1

AN APPLICATION OF URBANSIM TO THE AUSTIN, TEXAS REGION:

INTEGRATED-MODEL FORECASTS FOR THE YEAR 2030

Siva Karthik Kakaraparthi Graduate Research Assistant

Department of Civil, Architectural and Environmental Engineering The University of Texas at Austin

6.9 E. Cockrell Jr. Hall Austin, TX 78712

siva_karthik19@yahoo.co.in

Kara M. Kockelman (corresponding author)

Associate Professor and William J. Murray Jr. Fellow Department of Civil, Architectural and Environmental Engineering

The University of Texas at Austin 6.9 E. Cockrell Jr. Hall

Austin, TX 78712 kkockelm@mail.utexas.edu

Submitted for Presentation at the 88th Annual Meeting of the Transportation Research Board and

Publication in Transportation Research Record

ABSTRACT This work describes the modeling of year-2030 land use patterns of the Austin, Texas region using UrbanSim, an open-source model for microscopic simulation of land development, location choices and land values, at the level of grid cells (typically 150 m x 150 m). Every five years a travel demand model was run, resulting in accessibility indices for use in UrbanSim. A business-as-usual trend scenario was compared to urban growth boundary (UGB) and added transport-cost-sensitivity scenarios, in order the appreciate UrbanSim’s performance and the potential land use and travel impacts of such policies. As expected, several land use results (e.g., population densities) appear highly responsive to both scenario contexts, and travel patterns are responsive as well. Local access variables (within 600-meter Euclidean distances) also enjoy significant relevance in this implementation of UrbanSim. While UrbanSim specification limitations are multiple and its data requirements are serious (and may be impossible for almost any planning agency to meet − even after substantial effort), the model does run reasonably fast and may make good sense over the longer term for interested regions with sophisticated planning staff on board to pursue. Enhancements will emerge over time, rendering the model more user-friendly and, hopefully, more accurate in prediction.

2

INTRODUCTION Population and job growth, demographic shifts, and transportation system investment and policy are expected to impact future land use and traffic patterns, and therefore air quality, congestion, energy use, urban sprawl, housing affordability, and many other facets of our communities. It is useful for transportation planners and policy makers to anticipate such shifts, for timely implementation of suitable policies. Land use patterns emerge from the developer decisions in concert with the location choices of households, industrial production, retail stores, offices, and other activities. Agent actions depend on land prices, market regulations, economic conditions, and other policies. The resulting activity location patterns are fundamental to travel demand prediction. And to some extent, transportation system investments and trip costs are relevant in the land development and location choices decisions agents make. In this study, year-2030 land use patterns of the three-county Austin, Texas region were forecast using Waddell et al.’s (2003) microsimulation model, UrbanSim. The study area consists of 1,074 traffic analysis zones (TAZs) subdivided into 328,318 grid cells (each measuring 150m × 150m). The following sections describe relevant literature, data sets and models used, along with application results. LITERATURE REVIEW A variety of land use models (LUMs) now exist1, and UrbanSim enjoys some valuable features. While most LUMs rely on neighborhood zone units (such as TAZs) and aggregations of actors (Dowling et al. 2000), UrbanSim emphasizes relatively small grid cells while tracking the grid cell locations of individual households and jobs.2 A 150m x 150m grid cell is 5.56 acres, while the average Austin TAZ is 1691 acres − or 300 times larger. While many models either ignore constraints on land use and built-space availability, UrbanSim emphasizes these key facets of urban form. (Waddell et al. 2003) While many models are highly deterministic in their predictions, UrbanSim does allow for the option of new seeds3. UrbanSim also uses a dynamic disequilibrium approach to forecast land use patterns, in contrast to many past models, which tended to rely on cross-sectional equilibrium approaches (Waddell et al. 2003). Of course, land use modeling is a complex endeavor, and the UrbanSim modeling system, like any abstraction of reality, exhibits many limitations. For example, households and firms do not evolve, workers are not linked to job sites, jobs are not linked to firms, economic interactions are 1 Relatively common LUMs include Putman’s (1983) gravity-based ITLUP, and models based on spatial input-output specifications, like de la Barra’s [1989, 1984] TRANUS, Kockelman et al.’s [2002] RUBMRIO, and Echenique’s [1990, 1969] MEPLAN, Others include Simmond’s [2001, 1999] DELTA, and Martinez’s [1996] MUSSA. 2 Other microsimulation-based LUMs are Landis and Zhang’s (1998a, 1998b) California Urban Futures (CUF) model, Miller et al.’s (2001) Integrated Land Use, Transport and Environment (ILUTE) model, Timmermans’ (2000) ALBATROSS model, Straunch et al.’s (2005) ILUMASS and portions of Hunt and Abraham’s (2003) PECAS model. 3 Here, casual experimentation with several random seed values suggested roughly 5 to 10 percent changes in predicted values of job and household counts at the gridcell level after a five-year period.

3

neglected, and job growth and economic conditions are exogenous. Moreover, the travel demand modeling process is external (with only a relatively weak link, through regional accessibility terms that may or may not make it into the location choice specification), the land price model is not generally fully estimated (since dynamic land price and vacancy data are so difficult to come by), and jobs and households are assigned one by one and do not compete for space in any given year4. In practice, UrbanSim’s location choice models are typically calibrated using cross-sectional data sets (rather than those of recent movers) and population synthesis is up to the analyst. There are many computing challenges for new users of UrbanSim, and data set acquisition will always pose a major issue for models as detailed as UrbanSim. Nevertheless, UrbanSim pays attention to key features of land markets; and, once it is running properly, it runs quickly. Moreover, information technologies are evolving and such challenges will be moderated (to some extent) over time. Finally, UrbanSim now permits estimation of most sub-model parameters within the OPUS environment, rather than requiring that users enter the model with parameters in hand. UrbanSim uses independent logit models to simulate the relocation decisions of existing households and firms, place households and jobs in grid cells5, and anticipate grid-cell-level changes in development type (Waddell 2004). Continuous logistic expressions provide estimates of residential land shares across zones, and land price estimates are based on a hedonic regression equation calibrated using ordinary least squares techniques. New transport infrastructure and local use restrictions are coded in at proper time points over the multi-year modeling process. Monte Carlo techniques are then used to simulate future year forecasts. Figure 1 illustrates UrbanSim’s sub-model and data set interactions, along with user-specified events (such as road building and changes in zoning policy) and scenario details. DATA SETS USED As described below, calibration and application of the UrbanSim model is a highly data intensive process, particularly for a large multi- county region. UrbanSim can simulate land use patterns at any gridcell resolution (Waddell 2001); a typical resolution is 150m × 150m and so was used here. Many assumptions had to be made, in placing and defining individual households and buildings at the grid cell level, in order to get the model to run. Few regions are likely to have such data and will need to resort to some sort of reasonable rules for data generation. Household Data UrbanSim’s household data set consists of a list of all households, with current locations (by gridcell), household size (number of members), age of the household head, race, and number of workers, children and autos. Household data was synthesized using iterative proportional fitting techniques at the level of year-2000 Census block groups (as described in Lemp et al. [2007] and McWethy [2006]).

4 Market response emerges via a land price shift in the following year, based on prior year vacancy rates in each cell. 5 According to Waddell (2008), the number of sampled alternatives is user defined, and is available as an option in a new version of UrbanSim. They have done some testing and believe that allocation results generally stabilize by around 30 alternatives.

4

Exogenous regional household control totals were obtained from Capital Area Metropolitan Organization (CAMPO) and used as model inputs. Annual relocation probabilities for households and the vacancies in residential units were imported from Eugene datasets and assumed to hold for Austin.

Employment Data The employment data consist of point locations for all firms6 and the number of jobs (by sector). Each job is given a unique ID number, to allow it to move independently of others. (While this greatly simplifies the modeling process, it is an important limitation of UrbanSim’s specification. In reality, firms make the great majority of job-site location choices, rather than individual jobs.) Several employment types are modeled separately; these are basic, retail, service, and education sectors, along with home-based jobs7. Annual relocation rates (for each job type) and non-residential vacancies also are required inputs.

Built Space and Transportation Data Austin is fortunate to now have much geographic data at parcel level resolution. Map layers (of parcel land uses [e.g., single-family and multi-family uses] were obtained from the Capital Area Council of Governments (CAPCOG) for the year 2005. Total housing units for year 2000 at the Census block-group level were obtained from TransCAD data CDs. These data sets were merged using simple rules to occupy the homes and generate multi-family units of specific sizes − all at an assumed 90 percent rate of occupancy (based on UrbanSim default targets). The parcel data give information on land use, land value and improvement value of all parcels in the three-county model region. The exact location of environmentally sensitive areas was assumed based on the percentages of sensitive lands in each TAZ. UrbanSim requires network travel times to the region’s CBD and major airport from each TAZ centroid, along with Euclidean distances to the nearest arterial and freeway from TAZ and gridcell centroids.8 MODEL SPECIFICATION Future land use patterns depend on household and job location choices, which in turn depend on the supply, quality, and price of built space, access to jobs and other destinations, household income, industry sector, and so forth. The following discussion describes the model estimation process for key sub-models. Household Location Choice Model (HLCM) UrbanSim’s household location choice sub-models are based a transition model and a relocation model. The household transition model generates a list of households to be added to or

6 Point locations of employment data were obtained as GIS layers from the Texas Work Commission via the Texas DOT and then cleaned/enhanced some by CAMPO. 7 Since home-based job counts were not available, two percent of total employment in each zone was assumed to be home-based. 8 Interestingly, UrbanSim does not call upon network files directly; all distances must be computed externally and provided (or coded internally by the analyst). This is an opportunity for relatively easy model improvement.

5

subtracted from the set of existing households. These shifts result from various demographic processes, like aging, marriage, divorce, births, and deaths. UrbanSim uses exogenous control totals to account for such demographic changes and gives an approximate measure for the number of households existing in a future year. The household relocation model considers the individuals moving from one location to another. It generates vacant spaces when the households move and adds the movers to the list of unplaced households. The probability that a household moves is dependent on the relative attractiveness of the new locality compared to the current locality. However, such data is unavailable for calibration. Therefore, exogenous relocation probability rates obtained from the Census current population survey are used. In the household transition model, the new households resulting from births, aging, marriages and divorces are added to the list of unplaced households and the vacant spaces are added when households are removed resulting from deaths. The set of unplaced households generated are placed into the vacant spaces obtained from the two component models by the household location choice model. The household location choice model is based on multinomial logit (MNL) framework and hence places the households into suitable gridcells, based on the composite utility of the gridcell. A gridcell of size 150m × 150m is considered as the unit of analysis. Explanatory variables that affect the household location choice reflect elements of urban economic theory and sociology (CUSPA 2006) (e.g., regional and local accessibilities, race, income levels, and land rents [see, e.g., Handy 1993 and Waddell 2002]). HLCM’s estimation results suggest that an increase in housing price, distance to the nearest freeway, number of residential units in the grid cell, population in the grid cell, number of vacant dwelling units in the grid cell, and travel time to the Austin’s CBD are estimated to have a negative impact on the residential location utility of a grid cell, everything else constant. In contrast, the average size of households living in the gridcell, distance to non-freeway arterials, access to workplaces, residential units within walking distance (600 meters, Euclidean distance), (model-estimated) land value, home values and rent, and number of residential units within walking distance are estimated to have a positive impact on such utilities. Also, it can be observed that households of similar income and those of minority race tend to co-locate in Austin. Employment Location Choice Models (ELCMs) UrbanSim’s employment location choice model (ELCM) is analogous to its household location choice model. A transition model generates or removes the newly created jobs in each sector, depending upon the growth or decline of employment in that sector as compared to the prior year. Such input assumptions generally are obtained from the state economic forecasts and/or commercial and in-house sources (CUSPA 2006). An employment relocation model then determines which individuals will change jobs in any given year. Employment relocation probabilities are obtained exogenously and given as inputs to UrbanSim. These probabilities depend on worker attrition rates and the utility of potential new jobs relative to a worker’s present job. A Monte Carlo sampling process, based on such relocation probabilities, determines whether a worker/job moves in the current year.

6

The jobs that are removed from the market and the jobs that are relocated are added to the database of job vacancies and the list of jobs that are created from the employment transition model and the employment relocation model are added to the database of unplaced jobs. The unplaced jobs are allocated into the gridcells by sector, based on the composite utilities of each gridcell. The employment location choice model is also based on a multinomial logit specification. The only difference between household and employment location choice models is that the latter allocates jobs using sector-specific preference functions (MNL model estimates), while all households are allocated here using a single MNL specification − with indicator variables for variables like race and income, to accommodate some forms of preference variation. The sectors considered in this study are industrial9, commercial10 and home-based employment11 sectors. The Industrial Employment Location Choice Model locates new and relocating industrial jobs based on the relative values of all grid cells’ composite utilities. Estimation results suggest that vacant industrial job space (which had to be “faked” here, due to data limitations), travel time to the region’s CBD, and job access (by zero-car households12) tend to attract industrial workers/firms, whereas the number of retail jobs within a gridcellhas a negative parameter estimate (ostensibly since industrial job sites tends to be segregated, away from retail uses). The travel-time-to-CBD variable is to a fair extent offset by the regional accessibility measure (which proxies for overall access of a neighborhood); and its positive coefficient reflects, to some degree, the nature of industrial activities to choose more peripheral locations, where zoning permits such uses and negative impacts on nearby households and businesses are moderated. The Commercial Employment Location Choice Model locates the commercial employment generated from the employment transition and relocation models. Commercial employment location choice differs from industrial location choice because the former behaves differently from the later. All the jobs that are allocated in commercial buildings are modeled using commercial employment location choice model.

Commercial job estimation results indicate that the utility of commercial employment location choice increases with the increase in industrial sq ft., distance to highways, residential units, service sector employment within walking distance, total value (cost of land and the improvement value), accessibility to work from households with one car, presence of non-residential sq ft., and number of high income households. The utility of commercial employment location choice decreases with the increase in average land value per acre, total land value, presence of retail sector employment within walking distance and the number of retail sector jobs.

9 Industrial employment refers to the employment placed in industrial buildings. 10 Commercial employment refers to the employment located in commercial buildings. 11 Home based jobs are those where workers work from home. Buildings are classified into industrial, commercial, governmental and home-based buildings. This is an input dataset but since data is unavailable in our case, this data is faked. 12 The zero-car and one-plus-car-household job accessibility values are highly correlated (ρ = 0.81), so either probably could have been used (but generally not both, to avoid certain issues of multicollinearity).

7

As described above, the home-based jobs data set was manufactured (using a two-percent-of-jobs-per-zone assumption), since Austin does not have records of such information at this time. The Home-based Employment Location Choice Model seeks to anticipate the distribution of such workers. Home-Based model results indicate that such jobs more often exist in older buildings and in neighborhoods with higher residential, non-home-based and same-sector employment densities, and less often in neighborhoods exhibiting a higher share of vacancies. This makes good sense given that the distribution simply mimics existing jobs distributions.

Modeling Land Prices UrbanSim’s Land Price Model provides a key input to the household and employment location choice models. Firm location determines employment location choice. Land price is modeled by UrbanSim using hedonic regression on attributes such as land use, site characteristics, accessibility variables, and neighborhood and zoning characteristics. The assumptions (CUSPA 2006) made in this model are:

1. Households, businesses and developers are all price takers and market adjustments are made to aggregate the demand and supply relationships. Each of these agents responds to the previous year’s price.

2. The location preferences and demand and supply imbalances alter the land price. 3. Land price varies with the variation in the current vacancy rate in comparison with long

term structural vacancy rate.

As the current vacancy rate falls below the long term structural vacancy rate, the land price increases and vice versa (DiPasquale 1996). The land price values are updated annually after all the construction and development is undertaken and after vacancy rates are computed.

iltsi

cit

si

ilt XV

VVP βδα +−+= )(

where Pilt is the land price per acre of development type i at location l and at time t, Vi

s is the long term structural vacancy rate, Vit

c is the current vacancy rate at time t, Xilt is the vector of site attributes, and α, β, and δ13 are parameters to be estimated. The prices computed in the current year are used to compute the next year’s market activity (Waddell et al. 2003). Land Price estimation results indicate that prices rise with increases in residential property values (per residential unit within walking distance), year in which the housing units are built (so older homes are more valued, due perhaps to more mature trees or more built-ins), presence of arterials and highways in the vicinity of the site, commercial space in the gridcell,, accessibility of jobs from households with one car, residential units, non-residential sq ft. within walking distance, percentage of developed lands and high income households within walking distance. The land price decreases with the increase in the industrial sq ft. within walking distance, percentage of open space within walking distance and residential area and population.

13 Due to a lack of occupancy rate data for two time points across zones, δ could not be estimated and was simply set equal to 1.0 here, to be able to run the model and evaluate performance.

8

The Residential Land Share Model is used to compute the residential land share in a gridcell. The residential land share is computed using:

y

y

eey+

=1

'

where y is obtained from ordinary least square (OLS) regression of residential land use shares in each grid cell. Essentially, the model assumes a logistic specification for the fractional shares, but OLS methods can be used in estimation. From Table 6, it can be observed that the residential land share in a gridcell increases with the increase in average income of the households living in the gridcell, service sector employment present within walking distance, residential density and the percentage of residential land in the lag year. The residential land share in a gridcell decreases with the increase in the average residential value per residential unit, developable residential units and basic sector employment within walking distance. Introduction of a Travel Demand Model (TDM) Figure 1 illustrates the external linkage of UrbanSim to a travel demand model (TDM). UrbanSim has been successfully interfaced with TP+, MinUTP, EMME/2, and other modeling systems (according to OPUS 2006 workshop documents), and Caliper’s TransCAD software was used here. Using 1996/1997 Austin Travel Survey (ATS) data, the TDM employed here was estimated by Lemp (2007) and relies on fairly standard techniques. Regression models are used for trip generation, at the household level for home-based trips and at the zonal level for non-home-based trips. An MNL model of destination choice is used for trip distribution, and includes a logsum parameter measuring the maximum expected utility achieved over all modes and times of day (TODs). And a joint MNL model of mode choice (drive alone, shared ride, transit, and bike/walk) and four TODs was used. Separate models were used for each of four trip types (home-based work, home-based non-work, non-home-based work, and non-home-based non-work). Finally, deterministic network assignment routines were used for each TOD period, and four feedback iterations (from network-equilibrium travel times and costs to trip distribution) were performed, in order to obtain estimates of inter-zonal travel times, trip distances and travel costs − specific to each of the four modes at each of the four TODs14. Using this information, logsum accessibility indices were computed and then input into UrbanSim, in order to anticipate the next five years’ land use patterns. UrbanSim was run every year, for 30 years (for 2001 through 2030), while the TDM was applied six times (2005 through 2030, at five-year time points). The logsum values are computed as follows (see, e.g., Ben-Akiva and Lerman [1985]):

14 Truck and external trips were not modeled explicitly. Instead, CAMPO's estimates of such trips were added to TOD-specific trip tables before traffic assignment. While not ideal, such methods are not uncommon and provide a simple way for dealing with travel of these types.

9

⎟⎠

⎞⎜⎝

⎛= ∑∈Cm

nijmtij UL )exp(ln

nijmtn

nijmtnGCnmtnijmt TT

VOTTCost

U ++= )(,1 ββ

And these are used in somewhat simplistic accessibility indices, as proscribed by UrbanSim documentation (CUSPA, 2006):

∑=

=J

j

Lji

ijeDIA1

..

where i and j index origin and destination zones, n indicates trip type (e.g., HBNW), m indicates mode, t denotes TOD, β1nmt is the alternative specific constant from the joint mode-TOD choice model, βGC is the coefficient of generalized trip costs, TT is travel time, VOTT is the assumed value of travel time ($9 per person-hour for work trips and $4.5 per person-hour for non-work trips), and COST indicates trip cost (assumed to be $0.20 per vehicle-mile). For purposes of TDM feedback to UrbanSim in this study, accessibility indices (AI’s) were computed only for home-based work (HBW) trips made during the AM peak period, for zero-vehicle and 1+-vehicle-owning households, separately15. As expected, these twin AI values were highly correlated, and would be highly correlated with other AIs at other times of day by other modes, so only the zero-vehicle-household AI values were controlled for in the location choice models described earlier. RESULTS OF MODEL TESTING To what extent do transportation decisions impact land use patterns, or at least land use pattern predictions? Guiliano (1995) argued that any feedback from transport to land use had weakened, and location choice appears largely independent of job access within a region. Cervero and Landis (1995) agreed that falling transport costs had weakened the connection, but transportation investments and policies still play a key role in land use patterns. And Handy (2002) concludes that highways still determine the location of new development. Of course, worsening congestion, recent jumps in gasoline prices, and the emergence of tolled highways in many U.S. states may strengthen such transport-land use relationships. Several distinct scenarios were examined here, to get a sense of UrbanSim’s performance under various designs, as well as possible Austin futures. A Business as Usual (BAU) scenario simply held the network constant over the modeling horizon, but added new households and jobs at a rate of roughly 2.8 percent per year, respectively − a rather hearty growth rate. Using the same network and control totals on jobs and households, two more scenarios were tested, to examine the effects of land use policies and added travel costs.

15 While the TDM was not segmented on the basis of vehicle ownership levels, transit and bike/walk mode choice model specifications were used for the zero-vehicle households’ accessibility index calculations and drive times were used for the 1+ vehicle-owning households’ index.

10

As alluded to above, UrbanSim allocates households and firms on a yearly basis, as it updates land prices, built space, and government policies across gridcells. Here, land use patterns were forecasted through the year 2030, by updating land use patterns every year (using the sub-models described above) and updating regional accessibility values by running a travel demand model every five years. Figures 2(a) and 2(b) show predicted household and employment densities for the BAU scenario across gridcells in year 2030. Surprisingly, households and jobs are predicted to move closer to the CBD zones, in part because the current transportation network cannot handle the exponential growth in households and jobs, making congestion almost unbearable. The count-weighted average16 household and employment densities are 1,568 households and 7,148 jobs per square mile, and regional VMT during the AM peak period is estimated to be 9.1 million (or 7.33 miles per capita). Urban Growth Boundary (UGB) Scenario To keep things interesting, a striking urban growth boundary (UGB) policy was tested, with 453 of the region’s 1,074 TAZs falling outside the boundary (as shown in Figure 3(a)), and thus unable to receive any new development. The UGB was defined as follows: Any TAZs with more than two job equivalents per acre17 and those sharing an edge with these were permitted development. Figures 3(b) and 3(d) illustrate household and job densities after year-2000 implementation of the UGB policy, while Figures 3(c) and 3(e) present differences in densities relative to the BAU scenario. Dramatic increases in land use densities result from this UGB scenario: count-weighted average densities for households and jobs are estimated to be 13,951 and 48,681 per square mile, respectively. sDensification is estimated to occur across the growth-bounded sub-region, but is particularly noticeable in the CBD zonesThe UBG scenario’s VMT is estimated to be 8.7 million during the AM peak, or 4.1 percent less than that of the BAU scenario. Added Travel Cost Sensitivity (TCS) Scenario In order to get a sense of the practical effect of travel cost changes, the sensitivity of household and employment location choices to increased travel costs was determined. Essentially, the βGC parameter for mode choice (in Eqn. 1) was doubled, resulting in shorter trip making and, ultimately, lower accessibility indices and higher density and more clustered development, as shown in Figures 4(a) and 4(c). As with the UGB scenario, Figures 4(b) and 4(d) present density differences, relative to the BAU scenario. As expected, heightened sensitivity to travel costs reduced travel distances and thus network travel times while raising the relative utility of more centrally located/more regionally accessible zones, resulting in a far more clustered pattern of year 2030 development, as suggested by the ten-fold rise in weighted-average household and jobs densities (to 16,404 and 48,896 per square

16 The number of households and jobs are used as weights in calculating the count-weighted average densities, reflecting average density as experienced by the average worker or household. If simple averages are used, one will simply get a regional density value that is constant across scenarios. 17 1 household equals 0.7143 job equivalents since total regional employment implies 1.4 jobs per household.

11

mile, respectively), relative to the BAU scenario. AM peak VMT is estimated to be 8.2 million miles, or almost 10 percent less than that of the BAU scenario. CONCLUSIONS AND RECOMMENDATIONS Here, Austin year 2030 land use patterns were forecasted using UrbanSim at 150 meter resolution, primarily to get a sense of modeling challenges and model performance. As detailed as Austin’s current data sets are, various assumptions on building age and floor space, occupancy values, agent relocation probabilities, development decisions, demographics, environmental layers, and synthetic household demographics were required in order to fulfill UrbanSim’s heavy data demands, and many values simply had to be manufactuered or borrowed. Even with a variety of heroic assumptions in generating required variables for a region as well versed in spatial data acquisition and assembly as Austin, the data formulation stage of the entire one-year modeling process required at least 50 percent of a dedicated analyst’s time (e.g., 4 to 5 person-months). Model estimation, definition of scenario assumptions and actual forecasting tasks required much smaller shares of the total one-year process (estimated to be 15%, 10% and 25% of total effort, respectively). While the model predictions may be highly inaccurate at this stage, it is hoped that improvements to data sets and associated parameters will remedy a reasonable share of this. The main challenge remains in the data arena, but model code enhancements also emerge as candidates for any new work in UrbanSim. ACKNOWLEDGEMENTS The authors wish to thank the Texas Department of Transportation for funding this research, under project 0-5667 (titled “Analysis and Guidelines for Establishing Unified Urban Land-Use and Transportation System Planning Framework and Procedures”) and U.T. Austin graduate students Brenda Zhou, Jason Lemp, and Jen Duthie for providing parcel maps, travel demand model details, and all-around support of this effort. We also appreciate Houston-Galveston Area Council’s Dmitri Messen’s and the University of Washington’s Hana Ševčíková’s assistance in running UrbanSim, via helpful email correspondence. REFERENCES Ben-Akiva. M., and S. Lerman. 1985. Discrete Choice Analysis, The MIT Press, Cambridge Massachusetts. Cervero, R., and J. Landis. 1995. “The Transportation-Land Use Connection Still Matters,” Access 7: 2-10. Center for Urban Simulation and Policy Analysis (CUSPA). 2006. Opus: The Open Platform for UrbanSim Version 4.0 Reference Manual and Users Guide. URL: http://www.urbansim.org/opus/releases/opus-4-1-2/docs/opus-userguide.pdf. Accessed June 27, 2008.

12

DiPasquale, D., and W. Wheaton. 1996. Urban Economics and Real Estate Markets. Prentice Hall, Englewood Cliffs, New Jersey. Dowling, R., R. Ireson, A. Skabardonis, D. Gillen, P. Stopher,.A. Horowitz, J. Bowman, E. Deakin, and R. Dulla. 2000. Predicting Short-Term and Long-Term Air Quality Effects of Traffic-Flow Improvement Projects. National Cooperative Highway Research Program 25-21. Transportation Research Board, Washington D.C. Echenique, Marcial H., D. Crowther, W. Lindsay. 1969. A Spatial Model of Urban Stock and Activity. Regional Studies 3: 281-312. Echenique, Marcial H., A.D.J. Flowerdew, John D. Hunt, T.R. Mayo, I.J. Skidmore, David C. Simmonds. 1990. The MEPLAN models of Bilbao, Leeds and Dortmund. Transport Reviews 10 (4): 309-322. Guiliano, G. 1995. The Weakening Transportation – Land use Connection. Access 6: 3-11. Handy, S. 1993. Regional Versus Local Accessibility: Implications for Nonwork Travel. Transportation Research Record Number 1400: 58 – 66. Handy, S. 2002. Smart Growth and The Transportation-Land use Connection: What Does the Research Tell Us? Prepared for New Urbanism and Smart Growth: A Research Symposium. URL: http://www.des.ucdavis.edu/faculty/handy/MD_paper.pdf. Accessed June 27, 2008. Hunt, J.D., and J. E. Abraham. 2003. Design and Application of the PECAS Land Use Modelling System. Proceedings of the 8th Conference on Computers in Urban Planning and Urban Management. Sendai, Japan. Landis, J., and Zhang, M. 1998a. The second generation of the California Urban Futures model. Part 1: Model logic and theory. Environment and Planning B: Planning and Design 25: 657 - 666. Landis, J., and Zhang, M. 1998b. The second generation of the California Urban Futures model. Part 2: Specification and calibration results of the land use change module. In: Environment and Planning B: Planning and Design 25: 795 – 824. Lemp, J.D, McWethy, L., and Kockelman, K. 2007. From Aggregate Methods to Microsimulation: Assessing the Benefits of Microscopic Activity-Based Models of Travel Demand. Transportation Research Record 1994: 80-88. Lemp, J.D. (2007) Travel Demand Forecasting Models: Development, Application, and Comparison of Aggregate and Activity-Based Approaches for the Austin, Texas Region. Master’s Thesis, Department of Civil, Architectural, and Environmental Engineering, University of Texas at Austin.

13

Martínez, Francisco J. (1996) MUSSA: Land Use Model for Santiago City. Transportation Research Record No. 1552: 126-134.

McWethy, L. M. 2006. Comparing Microscopic Activity-Based and Traditional Models of Travel Demand: An Austin Area Case Study. Master’s Thesis. Department of Civil Engineering, University of Texas at Austin.

Miller, E.J. 2001. Integrated Land Use, Transportation, Environment (ILUTE) Modelling System. URL: http://www.civ.utoronto.ca/sect/traeng/ilute/ielute_the_model.htm. Accessed June 28, 2008. Putman, S.H. 1983. Integrated Urban Models: Policy Analysis of Transportation and Land Use. London: Pion. Simmonds, D. C. 1999. The design of the DELTA land-use modelling package. Environment and Planning B: Planning and Design 26: 665-684. Simmonds, David C. 2001. The Objectives and Design of a New Land-use Modelling Package: DELTA. In: G. Clarke and M. Madden (Eds.) Regional Science in Business, Advances in Spatial Sciences. Berlin: Springer. Pp. 159-188. Waddell, P. 2001. Analytical Tools for Land Use, Transportation and Growth Management. URL: http://www.urbansim.org/papers/Analytical_Tools.pdf. Accessed June 27, 2008. Waddell, P. 2002. UrbanSim: Modeling Urban Development for Land Use, Transportation and Environmental Planning. Preprint of an article that appeared in the Journal of the American Planning Association: 68. 297-314. URL: http://www.urbansim.org/papers/UrbanSim-JAPA.pdf Accessed June 28, 2008. Waddell, P., A. Borning, M. Noth, N. Freier, M. Becke and F. Ulfarsson. 2003. Microsimulation of Urban Development and Location Choices: Design and Implementation of UrbanSim. Networks and Spatial Economics 3: 43-67. Waddell, P., and F. Ulfarsson. 2004. Introduction to Urban Simulation: Design and Development of Operational Models. Transport Geography and Spatial Systems 5: 203-236. Waddell, P., G. Ulfarsson, J. Franklin, and J. Britting. 2007. Incorporating Land Use in Transportation Planning. Preprint of paper published in Transportation Research Part A 41: 382-410. Waddell, P. 2008. Email correspondence with Kara Kockelman on July 29. Zhou, B., and K.M. Kockelman. 2008. “Predicting the Spatial Distribution of Households and Employment: Application of A Gravity-based Land Use Model”. Under review for presentation at the Transportation Research Board’s 88th Annual Meeting. Washington, DC, January 2009.

14

15

LIST OF TABLES and FIGURES Table 1. VMT, and Count-weighted Average Densities for Household and Jobs in BAU, UGB and Travel Cost Sensitivity (TCS) Scenarios, for Year 2030 Figure 1: Interaction between Various Datasets and Models in UrbanSim Figure 2: Household and Employment Locations across Gridcells in 2000 and 2030, under the BAU Scenario Figure 3: Household and Employment Locations in 2030 under the UGB Scenario, with Differences from BAU Results Figure 4: Household and Employment Locations in 2030 under the Increased Travel Cost Sensitivity Scenario, with Differences from BAU Results

Table 1. VMT, and Count-weighted Average Densities for Household and Jobs in BAU, UGB and Travel Cost Sensitivity (TCS) Scenarios, for Year 2030

BAU UGB TCS

VMT AM Peak Hour 9,107,379 8,725,855 8,218,373 Count-weighted Average Household Density (households/sq.mi.) 1,568 13,951 16,404 Count-weighted Average Jobs Density (jobs/sq.mi.) 7,148 48,681 48,896

16

Fig 1: Interaction between Various Datasets and Models in UrbanSim

(Source: Waddell et al., 2007, Fig. 1)

17

2(a): Households in 2030 (Business as usual) 2(b): Jobs in 2030 (Business as usual) Figure 2: Household and Employment Locations across Gridcells in 2000 and 2030, under

the BAU Scneario

18

3(a): Urban Growth Boundary (Blue zones allow new growth)

19

3(b): Households in 2030 (UGB policy) 3(c): Differences in Households due to

UGB Policy (UGB - BAU)

3(d): Jobs in 2030 (UGB policy) 3(e): Differences in Jobs due to UGB

Policy (UGB - BAU)

Figure 3: Household and Employment Locations in 2030 under the UGB Scenario, with Differences from BAU Results

20

4(a): Households in 2030 (Increased travel cost 4(b): Difference in Households 2030 sensitivity) due to increased TCS (TCS - BAU)

4(c): Jobs in 2030 (Increased travel cost sensitivity) 4(d): Difference in Jobs 2030 due to Increased TCS (TCS - BAU)

Figure 4: Household and Employment Locations in 2030 under the Increased Travel Cost Sensitivity Scenario, with Differences from BAU Results