) also evaluated large-truck transportation alternatives with safety, mobility, energy, and emissions analysis using TRANSIMS micro-simulation. They concluded that left-lane restrictions were the most statistically significant and beneficial treatment strategy; decreasing the likelihood of a rear-end crash by about 2% for the off-peak and peak hours. Supplementary analysis with the emissions model indicated that left-lane restrictions also had marginally positive effects on fuel consumption and emissions. 147

161 Since the FDOT is planning on prohibiting heavy duty trucks from using the I-4 corridor completely especially after the implementation of ML, the model application presented in this section is hypothetical. The main idea is to study the environmental impacts of RTL implementation on the I-4 corridor during the peak hour and evaluate the potential benefits of mitigation strategies in terms of vehicular emissions. Although, the truck percentage on I-4 did not exceed 5% during the peak hour, environmental benefits still need to be studied and evaluated in light of the literature findings. 8.4 Evaluation of Scenarios As mentioned earlier, the I-4 VISSIM model for the ML ultimate project obtained from the FDOT was run and the results of both the managed lanes (ML) and the general use lanes (GUL) were compared with the existing conditions (EX) for CO, NOx, CO 2 as well as PM10 and PM2.5 emissions. In addition, the restricted truck lane (RTL) hypothetical scenario on I-4 was also compared with the existing conditions and assuming an additional lane for trucks, thus having 0% trucks on the mainline while 5% of the total I-4 vehicular traffic is assumed as trucks on the truck only lane. The analysis was conducted for the peak hour (5-6) pm utilizing the OPMODE approach and the VISSIM/MOVES integration as explained in Chapter 7. Traffic composition in all scenarios included 60% passenger cars (PC), 37% passenger trucks (PT) and 3% heavy duty diesel trucks (HDDT) except in the RTL scenario, HDDT were assumed as 5% 148

162 (about 425 vph) as more truck traffic would be induced and therefore PT were decreased from 37% to 35%. The assumption was based on the FDOT AADT of 200,000 vpd and using the traffic characteristics of the I-4 corridor (k=8%, D=53%). It should be noted that MOVES project level data were the same as in Table 6-3 except for 3 inputs. The temperature was set at 85F for the month of June and calendar year 2011 since the I-4 traffic counts and calibrated data was for that date. Emissions analysis was conducted for total emissions as well as for each vehicle type as shown in the following figures and table. Furthermore, emission rates on the I-4 corridor were calculated in each scenario and compared with the base scenario (EX) to determine the effectiveness of the mitigation application. As shown in Table 8-1 and illustrated in Figure 8-1(a), (b), (c), overall total emissions were reduced in all scenarios and for all pollutants when compared with the EX scenario except for the CO emissions in the GUL scenario and slightly in the RTL scenario. 149

163 Table 8-1: Pollutant Emissions Comparison by Scenario Pollutant Scenario PC-Gas (kg) PT-Gas (kg) HDDT (kg) Total (kg) EX CO GUL ML RTL EX NOx GUL ML RTL EX 16,757 12,048 4,136 32,941 CO 2 GUL 11,085 8,932 3,354 23,371 ML 3,548 2,956 1,061 7,566 RTL 13,864 10, ,840 EX PM10 GUL ML RTL EX PM2.5 GUL ML RTL

164 (a) CO 2 (b) CO & NOx 151

165 (c) PM10 & PM2.5 Figure 8-1: Pollutant Emissions Comparisons by Scenario 152

166 When examining emissions in each scenario by vehicle type as shown in Figure 8-2(a) through (e), we can see that total emissions were reduced for all pollutants for all vehicle types in all scenarios except for the CO and NOx emissions especially in the GUL and RTL scenarios. This was attributed to two main reasons. First, the behavior of the pollutant is different with each vehicle type and second, more induced demand due to improved operations or sometimes known as latent demand that was restrained due to capacity restrictions. Recall that, CO results from the vehicle s incomplete combustion of fuels. Gasoline engines emit higher amounts of CO than diesel engines, due to their lower combustion temperature compared to diesel. Furthermore, CO increases at very high speeds which is also attributed to relatively cooler engine operations. On the other hand, NOx is totally opposite in performance to CO. NOx are mainly created during fuel combustion where a small amount of the nitrogen in the air along with nitrogen compounds from the vehicle fuels is oxidized at high temperatures. Diesel engines generally produce greater amounts of NO x than gasoline engines due to their higher combustion temperatures which can be observed in the amount of NOx emissions generated from only 3% of trucks (almost equal to the emissions released from 60% PC or 37% PT). Conversely, maximum benefits were observed in the ML scenario for gasoline vehicles and in the RTL scenario for diesel trucks. 153

167 (a) CO 2 (b) CO 154

168 (c) NOx (d) PM10 155