National Academies Press: OpenBook

Integrated Transportation and Land Use Models (2018)

Chapter: Chapter 5 - Microsimulation Discrete Choice Land Use Models

« Previous: Chapter 4 - Sketch Planning Land Use Models
Page 35
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 35
Page 36
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 36
Page 37
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 37
Page 38
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 38
Page 39
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 39
Page 40
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 40
Page 41
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 41
Page 42
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 42
Page 43
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 43
Page 44
Suggested Citation:"Chapter 5 - Microsimulation Discrete Choice Land Use Models." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 44

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.

35 In contrast to sketch planning models, microsimulation discrete choice models are based on comprehensive behavioral theories of decision making represented in more or less complex systems of mathematical equations. Microsimulation discrete choice models tend to have much higher requirements for the model user to understand how each submodel is handling decision making in the model. At the same time, such models allow for a much larger range of scenarios to be analyzed. It is not uncommon to take advantage of both types of models, with sketch planning models used to test many big-picture scenarios and microsimulation discrete choice models used to explore the most promising scenarios in more detail. The distinction between microsimulation discrete choice models (this chapter) and spatial input-output models (Chapter 6) is somewhat arbitrary. The land use model described in this chapter (called UrbanSim) applies bid-rent theory that is commonly associated with spatial input-output models. A model described in Chapter 6 (called PECAS) also applies micro- simulation to simulate land development. Nevertheless, separating Chapters 5 and 6 seemed reasonable, given that these models differ in terms of the fundamental theory applied. 5.1 Model Concept Discrete choice models are based on random utility maximization. The concept is derived from the Law of Comparative Judgment (Thurstone 1927) that describes the choice between two alternatives. The choice is based on the comparison of the alternatives’ utilities. Domencich and McFadden (1975) used this random utility maximization concept to develop the logit model that forms the core structure of discrete choice land use models. The logit model is based on psychological theory and was very influential for the field of choice analysis, which led to Daniel McFadden receiving the Nobel Prize in 2000. The binomial logit model describes the decision between two choices, such as the decision to relocate or stay in the current dwelling. Figure 5-1 shows the shape of the probability distri- bution. The higher the utility (e.g., the higher the utility of moving to another dwelling), the higher the probability will be (here: the probability to choose to move). Probability does not grow linearly with increasing utility. Instead, the logit model uses the S-shaped curve shown in Figure 5-1. Although changes in utility (x-axis) have the most effect on the choice probability (y-axis) near the central utility, a richer diversity of choices is enabled by including small probabilities at the left “tail” of very low utility and large (but below 1) probabilities at the right “tail” of high utilities. The S-shaped curve assumes that only a few people will decide to move with low utili- ties, even though there might be a positive utility associated with a move. Transaction costs to move are so high that most households will decide to stay in their current dwelling even if there C H A P T E R 5 Microsimulation Discrete Choice Land Use Models

36 Integrated Transportation and Land Use Models is a small benefit in moving. Moving tends to be expensive, and relocating households need to reacquaint themselves with the new neighborhood, which is why households tend to avoid relocation if there is not a substantial increase in utility. The S-shaped curve in Figure 5-1 suggests that, with a larger degree of utility improvement, most households would decide to relocate. For example, if cheaper and larger apartments in the same neighborhood were intro- duced, many households would decide to relocate. However, the S-shaped curve never reaches 100%. Even with the most striking improvement in utility, a few households would hesitate to relocate. Reasons may be to save the money for the move, to avoid the personal upset of getting used to a new dwelling, or being emotionally attached to the current dwelling. The logit model is particularly suited to represent such decision making. The companion of the binomial logit model is the multinomial logit model, which models choice among more than two options. In discrete choice land use modeling, multinomial logit models are commonly used to select a dwelling among several vacant dwellings or to choose a city to move to. The change in housing prices commonly is modeled with regression analysis using hedonic price models (Waddell and Ulfarsson 2003). Logit models are also common prac- tice in transport modeling, in particular for mode choice and increasingly for destination choice. Some of the most relevant operational discrete choice land use models include DELTA (Simmonds 1999), ILUTE (Miller and Salvini, 2001), IRPUD (Wegener 1982), MetroScope (Conder and Lawton 2002), Silo (Moeckel 2017), and UrbanSim (Waddell 2000). In microsimulation models, every household (and sometimes every person) is modeled indi- vidually. By contrast, aggregate models treat entire groups of households at the same time. An aggregate model could, for example, take all single-person households with low income in one zone as a group and calculate the share of households (which may be a non-integer number of households) who move to another zone. In microsimulation models, in contrast, household relocation is modeled explicitly for every single household. One important benefit of micro- simulations is that data for households are stored as individual household records. This makes it rather easy to add another household attribute (such as availability of e-bikes or access to autonomous cars). Not all discrete choice models are built as microsimulations (of the list in the paragraph above, ILUTE, Silo, and UrbanSim are microsimulations). The applications described in this chapter, however, are microscopic. Due to the limitations of treating entire groups in household relocation, it has become rather uncommon in the United States and Canada to use aggregate discrete choice models for land use modeling. 0 1 P ro ba bi lit y Utility Figure 5-1. Shape of the probability distribution of the binomial logit model.

Microsimulation Discrete Choice Land Use Models 37 5.2 Interviews Two in-depth interviews were conducted to learn about experience with a microsimulation discrete choice model. Staff at the Wasatch Front Regional Council (WFRC), the MPO of the Greater Metropolitan Area of Salt Lake City, Utah, and the Metropolitan Transportation Com- mission (MTC), the MPO of the nine-county region of San Francisco, California, were inter- viewed. Both agencies use the UrbanSim model (Waddell 2000), which has been developed for more than 15 years by Dr. Paul Waddell, University of California of Berkeley. UrbanSim appears to be the only microsimulation discrete choice model actively used in more than one region in the United States and Canada. Other influential discrete choice land use models, such as MetroScope in Portland, Oregon (Conder and Lawton 2002), or Silo in the state of Maryland (Dawkins and Moeckel 2016), are used by single agencies only and, therefore, are less representative of the state of practice in dis- crete choice modeling than is UrbanSim. A few decades ago, the most common land use model was DRAM/EMPAL (Putman 1983 and 1991). Today, the MPO for Dallas/Fort Worth (NCTCOG) may be the only agency that still uses DRAM/EMPAL. Since the model developer retired, and thus, support for model updates and revisions ended, most agencies have moved to models that are still supported. Although one might regret that models disappear when their developer retires, this may also be seen as a chance to advance the state of the art for land use modeling. Integrated models have advanced over the past 50 years, and newer models have learned from predecessors that have eventually disappeared. Model developers may be disappointed that their models are phased out, but this evolution has benefited the state of the art. 5.3 Model History Both WFRC and MTC started implementing an UrbanSim (Version 2) model in the early 2010s. As different as the two study areas are in geography and population, the experience using UrbanSim has been comparable. Both have worked with the model developer, and both integrated UrbanSim with their respective transport models. Although WFRC calls their land use the Real Estimate Market Model (REMM), the software behind this model is UrbanSim, which is the name used in this report to describe WFRC’s land use model. WFRC WFRC, an early adopter, implemented UrbanSim in the early 2000s. This model was over- hauled with an implementation of UrbanSim2 in 2012. Four months were spent researching appropriate tools. Taking into account advice from other agencies using different land use models, UrbanSim was selected as the most promising tool for implementation (within avail- able resources) to provide model sensitivities required by WFRC. The data preparation for the model took 2 years of staff time. During this time, WFRC hired many interns to clean up data, survey streets, review aerial photographs, and knock on doors to verify data (such as number of employees, number of bedrooms, year built, elevation, and proximity to amenities). Similar to TJCOG, WFRC implemented an expert panel to advise on model development. The panel, which reviews model results and sensitivities, consists of real estate professionals and professors in land use modeling, demography, and urban planning. The panel found UrbanSim performed reasonably well and provided useful policy insights. WFRC decided to simplify the structure of UrbanSim and forego the full capabilities of the model. Staff members say that this might have been key to implementing the model. Keeping the

38 Integrated Transportation and Land Use Models model simple and not adding all the detailed features that the model offers initially resulted in some minor conflict with the model developer who was interested in providing the most power- ful model possible. At the same time, the agency’s demographers asked for sufficient detail to ensure that UrbanSim would predict demographic change at least as well as previous methods. Otherwise, WFRC attempted to keep the initial implementation simple. WFRC’s approach was to get a simplified model to run, look at results, and add enhancements as needed. For example, the initial model setup did not include redevelopment. Recently, this was identified as an important improvement and its implementation is underway. According to WFRC staff members, users want more functionality, but one needs to push back to keep the model simple initially. Staffers concluded that if the model needs to be operational in time for a deadline, some detail will have to be sacrificed initially. Today, WFRC has an operational UrbanSim model. Staff members report that model imple- mentation is an ongoing effort to continuously improve the model results and add model sensi- tivities. For the tasks to be completed right now, the model is sufficiently implemented. MTC In California, Senate Bill (SB) 375, The Sustainable Communities and Climate Protection Act of 2008, recommended transport/land use model updates, and the four largest MPOs in California were advised to move toward the state of the art in land use modeling. An advi- sory panel of four scientists recommended the PECAS model (compare Chapter 6) be used to comply with SB 375. Accordingly, MTC started implementing PECAS for the nine-county Bay Area in 2008. Only 1 year later, MTC abandoned PECAS and switched to UrbanSim for several reasons. First, the main consultant providing support for PECAS did not make progress with the MTC implementation at the pace needed by the MPO. In addition, PECAS was a complex model to calibrate. It was difficult to find staff adequately trained to work with input-output models, which form the core concept for PECAS. Discrete choice models, on the other hand, are estimated based on methods more commonly taught at universities, so it is easier to find staff trained to work on them. Given the training of most MTC staff members, it appeared more feasible to work with discrete choice models than with spatial input-output models. Last but not least, a key developer of UrbanSim moved to the Bay Area, and the proximity of this developer was deemed to be a crucial benefit for successful model implementation. Since 2015, a renewed effort to enhance the UrbanSim implementation in the Bay Area is underway. A new RTP and Environmental Impact Report (EIR) using UrbanSim throughout were approved in summer 2017. 5.4 Model Implementation and Application Effort WFRC WFRC has been fortunate that much model implementation work could be done in house. One employee works on this model full time, and two others spend part of their time working on this project. In addition, one person is concerned with coordinating and managing expectations within the agency, thus enabling the technical staff to entirely focus on model implementation. Two consultants were hired to support model implementation: one was under contract to get the software architecture in place, and the other helped with model specifications and estimations. The costs for staff time amounts to about US$ 150,000 to 200,000 per year. Consulting costs were US$ 200,000 for the first year, and US$ 50,000 to 80,000 per year in the following 5 years. In addition, US$ 60,000 were spent on university interns. No money had to be spent on additional

Microsimulation Discrete Choice Land Use Models 39 data. Much data were available online, and employment data were provided by the state. Because WFRC was responsible for data development, the agency is intimately aware of the strengths and weaknesses of the input data. In addition, workstations had to be purchased. To save on runtime, several workstations are used in parallel if multiple scenarios need to be run. Because the software architecture is not designed to be distributed across multiple machines, each work- station is used to run one scenario at a time. WFRC collaborates closely with the Mountainland Association of Governments (MAG), which covers counties south and east of Salt Lake City. WFRC and MAG use the same transport and land use models, and costs are shared by a ratio of 1 to 2. MAG has dedicated up to half a full-time equivalent employee on this project since inception. MTC At MTC, one staff member devotes about 90% of his working hours to implementing and updating UrbanSim. In addition to staff time and hardware funding, MTC has a budget of approximately US$ 150,000 per year for the land use model, most of which is used for consulting fees. After the consultancy that supports UrbanSim implementations was bought by a competi- tor, MTC hired an onsite consultant, who now spends 2 to 3 days per week at the MTC offices to work on their UrbanSim implementation. Given that the consultant had written much of the current UrbanSim code, his presence was comparable to the benefit of working with the original model developer. MTC considers this setup as a key aspect for its successful model implementation. MTC serves nine counties in the Bay Area, each of which has different parcel datasets in varying formats. This fact makes it labor intensive to reconcile input data for UrbanSim, which works at the parcel level in the MTC implementation. As an additional dataset, MTC has bought the COSTAR Real Estate database, which costs about US$ 15,000 per year. This data- base provides rents and building attributes for almost every commercial site and most rentable apartments. This database is used for estimating and calibrating UrbanSim for the Bay Area. The base year is 2010, and MTC models in 5-year increments to 2040. For 2015, some over- rides were used to replicate observed developments in the model. The runtime of the land use model is 2 h and 30 min for all seven time periods, making it feasible to run several scenarios per day, if only land use changes are of interest. The travel demand model, in contrast, takes 30 hours to run for one model year. For this reason, the model integration with the transport model is completely manual at this point. In fact, MTC does not run the transport model very often, but rather uses base year logsums during the development of the land use scenarios. Only future year logsums are used for the final production model run used for the RTP and EIR. 5.5 Land Use/Transport Model Integration WFRC The WFRC model is fully integrated with a four-step transport model. Every 5 to 10 years, the four-step transport model is run. Logsum travel times, transit travel times, and automobile travel times are fed from the transport model to UrbanSim to calculate accessibilities. UrbanSim operates at the parcel level, and land use results are aggregated to TAZ and fed into the transport model. An air quality model is linked to the transport model. Land use forecasts are constrained by county control totals developed in collaboration with the state. Although county control totals were respected by UrbanSim for the LRTP, it is pos- sible to aggregate county-level forecasts to regional forecasts and turn off the restriction at the county level.

40 Integrated Transportation and Land Use Models MTC MTC has implemented two-way feedback between UrbanSim and the MTC travel model (refer to Figure 2-1). The data transfer is manual at this point, and draft land use model runs are developed using base year logsums. MTC’s UrbanSim uses both mode choice logsums and more specific measures (such as distance to closest transit station) to represent accessibility. These variables change every 10 years or so during the model forecast. During early runs, while land use policies are still being tested, the transport variables are from a base year travel model run, and they are not modified to capture changes resulting from the different land use patterns and transport investments. This is because most scenarios assessed have fairly similar accessibility because investments and land use differ modestly in relation to the existing situation. Logsums largely aggregate changes in travel time of individual zone-to-zone pairs. Even if a travel time between two neighborhoods improves significantly, these logsums do not show as much change because they aggregate travel times by all modes to all potential destinations. For scenarios that implement marginal changes, this appears to be an appropriate simplification. For more radical scenarios that affect travel time throughout the study area, using base year logsums might be an issue. Therefore, the travel model is run and provides slightly different accessibility variables over different model years once the various scenarios have been firmed up. Moving in the other direction, UrbanSim provides a complete set of socioeconomic data for the transport model every 5 years, and these serve as the basis for travel behavior. To date, data transfer between the two models is manual. In the near future, the handoff is planned to be automated for all model runs. MTC’s activity-based model is called CT-RAMP (Vovsha et al. 2011) and is being updated from 1,500 zones to 45,000 zones. In parallel, MTC is moving toward ActivitySim (http:// udst.github.io/activitysim), an open-source activity-based model that shares its code base with UrbanSim. The long-term plan is to use ActivitySim with CT-RAMP coefficients. As UrbanSim runs at the parcel level, spatially more detailed logsums are expected to further improve the land use model. Also, the parcel-level output is sufficient to populate the higher resolution zone system. Control totals for 2040 for the entire nine-county study area are provided by a REMI economic model and a cohort-survival model. Forecasts are provided at the county level, but UrbanSim is only constrained to match aggregate forecasts for the entire nine-county study area. Within the study area, UrbanSim can freely reallocate households, subject to available housing and demand at given prices. The EMFAC emissions model, developed by California’s Air Resources Board (CARB) and used in all urban models in the state, estimates traffic pollutants and greenhouse gas (GHG) emissions per capita. Furthermore, an Excel spreadsheet health model is used to estimate the fatalities from automobile travel. 5.6 Model Application WFRC The WFRC base year is 2011, and the UrbanSim model runs in 1-year increments to 2050. UrbanSim takes 15 to 25 minutes per year, and the official transport model takes 8 to 9 hours per model year. A simplified version of the transport model that runs in about 5 hours is used for the integration with UrbanSim. Using the simpler transport model, a fully integrated model run from 2011 to 2050 takes about 2½ to 3 days. The fully integrated model was used for the official LRTP. The focus was on developing regional centers to foster polycentric development. Scenarios with varying transport investments

Microsimulation Discrete Choice Land Use Models 41 were run, including adjustments to capital investment, bus service, highway expansions, better local street connectivity, zoning, and capacities for redevelopment. No consultants are needed to run scenarios, and scenarios can be implemented and analyzed in house. It is planned to show the scenarios to all member cities to select the preferred scenario. About every 10 years, projects will be prioritized. Many scenarios showed less effect than anticipated. WFRC’s staff recognizes, however, that most of the development area in 2050 is already built in 2011. Thus, smaller effects seem rea- sonable. In addition, staff are scrutinizing the model to ensure that it is properly sensitive to transportation investment. Some manual adjustments of UrbanSim results are necessary. Only results for 2050 are checked for reasonability. Once a final scenario was agreed on among the member cities, a proposal was accepted to scrutinize every intermediate year. The implementation of UrbanSim is starting to affect decision making in the region. Because the model helps develop better understanding of urban development, it has initiated useful conversations among staff members and elected officials in the region. There is general support from upper management. According to WFRC staff members, implementing UrbanSim and integrating it with the transport model has been a game changer in terms of recognizing inter- actions between land use and transport. MTC At MTC, four scenarios (as well as ABAG’s preferred plan) were modeled with UrbanSim. Each scenario had different land use policy assumptions. In the preferred plan, zoning was made stricter in certain areas and subsidies for construction were added to the model to match ABAG’s expected development. These five comparative scenarios were required for the Environmental Impact Report (EIR). A dozen environmental qualities were analyzed for the EIR. UrbanSim was used for the 2013 EIR process, but results were not used in MTC’s LRTP in 2013. UrbanSim was used for both the RTP scenarios and the EIR alternatives in 2017. Scenarios at MTC always start with a big vision for future development, which is often drawn in a diagram. Based on such a vision, policies are developed. The Big Cities Scenario (Figure 5-2), for example, combines transport policies (such as a VMT tax and train expansions) with land use policies that tie in with transport plans (such as urban growth boundaries, subsidies for development, zoning for denser development near transit stations, or taxing of construction of offices in inefficient locations). MTC always tested a bundle of multiple policies—they did not have the time to test individual policies. For the 2013 RTP, the model was run most of the time at the consultant’s offices at UC Berkeley and rarely at MTC. Since 2015, the model has run on a server with 100 GB memory, but only a quarter of the computer capacity is used by a single UrbanSim run. The model usually runs in Linux in a Virtual Machine on a Windows-based computer. The model has also been run in the cloud of a major virtual computer power provider, which had the benefit that the model was accessible from anywhere. Also, the model was run on a Mac Pro desktop computer with 64 GB memory. MTC tries to avoid adjusting model results manually. A script was written that performs a series of reality checks, which may lead to selected manual adjustments if some results are deemed unlikely. For the most part, the script serves as quality control. In the past, when issues in model results were found by this script, input data was cleaned up, which often fixed the issue. The script has also served to improve the land use model. For example, the script helped MTC staff members reveal that some governmental employees needed to be in general office

42 Integrated Transportation and Land Use Models space and not just in governmental buildings. But this employment should not “convert” into non-governmental employment just because it is in a general office building. Using this script also helped improve UrbanSim. Although manual adjustments of UrbanSim results could be avoided for the most part, some k-factors were implemented to comply with political expectations. K-factors are adjustments to the model implemented to nudge model results, without a consistent theoretical explanation Figure 5-2. Land use model results of the Big Cities Scenario (Source: MTC ABAG 2017).

Microsimulation Discrete Choice Land Use Models 43 of why the adjustment was needed. For example, no city was politically allowed to shrink in population or lose employment over 30 years, and k-factors ensured that this would never happen in the model—accordingly, this practice had to be implemented for small towns with less than a thousand households. According to MTC staff members, such k-factors were only minor adjustments to the model. Although UrbanSim results were not used in the 2013 LRTP, UrbanSim results helped offi- cials consider alternative policies. UrbanSim also has helped reveal what is happening in urban development and how policies may affect urban processes. Showing these effects has stimulated useful discussion about cause and effect in urban development. In the 2017 LRTP, UrbanSim results were used during the public planning process and the formal creation of RTP and EIR scenarios. This focused the discussion on potential policies as opposed to static visions of the future. At least two of the policies examined in the RTP scenarios continue to be discussed politically after the conclusion of the plan-making process: (1) a fee for developing office space in areas with poor job-housing ratios and (2) a policy for inclusionary zoning, which requires developers to include low-income housing in every new housing project. 5.7 Lessons Learned WFRC Back in 2012, WFRC had to make some difficult choices about whether they should invest in improving the transport model or implement a new land use model from scratch. The limited effects of transport investments on traffic flows convinced staff members to focus on the latter. To some degree, WFRC was disappointed to not find more effect on transport flows after the land use model was added. But after carefully reviewing results, the small effects appear to be probable, given that cities are not rebuilt from scratch over 39 years but merely extended and revised in detail. However, it has also been shown that some land use policies are even more effective than transportation investments. Zoning, in particular, showed effects more pro- nounced than many highway expansion projects. A local panel of land use and transport experts was consulted. The panel was established in a different project, but now reviews all model results for plausibility. The panel generally con- firmed the reasonableness of model output. This panel has added credibility to the modeling endeavor and a public voice in support of policies based on model results. A rather long runtime of the integrated model presents a challenge. Given the stochastic nature of the UrbanSim model, every model run produces different model results. Although those differences are negligible or even barely detectable at the regional scale, stochastic varia- tions can dominate model results when detailed results are used. For further discussion on the level of detail affected by stochastic variations, see Wegener (2011). To overcome the influence of stochastic variation, several runs with identical settings need to be run. Either each model result may be considered as a possible future development (as done by Gregor [2006] to show the uncertainty of model results), or an average of all model runs is taken (as done by Donnelly [2009] to average out stochastic differences). Given the relatively long runtime of the integrated model, multiple workstations would be required to run the model several times in parallel to systematically analyze model output in detail. WFRC was satisfied starting relatively simple and gradually improving the model. Over the course of 4 years, model developers created increasingly plausible model runs. Preconceived notions of how the model “should be built” were overcome. Of great value was the in-house exper- tise to run the model. Although initial training may be required, it pays off in the long run to reduce dependency on consultants for preparing and running scenarios as well as analyzing the results.

44 Integrated Transportation and Land Use Models Equally important were having (1) an ongoing budget for this effort and (2) staff authorized to spend time on this task. Developing a long-term plan for how resources are handled within the agency helped create working conditions in which staff members knew their efforts were valued. MTC An important lesson for MTC staff members was how essential data collection and processing are. In retrospect, they should have budgeted more time and money for data preparation. They also accepted using synthesized data where necessary, because it turned out to be impossible to collect observed data for all inputs to UrbanSim. Hence, it worked reasonably well for MTC to impute some data synthetically. At the same time, quality control of input data was crucial. MTC staff sought to make the land use model as transparent as possible and avoid creating black boxes. That UrbanSim is an open-source tool widely accepted as being theoretically sound has helped in this regard. Having the developer on site has been helpful to set up the model. Model implementation would probably have taken much longer if MTC had not contracted with the model developer to help setting up the model. MTC staff also recognized how important it is to reserve sufficient time for scenario analyses. A model that simply validates well and replicates reality is a scientific achievement but has little value for an MPO. To show what the model can do and what the sensitivities of the model are, it is crucial to market the implementation by presenting scenario results. Model developers tend to be more interested in model development than model application. This makes it even more important to budget time for someone who uses the model for sensitivity analyses, model test- ing, and scenario implementation. The true value of the model can be explained more easily if successful model applications can be shown.

Next: Chapter 6 - Spatial Input-Output Land Use Models »
Integrated Transportation and Land Use Models Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB's National Cooperative Highway Research Program (NCHRP) Synthesis 520: Integrated Transportation and Land Use Models presents information on how select agencies are using sketch planning models and advanced behavioral models to support decision making. The synthesis describes the performance of these models and the basic principles of land use/transport integration.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!