A Study of Floor Space Synthesizer Behavior. Kristen Andersen CPLN

Transcription

1 A Study of Floor Space Synthesizer Behavior Kristen Andersen CPLN

2 TABLE OF CONTENTS 1. INTRODUCTION PURPOSE OF STUDY CASE STUDY BACKGROUND DATA Inputs Comparison Map SYNTHESIZER STRUCTURE Scoring System Overview Scoring System and Match Coefficient Table Floor Space Inventory Table Parcel Table METHODS RESULTS AND ANALYSIS Preference to Actual Built Parcels Floor Space Inventory Chunk Sizes FAR Targets Match Values More on Calibration of the Match Coefficient Table FUTURE RESEARCH Data Availability ii

3 8.2 Additional Input Attributes Floor Space Inventory Improvements Alternative Calibration Approach Improvements to Floor Space Synthesizer Design CONCLUSION REFERENCES iii

4 LIST OF TABLES Table 1: Zoning Classifications Used as Floorspace Synthesizer Inputs... 6 Table 2: Ranges of Built Space Used to Assign Development Types to Built Parcels in Unincorporated Autauga County... 9 Table 3: Match Coefficient Table Table 4: Portion of Floor Space Inventory Table Table 5: Portion of Parcel Table Table 6: Floor Space Synthesizer Output Analysis Worksheet Table 7: Effect of Chunk Size Change on Proportions of Parcels with Actual and Assigned Space Types LIST OF FIGURES Fig. 1: Montgomery MPO Area... 3 Fig. 2: Actual and Synthesized Space Distribution in a Portion of the Study Area iv

5 1. INTRODUCTION Integrated land-use and transportation models have significant data needs, requiring baseyear information on population, land use, zoning, employment, housing, household income, travel accessibility, and floor space inventory and price. When data of sufficient quality or quantity cannot be found, a model s ability to generate usable results is compromised, and data weaknesses may derail a modeling project altogether. In particular, data on floor space inventory describing the quantity, type, price, and cost of development of built space found in a metropolitan area is often inconsistent with other regional data or not available at all. 1 Floor space synthesizers, which simulate a region s base-year built form, can help compensate for deficiencies in floor space data, allowing modeling projects to proceed. The contribution of floor space synthesizers to modeling efforts in Oregon, Oahu, and Sacramento, has been documented PURPOSE OF STUDY The purpose of this study is to analyze the behavior of the floor space synthesizer developed for the Sacramento project, devised by John Abraham, Research Fellow at the University of Calgary s Institute for Advanced Policy Research. In Sacramento, the floor space synthesizer was part of a full-fledged modeling effort, using 22 land use categories and 13 development types to represent land cover across four million acres and six counties. 3 In this study, the conditions represented by the floor space synthesizer were considerably less complex. Calibrating the synthesizer on a simplified scale made it 1 J.E. Abraham, T. Weidner, J. Gliebe, C. Willison, and J.D. Hunt, Three Methods for Synthesizing Base- Year Built Form for Integrated Land Use-Transport Models, Transportation Research Record: Journal of the Transportation Research Board 1902 (2005): Ibid., Ibid.,

6 possible to closely examine the synthesizer s response to various inputs and gauge the value of different data sets. This study required the synthesizer to be run numerous times, and since a relatively small area was considered, the computer could complete each test quickly. The findings from this study will help streamline calibration of the synthesizer in future applications. 3. CASE STUDY BACKGROUND The floor space synthesizer study is an outgrowth of a PECAS model development effort for the Montgomery (Alabama) Area Metropolitan Planning Organization (MPO), currently in its second year. 4 The Montgomery Area MPO directs transportation planning in portions of three counties in central Alabama: Autauga, Elmore, and Montgomery. Figure 1. In 2000, the MPO area had approximately 300,000 residents, and approximately two-thirds of this population lived in Montgomery, the state capital. As one of the first integrated modeling efforts in a small-to-medium sized metropolitan area, one of the project s goals is to determine whether integrated models with their extensive manpower and data requirements can be adapted to the needs of smaller regions. 4 Michael J. Clay, Kristen M. Andersen, Kim Mullins, Nicole Woerner, John Abraham and Doug Hunt. (submitted) Developing an Integrated Land Use and Transportation Modeling System in Small to Medium Sized MPOs: Experiences from the Montgomery, Alabama, Model Development Project. Publication decision pending in Transportation Research Record, Journal of the Transportation Research Board. 2

7 FIGURE 1 Montgomery MPO Area. In this study, the floorspace synthesizer will be tested on the Autauga County portion of the Montgomery Area MPO. This area s relatively small size and concentrated population simplifies the evaluation of synthesizer results, making it an ideal case study. Complete and reliable data on built space in Autauga County has been obtained, making it possible to compare synthesizer results with actual conditions. The study area consisting of approximately 123,000 acres contains a range of land use and housing types. Most of the study area is rural, and the vast majority of Autauga County s approximately 50,000 inhabitants reside within the city limits of Prattville (2006 population approximately 31,000). A small portion of the incorporated area of Millbrook, a neighboring municipality, extends into the study area. As well, a small 3

8 portion of Prattville extends into adjacent Elmore County. Prattville, about 13 miles north of Montgomery, is one of the fastest-growing cities in Alabama, with a 28% increase in population between 2000 and New residents are attracted by retail growth, a highly-ranked public school system, low crime rate, and relatively high levels of accessibility to Montgomery employment centers and amenities. PECAS space development module is a microsimulation model, which uses a random number generator to simulate the development on each unit of land represented by the model. Individually, the development simulated on each land unit should be of small importance to the overall model results, so relatively small geographic divisions of land must be used. 6 The study area is divided into 19,970 units, following actual parcel boundaries obtained from the Autauga County tax assessor. The floor space synthesizer can also process land use information with gridcell-based geographic divisions, as demonstrated in Oregon and Oahu. 7 This study used a parcel database structured according to parcel boundaries, because this format was adopted by the Montgomery modeling effort. Project managers wanted outputs to reflect actual political divisions, to make model results more believable and relevant to official and citizens. Autauga County s parcel database (taken from the three-county parcel database used in the Montgomery project) contains a wide range of parcel sizes, from.1 acre to 624 acres, with a mean parcel size of 6.2 acres. In Sacramento, where the floorspace synthesizer was applied to a parcel-based database, project managers limited and standardized the sizes of 5 U.S. Census. 6 Abraham, et. al, Three Methods, Ibid.,

9 land divisions. Using a parcel subdivision algorithm, parcels over five acres were divided into pseudoparcels no larger than five acres. This was not done in Montgomery because improvements to the PECAS model cause it to build space.1 acre at a time, regardless of the size of the parcel. 4. DATA 4.1 Inputs Data about attributes of parcels in the study area were used to influence the floor space synthesizer s space assignment decisions. Zoning classifications provided a key attribute used to guide the synthesizer. Due to geography and land development patterns, certain types of development are more suited to certain areas than others, and municipal zoning boundaries usually reflect this. Including zoning discourages the synthesizer from assigning type and quantity of space in a way that violates existing zoning regulations. Including zoning as an input helps the synthesizer replicate the clustering of development types that occur in the real world, as businesses take advantage of economies of agglomeration and advantageous traffic patterns and similar housing types cluster in order to maximize property values and avoid noxious land uses. Parcels processed by the floorspace synthesizer were associated with a set of ten zoning classifications (see Table 1). These categories were the same used by the Montgomery PECAS model, and the set was created by consolidating the myriad zoning categories used by individual municipalities in the three-county area into a uniform set applied to the entire region. (Only two municipalities have land within Autauga County, making the 5

10 zoning of parcels in the study area relatively uniform. However, this study shares the parcel attribute database used by Montgomery PECAS model, and uses the same set of zoning categories. Not all the categories included in the three-county model are represented in Autauga County.) Prattville and Millbrook officials provided zoning maps in digital format, and municipal zoning ordinances were studied to determine which consolidated category best matched the regulations applicable in a particular municipal zone. TABLE 1 Zoning Classifications Used as Floorspace Synthesizer Inputs R-1 Low Density Residential (minimum lot size > 1 acre) R-2 Med Density Residential (minimum lot size.25-1 acre) R-3 High Density Residential (minimum lot size <.25 acre) R-4 Mobile Homes- Rural (minimum lot size >.5 acre) R-5 Mobile Homes- Urban (minimum lot size <.5 acres) R-6 Multi-family Dwellings and Apartments O-1 Lower Density Office/Business O-2 Higher Density Office/Business I-1 Light Industrial/Manufacturing I-2 Heavy Industrial/Manufacturing PUD-2 Single and Multi-family Residential; Lower Density Office Other parcel attributes were used as inputs to the floor space synthesizer. The synthesizer steered development to parcels actually containing built space, and data from the Autauga County tax assessor s office was used to identify such parcels. Information about parcel numbers, boundaries, and area were also obtained from the assessor. Assigning parcels to 6

11 the correct TAZ used by the Montgomery Area MPO was an important step, since the synthesizer works one TAZ at a time, assigning the inventory of space for a particular zone before progressing to the next one. Though other types of spatial divisions (e.g. Census boundaries) could have been used by the synthesizer, TAZ geography matched the land divisions used by the Montgomery PECAS model. Since this study shares the parcel information used by the integrated model, TAZ boundaries are used by the synthesizer. Information about the quantity of space to be assigned within a particular TAZ was also input to the synthesizer. For this study, the synthesizer assigned space in four categories: single-family dwelling, apartment dwelling, mobile home dwelling, and non-residential space. Though this is an extreme simplification of real world conditions, using a condensed set of development types makes it possible to closely examine the synthesizer s behavior and its responsiveness to various inputs. Ratios were applied to Census housing and employment data to generate estimates of quantities of space by type in each TAZ (these ratios are discussed further in 5.3). 4.2 Comparison Map Information on the type and quantity of actual built space was needed to produce an official map of space development in the study area, which was used during calibration to facilitate the evaluation of synthesizer results. The amount of built space on each parcel in the study area was provided by the Autauga County tax assessor. Though this data was disaggregated by building, amounts were added to produce a single figure 7

12 representing the total square footage of built space for each parcel. Obtaining information on the type of built space in the study area was not as clear-cut. Ideally, each built parcel would have been assigned to one of the four development types based on the land use of the parcel. Land use data, provided by the Autauga County assessor was already part of the parcel database used by the Montgomery PECAS model. However, after examination of this information for parcels in Autauga County, significant weaknesses were found, making the data unsuitable to guide space type assignment. Though zones permitting a variety of residential development types (including single family dwellings, apartments, townhomes, patio homes, and mobile homes) exist in Prattville and Millbrook, the only residential land use type accounted for in Autauga County s parcel database was singlefamily residential. Using this dataset, parcels with multi-family housing or mobile homes would not have been able to be identified. In addition, a large number of parcels containing floor space (according to the assessor) were listed as Undeveloped in the land use data. The land use data is not as current as the assessor data, but this kind of mismatch occurred with unacceptable frequency. Instead, space types were assigned to built parcels based on applicable zoning. Mixed-use zoning applied to only one built parcel in the study area, and remaining zones could be easily associated with one of the four space types. However, this method presented its own difficulties. While zoning limits intensity and constrains where particular types of development can locate, zoning adjustments are common. Moreover, no zoning controls exist in unincorporated areas of Autauga County, making it impossible to use this technique to identify the type of development on a particular parcel. Using assessor data, 8

13 it was determined that 3058 developed parcels (of the 13,828 developed parcels in the study area) were located in unincorporated Autauga County. These parcels were assigned a space type according to the amount of space they contained (see Table 2). While based on some considerable assumptions, this method made it possible for the space type results generated by the synthesizer for rural Autauga County to have a point of comparison. TABLE 2 Ranges of Built Space Used to Assign Development Types to Built Parcels in Unincorporated Autauga County Amount of built space on parcel (in square feet) Space Type Assigned Mobile Home Dwelling Single Family Dwelling > 4001 Non-Residential Space 5. SYNTHESIZER STRUCTURE 5.1 Scoring System Overview The floor space synthesizer works through each TAZ in a region, examining various attributes of the parcels in a particular zone and evaluating each parcel s suitability for development. To influence the synthesizer s operation, rewards and penalties are set by the user to encourage or discourage the type or amount of space assigned to a particular parcel. Then the synthesizer uses an algorithm to assign a set of scores to each parcel, and space is distributed among parcels based on their score for a particular type of space. As parcels approach capacity they receive a penalty, preventing them from receiving additional space until all other parcels in the TAZ have a lower score for that space type. 9

14 Parcels ranked so low that they are not considered by the synthesizer before all space has been distributed are left vacant. The scoring algorithm works to replicate a real-world phenomenon: land competes for development, and each piece of land tends to attract the type and quantity of development to which it is best suited. 8 When the synthesizer is calibrated properly, space is assigned to parcels in a way that approximates reality. In order to assign space, the floor space synthesizer draws information from three separate tables. The synthesis software uses a Java database connectivity interface and structured query language to access the inputs that influence its operation. 5.2 Scoring System and Match Coefficient Table The information contained in the Match Coefficient table (Table 3) influences the score given to each parcel by the scoring algorithm. To change the likelihood that the synthesizer will assign space to parcels with a particular attribute, users modify the Match value associated with that attribute. Scores are increased by positive Match values, and decreased by negative Match values. If more than one attribute in the Match Coefficient table applies to a particular parcel, the Match values associated with applicable attributes are added together (e.g. In Table 3, if a Built parcel is located in zone R-1, its Match value is 0.) A penalty may wipe out the bonus conferred by possession of another attribute, and vice versa. 8 Ibid.,

15 TABLE 3 Match Coefficient Table PECASTYPE FIELDNAME FIELDVALUE FAR Target Match Single Family Dwelling Built/Vacant Built 0 5 Single Family Dwelling Built/Vacant Vacant Apartment Dwelling Built/Vacant Built 0 5 Apartment Dwelling Built/Vacant Vacant Mobile Home Dwelling Built/Vacant Built 0 5 Mobile Home Dwelling Built/Vacant Vacant Non-residential space Built/Vacant Built 0 5 Non-residential space Built/Vacant Vacant Apartment Dwelling Zoning Category PUD-2: Single and Multi-family Res; Lwr Density Office 1-5 Apartment Dwelling Zoning Category R-6: Multi-family Dwellings and Apartments 1-5 Mobile Home Dwelling Zoning Category R-1: Low Density Residential Mobile Home Dwelling Zoning Category R-5: Mobile Hms- Urban Non-residential space Zoning Category I-1: Light Industrial/Manufacturing 1 0 Non-residential space Zoning Category I-2: Heavy Industrial/Manufacturing Non-residential space Zoning Category O-2: Higher Density Office/Business Non-residential space Zoning Category PUD-2: Single and Multi-family Res; Lwr Density Office Non-residential space Zoning Category O-1: Lower Density Office/Business Single Family Dwelling Zoning Category R-1: Low Density Residential Single Family Dwelling Zoning Category R-2: Med Density Residential Single Family Dwelling Zoning Category R-3: High Density Residential Single Family Dwelling Zoning Category PUD-2: Single and Multi-family Res; Lwr Density Office 1-5 To influence the quantity of space assigned to parcels, users provide FAR Targets associated with particular attributes (e.g. the FAR Targets for non-residential space should be higher than those for single family dwellings, so that synthesized space will reflect the increased densities in which non-residential space is built). These values should approximate actual development densities in the area for which built space is being simulated. FARs are a widely available input, since maximum allowable FARs are published in municipal zoning ordinances, or actual FARs can be estimated by those with knowledge of the area. When assigning floor space, the synthesizer calculates a parcel s FAR the quantity of space assigned to the parcel divided by the size of the parcel. (This amount only a true Floor Area Ratio if both values are in the same unit; in this study, both values are in square feet). When the quantity of floor space assigned to a parcel exceeds 70% of its FAR Target, its score is reduced in the amount of: 11

16 (FAR-farTarget s * 0.7) / fartarget s * 3.0 where: fartarget s = the total of all FAR Targets applicable to a particular parcel (Like Match values, the effects of FAR Targets are cumulative if a particular parcel has more than one attribute included in the Match Coefficient table) Within a particular TAZ, the parcel deemed most suitable for a particular type of space will be given all of that space, until the parcel fills to the point where it is assigned a penalty. Smaller parcels reach capacity more quickly than larger parcels having the same FAR Target. As the parcel having the highest score for the space type being assigned approaches capacity, the FAR penalty acts to direct the assignment of space to parcels that would otherwise have lower scores. By distributing space to a greater number of parcels, the FAR Target component prevents the floorspace synthesizer from concentrating built space on a few parcels with the highest scores, while remaining parcels stay vacant. Once penalized, parcels will only receive more space if their scores for that particular space type exceed those of other parcels in the TAZ. In this way, FAR Targets ensure that the synthesizer replicates realistic development intensities. According to Abraham, Weidner, Gliebe, Willison, and Hunt, Development will show up at typical intensities on the best parcels before the not-quite-so-good parcels show any development. And very intense development will not show up until a shortage of 12

17 suitable land is evident in a zone. 9 Since their FAR Targets and Match values are effectively zero, parcels having none of the attributes that are part of the Match Coefficient table will not receive any space until the rest of the parcels in the TAZ are filled to capacity. As the synthesizer assigns space within a TAZ, the scores of parcels within the zone are constantly in flux. When the synthesizer attempts to assign space of a particular type within a TAZ, parcels assigned that space type earlier in the process are given a bonus to their score for that space type of.5. This increases the likelihood that parcels will continue to receive space of the same type, and that space will accumulate on parcels that have already been assigned space, instead of being assigned to vacant parcels (at least, until a large number of built parcels approach capacity). Conversely, if a parcel has previously been assigned a different space type, it receives a penalty of 5.0. However, if the penalized score for such a parcel still exceeds the scores of other parcels in the TAZ, its existing space type will be swapped with the conflicting type. The synthesizer assigns space in proportion to the total amount of space to be distributed within the TAZ (i.e. when the synthesizer is halfway through distributing the total amount of space alotted to a particular TAZ within the Floorspace Inventory table, the synthesizer will have assigned 50% of each space type.) Parcels with identical scores within a particular TAZ are assigned space in numerical order according to their parcel number, producing a distribution of space that appears random. This ensures that parcels 9 Ibid.,

18 of a certain space type are not clustered together in unrealistic ways, and that the outcome appears random. 5.3 Floor Space Inventory Table As the scoring system works, it draws information about the amount and type of space it is distributing from the Floorspace Inventory table (see Table 4). Since synthesizer users are not likely to have access to quality data of this kind, a method of estimating these amounts was developed. Census data was used to determine the number of dwellings of each type in a zone, as well as the number of employees working in the TAZ. For each dwelling or employee, a certain amount of square feet was allocated to the TAZ for different types of development. While the amount of space allocated to each space type varied (according to the number of dwellings and employees in a particular zone), overall approximately 60% of square footage in the Floor Space Inventory table is designated for single-family dwellings, 14% for apartments, 8% for mobile homes, and 19% for non-residential space. 14

19 TABLE 4 Portion of Floor Space Inventory Table TAZ Commodity Quantity Chunk Size 269 Single Family Dwelling Apartment Dwelling Mobile Home Dwelling Non-residential space Single Family Dwelling Apartment Dwelling Mobile Home Dwelling Non-residential space Single Family Dwelling Apartment Dwelling Mobile Home Dwelling Non-residential space Single Family Dwelling Apartment Dwelling Mobile Home Dwelling Non-residential space By manipulating values in the Chunk Size field, users change the square footage of built space the synthesizer allocates to parcels at a time. The use of larger space chunks allows the synthesizer to run more quickly, but smaller units of space increase the synthesizer s ability to assign space to a parcel before exceeding the parcel s FAR Target. 5.4 Parcel Table Information about the attributes of individual parcels is contained in the Parcel table (see Table 5; See 4.1 for more information about the data stored in the Parcel table.) Field names referring to specific attributes must appear in both the Parcel and Match Coefficient tables to be considered by the floor space synthesizer. The synthesizer also 15

20 draws information on parcel size used to calculate FARs from this file. In this study, the Parcel table is by far the largest source of synthesizer inputs, containing 19,970 records one for each parcel in the study area. The Parcel table used for this project is an abbreviated version of the one storing the inputs to the space development module of the three-county Montgomery PECAS model. TABLE 5 Portion of Parcel Table Parcel Number TAZ City Zoning Category Built/ Vacant Parcel Area (sq ft) Actual Built Area (sq ft) Assigned Built Area (sq ft) Actual Type of Built Space Space Type Assigned by Synthesizer Prattville R-2: Med Density Residential Built Single Family Dwelling Non-residential space Prattville R-2: Med Density Residential Built Single Family Dwelling Apartment Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Non-residential space Prattville R-2: Med Density Residential Vacant None None Prattville R-2: Med Density Residential Vacant None None Prattville R-2: Med Density Residential Built Single Family Dwelling Single Family Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Single Family Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Apartment Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Non-residential space Prattville R-1: Low Density Residential Vacant None None Prattville R-3: High Density Residential Built Apartment Dwelling None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville R-3: High Density Residential Built Apartment Dwelling None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville R-3: High Density Residential Built Apartment Dwelling None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville R-3: High Density Residential Built Apartment Dwelling None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville R-3: High Density Residential Built Apartment Dwelling None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville R-2: Med Density Residential Built Single Family Dwelling Non-residential space Prattville R-2: Med Density Residential Built Single Family Dwelling Apartment Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Non-residential space Prattville R-2: Med Density Residential Built Single Family Dwelling Non-residential space Prattville O-2: Higher Density Office/Business Built Non-residential space None Prattville R-6: Multi-family Dwellings and Apartments Built Apartment Dwelling Non-residential space None None Vacant None None Prattville R-2: Med Density Residential Built Single Family Dwelling Mobile Home Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Mobile Home Dwelling Prattville R-2: Med Density Residential Built Single Family Dwelling Mobile Home Dwelling During the tests run as part of this study, the Match Coefficient table was changed to represent various combinations of parcel attributes, meaning that different Parcel table fields were used at different times by the floor space synthesizer. Once a synthesizer run is complete, outputs are written to the Parcel table. Data about the amount and type of development located on actual parcels are also stored in the Parcel table, facilitating evaluation of the validity of synthesizer results. 16

21 6. METHODS Calibration efforts focused on three tasks: 1) directing synthesized development to actual built parcels; 2) directing the correct space type to as many of these developed parcels as possible; and 3) directing synthesized built space to developed parcels in amounts resembling actual quantities. FIGURE 2 Actual and Synthesized Space Distribution in a Portion of the Study Area. During this study, the floor space synthesizer was run over 100 times, using different inputs. After each run, the table of outputs was imported into ArcGIS and joined with the Parcel table, allowing results to be viewed spatially (see Figure 2, showing central 17

22 Prattville). The synthesizer-generated map was compared with the official map showing actual space development in the study area. This technique provided a convenient though superficial view of how input changes had been processed by the synthesizer. A scoring system was developed to measure how well the amount and type of space assigned to parcels corresponded to development shown on the official map. When a more intensive, quantitative analysis was required, results were transferred to an Excel spreadsheet providing a detailed breakdown of the square footage and number of parcels assigned to different space types (see Table 6). After the results were analyzed and recorded, inputs were adjusted accordingly, and the synthesizer run again. Since fields in the Match Coefficient table have the most influence on the synthesizer s scoring system, this file was modified extensively during the calibration process. However, changes to the Floorspace Inventory table were made as well. 18

23 TABLE 6 Floor Space Synthesizer Output Analysis Worksheet Square Feet Actual Proportions Assigned Assigned Assigned Assigned Assigned Assigned S A M N V TOTAL S 19,461, % 5,770, , , , ,149,445 A 4,798, % 717, ,734 51, , ,018,053 M 1,190, % 315,000 68,500 57,233 23, ,833 N 6,976, % 325,500 37,766 8,771 2,039, ,411,597 V 0 0.0% 4,569,229 1,427, , , ,244,678 TOTAL 32,426, % 11,698,184 2,503,019 1,362,573 3,723, ,287,607 Assigned 60.7% 13.0% 7.1% 19.3% 0.0% 100.0% Proportions Parcels Actual Output Score: 3924 Proportions Assigned Assigned Assigned Assigned Assigned Assigned S A M N V TOTAL S 9, % 1,579 1, , ,520 A 2, % 1, ,455 M % N % V 6, % ,020 6,142 TOTAL 19, % 4,219 2,015 1,030 6,363 6,343 19,970 Assigned Proportions 21.1% 10.1% 5.2% 31.9% 31.8% 100.0% 7. RESULTS AND ANALYSIS (Note: in this section, the term assigned refers to the synthesizer-generated simulation of built space, and actual refers to the quantity and type of built space currently existing in the study area according to assessor data used for comparison purposes.) During this study, the synthesizer was never able to simulate the quantity and type of built space on parcels in the study area in a reasonably accurate way. Table 3 shows the inputs that produced the best development type results, Table 6 shows the quantitative results used to objectively measure the success of the test, and Figure 2 shows these results displayed spatially, for a portion of the study area. 19

24 7.1 Preference to Actual Built Parcels In order to guide development toward parcels within urbanized Prattville, it was necessary to include an input in the Match Coefficient table causing the synthesizer to give preference to parcels with actual built space. Without this component, the synthesizer could not simulate the disparity in number of developed parcels that exists between urban and rural areas. The addition of this component ensured a realistic distribution of developed parcels during all the tests that followed, though space was assigned to fewer parcels after the addition of other attributes to the Match Coefficient table. Giving preference to parcels with actual built space ensures that only a few actual vacant parcels are assigned space by the synthesizer. However, this component also caused actual vacant parcels that were given space to receive a large amount of it, relative to the amount of space given to parcels that, according to assessor data, have other actual space types. In one test, the mean square footage assigned to an actual built parcel was 2345, while the mean square footage assigned to a parcel identified as vacant by assessor data was 59,383 (the Match Coefficient table for this test is shown in Table 3, and outputs for the test are shown in Table 6). This phenomenon was apparent in all tests with inputs giving preference to parcels that have built space according to assessor data. 7.2 Floor Space Inventory The proportions of assigned square footage of each type are strongly influenced by the space type proportions used in the Floor Space Inventory (these are discussed in 5.3). 20

25 Though developed using ratios, the overall proportions of each type of space distributed by the synthesizer correspond closely with actual proportions of built space of each type in study area. In all tests, proportions of each type of assigned space varied from Floor Space Inventory proportions by only a few percentage points. In addition, the total amount of square feet assigned during each run remained consistent, despite changes to a variety of inputs. In every test, the amount of square feet assigned hewed closely to the total amount listed in the Floor Space Inventory, another indication of this table s influence. However, the total amount of square feet listed in the Floor Space Inventory table was significantly less than the actual built space in the study area (as can be seen in Table 6). According to the assessor, ~ 35.5 million square feet of built space exists in the study area. Yet in every test, only ~ 20 million square feet was assigned. Since none of the changes made to synthesizer inputs during this study had a significant effect on the amount of space assigned, it appears that space quantities listed in the Floor Space Inventory must be increased for these results to better resemble actual conditions. 7.3 Chunk Sizes Several runs tested the effects of modifying the size of space chunks assigned by the synthesizer. Changing the residential chunk size from 500 to 50 square feet, and the nonresidential chunk size from 150 to 50 square feet dramatically reduced the Output Score (this evaluation tool not to be confused with the score assigned to parcels during space assignment is discussed in 6), meaning that the amount and type of space assigned to 21

26 parcels decreased in similarity to development shown on the official map. As discussed in 5.3, reducing chunk sizes makes it easier for the synthesizer to stop assigning space to a parcel before exceeding the FAR Target, which spreads space to a larger number of parcels. Though distribution of space to more parcels causes an increase in space type matches to occur between actual and assigned parcels, reducing chunk sizes seems more likely to result in space type mismatches, as evidenced by the lower Output Score. Every parcel without a matching attribute in the Match Coefficient table has a default FAR Target of zero. Therefore, any space assigned to these parcels exceeds their FAR Target, and it is hypothesized that the drop in Output Score following a reduction in chunk size results from this phenomenon. As seen in Table 7, the decrease in percentage of vacant parcels following the chunk size change shows how total floor space was distributed to a larger number of parcels. In particular (and for unknown reasons), the chunk size change spread Mobile Home space to more parcels (which can also be seen in Table 7). In the study area, only approximately 5% of parcels contain mobile homes, but the chunk size change increased the proportion of parcels assigned mobile home space from 7.6% to nearly 27%, meaning that the mobile home space being assigned to parcels does not reflect realistic intensities. Reducing chunk sizes increased the density of single-family dwellings, since more space of this type was concentrated onto fewer parcels. 22

27 TABLE 7 Effect of Chunk Size Change on Proportions of Parcels with Actual and Assigned Space Types Output Score Single Family Dwelling Apartment Dwelling Mobile Home Dwelling Non-residential Space Vacant Total Actual proportions 47.7% 12.3% 4.9% 4.4% 30.8% 100.0% Assigned proportions: residential chunk size 500; non-residential chunk size 150 Assigned proportions: all chunk sizes 50 10, % 7.2% 7.6% 13.3% 44.9% 100.0% 5, % 10.7% 26.9% 10.0% 34.3% 100.0% 7.4 FAR Targets Many tests involved the manipulation of FAR Targets in order to change the proportion of parcels assigned a particular type of space relative to another type. However, the ability of FAR Targets to control these proportions was limited, since the Floor Space Inventory also influences the proportions of space in TAZs assigned to a particular type. In one test, FAR Targets for Mobile Home Dwellings were increased (from.5 to 1.5), in an effort to concentrate this type of development on fewer parcels. Though this adjustment should have allowed parcels assigned Mobile Home space to receive more square footage before the FAR penalty was activated, this was not the case. However, a test that reduced the FAR Target for Single-Family Dwellings, in order to distribute this type of development to more parcels, had the desired effect. 7.5 Match Values Like decreasing chunk sizes and FAR Targets, reducing Match values also spread assigned space to a greater number of parcels. Reducing Match values decreases the relative attractiveness of any particular parcel, distributing space more evenly among 23

28 parcels in a TAZ. Despite this, modification of Match values could not achieve a realistic distribution of assigned space according to type. In terms of the number of parcels assigned space, the synthesizer was much more responsive to changes in FAR Targets than Match values. 7.6 More on Calibration of the Match Coefficient Table As discussed in 7.4 and 7.5, decreasing FAR Targets and Match values spread development to a larger number of parcels. Lower FAR Targets allowed individual parcels to reach capacity quicker than parcels with higher FAR Targets, and the total inventory of space was assigned to a larger number of parcels. The use of negative FAR Targets and Match values continued this trend. Obviously, negative FARs are not based in reality, and are of limited value when legitimate FAR Targets are being used as synthesizer inputs. During the calibration process it was noticed that very large or small FAR Targets and Match values did not produce extreme results (e.g. an FAR Target of 100 did not concentrate all space on a single parcel in a TAZ). As FAR Targets and Match values are increased, the synthesizer becomes less responsive to the change (i.e. most significant effects to the scoring system are caused by the smallest adjustments to FAR Targets and Match values). Also, in synthesizer runs when a relatively small adjustment to FAR Targets and Match values had no effect on the outcome, it was observed that larger changes to these numbers also had no effect. 24

29 8. FUTURE RESEARCH Though a reasonable simulation of built space in Autauga County was not produced during this study, much was learned about the behavior of the floor space synthesizer that will improve the accuracy of synthesizer results when further tests are done. 8.1 Data Availability Throughout this study, the general availability of data used as synthesizer inputs was kept in mind. Since the most common use of the synthesizer is to compensate for data weaknesses, the synthesizer was calibrated using data that can be obtained by many users. As discussed in 7.1, the Match Coefficient table must include an input to increase the number of parcels developed in urban areas relative to rural areas. This study used assessor data indicating parcels containing built space. However, this data may not be available in regions needing the synthesizer to compensate for a lack of floor space data. In these regions, aerial photography could be used to estimate areas of developed parcels. Similarly, identifying parcels as urban or rural in the Parcel database, with an input in the Match Coefficient table giving preference to urban parcels, would also direct development to urban areas. Users can apply their own knowledge of the regions where they live and work to improve synthesizer calibration. They may be able to create new datasets for input to the synthesizer, beyond what is available from official sources. Inevitably, users 25

30 expectations of the results a synthesizer should produce guide them to modify inputs in a certain way. The floor space synthesizer enables users to apply information from any source about the type and quantity of built space in an area to individual parcels in a database, preparing a data set that can be utilized by a land-use and transportation model. 8.2 Additional Input Attributes In this study, reducing the number of inputs considered by the floor space synthesizer facilitated a deeper analysis of the synthesizer s behavior. Numerous other attributes could be added to the Match Coefficient table in order to guide space assignment, varying according to the data that can be obtained for a particular region. Distance-to-roadway is an input that is widely available, and this data would cause the synthesizer to consider a particular parcel s distance from major roads, better simulating actual space development patterns (e.g. encouraging the assignment of non-residential space within a certain distance from a highway or discouraging the assignment of residential space there). Data on the age of structures can be a valuable input, allowing integrated models using synthesizer results to simulate the cycle of redevelopment as accurately as possible. 10 The lack of utility infrastructure on a particular parcel could be used to discourage space assignment. Development could be diverted from wetlands, floodplains, parks, schools, airports, and areas of excessive slope unless the synthesizer calculated a demand for space on such parcels strong enough to exceed assigned penalties. In areas such as rivers and lakes, development could be prohibited altogether. 10 Ibid.,

31 Data with known weaknesses such as Autauga County s land use dataset could be used as inputs, but given less importance relative to other attributes. If quality floor space data is only available for a portion of a region, the Match Coefficient table could be used to preassign floor space to parcels in this area, with the synthesizer distributing space only in remaining parcels. 8.3 Floor Space Inventory Improvements This study has established that, within particular TAZs, it is impossible to achieve correct space distribution by type using current Floor Space Inventory proportions. New methods of apportioning space by type within TAZs must be developed, for the synthesizer to be successful. Another problem encountered during this study was an insufficient total amount of floor space assigned to the study area, because the amount of space in the Floor Space Inventory was too low relative to assessor data. Those in need of a Floor Space synthesizer probably would not have access to correct square foot totals available for input to the synthesizer, as were available for the study area. However, users could run the synthesizer using preliminary estimates of floor space sq footage to be assigned, adjusting coefficients until space was assigned to parcels in reasonable densities (meaning that FARs of space on individual parcels were appropriate). Once this has been accomplished, synthesizer calibrators could work on finding the total amount of square footage that needed to be input to the Floor Space Inventory to get space distributed to 27

32 the right number of parcels during a synthesizer run. In this way, local knowledge of floor space in an area is used to estimate an unknown quantity. 8.4 Alternative Calibration Approach A similar process could be used to guide future floor space synthesizer research. In this study, efforts focused on achieving a realistic distribution of space in the study area according to quantity and type assigned to parcels, taking the quantities of space in the FI for granted. Future calibration efforts could take an alternate approach, focusing on modifying coefficients until the synthesizer assigns space in reasonable densities/fars, then adjusting the square footage in the Floor Space Inventory so that space is distributed to the right number of parcels (according to assessor data). 8.5 Improvements to Floor Space Synthesizer Design Changes could be made to the design of the synthesizer itself that may increase the validity of results and ease of calibration efforts. Currently, the synthesizer is not responsive to numeric fields in the Parcel table, meaning it cannot assign space based on whether a parcel attribute in the Match Coefficient table fits in a specific numerical range. To allow numerical attributes (e.g. distance-to-roadway and percentage of slope) to be considered by the synthesizer, users must first create a separate field in the Parcel table to identify parcels fitting into a specific range. Adding the ability to process numerical fields would make it easier for a wider variety of physical attributes to be used as synthesizer inputs. 28

33 The synthesizer also lacks a mechanism favoring assignment of the same space type to neighboring parcels, so simulated floor space fails to reflect the clustering of like development types existing in real life. The score applied to individual parcels by the scoring algorithm used to determine the quantity and type of space assigned to parcels could be provided by the synthesizer as an output. This would make it possible for users to compare scores among parcels within a TAZ, facilitating better adjustment of inputs. In addition, increased knowledge about the synthesizer s operation could be gained through analysis of the scores. 9. CONCLUSION In this study, much was learned about the behavior of Abraham s floor space synthesizer, and many issues were raised that merit further investigation. The large number of tests completed, coupled with the limited number of inputs and relatively small study area, facilitated in-depth study of synthesizer results. Using inputs that are widely available, I saw how the synthesizer functioned when used in an actual region. The Output Score developed during this study provided a convenient, objective way to gauge the success of tests, and will help future users of the synthesizer. Many of the inconsistencies between actual and assigned space that were encountered during this study could be traced back to the Floor Space Inventory. I am optimistic that improvements to this and other components will greatly improve the ease and accuracy of synthesizer calibration, and increase the usefulness of this tool. 29

34 REFERENCES Abraham, J.E., T. Weidner, J. Gliebe, C. Willison, and J.D. Hunt. Three Methods for Synthesizing Base-Year Built Form for Integrated Land Use-Transport Models. Transportation Research Record: Journal of the Transportation Research Board 1902 (2005): Clay, Michael J., Kristen M. Andersen, Kim Mullins, Nicole Woerner, John Abraham and Doug Hunt. (submitted) Developing an Integrated Land Use and Transportation Modeling System in Small to Medium Sized MPOs: Experiences from the Montgomery, Alabama, Model Development Project. Publication decision pending in Transportation Research Record, Journal of the Transportation Research Board. 30