Introduction
Planners, developers, real estate brokers, and anyone else whose job depends on keeping a close watch on real estate development can give you a pretty good educated guess about what land in a region is likely to be built, and what land is likely not to be built, based on their knowledge of land costs, zoning factors, infrastructure availability, water and utility access, site access and visibility, environmental conditions, city development review practices, and local market conditions. Or, as they say, there are three factors that dictate the success of a business: "location, location, and location."
Quantifying opinions about future growth patterns is certainly impractical on a massive scale such as in Maricopa County, where there are over a million parcels covering more than 9,000 square miles and land development forecasts are needed to place over 1.5 million new people over the next 25 years. Even if land use forecasts are needed at five year intervals over the next four decades, urban growth models should still take into consideration the same factors that interest developers
SAM-IM is a rule-based urban growth model. It simulates both short-term and long-term urbanization of a region by reacting to any set of factors and conditions that the planner wishes to express. The model is completely embedded in a Geographic Information System (GIS) – it runs on ArcView GIS using the Spatial Analyst extension. The concept of the model is entirely GIS oriented – all of the data that drives the model, whether it be existing land use distributions, future market conditions, adopted planned land use, developments already approved and underway, or local land conditions, are expressed geographically, in the form of ArcView shape files. Anything that can be expressed geographically can be taken into consideration in the model.
This paper describes the SAM-IM suite of modeling applications that have been developed to support land use forecasting at the Maricopa Association of Governments, the Metropolitan Planning Organization (MPO) and regional planning agency for the Phoenix metropolitan area.
GIS at MAG
The Maricopa Association of Governments (MAG) is responsible for regional land use, transportation, and air quality modeling for the Phoenix metropolitan area. Like other regional governments with similar MPO responsibilities, MAG has been seeking to improve the performance of these models, thereby making the regional planning process more effective. GIS has played a major role at MAG for the past twelve years. This technology has significantly enhanced MAG’s ability to manage dozens of planning databases: databases describing census statistics, employment inventories, land use, general plans, parcels, building permits, and highway infrastructure – all sharing one common characteristic -- they are essentially spatial in nature. GIS helps MAG assume an important responsibility in the region – the role of regional information agency for public planning.
GIS has capabilities that run far beyond just being a platform for planning and mapping. In recent years, MAG has been developing new classes of planning models that run directly on top of GIS. This new class of models works directly from GIS databases. More importantly, these models tap a wide variety of powerful spatial analysis methods that are found only in GIS and can not be even remotely matched by older custom-written programs developed over the last three decades. In the long term, this new class of GIS-based models offers the potential for completely replacing older, non-spatial planning models.
The SAM-IM simulation generated this forecast of growth
and development in Maricopa County over the next 20 years.
Shown in gray are existing "built" uses (in 1995).
Shown in black are new developments projected to occur by the year 2020.
(The proposed 2020 freeway system is also shown).
This paper describes one such model: the Subarea Allocation Model (SAM), developed and run at the Maricopa Association of Governments. SAM forecasts land use and development throughout the MAG planning region. It does this by simulating factors that influence the value of land and the likelihood that land will be built, based on those factors. It observes planning policies – general plan designations controlling the use of land (approved by municipalities in the region), for example, are observed by the model.
The land use, population and socioeconomic modeling at MAG is based on a three-tier modeling process. The first tier is a demographic model that is used to produce county control totals. The second tier involves using a spatial interaction model to allocate the county control total population and employment to subregions. The third tier allows for the allocation of the subregional population to smaller areas drawing upon GIS representation of land use plans and local policies of MAG member agencies.
The first tier model is a county level model. In accordance with Executive Order 95-2, the preparation of county and state level population projections is the responsibility of the Arizona Department of Economic Security (DES). This model is a demographic model, projecting births, deaths and net migration in each county for a fifty-year time horizon. This model incorporates population by age and sex, birth rates, death rates and net migration trends. The model takes into account short term economic conditions, but not long-range employment trends.
For the second tier process, MAG is using DRAM/EMPAL. DRAM and EMPAL are registered trademarks of S.H. Putman Associates. The two models, DRAM (Disaggregated Residential Allocation Model) and EMPAL (EMPloyment Allocation Model), forecast household location, and employment location. These models are being used by a number of major metropolitan areas. DRAM/EMPAL projects the spatial patterns of households and employment in the MAG region. The forecasting procedure starts with regional trends, transportation facility descriptions and data on the current location of employment by sector. This information is then used to project the future location of households. The projections are done for five-year intervals. Each five-year step begins with the EMPAL model to project employment by sector by zone. DRAM modeling to project households by income category follows the EMPAL run for that time period.
The third tier Subarea Allocation Model (SAM) allocates population and employment from Regional Analysis Zones (RAZ) to one acre grids which are then aggregated to Traffic Analysis Zones (TAZ). SAM generates simulations of how Phoenix will grow as a region over time. Its current role is to bridge the gap between the spatial allocation model, DRAM/EMPAL, and EMME/2, MAG’s transportation model that works on much more detailed geographies. So, SAM’s official role is an "allocation" model in that it disaggregates land use forecasts generated by DRAM/EMPAL for large statistical areas to smaller TAZs that are needed to drive the transportation model.
Model Integration
The sequence of land use models and transportation models run at MAG is recursive. DRAM/EMPAL requires assumptions about future travel times between RAZs across the transportation network in order to generate forecasts of land use. These travel time forecasts are available from the transportation model. In turn, the EMME/2-based transportation model requires forecasts of future land use in order to generate estimates of travel times between RAZs. This has led to the development of a MAG "feedback loop," whereby the DRAM/EMPAL and EMME/2 model chains are executed a number of times, feeding the results of one model back into the other, until "closure" is achieved. This is necessary to assure that the results generated by one model are consistent with the assumptions made by the other.
SAM-IM provides a bridge between forecasts generated
by DRAM/EMPAL at a relatively coarse geographic
scale to the transportation model EMME/2, which operates
with much greater geographic detail.
From a modeling standpoint, the integration of two independently built models, DRAM/EMPAL for land use and EMME/2 for transportation, presents two significant database issues:
Subarea Allocation Models
The Subarea Allocation Model (SAM) was first posed as a land use disaggregation model. The objective was to develop an automated procedure to bridge the geographic gap between DRAM/EMPAL forecasts generated for 147 RAZs and the needs of the transportation model which is driven by a 1,549 TAZ system. Also, the automated procedure had to inject other variables needed by the transportation model, some of which are highly correlated with DRAM/EMPAL forecasts and therefore should be derived from them.
Before embarking upon the development of SAM, MAG’s level of investment in GIS capability was sufficiently advanced to support the model's requirements. MAG had developed existing land use covers for all of Maricopa County, and it had assembled and unified the general plan covers adopted by municipalities throughout Maricopa County. MAG had also mounted a number of other efforts germane to SAM-type model concepts. For example, MAG was tracking, on GIS, active developments throughout the region as well as planned developments in various stages of the approval process. Retirement communities had been mapped, redevelopment district coverages were available and MAG had a capability to represent future EMME/2 transportation networks in GIS. So, in short, the database support needed for a SAM-type model was already available.
The original SAM model was implemented as a set of ARC/INFO AMLs running on UNIX, using the ARC/INFO GRID module to convert feature coverages to grids. Much of the logic associated with allocating growth capitalized on spatial analysis functions available in ARC/INFO GRID. In fact, MAG had participated in the development of a grid-based model with a 40-acre resolution in the late 1980's, but ARC/INFO GRID allowed MAG to consider modeling at a much finer resolution.
The implementation of SAM was completed in 1996/97 and was used for the purposes of generating land use forecasts for the region for five-year intervals, 2000 through 2020. These forecasts were reviewed with the member governments and were ultimately adopted by the Regional Council in 1997. Some of the key characteristics associated with the SAM model are:
SAM-IM simulates the growth of regions by looking for land that is best suited for absorbing development.
All of these computational operations are not actually carried out on feature themes; instead, land covers and site suitability evaluations are represented as grids. SAM-IM converts them automatically. Grid representation is used for a number of reasons.
SAM-IM predicts future growth and urbanization by simulating
development. The allocation mechanism evaluates the
suitability of land to absorb development and chooses
the most appropriate sites in conformance with general plans.
The actual implementation of the allocation mechanism in SAM-IM offers a number of other add-ons and features not mentioned above. These include:
Users can define how allocations should be performed
for each of the 14 different forecasting sectors.
SAM-IM provides a variety of methods for allocating
growth, the most sophisticated of which involves an
appraisal of the suitability of land to absorb different
kinds of development.
Site Suitability Evaluation
The growth allocation mechanism in SAM-IM is driven by the evaluation of site suitability. Growth is allocated to sites (grid cells) that are considered to be more suitable than their competing neighbors. This is accomplished by representing the evaluation of land to absorb development by "site suitability scores". Growth is allocated to land on a rank-order basis.
The model itself provides this mechanism. It will allocate according to any "site suitability evaluation" system defined for it, linear or non-linear, normalized or non-normalized -- the only requirement is that it must be expressed as a grid. The modeler, however, can create any kind of site suitability scoring system, reflecting any set of site characteristics or conditions, for which there is information.
The original site evaluation methodology incorporated these measures:
This is a composite image of the overall evaluation
of sites for residential land, with darker shades reflecting
better locations (in areas not already built). SAM-IM
uses this land evaluation theme to target growth.
Calibrating Site Evaluation "Scoring" Equations
The original set of site evaluation measures hypothesized for Maricopa County has not been formally calibrated. However, this calibration is scheduled to occur during the summer of 2000. The calibration procedure will be based on an analysis of land use change between two different points in time. Development and growth occurring over this time frame can be identified on a cell-by-cell basis, along with various measures of the features and characteristics associated with growth cells.
SAM-IM Trip Generation
One of the principal goals of SAM-IM is to provide forecasts for the 1,549 TAZs required by the transportation model. In addition, though, MAG’s transportation model requires TAZ-level forecasts for variables other than those that come from the DRAM/EMPAL and SAM-IM land use modeling environments. Examples are projections of enrollments at post-high school campuses and total air traffic embarkations at Sky Harbor Airport. Projections of both of these variables are geographic-specific but are prepared completely outside of the land use forecasting environment.
SAM-IM contains a module that provides features for generating trip generation data sets from SAM-IM forecast land use covers. These features provide for the estimation of parameters "derived" from the land use forecasts themselves, as well as for the insertion of forecasts of other independent variables acquired from other independent sources. The module will assemble these various sources, will do the necessary computations, and will generate the results according to any user-specified TAZ geography. Furthermore, SAM-IM will let the user specify the output format of the data.
Historically, TAZ geographies and data requirements change with the evolution of the transportation model. Consequently, it is important for SAM-IM trip generation functions to be programmable.
Similar in concept to Esri’s new "Model Builder" extension (but predating it), SAM-IM allows users to write "programs" that express how the trip generation database is to be created. The "program" has these features:
The Trip Generation feature of SAM-IM combines variables
drawn from the land use allocation and summarizes
them for any TAZ system required by the transportation
model. In addition, users can write expressions to
generate estimates of any derived variable drawn from
factors represented by other GIS covers.
Historically, much of the tedious, labor-intensive work with land use and transportation databases is driven by the need to operate models with inherently different geographical zone structures. Furthermore, much of the source data that is used to generate estimates of many variables comes from yet other geographies.
Zone system database maintenance is one of the major impediments to migrating data sets created from one modeling environment to another, and therefore is a major concern. Historically, much attention was focused on maintaining a variety of geographic zone structures, making sure that zone structures are "upwardly" compatible. Changes made in one zone structure impact all of the others for which compatibility is being maintained. Every change to zone geography carries with it attendant database issues – households, employment, resident population, visitor population, and other attributes all have to be re-estimated for the new zone system, often requiring a retreat back to the original source data.
Consequently, one of the principal objectives of SAM-IM was to build a model and a database structure that was fundamentally "zone-less" in nature -- where land and development, both now and in the future, was not dependent on any geographic system of zones. Consequently, aggregating land use information to any zone system that might be needed to drive a land use or transportation model could be done fully automatically and routinely.
The best way to represent land use at a micro-level and on a "zone-less" system is with grid cells ("raster geography"), such as those inherent to ARC/INFO GRID and the ArcView Spatial Analyst. The finely-grained, most detailed, level of geography typically possible with conventional vector (polygon) representation presumably would be the "parcel" level – but this is not good enough. First, "parcel" coverages are simply too big – processing a parcel database for Maricopa County of over one million records is too much for ArcView to perform effectively. Also, parcels themselves can easily cross zone boundary lines thereby introducing the very same zone-boundary problem that the model is designed to resolve. And finally, land use development that is being simulated primarily occurs in undeveloped and unsubdivided regions – the exact places that the parcel coverage does not support effectively.
Grid representation also gives the model access to a number of useful spatial analysis functions that do not exist for conventional vector geographies. A proximity calculation, for example, cannot be effectively measured for polygons in an unbiased way because of the variation in size in polygons. Grid representations of geography, on the other hand, can treat land based on its specific cell location, regardless of what polygon it is a part of. Grid representations also offer an entire family of raster functions and operations, such as "proximity", "neighborhood", "partial selection", "compute" and "combine" that simply do not exist for vector geographies.
The SAM-IM allocation mechanism depends heavily on a grid representation of land use geography. The basic premise of the model would not even be possible without the capabilities of ArcView Spatial Analyst or something comparable. This being said, the implementation of the model using ArcView grid did present us with several technical issues. These are covered next.
Single Attribute Grid Representation
Much of the grid representation of geography offered by the ArcView Spatial Analyst seems to have been driven by natural resource applications, where each grid represents a single specific attribute of land: for example, soil type, permeability, vegetation classification, etc. The data structures needed for urban land use modeling are much more complex – they need to carry a number of different attributes which a single grid cell can exhibit, including land use code, dwelling unit density, employment density, and densities associated with any one of a number of different use characteristics (nursing homes, hotel and motel rooms, housing classes, etc.). A "single attribute" representation leads to a proliferation of grid databases and high execution time needed to build and process the databases. Fortunately, ArcView does provide mechanisms to associate multiple attributes with a single cell through the "VAT," which is virtually the same as a standard dBase or INFO table. If land use grids are built based on a "polygon identification number", then all of the attributes associated with the original polygon can be migrated to the pertinent grid cell and the "VAT" of attributes can be manipulated with conventional database management functions (i.e., fields can be added, deleted, etc.).
Issues with Spatial Analyst
The ArcView Spatial Analyst proved to be sufficiently stable to support the SAM-IM model, a model that has grown significantly in computational complexity and sophistication. Still, there are several aspects associated with the Spatial Analyst that posed serious problems and limitations. These are:
For example, SAM-IM users are limited when designing site selection scoring equations or reflecting the results of a statistical calibration. Our practical experiences at MAG are that site scores have to be "normalized" within certain restricted numeric ranges, such as integers between 0 and 100. This is not an absolute rule, but one that applies specifically to Maricopa County – the size of the grid cell used there, along with the number of land use and general plan polygons, all conspire to fail due to size restrictions if the scoring theme has more than 100 individual scores. The limitation can not be determined in advance – it can only be determined by trial.
Variable Grid Cell Sizes
MAG’s legal jurisdiction for regional land use, transportation, and air quality modeling covers all of Maricopa County. Maricopa County, like other planning areas in the Southwestern United States, is large (9,000+ square miles). And therefore, it is appropriate that the scope of the modeling area should match it even though a large part of land in Maricopa County is undeveloped desert.
The ArcView Spatial Analyst requires grids to be uniform with respect to grid cell size. This seems entirely natural (compute screens don’t have pixels of variable dimensions), but for land use modeling it would be helpful if grid cells could be variable in size. The size of grid cell sizes for representing land use such as SAM-IM is driven by the need for resolution.
SAM-IM currently runs at grid cell sizes of 220 feet on a side, or approximately 1.11 acres. This appears to push ArcView to the very limit of its constraints on grid sizes and yet is insufficient for densely developed areas, such as downtowns. There are highly developed uses in downtowns, such as a 20-story building on a half-acre parcel, that literally "disappear" during the conversion from vector representation (the land use polygon cover) to grid. A mechanism to exchange higher resolution in densely developed areas, such as downtowns, for lower resolution in undeveloped deserts and forests, would be helpful.
Loss in Information During Database Conversions
The normal representation of urban land use is the conventional vector feature database of polygons – each representing "generalized" land use (that is, a "shape file"). Land use data is typically digitized this way and is maintained this way. Therefore, SAM-IM automatically converts land use data from its normal vector form to grids in order to perform the growth simulation. Once accomplished, the resulting "growth" grid is converted back to its vector form.
Land use modelers should expect, though, that there is a natural loss in information associated with land use through the conversion process. Anyone who has spent time looking at aerial photographs knows that you can see things in aerial photo-mosaics shot at sub-meter resolution (e.g., buildings) that you can not see in imagery shot at higher resolutions (e.g., 30-meter satellite images). The same holds true for grid representations – the simple conversion of a land use "shape file" to grid and back again will not give you the same polygon representation that you started with, except at very high resolutions.
The problems are:
During the next phase of work, a number of enhancements are planned. These are:
Current and Future Applications
SAM-IM is becoming an integral part of the forecasting process at MAG. The predecessor model SAM played a principal role in the development of the official population and employment projections by TAZ which were adopted by the MAG Regional Council in June 1997. Data being collected by the 2000 Census and the ongoing MAG Database Enhancement Project will be input to SAM-IM to prepare the next set of TAZ projections for Maricopa County during 2002.
In the meantime MAG is testing a number of land use development scenarios. For example, during the early part of 2000 MAG applied SAM-IM to produce alternative population and employment allocations for:
Because the model has proven to be both useful and configurable, next year MAG intends to provide SAM-IM, with appropriate manuals and training, to its member agencies, i.e. 24 cities and towns and the County, many of which already use ArcView. It is anticipated that SAM-IM will be used by these agencies to evaluate the impact of local development policies and plans. Some may choose to focus the model within their geographical boundaries and/or use smaller spatial units than TAZs. SAM-IM’s structure will allow users to tailor the modeling domain, grid-size, and aggregation of model output to local land use planning needs.
MAG’s experience with SAM over the last five years indicates that the model’s greatest asset may be its flexibility. It can be used to prepare socioeconomic forecasts, as well as to analyze alternative development patterns, perform what-if analyses, at varying spatial scales.
Due to concerns about the impacts of growth on infrastructure such as roads, utilities, and schools, and the overall quality of life, the Arizona legislature has recently passed "growing smarter" laws, requiring jurisdictions to develop new land use plans. This increased emphasis on accurate and potentially-binding plans, along with the looming possibility of growth boundaries, is sparking an interest in land use models. Fortunately, MAG and its member agencies are well-positioned to address future growth and other regional and local land use issues, thanks to SAM-IM.
Rita Walton
Information Services Manager
Maricopa Association of Governments
302 North 1st Avenue, Suite 300
Phoenix, AZ 85003
rwalton@mag.maricopa.gov