Integrated Transportation and Land Use Models (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.

6 The transportation system is closely linked to the locations of households and firms. Every- thing else being equal, more accessible neighborhoods are more desirable than less accessible neighborhoods. Location choice and transportation are interdependent. For example, if more households and/or firms decide to locate in the suburbs, the origins and destinations of trips change and levels of congestion are affected. That land use and transport are interdependent is not new, but many transportation models ignore this link to land use models. Several agencies, having implemented integrated land use/transport models, can fully model the effects of transportation on location choice and the effects of location choice on the trans- portation system. These models tend to present behavior more likely to align with reality. A new highway, built to alleviate congestion, might attract other households and firms to relocate near this new highway. This might then result in longer trips, which might use up the added capacity of this new highway. Integrated land use/transport models can model such possibilities to help decisionmakers make more informed assessments of infrastructure investments. In addition, integrated land use/transport models enable users to model a whole range of additional scenarios (such as zoning policies, affordable housing policies, user profiles of residents in transit-oriented developments, and the effect of traffic noise on location choice). Failure to represent this land use/transport feedback cycle has varying effects, depending on policies, congestion levels, and remaining land use capacities in the area. For example, a region without an integrated model might miss the effect that a new highway might encourage more development along the highway corridor, which might lead to additional traffic that might not have been accounted for during the highway planning process. Although the relevance of integrating land use with transportation analyses has been described for decades (Forrester 1969), modeling land use and transport in an integrated fashion is far less common than traditional transport modeling. Forrester described how the location of population and employment affect travel demand and how travel times on the transport network influ- ence location decisions of both households and firms. Hansen (1959) showed empirically that New York City neighborhoods with subway access developed much faster than other neighbor- hoods. Wegener (2004) proposed that the land use/transport feedback cycle should be accounted for in transport modeling. Despite the theoretical work so far, practical applications are not common—many integrated land use/transport models remain in a research exploration stage and some have been abandoned after a few years due to complexity or non-operationality. There are, however, several useful implementations of integrated land use/transport models in the United States and Canada. Some agencies have decided to expand their transport model with an integrated land use model so as to fully represent the land use/transport feedback cycle. Such models have been implemented successfully both at the level of Metropolitan Planning Organizations (MPOs) and at the state C H A P T E R 1 Introduction

Introduction 7 level. This report discusses how these models have performed and how an integrated land use/ transport model has affected decision making in the region. This report focuses on model implementations applied in the United States and Canada and how integrated land use/transport models have affected decision making. Scientific model devel- opment is assumed to be a necessary precursor for future model applications in practice and is not further explored herein. There is a renewed interest in overcoming implementation issues with integrated land use/ transport models, because new trends, such as telework or driverless vehicles, are expected to affect land use patterns substantially. Modeling the effect of such scenarios simply cannot afford to leave the land use forecast static and unaffected by changes in the transportation system. Simi- larly, pricing studies will benefit from considering the effects on location choice. The relatively high toll to cross the Chesapeake Bay Bridge near Annapolis, MD, encourages people working in the Washington, DC, metro area to live west of the Chesapeake Bay, thus reducing the commute shed eastward much more than in other directions. Even though pricing predominately aims at affecting transport behavior, accounting for land use choices in such studies is likely to generate substantially more realistic model sensitivities. Other agencies have used integrated models for enhanced equity analyses. A new subway line, for example, could increase land prices near the subway stations and price out current residents. In this case, a traditional transport model would show that current residents benefit from the subway, while an integrated land use/transport model might show that current residents will not be able to afford living there anymore. In land use models, land use is only one of the variables modeled. Land use describes the type of usage of a parcel, such as residential, industrial, commercial, or recreational. This is distinguished from land cover, which may include forests, wetlands, impervious surfaces, agriculture, and others. Urban Ecology models tend to work with land cover definitions (Alberti 1999), while integrated land use/transport models tend to define land uses instead. In addition to land use, such models include the distribution of population and employment. Sometimes, the distribu- tion of dwellings and non-residential floor space is represented explicitly as well. This report focuses on land use models integrated with transport models. The report has eight chapters and appendix materials. Chapter 1 provides a brief introduction to the history of land use/transport modeling, an overview of the three most relevant model types for integrated land use/transport modeling, and a description of the methods used to create this report. Concepts of integrating land use with transport models are discussed in Chapter 2. Chapter 3 describes a screening survey conducted to identify the most relevant implementations of integrated land use/transport models. Chapters 4, 5, and 6 each describe one of the three pre- dominant land use model types, namely sketch planning, microsimulation discrete choice, and spatial input-output models. In-depth interviews were conducted for each model type, and these three chapters summarize knowledge gleaned from these surveys. After explaining the gist of each case study implementation, particular attention is given to the effort to develop the model, the effect on decision making in the region, and lessons learned by each agency. Chapter 7 synthesizes findings of this research, provides an evaluation matrix to help agencies select the right model type for their particular needs, and identifies research needs to further strengthen integrated land use/ transport modeling. Chapter 8 presents the conclusions of the research team. 1.1 Short History of Integrated Land Use/Transport Modeling Although this report cannot provide a thorough review of the literature on integrated land use/transport modeling, a brief introduction to the history is provided to help the reader to understand the state of the art of this kind of modeling.

8 Integrated Transportation and Land Use Models The interest in land use modeling comes in waves (see Figure 1-1). After peaks of interest in the 1960s (Echenique, Forrester, Lowry), 1980s (Anas, Putman, Wegener) and 2000s (Hunt, Miller, Waddell), practitioners’ interest in integrated land use/transport modeling is on the rise again. New legislation being adopted in California (Senate Bill 375 in 2008) and Oregon (Senate Bill 1059 in 2010) has created momentum for improved land use/transport integration. Many transportation agencies have recognized that they cannot sufficiently test policies under congested conditions or in tight housing markets with transportation models alone. New trans- portation infrastructure changes accessibilities, which may change preferred housing locations, thereby indirectly changing origins and destinations of travel. A well-intended new highway project may fill up quickly because of new subdivisions being built near that highway. In 1973, Lee’s Requiem for Large-Scale Models was published. This document criticized models for committing seven “sins” (Lee, Jr., 1973). His conclusions affected researchers in the field of integrated land use/transport modeling and led to reduced activity in this field. The academic retreat in the 1990s might be attributed to the debate about whether or not the free market generates the most favorable land use patterns. At least in part, concerns about the environ- mental impact of growing urban sprawl led to a renewed interest in integrated land use/transport modeling. Since the publication of TCRP Report 48: Integrated Urban Models for Simulation of Transit and Land Use Policies: Guidelines for Implementation and Use (Miller, Kriger, and Hunt 1998), land use modeling has developed in two opposite directions. On the one hand, so-called sketch planning land use models (such as CommunityViz, I-PLACE3S, or UPlan) are considered easy to implement and provide results under different scenarios almost instantaneously. However, they lack the behavioral theories that lead to land use decisions, thereby limiting the ability to realistically represent land use changes for some scenarios. On the other hand are complex discrete choice and spatial input-output models (such as PECAS, TRANUS, or UrbanSim) that incorporate behavioral, economic, and social science theory. Although those models are power- ful enough to model a wide range of scenarios, users have reported implementation challenges and long model runtimes. In addition, several hybrid products (such as DELTA, MEPLAN, or Silo) try to balance theoretical perfection and shorter runtimes. For State DOTs and MPOs, it is inherently difficult to evaluate these options against one another and select the right approach for their particular needs. Lowry’s Model of Metropolis (1964) is often considered to be the first computer model that truly integrated land use and transportation. The Lowry Model assumed the location of Figure 1-1. Waves of attention to integrated land use/transport modeling over time.

Introduction 9 basic employment exogenously and generated an equilibrium for the allocation of non-basic employment and population. Over the last five decades, this spatial interaction model has been implemented many times (such as Batty 1976, Mishra et al. 2011, Wang 1998). At least equally influential was Forrester’s theory of urban interactions (1969). Even though it was an aspatial model, his research on interactions between population, employment, and housing influenced the design of many spatial land use models developed since. This synthesis report focuses on land use models and their integration with transport models. Putman developed the gravity-based Integrated Transportation and Land Use model Package (ITLUP) (Putman 1983), which led to the frequently applied DRAM and EMPAL models. Wilson’s Entropy Model (1967) generated an equilibrium by maximizing entropy of trips, goods flows or the distribution of population. The MEPLAN model developed by Echenique, Crowther, and Lindsay (1969) is an aggregated land use transport model that used the basic concept of the Lowry Model as a starting point. The model can simulate various land use and transport scenarios. Another modeling approach using the Lowry Model as a starting point is the TRANUS model (de la Barra and Rickaby 1982)— this simulates land use, transport, and interactions at the urban and regional scale. Martínez (1996) developed a land use model (called MUSSA) in which location choice is modeled as a static equilibrium. MUSSA used the bid-auction approach based on the bid-rent theory where consumers try to achieve prices as low as possible and not higher than their willing- ness to pay. In the bid-rent theory, first introduced by Alonso (1964: 36 ff.), land prices are the immediate result of the bid-auction process. In contrast, the discrete choice approach—initially developed for housing choice by McFadden (1978: 76 ff.)—models land being bought or rented with no instant effect on the price. Wegener (1982) developed the IRPUD model as a fully integrated land use transport model. The household location choice is microscopic, simulating every household individually. The IRPUD model was one of the few early approaches that contradicted the common assumption that land use models should reach an equilibrium at the end of each simulation period. Land use development aims at equilibrium constantly, but given a continuously changing environment and slow reaction times of households, businesses, developers, and planners, this equilibrium stage is never reached. The price of a new dwelling and the commute distance to the household’s main workplace are accounted for as true constraints in location choice. Similarly, the MetroScope model for Portland, OR (Conder and Lawton 2002), compares expenditures for housing, trans- portation, food, health, and all other expenses to ensure that household budgets are not exceeded. PECAS (Hunt and Abraham 2003) is a market-based land use model that represents an equilibrium of competing demand for developable land as one input to household and business location choice. Households relocate based on available floorspace, prices, accessibilities and other location factors. PECAS combines this bid-rent approach in a spatial input-output model with a microscopic land development model. DELTA (Simmonds 1999) combines a spatial-economic model with households and job location model and a long-distance migration model. Orcutt et al. (1961: 45 ff.) proposed to simulate individuals (referred to as microsimulation), rather than modeling population at the aggregate. A few influential microscopic land use models have been developed, including the ILUTE (Miller and Salvini 2001), UrbanSim (Waddell 2002), ALBATROSS (Arentze and Timmermans 2000), PUMA (Ettema et al. 2004) and SimDELTA (Simmonds and Feldman 2007) and LUSDR (Gregor 2006). All these models use an aggregate transport model. An exception is the ILUMASS model (Wagner and Wegener 2007). However, the microscopic integration between land use and transport never became operational. Probably, the first true integration of microsimulation land use and transport models was developed by Waddell et al. (2010) for San Francisco. Kii et al. (2016) identified new challenges for integrated

10 Integrated Transportation and Land Use Models land use/transport models—climate change mitigation, energy scarcity, and social conflicts as well as new technologies, such as autonomous vehicles or shared mobility services. Good overviews of operational land use/transport models are given by Acheampong and Silva (2015); Hunt, Kriger, and Miller (2005); and Wegener (2004 and 2014). 1.2 Model Types Land use models can be classified as rule-based models, behavioral models, and hybrids thereof. Rule-based land use model are commonly called sketch planning models (Figure 1-2). Sketch planning models are excellent tools for long-term visioning, because they tend to run faster and allow users to assess the development capacity of considered land use scenarios. Such models do not attempt to model human behavior but rather develop rules of development inter- actions. Behavioral models, on the other hand, try to represent human behavior, based on behav- ioral theory, at the expense of (commonly) longer runtimes and larger data requirements. Such advanced models can be further categorized into two types: microsimulation discrete choice models and spatial input-output models. Both types use advanced theory to explain human behavior, although the approaches are rather different in model implementation. The former reflects utility-based or behavioral theory, while the latter is derived from economic theories. Hence, three major land use model types are presented in this report; a separate chapter is dedicated to each of them as follows: • Sketch planning land use models, such as CommunityViz, Envision Tomorrow, I-PLACE3S or UPlan are presented in Chapter 4. • Microsimulation discrete choice models, such as MetroScope, Silo, or UrbanSim are discussed in Chapter 5. • Spatial input-output models, such as CUBE Land, PECAS or TRANUS, are presented in Chapter 6. The entire report is organized by these three modeling paradigms. Chapters 4, 5, and 6 provide more detailed explanations on how these models work and how they differ from one another. This classification into three types is somewhat arbitrary—many hybrid models use elements of all three model types. Nevertheless, it is useful to classify models by their main theoretical frame- work when describing modeling applications. By emphasizing differences between models, it will be easier to identify the strengths and weaknesses of various modeling approaches. For all three model types, it is common to exogenously define larger events expected to change land use in the future. For example, it may be known that a new shopping mall will be opened at a given location within the next 10 years. Land use models cannot predict the exact location of this mall. Given that the location and opening year are already known, the location and size of this mall can be provided as an override to the model. The same is true for housing Sketch Planning Models Rule-Based Models Behavioral Models Microsimulation Discrete Choice Models Spatial Input- Output Models Figure 1-2. Land use model type classification.

Introduction 11 projects, such as the “known” future conversion of a warehouse into condominiums or the “known” development of transit-oriented development next to a subway line extension. The use of overrides is generally assumed to improve the model results, because expected developments are reflected in every scenario. Cellular Automata (CA) models are not part of this report. Such models rasterize the research area into grid cells of equal size (often 100 by 100 meters) and calculate the probability that one cell changes from one land cover type to another. Although these models are powerful at representing how urban development may grow organically around infrastructure provided, such models are rarely used to enhance transport models. Most CA models do not consider travel times which could be provided by a transport model, nor do they provide input data for transport models. Excellent examples of CA models have been developed in the United States and Canada [such as the model LEAM published by Deal and Sun (2006) or the Chesapeake Bay Land Change Model published by Shahumyan et al. (2016)], but none of those was designed to be integrated with transportation models. A good overview of the state of the art of CA models can be found in Berling-Wolff and Wu (2004) and the Committee on Needs and Research Require- ments for Land Change Modeling et al. (2014). 1.3 Report Method In addition to the literature, two sources of information were used to develop this report. First, an online screening survey was conducted. This short survey was distributed widely to screen for possible land use/transport model applications. The survey was sent to all U.S. MPOs and State DOTs as well as Provincial Ministries of Transport in Canada. The screening survey is described in more detail in Chapter 3, and the survey form is provided in Appendix A. In addition, all model developers of major land use models applied in the United States and Canada (namely CommunityViz, CUBE Land, I-PLACE3S, MEPLAN, PECAS, TRANUS, UrbanSim, and What If?) were contacted and asked to name the most successful implementations of their respective model. The results of the screening survey, suggestions of model developers, and prior knowledge of the author helped identify agencies to be contacted for in-depth interviews. Instead of standardized surveys with multiple choice questions, in-depth structured inter- views were conducted with agencies in the United States that have used integrated land use/ transport models as a means to collect complex and qualitative information. Although the number of agencies interviewed in this was limited, this interview format provided benefits deemed particularly relevant for this topic: • Slightly varying questions could be asked for each particular model implementation. For example, mathematical models, such as PECAS or UrbanSim, allow for detailed represen- tation of land use/transport feedback loops. In rule-based models, such as CommunityViz or I-PLACE3S, this integration necessarily is simpler. Hence, agencies using a mathematical model were asked detailed questions on model integration, while this topic was not a central issue for surveys with agencies that use a rule-based model. • In-depth interviews allow for clarification of questions. Prior experience suggested that describ- ing complex modeling structures requires the option to ask about specific interpretations of questions. For example, rule-based models often define the term land use as land cover, while mathematical models tend to define land use as allocation of population and employment. The in-person interviews clarified which land use definition was used by the respondent. • In-depth interviews allow adjusting questions subject to answers given in the interview. For example, the Sacramento Area Council of Governments (SACOG) was interviewed to under- stand their use of I-PLACE3S. SACOG has used this model extensively and provided valuable

12 Integrated Transportation and Land Use Models feedback on this particular model. However, SACOG is transitioning to the model Envision Tomorrow, and the in-person interview allowed for asking questions about the reasons for changing. In-depth interviews were conducted with seven agencies that use integrated land use/transport models (see Table 1-1). Figure 1-3 shows the geographic distribution of surveyed agencies. These agencies are scattered all over the United States, with no model types being more representative in some parts of the United States than others. Although the seven interviews cannot be fully rep- resentative of the whole field of integrated land use/transport modeling in the United States and Canada, the agencies were selected based on geography and theme so as to broaden the informa- tion base from which conclusions were drawn. The results of the in-depth interviews are presented in Chapter 4 for sketch planning models, in Chapter 5 for microsimulation discrete choice models, and in Chapter 6 for models that use the spatial input-output approach. A list of guiding questions for these in-depth interviews is provided in Appendix A. The in-depth interviews were complemented by a review of the agency’s model documentations and user guides where available. Agency Model in use Interviewed ARC (Atlanta) PECAS 10 Jan. 2017 ODOT (Ohio) PECAS and SEAM/SLUM 11 Jan. 2017 WFRC (Salt Lake City) UrbanSim 21 Mar. 2017 MTC (San Francisco) UrbanSim 22 Mar. 2017 TJCOG (Raleigh) CommunityViz and Envision Tomorrow 5 Apr. 2017 SACOG I-PLACE3S 7 Apr. 2017 MetCouncil (Minneapolis) CUBE Land 26 Apr. 2017 Table 1-1. In-depth survey interview partners. Figure 1-3. Geographic distribution of agencies interviewed for the in-depth surveys.