Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles, Volume 2: Guidance (2018)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

23 This chapter builds on the information presented in the preced- ing chapters and provides a conceptual framework for planning and modeling CAVs. Subsequent chapters provide high-level guidance for planning and modeling practice. Elements of a CAV Planning and Modeling Framework A framework for forecasting CAVs and the technologies and associ- ated changes in travel behavior that might occur includes five elements: • Data, • Planning context, • Modeling, • CAV adoption timeline, and • Communication of uncertainty. The first three elements—data, planning, and modeling—combine to create a forecasting environment. The typical planning process includes developing a vision for transportation in a region, setting goals and performance measures as targets, collecting data, building models from the data, and using models in either a predictive or an exploratory mode to evaluate alternative transportation investments. The planning context depends on the timeframe and the level of uncertainty that stakeholders and planners have about the future within a specific time in the future. More certainty leads to the application of predictive models to analyze facility alterna- tives (e.g., capacity, alignment, and mode), while greater uncertainty leads to the application of scenario-based planning or other methods for addressing deep uncertainty. The basic tenet of the CAV planning and modeling framework is the uncertainty that is fundamental to forecasting. Data, by definition, always describe a past condition; they relate information from the time the data were collected. Models that are calibrated to data describe the relationship between independent variables, such as households and employment, and the impact those independent variables have on dependent variables, such as total VMT. The models describe the conditions depicted by the data. A forecast application of data-calibrated models changes the value of the independent variables by adding growth but keeps the relationship to the outcome—the dependent variables—the same. For instance, a trip rate is calibrated to observed trip-making frequency based on a house- hold travel survey. A forecast application of the rate of travel may remain constant, even while the overall population grows, and the outcome of the application of the fixed trip rate is a C H A P T E R 4 Framework for Planning and Modeling CAVs Chapter Highlights • Provides a conceptual framework for planning and modeling CAVs. • Discusses the individual elements of the framework—data, model- ing, planning context, timeline for CAV adoption, and communicating uncertainty.

24 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles greater number of trips, even though the rate of trip making stays the same. Data-supported modeling does not forecast behavioral changes; instead, the process forecasts growth in presently observed behavior— as it should. The further out in time a forecast application is made, the less cer- tainty an analyst has about the accuracy of the modeling outcome based on the relevancy of the data used to calibrate the model. Data relevancy is the inherent validity of the data as time passes. In a rap- idly changing transportation environment, models calibrated to data become less useful for longer-range planning. Conversely, in a stable and unchanging environment, data and models built upon the data retain their relevancy for a longer period. The relevancy of data and models is a subjective determination. No fixed guidelines exist to determine the amount of time that data remain relevant and can be used for predictions to sup- port transportation plans. Decisions regarding data relevancy and the use of predictive modeling versus exploratory modeling need to be made at the outset of a planning process. As data and models become less predictive, the planning process used needs to change. Alternative analysis (choosing one outcome as the best) can be done by using predictive models, while exploratory modeling can support scenario-based planning, which keeps several future scenarios in play when making decisions. The fourth element of the framework for forecasting CAVs concerns the timeline within the planning horizon and the level of advancement and adoption of automated transportation technologies. The level of advancement and adoption can be built into the definition of scenarios in long-range analysis or viewed as a static prediction for predictive analysis of alternatives. The last element, communication, involves the need for the analyst to convey the level of uncertainty associated with the model results to decision makers and stakeholders. A situation that analysts commonly face in using this framework is a de facto interpretation of results as predictive. A better method is to determine at initial project scoping whether the analysis is going to be based on predictive, data-supported modeling or on exploratory techniques, which should not be taken as a prediction. Framework for CAV Planning and Modeling Figure 3 depicts the framework for a system of planning and modeling for CAVs. The frame- work displays a planning and modeling timeline at the top showing that data are collected in the past and planning occurs in the present. The CAV adoption timeline arrow indicates that an agreement needs to be made on the level of adoption/advancement of CAV technology—and the rate of adoption of the technology—over the planning/modeling timeline period. No assumed time period is indicated in the figure. The total time period could be 20 years or 50 years, depend- ing on the application. The assumption in the diagram is that all modeling is done in the present time. As time passes, the relevancy of data becomes less. Therefore, predictive modeling is more valid for the shorter term, while exploratory modeling is useful for long-term planning. However, the diagram does not indicate the time when modeling should be done within the planning timeframe. The intent of the diagram is to show that predictive modeling can be used when more confidence in the relevance of the data occurs, while exploratory modeling should be used if greater uncertainty exists about the relevance of data and the models that are calibrated to the data. Data-supported modeling does not forecast behavioral changes; instead, the process forecasts growth in presently observed behavior— as it should. Decisions regarding data relevancy and the use of predictive modeling versus exploratory modeling need to be made at the outset of a planning process.

Framework for Planning and Modeling CAVs 25 Framework: Data Data are not available to build predictive long-range travel models of CAVs, or any of the technologies and applications that will be pro- mulgated in the future. However, deployment of preliminary versions and types of CAV technology is expected to increase within just a few years. As data that include AVs, CVs, and related applications become available, predictive models may be constructed. A move is occurring in many cities to develop smart city technol- ogy. A smart city is predicated on generating and sharing data with the desire to develop greater efficiencies with physical infrastructure, citizens, multiple levels of government, and business crossing all economic sectors including transportation. Municipal leadership hopes that opening data access will stimulate creativity that will result in more efficient and equitable cities with a reduced impact on the environment and a better quality of life for regional residents and visitors. Two main types of data will become important as CAV technology gains a foothold: archival data and real-time data. Archival data can be used to build informative descriptions and models about the existing transportation choices people are making and the trends in those choices. Real-time data describing immediate changes in system performance, vehicle tracking, route planning, infrastructure status (such as traffic signal timing), speed, direction, occupancy, and other statistics are being used to enhance the efficiency of transportation system operations. Because data to build forecast models of CAVs will be scarce in the short term, models of longer-range futures must be used in exploratory mode by imposing reasonable changes to the parameters that describe travel behavior and choices based on the judgment of analysts, leader- ship, and stakeholders. Rather than rely solely on the judgment of analysts, however, the planning process must also include a method to define which changes are reasonable and which are not acceptable to the planning community. A scenario-based planning process can accomplish this. Note that the quality of the data, such as avoiding sampling bias and the underreporting of trips, remains important. A judgment about the relevancy of data must be made, based not solely on the age of the data but also on the applicability of the data to the questions being asked. For example, data collected currently (in 2018) from TNCs may inform, but may not reflect, travel behavior changes of later, more intense market penetration of shared CAVs and an inter- connected trip-planning environment. Data are not available to build predictive long-range travel models of CAVs, but as data become available, predictive models may be constructed. Figure 3. Framework for CAV planning and modeling.

26 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles Similarly, stated preference data should be caveated, particularly in the case of CAV technologies, because respondents probably do not have any experience with the technology. Models calibrated to data collected in today’s transportation environment cannot simply be applied with an additional mode representing the CAV technology because the chosen mode needs to be present in the environment at the time the data are collected for the individual choice to be valid in the model. There is a difference between making a choice on the basis of a lived experience versus the imagined qualities of an experience. Data quantity and quality will develop alongside smart mobility advances. Most models are built from data that are collected by sur- vey sampling. This method is useful to modelers, because individual characteristics, such as income level, can be used as a stratification of the independent variable (person or household). Passively collected movement data from connected devices such as smartphones are gaining in popularity because the data are broad—meaning the data are collected across a wide spectrum of the sample universe—but individual characteristics are suppressed to protect privacy. Survey sampling is expensive but valuable because it gathers data that are “deep” in terms of joining respondent characteristics with travel behavior. In the future, as automation and digitalization become commonplace, data that are both broad and deep will likely become available, making the data much more useful for calibration of models. Planners and modelers need both broad data—or big data—to capture small but impactful nuances and deep data—or survey data— to describe motivational factors (independent variables) that can be used to predict behavior. However, the basic tenet of the planning and modeling framework—that data relevancy passes with time—still applies, regardless of the span or depth of the data collection methods. Framework: Planning Context The purpose of modeling systems and other tools and processes used for travel forecasting is to inform analysts, agency leadership, stakeholders, and the public about potential outcomes of planning decisions. The context of planning for CAV technology is one of deep uncertainty about the impacts, infrastructure needs, deployment timeline, market penetration, design, engi- neering, and many more aspects of automation. CAVs are expected to be disruptive and impact- ful, but that is the only certainty about which the planning community generally agrees. The framework for CAV modeling and planning suggests a simple idea within the context of deep uncertainty: the longer the planning horizon, the less certainty about predictive processes. The basic method for planning under deep uncertainty across many fields of study, including transportation, has been scenario-based planning. Scenario planning can be applied to transpor- tation as a step incorporated in performance-based planning at various stages throughout the process (Twaddell et al. 2016). Also, many MPOs, DOTs, and other planning agencies have con- ducted visioning processes for the specific purpose of evaluating a larger set of alternative futures than those that are typically studied in federally required metropolitan transportation plans. Scenario planning arose from the business problem of products becoming obsolete as mar- kets and technology changed over time. Businesses needed a process to protect their business lines against disruptive changes in the future. They realized that prediction of only one future, or selecting one alternative from a set of futures and pursuing only that one, was causing a sort of blindness to change. Being prepared for change was the key to being able to adapt quickly to changes in the markets. To prepare for change, a set of plausible scenarios had to be developed as a part of the planning process, which enabled adaptability and lowered the risk associated with changes in the marketplace. Stated preference data should be caveated, particularly in the case of CAV technologies, since respondents probably do not have any experience with the technology.

Framework for Planning and Modeling CAVs 27 Similarly, the transportation planning community is facing the problem of adaptability and risk associated with an uncertain future, one based on advanced technology that is not currently well defined. While the one-best-alternative method that has been established over decades in transportation planning may suffice for short-term plans, it is becoming abundantly clear that long-range plans are facing increas- ing risk of being invalid because of CAVs and other disruptions to the transportation space. The Federal Highway Administration (FHWA) recently published a scenario-planning guidebook for practitioners titled Next Generation Scenario Planning: A Transportation Practitioner’s Guide (Ange et al. 2017), which describes the process of scenario planning for transportation. FHWA is also pending production of report guidance on scenario planning for CAVs. Framework: Modeling Modeling in this framework is divided into two parts: predictive modeling and exploratory modeling. Predictive modeling is used to describe models that are calibrated with historical data, validated by system performance (traffic counts) for a past year, and tested for short-range forecasting accuracy. Predictive modeling is the way travel forecasting has been done tradition- ally because past data have usually been good for predicting future outcomes, even in the longer term, under a stable transportation environment. In fact, the ability of travel demand models to forecast in the short term, and in most cases simply validated by the parameter calibration year without forecasting at all, has been commonly held as the indicator of the accuracy of the model for forecasting. This is patently a false assumption. Exploratory modeling is the use of models for testing various scenarios that do not match the trends seen in the historical data archive. Models calibrated to recently measured travel behavior in surveys or from passively collected data are used in predictive mode. When analysts change the calibrated parameters to reflect a change in behavior—as is expected with CAV impacts— they are using exploratory models. However, the changes in parameters are not done for the purposes of testing the sensitivity of the calibrated model to changes. Sensitivity testing is done by changing the independent input variables, such as forecast households. Analysts often cre- ate ranges of values for input parameters. For CAVs, the ranges would need to be plausible. For instance, it would not be plausible to increase trip generation per household to a rate, which would occupy all household members all day in travel activity. Analysts may use professional and rational judgment to set ranges for exploratory modeling. Several types of modeling processes exist for conducting exploratory modeling and plan- ning. Scenario-based planning may require simply adjusting models to match the assumptions generated as part of a workshop process. These scenarios may differ widely in model inputs, such as demographic forecasts, but they may also assume varying levels of technology adoption. Assumption-based planning, quantitative risk analysis, and exploratory modeling and analysis/ robust decision making (RDM) are other techniques that may prove useful to CAV modeling efforts. These techniques are discussed in detail in Chapter 5 of this report. Exploratory modeling can be done with process models (such as trip-based and AB mod- els), or it can be done with various strategic models that use a wide range of plausible inputs, distributions, elasticities, and outputs. While process models use a finer level of detail, the time required to process details may limit the usefulness of these models for exploratory analysis. More generalized strategic models may be more useful because of the ability to run many itera- tions and create distributions of outputs that are helpful in analyzing scenarios in the several It is becoming abundantly clear that long-range plans are facing increasing risk of being invalid because of CAVs and other disruptions to the trans- portation space.

28 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles scenario-planning processes. At an appropriate point in time, planners and modelers in a region should agree on whether models can be taken as predictive tools that extend trends found in observed data or should be discussed as exploratory tools that are not based on observed data. The framework provides three types of modeling systems that can be applied in an exploratory modeling context: • Trip-based models developed as aggregate models of population and employment in a region with disaggregate measures of transportation supply and an aggregate assignment process, • Activity-based and dynamic traffic assignment models developed as disaggregate models of persons and firms in a region with disaggregate measures of transportation supply, and • Strategic models developed as disaggregate models of persons and firms in a region with aggregate measures of transportation supply. Strategic models are intended to be applied in a scenario planning context to evaluate the impacts of a variety of policies and investments. Often, these are used as a screening analysis, where hundreds of combinations of different policies can be tested and prioritized. Trip-based and AB or DTA models are applied for a more limited set of scenarios to explore the more detailed impacts of policies and investments on the transportation system. Framework: Adoption Timeline The timeline for adoption of CAV technology is debatable and is complicated by the defini- tion and functionality of the technology. However, in this planning and modeling framework, it is important to include a description of potential phases of deployment where specific modeling and planning tools may be warranted. For example, early adoption may be with shared auto- mated fleet services operating in a limited range or geography, and may involve relatively little private ownership of highly automated AVs. Over time, as the transportation fleet transitions to high automation and reaches a tipping point, more data about behavior will become available. The overall shift is expected to be toward exploratory modeling and planning in the early years of deployment, and then back to predictive modeling and planning as the fleet becomes saturated with AVs and behavioral outcomes can be measured. Predicting CAV adoption involves many uncertainties. First, technology develops in con- trolled laboratory conditions but needs extensive real-world testing to bring it to fruition. Many individual technologies need to converge and be tested for driverless cars, and there have already been setbacks from crashes. Computing and communications systems that CAVs depend on are advancing in parallel. CAV system adoption is also dependent on public acceptance and desirability, most notably operating cost, which has not yet been determined in an open market. Public policy and governance may eventually adapt and control the technology to address soci- etal health, safety, and welfare issues, including equitable access to services. The framework for CAV modeling and planning suggests that the planning process include a thoroughly developed timeline. Consent among stakeholders about the timing of deployment will ease the process of scenario development. Framework: Communicating Uncertainty Framing the conversation about uncertainty is part of the CAV planning and modeling framework because in transportation planning, decision makers typically look to transporta- tion planners to provide robust predictions. With so much uncertainty surrounding CAV and its potential impacts, planners need to be knowledgeable about how to communicate uncer- tainty without introducing doubt and a lack of confidence in forecasts. In this framework, it is

Framework for Planning and Modeling CAVs 29 critical to communicate with leadership about uncertainty but without under- or overstating outcomes. Personality preferences vary among planners, modelers, and leadership. Analysts present- ing information about CAVs to leadership and to other planners and modelers will need to understand various perspectives and preferences when collecting information describing deep uncertainty. Most leaders feel it is their responsibility to act, but with the inherent risk that automated technology poses about the long-term future, it may be a time for contemplation, not action. Planners need to learn to talk confidently without misleading the audience about the level of certainty. Plans and models are communication tools. Understanding how to communicate under uncertainty is fundamental to a conceptual framework for planning and modeling CAVs. The processes and tools are used to inform leadership about the usage and character of transpor- tation facilities and the development of and impacts to the urban and rural transportation environment. Today, CAVs loom large as deeply uncertain and potentially transformative transportation technologies. Deep uncertainty exists when parties to a decision do not know, or do not agree upon, the system model that relates action to consequences, the probability distributions to place over the inputs to these models, or the relative importance of different consequences (Lempert et al. 2003). The ultimate design of the technologies, the timing and pace of their adoption, and their impacts on transportation goals are unknown but need to be included in forward-looking planning efforts, since they will become fully mature within most planning horizons.