This item is only available as the following downloads:
1 EVALUATING ACCESSIBI LITY AND TRAVEL COST AS SUITABILITY COMPONENTS IN THE AL LOCATION OF LAND USE, A CASE STU DY OF IDENTIFYING LA ND FOR AFFORDABLE HO USING IN THREE COUNTIES IN FLORIDA By ABDULNASER A. A. ARA FAT A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011
2 2011 Abdulnaser A. A. Arafat
3 To my wife and children
4 ACKNOWLEDGMENTS I thank the faculty and colleagues in the Department for Urban and Regional Planning and the Shimberg Center for Affordable Housing Studies for their continuous support during my studies. I would like to extend my thanks to my committe e members for their academic and professional support. My gratitude and thanks for my committee chair Prof. Ruth Steiner and my co chair Prof. Paul Zwick for their guidance during the research and writing stages and for their continuous support for me acad emically, professionally and financially. I thank Prof. Ilir Bejleri for his professional support especially his guidance in the automation of GIS and the creation of tools using programming. I also thank Prof. Sivaramakrishnan Srinivasan and I appreciate his guidance in my di ssertation which was very helpful in the statistical and transportation modeling. ell, the interim Director of the Shimberg Center for his support and guidance for me in the affordable housing research. I also acknowledge my colleague Elizabeth Thompson for helping me in editing my work. I also thank the Palestinian Facul ty Development Program supported by the Open Society Institute, USAID, and AMIDEAST for sponsoring me in my Ph.D. program and for their finan cial support for me during my four years of study. Finally, I thank Birzeit University in my home country. I extend my thanks and gratitude to Prof. Kamal Abdul Fattah and my colleges in the Geography and Civil Engineering Departments for their support.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 10 LIST OF ABBR EVIATIONS ................................ ................................ ........................... 15 ABSTRACT ................................ ................................ ................................ ................... 17 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 19 Research Purpose ................................ ................................ ................................ .. 19 Dissertation Objectives ................................ ................................ ........................... 23 Research Questions ................................ ................................ ............................... 24 What is The Contribution of This Research? ................................ .......................... 24 2 LITERATURE R EVIEW ................................ ................................ .......................... 27 Evaluation of Location Choice Models ................................ ................................ .... 27 Suitability Models ................................ ................................ ................................ .... 33 Incorporating Transportation Variables in Suitability Modeling ............................... 42 Evaluating Methods of Estimating Accessibility ................................ ................ 44 Evaluating Distance Measurement Methodologies ................................ ........... 48 Evaluating Literature on the Impact of Land Use on Transportation ................. 49 Advanced Methods of Capturing Travel Behav ior ................................ ............ 57 Evaluating Methods for Predicting Travel Cost ................................ ................. 61 Using Conflict Identification Strategies to Identify the Conflict between Transportation and Land Use ................................ ................................ .............. 62 Building Allocation Scenarios ................................ ................................ .................. 63 3 RESEARCH METHODOLOGY ................................ ................................ ............... 73 Study Design: ................................ ................................ ................................ ......... 73 Model Structure ................................ ................................ ................................ ...... 75 Computer Req uirement and Data Sources ................................ ............................. 76 Level of Analysis and Selection of Research Areal Units ................................ ........ 77 Introducing Transportation Variables as Suitability Surfaces: ................................ 81 Investigating Neighborhood Accessibility as a Suitability Surface: ................... 82 Generating Neighborhood Acce ssibility as a Suitability Surface ...................... 83 Distance proximity s urface ................................ ................................ ......... 83 Opportunity s urfaces ................................ ................................ .................. 84
6 Gravity and opportunity d istanc e s urfaces ................................ ................. 84 Investigating Travel Cost ................................ ................................ .................. 86 Travel survey d ata ................................ ................................ ..................... 86 Spatial i n terpolation ................................ ................................ ................... 87 Statistical a pproach ................................ ................................ .................... 88 Research Automation Tools ................................ ................................ .................... 92 Th e A4 Suitabil i ty Tool ................................ ................................ ..................... 93 The A4 Community Values Program ................................ ................................ 95 The A4 Layer Weighting Tool ................................ ................................ ........... 96 Affordable Housing Opportunity Surfaces ................................ ............................... 97 Introducing Housing Cost as a Preference Surface ................................ .......... 97 Introducing Transit Access as a Preference Surface ................................ ........ 98 Downstream stations s core ................................ ................................ ........ 99 Es timating network distance/t ime ................................ ............................... 99 Creating transit accessibility suitability surface ................................ ........ 100 Automatic Allocation for Affordable Housing ................................ ......................... 100 The A4 Allocation Tool ................................ ................................ ................... 100 Buil d ing Affordable Housing Scenario ................................ ............................ 101 Model Validation Methods ................................ ................................ ..................... 102 4 CHOICE OF ANALYSIS UNITS ................................ ................................ ............ 115 The Importance of Defining the Areal Unit ................................ ............................ 115 Entropy Test ................................ ................................ ................................ ......... 119 Density and Connectivity Tests ................................ ................................ ............. 120 Shape of Neighborhood Test ................................ ................................ ................ 121 Reducing the Modifiable Areal Unit Problem ................................ ........................ 122 5 ESTIMATING ACCESSIBILTY AND TRAVEL COST ................................ ........... 135 Neighborhood Accessibility ................................ ................................ ................... 135 Opportunity Access Measures ................................ ................................ ........ 136 Combined Opportunity Distance Access Measures ................................ ....... 137 Estimating and Predicting Travel Cost ................................ ................................ .. 138 Estimating Travel Cost by Spatial Interpolation ................................ .............. 139 Predicting Travel Cost from Location and Urban Form Characteristics .......... 141 Predicting travel cost by ordinary least s quares ................................ ....... 142 Predicting travel cost by geographically weighted r egression .................. 144 Comparison Between GWR and OLS Results ................................ ............... 145 6 INTR ODUCING SUITABILTY A UTOMATION TOOLS ................................ ......... 156 Suitability Assignment Tool ................................ ................................ ................... 158 Overlay and Weighting Tools ................................ ................................ ................ 160 Allocation Tools ................................ ................................ ................................ .... 162
7 7 AFFORDABLE HOUSING A LLOCATION ................................ ............................. 179 Generating the Affordable Housing Opportunity Surface ................................ ...... 179 Physical and Neighborhood Characteristic Preference Surface ..................... 180 Rent Preference Surface ................................ ................................ ................ 180 Travel Cost Preference Surface ................................ ................................ ..... 181 Transit Access Preference Surface ................................ ................................ 182 Access Rent Driving Transit Opportunity Surface (ARDT) ............................. 182 Refined ARDT Opportunity Surface ................................ ............................... 183 Impact of Travel Cost and Transit Accessibility on Affordable Housing Opportunity ................................ ................................ ................................ ........ 184 Affordable Housing Allocation ................................ ................................ ............... 189 Creating the Combine Grid for the Allocation ................................ ................. 191 Under utilized density g rid ................................ ................................ ......... 191 Livability g rids ................................ ................................ .......................... 192 Land c haracteristics ................................ ................................ ................. 192 Proximity g rids ................................ ................................ ......................... 193 Policy g rids ................................ ................................ ............................... 194 Zoning g rids ................................ ................................ ............................. 195 Compact Development Scenarios ................................ ................................ ........ 195 8 CONCLUSION, RECOMMEN DATIONS AND LIMITATI ONS ............................... 231 Accessibility and Travel Cost as Suitability Surfaces in an Affordable Housing Suitability Model ................................ ................................ ................................ 231 The Impact of Travel Cost and Transit Accessibility on the Allocation and Preservation of Affordable Housing Sites ................................ .......................... 234 Incorporating Multi modal Transportation and Sprawl Conceptualization Metrics in Allocating Land for Affordable Housing? ................................ ........................ 236 Introducing Parcel Level Analysis for Affordable Housing Sites ............................ 239 APPENDIX A THE A4 SUITABILITY TOOL SOU RCE CODE ................................ .................... 242 B WEIGHTING TOOLS ................................ ................................ ............................ 247 C ALLOCATION TOOLS ................................ ................................ .......................... 262 LIST OF REFERENCES ................................ ................................ ............................. 276 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 284
8 LIST OF TABLES Table page 2 1 Scale for pair wise comparison ................................ ................................ ........... 66 2 2 Preference value descriptions. ................................ ................................ ........... 66 2 3 Conflict score matrix (AHS goal 1 and travel cost). ................................ ............ 66 2 4 Variables used in accessibility equations. ................................ .......................... 67 2 5 Accessibility matrix (Merits and limitations for different ac cessibility estimation methods) ................................ ................................ ........................... 68 2 6 Elasticity values for VMT ................................ ................................ .................... 68 2 7 Trip length associated with residential parcels ................................ ................... 69 2 8 Travel cost matrix (Variables) ................................ ................................ ............. 70 2 9 Travel cost matrix (Merits and limitations). ................................ ......................... 71 3 1 Data sources ................................ ................................ ................................ .... 103 3 2 Percentage Trips by purpose ................................ ................................ ............ 103 4 1 Change in entropy mean and standard deviation ................................ ............. 125 4 2 Change in entropy mean and standard deviation for three counties ................. 125 4 3 Change in density mean and standard deviation ................................ .............. 125 4 4 Change in connectivity mean and standard deviation ................................ ...... 126 5 1 Explanatory variable coefficients and their significance using OLS .................. 149 5 2 R 2 values for OLS and GWR ................................ ................................ ............ 149 7 1 AR and ARDT equivalent categories ................................ ................................ 198 7 2 Tabulated total acre in each AR category ................................ ......................... 198 7 3 Zonal statistics for distance to CBD ................................ ................................ .. 199 7 4 Zonal statistics for distance to major activity centers ................................ ........ 199 7 5 Zonal statistics for density of surrounding ................................ ........................ 200
9 7 6 Zonal statistics for entropy of surrounding ................................ ........................ 200 7 7 Collective measurements for Duval County ................................ ...................... 201 7 8 Collective measurements for Orange County ................................ ................... 201 7 9 Collective measurements for Pinellas County ................................ .................. 201 7 10 Duval scenario table ................................ ................................ ......................... 202 7 11 Orange scenario table ................................ ................................ ...................... 203 7 12 Pinellas scenario table ................................ ................................ ...................... 204 7 13 Frequently used land uses that are suitable for affordable housing .................. 205 7 14 Not frequently used land uses that are suitable for affordable housing ............ 206
10 LIST OF FIGURES Figure page 2 1 Pair wise comparison matrix. ................................ ................................ .............. 72 3 1 Structure for the co nflict / opportunity process ................................ .................. 104 3 2 Variable size neighborhoods around the parcel ................................ ............... 104 3 3 Euclidean proximity model ................................ ................................ ................ 105 3 4 Opportunity model ................................ ................................ ............................ 106 3 5 Tr ip end points in Duval County ................................ ................................ ........ 107 3 6 Trips by purpose ................................ ................................ ............................... 107 3 7 Percentage of trips categorized by purpose and trip length .............................. 108 3 8 Local outlier identification tool ................................ ................................ ........... 108 3 9 Example local outlier identification for Orange County ................................ ..... 109 3 10 Cross validation of errors for interpolating work t rips in Orange County. .......... 109 3 11 Iterative model used to generate travel miles surfaces ................................ ..... 110 3 12 Estimation of travel miles and travel cost ................................ ......................... 110 3 13 Daily travel miles surface for Orange County ................................ ................... 111 3 14 Decreasing suitability indexing. ................................ ................................ ........ 111 3 15 Increasing suitability indexing. ................................ ................................ .......... 112 3 16 Manhattan buffers arround transit stops ................................ ........................... 112 3 17 Manhattan distance raster to transit stops ................................ ........................ 113 3 18 Transit access suitabilty surface ................................ ................................ ....... 113 3 19 Planning by Table tool ................................ ................................ ...................... 114 4 1 Neighborhood definition according t o t he Steiner and Srinivasan model ......... 127 4 2 Mean of entropy value change that corresponds to 0.5 mile change in unit size for Orange County ................................ ................................ ..................... 127
11 4 3 Standard deviation of entropy value change that corresponds to 0.5 mile change in unit size for Orange Count y ................................ ............................. 128 4 4 Mean of entropy value change that corresponds to 0.5 mile change in unit size for three counties ................................ ................................ ...................... 128 4 5 Standard deviation of entropy value change that corresponds to 0.5 mile change in unit size for three counties ................................ ............................... 129 4 6 Land use mix entropy based on 2.5x2.5 mile areal unit for Orange County ..... 129 4 7 Mean of density value change that corresponds to 0.5 mile change in unit size. ................................ ................................ ................................ .................. 130 4 8 Standard deviation of density value change that corresponds to 0.5 mile change in unit size ................................ ................................ ............................ 130 4 9 Density surface for Hillsborough County based on 2.5x2.5 mile areal unit ....... 131 4 10 Mean of connectivity value change that corresponds to 0.5 mile change in unit size ................................ ................................ ................................ ............ 131 4 11 Standard deviation of connectivity value change that corresponds to 0.5 mile change in unit size ................................ ................................ ............................ 132 4 12 Connectivity surface for Duval County based on a 2.5 x 2.5 mile areal unit ..... 133 4 13 Shapes for the tested areal units ................................ ................................ ...... 133 4 14 Land use mix entropy surface for Hillsborough County based on 2.5 mile diamond shaped areal unit ................................ ................................ ............... 134 4 15 Streets and land use mix entropy overlay surface for Hillsborough County based on 2.5 mile diamond shaped areal unit. ................................ ................. 134 5 1 Proximity access measures Euclidean distance ................................ .............. 150 5 2 Proximity access measures network distance ................................ ................. 150 5 3 Euclidean buffer opportunity access ................................ ................................ 151 5 4 Network service area opportunity access ................................ ......................... 151 5 5 Combine opportunity distance access based on Euclidean distance ............... 152 5 6 Combine opportunity distance access based on network distance .................. 152 5 7 ArcGIS cross val idation chart for inverse distance weighted interpolation using work trip length as the interpolation field ................................ ................. 153
12 5 8 Travel cost generated by spatial interpolation ................................ .................. 153 5 9 Travel cost smoothened within walking distance ................................ .............. 154 5 10 Prediction error as a percentage of travel cost ................................ ................. 154 5 11 Reclassification of the GWR density coefficient. ................................ ............... 155 5 12 Reclassification of the GWR connectivity coefficient ................................ ........ 155 6 1 A4 Suitability tool interface ................................ ................................ ............... 167 6 2 LUCIS model illustrating the economic suitability of singl e family residential land use to retail and shopping opportunities ................................ ................... 167 6 3 Integration of the A4 Suitability tool in a LUCIS model ................................ ..... 168 6 4 Sample zonal statistics attribute table ................................ .............................. 169 6 5 Output table of A4 Suitability tool ................................ ................................ ...... 169 6 6 Proximity to shopping centers in Duval County reclassified according to the average and standard deviation of the distance to multifamily residential parcels ................................ ................................ ................................ .............. 170 6 7 The interface of the A4 Community Values program ................................ ........ 171 6 8 Output weights table ................................ ................................ ......................... 171 6 9 The A4 Weighting tool ................................ ................................ ...................... 172 6 10 Input layers for the Weighing Tool ................................ ................................ .... 172 6 11 Output layer from the Weighing Tool ................................ ................................ 173 6 12 Trend Allocation tool ................................ ................................ ......................... 174 6 13 Planning by Table tool ................................ ................................ ...................... 175 6 14 Planning table ................................ ................................ ................................ ... 1 76 6 15 Pop ulation allocation map for Orange County transit scenario ......................... 176 6 16 Population allocation around transit stops ................................ ........................ 177 6 17 Summarizing results to block groups ................................ ................................ 177 6 18 Summarizing results to land use ................................ ................................ ....... 178
13 7 1 Suitability based on AHS goal 1 ................................ ................................ ....... 207 7 2 Preference based on AHS goal 1 ................................ ................................ ..... 207 7 3 Rent monthly estimation for Orange County ................................ ..................... 208 7 4 Rent preference surface for Orange County ................................ ..................... 208 7 5 Travel cost monthly estimation for Duval County ................................ .............. 209 7 6 Travel cost preference surface for Duval County ................................ .............. 209 7 7 Transit access suitability surface for Pinellas County ................................ ....... 210 7 8 Transi t access preference surface for Pinellas County ................................ ..... 211 7 9 ARDT opportunity surface for Orange County ................................ .................. 212 7 10 ARDT opportunity surface for Duval County ................................ ..................... 213 7 11 ARDT opportunity surface for Pinellas County ................................ ................. 214 7 12 Refined ARDT opportunity surface for Orange County ................................ ..... 215 7 13 Refined ARDT opportunity surface for Duval County ................................ ....... 216 7 14 Refined ARDT opportunity surface for Pinellas County ................................ .... 217 7 15 Refined AR opportunity surface for Duval County ................................ ............ 218 7 16 Refined AR opportunity surface for Orange County ................................ ......... 219 7 17 Refined AR Opportunity Surface for Pinellas County ................................ ....... 220 7 18 Refined ARD Opportunity Surface for Duval County ................................ ........ 221 7 19 Refined ART Opportunity Surface for Duval County ................................ ......... 221 7 20 Example combine grid for Orange County ................................ ........................ 222 7 21 Example underutilized grid for Duval County ................................ .................... 222 7 22 Example walkabilty bikabilty grid for Orange County ................................ ........ 223 7 23 Example of vacant parcel grid for Duval County ................................ ............... 223 7 24 Example land value grid for Pinellas County ................................ .................... 224 7 25 Example distance to CBD for Orange County ................................ .................. 224
14 7 26 Example distance to major activity centers for Pinellas County ........................ 225 7 27 Qualified Census tracts grid For Duval County. ................................ ................ 225 7 28 Transportation analysis zones for Orange County ................................ ............ 226 7 29 Affordable housing allocation model ................................ ................................ 227 7 30 Duval County allocation ................................ ................................ .................... 228 7 31 Orange County allocation ................................ ................................ ................. 229 7 32 Pinellas County allocation ................................ ................................ ................ 230 8 1 Example of parcel level allocation for affordable housing ................................ 241
15 LIST OF ABBREVIATION S AHP Analytical Hierarchy Process AHS Affordable Housing Suitability Model AMI Area Median Income AR Access Rent Opportunity Su rface AR D Access Rent Driving Opportunity Surface ARDT Access Rent Driving Transit Opportunity Surface AR T Access Rent Transit Opportunity Surface CBD Central Business District CNT Center of Neighborhood Technology CRA Community Redevelopment Act CTPP Census Transportation Planning Package ESRI Environmental Systems Research Institute FGDL Florida Geographic Data Library GIS Geographic Information Systems GWR Geographically Weighted Regression HBO Home based Other Trips HBS Home based Shopping Trips HBS R Home based Social and Recreational Trips HBW Home based Work Trips HT Housing Transportation Index HUD United States Department of Housing and Urban Development IDW Inverse Distance Weighted LUCIS Land Use Conflict Identification Strategies LUCI2 Land U se of Central Indiana
16 MAUP Modifiable Areal Unit Problem MCDA Multi Criteria Decision Analysis MCDM Multi Criteria Decision Making MUA Multiple Utility Assignment NHB Non Home Based Trips NHTS National Household Travel Survey OLS Ordinary Least Squares OW A Ordered Weighted Averaging QCT Qualified Census Tracts STD Standard Deviation SUA Single Utility Assignment TAZ Traffic Analysis Zones TCRP Transit Cooperative Research Program VBA Visual Basic for Applications VBEditor Visual Basic Editor VLI Very Low Income Group VMT Vehicle Miles of Travel WLC Weighted Linear Combination 5Ds Density, Diversity, Design, Destination and Distance
17 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillm ent of the Requirements for the Degree of Doctor of Philosophy EVALUATING ACCESSIBI LITY AND TRAVEL COST AS SUITABILITY COMPONENTS IN THE AL LOCATION OF LAND USE A CASE STUDY OF IDENTIFYING LAND FOR AFFORDABLE HOUSING I N THREE COUNTIES IN FLORIDA By Abdul naser A. A. Arafat August 2011 Chair: Ruth L. Steiner Co chair: Paul D. Zwick Major: Design, Construction and Planning The allocation of suitable land for future growth is based on either stochastic or deterministic land use modeling. The stochastic approach uses land use change data samples to predict the future land use in a process to extend the past trend into the future. The deterministic land use modeling uses the data for the whole population to provide a vision for the future. The stochastic a pproach lacks the flexibility available in deterministic models to include planning expertise and community preferences. Deterministic methods include Multi Criteria Decision Analysis (MCDA) and the Land Use Conflict Identification Strategy (LUCIS). In the se methods suitability is determined by criteria evaluations using MCDA or by the relative usefulness towards goals and objectives using LUCIS The allocation of future land use can be conflicting depending on the goal s and objectives of the planner or stakeholders. These goals and objectives are presented as suitability surfaces and are combined to generate one suitability surface using the Analytical Hierarc hy Process (AHP). LUCIS adds the ability to identify the land use conflict between the sets of goals and objectives. The conflict identification
18 strategies introduced by LUCIS can, for example, identify the conflict between three main land uses: agriculture, conservation and urban. The allocation of suitable land for affordable housing differs from the allocation of land for residential locations performed by land use suitability models. Housing affordability is evaluated by estimating the housing cost as a percentage of income. The Center of Neighborhood Technology ( CNT) redefined the affordability evaluation to include housing and transportation cost. The affordable housing model in this research is a suitability model that is sensitive to multimodal transportation systems. The research evaluates the use of accessibi lity and travel cost as components in the all ocation of affordable housing. Furthermore, the research introduces a set of new Geographic Information System ( GIS ) tools for the automation of the affordable housing allocation process This research applies LUCIS conflict and allocation methodologies for the allocati on and preserv ation of land for affordable housing in three counties in Florida. For that purpose, LUCIS models have been restructured by updating the goals and objectives in the suitability struc ture to allow the inclusion of housing cost, travel cost and transit accessibility in the affordable housing suitability model.
19 CHAPTER 1 INTRODUCTION Research Purpose The coordination between land use and transportation has been the focus of several research studies. Giuliano (2004) identifie s the relationship between land use and transportation as being bi directional ; the i mpact of transportation on land use is captured via an accessibility measure ( Hanson, 2004 ; Giuliano, 2004) and the impa ct of land use on transportation is captured via land use descriptors such as diversity, density, design, destinations and distance ( Ewing & Cer vero, 2001) With reference to location and residential location theories, the choice of a place of residence depends on the combination of housing cost, transportation cost and other location attributes such as the distance to central busin ess centers (C BD) (Steiner, 1996 ). Affordable housing is one of the most sensitive types of housing to transportation variables. Chapter 2 review s the literature related to land use modeling, land use transportation coordination as well as the literature relating the affordable housing allocation to land use and transportation variables. According to CNT (2007), t he relation ship between housing cost and transportation cost is a crucial element in deciding the location of affordable housing. A ccessibility, mobility and proximity to public transportation and employment centers are also important. Furthermore, t he choice of affordable housing locations depends on demographic and site characteristics and the demand for affordable housing in addition to transportation vari ables. The CNT (2007) report on affordability focused on housing and transportation cost ( HT ) and emphasized the importance of the HT index which is mainly travel cost and housing cost relative to income. The CNT model wa s the first model to predict trave l cost spatially. However,
20 their index was based on a regression model with a coarse spatial resolution (C ensus blocks 2000 ) and focuses on national data more than the data from local sources Studies on affordable housing location such as the HT model estimate travel cost based on statistical models. This is done by taking the travel cost as a dependent variable and the land use, transportation and socio economic variables as independent variables in ordinary least square r egression models. However, i n estimating travel cost and evaluating its burden on suitability, the following should be taken into consideration when using statistical models: 1) They should be applied to small data samples and not on whole populations; 2) They traditionally ignore spatial location; 3) The lack the intervention of planning expertise in deciding suitable locations; 4) They underestimate or overestimate the cost for locations that are under represented in the sample; and 5) They traditionally applied to political and natural areal units such as traffic analysis zones (TAZ) and Census block g roups Chapter 4 in this dissertation explains that using natural and political areal units to capture urban form character istics may lead to inconsistencies in the results The Modifiable Areal Unit Problem (MAUP) suggests that scale and zoning affect the aggregated values captured by the area unit C hapter 4 also introduces a method to reduce the scale and zoning problems by using a floating neighborhood that has an op timized s ize and shape Modeling and predicting travel cost spatially can be performed using a spatially discriminated approach based on the latest Florida geo coded trip ends data from the National Household Travel Survey (NHTS) 2009 The travel cost estimation in this research is explained in Chapter 5 and follows two main methods; an interpolation
21 approach and a statistical approach. In the interpolation approach, the travel cost is directly estimated from the spatial interpolation of the geo coded trip location data of the NHTS 2009. The statistical approach uses the trip data in a regression model that relates the travel cost as a dependent variable to land use and urban form variables as independent variables. These independent variables are cross sectional data sets that include residential density, retail density, connectivity diversity of land uses in the form of entropy and a n accessibility estimation that represent s access to major employment destination s Two regression methods are u sed in the stati stical approach; t he first method uses a regression analysis by ordinary least squares (OLS) which gives a global regression equation for a regional or county wide area. The second method us es a geographically wei ghted regression with different equations r epresenting different geographic locations and different goodness of fit for each location Both methods result in a predictive model that can be applied to calculate the travel cost depending on land use an d urban form characteristics. The regression mode ls can be also used longitudinally by applying them to a different year using different independent variable values due to the future land use change. Land use suitability surfaces are used in hierarchal structures that include constructing single utility and multiple utility surfaces in which the composition of these utilities are performed using preferences and community values ( Carr & Zwick, 2007). In these utility compositions accessibility is simplified as the proximity to facilities such as proximit y to highways and t ransit stations without taking other definitions of accessibility
22 into account. Chapter 5 also investigates and compares mathematical models that are used to estimate accessibility in the transportation and land use planning literature. Gravity, opportunity and distance models are used in statisti cal and stochastic models ( Waddell, 2002; Handy, 2004; Hanson 2004), however, they are applied on either a random data sample or on aggregate and zonal level such as TAZs This research uses a p arcel level suitability evaluation that captures a combination of access by opportunity and access by distance in a multiple utility assignment. Furthermore, the distance estimation component can be currently performed by network or Manhattan distance as a lternatives to using Euclidean distance. The network distance is, the distance travelled between two locations using the road network. The Manhattan distance is the distance travel between two locations following a grid network while the Euclidean distance is the straight line distance between two locations. The aforementioned distance estimation methods are used to evaluate the suitability in the proximity component, and in creating the capture area for the opportunity suitability component. LUCIS include s a land use modeling methodology that replaces the traditional multi criterion methodology. The LUCIS methodology depends on community preferences, suitability surfaces and pair wise comparisons to build conflict surfaces that are used to identify conflic t between different land use s in the process of allocating lands for future growth. This suitability approach is usually taken by composing multiple utility surfaces (MUA) ( Carr & Zwick 2005 ) These MUA surfaces are generated from a combination of different suitability surfaces that are weighted by community decision makers using Delphi or pair wise comparisons to assess community preferences using
23 automated t ools as explained in Chapter 6 Applying these methods in the choice of affordable housing locations can identify suitable places but it does not emphasize the reason that makes these locations suitable for affordable housing. This is because affordable housing locations are associated with a low travel cost, or because they are associated wit h better amenities. This tradeoff may lead to undesirable results in terms of allocation or preservation of land f or affordable housing because of the conflicti ng nature of elements that generate the suitability surfaces. Chapter 7 explains the allocation of affordable housing by LUCIS model s which take conflict into account ( Carr & Zwick 2007 ) and allow the allocation or selection of suitable lands more flexibility using the LUCIS automated allocation tools. The research conclusions are explained in Chapt er 8 In C hapter 8, the results of the affordable housing allocation model are related to the literature on affordable housing, compact development and sprawl. C hapter 8 also explains the research limitations, recommendations and the future research. Disse rtation Objectives The main objective of this dissertation is to identify the transportation variables that impact the location choice of affordable housing and use these variables in a land use conflict identification model for the allocation or preservat ion of affordable housing site. The main objective includes the following sub objectives: 1) Capture the physical relationship between residential locations and services by the use of destination accessibility as suitability surfaces instead of the tradit ional proximity surfaces used in land use models. This includes the creating these surfaces using different accessibility definitions such as proximity, opportunity and distance; 2) Investigate the use of network distance and network service areas in calculating accessibility and compare the results with Euclidean distance based accessibility;
24 3) Enrich the land use suitability model with the use of travel cost as suitability surfaces. This inc ludes creating these surfaces directly from travel surveys or other data sources; 4) Investigate methods for estimating or predicting future travel cost based on travel surveys and the relationship between land use and urban form characteristics; and 5) Introduce an updated conflict identification procedure with an automation tool that helps planners to choose locations for affordable housing in a more flexible and automated process than traditional suitability models. Research Questions The research i s about creating a deterministic suitability model for the allocation or preservation of land for affordable housing. The research answers the following questions: 1) What are the feasible methods to create and include accessibility and travel cost as su itability surfaces in an affor dable housing suitability model? 2) What is the impact of travel cost and transit accessibility on the allocation and preservation of affordable housing sites ? and 3) How to incorporate multi modal transportation and metrics f or conceptualizing sprawl in allocating land for affordable housing? What is Th e Contribution of This Research ? This research adds to the body of knowledge of transportation land use coordination in general and specifically to the literature on location choice for affordable housing. Traditionally, suitability models consider proximity to services as a utility function. This research incorporate s accessibility and travel cost as suitability surfaces in addition to of proximity. Chapters 7 and 8 show that using these transportation variables impacts the a llocation of affordable housing and provides multi modal transportation options The research also shows that the use of transit acc essibility and travel cost can identify locations for compact development i nstead of sprawl.
25 Euclidean distance is often the method used in deterministic suitability models. The literature review in Chapter 2 shows that network distance is a better estimation of travel distance than the Euclidean or Manhattan distance. This resea rch investigates and compares accessibility estimations based on network distance versus estimations based on Euclidean distance. In transportation research however, networks are used to estimate distances and travel times Building network distance suitab ility surfaces, estimating and creating accessibility surfaces using network distance, or creating opportunity suitability surfaces based on network service areas are not yet covered by the literature of land use and transportation, and is introduced in th is research. The literature identifies the impact of mobility on accessibility which increases at the regional level or with the use of other transportation modes such as transit. This research takes that into consideration by incorporating a multi sca le accessibility estimation that is sensitive to multimodal transportation systems such as walking biking, driving and transit. Accessibility is considered on the local scale by estimating neighborhood accessibility to transit stops and other local servi ces. On the regional scale, accessibility is considered in the estimation of travel cost burden and/or the estimation of transit accessibility to employment when transit is a strong competitor. The CNT (2007) ha s used travel and housing cost to build an in dex that is also dependent on income. The HT index created by the CNT aims to estimate the affordability of housing site locations while considering transportation costs for the people that live in those places. This research however, aims to allocate affo rdable housing land for low income population. Similar to the HT index, travel and housing costs are important factors in deciding residential location. The location in the HT index
26 combines housing and transportation cost. This combination applies a trad eoff between housing cost and transportation cost. Chapter 7 of this dissertation builds conflict surfaces that identify the conflict between travel cost, housing cost and transit accessibility and the urban residential suitability. The research also intro duces scenario building to refine the output of the Affordable Housing Suitability (AHS) model and to allocate the most suitable land for affordable housing according to scenario conditions. The travel cost estimation in CNT research is performed by ord inary least squares regression. This research also uses ordinary least squares regression to create a travel cost estimation model. Chapter 5 of this dissertation shows that the residuals for the ordinary least squares regression are clustered and introduc es using spatial interpolation and geographically weighted regression to estimate travel cost. This research also introduces geographically weighted regression as a method to study the relationship between land use and transportation variables, as it can c apture the relationship on a local scale and identifies the local anomalies that the least square analysis ignores. Chapter 4 explains the research conducted to choose the areal unit and introduces a new method to reduce the effect of the modifiable area l unit problem using an optimized size and shape floating neighborhood instead of using political zoning and natural neighborhood boundaries. The research applies the optimized neighborhood in estimating the aggregated values for the metrics that capture t he urban form characteristics. Metrics such as density, diversity and connectivity are used as variables in the travel cost regression and the affordable housing allocation scenarios.
27 CHAPTER 2 LITERATURE REVIEW Housing is considered to be affordable if its burden does not e xceed 30% of household income (US Department of Housing and Urban Development [ HUD ] 2011 a ). Other evaluations for affordable housing include a housing and transportation burden of 45% of house hold income (Center of N eighborhood Technology [ CNT ] 2011 ). Therefore, the definition of affordable housing in general contains all the income groups. This research focuses on affordable housing for very low income (VLI) groups which is defined as income groups below 50% of the area median income (AMI) (HUD, 2011 b ). Regardless of their income category, people generally look for the best location available that they can afford within their budget. As such, the location choice for affordable housing can be p erformed by the same methodology as residential location choice models. The residential location choice depends on transportation cost, housing cost and proximity to CBD (Steiner, 1996 ). Accessibility, environmental quality and space are also important for deci ding where to live (Yamada, 1972). The a ffordable housing choice imposes additional constraints on the transportation and housing cost s and can be seen as a special case of the residential location choice model Evaluation of Location Choice Models Th e choice of residential location depends on attributes that are derived from location theories. Attributes such as housing cost, transportation cost, distance to CBD, accessibility, space, leisure and environmental quality are frequently used in location choice models (Yamada, 1972). Historically people left the inner city to live in suburban locations that allows them to have more spacious and luxurious homes they can afford.
28 suburban residential locations available for people. These suburban locations are promoted by low pricing, low interest rates on mortgages. As a result of living far away from jobs, the use of cars increased. The use of cars is also encouraged by low gas prices and the increasing mobility due to the construction of highways and freeways. This car depe ndent suburban housing has caused sprawl, larger commuting distances and higher transportation costs associated with the rise in gas prices. In terms of affordability, suburban housing has become a heavy burden for householders and less affordable due to i ncreasing transportation cost (Lipman, 2006). CNT (2007) worked on evaluating the affordability of housing by estimating the housing and transportation cost with respect to income. Housing is considered affordable if the combined housing and transportatio n cost does not exceed 45% of income. However, the actual spending on combined transportation and housing nationally exceeds 50% of household income. CNT (2007) showed that increasing the commuting distance by living away from work reduces the housing cost but the total added an extra burden on people and reduced the affordability of residential location. Transit research on the national level shows an increase in tra nsit ridership in ridership trends increases for low income and older populations (TCRP, 1998). However, this also depends on the urban form and the level of service of t he transit system. Commuting to work by car or by transit for low income populations is a controversial and complex issue. Grengs (2009) shows that transit in a city like Detroit is not advantageous for low income populations where using cars may increase the
29 employment rate and increase accessibility to jobs. Research analysis shows the complexity of the relationship between accessibility, driving and transit especially for low income populations. This suggests that choosing between driving and transit f or the trip to work is not a sole choice of the traveler but depends on the accessibility to work and the transit level of service. For Detroit, transit is inferior compared with cars. Car usage adds flexibility especially when the trip chaining for daycar e or shopping is needed. Furthermore, transit is inferior in Detroit because the accessibility to employment by transit is very small (Grengs, 2009). The research on land use transportation coordination suggest the importance of accessibility and urban fo rm characteristics in reducing the cost of driving associated with residential location and increasing the use of other modes of transportation such as walking and transit (Cervero & Kockelman, 1997). This suggests locating residential land uses in places that have high accessibility to services and employment to reduce the travel cost when using cars and at the same time encourage the use of other transportation modes such as walking, biking and transit. Location choice modeling follows methodologies tha t are derived from location theories such as the residential location theory. Generally there two main modeling approaches; a deterministic approach such as the Land Use Conflict LUCIS (Carr & Zwick, 2007) and a statistical approach like that used by Wadd ell et al (2003) in the statistical compo nents of their UrbanSim models The deterministic approach assumes that the modeler understands the relationship between model variables and can apply his experience to predict the future. The statistical approac h relies on data samples to
30 capture the relationship between variables via regression or other statistically robust methods of analysis. In location choice models, accessibility to services and amenities is important. Accessibility is taken as a typologic al proximity estimation in deterministic models while many mathematical forms of typological and opportunity estimations are used in statistical and stochastic approaches. Also, statistical approaches usually do not deal with visual representation using GI S as is often the case with deterministic suitability app roaches. Statistical approaches also are applied on data samples unlike deterministic approach es which are typically applied on the general population The prediction of land use change does not depen d on ly on accessibility but also on land use and economic variables LUCIS models depend on E uclidian distance proximity measurement s to include the suitability of locations Additionally, LUCIS models include community preferences using pair wise comparis ons in deciding the potential of land use change on a small scale, ultimately construct ing a conflict raster that can be used in the allocation of projected population in different locations. The allocation of population follows the rules that are set by t he urban planner. For example, for areas around transit, the allocation is performed using a circular buffer. These are the places that have transit accessibility which is used as a shed in the allocation. Other economic, spatial and temporal variables ar e also used indirectly in the generation of suitability surfaces or in the land allocation process depending on the conflict raster surfaces being used ( Carr & Zwick 2007 ). In other land use prediction model s, cells that represent par cels are also used to construct probability functions that capture the existing land use pattern and project tha t to predict the future using Markov
31 Chain probability transformation (Levins on & Chen, 2004 ). The later method ignores the planners and community preferences includ ed in LUCIS models and also it assumes that the same relation between variables in the past will apply to the future which is only estimat es for future growth trend. Waddell (2002) used transportation and economic models to feed UrbanSim use parcel models by incorporating travel times and other variables from networks using the trip based four step transportation modeling. However, the UbanS im provided by GIS modeling. Furthermor e, using four step transportation modeling does not include the land use impact on transportation nor the representation of space time relationships that are found in activity based and trip chaining transportation models. This raises the need of activity simulators and the discrete choice models in location choice. It is very difficult therefore to decide which model gives more appropriate result s for land use change due to the lack of comparative scholarly research on different types of models. Location preferences and c hoice s are also modeled in different ways. In LUCIS models it depends on the community va lues and participatory sessions while in other land use models it is mostly probabilistic logit models such as Land Use in Central Indiana Model (LUC I2) (Ottensman, 2004). Location preference and choice dec isions can also be facilitated by proba bility matrices to capture the growth from two different years in the past and us ing a transformation matrix to calculate the probability of change in the futur e (Levins on & Chen, 2004 ). Waddell et al (2003) uses more sophisticated statistical methods such as discrete choice model s to construct utility and nested utility logit functions The discrete choice
32 model is an important approach to model location choic e (Waddell et al 2003 ). The p arcel level analysis and discrete choice models have been used effective ly in the building of the Urban Sim mod els. An example application of this methodology was performed on the Spring Field Metro politan A rea and Lane Count y in Oregon using a disaggregate database (Waddell, 1998). The Urban Sim model s are also used for growth management such as studying sprawl, congestion and affordable housing through the integration between transportation and land use models that can feed each other (Waddell, 2002). Land use affect s transportation system s and transportation system s affect land use. Therefore computer models like Urban Sim can help the urban planner in running different scena rios throu gh the use of different UrbanSim componen ts. Urban Sim is a model that simulates land use and transportation and has many components such as e conomic elements and transportation and utilized choice model s in tegrated within other models (Borning & Waddell 2004 ). However, several dis advantages of UrbanSim are its lack of the visual and GIS interfaces like those that exist in other land use models Also, its intensive use of disaggregated travel data precludes its use in states where that level of data may not be available such as in Florida. There are many models that deal with the effect of transportation on land use The traditional way is to use economic factors embedded in an accessibility index A review of comm on frameworks can be found in the work of Iac o no and Levinson ( 2008 ). Iacono and Le vinson also compared a new regression approach using additional variables for transportation and compared that to a model that does not include transportation
33 variables T hey found that transportation variable s exert some influence on land change and land use pa tterns The use of deterministic or stochastic approaches to model the effect of transportation on land use depends mainly on the available data and the need of planning intervention in the prediction or visioning process. This research relies on th e deterministic approach and investigates the role of transportation in deciding location choice. Statistical methods are also used in this research to understand and capture relationships that might be difficult to obtain using deterministic models. The r esearch will incorporate more transportation variables than what is currently used in suitability modeling. Suitability raster surfaces will be built out of accessibility estimations. However, because of the interdependency between accessibility and mobili ty, which increases when capturing regional accessibility or when using other modes such as transit ( Sa lomon et al 1998 ; Handy, 2004; Hanson, 2004 ) this research adds the ability to spatially modify the accessibility value according to mobility measurem ents in a suitability environment. Generally, the mathematical forms of accessib ility other than proximity have been used in statistical and stochast ic land use modeling and not in land use suitability analysis. This research will introduce these variables such as accessibility and travel cost as suitability raster surfaces in a land use suitability model structure for affordable housing location choice. S uitability Models Land use suitability analysis is an analytical process that combines inventory information to determine whether the requirements of particular land use s are adequately met by the characteristics of the land. The result is either tabular data or a single map or se ries of composite maps that display the relative suitability or
34 appropriateness for a specific land use useful for example in location choice studies, or for a number of land uses such as in comprehensive planning (Randolph 2004 p. 591). As landsca pe architects in the late 1800s, Charles Eliot and Warren Manning used suitability analysis in their environmental planning pursuits to measure the relative degree lands in Boston were fit for integration into the Bo ston Metropolitan Park System. Central t o this process was the develop ment of a systematic approach to the inventory of site resources and, through the use of overlay mapping, the anal ysis of the natural fitness of the land ( Carr, 2008, p. 5 ). Suitability techniques have evolved quickly during the twentieth century. In the 1960s Ian McHarg included an ecological inventory process into suitability analysis. During the late 1960s and early 1970s the advent and use of computers in land use suitability marked the beginning of a revolution expandi ng the capabilities of suitability analysis. With computers large amounts of information could be combined and overlays became more accurate. The most significant technological advance was the use of the comp uter to make simple grid maps. The grid cell al lowed more precise analysis of map factors between multiple m aps. In the 1980s map algebra was developed which allowed mathematical computation among several grid maps. In the early 1990s, GIS became a formal technology. According to Collins et al (2001 ) GIS is used to used to manage spatial and none spatial data (storing, analyzing and presenting). GIS also used to create new datasets by overlays and spatial operations. LUCIS illustrates the next era of suitability modeling. LUCIS is organized in a hie rarchical structure of goals, objectives, and sub objectives, for each respective objective and/or sub objective, a GIS model is developed. Each model is a sequence of
35 spatial data and geo processing tools that first assign an estimate of utility and then assigns a suitability value for that utility. In the higher orders of the hierarchy, suitability assignments are made for the development of land uses (i.e., agriculture, conservation, and urban) which are then combined in a single raster to identify the conflict between the land use preferences (Carr & Zwick, 2007). The suitability index is a value that represents the relative usefulness for a particular land use. In the LUCIS model values ranging from one to nine are assigned, where one represents the l owest suitability and nine the highest suitability value (Carr & Zwick, 2007). Classification into these value ranges occur using various methods depending upon the nature of the criteria to be evaluated or according to the utility to be classified as a su itability surface. Some of the procedures are simple (binary methods) and some of them have higher complexities. Regardless of whether the model measures a qualitative or quantitative process, the output of the LUCIS model employ s at least two values, 1 an d 9. GIS layer overlay is the core of suitability analysis. Even suitability analysis undertaken at the time of hand drawn maps was dependent on map overlay (Collins et al ., 2001; McHarg, 1969). The overlay procedure in GIS raster analysis depends upon th ree logical spatial overlay rules: enumeration, dominance, contributory and interaction. According to Carr and Zwick (2007, p. 50 57): Enumeration preserves all attribute values from multiple input layers. Enumeration creates an output layer that combines all attributes from the spatial input layers to provide a clear and distinct set of unique attribute combinations from the input. The dominance rule depends on the selection of a single value that is preferred over all other values found at the same spat ial location. The selection is defined or governed by external rules, not simply the combination of values. The contributory rule is applied by performing a group of operations [which are] values from one input
36 contributing to the results without regard for the values from other inputs. Lastly, the interaction rule, unlike the contributory rule, considers the interaction between factors. However, to consider interactions between factors, the factors must be translated into the same standard intervals. T hese rules represent logical operations that can be translated into equivalent functions in land use modeling such as layer weighting and the combination of different utility surfaces into a suitability layer. The dynamic relationship between land charact eristics and land use illustrates the complexity of land use suitability analysis (Driessen & Konijn 1992 ). Through interaction, utility is combined to create suitability. Single utility assignments (SUA), which are the assignment of utility values within an individual raster layer, are combined using weights to create multiple utility assignments (Carr & Zwick, 2007). However, utility is a measurement of human satisfaction and thus if applied to land use could represent how much a person can be satisfied by the land characteristics Utility can be easily connected to people characteristics while suitability is usually connected to the location characteristics. This explains why utility is usually used in statistical choic e models while suitability is used in c riterion evaluation models or land suitability models The various appr oaches to suitability analysis provide alternative ways to understand the interactions between human and nature in ecological planning (Ndubisi 2 002 ) Therefore, t he conflict between planning for urban expansion and ecological systems can be studied and identified by innovative methods in suitability modeling The GIS overlay techniques for the Multi Criteria Decision Making ( MCDM ) method which is another method for land use modeling that use the hierarchal structure, can be divided into two main methods: the multi objective method and the multi attribute method. The multi objective method depends on two or more objectives
37 to be combined using a set of constraints. This is always solved by standard linear programming methods. The problem in this method is that adding constraints will help the planner in decision making but will add computational complexity making it difficult to apply in a GIS env ironment. The multi attribute method is applied using GIS map algebra techniques. It uses weighted linear combination (WLC) and the Boolean T his process however, gives the same weight despite the geographi c location, as the WLC is based on the concept of a weighted average. In this method relatively more importance is given to the attributes because it is assumed that the importance of location is taken into account in generating each layer to be combined b y the Boolean operator. Ordered weighted averaging (OWA) has also been used in the MCDM method to overcome the disadvantages of WLC. The OWA method involves two set s of weights ; one is the criterion importance weight which is constant for the criterion a t all locations and the other is the order weight which is associated with the criterion on a location by location weight (geographic or spatial weights). Additionally, AHP is a method used in MCDM that incorporates the generation of the linear combinati on weights by aggregating the priority for each level in the hierarchy process. AHP is also used as a consensus building tool in situations involving gr oup decision making (Malczewski, 2004). Programmed and automated procedures as well as community parti cipation using Delphi or pair wise comparison methods (i.e., AHP), are used in ranking and ordering procedures to assess the importance of weights (Carr & Zwick, 2007; Malczewski, 1999, 2004). The pair wise comparison technique developed by Thomas Saaty in the
38 1970s and 1980s in the context of AHP multiple criteria evaluation methods, represent s the relative importance of criteria. According to Nyerges and Jankowski (2010, p. 140 141): Weights are not a derived from the eigenvector of the square reciprocal matrix used to compare all possible pairs of criteria. The advantage of this technique is that information can be used from handbooks, regressi on output, or decision modelers/experts can be asked to rank order individual factors. Malczewski defines weight as a value assigned to the output of criterion evaluation. The weight represents the relative important of that output. The criterion is more important if the weight is higher and less important if the weight is lower ( Malczewski 199 9 ). Both MCDM and LUCIS integrate weights into their methods. MCDM provide s four methods for assessing criterion weights: ranking, rating, pair wise comparison, and trade off analysis. Malczewski (1999) included the choice of method depends in the trade offs the modeler is willing to perform, the availability of software, and the method of incorporating GIS based criteria evaluation. Carr and Zwick (2007) calcul ate community preference using the more advanced pair wise comparison method of AHP. In the AHP procedure, a model is created and a project goal is stated. The goal for pair wise comparison is a statement defining pair comparisons. The objectives and sub objectives are treated as components of the overall goal. Then, each unique pair is compared for their usefulness in supporting the one to nine scales, ranging from equally important/useful to extremely mor e important/useful (Table 2 1 ). Next, the pair wise comparisons are evaluated within a matrix for all pairs of values to produce final pair wise utility values ( Figure 2 1 ). Lastly the final pair wise utility values are transformed into single utility ass ignment values ran ging from one to nine (Carr & Zwick,
39 2007). After completing the pair wise comparisons, the weight for each layer is calculated according to an eigenvalue / eigenvector procedure. LUCIS (Carr & Zwick, 2007) uses software packages externa l to the Geographic Information System Software ArcM ap environment to calculate the AHP layer combinin g weight s. ArcMap is GIS software created by the Environmental Systems Research Institute ( ESRI ) The software provides tools that are vector and raster b ased to spatial analysis and to manage geographic data. Multi Criteria Decision Analysis MCDA uses layer combinations according to the outcomes of AHP and the consensus of Delphi panels. The combination is mainly layer weighting using an interaction rule. However some of the weighting is done in the suitability assignment level in the hierarchy structure. A similar technique is utilized in the LUCIS model. The primary and most important difference is that MCDA uses alternative scenarios and the weights gene rated by AHP to evaluate the suitability for each scenario while LUCIS uses a conflict surface, which is a matrix that preserves the original preference values. This matrix consists of three or four digits, according to the number of preference surfaces co mbined to create the conflict surface. Preference applies community values to the cumulative suitability of land fitness. The aggregation of relative suitability surfaces for a goal can be seen as an opportunity surface even if it has some conflicting asp ects. For example, an opportunity surface for urban suitability may contain the complex MUA grids for commercial, industrial, multi family and single family land uses The generated opportunity raster surface identifies the conflict between the components of an urban environment yet maintains the original suitability for each individual component. The interaction between sets of goals within
40 each land use, illustrated at the highest level of the hierarchy, demonstrates conflict while preserving the suitabil ity of the generating surfaces (Carr & Zwick 2005). The purpose of the conflict surface is to generate a suitability matrix. Individually, suitability is determined and weights are assigned to suitability raster grids from AHP values exercises to create a complex MUA for each respective land use. Next, these land use raster surfaces are transformed from suitability into preference, which places each land use raster surface on the same scale from one (low preference) t o three (high preference) (Table 2 2 ). Using map algebra, each respective preference surface is combined to create a single conflict surface which is a suitability matrix that identifies the conflict between the preferences. The generation of the conflic t surface is performed by multiplying the first preference by one hundred, the second preference by ten, and the third preference by one (Table 2 3 ). The surfaces are then combined using additive sum. Multiplication is not performed according to the import ance as a weight but only to generate a two decimal index for identifying the conflict. There are three conflict classifications in LUCIS. No conflict is when a single land use type has the highest preference value and the other land uses in the conf lict score have lower values Minor conflict is when two goals have the same preference value and no other land use type has a higher value. Major conflict is when all land use types have the same preference values. For example if a conflict surface for thr ee land use types (i.e., agriculture, conservation, and urban) is created and the conflict values were arranged in the conflict matrix as the first, second and third digit resp ectively then for a given conflict value of 113 the specified location would be highly preferred for urban
41 land uses Whereas, a given conflict value of 221 would indicate a minor conflict between agriculture and conservation land uses as they both have the same preference for the specified location and urban prefer ence for the lan d has a lower value. The value of LUCIS is two tiered. The first tier consists of the process to determine land use conflict. As described above, the process includes 1) determining land use suitability based upon the pre determined goals and objectives; 2 ) determining land use preference; and 3) identifying conflict. The second tier illustrates alternative futures through the allocation of population and/or employment. As stated earlier, the conflict surface is a suitability matrix using the cumulative su itability of the goals within each land use. Early applications of LUCIS allocated people and employment according to a a conflict surface does not manipulate the original preference values; thus a conflict su rface can also be generated between goals for a more detailed analysis of land use preference. Therefore, allocation of urban uses has evolved from areas generally classified as urban to allocating projected residential populations into areas with high mu lti family and single family preference. The MCDA scenario building approach takes different alternatives and calculates the suitability for the model alternatives, which inherits a selection of the more appropriate scenario. However, in the LUCIS struct ure and the LUCIS allocation procedure, scenario building is performed on multiple levels. The first opportunity is when changing the weights upon combining suitability surfaces for each hierarchical level, which is the same analysis used in MCDA. The sec ond opportunity is in the flexible allocation scenario where the conflict and suitability assignment are used in a combined grid and the population allocation is performed according to priorities
42 specified according to different scenarios. The combine grid s join conflict and suitability values and preserve the attributes for these grids in the overlay. The tool is also useful for scenario building and testing of policies. In the allocation process, Carr & Zwick (2007 p. 167 ) identifies six general steps t o visualize future land use which are: 1. Allocation starts in the area that does not include conflict and where urban preference dominates. 2. Allocation continues if needed in moderate conflict and major conflict, if necessary, where the normal values for urba n are highest. 3. steps 1 and 2. 4. Allocate remaining cells for future agricultural land where it is not in conflict and the preference is greater than conservation or urban. 5. Allocate remai ning cells for future conservation land where it is not in conflict and the preference is greater than agricultural or urban. 6. Allocate remaining cells that are in conflict between agriculture and conservation according to the greater preference. Incorpor ating Transportation Variables in Suitability Modeling Generally, the LUCIS method provide s solutions for many of the shortcomings in a traditional MCDM method. For the processes for which MCDM is best known, LUCIS provides a decision analysis framework f or land use planners and modelers with knowledge of GIS technologies. Although the role of land use planners is shifting to include more physical and spatial planning analysis skills, LUCIS can facilitate this role change by automating key procedures in th e identification of land use conflicts and the process of allocating future population The LUCIS model structure is flexible. It may include any goal and objective the planner determines as important in the allocation process. However, transportation
43 vari ables are not yet fully incorporated in the current LUCIS model. Incorporating such variables would be useful in the allocation of future land uses especially if the model is to be used to allocate locations that are sensitive to travel cost such as in the locating affordable housing units. The main elements in research on the connection between transportation and affordable housing are accessibility and travel cost. These elements are used in statistical models on Census Block level data (CNT, 2007). Inco rporating these elements as suitability surfaces based on parcels level data for allocating and preserving affordable housing is not yet in the literature. The bi directional land use transportation research does however relate travel behavior indicators s uch as vehicle miles of travel (VMT) and trip generation to accessibility and other urban form characteristics. Investigation of these bi directional relationships has o n the one hand shown the impact of tr ansportation systems on land use change This has included using transportation descriptors in the prediction of the future land use. However, this can be described as a direct approach for the purpose of modeling location choices. On the other hand, the impact of land use change on transportation systems and travel behavior can be considered as a direct impact when evaluating transportation systems. It also can be seen as an indirect effect on the choice of locations in terms of land use policies, land use modeling and the design of new developmen t or urban forms Considering the transport ation impact on land use and f ollowing applications on the economic theories such as the location theory, c entral place theory, h edonic pricing models and residentia l location theory (Steiner, 1996 ), it could be said that housing cost, transportation cost and the spatial location are important factors for the decision of
44 where to live and work and that the growth pattern and the choice of location follows a utility maximization pattern for community in dividuals This may be simple when talking about household rent or housing cost but get complicated if we in troduce transportation cost. Taking accessibility and travel cost as the two main elements in the relationship between land use and transportation c an lead to a better understanding of this complex relationship. On the first hand we have accessibility and mobility and their impact on land use and location choice (direct impact). On the other hand, travel cost is the other way around (indirect impact). Different land uses affect our travel behavior which mainly increase or decrease the VMT or travel cost ( Steiner et al 20 10 ) Travel cost in turn affects our choices for housing and other land uses. Most of the research addresse s an accessibility type o f measure which is originally derived from growth theories (i.e. aggregated choice models such as gravity models ) Generally, the research on the effect of transportation on land use capture s micro level relations with macro level methodologies For examp le, the affordability index created by CNT (2007) is based on coarse spatial resolution such as Census blocks rather than parcels It is usually difficult to capture the impact of transportation on land use in terms of the complexity of the indicators that can capture the change in the urban forms. Therefore, in most c ases, the research studies that deal with the effect of transportation on land use estimate accessibility and use it along with other variables to predict land use change, growth patterns and future land uses. Evaluating Methods of Estimating Accessibility This section of the literature review evaluates the research on the direct impact of transportation on the choice of location as well as evaluating the methods of estimating accessibility and proximity which are considered as the most significant variables of this
45 impact. The impact of transportation is captured via a simplified accessibility measurement. The measurement of accessi bility varies in existing research ranging from linear distance to network distance, travel time and the number of activities within a distance from an attraction or a certain residential location. Accessibility is defined as the potential to interact. To differentiate between the accessibility and mobility, w e can say that mobility is the potential to move. In these terms, accessibility is connected to destinations and the mobility is connected to the networks and vehicles. Accessibility for example, measures the number of jobs in a certain area or the number of des tinations in a specified area o r the availability of choices between modes, while mobility deals with traffic delay and level of service (Handy, 2004). However, this explains that accessibility could be different between modes, but does not take cong estion into consideration because it is a mobility measure according to the definition. The same approach for accessibility is taken by Hanson (2004). According to Hanson, accessibly is t he number of opportunities with in a distance or travel time w hile mob ility refers to the ability t o move between different sites. Furthermore, Hanson explains that because of the distance between activities become larger as density decreases, accessibility becomes depe ndent on mobility (Hanson, 2004). This adds interdepende ncy between density, accessibility and mobility. It is clear that there is a relationship between accessibility and mobility and that this relationship is stronger for regional destinations other than neighborhood destinations particularly when different m odes of transportation are taken into consideration (Hanson, 2004; Sa lomon et al., 1998 ) In regional destinations, using highways and freeways increases the dependency of accessibility on mobility. Accordingly, when transit is involved
46 accessibility may d epend on the level of service and thu s on mobility In this research accessibility and travel cost are estimated on both the local level and regional level using multi modal transportation. Therefore mobility can be incorporated within accessibility and tr avel cost. Accessibility can be also defined as the ease with which a destination can be reached and it is on e of the important factors in location decision choices. This definition clearly connects accessibility a s a function of land use and transportatio n pattern s A ccessibility also can be defined as the ease with which people can participate in activities. Such a definition acknowledges that the destination activities and location properties are important factors in accessibility (Primerano & Tylor, 20 04) Accessibility measurements can also be divided into personal accessibility and place accessibility. Personal accessibility can be measured by counting the number of incl ude the magnitude of the distance for each location in a gravity cumulative approach. The accessibility for a place investigates the number of activities at a certain distance from a place. These are simple methods for calculating accessibility. More advan ce methods of time space analysis are needed to address the effect of time on accessibility (Hanson, 2004). However doing time space analysis on a disaggregate level of data at a dependable accuracy is not always possible in places with poor travel activi ty diaries. To simplify, choosing to use personal accessibility or place accessibility depends on whether personal characteristics are included in the estimation of accessibility. For example, it is possible to use personal characteristics to estimate acce ssibility for
47 existing urban development but it is more complicated to predict personal characteristics for new development. Generally, t he estimation for ac cessibility can be a topological or opportunity measure ment or both. The topological estimations a re an estimation of physical proximity from origins to destinations which inclu des the measurement of distance such as the distance to the nearest location The opportunity models measure a density or attraction of accessible places. Incorporating both giv es the relative accessibility which can be c learly shown in gravity models. This relative accessibility if accumulated for a large scale will result in a measure of an absolute accessibility (Levinson & Krizek, 2008). Land use change and land use predict ion models use accessibility estimations to model the change of land use over time. Topological accessibility using proximity is used generally in the suitability models of locations ( Carr & Zwick 2007 ) Gravity models, which can be based on the available opportunities and their travel distances to the location, are used in the modeling of location using statistical methods (e.g. modeling the employment opportunities for residential location, Waddell et al 2003). Opportunity and combined opportunity distance accessibility indicators using either Euclidian or network distances are not yet used in land use suitability analysis. Usually the simple accessibility estimation, as defined as the proximity or t he Euclidian distance measurement is used in LUCIS models (Carr & Z wick 2005 ) Many mathematical forms are used to estimate accessibility. Bhat et al (2002), summarized accessibility measurements into different equations for cumulative opportunity and gravity. However, they are applied on either a random data sample or on aggregate and zonal level such as TAZ level These estimations are classified as
48 Gaussian, composite impedance, activity distance and i n vehicle travel time. Table 2 4 compares the va riables used for accessibility estimation while T able 2 5 compares the merits and limitations in accessibility estimation using different methods. This research initiates the use of these accessibility estimations in land use modeling as explained in the C hapter 3 Evaluating Distance Measurement Methodologies In the absence of measured travel times, the accuracy of topological accessibility estimations depend on the method used to estimate the distance. Generally, distance measurement methods in land use r esearch are one of three methods; Euclidean distance, rectilinear distance (Manhattan)and network distance. N etwork distance can be obtained from a network property approach by measuring the length of street segments as a percentage of the whole street ne or by measuring the actual distance travelled (Zhao et al 2003). The use of travel time may be more sophisticated and take additional variables into consideration. However, i n measuring network distance barriers can be included to g ive a more accurate ind ication of travel distance. Arafat, Steiner and Bejleri (2008 ) compared network distance to Euclidean and Manhattan distance in research on school sitting Their research found that the use of network distance gives a better estimation for walking distance than Euclidean or Manhattan distance s. Additionally they found that the catchment area for population which is an accessibility indicator is exagger ated when using an Euclidean buffer. The network distance is used in transportation research to build accessibility indices in Texas (Bhat et al 2002), where, the travel distance had been obtained from travel surveys which may not be available on a disaggregate l evel. An alternative methodology can be used to generate the network distance at a parcel level using
49 ArcGIS network analyst which is software that can calculate distance from origins to destination following the road networks. In this methodology, the sho rtest network distance can be measured from each origin to each destination (Arafat et al 2008). It is difficult to compare the results of Euclidian or network distances in generating proximity surfaces for two main reasons: The first is that a network proximity surface is not yet used and proximity models use Euclidean distance. On the other hand, the se measures of accessibility used in land use models are normalized using a base cost, that is, for example, a suitable base time for walking or a certain Euclidian or network distance is regard ed as the base distance. This make s the comparison more complicated and increases the difficulty of dec ision making about the method that give s a better estimation. This base distance or travel time is different from one place to another and from year to year, but if the general trend is assessed it can be seen to be increasing over time. Furthermore, travel time is related to human behavior and motivation which is in turn related to social and technological changes (J anelle, 2004). It is clear, however, that the impact of networks and mobility increases as distances increase. Therefore, it is more useful to use network distance in capturing regional accessibility or travel cost. Evaluating Literature on the Impact of Land Use on Transportation The impact of land use on transportation is captured by research and used indirectly by planners and decision makers who are concerned about location choice. and land use modeling. It allows planners to solve transportation netwo rk problems and to make improvements on the transportation networks as well as calculating impact fees. Urban form and land use generally impact the cost of travel which in turn affects choices of where to live. The land use impact on transportation and tr avel behavior is
50 covered extensively in the literature. Most of the research focus es on measuring the urban form and its relation ship to the travel behavior. Steiner (1994 ) reviewed the literature on the effect o f residential density on travel behavior an d mentioned that most of the research is done on aggregate data in which there is difficulty in sepa rating the economic from the land use effect on travel behavior. The research typically used the effect of income and density on travel behavior without sep arating them and their typical conclusion will be that members of lower income households travel less than other type s of household (Steiner, 1994 ) One of the most important outcomes of research on the impact of land use on transportation is the what is called the 3D s (5Ds in later work) which refer to density, diversity and design indicators and their effect on travel behavior. Cervero and Kockelman (1997) u sed 1990 travel diary and land use records for the San Francisco Bay area and worked on non work trips to show that built environment affects the miles travel led per household as well as modal choice. Their research showed that the density, land use diversity or land use mix in addition to pedestrian oriented design reduce s the trip rate and encou rage s walking and transit use. They also emphasized that compact development affect s modal choice. For the design element the ir study show ed that a grid network and restricted parking reduce d the use of autos and increase d the level of transit and walking (Ce rvero & Kockelman, 1997 ). A n empirical study to test the impact of land use on transportation test ed the effects of land use mix, population density and employment density on the use of single occupant vehicles, transit use and walking in addition to mod al choice. In that study, however, the land use mix was measured at the trip ends and it was shown that
51 walking and transit use increase d when density and land use mix increased while the use of the single occupancy vehicles decrease d (Frank & Pivo, 1994 ) .. The research also shows that measuring land use mix at the trip ends gives a greater a bility to predict modal choices In addition to tha t, land use mix at the trip end also increase s walking and transit use and reduce s the use of the single occupancy v ehicles (Frank & Pivo, 1994 ). Ross and Dunning (1997) also found that increasing population density will decrease person trip and mileage. Steiner et al (2010) used the land use variables at trip ends to create a model for calculating trip length induced by new developments. Cervero and Radisch (1996) studied pedestrian activity and learned that modal choice for biking and walking increases in transit ori ented neighborhoods. Handy (1996) also concluded that urban form affects modal choice and increasing local accessibility will s modal choices (Handy, 1996). Zhang (2005) tested measuring urban form and non urban forms quant it atively on transit oriented development. In his research he used regression analysis and calculates the e ntropy value which is a test of land use heterogene ity. Ewing and Cervero ( 2001 ) summariz ed most of the literature on the effect of land use on transportation by assessing the research performed on the 4 Ds and 5Ds in later works. ensity, diversity and design. The fourth and fifth Ds are desti nations and distance to major transit stations such as rail stations The destinations are measured by regional accessibility indices or the accessibility to major destinations, such as activity centers and central business districts. Accessibility and dis tance are new variables considered to affect land use in addition to the 3Ds. Ewing and Cervero
52 (2001) summarized the research done in nearly 50 papers in that domain. Their summary culminated in the creation of elasticity values based on 5Ds Elasticity i s an estimator that is used to quantify the extent to which choice probabilities will change in response to changes in land use values Alternatively, it can be defined as the percentage change in the model response (i.e. VMT) with respect to a ch ange in the model input (i.e. 5Ds) ( Bhat & Koppelman, 1993 ). In general, in Ewing summary of the research, elasticity is mainly used to predict how much trip generation or trip length would increase if for example, density was doubled. E lasticity va lues of this kind are useful in land use modeling such as the smart growth index and future land use prediction and thus represent a feedback for the transportation model that is used for land use modeling (Ewing & C ervero 2001 ). The same land use variabl es included in the 5 are used to study modal split. Litman (2008) summarizes the impact of density, accessibility, land use mix; roadway connectiv ity to test how they affect none motorized travel behavior. summary show ed that feasible use of growth management strategies can affect the land use variables such as density, diversity, design and etc ., which in turn can reduce automobile travel by 20 to 40%. In his analysis Litman shows that the VMT per capita is reduced wh en density is increased. Furthermore, using cars as a mode is decreased when inc reasing the land use mix. C onnectivity of the street network also affects VMT which is reduced for higher connectivity networks For example, c hoosing walking as a mode is increas ed when the pedestrian netwo rk is highly connected. Further more the mode choice of walking is also affected by the attractiveness and completeness of the street network which support s the design
53 aspects of urban form Handy et al ( 2005 ) m entions that the res earch performed in studying the land use effect on travel behavior showed that increasing density and diversity of land use did make people drive less but these studies miss the ca usal relationship. Therefore, neighbor hood characteristic s travel prefe rence and residence preference are used in a quasi longitudinal design study to show the relationship between them and how neighborhood characteristics have a significant causal relation with travel behavior (Handy et al 2005) Urban f orm estimators oth er than the 5Ds are also used in researching the relationship between land use and transportation. D isaggre gate approaches are used to measure built environment indices that help in planning for future land use change. Rodriguez et al. (200 6 ) used the variety and spatial co variati on and their relation with non aut omobile travel to build and generate the Built Environment I ndex ( BEI ). A method for calcula ting the circuity index ratio (El Geneidy & L evinston 2007 ) which is als o known as Portla nd Pedestrian Ratio is adopted by Arafat et al (2008 ) to measure the connectivity of networks and the efficiency of using network distance to build a wal kability index to be used for school sitting based on a parcel level GIS and network analysis. These indices help also in predicting the travel behavior in addition to predicting location choice. Bejle ri et al (2008) us ed urban form indices to capture the urban form effect on student opportunity to walk or bike to school. Furthermore, Frank et al (2006) used a walkability index developed using density, connectivity, land use mix and retail floor area and applied that in King County, Washington to find out that a 5% increase in the walkability index decreased the VMT by 6.5% as well as similar reduction i n air pollution. Similarly, Pendall and Chen (2002) used a sprawl index based
54 on land use density, land use mix, street connectivity and commercial clustering to find that there is a high correlation between these factors and travel behavior as an increas e of the sprawl index decrease s the use of alternative modes Galster et al (20 01 ) conceptualized sprawl metrics that also can be used in suitability models to avoid sprawl. Many researchers of the impact of land use on travel behavior encounter similar i ssues regarding the research undertaken in this area thus far. Firstly, using aggregated data and macro scale of analysis where the unit of analysis might be larger than a county level. Secondly, the mix of economic variables with land use variables withou t controlling any of them mak es it difficult to identify the land use effect on travel behavior in a manner independent of the characteristics of the traveler Thirdly, the research effort to build indices using disaggregate approaches can be spatially rep resented as surfaces but these indices describe an existing condition and do not predict future trends. There is also some interdependency between urban form variables. For example, increasing the commercial density may increase the accessibility for shopp ing as the opportunity increases. Furthermore, the 5Ds concluded from Ewing and Cervero show that increasing density will also increase transportation options but at the same time decrease travel speed and increase travel congestion (Litman, 2008). The 5D s research shows that diversity and land use mix affect travel behavior. However, the literature shows differences between researchers in how they are measured. Cervero and Kockelman (1997) used entropy and dissimilarity to capture the land use mix of a ne ighborhood, in addition to the density and design variables used in
55 a regression analysis to find the impact on transportation through modal choice and VMT. Entropy and dissimilarity are indices that are calculated on a neighborhood scale to capture land u se mix. The entropy measure in general takes the percentages of land use mixes in a neighborhood to build an index. The entropy index developed by Frank and Piv o (1994) describe s the evenness of the distribution of built square footage among seven land use categories. The aforementioned dissimilarity index was developed by Cervero and Kockelman (1997 ) This index was based on the dissimilarity of a hectare use from the adjacent eight hectares that surround that specific hectare. The average of hectare accu mulations across all active hectares in a tract is the dissimilarity which is an indication of the land use mix in that tract. The reason for using the dissimilarity is that using entropy lacks the ability of capturing the distribution of land use mixes sp atially, while dissimilarity captures the variety of different land uses that surround a certain land use and thus captures the spatial pattern for different land uses. A Transit Cooperative Research Program [TCRP] (2003) report showed that increasi ng dens ity increase s modes of transportation other than cars An e ntropy index, accessibility and dissimilarity indices were also studied, with the accessibility and entropy indices being the most efficient in capturing the travel behavior. Steiner et al (2010) studied the effect of land use mix at trip end to come up with an equation to predict trip length for new development. They found that the entropy measurement proposed by Frank and Piv o (1994) was insignificant for their study in south east Florida and the y suggested a model for calculating the land use mix as a percentage between residential and non residential uses.
56 The literature indicates the importance of the impact of the 5Ds on vehicle trips and on the VMT. CNT (2007) used similar indicators in a re gression model for travel cost. Steiner et al (2010) also used similar variables to model trip length. However, it is not difficult to conclude from the literature that the 5Ds have an impact on travel cost. Generally, from the literature on the indirect effect several conclusions can be drawn. Firstly, the frequently used 5Ds, density, diversity, design, destinations and distance are the most important variables impacting transportation. Secondly, urban form indices other than the 5Ds can used to capture the impact of urban form on travel cost. Thirdly, many of the existing research has been performed using statistical approaches and performed on aggregate level data and analysis. Fourthly, it is clear that these variables affect the travel cost and VMT an d therefore can be considered important variables in a housing location choice model and in deciding the relationship between housing and jobs. This research will use statistical methods on parcel level and local data to show which variables impact travel cost and use them in the prediction of future travel cost values that can be modeled spatially. Low transportation cost and connectivity to transit are important factors in determining affordable housing locations, however, at the same time these may conf lict with other location costs such as the cost of housing. Therefore building index surfaces for these variables is important because it can work together effectively in the allocation of affordable housing. This research will use spatial interpolation an d regression methods to build travel cost surfaces based on travel surveys for a certain year and will modify the cost for a projected year according to the land use variables at the trip ends such as density, diversity represented by entropy values, desig n represented by
57 connectivity, destinations represented by the regional accessibility to employment and the distance which represent the distance to major transit stations. Advanced Methods of C apturing Travel Behavior The research trend for the coordination between land use and transportation aims to the highest integration of transportation and land use models at the highest possible disaggregation level. This can be seen from the trend to apply parcel level analysis and the use of activity bas ed models. Many effort s can be seen i n research approaches that try to use activity based models ( Ben Akiva & Bowman, 1998). Activity based modeling is a new trend in transportation modeling research even though it has existed since the 1980s. These model s process activities with their relevant time s and thus allow more consideration for air quality and congestion problems through the sp ace temporal dimension. If activity based models are integrated with disaggregate land use models the result will be a la nd use and travel model that corresponds to the behavioral integration of the choice s across their relevant times. Recent trends in transportation research also have seen a growing focus on multi modal transportation systems including interaction between all modes of transportation, such as driving, transit, bicycling and walking. Traditionally, transportation modeling is represented using a f our step transportati on model The first step is trip generation where trips are divided into home based and non home based trips produced from a place and attracted to another place. In the se cond step, trip distribution, t he trips are distributed between production zones an d attraction zones or the number of trips interchangeable between zones. The third step, moda l choice, involves the choice between driving, transit, bicycling and walking. The final step, network assignment, concentrates on routes and the transportation ne twork (Kutz, 2003).
58 Researchers disagree about trip generation in transportation models such as the four step model. Generating the production of trips generally depends on household characteristics such as income, sex, number of people in the household, n umber of cars etc. and depends on household surveys for the number of trips. A statistical regression model can be generated. The regression equation is used to forecast the number trips. However, this regression model is not sensitive to land use variable s. Iacono and Levinson (2008) investigated the land use and transportation network elements in generating trips and showed that the transportation network characteristics play a role in generating trips and the transportation network affect the location ch oice. The four step model is trip based and does not include trip chaining; therefore, the time of day is not scheduled and thus ignored. The use of time in the four step model is limited to certain uses in deciding ( i.e. the peak count of trips). The fo ur step model also does not take into account the interdependence between trips and it divides trip into home and non home based trips and thus does not distinguish between a single stop home based journey and a multiple stop journey (Bhat & Koppelman, 199 9 ). Ben Akiva and Bowman (1998) studied the behavior realism in urban transportation models. Their research gave a more realistic representation of travel patterns. Their paper also followed the evolution of transportation models from models that represen t the day schedule as isolated trips which is generally the original trip modeling of four step models to modified models that combine trips explicitly in tours or chains of trips and finally to more sophisticated models that combine the tours or chains to their respective times in a day schedule. In the trip based model s, the trips are schedule d as independent one way trips, with no relation ship between trips. In the tour based model
59 trips are connected in tours with spatial constraints an d direction of movement. For the chains or tours attached to a day schedule, the model links the sequence and t iming of activities across the chain. The development of such models also led to the inclusion of behavioral varia bles in the model (Ben Akiva & Bowman, 1998). The trip chaining and the time of day is important for studying congestion and to estimate travel cost that do not depend only on free flow conditions. However, the applications of trip chaining and activity models are not always possible in terms of data availability. Land use characteristics do not only affect travel distance as shown in VMT research. It also affects the number of generated trips (Ewing & Cervero, 2001). In the four step model, the trip attraction, which is a function of land use, is cal culated by an empirical equation generated using research results in each Metropolitan Transportation and Planning Organization (MTPO). In land use models like UrbanSim a modified five step model is used to replace the four step model. This five step mode l accounts for the interaction between the transportation modeler and the land use. This fifth step uses density and transit accessibility measures as a feedback loop between distribution and assignment. The model also uses a nested logit model for home based work (Franklin et al 2002) The modification of the four step models indicates the problems facing the MTPOs in using the four step model and the need for more advance and comprehensive transportation models. Pozsgay and Bhat ( 2001 ) used demog raphic characteristics as well as spatial indicators to identify the impact of spatial location on travel behavior using destination choice models. Destination choice models are also used by Arafat, Srinivasan and
60 Steiner (2009) to relate the choice of des tination to the urban form variables such density, accessibility and distance. Bhat and Lawton (1999) also addressed the importance of travel forecasting and the importance of integrating land use and transportation models, especially the movement from tri p based modeling to activity based modeling. This move includes treatment and scheduling of tours as a series of linked trips that the traveler performs as a series of activities from the moment he leaves home to the mom ent he returns to home (Bhat & Lawto n 1999). Waddell et al (2003) commented that the advancement in research on modeling the location of urban developments, real state and the analysis of travel behavior is to use the activity based approach. A major development in the UrbanSim model for Sa n Francisco was the shift in paradigm to use a tour based travel model instead of a trip based or four step model, originally integrated with UrbanSim The activity based integration for the short and long term is to be performed in broader future research projects (Waddell et al 2003). The literature shows clearly the new trend in transportation research as focusing on activity based modeling for transportation and the use of discrete choice modeling to relate the transportation variables to land use. U nfortunately, the use of activity based modeling will not always be pos sible in this type of research due to the data needed for the model. However, trip chaining is still needed because people may use more than one mode of transportation in their daily tr avel activities. In the absence of the data needed to run activity models and the poor handling of the four step model in prediction at the parcel level, this research uses a longitudinal approach involving regression to
61 relate travel survey data to urban form variables. It uses the relationship of these variables for future years by assuming that that the relationship between travel cost and urban form characteristics will remain constant. The next section will explain the literature for estimating and pre dicting travel cost. E valuating Methods for Predicting Travel Cost The interdependency between urban form variables and their relationship to trip length has been also the focus of VMT and travel cost research. Different methods have been used to generate trip miles or VMT using statistical approaches. One method used the Delphi panel consensus for elasticity values linked to the 4Ds (Lee & Cervero, 2007). According to the 5Ds research, the increase of any of the Ds values will decrease VMT, reduce the num ber of auto trips generated and increase the share of tr ansit and walking modes (Ewing & Cervero, 2001). Lee and Cervero (2007) summarized the value of elasticity for VMT. These values can be applied in a post processing procedure to update the predicted t rip according to the change in la nd use at the trip ends. Table 2 6 shows the elasticity values for the purpose of updating VMT. The other method to estimate and predict trip length is using regression, taking the trip length as the dependent variable and the land use on trip end as independent variables. Steiner et al (2010) applied this methodology on South East Florida travel diary data which included the positional coordinates at the origin and des tination of each trip. Table 2 7 shows the result of th e regression model to estimate the trip length of the home based work and non work trips and non home based trips. Back calculating of the trip length according to the previous model can generate a spatially discriminated prediction surface for the trip length on a parcel or neighborhood level.
62 Based also on the statistical models, CNT (2007) research developed the HT index, which is an equation for the combined location cost for housing and transportation with respect to income for the house hold. T he variables included in generating this cost are housing cost, travel system characteristics which include road and transit connectivity and density, and the distance from employment. However, they used the Census 2000, Census Transportation Planning Pack age CTPP 2001 and local data to statistically generate the travel cost which mainly includes travel miles, car ownership cost and transit rides. Furthermore they use general categories such as low, medium, and high to describe the access or transit frequen cy. The level of analysis can be d escribed as aggregate. Table 2 8 compares the variables used for the travel cost estimation between di fferent methods while Table 2 9 compares the merits and limitations for each method. The application of discrete choice or geographically weighted models is data dependable. This research will not focus on using discrete choice models in predicting trip cost or generating a cost suitability surface. Instead, this research will use ordinary least squares and geographically w eighted regression to predict travel cost and to generate a travel cost suitability and preference surfaces. Using Conflict Identification Strategies to Identify the Conflict between Transportation and Land Use The conflict identification strategy (Carr & Zwick, 2005), can be applied on any set of conflicting objectives and goals. It was used in LUCIS to identify the conflict in land use but the same methodology can be used to identify the conflict between transportation planning goals and environmental go als or between affordable housing goals and transportation goals.
63 In the affordable housing suitability structure four main objectives may conflict with each other. These ob jectives are: (1) Planning for housing sites in terms of land physical characteri stics as well as neighborhood socio economic characteristics; (2) The objective to relieve the housing burden for low income families; (3) The transportation planning objectives to increase access and reduce VMT and travel cost; and ( 4 ) The objective to increase transportation options like increasing transit access and walkability The process of conflict identification should consider the planning for each objective in isolation of the other objectives. That means when planning for hous ing preference the planner works on the site characteristics and socio economic characteristic. Therefore the planner will not take into consideration the cost of traveling to work from residential locations. The planner concerned with housing over burden will work to identify places that are of low housing burden. The transportation planner will plan to reduce VMT and travel cost as well as increasing accessibility without taking into consideration the site preference for affordable housing or the work of the first planner. The same applies to the fourth objective, planning for increasing transit accessibility without taking the other objectives into account The main criterion for using conflict identification strategies exists in the affordable housing st ructure. Building Allocation Scenarios LUCIS works to identify potential future land use conflict. This conflict can be used in various applications one of which is future land use allocation scenarios (Carr & Zwick, 2007). The future allocation process based on LUCIS outputs can be seen as identifying the land for future growth and the allocation of people to residential, retail,
64 and industrial land uses, based on densities and/or any other factors that a planner might wish to use in the allocation proce ss. The allocation procedures in the LUCIS model can be automat ed by introduc ing a set of GIS tools for future allocation, scenario buildi ng and testing of policies. For example, the automatic allocation of urban land using the LUCIS procedure can be easily understood according to the category it fits in. Three main categories can be identified for the allocation process. The first is an infill category where some of the vacant land should be allocated for use before using other land c ategories. The second is a redevelopment category where allocation is performed on existing uses that are convenient to redevelop ; and the third will be the use of agricultural land in a green field category which is based on the urban, conservation and ag ricultural conflict. Using GIS facilitates the allocation process where the identif ication of land and the allocation process will be performed according to priorities. These priorities may depend on growth patterns, proposed densities, transportation mas ks, etc. The complex procedure, the accuracy and the time spent in the allocation process generates the need for an automated procedure that can perform the allocation in a more feasible and flexible fashion. Similarly, the automatic allocation process can be performed using the affordable housing conflict surface, in addition to any other conditions and priorities in preparing scenarios for the allocation of affordable housing. The literature review summarizes the literature on the creating and evaluating the suitability components that are used in the affordable housing suitability models such as neighborhood access to services, transit accessibility and travel cost. The literature review also summarized the literature on location choice and suitability mo dels. The
65 review highlights the importance of urban form characteristics in reducing travel cost and the importance of accessibility in land use modeling. It can be concluded that these variables should be used in modeling land use. However, using transpor tation variables such as accessibility and travel cost as suitability surfaces has not been implemented yet in the literature of land use modeling. The literature review also summarizes the trend in capturing travel behavior through the use of activity mod els. Even though this dissertation does not focus on capturing travel cost using activity simulators, the suitability model structure adopted in this research is flexible and can be updated to take activity models into account when the activity diaries dat a for these models becomes available in Florida.
66 Table 2 1. Scale for p air wise c omparison ( Saaty 1980). Definition Intensity or Importance Extremely more important 9 Very strongly to extremely more important 8 Very strongly more important 7 Strongly to very strongly more important 6 Strongly more important 5 Moderately to strongly more important 4 Moderately more important 3 Equally to moderately more important 2 Equally important 1 Table 2 2. Preference v alue d escriptions. Cells with a value of: Indicate 1 low preference 2 medium preference 3 high preference Table 2 3. Conflict s core m atrix ( AHS goal 1 and travel cost ). Goal 1 Preference Travel Cost Low Preference 1 Travel Cost Medium Preference 2 Travel Cost High Preference 3 Low Preference 1 1 *10 + 1 = 11 12 13 Medium Preference 2 2* 10 + 1 = 21 22 23 High Preference 3 31 32 33
67 Table 2 4. Variables used in accessibility equations. Article Distance Network Distance Opportunity Gravity General Gravity Hanson Remarks Carr and Zwick (2007). D D Distance Levinson and Krizek (2008). D or TM O O, TM, TM Travel Miles Decay Factor Arafat, Steiner and Bejliri, (2008). ND ND Network Distance Bhat et al. (2002). TM, TT O O, TM, TT Travel Time Handy (2004). O Hanson (2004). O Waddell et al. (2003). O, TM, Grengs (2009). Srour et al (2002). D O (Driving time buffer) Bhat and Guo (2004). TT O,TM,
68 Table 2 5. Accessibility matrix (Merits and limitations for different accessibility estimation methods) Distance Network Distance Opportunity Gravity General Gravity Hansen Merits Easy to calculate. Spatial surface can be generated directly by raster analysis. Good access estimation for highly connected locations. Precise measurement of proximity distance. Easy to estimate on zonal level. Used by many articl es as accessibility estimation. Easy to estimate. Easy to estimate on zonal level. Value the attraction a nd distance in the model. Considered more precise estimation of accessibility. Easy to apply on zonal level Applies on travel time from travel survey or forecasts. Limitations Poor estimator for poorly connected places and neighborhood s with higher block sizes. Estimate only distance and does not include attraction in the estimation. Complex and time consuming to generate accessibility surfaces by this method. Estimate only distance and does not include attraction in the estimation. The equation wil l estimate that large attractions are more accessible even though they might be far away. Applied on zonal level and complicated to be applied on parcel level. Applied on zonal level if the travel time for each zonal pairs is estimated. Complicated to apply on parcel level. Table 2 6. Elasticity values for VMT (Lee & Cervero, 2007) Variable Elasticity Density 0.05 Diversity 0.04 Design 0.20 Access 0.05
69 Table 2 7. Trip length associated with residential parcels (Steiner & Srinivasan, 2009) Explanatory Variable Home based non work t stat. Home based work t stat. Non home based t stat. Constant 7.621 12.592 11.099 Developed area as a % of total neighborhood area 2.395 4.426 Residential area as % of developed area 1.682 Building square feet (retail commercial) 2.527 4.408 Building square feet (office/service) 2.527 2.445 4.408 Building square feet (industrial) 2.776 Road miles per developed area 3.015 3.793 Number of intersections per road mile 2.722 Distance to nearest regional residential center 3.925 1.908 Distance of nearest regional activity center 2.968 5.528 Range of distances to regional activity centers 1.330 2.175
70 Table 2 8. Travel cost matrix (Variables) Variable Steiner and Srinivasan (2009) Lee and Cervero (2007) CNT (2007) Density Ratio of d eveloped area Residential area per d eveloped area Residential Density Dwelling per square mile. Retail and c ommercial density is taken as floor area ratio. Net density Gross density Diversity Retail SQF Office SQF Industrial SQF Entropy v alue: R esidential / r etail/ commercial / services. Job d ensity. Design Road miles per developed area. Number of intersection per road miles Street grid density. Average block size. Destination Distance to nearest regional activity center. Distance to regional residential center. Range of distance to major activity centers. Regional a ccess t o major destination zonal gravity model. Distance to employment centers. Other Variables None None Neighborh ood access to amenities, household income, house hold size, Transit connectivity
71 Table 2 9 Travel cost matrix (Merits and limitations). Variable Steiner and Srinivasan (2009) Lee and Cervero (2007) CNT (2007) Merits Includes aggregate and disaggregate data Use overlapping neighborhood definition to limit the effect of the modifiable unit area problem. Can be replicated nationwide. Can use the output of the four step model in case of no travel survey is available. Calculates t he monetary cost Calculates the affordability index directly. Limitations Calculate t he trip length for work and non work trips without calculating the monetary cost The model depends on the travel surveys of South East Florida and is not tested for it is prediction accuracy in other regions. Course spatial resolution Does not include local access. Does not estimate travel cost directly but via post processors. Coarse spatial resolution neighborhood level data Few land use d escriptors. Datasets are on t he aggregate level of neighborhood and TAZ Very rough connectivity measurements for roads and transit.
72 Figure 2 1. Pair wise comparison matrix.
73 CHAPTER 3 RESEARCH METHODOLOGY Study Design: This research build s GIS simulation tools that integrate transportation variables within a land use suitability model. The suitability model is used for the identification and preservation of land suitable for affordable housing in three counties in Florida. The study desig n is mainly composed of five parts: 1) Evaluation of methods to create suitability based on neighborhood accessibility to services; 2) Empirical analysis to establish models to create or predict travel cost based on travel surveys; 3) Preparing and validating GIS and simulation tools that use accessibility and travel cost in the allocation of affordable housing; 4) Using the allocation model to test the impact of accessibility and travel cost on the affordable housing location choice : and 5) Evaluating the automation tools to allocate affordable housing in the three case study areas. The simulation tools are prepared to work within a ArcGIS interface using both an interactive user interface and an integrated model building interfac e supported by ArcGIS model builder. Generally tools and automated models are prepared to work under the GIS environment. LUCIS suitability models typically use proximity to services as suitability indicators for residential parcels. The nearer the parcel is to services, the more suitable the parcel will be for a residential location. This research, however, uses neighborhood accessibility to services suitability model. The selection of a neighborhood accessibility estimation method is based on comparing th e estimation results between different methods of accessibility estimation. Generally accessibility estimation in this research is
74 based on parcel level data and replaces the traditional proximity tool in LUCIS models. ArcGIS Model Builder is used to creat e an automated accessibility model which also includes new tools for building and reclassifying the final suitability surface for accessibility. These new tools are programmed using Python programming language and prepared to be used as tools in Model Buil der or the ArcGIS interface. Estimating travel cost is also an important component in the Affordable Housing Suitability model. Two methods are used in the estimation of travel cost spatially. The first is a spatial interpolation method. The interpolation works for geographically mapped trips taken from the NHTS (2009) dataset The second method involves the generation of a model that relates the travel cost to land use and urban form characteristics. This model uses ordinary least squares and geographical ly weighted regression to estimate the travel cost based on the NHTS and parcel data for the year 2009. LUCIS suitability models depend on visual simulation using ArcGIS and are based on a deterministic approach. A program for weighting utility surfaces based on pair wise comparison and Delphi panel scoring is prepared. This model is associated with ArcGIS Model Builder through the use of a special layer weighting tool prepared using Python programming. This program helps in scenario building where differ ent scenarios can be run interactively during expert or community meetings. However, a tool to help planners in the selection process for affordable housing locations is also prepared by Python programming. This tool works on the final conflict surfaces th at are generated using LUCIS. The goals that are used in the conflict surface may include site characteristics, neighborhood characteristics, travel cost, housing cost and transit access. Using additional conditions such as incentives and priorities, the t ool can help the planner
75 interactively in the selection of new affordable housing locations or in the analysis to preserve existing affordable housing sites. Model Structure The affordable housing suitability model is an automated GIS based suitability mo del that aims to help local communities in Florida plan for sustainable affordable housing. The AHS research project is intended to build The Affordable Housing plan for attractive, eq uitable, and sustainable affordable housing. The project goal is t o evaluate and identify lands in Florida communities that are suitable for the development and/or preservation of affordable housing based on local preferences and planning expertise (AHS, 2009, p. 3) To achieve the goal, the model uses LUCIS conflict identification strategy and the population allocation procedure using combined grids ( Carr & Zwick, 2007) The LUCIS l ayer overlay methodology uses local community preferences and includes i d entify ing lands that are suitable for residential development. However, additional overlay layers are used in the AHS models that are specific to affordable housing residential locations. The conflict surface is the second step in the process which again u ses the LUCIS methodology to find places that are suitable for affordable housing based on travel cost, housing burden and transit accessibility. The population allocation procedure refines those locations further by adding more affordable housing suitabi lity and cost indicators such as socio economic conditions, policies and incentives This research will focus on objectives that are linked to lowering the transportation burden such as compact development and/or Transit Oriented Developments (TOD). This includes the use of transit accessibility and location specific cost which includes the travel and housing costs. Figure 3 1 shows the structure for the suitability model and allocation procedure.
76 T he suitability model generates an opportunity surface by c ombining four affordable housing goals. The opportunity surface is then combined with surfaces that represent policy incentives and allocation conditions to generate the combine grid used in the allocation process. The allocation tool in Figure 3 1 is a ne w ArcG IS tool created using Python scripting language. This tool allocate s the affordable housing land based on instructions given in a scenario table Computer Requirement and Data Sources The computer requirements can be divided into two categories; co mputer requirements for development, and computer requirements for model application. In the model development, the analysis and model preparation include running network analysis to calculate distances on a parcel level and needs extensive computer comput ation for millions of records. Therefore the computer used for the model preparation is a quad pro Intel core computer Q9600. This computer has a processor speed of 2.5 GHZ for each of the four processors and has a memory is 8GB and 1TB of disk space. How ever, the model application does not need a high specification computer. The models and tools are tested to run on a DUO core II computer with a 2GB of memory. Running the tool on a lower efficiency computer is not tested and may have some limitations. In terms of data, the research used the most recent data sources from the Florida Geographic Data Library FGDL, data at d geographic levels from the U.S. Census Bureau, National Household Travel Survey 2009, data from local transit agencies and other local d ata sources. Table 3 1 shows some of the primary data sources used in the research.
77 Level of Analysis and Selection of Research Areal Unit s Many of the urban form characteristics are measured on an aggregate scale (Hess et al 2003) H owever, the new trend in transportation planning is to perform disaggregate analysis in time and space (Wegener, 2005; Johnston 2004). The characteristics of a predefined areal unit affect the values of an urban form measurement. This problem is known as the Modifiable Ar eal Unit Problem (MAUP) (Guo & Bhat, 2004; Kwan & Weber, 2008). The focus of this section is to explain the methodology on the use of an optimized zoning procedure to reduce the effect of the MAUP on the estimation of urban form patterns. To do that, an ex tensive GIS iterative analysis will be perform ed on three counties in Florida. To minimize the zonal boundary impact, every parcel will be taken as a center of a zone that will capture the urban form characteristic surrounding that parcel, and then the ana lysis will be repeated for different zone sizes and shapes to find an optimal size for the zones. The research show s that the use of natural neighborhood boundaries as an areal unit may lead to undesirable results because of differences in size and shape, and suggests the use of GIS tool s that use the optimal neighborhood in measuring urban form characteristics. Urban forms have impacts on transportation and the transportation systems have impacts on urban form and land use. The impact of transportation on land use is captured via a simplified accessibility measurement while the impact of urban form on transportation is captured by many urban form measurements such as density, land use mix, connectivity and accessibility. Some of these variables are better used on parcel level data while other variables are better used on an aggregate scale.
78 Density, d iversity (land use mix), design (connectivity), destination (accessibility) and distance (5Ds) are used to study the interaction between land use and transpo rtation, especially with regard to the impact of land use on transportation in terms of vehicle trips and VMT (Ewing & Cervero, 2001; Ewing et al 2007; Lee & Cervero, 2007). The five Ds in general represent some of the measurements for the urban form cha racteristics. Therefore, their impact is not isolated from scale and zoning problems. The new trend in transportation planning is to perform disaggregate analysis in both time and space. Disaggregate measurements can be performed on urban form measurement s such as accessibility. Urban form characteristics such as density, diversity and connectivity are usually measured on an aggregate scale (Hess et al 2003; Dill, 2004 ) In the MAUP, the scale and zoning of the areal unit affect the value and variation of the aggregated entity. The MUAP is a recognized problem that has faced researchers in the modeling and calculation of entities that have a spatial representation. The scale and zoning in this research are represented by the areal unit characteristics wh ich include but are not limited to the size, shape and location of the areal unit. GIS can be used to reduce scale and zoning issues and offers solutions to reduce the MAUP GIS Spatial Analyst offers the tools for defining the areal unit and running iter ative analysis on differe nt sizes and shapes. This research present s the impact of size and shape of the areal unit on the urb an form characteristics and present s a method to optimize the neighborhood size and shape. For the location of the areal unit,
79 th e research present s a floating areal unit method to reduce the effect of the zoning and adjacent neighborhoods limitation. A first step to obtain the optimal neighborhood for capturing land use and transportation characteristics is researching the effect of size and shape as well as the boundaries of neighborhoods used for the data aggregation. To do that, preliminary research is performed using variable neighborhood size around each parcel to study the effect of neighborhood size and shape on the land use mix entropy value. The same method is applied on other urban form characteristics such as connectivity and density. Th is methodology accounts for the scale component of the MAUP by taking a variable size neighborhood. The methodology also accounts for the zoning component by using a floating areal unit as well as different unit shapes. The land use mix in this research is captured by the entropy index Cervero and Kockelman (1997) used entropy on a neighborhood scale to capture the land use mix. In genera l, entropy measures the percentages of the land use mixes in the neighborhood to build an index. The entropy index developed by Frank and Pivo (1994) describes the evenness of the distribution of built square footage among seven land use categories. The fo llowing equation shows how entropy is calculated for several land uses. Where, p i is the proportion of specific land use to all land uses in each catchment area k the total number of land use categories.
80 The other two variables studied here are connectivity and density. For neighborhood connectivity, a simple measurement of road density is used as a connectivity indicator. This measurement is primarily miles of roads per unit of ar ea (square miles). Other connectivity measurements can be used too, such as the number of links divided by the number of intersections in the neighborhood (Dill, 2004). Arc y for a certain defined areal unit. For the density, this research uses a unit per acre variable from the property appraisal data. These density values are aggregated and averaged to the defined areal unit. The value of the average density is assigned to t he center point of the areal unit. To eliminate the zoning boundary problem, a floating areal unit is used where each point in the map is a center of a defined neighborhood. In summary, a multi level analysis for land use mix, density and connectivity is performed using a neighborhood size that is increased incrementally Because of the zoning component of the MAUP, each parcel will be taken as a center of this neighborhood and the analysis will be replicated for all parcels. The result surfaces will be c ompared by ArcGIS statistics by calculating the mean and standard deviation for the change in value and standard deviation due to the incremental change in size. Figure 3 2 shows the scale component by displaying the variable size unit area around the parc el. The smallest size is a 0.5 mile and the increment is taken as 0 .5 mile To study the effect of the areal unit shape, three different unit shapes are studied. These shapes are a square, a circle and a diamond. The square shape represents aggregated parc el blocks in a grid pattern system. The circle represents the Euclidian travel distance, and the diamond shape represents the Manhattan travel distance. The
81 results of the three shapes are overlaid with the street network to obtain the optimal areal unit s hape. ArcGIS Model Builder models and customized Python scripts are used to calculate and create surfaces for the land use mix (entropy), density and connectivity using floating areal units. Multiple shapes and sizes are used in the analysis. The models a nd scripts calculate the value for each parcel assuming that it is at the center of the areal unit. After generating the surfaces, descriptive statistical methods are used to obtain the optimal neighborhood size and shape. Introducing Transportation Variab les as Suitability Surfaces: Transportation land use coordination impacts on each other are identified by literature and were categorized in the literature review in terms of the impact on location choice as direct and indirect effects of transportation. A ccessibility is identified as the direct impact of transportation on location choice and specifically on the choice of land for affordable housing. There are many definitions for accessibility in the literature Many of these definitions are operationalize d and used in the estimation of the urban accessibility index ( Bhat et al 2002). This research uses and compares different accessibility measurements on local and regional levels. For the indirect impact direction in the land use transportation coordina tion, the variables are mainly characteristics of the urban form that can be seen to be of importance because they affect transportation which in turn affects the choice of location. These variables are density, diversity, design, destination and distance. There are three approaches that are taken in this research which are as follows: 1) Using travel trip data in spatial interpolation to generate a spatial distribution of travel cost;
82 2) Using urban form measurements and trip data in a global statistical model and use that statistical model to predict travel cost; and 3) Perform a geographically weighted regression to relate the travel cost to the urban form characteristics. Investigating Neighborhood Accessibility as a Suitability Surface: Traditionally, proximity as straight line distance has been used as an accessibility measurement in deterministic land use model ( Zwick & Carr, 2007) while gravity models have been used in statistical a nd stochastic models ( Waddell, 2002). There are also other measurements of acc essibility used in statistical models such a s opportunity access ( Handy, 2004; Hanson 2004). However, using gravity models on a parcel level scale requires generating huge origin destination matrices that contain billions of records for the study areas. Th ese origin destination matrices exceed the capacity of hardware and software used in the analysis. Therefore, a combination between estimating access by opportunity and distance is used in the research. The distance estimation component can be performed by network or Euclidian distance. These distance estimation methods can be also used to generate the capture area for opportunity estimation which has historically taken Euclidean distances. Arafat, Steiner and Bejliri (2008) discussed the differences betwee n capture areas based on network and Euclidean distances in terms of the number of students for within a walking distance to schools and recommended the use of network distance to generate the capture area. However, the use of network distance is time cons uming and may not be practical for all planners in terms of hardware and software limitations. Therefore, this research agrees that the network distance based estimation is better than other estimation methods and, in terms of accessibility, both proximity and opportunity are used as estimators as shown in the literature review. However, multiple utility
83 assignments give a method to combine the suitability based on proximity and the suitability based on opportunity in a multiple utility surface. Therefore, in terms of suitability assignment and based on accessibility measurement methods, it is not difficult to conclude that the network based combined opportunity distance estimation can be regarded the best possible suitability assignment for accessibility t o services on a parcel level. However, there is no right or wrong method for estimating suitability based on accessibility to services. This research investigates and compares the result of neighborhood accessibility estimation methods that the planner may decide to use. Generating Neighborhood Accessibility as a Suitability Surface Distance p roximity s urface This is the simplest model to measure accessibility. The Euclidean distance is reclassified according to standard deviation of the normalized distanc e by the mean distance away from the multifamily parcels which is regarded as the base for suitability reclassification. The ArcGIS model in Figure 3 3 generates this suitability surface. In the model, an ArcGIS customized tool (A4 Suitability ) is prepared using Python programming to reclassify the Euclidean distance raster and assign the suitability values based on their distance to services referenced to the average Euclidean distance to multifamily parcels. However the shown model measures and reclassifi es the Euclidean distance. A network distances suitability surface is also generated using the Nearest Neighborhood tool (Arafat et al 2008) which is an ArcGIS customized tool that measures the distances from all parcels to their nearer destinations and generates a network distance raster instead of a Euclidean distance raster. The same reclassification procedure is used for the network distance. This research generates and compares both Euclidean and network distance proximity surfaces.
84 Opportunity s urfaces The opportunity access can be defined as the number of opportunities within a defined area. It is also can be weighted to include the floor area of services according to the following equation: Where: O i = the weight or the attraction for a facility C i = 1 for Di < buffer distance and zero else where Therefore, the opportunity is calculated by accumulating the attraction value of the service which depends on the type of service. This attraction could be the square feet for retail services, the number of beds for hospitals or any other criteria t hat can be used to discriminate our preference of one service to another. The model captures all th e opportunities that lie within the buffer which here is used as the walking and biking distances Th e two output surfaces a re combined to get a multiple uti lity assignment (MUA) surface that includes suitability based on opportunity within walking and biking distances The model in Figure 3 4 is prepared to generate that surface. However, the model in Figure 3 4 uses a Euclidean buffer which is an exagerated buffer to be used as a capturing area for oppurtunities. The model is modified to take the cummulative oppurtunity inside a dynamic network buffer for both walking and biking and use the opportunity value to generate the suitabilty raster. Gravity and oppo rtunity distance s urfaces According to Bhat et al (2002), the accessibility measure is summarized into different equations for cumulative opport unity and gravity measures that are classified
85 as Gaussian, composite impedance, activity distance and in vehicle travel time measures. The equation for c al culating access from gravity is: Where O j is the weight or the attraction for the facility d ij is the distance from each to each destina tion J is the number of destinations connected to each origin is the distance decay factor The distance as impedance estimation is a gravity estimation. However, the gravity model is a simple model for use on a zonal level such as TAZs. On a parcel le vel, the model will generate an origin destination matrix that contains hundreds of millions of trip combinations for one county. Applying the gravity mathematical equation at that level is impractical in terms of hardware and software limitations. However in this research the gravity approach is used to capture the regional accessibility to major destinations used to estimate the travel cost in this research. This is because the limited number of destinations will dramatically reduce the number of trip co mbinations. This will be further explained under the travel cost estimation section. The combined opportunity distance surface can be regarded as a modified gravity approach and is generated as an MUA surface that combines the aforementioned distance and opportunity surfaces. The surface will have the same gravity elements which are the re tail density and distance but each generated as suitability surfaces and combined deterministically to generate the final suitability surface. This surface can be
86 generated based on network distance using ArcGIS network distance estimation method. However, because estimating the network distance is time consuming, planners may avoid using the network base and prefer to generate a Euclidean distance based opportunity distance access estimation. Investigating Travel Cost Travel survey data Travel cost is imp ortant in deciding residential location. CNT (2007) have used travel and housing costs to build a HT index for each income group. The HT index is used by CNT to determine the affordability of housing site locations. In this research travel cost will be use d as a suitability surface in a model to find the suitable places for affordable housing. Many research approaches can be taken that include travel cost as shown in the literature review. The first method used in this research is to estimate the travel cos t spatially is a spatial interpolation method based on the geo coded data for trips in the NHTS 2009 trip diary. Figure 3 5 shows sample trip end points for Duval county. These trip ends are geo coded trip end locations for the trips included in the travel diary for that County. In addition to the geo coded location, The trip diary includes the purpose, mode and the reported trip miles for each trip. Investigating NHTS 2009 trip frequency data shown in Table 3 2 in terms of trip length, we can find that nea rly 95% of the trips are less than sixty miles long. Figure 3 6 shows the distribution of those trips according to the trip purpose while Figure 3 7 shows percentages of trips classified according to their length and type. From Fi gure 3 7 we can conclude t hat more than 95 % of the trips are of lengths less than 60 miles, except for work trips and non home based trips where 95% of the trips are of lengths less than 70 miles. This threshold value for trip length is used to remove the outliers
8 7 from the travel s urvey. Additional outlier analysis is performed to identify the local outliers using AcrGIS. An ArcGIS tool to identify the local spatial data outliers (Anselin Local Moran I) (ESRI, 2011) is used. This method depends on the clustering within a certain nei ghborhood size to find the local outliers. Figure 3 8 shows the interface for the tool while Figure 3 9 shows an example for the outliers identified by tools for Orange County. Spatial i nterpolation This research focuses on the impact of location on travel cost. Therefore, variables considering individual travel characteristics will not be taken into consideration in the research. All the analysis and results are generated for an assumed household of size that is equal to the average house hold size. Other household characteristics, if used, are taken from the average surrounding area characteristics. Generally, after removing the outlier from the data, the trips are classified into five categories as follows: 1. Home based w ork ( HBW ) 2. Home based s hopping ( HBS ) 3. Home based social and recreation ( HBSR ) 4. Home based o ther ( HBO ) 5. Non home based t rips ( NHB ) For generating values for travel cost by spatial interpolation we assume that the spatial location is the most important discriminator in calculating the travel cost. Because the analysis is not performed for a certain neighborhood, the state of Florida NHTS 2009 summary statistics was used in estimating the combination between all trip categories. However for a certain traveler at the travel survey day, the household may not have certain types of trip. Therefore spatial interpolation is used to estimate values for the missing trips using the nearest surrounding trip data. The combination between the trips is performed when all the missing trips are estimated.
88 Different interpolation methods were investigated to estimate the missing trips in each category. These methods include deterministic models such as inverse distance weighted (IDW) and stochastic meth ods such as kriging interpolation. Figure 3 10 shows an example of the IDW cross validation estimation for the interpolation of work trips in Orange County. The interpolation procedure is used in an iterative ArcGIS Model Builder Model environment to creat e continuous prediction surfaces for all the trip categories as shown in Figure 3 11. Figure 3 12 shows the ArcGIS model used to establish the trip mile and the trip cost out of the trip length data for each category. The trip length is transferred to a do llar value using a per mile cost of driving. This per mile car driving cost is posted by the American Automobile Association [ AAA ] (2011) and includes ownership and maintenance costs. The final values of travel cost are represented into two different ways: The first taken is the parcel location travel as a direct output of the model. The second output is to assume that within a walking distance from a parcel, travel costs should be similar and therefore the output will average the travel cost within the wal king distance and assign that value to the parcel in the center. The walking distance is taken as 0.25 mile using a Manhattan buffer. Figure 3 13 shows an example daily travel cost surface for Orange County. Statistical a pproach This approach relates the travel cost as a dependant variable w ith the land use and urban form characteristics as independent variables. These independent variables are timely data sets that include but not limited to residential density, retail density, co nnectivity measurement, a diversity measurement such as entropy and a regional accessibility measurement that represent s the regional employment destinatio n The regression analysis for such a model can be performed by OLS which gives a global
89 regression e quation for the whole county or can be performed using geographically weighted regression which will have different equations representing different geographic regions with different intercept and goodness of fit. However both methods give a predictive mo del that can be applied to find travel cost depending on land use and urban form characteristics or longitudinally by applying the model to a different year using different independent variable values for the specified year. The aforementioned methods ar e used and compared to each other. Global r egression In the global regression approach, urban form and land used characteristics were used as independent variables and the travel cost as a dependent variable. The data set is generated by taken three tho usand random points in each county and creates a table of the estimated travel cost and the corresponding urban form characteristics. This table is used for the OLS regression and was prepared for the three counties. The final set of independent variables may vary from one county to another because of statistical tests and the sign ificance of variables. Generally, the following variables were tested in the global regression model: 1. Density 2. Land use mix 3. Connectivity 4. Network accessibility to employment 5. Transit connectivity 6. Income 7. H ousehold s ize Density is defined as the number of residential units in one acre. However, there are two different density variables tested in the regression model. The first is the parcel density which represents the number of residential units in an acre area of each p arcel. The second is the average surrounding density which is taken as the average parcel
90 density within a 1.25 mile Manhattan distance from the center parcel. This value was estimated for each parcel assuming that it is the center of the surrounding are a. The value of 1.25 mile is chosen because it is the optimal size that reduces the effect o f modifiable areal unit problem as shown in Chapter 4 The land use mix variable is captured by the land use entropy estimation. Land uses are re categorized into f ive categories; residential, retail, services, industrial and other. The entropy equation shown earlier is then applied to all parcels in the defined neighborhood. However, because the entropy is an aggregate estimation method, the neighborhood size, shape and location followed the results of the optimal size nei ghborhood and therefore uses a 1.25 Manhattan distance from the parcel, which is taken in the center of the floating or roving neighborhood. The measurement is then replicated for all the parcels in the county of study. The connectivity measurement captures the design elements of the street network. Three connectivity variables are tested and the variable of the strongest correlation with the travel cost are used. These three connectivity variables a re street density, intersection density and the links/intersection connectivity measurement which is defined as the number of links divided by the number of intersections in the neighborhood. The larger this ratio is, the larger the network connectivity. T he neighborhood size for connectivity is taken the same as for density and entropy to reduce the effect of the modifiable areal unit problem. Network accessibility is taken as a gravity measurement as explained in the accessibility section. In this variabl e estimation, the major activities are identified, which are the major activities in retail and services. Then the resultant activities are integrated
91 within walking distances to reduce the number of major activities. However, the value of attractions is a ccumulated in the integrated points. The major activities are then selected, which are the points that have aggregated activities of one million or more square feet. To reduce the size of the origin destination matrix, the random points taken in the genera tion of the data set for regression are used as origin points and the major activities are taken as destination points. The resultant origin destination matrix that contained the network distances generated by ArcGIS Network Analyst is joined to major acti vities to calculate the attraction for each trip. The trips of each parcel to all activities are used to estimate the accessibility of the parcels to major activities using the gravity access mathematical equ ation explained earlier The transit connectivi ty score is a simple measurement to indicate how connected a parcel is to transit stops. The estimation is performed using a distance weighted density variable taken by the ArcGIS Kernel Density tool to estimate the transit stop density surrounding the res idential unit. For consistency with the density and connectivity variables the same neighborhood size is taken and the density of transit stop is calculated per square mile. The income and household size variables are based on the Census block group level. The Census block group data is intersected with the random points dataset to establish values for the income and household size in that dataset Because the urban form characteristics are different from one county to another, additional independent and bi nary variables are tested for each county separately. For example, the land value taken from the property appraisal data on the parcel level is used in the regression model for Duval County. This variable is initiated as the dollar
92 amount per acre with a l ocal interpolation from the surrounding for missing data. More variables can be also used on a case by case situation. These variables aimed to increase the goodness of the regression model by adding some local variable that can be significant in a county and not significant in another county. Geographically weighted r egression The variables that proved to be statistically significant in the global regression are used in the geographically weighted regression (GWR). The dataset that relates the travel c ost to urban form characteristics can be also geo coded as points distributed spatially in each county. The geographically weighted regression can be based on a fixed or on adaptable neighborhood. The choice between these two alternatives depends on the di stribution of data spatially. Because the distribution of the data is random, taking a fixed neighborhood kernel failed because the kernel may have a large number of points in one place and less points or even no points in another place. The number of need ed points for regression is related to the number of variables which makes the regression fail when using a fixed kernel. In this research an adaptable kernel with a fixed number of points is used. To be consistent between the three study areas, the number of points is taken as the minimum number of points of which the regression succeeds in the three study areas, which is 700 points. The geographically weighted regression creates an equation for each point in the map. Each location (point) equation is gene rated using the surrounding 700 points. This is a sufficient number to perform regression even if one is to use global regression on that location. Research Automation Tools The research automation tool box is composed of four tools for use in the LUCI S model environment. However, these tools can be adapted for any spatial analysis
93 method performed within a GIS environment. These tools have been developed to automate standard processes imperative to developing a LUCIS conflict surface and in the alloca tion process. The first, the A4 Suitability tool, is a utility reclassification tool that reclassifies a utility surface according to a reclassification table or according to statistics based on geographic zones (i.e., zonal statistics). The second tool, the A4 Community Va lues Calculator integrates pair wise comparison calculations into the ArcM ap environment as a Visual Basic for Applications ( VBA ) program. The third automation tool is the A4 Layer Weighting tool. This tool uses the output table gener ated by the A4 Community Values Calculator to execute a map overlay. The fourth and final toolset, the A4 Allocation tools is a set of three tools This toolset automatically allocates land, population and employment based on different scenarios However this section will focus on the first three automation tools. The allocation tool methodology will be explained in a separate section The A4 Suitability Tool Proximity based indicators of change are probably the most important in land use analysis as they integrate transaction costs in determining land use opportunity. Prior to the introduction of the A4 Suitability Tool the planner would take the mean (MEAN), standard deviation (STD), and minimum (MIN) or maximum (MAX) statistics generated from Zonal Statistics to manually calculate the suitability intervals for non binary classifications. Once the values for each interval were determined these values would be manually input into the Reclassify tool. This method proved to be time consuming, cumbersome and prone to error. The A4 Suitability tool functions as a standalone tool available within a custom ArcToolbox or can be seamlessly integrated into a model facilitating a continuous
94 automation procedure. Additionally, the A4 Suitability tool automatic ally generates the reclassification table and output raster. The reclassification table is a listing of the LUCIS suitability index assignments and the ranges of values to which the tool assigns the specific utility value. To determine this utility value, either the average of the mean values for all zones acts as the baseline for suitability and one quarter standard deviation ranges; or data from a table introduced by the user is used to determine the remap ranges. The user can manually modify the remap table produced by the A4 Suitability tool and use the modified table for subsequent model analysis. The A4 Suitability tool output raster is based upon the suitability index values listed within the reclassification table. LUCIS employs two possible suitab ility index classification value ranges increasing suitability (ranging from one to nine) or decreasing suitability (ranging from nine to one). Increasing suitability is best described as the further away a feature (i.e. noise sources) is from its object ive (i.e. residential development) the more suitable the land. Decreasing suitability is best described as the closer a feature (i.e. roads) is to its objective (i.e. residential development) the more suitable the land. The A4 Suitability tool allows the user to indicate the suitability index as decreasing or increasing within the A4 Suitability tool interface. If the user chooses the decreasing suitability option, the tool will use the mean and a one quarter standard deviation to compose ranges that corre spond to the suitability index values from nine to one starting with a suitability index of nine for all values up to the MEAN value and decreasing by one quarter standard deviation increments for eight intervals between the MEAN and MAX value (Figure 3 1 4 ). Since the suitability index one is the last value calculated this value
95 range may be larger or smaller than the other eight suitability index ranges. If the one quarter standard deviation value is less than the cell size then the suitability index v alues will be divided into equal intervals between the MEAN and MAX value. Increasing suitability is calculated in a similar manner. If the user chooses the increasing suitability option, the tool will prepare suitability index values from one to nine, st arting with a suitability index of nine for all values above the MEAN and decreasing by one quarter standard deviation increments for eight intervals between the MEAN and MIN (Figure 3 15 ). Since a suitability index of one is the last value calculated thi s value range may be larger or smaller than the other eight suitability index ranges. If the one quarter standard deviation value is less than the cell size then the suitability index values will be divided into equal intervals between the MEAN and MIN. The A4 Community Values Program Once the suitability of each objective and/or sub objective is determined, they are combined according to their hierarchical level using utility values (i.e. weights) that equal 1.0 (100%). The weights at the objective and sub objective level s are citizen driven; meaning the weights obtained at this level reflect localized knowledge of community values. These weights are obtained from existing plans, community meetings, or focus groups. Often surveys are used to gauge com munity values. To determine the numeric weight, particularly between goals, the A4 Community Values Calculator is developed. The A4 Community Values Calculator is initiated by ins talling the program as an ArcMap macro in the Visual Basic Editor ( VBEdito r ) Based upon pair wise comparison methods, this program blurs the line between planner and land use modeler. A planner with minimal experience in modeling can easily use this program within a GIS environment to complete a values survey among stakeholder s.
96 When evaluating the importance between objectives/alternatives the A4 Community Values Calculator integrates any number of objective and/or sub objective raster suitability surfaces as inputs. The A4 Community Values Calculator interface prompts the us er to specify the usefulness of each pair of raster surfaces and dynamically compares the raster pair. As the user indicates values for each pair, the A4 Community Values Calculator automatically populates a pair wise comparison matrix. The calculator the n outputs a parameter table of the raster names and their corresponding relative weights. As a way to reflect community participation, the tool also uses an algorithm to update the weights based on the different pair wise comparison assignments for a grou p of people or a panel meeting. The result is a table of weights reflecting group values which is then used as an input for the A4 Layer Weighting Tool, the tool used to create complex MUAs Although there are many multi criteria decision support tools a vailable, having this tool available within the GIS saves time, eliminates the expense of purchasing a third party software package, and reduces error when inputting values from a standalone software package. The tool is used to determine the weights to c ombine the suitability in the affordable housing model depending on expert values for the three counties in the study area. Once the expert values are completed, the A4 Community Values Calculator generates a table of weights used by the A4 Layer W e ighting Tool to combine the suitability layers. The A4 Layer Weighting Tool When determining the final suitability for each land use, the degree of interaction between each goal MUA is measured by the weights generated from the A4 Community Values program. The A4 Layer Weighting Tool is similar to t he Weighted Sum tool
97 available in the Spatial Analyst toolbox. Both tools can multiply multiple raster su rfaces by a specified weight then sum the surfaces together. Instead of manually entering the weights for each goal surface, the A4 Layer Weighting Tool uses the parameter table generated from the A4 Community Values program or a table of similar structure gene rated outside the A4 Community Values program as an input to the A4 Layer Weighting tool Affordable Housing Opportunity Surfaces The automation tools help in the automation of the suitability structure which are divided into goals, objectives and sub objectives. In LUCIS the conflict Identification strategies are used to identify the conflict between the agricultural, conservation and urban goals. The conflict is a suitability matrix that combines the preference of the three goals. The preference n umber is a number between one and three. The number three represent the highest preference. Two represent moderate preference, and one is the lowest preference. The combination number of 333 means that the area that has the mentioned preference is identif ied with a major conflict because it is preferred for agriculture, conservation and urban use at the same time. 113 mean that the area has a high preference for urban and low for agriculture and conservation. LUCIS Conflict Identification Strategies are us ed to generate the conflict surface to identify the conflict between the affordable housing objectives, which are accessibility and neighborhood characteristics, housing burden, transportation burden, and transit access. Introducing Housing Cost as a Prefe rence Surface CNT (2007) have used travel and housing cost to build an index that is also dependent on income. The HT index created by CNT aims to estimate the affordability of housing site locations to the people that live in that place, while the case st udies in
98 this research aim to allocate affordable housing land for very low income population. Similar to the HT index, travel and housing costs are important factors in deciding the residential location in the model. Generally, housing costs are incorpor ated as a preference surface in the generation of the final conflict and opportunity surfaces which are the surfaces used in the allocation of affordable housing sites. There are three ways to incorporate the housing cost in the affordable housing model. The housing cost could be a rent cost or a mortgage cost and depends on the objective of the affordability studies surfaces that can be generated for mortgage or rent. These surfaces could be weighted according to the number of rent or owned units in the C ensus data to get an estimation or proxy for the housing cost. However, the research uses the rent data for the year 2009 prepared by the Shimberg Center for Housing Studies to estimate the location cost. Introducing Transit Access as a Preference Surface The transit accessibility adds a very important transportation option for low income and very low income populations especially when the driving cost gets higher. This research creates a parcel level transit accessibility score for residential location bas ed on transit stops routes and employment. These transit scores are used to create a preference surface that is used in the generation of the affordable housing conflict/opportunity surfaces. Therefore, if the driving preference is low and the transit acce ssibility is high this indicates that the place is also eligible to be allocated for affordable housing. The choice of preference number depends on the allocation scenario. The methodology for creating the transit scores can be divided into three categorie s:
99 1. Downstream stop score estimation based on employment opportunities; 2. The network distance or time estimation including time spent in transfers or delays; and 3. The upstream transit opportunities and their distances from residential parcels. Downstream s t ations s core The creation of the downstream stop score starts by identifying the activities within the walking distance between each stop and assigns the distance weighted employment to the transit stop. The walking distance buffer is taken as a 0.25 mile Manhattan distance taken from the transit stop at the center of the Manhattan buffer. Two tools were created using ArcGIS customized Python programming for that purpose. The first tool, The Manhattan buffer, created a diamond shaped Manhattan buffer around point feature classes. Figure 3 16 shows an example of Manhattan buffer features created around transit stops. The second tool measures the Manhattan distance to point features and creates a raster similar to the Euclidean distance raster tool in ArcGIS. Figure 3 17 shows an example Manhattan distance raster created around transit stops. The downstream transit stops could also be upstream transit stops for different transit trips. The transit accessibility measurement depends on creating stop to stop, orig in destination matrices which may have some size limitations depending on the computer and the size of the study area. However, some of the downstream stops could be ignored in the analysis if they are not significant in serving employment. Estimating netw ork distance/t ime Depending on the data for bus routes and stop, the distance or time for a transit trip can be calculated using ArcGIS Network Analyst. However, transfer stops should be taking into consideration if the distance is calculated. If time is c alculated, the trip time
100 should also include delays at the bus stops and during the transfer. ArcGIS origin destination matrix time or distance estimation using network analysis are used to generate all the alternative trips using transit from each upstrea m stop. These alternative distance measurements or time measurements are used in addition to the downstream score based on employment in the gravity accessibility estimation equ ation This accessibility estimation will generate an accessibility score for e ach upstream stop. Creating transit accessibility suitability s urface: The transit access tool is an ArcGIS tool that c alculates the gravity acce ss for each trip using the down stream score as an attraction, the trip distance and the route frequencies and a ggregate s the calculated values of all the trips connected to each upstream station. The tool also uses these upstream transit scores to create the final transit accessibility multiple utility assignment surfaces. First a Manhattan raster is created around each upstream stop. This surface is reclassified according to the distance from residential parcels and represents the walkability to transit stops based on the distance. The second surface aggregates the upstream transit scores that are within the walkin g distance from residential parcels. This surface is also reclassified according to the mean and standard deviation to create the suitability surface. The final transit access surface is generated by combining the two mentioned suitability surfaces. Figure 3 18 shows an example transit access suitability surface generated for Orange County. Automatic Allocation for Affordable Housing The A4 Allocation Tool The new allocation procedures in LUCIS model s adopt automation tools for the future allocation of population and employment, scenario building, and testing of
101 policies. This tool can be also used to build scenarios for the allocation of affordable housing based on prioritizing allocation conditions and policy initiatives. This is done by the A4 A lloca tion toolbox which consists of three allocation tools The first tool in the allocation toolset is the Trend Allocation Tool. The A4 Allocation toolset provides tools for an automated allocation of new urban populations and accommodates for spatial constraints and variable density allocations. The foundation of the allocation tools are combine grids The combine grid is prepared by an enumeration rule that combines all of the grids needed in the allocation process while maint aining their attribute values. The Allocation by Table tool can also be seen as a planning table or a scenario builder where the planner enters the conditions for an allocation depending on each conflict score or on multiple sets of score. Using this tool the planner can perform the allocation using different conditions and priorities as iterations simultaneously. Figure 3 19 shows a sample planning table used for a n allocation process as well as the Planning by Table Tool interface. However, t he previous tools work for eight masks and/or conditions for an allocation. A detailed tool can be also used in the allocation process. This detailed allocation can work on tw elve different masks and/or additional conditions for an allocation procedure and can also be incorporated into a model that has iterative procedures. Building Affordable Housing Scenario The output of the affordable housing model is the conflict/opportuni ty surface that contains neighborhoods characteristics, rent burden, travel cost and transit accessibility. This combine grid combine the opportunity grid to additional grids that represent walkability, underutilized density, policy incentives areas and ot her grids that are necessary for the allocation process. The A4 A llocation tools are used to run different
102 scenarios for the allocation of land for affordable housing. These scenarios are based on the affordable housing combine grid which helps the planner to run automated queries for allocation based on a specific scenario. This research will utilize under utilized density values in a compact development scenario to find places that have location efficiency and at the same time have densities much lower th an its surrounding area. The basic element of the allocation will be the conflict surface which will include the four objectives in a 4 digit conflict matrix. This conflict surface will be combined with the under utilized density raster in addition to othe r masks and conditions to allocate the land for affordable housing. The full list of conditions utilized in the allocat ion is s hown in C hapter 7 Model Validation Methods The validation of the model s is included in detail in C hapter 7 In general there are different levels of model validation. The first level is the validation of the automation tools. This is performed by using the ArcGIS user interface in a step by step procedure n the model output level by doing sensitivity analysis. This is performed by creating the output affordable housing opportunity surface using travel cost and transit accessibility and the other opportunity surface without travel cost and transit accessibil ity. The output opportunity surfaces are compared using descriptive statistical indicators derived from the 5Ds (Ewing & Cervero, 2001) and the sprawl conceptualization metrics (Galster et al 2001). The third level of validation is performed by comparing the allocation sites and overlaying them with the Assisted Housing Inventory data (AHI) to analyze the surrounding locations.
103 Table 3 1. Data sources No. Description Source Year 1 Parcel data FGDL 2009 2 Census Blocks, Block Groups and Tracts C ensus 2000 3 Hosing Cost Shimberg Center 2009 4 Trip Data NHTS 2009 5 Transit Data Transit Agencies Varied Table 3 2 Percentage Trips by purpose (National Househ old Travel Survey, 2009) Trip Distance Home Based Other Home Based Shop Home Based Soc ial Home Based Work Non Home Based Total Less than 10 miles 16.44 18.37 11.25 5.70 22.06 73.82 10 and 20 miles 3.22 2.31 1.78 2.87 3.64 13.82 20 and 30 miles 1.02 0.58 0.75 1.16 1.25 4.77 30 and 40 miles 0.37 0.21 0.37 0.49 0.55 1.99 40 and 50 miles 0.18 0.16 0.15 0.20 0.24 0.92 50 and 60 miles 0.12 0.04 0.06 0.09 0.17 0.48 60 and 70 miles 0.04 0.04 0.08 0.04 0.09 0.29 70 and 80 miles 0.02 0.02 0.03 0.03 0.07 0.16 80 and 90 miles 0.03 0.01 0.03 0.01 0.08 0.17 90 and 100 miles 0.03 0.03 0.03 0.01 0.04 0.13 100 and 110 miles 0.02 0.02 0.02 0.00 0.03 0.08 110 and 120 miles 0.01 0.01 0.01 0.00 0.03 0.05 120 and 130 miles 0.01 0.01 0.01 0.00 0.03 0.06 130 and 140 miles 0.00 0.01 0.01 0.00 0.03 0.05 140 and 150 miles 0.01 0.00 0.01 0.00 0.02 0.04 150 and 160 miles 0.01 0.00 0.00 0.00 0.01 0.02 160 and 170 miles 0.00 0.00 0.00 0.00 0.01 0.02 170 and 180 miles 0.00 0.00 0.02 0.00 0.01 0.03 180 and 190 miles 0.00 0.00 0.01 0.00 0.01 0.02 190 and 200 miles 0.00 0.00 0.00 0.00 0.01 0.01 More than 200 Miles 0.02 0.02 0.04 0.00 0.16 0.24 Missing 0.80 0.53 0.33 0.38 0.78 2.82
104 Figure 3 1. Structure for the conflict / opportunity process Figure 3 2 Variable size neighborhoods around the parcel
105 Figure 3 3. Euclidean proximity model
106 Figure 3 4 Opportunity model
107 Figure 3 5 Trip end points in Duval C ounty Figure 3 6. Trip s by purpose 0.00% 5.00% 10.00% 15.00% 20.00% 25.00% 30.00% 35.00% 40.00% 45.00% 50.00% 55.00% 60.00% 65.00% 70.00% 75.00% 80.00% 85.00% 90.00% 95.00% 100.00% 50 and 60 miles 40 and 50 miles 30 and 40 miles 20 and 30 miles 10 and 20 miles Less than 10 miles
108 Figure 3 7. Percentage of trip s categorized by purpose and trip length Figure 3 8. Local outlier identification tool 93.50% 94.00% 94.50% 95.00% 95.50% 96.00% 96.50% 97.00% percentage
109 Figure 3 9. Example local outlier identification for Orange Cou n ty Figure 3 10. Cross validation of errors for interpolating work trips in Orange County.
110 Figure 3 11. Iterative model used to generate travel mile s surfaces Figure 3 12. Estimation of travel miles and travel cost
111 Figure 3 13. Daily travel miles surface for Orange County Figure 3 14. Decreasing suitability indexing.
112 Figure 3 15 Increasing suitability indexing. Figure 3 16. Manhattan buffers arround transit stops
113 Figure 3 17. Manhattan d istance raster to transit stops Figure 3 18. Transit access suitabilty surface
114 Figure 3 19 Planning by T able tool
115 CHAPTER 4 CHOICE OF ANALYSIS UNITS The choice of the areal unit in this research focuses for the research on land use and transportation coordination. Many urban form indices can be calculated on a disaggregate level such as at a parcel level analysis but results are usually aggrega ted to a neighborhood or any other aggregate scale such as TAZs. This research studies the effect of scale and zoning of the areal unit by investigating the impact of size, shape and location of the areal unit on the most commonly used urban form character istics These urban f orm characteristics are diversity, density and connectivity. The areal unit in this research is chosen after carrying out area l unit optimization research. This optimization is conducted by calculating the urban form metrics on differe nt scales and studying the effect of scale on these metrics. The Importance of Defining the Areal Unit The 5Ds (Lee & Cervero, 2007) are generally urban form measurements. Usually, these measurements are obtained after defining the areal unit of analysis such as the neighborhood or any type of gridding. However, not all urban form characteristics should be aggregated to a defined areal unit. Sometimes it is better to understand the characteristic values on a more disaggregate level such as at the parcel level. A ccessibility and distance estimations are r ecommended by many researchers to be performed on a disaggregate d and parcel level. Common methods of capturing accessibility are based on aggregate analysis zones such as TAZs (Levinson & Krizek, 2008). However, it is more helpful and more accurate to undertake estimations at the parcel level. Disaggregate and parcel level research can reduce the shortcomings of
116 traditional models by capturing the fine land use effect on transportation or vice versa (Lee, 2004). Primarily there are two approaches in the level of analysis used to delineate or determine the connection between land use and transportation. One of them is an aggregate approach that uses zones such as TAZs and calculates accessibility indices based on the analysis zones. The other approach uses parcel level analysis. Johnston (2004) mentioned that future land use and transportation modeling should be discrete in both time and location and should be based o n GIS tract s or street address es This shows the common ground for future land use and transportation modeling. The same GIS approach has been recommended by Wegner (2005) to deal with disaggregate data in activity based models, which is fundamental in th e latest trend in transportation modeling. However, discrete analysis may not always be possible due to lack of data. Therefore, TAZ and neighborhood urban form aggregation is used to conduct land use and transportation analysis. Nevertheless if TAZs are used, the effect of the MAUP should be taken into account for calculating aggregate accessibility value s as well as its spatial variation. In practice, most of the mathematical forms for accessibility are used on aggregate levels such as TAZ. The same equa tions can be applied on a parcel level to give more accurate accessibility measurements and eliminate the effect of the MAUP for accessibility estimation. Sometimes accessibility is applied in an equation on a parcel level and the resultant accessibility i ndex is aggregated to a neighborhood level. In this case the areal unit problem should also be taken into account.
117 Several land use and transportation coordination variables are urban form characteristics that are gross measures by definition, and thus measured on a neighborhood scale or other defined analysis unit areas. The neighborhood scale primarily will affect the value of measurement. Ewing et al (2008) found the impact of the 5Ds on VMT is less for small areas, thus the elasticity values should different. The differences between values of land use and transportation characteristics aggregated to the neighborhood level are caused by the MAUP. According to Jelinski and Wu (1996) the MAUP has two problem components which are scale and zoning problems. In the scale problem, the value and the statistical variation of the aggregated smaller units will be different depending on the size of the areal unit. The zoning problem however, deals with the location of the areal unit and the choice of zoning which also affect s the value and the variation of the aggregated entities. Due to the zoning problem, the same size d aggregated units may give different results due to differences in location. Researchers frequently suggest grid based approaches to do spatial aggregation. However, little attention is paid to the size of the grid cells and the method of aggregation to reduce the MAUP. Reynolds (1998) mentioned the MAUP as a problem in spatial research as early as the 1930s bu t indicated it could be reduced by the use of GIS. He also mentioned that many of the data are collected on a disaggregate level and aggregated to coarser resolution for different reasons such as privacy. However, most time data is measured at a disaggrega te level but mapped spatially on an aggregate level. Therefore, the aggregation process will be necessary to map the data. Reynolds also explained the two sides of the areal unit problem which are scale and zoning.
118 Tomoki (1999) explained the dependence o f the map interpretation on the areal unit and emphasized that currently there are only a few criteria on the choice on the areal units. Tomoki explained the effect of the areal unit size on mortality indicators where the variation in the value decreases f or larger units. Therefore taking larger units will be more statistically stable but at the same time the results will be more ambiguous. In summary, the size, shape and boundaries of neighborhoods affect the value of any aggregate measurement based on these neighborhoods. General gridding procedures and naturally defined neighborhoods are used in the land use and transportation research to define unit areas. If one is to research the land use mix variations using an entropy value on a parcel adjacent to the boundary in a naturally defined neighborhood this parcel may have, for example, retail and service opportunities surrounding it and the entropy value for that neighborhood will not capture that land use mix. This is mainly a form of the zoning effec t induced by the MAUP. Steiner and Srinivasan (2009) proposed a solution for this problem by using overlapping 2 by 2 mile square neighborhoods for calculating the land use and transportation characteristics for their trip length model. To reduce the edge e ffect of the MAUP, a methodology to overlap neighborhoods is used where the neighborhoods were duplicated and shifted one mile east and the resultant duplicated neighborhood are duplicated again and shifted one mile north. Therefore, a parcel at the edge o f a neighborhood will be at the center of an overlapping neigh borhood. Figure 4 1 shows the neighborhood overlay according to their methodology (Steiner & Srinivasan, 2009). Taking into consideration the scale and zoning components of the MAUP, it is clear that the size, location and shape of the areal unit are the main component s that
119 may affect the value and the variation of the urban form measurements aggregated to that area. Many of the urban form measurements such as accessibility are recommended to be performed on a parcel level and therefore there is no need to consider the modifiable area problem. However, other urban form measurements such as density, land use mix and connectivity are needed on an aggregate scale. Therefore, the effect of the modifi able areal unit should be taken into account when dealing with those variables. Entropy Test The first experiment is performed on a land use mix index represented by entropy values based on different areal unit sizes. The same experiment is then applied fo r density and connectivity metrics Generally, the analysis follows an iterative procedure taking unit sizes of (0.5 x 0.5 mile 5 x 5 mile) using a 0.5 mile incremental increase. The optimal neighborhood is taken as the minimum size where increasing the neighborhood no longer has a significant effect on the entropy value. Table 4 1 and Figure 4 2 show the change in entropy value that c orresponds to an increase of 0.5 mile in the unit size for Orange County Figure 4 2 shows clearly that the mean entropy values for neighborhoods sizes of 2.5 mile x 2.5 mile are not significantly different from the values for a 3 mile x 3 mile neighborhoo d. The 2.5 mile unit size is the smallest size that can be chosen as an optimized unit. Figure 4 3 shows that the standard deviation of the entropy value change is very small for unit areas of 2.5 mile or more. Table 4 1 demonstrate s a minimum change in m ean and standard deviation for a 2.5 mile x 2.5 mile neighborhood and thus conclude s that the optimal neighborhood size for the entropy value is 2.5 x 2.5 miles. The conclusion is also supported by the mean change chart ( Figure 4 2 ) and the standard deviat ion change chart ( Figure 4 3 ) The
120 analysis is applied to three counties of urban context in the state of Florida The results of analysis were comparable. Table 4 2 shows the effect of changing neighborhood size on the entropy value for the three counties Figure 4 4 shows clearly that the minimal change in the entropy mean value occur s when the neighborhood sizes are set to 2.5 mile x 2.5 mile or more. Figure 4 5 shows that the standard deviation of the change in value is generally decreasing when increas ing the neighborhood size and the curve is more flat for the size value of 2.5 mile or more The result shows also that the variation in entropy values are less for larger neighborhoods which agrees with the literature on the scale component of the MAUP. F igure 4 6 shows the entropy surface based on a 2.5 x 2.5 mile neighborhood for Orange County. Conducting the analysis on a roving unit where the calculated cell value is always in the center of the neighborhood minimizes the zoning problem and creates a su rface of a finer resolution. Density and Connectivity Tests Density is usually calculated on multiple scales depending on it is use. However, in land use and transportation research the neighborhood density is used to identify the impact of density on tra v el cost or VMT. In this research a si mple density measurement is used which is the gross population density per acre averaged for each the areal unit s and assigned to the point at the center of the neighborhood. The same method is conducted for connectivi ty where a simple connectivity measurement of street density is taken to study the effect of unit size on urban form measurements. Table 4 3 shows the mean value and the standard deviation of the change in density values result ing from a gradual increase o f 0.5 miles in each areal unit size. Figures 4 7 and 4 8 show that the minimal changes in density mean corresponds to the unit size of 2.5 x 2.5 mile or more.
121 The analysis has been performed on three different counties and the result was that the neighborh ood size of 2.5 mile x 2.5 mile can be used as an optimal neighborhood size for density calculation s for the three counties. Figure 4 9 shows a density surface based on 2.5 mile unit. For the connectivity measurement, the street density is another urban form measureme nt to be tested. T able 4 4 shows the mean value and the standard deviation of the change in road density values resulting from a gradual increase of 0.5 miles in unit size. Figures 4 10 and 4 11 show that the minimal change in road density me an s value s correspond to a unit size of 2.5 x 2.5 miles or more The analysis is replicated on three urban counties and it was found that the neighborhood size of 2.5 mile x 2.5 mile can be used as an optim al neighborhood size for connectivity calculation s for the three counties. Figure 4 12 shows the connectivity surface for Duval County based on a 2.5 mile areal unit. Shape of Neighborhood Test T he optimal neighborhood size for the three counties is tested for a change in shape. A circle and diamond shape d boundary instead of a square shaped boundary for the neighborhood analysis are used to test the impact of the neighborhood shape on entropy values Figure 4 1 3 shows the different areal unit shapes used in the analysis. Figure 4 14 shows a land use mix entropy surface based on a 2.5 mile corner to corner diamond areal unit. Figure 4 1 4 is the output entropy surface using a 2.5 mile of a diamond shaped unit which represents a driving / biking distance of 1.25 miles. To understand the results of changing the neighborhood shape, the entropy layer is overlaid with other layers such as street networks as shown in Figure 4 1 5 It was found that the land use
122 mix layer resulting from the shape analysis using the Manhattan diamond shape is more subjectively jus tified than the result surfaces using square and circle neighborhoods. This is because the diamond shaped neighborhood gives more mixed use density in areas that have more street networks while the other surface s do not distinguish these areas from others Reducing the Modifiable Areal Unit Problem In c onclusion t he analysis in this research proved that changing the areal unit size of a neighborhood changes the aggregated value of the variable being measured. The generated surfaces also show that the v ariation in aggregated value increases for smaller zones and decreases for larger zones. Therefore the results show that larger areal units are more stable but the values could be more ambiguous which is corroborated by the literature on the MAUP. The results also show that the neighborhood size could be optimized for the tested urban form variables which are land use mix, density and connectivity. The graphs for the change in aggregated values c orresponds to the change in neighborhood size and shows that the curve is nearly flat for neighborhoods sized 2.5 mile and more, and the standard deviations for the change are very low. This means that the optimal size of the neighborhood is the minimum un it size where the curves began to be flat which is 2.5 mile x 2.5 mile. For the scale issue of the MAUP, the results also show that the change in aggregated urban form value is very large for small neighborhood sizes, which means that using small unit siz es as area units may lead to undesirable results. On the local level, the entropy value change is large and unpredictable as shown in F igure 4 1 Therefore changing from a 0.5 mile to a 1 mile size could positively or negatively change the aggregated valu es. It is true that the variation is less for larger
123 neighborhood, but from the results, it is not difficult to conclude that the change in aggregated values due to changing neighborhood sizes in small neighborhood is not predictable. However, this will ra ise the question of how applicable the natural neighborhood will be as an area unit where the unit sizes are always different between zones. From the results of the analysis we can conclude that it is better to use the same size of area unit for the whole spatial environment and changing the unit size may lead to undesirable results. Taking entropy as an example, the magnitude and direction of change in entropy values for small neighborhoods gives us a hint that the use of an entropy value, in a naturally defined neighborhood may have inconsistencies due to the MAUP. These inconsistencies maybe reduced if the same neighborhood size for the entire model is used. These inconsistencies are also reduced if the optimized neighborhood size is used as well as the floating area methodology. For the zoning issue in the MAUP, the results show that changing the boundary shape leads to different aggregated results, which confirm the MAUP zoning problem. The results show that the shape can be optimized by analyzing the urban form and the spatial distribution of activities. However, analyzing the land use mix relationship with the street network, we can see that the diamond shape which represents a Manhattan grid driving or biking distance gives better aggregate results than the Euclidean distance (circle) or the square neighborhood. The zoning problem in this research is also reduced by the use of the floating neighborhood where each point in the map is taken as the center of the aggregated
124 zone. However, this unit area definition works for the purpose of tested variables in this research and may not be valid for other types of neighborhood aggregation processes. The selection of the areal unit research has limitations that can be summarized as follows Firstly, the rese arch is limited to counties that have similar urban context. The tested counties may have different spatial patterns but the three counties have large cities and CBDs. The research is not applied to rural counties which will be left for future research. S econdly, the change in aggregated value is compared by the mean and the standard deviation of change. More statistical indicators could be used in the optimization process. This also will be performed in future research. Thirdly, the research is limited t o the tested variables for land use mix, density and connectivity. Therefore the results should not be generalized. Finally, the research is limited to the aggregation of urban form measurements that can be used to test the impact of urban form on VMT or t ravel cost The optimized unit should not be used for the aggregation of values that depend on small neighborhood sizes by definition such as walking or transit zones
125 Table 4 1. Change in entropy mean and standard d eviation Change in Neighborhood Size in Miles Entropy Minimum Change Entropy Maximum Change Entropy Mean Change Entropy STD of Change 0.5 1 0.2296 0.5921 0.0784 0.0911 1.0 1.5 0.1684 0.4024 0.0199 0.0484 1.5 2.0 0.2979 0.1776 0.1126 0.0924 2.0 2.5 0.0764 0.1537 0.0038 0.0204 2.5 3.0 0.0645 0.1205 0.0035 0.0165 3.0 3.5 0.0458 0.0947 0.0031 0.0139 3.5 4.0 0.0395 0.1016 0.0028 0.0124 4.0 4.5 0.0309 0.1115 0.0026 0.0110 4.5 5.0 0.0273 0.0752 0.0025 0.0097 Table 4 2. Change in entropy mean and standard deviation for three counties Size From Size To Mean Change Orange Mean Change Duval Mean Change Hillsborough STD Change Orange STD Change Duval STD Change Hillsborough 0.5 1 0.0784 0.0484 0.0309 0.0911 0.0832 0.0685 1 1.5 0.0199 0.0267 0.0006 0.0484 0.0558 0.0387 1.5 2 0.1126 0.0177 0.0051 0.0924 0.0406 0.0281 2 2.5 0.0038 0.0126 0.0036 0.0204 0.0315 0.0221 2.5 3 0.0035 0.0097 0.0028 0.0165 0.0257 0.0181 3 3.5 0.0031 0.0077 0.0027 0.0139 0.0213 0.015 3.5 4 0.0028 0.0064 0.0022 0.0124 0.0182 0.0112 4 4.5 0.0026 0.0054 0.0032 0.011 0 0.0158 0.0132 4.5 5.0 0.0025 0.0046 0.0023 0.0097 0.0138 0.0102 Table 4 3. Change in density mean and standard d eviation Size From Size To Mean Change Orange STD Change Orange 0.5 1 0.337 1.286 1 1.5 0.012 0.502 1.5 2 0.003 0.349 2 2.5 0.002 0.259 2.5 3 0.001 0.203 3 3.5 0.001 0.168 3.5 4 0.001 0.145 4 4.5 0.001 0.130 4.5 5.0 0.001 0.116
126 Table 4 4. Change in connectivity mean and standard d eviation Size From Size To Mean Change Duval STD Change Duval 0.5 1 4.437 3.529 1 1.5 4.384 1.197 1.5 2 0.043 1.351 2 2.5 0.043 1.027 2.5 3 0.042 0.82 0 3 3.5 0.041 0.671 3.5 4 0.040 0.572 4 4.5 0.039 0.491 4.5 5.0 0.038 0.429
127 Figure 4 1. Neighborhood d efinition a ccording to the Steiner and Srinivasan (2009) m odel Figure 4 2 Mean of entropy value change that corresponds to 0.5 mile c hang e in unit s ize for Orange County -0.15 -0.1 -0.05 0 0.05 0.1 0 1 2 3 4 5 Mean of Entropy Value Change Unit Size
128 Figure 4 3 Standard deviation of entropy value change that corresponds to 0.5 mile change in unit s ize for Orange County Figure 4 4 Mean of entropy value change that corresponds to 0.5 mile change in unit s ize for three counties 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 1 2 3 4 5 SD of Entropy Value Change Unit Size -0.15 -0.1 -0.05 0 0.05 0.1 0 1 2 3 4 5 Mean of Entropy Value Change Neighborhood Size Mean Change Orange Mean Change Duval Mean Change Hillsborough
129 Figure 4 5 Standard deviation of e ntr opy value change that corresponds to 0.5 mile change in unit s ize for three counties Figure 4 6 Land use mix entropy based on 2.5x2.5 mile areal u nit for Orange County 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 1 2 3 4 5 STD Value of Change Neighborhood Size STD Change Orange STD Change Duval STD Change Hillsborough
130 Figure 4 7 Mean of density value change that corresponds to 0.5 mile change in unit s ize Figure 4 8 Standard deviation of density value change that corresponds to 0.5 mile change in unit s ize -0.4 -0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0 1 2 3 4 5 Mean Change Density Neighborhood Size Mean Change Orange 0 0.2 0.4 0.6 0.8 1 1.2 1.4 0 1 2 3 4 5 Std of Change Neighborhood Size STD Change Orange
131 Figure 4 9 Density surface for Hillsboroug h County based on 2.5x2.5 mile areal u nit Figure 4 10 Mean of connectivity value c hange that corresponds to 0.5 mile c hange in unit size -5 -4 -3 -2 -1 0 1 0 1 2 3 4 5 Mean Change Connectivity Neighborhood Size Mean Of Connectivity Change in Duval
132 Figure 4 1 1 Standard deviation of connectivity value change that corresponds to 0.5 mile change in unit s ize 0 0.5 1 1.5 2 2.5 3 3.5 4 0 1 2 3 4 5 STD Change Connectivity Neigborhood Size STD of Connectivity Change in Duval
133 Figure 4 1 2 Connectivity s urface for Duval Co unty based on a 2.5 x 2.5 mile areal u nit Figure 4 1 3 Shapes for the tested areal units
134 Figure 4 1 4 Land use mix entropy s urface for Hillsbo rough County based on 2.5 mile diamond shaped areal u nit Figure 4 1 5 Streets and land use mix entropy overlay s urface for Hillsbo rough County based on 2.5 mile d iamond shaped areal u nit.
135 CHAPTER 5 ESTIMATING ACCESSIBI LTY AND TRAVEL COST Neighborhood Accessibility Neighborhood accessibility is estimated using differ ent methods. The first and simplest method to estimate accessibility to services is to use proximity by estimating the Euclidian distance from an origin to a destination. In raster land use modeling the Euclidean raster is used by many researchers. Land us e modelers however, usually use the term proximity and not accessibility when they are using Euclidean distance. It should be noted tha t the Euclidean raster estimates the distance between any cell in the raster to the nearest facility or destination feat ure (ESRI, 2011 ). In suitability analysis, the distance value is transformed into a suitability score. This score will be an increasing suitability score or decreasing suitability score depending on the type of feature. Facilities like schools and hospitals are usually favorable locations in terms of proximity while facilities like prisons and noise sources are regarded as a disadvantage in terms of proximity. For example, distances to am enities will generally have decreasing suitability meaning that if the distance away from the amenity increases, the suitability of a location decreases. Conversely, for facilities that are less desirable to be located near, suitability will increase as the distance away from the facility increases. The reclassification to decreasing and increasing suitability usually follows a sequential process including descriptive statistics. An example of descriptive statistics used in the process is the zonal stati stics of the Euclidean distances for all the residential parcels. From the zonal statistics, the mean and standard deviation is calculated and their values are used to assign the suitability values. It should be noted however, that proximity does not repr esent the actual travel distance from an origin to a
136 destination. It also does not discriminate between destinations according to their size or attractiveness. The proximity suitability surface for distances away from shopping center s (Figure 5 1) is recla ssified using decreasing suitability based on the mean and the standard deviation o f the distance away from multif amily parcels. Proximity can be also estimated by estimating the network distance from any cell in a raster to the nearest facility or destin ation feature. However, the estimation process is not a direct output from ArcGIS tools. Network analyst can be used to locate the nearest network points for each origin and its nearest destination. Using the shortest path method, the network distance is a lso calculated. Then the values for the distance of the whole trip is summed and assign ed as a score for the origin in the raster allow ing the user to create a network distance raster. This procedure is automated using a customized P ython tool created for this research. T he net work distance raster ( Figure 5 2 ) is created b y the tool It should be also noted that the raster may be a better estimation of travel distance to services but it does not discriminate between destination s according to thei r size or attraction. The raster only estimates the distance to the nearest facility. Furthermore, t he method is used for destinations that can be reached by walking or driving using street networks. Therefore, it is not used to estimate the proximity to noise gener ators or sources of pollution. The distance to shopping centers ( Figure 5 2 ) is reclassified to a suitability surface using a decreasing suitability based on the mean and standard deviation of the network distance summarized by z onal statistics using the multi family parcels as zones. Opportunity Access Measures As mentioned before, proximity does not discriminate between services that a person may choose as a destination. It only estimates the distance to the nearest
137 facility. However, a person may travel more for shopping if the destinations have more shopping options (Arafat et al 2010) The opportunity accessibility measurement captures how many services are within a specified distance. The score for services can be different depending on the type of service such as the area of a retail store or the n umber of beds for a hospital. The neighborhood or the surrounding distance can be Euclidean distance, Manhattan distance or network distance. The cumulative opportunity score within the specified distance from a parcel is assign as an opportunity score for that parcel (or cell in the raster). A suitability surface is created by the reclassified opportunity surface. Figure 5 3 shows a suitability surface depending on the opportunity access score. The surface is composed by combining two suitability surfaces that represent the opportunity in walking and biking distances. The surface is generated using a Euclidean buffer distance. Figure 5 4 shows the suitability based on the opportunity using a network buffer instead of a Euclidean buffer. The generation of the raster using a network buffer is a sophisticated iterative procedure that needs programming in addition to the use of ArcGIS Network Analyst. A customized GIS tool model was created to estimate the oppor tunity within the network service areas and to construct the output raster. Combined Opportunity Distance Access Measures The opportunity raster does not discriminate between opportunities according to their distance from a parcel. Therefore, the opportuni ty suitability estimate is combined with the distance suitability estimate to generate an opp ortunity distance MUA. Figure 5 5 shows the MUA surface based on combined opportunity and distance using a Euclidean distance estimation and buffer while Figure 5 6 shows the combine opportunity and distance MUA based on network distance and network service area.
138 However, the combined distance opportunity measurement is different from gravity access. Gravity access can be estimated by generating origin destination matrices for the entire area either by using network or Euclidean distances. The estimation proves efficient for small areas or using zonal level analysis such as with TAZs. On a parcel level analysis the resultant origin destination matrix could be more t han a billion records for a single county which imposes a limitation on the use of gravity access at that level. The gravity estimation in this research is used to estimate transit accessibility. The gravity estimation also is used to estimate accessibilit y for a limited number of origins picked randomly. This estimation is used to calculate the accessibility to major activity agglomerations which is important in estimating travel cost. Estimat ing and P redicting Travel Cost Travel cost estimation is complex in its nature. The vehicles miles of travel are dependent on urban form characteristic as indicated in many studies (Ewing & Cervero, 2001; Ewing et al 2007; Lee & Cervero, 2007). However, travel behavior depends also on traveler characteristics. Land use models usually focus on allocating future population in places that are suitable for housing. These models usually do not focus on the household characteristics for the people who will reside in these locations. Neighbo rhood aggregated values such as income and poverty may be used, however, in the modeling procedure, the use of these aggregated values helps the modeler take into account neighborhood change and social justice issues and to plan for more sustainable commun ities. This research is concerned about a spatial discrimination of travel cost. Therefore, it focuses on spatial location more than traveler characteristics. However, aggregated surrounding values such as income and household size are used in the research
139 Taking the location and neighborhood characteristics into account, the travel cost can be estimated depending on the National Household Travel Survey 2009. The travel cost can be estimated by the spatial interpolation of the geo coded trip data or by rel ating the travel cost to the urban form and neighborhood characteristics and use the output relation to estimate the cost. However, this assumes that there is a travel cost estimate in the data. The NHTS data does not have a travel cost field in its trip d iary and it should be estimated indirectly using other fields in the survey. Therefore, trip length is used to generate a travel cost depending on a per mile travel cost and the household locations provided in the travel survey. Spatial interpolation provi des a method of estimating values for missing trip length values for trip categories not reported in the survey. These values are estimated depending on the nearest neighboring values as explained in Chapter 3 The following sections show the results of th e travel cost estimations as well as the validation process for the estimation. Estimating Travel Cost by Spatial Interpolation S patial interpolation is used to estimate values for the trip length of a certain category depending on the neighboring values in the travel survey. In the travel survey, a person may have reported a work trip but did not report shopping trips The proxy of travel cost estimation assumes that each household does all the trip categories each month depending on summary statistics f rom the travel survey. Therefore, missing trip categories are estimated for each location by spatial interpolation of each trip category independently. Different interpolation methods were investigated to come up with the method with least errors in the c ross validation procedure. In investigating interpolation methods, it should be noted that there are two main categories for spatial interpolators; these are
140 deterministic interpolators and stochastic interpolators. For estimating trip length based on repo rted values for neighboring points, deterministic methods of interpolation such as inverse distance weighting is more preferable than stochastic methods of interpolation such as kriging. The reason is that deterministic interpolation does not change the re ported values during the interpolation. However, these reported values could be changed during the stochastic interpolation procedure. This research tested interpolation methods using both deterministic and stochastic methods such as kriging and inverse di stance weighting and it was found that inverse distance weighting gives the lowest estimated cross validation errors (Figure 5 7). The trips in the travel diary were categorize d into home based and non home based trips. The home based trips are categoriz ed as work, shopping, social and entertainment and other. The non home based trips are represented as one category. Spatial interpolation is performed on each category independently and the result is combined to give the final travel cost using the method explained in Chapter 3 The output surface shown in F igure 5 8 is an estimation of travel cost for Orange County. The surface shows, however, that the variation of travel cost in some locations is high. To reduce the variation, an average surrounding trav el cost surface can be generated by taking the mean surrounding travel cost within a walking distance. This value is assigned to the center of the neighborhood as the average surrounding travel cost (Figure 5 9). The average surrounding travel cost surface is generated by focal statistics tool (ESRI, 2011) using a Manhattan neighborhood size of eight hundred meters, which represents the walking distance in Transit Oriented Development (TOD) research. The
141 resulting surface is smoother due to the averaging pr ocedure performed by focal statistics. Predicting Travel Cost from Location and Urban Form Characteristics The spatial interpolation method works for places that are highly represented in the geo coded trip ends data of the NHTS 2009. The spatial estimati on also assumes cross sectional data, which means that the spatial interpolation output works for the year 2009. However, any future change in land use will have an effect on increasing VMT as well as travel cost as shown in C hapter 2 Therefore, a future travel cost prediction procedure has to include urban form characteristics. This includes capturing a longitudinal relationship by regression. In the longitudinal relationship, travel cost is sensitive to land use change and not limited to spatial location The travel cost estimation resulting from the spatial interpolation using the NHTS 2009 trip data is used in the regression model to capture the relationship between travel cost and urban form characteristics as well as aggregated neighborhood characteri stics. A random data set is generated for each county in the study area that includes approximatly three thousands points. The number of points, however, is reduced due to missing data at certain locations. The data includes the estimated travel cost at each of the random location as well as the following metrics that are taken from the land use transp ortation coordination litrature. 1. Density 2. Land use mix (Entropy) 3. C onnectivity 4. A ccessibilty to major retail and service activity agglomerations 5. Transit conne ctivity 6. Household size 7. I ncome
142 The estimation methods of these metrics are exp lained in Chapter 3 Other variables that are specific to the county location are also used for each county. Predicting travel cost by ordinary least s quares The first method used in the regression is the OLS method This regression is a global regression that does not take into account the spatial location of the points used in the regression table. However, spatially differing variables can be used in the reg ression table. OLS regression is the most common regression method used in relating transportation and land use variables (Ewing & Cervero, 2001 ; CNT; 2007). It uses a global regression function and may not capture the local variation which usually leads t o lower goodness of fit of the regression model. OLS regression is performed on the random data sets generated for Duval, Orange and Pinellas counties. The results of OLS models for the three counties show that different counties have different relationsh ips between travel cost and urban form charactaristics. The OLS model result for Orange County has an R 2 value of 0.640 while Duval County has a value of 0.450 and Pinellas County a value of 0.388 (Table 5 1). This means that the urban form charactaristics relationship to travel cost is strongest in Orange County and the weakest in Pinellas County. The result shows that the residuals have a normal distibutions. However, the spatial autocorrolation check proved that the regression models for the three counti es have cluster residuals This raises questions about the validity of the global regression for these counties and the need for other regression methods. Furthermore, the OLS regression model proved that the same urban form variables are important in Oran ge and Duval counties but most of the variables where insignificant for Pinellas County which also suggests that the geographic location and
143 local variables are important in finding the relationship between travel cost and urban form (Table 5 1). A tradit ional method of increasing the goodness of a model and reducing residual clustering is to use spatial discriminator such as binary variables. This was tested in Duval County. The new OLS model for Duval County includes additional variables such as land us e category and land value. Using additional variables slightly increased the goodness of the model and slightly reduces the residual clustering. The additional variables were not significan t in Orange or Pinellas counties which again suggests that the glob al regression using OLS is differet from one geographic region to another. The coefficient of the OLS indicates that increasing the surrounding density will decrease the travel cost. The travel cost also decreases if the land use mix, connectivity, access ibilty to major actvity agglomerations and transit connectivity increase. This result matches the literature on compact development. This is not true, however, for Pinellas County where the result indicates that increasing density and accessibilty will dec rease travel cost. The other variables such as land use mix, connectivity and transit connectivity however, are not significant. The clustered residuals indicate that the clustering is not random meaning that there are missing variables in the model. Due to the complex nature of the relationship between urban form and travel cost, it is extremely difficult to identify these missing variables. To validate the goodness of the global model, the percentage prediction error is calculated and mapped for the cou nties. Figure 5 10 shows the spatial distribution of the prediction error and that it is always less than 10% of the estimated travel cost. The distribution of the prediction error is low in rural areas which have have high travel cost,
144 and higher in the places that have low travel cost such as the CBD area. This can be an indication that some people are living in downtown areas but have large travel costs due to anomalies in the location such as bridges or avoiding congested roads. The prediction error ma pping shows that even though the residuals are clustered, the prediction error is low and the model can still be used for the prediction of the travel cost. The OLS model explains the global relationship between travel cost and urban form charactaristics. For example, increasing density will decrease travel cost and increasing connectivity will decrease travel cost. However, in areas that have geographic anomalies such as bridges that are used for travel, increasing density may increase congestion and as a result increase travel cost. Also, adding street links to the network will increase connectivity but eventually cars will need to travel through the bottle neck of the bridge. The global regression lacks the ability to capture these anomalies and other loc al spatial charactaristics. Therefore, this research suggests the use of geographically weighted regression in the estimation of the travel cost. Predicting travel cost by g eograp hically weighted r egression The same dependent variable and the significant i ndependent variables used in the OLS model are used in the GWR model. The dataset for the GWR model is also the same dataset used for the OLS regression model. However, the GWR model depends on the other data points that spatially surround each of the poin t in the dataset. This generates a neigborhood or a kernel around each point in the data set. There are two types of neigbo rhoods that can be used in geographically weighted regression. The first is the fixed neighborhood in which a distance is set around each point and the points that are inside the neigborhood are used in the regression. The
145 second type is the adaptive neigborhood where the user can decide the number of point that will be used in the regression. The number of points needed for the regre ssion in GWR depends on the number of explanatory variables. GWR regression fails if that minimum numbr of point is not satisfied, which places a restriction on the use of the fixed kernel in the travel cost regression model. Therefore, the adaptive kerne l is used. The choice of numbers of points was performed on a trial and error basis to identify the minimum number of points at which the regression will not fail in any of the three counties. This number was found to be 700 points. GWR models are tested f or the three counties. For Orange County, the GWR model has nearly the same R 2 as the OLS model, which is an indication that the global model of regression is a good model in that county. However, the explanitory variable coefficients are spatially distr ibuted and their values are not constant indicating differences between the GWR and OLS regression models. The results for Duval County were different. The GWR increased R 2 from 0.45 to 0.95 indicating that the GWR model has less prediction error than the OLS model. Table 5 2 shows the difference between the OLS and GWR regression models in terms of their R 2 values for the three counties. Comparison Between GWR and OLS Results The regression results using both methods show clearly that GWR models generally have larger R 2 than OLS models (Table 5 2), indicating that the relationship is stronger and the prediction errors are much less. The GWR model also solved the problem of the clustered residual and the need to find more explanatory variables, as explained in the aforeme ntioned OLS section. Therefore, one can say that the GWR model is a better predictor of travel cost. Even though this research focuses on methods
146 of predicting travel cost, explaining the meaning of the explanatory variables is also important to the resear ch of coordinating land use and transportation. OLS regression for Duval County gives a coefficient of the density explanatory variable 0.062. However, the coefficient for density using GWR for Duval County is not constant and varies spatially. The mean value of the density coefficient is 0.052. The mean coefficients have the same direction and means that, on average, increasing density will decrease the travel cost. However, some areas have the opposite direction of the relationship (Figure 5 11) meanin g that increasing the density in the positive relationship area will increase the travel cost. The logical explanation of that can be understood relating the variables in the regression together. Land use variables are complex in nature. The autocorrelatio n test shows that the variables are independent. On a local scale, these variables may not be absolutely independent. Figure 5 1 1 shows that the positive density relationship occurred in undeveloped areas with low connectivity and low land use mix. This me ans that in areas at the fringe of the county, increasing density without increasing land use mix and connectivity will not reduce the travel cost. The explanatory variable for connectivity also has a spatially discriminated distribution. The negative dir ection means that increasing connectivity means decreasing travel cost, which matches the results of research relating connectivity to VMT (Ewing & Cervero, 2001). However, there are also areas that have positive directions (Figure 5 12). The positive rela tionship occurs in two different places. The first is a place of low land use mix and low accessibility at the fringes of the county. The results show that increasing connectivity without increasing land use mix and
147 accessibility to destinations will not h elp to reduce travel cost. It is reasonable therefore to conclude that reducing travel cost requires adding retail and employment opportunities in these areas, in addition to increasing connectivity. That will increase connectivity, accessibility to destin ations and land use and at the same time decrease travel cost. The second place that has a positive direction in the relationship between connectivity and travel cost is in downtown Jacksonville, at the heartland of Duval County. The result shows an impor tant relationship captured by the GWR model that is not captured by the OLS model. This area has a bridge and increasing the connectivity by adding road links to the network without increasing the bridge capacity will not reduce travel cost. The GWR model captured the anomaly of the bridge that the OLS model could not capture. In conclusion, both the OLS and GWR models are good models for prediction. The importance of OLS is that it can explain what happens globally if the urban form characteristics or the land use is changed. The results were in agreement with the research with the land use and transportation research mentioned in the literature review. Therefore it can be used for policy makers at the county, state or even national scale. However, it does not capture what happens locally. The urban form has many anomalies that cannot be generalized. GWR can capture some of these anomalies. It can also predict values with less prediction errors. However, it is a local tool and it can be only used for the are a for which the model is generated. This conclusion is not different for OLS. The result for OLS for the three counties also proved that the significant explanatory variables are different from one county to another. The
148 explanatory coefficients are differ ent too, meaning that the OLS model also cannot be generalized and used in different geographic locations.
149 Table 5 1 Explanatory variable coefficients and their significance using OLS County Variable Duval Coef ficient Duval t stat Orange Coef fi cient Orange t stat Pinellas Coef ficient Pinellas t stat Density 0.062 11.82 0.051 17.42 0.099 33.54 Land Use mix (Entropy) 0.400 13.07 0.179 16.94 ------Not Significant Connectivity (link s /Node s ) 0.121 2.71 0.017 4.68 -------Not Significant Accessibility To Major Destination 0.012 7.82 0.043 5.79 0.005 2.51 Transit Connectivity 0.001 2.04 0.004 10.74 --------Not Significant House hold size 0.026 3.34 0.049 4.47 ---------Not Significant Income 0.002 10.34 0.001 2.54 0.001 6.56 R 2 0.452 0.640 0.388 Table 5 2 R 2 v alues for OLS and GWR County OLS R 2 GWR R 2 Orange 0.640 0.645 Duval 0.452 0.952 Pinellas 0.388 0.757
150 Figure 5 1. Proximity a ccess measures E uclidean distance Figure 5 2. Proximity a ccess measures network distance
151 Figure 5 3. Euclidean buffer opportunity access Figure 5 4. Network service area opportunity access
152 Figure 5 5. Combine opportunity distance access based on Euclidean distance Figure 5 6. Combine opportunity distance access based on network distance
153 Figure 5 7. ArcGIS cross validation chart for inverse distance weighted interpolation using work trip length as the interpolation field Figure 5 8. Travel cost generated by spatial interpolation
154 F igure 5 9. Travel cost smoothened within walking distance Figure 5 10. Prediction error as a percentage of travel cost
155 Fi gure 5 11. Reclassification of t he GWR density coefficient Fi gure 5 12. Reclassification of t he GWR connectivity coefficient
156 CHAPTER 6 INTRODUCING SUITABIL TY AUTOMATION TOOLS Land use planning has become more complex as the incorporation of sustainable development goals has increasingly taken place. Therefore, the modern planning process involves conflicting and contradi cting interests between conservation and economic development. In this new situation, planners face challenges as they attempt to design their projects while taking into account the balance between ecological e conomic development at the same time. This adds to the complexity of how planners handle suitability analysis. This complexity has been reduced by the development of GIS analysis, as well as the development of procedures that help to plan for several land use alterna tives before making a decision. ArcGIS software is equipped with analysis tools that can help planners in land use modeling and in making decisions concerning land use. Typically these tools have been designed to be used in many ways. One way t o use the tools is interactively, where the user uses icons and menus to initiate the tool. An other way is through the use of toolboxes, organized in a tree structure that classifies the tools into different categories. However, land use modeling is a sequ ential procedure (Carr & Zwick, 2007). This sequential procedure requires the user to use the GIS tools one after the other, which also allows the process to be automated. The automation process is supported in ArcGIS. The software allows the user to use p rogramming to prepare their customized tools or to incorporate the software built in tool in a batch process. The programming in ArcGIS is supported by many programming languages such as, Python, Visual Basic and C. It is also supported with graphical prog ramming software called Model Builder, where graphical symbols like circles, ovals, squares and rectangles, represent the input
157 and output data sets as well as the tools. The processes are usually represented by arrows or connectors that join the shapes to generate the process sequence. Model Builder therefore can do programming in a flowchart graphic pattern (ESRI, 2011). ArcGIS tools and programming methods are used intensively in this research. The aim of using programming is to create customized automat ed tools that can run complex procedures that are difficult to run using the built in interactive tool. Model Builder, Python and VBA are used to automate suitability modeling as well as in estimating urban form metrics an d other procedures. This research will introduce the tools that are used in suitability modeling. The testing of these tools was performed in land use modeling. However, these tools are used also to allocate land that is suitable for affordable housing as shown in Chapter 7 There are thr ee types of automation tools used in this research, they are: automation tools for suitability assignment, automation tools for suitability overlay and automation tools for the population allocation based on the combined grids. The use of automation tools in the research exceeds the three mentioned categories. However, the tool discussion only focus on the tools that can be used in most of the LUCIS land use models and the AHS model. The tool that works for the suitability assignment (A4 Suitability) is a r aster reclassification tool that reduces the manual calculation needed outside the GIS environment. The second category of tools is the suitability overlay toolbox which contains two tools. The first tool is for calculating suitability overlay weights acc ording to community values (A4 Community Values). The second tool is used to perform the overlay betw een suitability assignments (A4 LUCIS Weights).
158 The third tool category is the allocation toolbox which contains three tools, the trend allocation, the allocation by table tool and the detailed allocation tool (A4 Allocation tools). Suitability Assignment Tool The suitability index is a value that represents the relative usefulness for a land use. These values are assigned from one to nine in the LUC IS model where one represents the lowest suitability and nine is the highest suitability value (Carr & Zwick, 2007). The utility values are classified from one to nine using different methods according to the nature of the criteria to be evaluated or accor ding to the utility to be classified as a suitability surface. This classification procedure can be performed using interval, ratio, nominal and ordinal data (Carr & Zwick, 2007). Some of the procedures are simple and some of them have higher complexities. accurate, legally defensible, technically valid, ecologically sound, and open to scrutiny by the public 102 ). Prior to the introduction of the A4 Suitabilit y Tool the planner would take the MEAN, STD, and MIN or MAX statistics generated from Zonal Statistics as Table tool to manually calculate the suitability intervals for non binary classifications. Once the values for each interval are determined these val ues wo uld be manually input into the r eclassify tool. This method proved to be time consuming, cumbersome, and prone to error. This research introduces the A4 Suitability tool that performs these calculations for the planner and creates nine intervals betw een the MEAN and MIN or MAX values, depending upon whether the suitability is increasing or decreasing, from the input distance surface. The interface for that tool (Figure 6 1) shows that the tool is prepared to work in the GIS environment as a stand alon e tool
159 The tool can be also inserted as a customized tool usi ng ArcGIS Model builder. Zonal statistics as Table and Euclidean Distance (Figure 6 2 ) are typical tools that are used in LUCIS suitability models. T he A4 S uitability tool is inserted within the se models in Model Builder to automate the reclassification process (Figure 6 3). The output of the Z onal S tatistics as T able can be easily inserted into the A4 Suitability tool The tool uses the values from that table and do es all the required calculatio ns to generate the suitability table ranges between the minimum and maximum values. The tool will also generate the reclassification table for the raster. Figure 6 4 shows an example zonal statistics table for distance to shopping centers for Duval County, while F igure 6 5 shows the output reclassification table that is internally generated by the tool This table is automatically used to reclassify the input raster. However, it shou ld be noted that the zonal statistics table does not necessary contain on e row. In case of a multiple rows the zonal statistics table additional calculation is performed automatically by the tool. The tool can also accept a user generated reclassification table. The user can create any database format DBF table containing the reclassification ranges and the suitability score in the same format that is used to create the output table (Figure 6 5). This table can be generated using Microsoft Excel or Access software. The A4 Suitability tool has been verified many times by comparing the results of using the A4 Suitability tool with the results of the same raster reclassified by doing careful and time consuming manual calculations and using the ArcGIS raster reclassification to ols. The result is that the out put reclassified raster layers are always identical and the A4 Suitability tool work s only on the automation of the proces s within
160 the ArcGIS environment. It reduces the chance of the errors that may be caused by the user in the manual calculations. The tool also significantly reduces the time of the process execution Figure 6 6 shows an example verification raster for C ent ral Florida generated by the two mentioned methods. The tool code was programmed using Python pro grammin g language. Appendix (A ) contains the full listing of Python code for the tool. Overlay and Weighting Tool s This r esearch also introduces another new GIS tool the A4 Community Values Calculator. This tool integrates pair wise calculations into the ArcM ap environment as a VBA program to be used to support land use and affordable housing suitability models. The A4 Community Values Calculator is initiated by installing the program as an ArcM ap macro in VBEditor (Appendix B ) When evaluating the importance b etween goal, objectives or alternatives the A4 Community Values Calculator integrates any number of objective and/or sub objective raster suitability surfaces as inputs. The A4 Community Values Calculator interface prompts the user to specify the usefulne ss of each pair of raster surfaces and dynamically compares the raster pair (Figure 6 7) As the user indicates values for each pair, the A4 Community Values Calculator automatically populates a n internal pair wise comparison matrix. When the user enters the values for all pair wise comparisons, t he calculator then outputs a parameter table of the raster names and their corresponding relative weights (Figure 6 8) As a way to re flect community participation, the tool can accept participation from many par ticipant s at the same time in a community participation field (Figure 6 7) and
161 use their votes to generate weights. The tool uses the geometric mean to calculate the pair wise comparison scores generated from the votes of all participants. There are also methods to update existing weight tables. The first is to input the table of weights into the tool under the base table box. Then the user can either run the pair wise comparisons again and update the table using the update table button instead of the gene rate table button (Figure 6 7). The other way is to manually enter the new weights in the table provided by the tool interface and then select the update table button. This manual procedure fits if t he tool is used to update the weights according to expert or decision maker values (Figure 6 7) The weights generated by the A4 Community Values Calculator are automatically used by a new weighting tool (A4 Weighting Tool ) This tool is used in the LUCIS model and the AHS models within the Model Builder environment. The weighting tool depends on the DBF output table that is generated by the C ommunity V alues program (Figure 6 8) Therefore, updating this table at any time updates the weights associated in the suitability model whi ch allows the user to perform weighting scenarios and run the models according to the new weights within a community meeting Figure 6 9 shows the weighting tool in the Model Builder environment The input layers to the tool (Figure 6 10) should be in a raster dataset format. The tool uses map alge bra to combine the input raster grids and to generate the output raster according to the weights in the input table (Figure 6 11). The output raster is placed in the workspace directory specified in the input bo x for the workspace of that tool. The model (Figure 6 9) shows the input weighting table as a variable added in Model Builder. The tool is prepared to be used to automate models in the Model Builder
162 environment (Appendix B ). However, the tool also can be u sed as a stand alone tool and it has a user interface which allows the tool to be run interactively and in a n ArcGIS user friendly environment. It is also worth mentioning that the output of the tool was also verified by generating the weights manually usi ng the AHP procedure and combing the layer using the weighted sum tool provide in the ArcGIS environment and the results were identical. Allocation Tool s The A4 Allocation toolbox contains tools that allocate future population in land use models such as LUCIS. The tools are also used to allocate land for affordable housing as shown in Chapter 7 The new allocation tool box contain s three tools which are the Tren d Allocation tool the Allocation by Table s tool, and the Detailed Allocation t ool. The idea of the allocation tools is to generate an iterative environment that runs and display queries used to find the most suitable land for the allocation of future popu lation and populate them by proposed population values. The tool runs on combine grids which act like an enumeration container made up of values of many participating grids that the combine grid is composed of. The tools read the raster attribute table and use that table as a database where the queries can be applied. However, because the combine grid usually is a large file that may contain millions of records, the allocation process may take several days to be completed. Therefore, programming methods wer e used to reduce the size of the combine and increase the queries processing speed. Appendix (C) shows the source code for the three allocation scripts in Python. The Trend Allocation t ool (Figure 6 12 ) works on the conflict surface as well as other mask s and/or constraints to prioritize the allocation process using combined grids. The combine grid is prepared by an enumeration rule that combines many grids and
163 keeps their attribute values. The T rend Allocation tool works on an iterative procedure to allo cate all the available spaces specified by the conditions or constraints which could be the suitability values and the conflict scores. The Trend Allocation tool works on two conditions and six masks. The condition is mainly a query that can hold any num ber while the mask query only hold ones or zero s user leaves a field empty it will be considered optional and removed from the query. Understanding, the concept of generating the query is a first step to using the tool. However, the next important step is to understand the internal iteration provided by the trend tool. The input of the co nditions can be a multivariate separated by semicolons (Figure 6 12). The tool takes the value of the first number of the first condition and generates queries that iterate through all the numbers in the condition generating iteration query statements that are used to allocate population. The tool will also loop to all the values of the first condition which acts like an external or an outside iteration loop. The output of the tool can be taken by two different ways. The first is that the tool generates an output population raster and the second is that the tool populates year and population fields in the combine. If the output is to be extracted from the combine, the combine can be used in different reclassifications or summary methods that allow the displa y of allocation results. The detailed allocation tool has the same function as the trend tool but with more conditions added. The previous trend tool work ed for six masks and two conditions for an allocation. Th e detailed allocation can work on six different masks and six additional
164 conditi ons for an allocation procedure. These added conditions act like inner loops for the iteration. Therefore, careful use of this tool can allocate the population in a region in one step using the power of iterations provided by the tool. However, the detailed tool has a user friendly interface. It is a complex tool and the user should understand exactly how the iteration is performed in the tool before using it. The simplicity of the Allocation Trend tool and t he det ails in the detailed tool have been compromised to create a cross tool (The Allocation by Table Tool) that allow the user to perform simpler iteration inside the tool and other external iterations in the Model Builder environment. The Allocation by Tabl e tool is more sophisticated than the Trend Allocation tool and simpler than the detailed tool. The tool can work with up to eight conditions of which two are iterative inside the tool. Larger iterations external iterations are performed in the model build er environment and are directed by the scenario table which adds more flexibility to the process. The conditions used by the tool vary in complexity depend on the user and what they are trying to do in their scenario. The model iteration in Model Builder is performed by a list (Figure 6 13). These iteration lists are taken from a scenario table (Figure 6 14). Therefore the tool uses the values of the iteration table row by row to allocate the population using the conditions specified by the fields of the t able. The tool also populates a field in the scenario table containing the summary of population allocated in each of the iterations (rows). The row of the scenario table is an external iteration performed by model builder. The internal iterations are perf ormed by the values of condition separated by semicolons, as explained in the Trend Allocation tool.
165 The most important development in the tool is to use conditions instead of masks. Masks are a one or zero value raster while the condition can be any quer y (Figure 6 14). The allocation by table can also be seen as a planni ng table or a scenario builder where the planner enters the conditions for an allocation depending on each conflict score or on multiple sets of score. Using this tool the planner can per form the allocation for a specified year or for many year increments at the same time. This tool is tested and used to allocate the future population in a transit scenario for Orange County using a transit accessibility grid, land use mix and other proxim ity conditions. The layers (Figure 6 15 and Figure 6 16) show that the allocation is compacted around transit lines and mixed used areas, which was expected from the scenario. The tool was checked and verified by using the query statements generated by the tool against manually generated queries depending on the same allocation conditions. Both the automatically generated and the manually generated queries are used to allocate population. Identical results were obtained in the validation procedure. Using c ombine grids facilitates the automation procedure of the allocation process. However, the tools give a raster output file and at the same time update the input combine grid. The combine grid can be also used to present output results by doing different rec lassifications or summaries. Additional fields that are not used as allocation conditions are placed in the combine. Examples on these grids are TAZ, Census tracts and Census block groups. The idea of using these grids in the scenario is to summarize the o utput of the allocation to their zones. Figure 6 17 summarizes the scenario population to the corresponding block groups while Figure 6 18 summarizes the output to different land uses
166 The A4 Allocation tools are used also to run scenarios for affordable ho using as explained later in Chapter 7 LUCIS conflict strategies and the use of combine grids are used for the allocation of affordable housing. All the tools explain earlier are adopted in the allocation procedure of affordable housing.
167 Figure 6 1. A4 Suitability tool interface Figure 6 2 LUCIS model illustrating the economic suitability of single family residential land use to retail and shopping opportunities
168 Figure 6 3 Integration of the A4 Suitability t ool in a LUCIS model
169 Figure 6 4 Sample z onal statistics attribute t able Figure 6 5 Output table of A4 Suitability t ool
170 Figure 6 6 Proximity to shopping centers in Duval County reclassifie d according to the average and standard deviation of the distance to m ultifamil y r esidential parcels
171 Figure 6 7 The interface of the A4 Community Values p rogram Figure 6 8 Output weights table
172 Figure 6 9 The A4 Weighting t ool Figure 6 10 Input layers for the Weighing Tool
173 Figure 6 11 Output lay er from t he Weighing To ol
174 Figure 6 1 2 Trend Allocation t ool
175 Figure 6 1 3 Planning by Table t ool
176 Figure 6 1 4 Planning t able Figure 6 15 Population allocation map for Orange County transit scenario
177 Figure 6 16 Population allocation around transit stops Figure 6 17 Summarizing results to block groups
178 Figure 6 18 Summarizing results to land use
179 CHAPTER 7 AFFORDABLE HOUSING ALLOCATION Generating the Affordable Housing Opportunity Surface The affordable housing opportunity surface is composed of four goals. The first goal is to evaluate suitability of land parcels for affordable housing based on physical and neighborhood characteristics using the LUCIS suitability (Carr & Zwick, 2007). The second goal is to evaluate travel cost in terms of it is burden on low income population. The third goal is to evaluate household rent and it is burden on low income populations. The last goal is to evaluate land parcels in terms of their transit accessibi lity. This opportunity surface represents a combination of these goals. Taking into account that the general objective is to allocate suitable land for affordable housing, the preference for affordable housing is set as high if the suitability of the first goal, which is based on physical and neighborhood characteristics, is high. When the cost of travel is low the travel cost preference, i.e. the second goal, is set as high. The same is applied to the third goal, household rent, where the preference is se t to high when the rent is low. The fourth goal preference is set to high where the transit accessibility is high. The opportunity surface is thus a preference matrix that is a combination of the objectives using LUCIS conflict strategies (Carr & Zwick, 20 07). LUCIS methods are used to evaluate the opportunity of a location for affordable housing. For example, a combination of low travel cost and low rent associated with good physical and neighborhood characteristics could be identified as affordable and a t the same time another location of higher travel cost and high transit accessibility could be regarded also as affordable. The use of the conflict strategies adds flexibility to the
180 location choice of affordable housing without the tradeoff between goals that is usually performed in suitability models. Physical and Neighborhood Characteristic Preference Surface One of the goals of the AHS model is to generate suitability according to physical and neighborhood characteristic and then find suitable locatio ns for allocating residential housing in general (Figure 7 1) The structure of the suitability model contains many SUA and MUA surfaces that are weighted according to community preferences. During the AHS project execution, many webinars were held that in cluded planners from the case study areas, which were Duval, Orange and Pinellas counties. These webinars had two aims; firstly to include the local planners in the weighting process of the suitability model; and secondly to generate weights that were then used to update the model table of weights and to generate the final AHS suitability surface. Figure 7 1 shows an example of an AHS Goal 1 surface for Duval County. The output surface of the affordable housing Goal 1 is reclassified based on the mean and standard deviati ons to three preferences high, medium and low (Figure 7 2). The high preference areas are mainly areas those that are preferred for residential development based on parcel and neighborhood characteristics as well as parcel accessibility to services. Many of these places, however, may not be affordable for very low income families. Rent Preference Surface The updated 2009 Census M ortgage and Rental data taken from the Shimberg Center for Affordable Housing at the University of Florida is us ed to create surfaces that represent rent, mortgage or a weighted combination of both rent and mortgage to obtain the housing cost. In the case of very low income population s that are investigated in the
181 affordable housing scenario, however, rent alone is used to create the housing cost preference surface. This surface is a spatial model representing housing rent prices across all of the individual count ies, reclassified according to the threshold used in affordable housing research. According to HUD ( 2011 a ), the affordability threshold is 30% of VLI which is approximately 50% of the AMI The housing cost preference surface is created directly from the rent surfaces by reclassifying the rent burden values. High preference is assigned to areas where the rent is lower that 30% of the VLI M oderate preference is assigned to areas that have rent lower than 40% of VLI. Areas with a rent burden of more than 40% VLI are considered low preference surfaces. Figure 7 3 shows the rent surface for Orange County and F igure 7 4 shows the rent preference surface obtained by reclassifying the rent surface. Travel Cost Preference Surface The travel cost surfaces were created as explained in C hapter 4 by either spatial interpolation or regression using urban form and land use characteristics taken from the 5Ds research (Ewing & Cervero,2001; Ewing et al 2008; Lee & Cervero, 2007). However, these surfaces are transformed to monthly values to match the rent surface which is estimated based on monthly rent burden (Figure 7 5). The travel cost surface is also reclassified according to transportation cost thresholds used for affordable housing research CNT (20 11 ) used the combined percentage of 45% as the combined transportation and housing burden out of income. This means th at the transportation cost should be less than 15% of income with regard to location affordab ility Therefore, the travel cost surface is reclassified to give a high preference for areas of transportation cost that do not exceed 15% of VLI. The second pref erence of 25% VLI is used to decide the moderate preference. Any location of transportation cost more than
182 25% VLI is considered to be a low preference location. Figure 7 5 show s an example travel cost surface for Duval County and F igure 7 6 shows the trav el cost preference surface for that county using the mentioned threshold values Transit Access Preference Surface The transit access methodology explained in C hapter 3 is used to create transit access surfaces for the three co unties based on the actual tr ansit stop route frequency combination in addition to the parcel data from which the employment opportunities were derived The estimated transit accessibility scores for residential parcels are reclassified to create the transit access surface based on th e mean and standard deviations obtained from the zonal statistics for residential parcels (Figure 7 7). The suitability surface for transit accessibility is then reclassified using the mean and standard deviations into three preferences that are high mediu m and low (Figure 7 8) Figure 7 7 shows an example transit access surface for Pinellas County while F igure 7 8 show s the preference surface obtained from the transit access surfaces for the same county. Access Rent Driving Transit Opportunity Surface (ARDT) The four preference surfaces are combined in a conflict surface using LUCIS conflict strategies (Carr & Zwick, 2007) The first surface is the preference based on physical characteristics and neighborhood accessibility and relates to the letter A in the ARDT opportunity surface The second surface is the rent preference and corresponds to the letter R. The third is the driving travel cost D and the final surface is the transit access T. The LUCIS methodology as applied in this research deals wi th conflict between these four components, accessibility, rent, driving cost and transit access. This research investigates the opportunity by having the components preference
183 combination as an opportunity for affordable housing. The generated ARDT opportu nity surface is shown by Figures 7 9, 7 10 and 7 11, which correspond to O range, Duval and Pinellas cou nties respectively. The opportunity surface s can identify area s that have high preferences in location and transportation cost and also identifies places that may not have low driving cost but have alternative transportation options like high transit accessibility. Refined ARDT Opportunity Surface In the case of affordable housing, the opportunity surface is easier to understand compared with the conflict surface. The opportunity components A, R, D and T have the same preference direction when investigating affordable housing opportunity. Having the number 3333 in the opportunity surface means that the location is highly preferred for affordable housing bas ed on the evaluation of the four A, R, D and T components. The first digit refer to location and indicates in this example that the location is highly preferable in terms of physical characteristics and neighborhood accessibility to services. The second di git indicates that the location is highly preferred for affordable housing based on rent values. The third digit shows that the location is highly preferred for affordable housing based on driving cost, and the fourth digit shows that the location is highl y preferred for affordable housing based on transit accessibility. A digit of 2 in the combination indicates that a location has a moderate preference in the component that corresponds with the 2 digit. For example, 3223 indicates moderate preference based on the rent value and also based on the driving cost. A digit of 1 in the combination shows a low preference in the corresponding component. However, for affordable housing allocation values 2 and 3 are considered places of opportunity while a value of 1 in any of the categories collapses the opportunity for affordable housing. High or
184 moderate transit accessibility does not trade off the need for good neighborhood characteristics. Therefore, for affordable housing scenarios, the opportunity surface can be refined by masking out locations that have low preference in one of the four components. Respectively, the Figures 7 12 through 7 14 show the refined opportunity surface for Duval, Orange and Pinellas counties. All of these surfaces have moderate and high preferences and do not contain the low preference categories. Impact of Travel Cost and Transit Accessibility on Affordable Housing Opportunity The ARDT opportunity surface is combined from preferences of the affordable housing preference components based on physical and neighborhood characteristics, rent burden, travel cost and transit accessibility. Another opportunity surface can be generated from the affordable housing preference based on the physical and neighborhood characteristics (A) and rent burde n (R) without taking into account the driving cost or the transit accessibility (the AR opportunity surface). Respectively, the Figures 7 15 through 7 17 show the AR opportunity surfaces for Duval, Orange and Pinellas counties. A comparison between locatio ns that is preferred for affordable housing based on the two opportunity surfaces has been conducted. The first comparison is performed by mapping the areas in each opportunity combination. To do that the ARDT categories are collapsed and summarized into A R categories (Table 7 1). The ARDT surface is then reclassified using the collapsed category. The result is two surfaces that have the same opportunity combination. However, these surfaces are different because one of them includes the goals of travel cost and transit accessibility. The total number of opportunity acres in each surface had been compared. Table 7 2 shows that in Duval County, the total number of preferred acres for affordable
185 housing is reduced by 76.7% if we include the transit accessibili ty and travel cost. Many of the preferred locations based on the AR opportunity surface are less preferred or even not suitable when travel cost and transit accessibility are included in the ARDT surface. However, this percentage is varied among different categories as we can see that the impact of travel cost and transit accessibility is lower for the most preferred land depending on physical characteristics and neighborhood accessibility such as 33 and 32. This is because these areas have good proximity t o neighborhood amenities such as shopping centers which in turn reduces travel cost and eventually results in many of these places already have good transit access. The same result is obtained for Orange County. However, the impact of travel cost and trans it access is much stronger in Orange County (Table 7 2) where the preferred land for affordable housing is reduced by 89.5% due to incorporating transit accessibility and travel cost. Pinellas County has the least impact of the three counties, where includ ing travel cost and transit access decreases the preferred land for affordable housing by only 59.2%. Metrics that are derived from conceptualizing the sprawl and compact development research (Galster et al 20 01 ; Ewing & Cervero 2001) have been used to compare the two surfaces. Distance to is compared between AR and ARDT (Table 7 3). The distance to CBD is also one of the metrics derived from conceptualizing sprawl that captures the centrality of the allocations. The results show that including tr ansit access and travel cost reduces the distance to CBD. This means that the ARDT opportunity surface produces a more central allocation of affordable housing than does the AR opportunity surface, which results more sprawl. However, the degree of change i n centrality is different among the three counties. Table 7 3 shows
186 the impact of transportation cost and transit access on centrality is the highest in Orange County and the lowest in Pinellas County. This matches the result of surrounding density as wil l be shown later and is a sign that the impact of transportation and transit access is indicative of a more compact development pattern that has access to the high employment densities. Furthermore, the impact of travel cost and transit access is lower in the high suitability areas depending on proximity, such as 33 and 32. The distance to activity center agglomerations is compared between AR and ARDT surfaces (Table 7 4). This is a metric that also captures the poly nuclear pattern of development and can be used as a metric for sprawl conceptualization and/or compact development patterns. The mean distance to activity centers is also lower in the ARDT surface which also indicates that the allocation of affordable housing using ARDT is less sprawling and te nds to be a more compact development pattern. Density is tested on the areas surrounding locations identified in the opportunity surface as being suitable for affordable housing, using a walking distance neighborhood of 0.5 miles. A compact development trend should, in theory, increase the surrounding density. The effect of existing density depends on how many green field, low density parcels are in the surrounding neighborhood and theoretically should have the effect of increasing density, indicating the tendency toward compact development (Table 7 5) The results of surrounding density show that in general the preferred lands have higher surrounding densities if we take transit access and travel cost into account. The result also shows that the change in the mean density is the highest in Orange Count y and the lowest in Pinellas County which again matches the same result taken from the allocated acres. The result also shows that the least change in the mean density occurs for the
187 preferred categories based on physical and neighborhood characteristics, which indicates that the areas that have good proximity to amenities usually have higher densities. Surrounding land use mix, represented in entropy values, is also compare d between AR and ARDT surfaces. The results in the T able 7 6 show that the ARDT surf ace has larger land use mix values than the AR surface. Land use mix and density have been used in the literature for land use transportation coordination and Transit Oriented Development TOD, as shown in the literature review. The impact of land use mix a nd density have been proved by different research t o lower VMT and increase transportation options other than driving (Ewing & Ce rvero, 2001; Ewing et al 2008; Lee & Cervero, 200 7). From the result in Table 7 6 we can conclude that the ARDT leads to less travel miles and more use of transportation options, which also corresponds with the compact development literature. Tables 7 7 through 7 9 summarize the comparison results for the three counties and shows the mean and standard deviat ion for all the test s performed to compare the AR and the ARDT tests. The tables also compare the AR, ARDT surfaces to the Assisted Housing Inventory (AHI). The AHI is 2009 data is taken from FGDL The results show that comparing AR and ARDT surfaces to the AHI in terms of de nsity, land use mix, distance to CBD and distance to activities suggests that the ARDT surface is more comparable to AHI than the AR surface. However, the results show that in many instances, the model gives less distances to CBD and activity centers when com pared to the AHI data. The results also show that transit access and travel cost impact the allocation of affordable housing and
188 that including these transportation variables results in a more compact development pattern. The use of ARDT surface s may have some political issues. The refined ARDT surface identifies the opportunity for low and moderate driving cost. This surface also identifies high to moderate transit accessibility. This suggest s that both driving cost and transit accessibility are important for the affordable housing allocation. Historically, housing affordability was connected to the driving cost and no t transit in what is known 19 shows the ARD surface where neighborhood accessibility and phys ical characteristics corresponds to the letter A, rent preference corresponds to the letter R, and travel cost preference corresponds to the letter D This surface include s the low and moderate travel cost and does not contain the transit accessibility. Fi gure 7 1 8 shows locations that are suitable for affordable housing and does not have any transit accessibility. People living in these areas depend on cars for commuting to work as well as other household trips. Figure 7 19 shows the ART surface where neig hborhood accessibility and physical characteristics corresponds to the letter A, rent preference corresponds to the letter R, and transit accessibility preference corresponds to the letter T. This surface include s high and moderate transit accessibility bu t does not include driving cost. In the ART surface, all of the affordable housing site s are within a walking distance from transit stops. However, because the driving cost is not included, the allocation of places that have high transit accessibility and high driving cost is expected. The allocation of these places assumes that people are using transit in their daily commute.
189 The ARDT surface has both transit accessibility and driving cost in an opportunity surface which suggest s that people have the oppo rtunity to decide on the transportation mode they want to use. The travel er may consider using transit if it is more affordable for the work commute. However, the traveler may also prefer to use the driving mode in situation s such as where trip chaining is required or driving is more affordable The aforementioned comparison between ARD and ART surfaces suggests the importance of both driving cost and transit accessibility in the opportunity surface. Therefore, using the ARDT surface is better than using the ARD or ART surface for allocating affordable housing. T he allocation of affordable housing using the ARDT surface has also more steps and includes the use of more variables and con straints other than the transit access and travel cost. The process of a llocation of affordable housing is performed using the allocation scenarios and will be explained in the next section. Affordable Housing Allocation The results in the aforementioned sections suggest that the travel cost and transit accessibility have an i mpact on the affordable housing opportunity surface. Therefore, this research uses the ARDT opportunity surface for the allocation process for affordable housing. The ARDT opportunity surface includes the land that is moderately or highly preferred for aff ordable housing based on the four goals that generate the opportunity surface. However, more conditions are needed in the selection of land for affordable housing. These additional conditions may address a certain policy or priorities based on a certain s cenario. An affordable housing final preference score can be assigned using the allocation tools created for the allocation of land and people in land used models such as LUCIS. Acres for affordable housing are used as the allocation field using the A4 Al location
190 tools presented in C hapter 5 This allocation process is based on iterations. The first iteration targets locations with the highest preference while the last iteration the lowest preference. Parcels and locations that do not meet the criteria set in the conditions are left unranked which means that their preference is lower than the lowest preference in the ranking ladder. However, these unranked parcels are still suitable for affordable housing but they did not satisfy the ranking conditions used in the specified scenario. The allocation tools work on a combine grid. The combine grid is composed of several grids that represent the opportunity surface as well other grids that can work on refining the places for affordable housing or adding restric tions or constraints on the process. The grids can also represent a change in policy that need to be tested such as new transit lines. The scenario set in the research mainly investigates the opportunity for allocating affordable housing based on the ARDT and additional compact development constraints. The scenario looks at livability indicators such as walkabilty and refines these places according to their density and other variables that are important for compact development. Here, the scenario will targe t areas that are underutilized when compared to their surroundings. Other variables such as Enterprise Zones, Qualified Census Tracts and places qualified for Community Reinvestment Act funding (CRA) can be used in the allocation process. The scenario in t his research, however, focuses on compact development and reducing travel cost and does not address areas of distress or other policy incentive areas. The compact development scenario may differ from one county to another depending on the availability of d ata. For example, the livability index includes walkabilty and crime for Orange County while no data for walkability exists for Duval or
191 Pinellas counties. The following section will explain the grids used for allocation in addition to the ARDT opportunity grid. Creating the Combine Grid for the Allocation The combine grid tool (ESRI, 2011 ) is used to create the base combine grid for each county. The combine grid basically is a grid enumeration tool that can hold multiple grid values for each cell. The grid s that are used to create the combine may be different from one scenario to another and will also depend on the political region and boundary. This research used a standardized format for the layers included in the grid Table s 7 10 through 7 12 show the a llocation scenario conditions for Duval, Orange and Pinellas counties Differences in the grid used for allocation may occur only on the livability grid which will be a walkability bikabil i ty grid for Orange County and a crime density suitability surface f or Duval and Pinellas counties. The generated raster will have no significant meaning for the value field. However every grid in the combine is represented as a field in the attribute table for that combine (Figure 7 20 ). The following sections will expla in the grids used in the combine grid in addition to the ARDT opportunity grid. Underutilized density g rid The underutilized value for land in this research is a number that compares the cell density for a location to the surrounding area density in resid ential units. The surrounding neighborhood is taken as a quarter mile Manhattan distance that surrounds each cell. The underutilized density is the defined as the number of residential units that could be added to the cell to match its average surrounding density value. Cells that have a density value more than its surrounding neighborhood is assigned as a zero underutilized density. The importance of this surface is to capture the parcels that might
192 be preferab le for redevelopment. Figure 7 21 shows an exa mple underutilized density sur face for Duval County Livability g rids The livability term here is a general term for a category of grids that represent walkability and safety. This grid, however, may be different from one county to another depending on the availability of data. The grid is a utility assignment grid. Therefore it could contain one grid as a single utility assignment, such as a crime avoidance suitability grid (an SUA), and it may contain another grid to estimate a walkability or a bikability MUAs The crime suitability assignment uses the local crime incidents for the year 2006 in a kernel density estimatio n (ESRI, 2011 ) using a 0.25 mile radius which is the walking distance to create a crime density raster. The raster is then transformed to a suitability surface based on the mean and standard deviation values for the residential parcel zonal values. The reclassification uses a decreasing suitability reclassification using the A4 Suitability tool. The walkability and bikability surfaces are a multiple utility surface that incorporates other walkability and bikability indicators such as sidewalks, bike lanes, transit stops and crime densities within the walking and biking distances to generate a multiple utility assignment (Figur e 7 22 ). Land c haracteristics The land characteristics are any parcel characteristics grids that are important in the allocation of affordable housing. For example, a reasonable step is to allocate into vacant parcels, which is regarded a s an infill proces s (Figure 7 23 ). However, not all of the allocations are looking for vacant parcels. There are also parcels that are good for redevelopment and may have a priority in the allocation process based on the conditions set in the scenario. Therefore other parce l variables are also used in the
193 allocation. This includes land values. Land values are based on the Just Value per acre calculated from county property appraisal data for the year 2009. The grid is generated after cleaning the data which includes removin g outliers. The grid is then reclassified into user defined categories according the price ranges (Figure 7 24 ). Proximity g rids The distance to the CBD has been used in studies either as an indication of compact development or as a measure of sprawl. The distance to CBD is a metric of mono nuclear urban pattern and represents the centrality of the mono nuclear urban areas (Galster et al 20 01 ). However, the distance to CBD will not work independently and lead to compact development. It should be combined with a density or concentration and other compact development variables for that purpose. In this research a distance to CBD raster is generated for the allocation of affordable housing in the compact development scenario. The surface is generated by getti ng the central district feature data sets from the FGDL and finding the central feature of that dataset. The distance to CBD is mainly the distance to that central feature. The raster is a Euclidean distance raster that is reclassified into equal intervals of two miles and u sed in the scenario (Figure 7 25 ). The allocation prioritization depending on this raster is that the scenario will look in the first two miles away from the CBD in the first iteration and then increase the distance by two miles for othe r iterations in the same conflict/opportunity category. The other proximity raster used in building the scenario is the distance from activity agglomeration centers. The inclusion of this surface also corresponds with the compact development or the Trans it Oriented Development patterns. This also, if combined with clustering or density, leads to a poly nuclear pattern of development
194 (Galster et al 20 01 ). The proximity grid is created by identifying the larger agglomeration of activities as major activit ies. This is done by the agglomeration of the activity values within a walking distance. By using spatial overlay, the values of the activity square footage are aggregated and the larger values are selected. In this research a value of one million square f eet or more is selected as a major activity. The proximity grid is created by taking the Euclidean distance away from these major activity centers and then reclassifying the grid into equal inte rvals of half miles (Figure 7 26 ). During the scenario iterat ions, the allocation prioritization, depending on this grid, is that the scenario will look in the first half mile away from the major activities in the first iteration and then increase the distance by half mile for subsequent iterations in the same confl ict/opportunity category. Policy g rids The policy grids are prepared to test a certain policy in the scenario. For example, testing a proposed bus route on the allocation will include preparing a new transit access surface and use it as a policy grid. In this research, incentive zones such Enterprise Zones, Qualified Census Tracts (QCT) and Community Reinvestment Act (CRA) areas are regarded as policy grids. These areas are mainly distressed areas that people will avoid if there are no incentives. The generation of grids for these areas is done by assigning a constant value of one for areas outside the zones and a zero for inside t he distressed areas (Figure 7 27 ). The grids can be used by allocating away from distressed areas in a poverty de concentrat ion scenario by allocating in the areas that have a value of one or by allocating inside of these areas in a policy incentive scenario.
195 Zoning g rids Census Tract s and TAZ are grids that are used for summary purposes. This research does not use Census Tract s or TAZs in the allocation of affordable housing in general. However, that does not mean they are not useful in the allocation for certain scenarios. In this research, these zones are used for summary purposes only. The generation of these grids is by tr ansforming the zonal feature data into a raster grid data (Figure 7 28 ). Compact Development Scenarios The compact development scenario is based on using the allocation tools on combine grids. Therefore, a combine grid is generated for each county in the study area, namely, Duval, Orange and Pinellas counties. The combine grid is a raster grid that has a large attribute table showing the values of each of the combine grids for a cell or a group of cells (Figure 7 20 ). Other than the livability field diffe rence between counties mentioned earlier in this section, the fields used in the combine are the same. These fields include underutilized density, opportunity/ conflict, land values, proximity grids, vacant lands and other zoning grids. The affordable hous ing allocation model (Figure 7 2 9 ) uses the Allocation by Table tool that was explained in Chapter 6 Generally, the scenario table used for the allocation (Table 7 10) is replicated for the three counties. However, the priorities may differ slightly from county to county. The scenario table summarizes the conditions set for the scenario. The first two digits in the ARDT conflict/ opportunity are used to prioritize the process. The zonal statistics table for existing density, distance to CBD, and distance t o activity center agglomerations are used to set the general direction of priorities which was 33; 23; 32; 22 for Duval and
196 Pinellas and 33; 32; 23; 22 for Orange. Tables 7 10 through 7 12 are the scenario tables for Duval, Orange and Pinellas counties res pectively. Using the allocation tool and the scenario table, allocation maps that show the allocation of affordable housing are generated for Duval, Or ange, Pinellas counties (Figure 7 30 ), (Figure 7 31 ), (Figure 7 32 ). The maps have the ranking scores of 1 to 16 where one is the most preferred land and 16 is the lo west preferred land. The maps also show a 4 digit opportunity number for the areas that do not meet the ranking criteria in the table. However, the opportunity digit is assigned to these areas in case the planner wants to investigate the opportunity for affordable housing in these are as knowing that these areas are of a lower rank than the ranked areas depending on the scenario table. The maps also show line feature classes for the transit lines which suggests that most of the allocations are in areas that have strong transit opportuni ty. The resulting allocations for affordable housing based on the compact development scenario are tabulated according their original land uses (Table 7 13), (Table 7 14) and the acres allocated associated with each land use. Table 7 13 shows the major lan d uses while Table 7 14 shows the minor land uses. The allocation results show that a large percentage of single family residential areas are qualified for affordable housing in addition to multi family residential areas and other categories like commercia l uses. The single family residential areas are in locations that have high densities in their surrounding neighborhoods and they are close to transit and employment opportunities which make them suitable for affordable housing. For further analysis of aff ordable housing on a parcel level, the output of the model can be tabulated to the parcels according to the parcel identification number. The output of the
197 tabulation is then joined back to the parcels and more analysis can be performed on the parcel level to decide on the final suitability of a specific parcel for affordable housing. Chapter 8 will emphasize the use of parcel analysis for housing affordability as part of the research conclusions and recommendations.
198 Table 7 1 AR and ARDT equivalen t categories ARDT category Equivalent AR category 3333 33 3332 33 3323 33 3322 33 3233 32 3232 32 3223 32 3222 32 2333 23 2332 23 2323 23 2322 23 2233 22 2232 22 2223 22 2222 22 Table 7 2 Tabulated t otal acre in each AR c ategory Equivalent AR category Acres in AR Acres in ARDT Duval County 33 31034.73 18681.87 32 26274.20 9145.85 23 78198.31 15017.97 22 64430.14 3861.95 Total 199937.38 46707.64 Orange County 33 3725.17 1701.46 32 17733.89 4824.88 23 28803.24 1242.20 22 78942.30 5731.06 Total 129204.60 13499.60 Pinellas County 33 4258.99 2744.18 32 10360.98 4632.06 23 12461.62 5424.25 22 22657.09 7449.38 Total 49738.68 20249.87
199 Table 7 3 Zonal statistics for distance to CBD Equivalent AR category AR Mean Distance AR STD Distance ARDT Mean Distance ARDT STD Distance Duval County 33 3.100302 2.578092 2.802506 2.390863 32 7.785642 3.404559 6.849874 3.769736 23 4.9596 60 3.863394 2.212663 2.656914 22 10.856486 3.369811 7.964609 3.718822 M ean 6.942815 3.832192 Orange County 33 2.014736 1.489937 1.359414 1.205063 32 5.121713 3.417311 2.987531 1.834654 23 13.98262 5.341965 3.365938 2.892298 22 10.98086 0 5.357544 5.299049 2.069534 M ean 10.58732 0 3.798471 Pinellas County 33 2.964458 2.780636 2.321152 1.973747 32 5.303665 2.377305 4.524874 2.106511 23 5.035321 3.275428 3.371814 2.046297 22 5.971839 2.910384 4.665094 1.937563 M ean 5.341206 3.968888 Table 7 4 Zonal statistics for distance to major activity ce nters Equivalent AR category AR Mean Distance AR STD Distance ARDT Mean Distance ARDT STD Distance Duval County 33 2.078394 1.489471 1.950296 1.441228 32 2.862418 2.225353 3.052764 2.62364 0 23 4.990358 3.604055 2.744263 1.872097 22 3.522871 2.073929 4.653375 2.932191 M ean 3.785822 2.644955 Orange County 33 1.511171 0.65961 0 1.296026 0.67313 0 32 1.956474 1.358564 1.182909 0.682695 23 8.738854 4.378703 2.047809 0.862708 22 4.761751 4.130088 1.328445 0.878023 M ean 5.169607 1.338536 Pinellas County 33 0.840397 0.53267 0 0.787475 0.501397 32 1.259383 0.731996 1.141977 0.688911 23 1.006272 0.590087 0.991877 0.595499 22 1.305712 0.83728 0 1.020384 0.588548 M ean 1.1812 00 1.008991
200 Table 7 5 Zonal statistics for density of surrounding Equivalent AR category AR Mean Density AR STD Density ARDT Mean Density ARDT STD Density Duval County 33 3.2239 2.0765 3.3695 2.2164 32 2.7130 2.0152 2.8720 2.1883 23 1.7555 1.7605 2.6443 1.9207 22 1.2838 1.6493 2.7215 2.3418 mean 1.9573 2.9853 Orange County 33 13.2905 17.9955 17.5090 22.9969 32 7.7478 6.7021 10.2031 7.8328 23 3.0430 2.9497 6.4122 2.4358 22 5.4815 7.0586 10.4427 7.6341 mean 5.4741 10.8769 Pinellas County 33 3.5393 2.3712 3.6140 2.4247 32 2.5754 1.3452 2.6282 1.3762 23 2.6566 2.0439 3.3789 1.9898 22 2.0865 1.5662 2.3728 1.6562 mean 2.4552 2.8690 Table 7 6 Zonal statistics for entropy of surrounding Equivalent AR category AR Mean Entropy AR STD Entropy ARDT Mean Entropy ARDT STD Entropy Duval County 33 0.5282 0.1294 0.5486 0.1229 32 0.4770 0.1450 0.5227 0.1486 23 0.3860 0.1796 0.5090 0.1216 22 0.2558 0.2033 0.4722 0.1586 mean 0.3781 0.5244 Orange County 33 0.4412 0.1184 0.4462 0.0987 32 0.3337 0.1781 0.4514 0.1123 23 0.2693 0.1346 0.4663 0.1117 22 0.2616 0.1714 0.4609 0.1180 mean 0.2783 0.456 2 Pinellas County 33 0.2169 0.0213 0.2196 0.0197 32 0.2137 0.0223 0.2168 0.0199 23 0.2033 0.0314 0.2127 0.0250 22 0.2063 0.0360 0.2157 0.0263 mean 0.2079 0.2157
201 Table 7 7 Collective measurements for Duval County Measurement Statistic AR ARDT AHI Acres Mean 199937.38 46707.64 Distance to CBD Mean 6.9428 3.8321 3.9282 STD 3.4445 2.8562 3.7573 Distance to Major Activity centers Mean 3.7858 2.6449 2.5935 STD 2.601 6 1.9345 2.4658 Surrounding Entropy Mean 0.3781 0.5244 0.5362 STD 0.174 9 0.1304 0.1097 Surrounding Density Mean 1.9572 2.9853 4.3606 STD 1.8071 2.1262 1.9391 Table 7 8 Collective m easurements for Orange County Measurement Statistic AR ARDT AHI Acres Mean 129204.60 13499.60 Distance to CBD Mean 10.5873 3.7984 5.7077 STD 4.9762 1.9523 3.4198 Distance to Major Activity centers Mean 5.1696 1.3385 1.5121 STD 3.7050 0.7809 1.6474 Surrounding Entropy Mean 0.2783 0.4561 0.4249 STD 0.1626 0.1129 0.1026 Surrounding Density Mean 5.4740 10.8768 16.7166 STD 6.4089 9.1630 12.0403 Table 7 9 Collective m easurements for Pinellas County Measurement Statistic AR ARDT AHI Acres Mean 49738.68 20249.87 Distance to CBD Mean 5.3412 3.9689 3.4538 STD 2.8802 2.0102 2.9329 Distance to Major Activity centers Mean 1.1812 1.0090 1.0432 STD 0.7273 0.6016 0.9143 Surrounding Entropy Mean 0.2079 0.2156 0.2148 STD 0.0308 0.0236 0.0155 Surrounding Density Mean 2.4551 2.8689 3.2606 STD 1.7088 1.7856 1.9697
202 Table 7 10. Duv al scenario t able C onflict Mask1 Mask2 Mask3 Mask4 Mask5 Region 3333 ACT_1_TENTH <= 16 <= 30 >= 3 = 1 = 1 >= 1 3333 <= 32 <= 60 >= 3 = 1 = 1 >= 1 3333 <= 48 <= 90 >= 3 = 1 = 1 >= 1 3333 <= 64 <= 120 >= 3 = 1 = 1 >= 1 3332;3323;2332; 2323;2322;22 32;2223 <= 16 <= 30 >= 3 = 1 = 1 >= 1 3332;3323;2332; 2323;2322;22 32;2223 <= 32 <= 60 >= 3 = 1 = 1 >= 1 3332;3323;2332; 2323;2322;22 32;2223 <= 48 <= 90 >= 3 = 1 = 1 >= 1 3332;3323;2332; 2323;2322;2232; 2223 <= 64 <= 120 >= 3 = 1 = 1 >= 1 3322;3232;3223; 3222 <= 16 <= 30 >= 3 = 1 = 1 >= 1 3322;3232;3223; 3222 <= 32 <= 60 >= 3 = 1 = 1 >= 1 3322;3232;3223; 3222 <= 48 <= 90 >= 3 = 1 = 1 >= 1 3322;3232;3223; 3222 <= 64 <= 120 >= 3 = 1 = 1 >= 1 2222 <= 16 <= 30 >= 3 = 1 = 1 >= 1 2222 <= 32 <= 60 >= 3 = 1 = 1 >= 1 2222 <= 48 <= 90 >= 3 = 1 = 1 >= 1 2222 <= 64 <= 120 >= 3 = 1 = 1 >= 1
203 Table 7 11 Orange scenario t able Conflict Mask1 Mask2 Mask3 Mask4 Mask5 Region 3323 "ACT_1_TENTH <= 16 "CBD_1_TENTH" <= 30 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 50 3323 "ACT_1_TENTH <= 32 "CBD_1_TENTH" <= 60 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 40 3323 "ACT_1_TENTH <= 48 "CBD_1_TENTH" <= 90 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 30 3323 "ACT_1_TENTH <= 64 "CBD_1_TENTH" <= 120 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 20 3322;3223;3222 "ACT_1_TENTH <= 16 "CBD_1_TENTH" <= 30 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 50 3322;3223;3222 "ACT_1_TENTH <= 32 "CBD_1_TENTH" <= 60 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 40 3322;3223;3222 "ACT_1_TENTH <= 48 "CBD_1_TENTH" <= 90 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 30 3322;3223;3222 "ACT_1_TENTH <= 64 "CBD_1_TENTH" <= 120 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 20 2323;2322;2223 "ACT_1_TENTH <= 16 "CBD_1_TENTH" <= 30 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 50 2323;2322;2223 "ACT_1_TENTH <= 32 "CBD_1_TENTH" <= 60 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 40 2323;2322;2223 "ACT_1_TENTH <= 48 "CBD_1_TENTH" <= 90 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 30 2323;2322;2223 "ACT_1_TENTH <= 64 "CBD_1_TENTH" <= 120 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 20 2222 "ACT_1_TENTH <= 16 "CBD_1_TENTH" <= 30 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 50 2222 "ACT_1_TENTH <= 32 "CBD_1_TENTH" <= 60 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 40 2222 "ACT_1_TENTH <= 48 "CBD_1_TENTH" <= 90 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 30 2222 "ACT_1_TENTH <= 64 "CBD_1_TENTH" <= 120 "UDENSITY" >= 3 "ORANGE_VACANT = 1 "ORANGE_VACANT = 1 "WALK_BIK" >= 20
204 Table 7 12 Pinellas scenario t able C onflict Mask1 Mask2 Mask3 Mask4 Mask5 Region 3323 "ACT_1_TENTH" <= 16 "CBD_1_TENTH <= 30 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3323 "ACT_1_TENTH" <= 32 "CBD_1_TENTH <= 60 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3323 "ACT_1_TENTH" <= 48 "CBD_1_TENTH <= 90 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3323 "ACT_1_TENTH" <= 64 "CBD_1_TENTH <= 120 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2323;2322;2223 "ACT_1_TENTH" <= 16 "CBD_1_TENTH <= 30 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2323;2322;2223 "ACT_1_TENTH" <= 32 "CBD_1_TENTH <= 60 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2323;2322;2223 "ACT_1_TENTH" <= 48 "CBD_1_TENTH <= 90 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2323;2322;2223 "ACT_1_TENTH" <= 64 "CBD_1_TENTH <= 120 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3322;3223;3222 "ACT_1_TENTH" <= 16 "CBD_1_TENTH <= 30 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3322;3223;3222 "ACT_1_TENTH" <= 32 "CBD_1_TENTH <= 60 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3322;3223;3222 "ACT_1_TENTH" <= 48 "CBD_1_TENTH <= 90 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 3322;3223;3222 "ACT_1_TENTH" <= 64 "CBD_1_TENTH <= 120 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2222 "ACT_1_TENTH" <= 16 "CBD_1_TENTH <= 30 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2222 "ACT_1_TENTH" <= 32 "CBD_1_TENTH <= 60 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2222 "ACT_1_TENTH" <= 48 "CBD_1_TENTH <= 90 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1 2222 "ACT_1_TENTH" <= 64 "CBD_1_TENTH <= 120 "UDENSITY" >= 3 "PINELLAS_VACANT1 = 1 "PINELLAS_VACANT1 = 1 "CONREGION >= 1
205 Table 7 13. Freque ntly used land uses that are suitable for affordable housing Duval Duval Orange Orange Pinellas Pinellas Descript ion Ranked Suitable Ranked Suitable Ranked Suitable Single Family 3287.74 7748.14 640.63 1529.04 1340.63 2332.87 Vacant Comm ercial 666.76 449.70 308.40 181.69 527.95 582.06 Other Municipal 2.57 56.58 379.07 148.94 417.70 466.66 Condominium 25.21 61.77 478.37 44.09 370.84 391.52 Golf Courses 32.62 7.77 14.66 0.15 331.59 336.60 Vacant Resident ial 737.99 425.47 173.50 44.51 309.99 341.03 Public Schools 416.61 293.88 188.78 223.11 254.57 354.23 Vacant Indust rial. 335.42 39.53 51.04 11.37 213.30 220.54 Churches 476.72 580.40 152.94 175.41 207.12 322.17 Parking Lots 178.87 264.48 89.67 92.91 182.45 190.04 Utilities 133.39 43.68 16.21 6.80 175.66 196.22 Automotive Rep air 334.13 366.54 37.84 104.99 170.51 261.42 Parcels With No Value 910.46 375.02 29.41 10.95 156.77 170.49 Mortuaries 104.10 16.29 32.90 3.98 153.36 159.66 Other Counties 727.31 365.92 156.10 81.51 152.52 160.78 Light Manufact ure 218.07 165.50 27.11 36.21 152.50 269.58 Multi Family 149.41 413.83 46.07 100.22 138.93 217.70 Warehouses 705.39 971.26 166.73 305.78 134.48 184.70 Mobile Homes 136.38 122.70 4.79 2.84 118.31 124.82 One Story 189.72 865.91 60.33 362.04 103.51 132.38 Multi Family 338.11 1217.63 104.30 744.19 85.84 153.61 Outdoor Rec reation 6.25 0.03 0.15 0.17 77.78 80.77 Sewage Disposal 203.53 57.74 6.55 0.84 75.56 110.60 Forest and Park 314.68 91.48 10.48 0.47 66.96 69.13 Stores One Story 83.81 322.83 51.33 399.60 59.99 99.41 Acreage Not Zon ed 31.09 16.37 53.21 30.13 49.75 54.40 Vacant Institut ional 250.77 86.94 15.57 4.10 38.68 42.02 Community Shopping 226.99 1260.61 61.52 425.58 38.43 83.98 Clubs and Lodges 52.92 56.98 14.56 3.81 38.38 51.88 Heavy Manufact ure 24.35 12.91 17.80 18.69 34.95 80.50 Rights Of Way 29.49 21.34 1.78 3.02 30.87 32.30 Rivers and Lakes 4.60 1.66 7.56 0.12 29.24 29.78 Improved Agr iculture 0.00 0.00 0.00 0.00 25.04 30.97 Supermarket 33.76 206.14 2.77 21.11 23.11 64.95 Mineral Process ing 24.15 2.50 3.19 0.00 22.84 26.92 Restaurants 26.05 173.17 1.26 35.99 20.71 31.39 Homes For Aged 17.55 78.92 1.51 20.54 19.15 34.18 Industrial Storage 188.21 68.19 16.04 22.02 19.03 20.71
206 Table 7 14 Not frequently used land uses that are suitable for affordable housing Duval Duval Orange Orange Pinellas Pinellas Descript ion Ranked Suitable Ranked Suitable Ranked Suitable Professional Ser vice 59.66 259.01 5.29 89.52 15.79 43.99 Financial Intuit ional 7.96 156.13 1.61 47.58 12.93 28.40 Other State 62.61 178.49 38.90 25.06 11.44 14.85 Private Schools 101.06 142.83 2.97 50.00 10.16 25.83 Other Food Pro duction 20.17 9.00 3.36 3.21 9.89 10.41 Boarding Homes 2.13 0.37 0.00 0.00 9.12 10.01 Orphanages 57.34 37.29 7.02 19.25 7.76 8.82 Sanitariums 0.00 0.00 0.00 0.00 5.44 12.51 Hotels and Motels 23.33 204.33 5.29 141.62 5.14 7.98 Florist and Gree n houses 3.09 2.97 0.00 0.00 4.84 7.74 Private Hospital 17.40 148.57 0.00 0.00 4.60 5.24 Bowling Alleys 14.21 8.18 5.09 10.33 4.37 18.14 Multi Story 13.37 540.07 1.68 97.48 3.93 23.06 Drive In Res t. 8.77 153.81 3.98 94.07 3.16 7.74 Mixed Use 85.49 82.35 24.74 63.22 2.25 3.56 Lumber Yards 11.86 2.50 0.00 1.75 2.00 2.00 Night Clubs and Bar s 14.80 23.19 1.41 7.61 1.78 2.25 Cultural Org anization 1.58 3.71 0.00 1.04 1.36 7.37 Service Stations 13.27 37.32 0.00 1.04 1.14 1.26 Centrally Assess ed 0.00 0.00 4.18 1.31 1.11 1.48 Canneries 44.07 16.52 0.02 0.47 0.82 0.82 Cooperatives 0.17 2.10 0.10 0.00 0.77 0.96 Airports and Marina 4.37 2.99 0.00 0.00 0.74 0.74 Insurance Com panies 0.00 0.00 0.00 0.00 0.59 0.91 Repair Service s 3.81 10.88 19.55 88.78 0.59 0.67 Other Federal 14.76 106.21 0.00 7.39 0.57 42.39 Regional Shopping 0.25 94.19 0.67 126.42 0.54 0.94 Fruit and Vegetable 0.00 0.00 0.37 28.50 0.44 0.44 Wholesale 0.00 0.00 0.00 0.15 0.22 0.22 Colleges 56.67 249.42 57.71 6.97 0.10 0.10 Cropland Soil 18.04 0.00 3.06 0.30 0.00 0.00 Dairies 0.00 0.00 0.00 3.61 0.00 0.00 Department Sto res 26.25 411.10 0.02 80.50 0.00 0.00 Drive In Theater 0.20 6.15 0.00 0.00 0.00 0.00 Enclosed Theat er 0.89 24.25 0.00 8.85 0.00 0.00 Government Owned Lease 46.02 19.07 93.45 67.80 0.00 0.00 Grazing Land 21.87 9.69 24.57 11.00 0.00 0.00 Military 4.87 149.80 4.87 3.48 0.00 0.00 Orchard and Groves 0.00 0.00 5.93 18.12 0.00 0.00
207 Figure 7 1 Suitability based on AHS g oal 1 Figure 7 2 Preference based on AHS g oal 1
208 Figure 7 3 Rent monthly estimation for Orange County Figure 7 4 Rent preference surfac e for Orange County
209 Figure 7 5 Travel cost monthly estimation for Duval County Figure 7 6 Travel cost preference surface for Duval County
210 Figure 7 7 Transit access suitability surface for Pinellas County
211 Figure 7 8 Transit access preference surface for Pinel las County
212 Figure 7 9 ARDT opportunity s urface for Orange County
213 Figure 7 10 ARDT opportunity surface for Duval County
214 Figure 7 11 ARDT opportunity s urface for Pinellas County
215 Figure 7 12 R efined ARDT opportunity surface for Orange County
216 Figure 7 13 Refined ARDT opportunity surface for Duval County
217 Figure 7 14. Refined ARDT opportunity surface for Pinellas County
218 Figure 7 15 Refined AR opportunity surface for Duval County
219 Figure 7 16 Refined AR opportunity surface for Orange County
220 Figure 7 17 Refined AR Opportunity Surface for Pinellas County
221 Figure 7 18. Refined ARD Opportunity Surface for Duval County Figure 7 19. Refined ART Opportunity Surface for Duval County
222 Figure 7 20 Example combine grid for Orange County Figure 7 21 Example underutilized grid f or Duval County
223 Figure 7 22 Example walkabilty bikabilty grid f or Orange County Figure 7 23 Example of vacant parcel grid for Duval County
224 Figure 7 24 Example land value grid for Pinellas County Figure 7 25 Example d istance to CBD for Orange County
225 Figure 7 26 Example distance to major activity centers for Pinellas County Figure 7 27 Qualified Census tracts grid For Duval County.
226 Figure 7 28 Transportation analysis zones for Orange County
227 Figure 7 29 Affordable housing allocation model
228 Figure 7 30 Duval County a llocation
229 Figure 7 31 Orange County a llocation
230 Figure 7 32 Pinellas County a llocation
231 CHAPTER 8 CONCLUSION, RECOMMEN DATIONS AND LIMITATI ONS The research developed a hierarchical automated suitability for the allocation of affordable housing that takes into account transportation variables such as accessibility and travel cost. During the pr ocess of developing the model, the research answered the following research questions: 1) What are the feasible methods to create and include accessibility and travel cost as suitability surfaces in an affordable housing suitability model; 2) What is the impact of travel cost and transit accessibility on the allocation and preservation of affordable housing sites; and 3) How to incorporate multi modal transportation systems sprawl conceptualization metrics in allocating land for affordable housing? Th e following sections will discuss the extent to which the research answered recommendations, for future research. Accessibility and Travel Cost as Suitability Surfaces in an Affordabl e Housing Suitability Model The research established methodologies for creating accessibility and travel cost surfaces to be incorporated as suitability components in a hierarchical suitability model for the allocation of affordable housing. In terms of ac cessibility the research investigated different methods that in the literature are used to estimate accessibility (Handy, 2004; Levinson & Krizek, 2008; Bhat et al 2002). However, none of the mentioned methods in the literature, other than the simple pro ximity metric, had been used in a suitability model. This research introduced these accessibility metrics to be used as suitability surfaces that represent accessibility. Chapter 3 established the
232 methodology with which these suitability surfaces were gene rated and Chapter 5 established automation tools that were used to create these suitability surfaces. The research discussed the methods and formulas to be used in estimating accessibility. The choice of suitable accessibility estimation depends on many va riables. For example, the best estimation method might be to use a gravity accessibility measurement at a parcel level. However, such an approach needs the creation of an origin destination matrix that would contain billions of records, making it an unreal istic method considering the capabilities of a personal computer. The research compared other methods that can be operated on a personal computer. The comparison served to explain the merits and limitations of each method and allow the planner to choose a suitable model to estimate accessibility in terms of accuracy, limitations and model running time. In terms of travel cost, the research established statistical models to generate travel cost from travel survey data. Suitability models are deterministic models. Therefore, the combination of different utilities that compose a multiple utility assignment is usually done by a weighting process. This research generated some of these weights using statistical models showing that suitability can be generated by regression. However, the only deterministic steps in generating the travel cost were performed by transforming trip miles to cost and in reclassifying the final raster according to preference thresholds that were informed by other research as shown in the literature review (HUD, 2011 a ; CNT, 2011 ). Chapter 4 shows the statistical models that were used to generate the travel cost suitability surfaces.
233 The research showed that travel cost can be generated by spatial interpolation. However, this method only w orks to generate the present travel co st. Future visioning of travel cost or predicting travel miles requires travel cost to be related to urban form characteristics, such as the 5Ds (Ewing & Cervero, 2001). These land use and urban form characteristics ch ange in the future and lead to a change in travel cost. For that, two statistical models were estimated and compared in Chapter 4 These models were the OLS and GWR models. The comparison concluded that the GWR model had the least residual errors and a hi gher goodness of fit than the OLS model. The comparison of the two statistical models also showed that the residuals for the OLS model were spatially clustered. The model, however, is still useful because the residuals are small. The OLS model gives cons tant parameters that can help us to understand global variables, for example, how increasing density reduces travel cost. However, the GWR model shows that this is not always the case and increasing density in some areas may not decrease travel cost becaus e of local anomalies that changed relationship. There are also limitations of using GWR such as the minimum number of points and distance variables that could lead to the f ailure of GWR estimations. The travel cost generated by any of the three mentioned methods can be reclassified and transformed to suitability and preference surfaces and be included in the conflict/ opportunity surface. However, it should be noted that th e travel cost surface does not take personal characteristics of the driver into account. Household characteristics other than household size and income by Census block were not included in this research and will be left for future research. Additionally, t he output
234 travel cost was generated on a cell level scale, an area of approximately a quarter of an acre. For future research however, it is recommended that a travel cost value be assigned according to the average cost within a surrounding neighborhood co mprised of the cells within a walking distance. The cell at the center of each neighborhood would be assigned a value for the average travel cost for its surrounding neighborhood. The Impact of Travel Cost and Transit Accessibility on the Allocation and Preservation of Affordable Housing Sites The research compared the opportunity surface that contained transit accessibility and travel cost to an opportunity surface without transit accessibility or travel cost. Chapter 7 showed clearly that in the affordable housing opportunity surfaces, travel cost and transit accessibility reduced the mean distance to activities, reduced the mean distance to CBD, increased the surrounding density and increased the land use mix. Chapter 7 also compared these values to those generated for properties in the Assisted Housing Inventory (AHI) and showed that the results of the ARDT surface were more comparable to the AHI than the AR surface. The results also showed that the ARDT surface genera lly had lower distances to the CBD and activity centers while the AHI were generally in slightly denser places. In summary, it can be concluded that using transit accessibility and travel cost as suitability components leads to a more compact development pattern and that sprawl is reduced in the allocation of affordable housing development. Furthermore, the comparison of the AHI with the ARDT surface is useful in efforts aimed at the preservation of existing affordable housing, where more analysis could b e performed on the AHI properties that included categorizing them according to the year built and the type of assistance they are receiving. Categorizing AHI by year can help the
235 preservation efforts according the age of the buildings. However, the year ca n be also important to study the change in affordable housing allocation policies especially if the location and proximity to services had been considered in funding these sites. In terms of affordable housing program types, there are many funding categori es, such as public housing, assisted housing and housing vouchers. There are also programs that support the production of affordable housing such as capital financing, rental subsidies and tax credits (Ray et al 2009). The potential outcome from categori zing the AHI units according to funding or program type and comparing them to the affordable housing suitability sites is evaluate the efficiency of the program and preservation effort feasibility. Using AHI in preserving affordable housing units and in ev aluating affordable housing programs will be investigated in the future research. Creating the opportunity surface was performed by an automated sequential process facilitated by the suitability tools introduced in Chapter 6 These automation tools include d raster reclassification, raster weighting and the analytical hierarchy tools. The research showed how programming languages can be em ployed to automate the land use suitability models. In earlier LUCIS models, the user had to do some of the analysis outs ide of the GIS environment. The tools introduced by this research made the process automatic and can be done entirely in the GIS environment. The appendices of this dissertation show the source code of these tools. The tools that were generated in this res earch were also employed by LUCIS to produce a new version of
236 on raster suitability mod els. Creating tools that work in a vector GIS environment is outside the scope of this research and is left for future research. The research also measured the collective impact of transit accessibility and travel cost on the opportunity for affordable hou sing. Measuring the impact of each of the variables independently requires further sensitivity analysis that is not covered by this research and will be addressed in future research. The application of the models was also performed on three urban counties Despite the differences between these counties, all of them have urban cores of a large city and each has significant transit routes. The application of the model on rural areas may present some challenges especially in transportation modeling. Rural are as may not have established broad transportation network and transit accessibility. Therefore, some of the affordable housing transportation models and methodologies should be change to apply in rural counties. Applying the models on rural counties will be performed in future research. Incorporating Multi modal Transportation and Sprawl Conceptualization Metrics in Allocating Land for Affordable Housing? A further step in the allocation of affordable housing was the ranking of the affordable housing sites using extra conditions and restraints. The Allocation by Table tool was used in Chapter 7 The table represented a scenario that contained the extra conditions for the ranking procedure. The ranks started at 1 and ended at 16. All the locations that had an opportunity for affordable housing according to the opportunity surface and were not ranked were assigned the values of the four digit opportunity number. These lands did not satisfy the conditions set in the allocation table. However, they might have bee n useful for affordable housing under a different scenario. The
237 ranking process took into account the continuity of the affordable housing allocation in the assigned opportunity sites and the areas that were not ranked were only the areas that did not matc h the scenario. This research suggested a compact development scenario that also had a certain degree of underutilization in the land use. Displaying scenarios other than the compact development scenario will be performed in future research. The allocation procedure, which included the LUCIS suitability structure to create an opportunity surface and the use of a scenario table to rank the land for affordable housing, incorporated a multimodal transportation system in the process. The multi mod al modal transportation system included walking, biking, transit and driving. The suitability structure had the neighborhood access explained in Chapter 5 The suitability models suggested that access promotes more walking and biking and reduces the use of driving. This effect is explained in the literature review of this research and also explained in the research covered by Handy (2004) The use of a network based accessibility estimation, however is more effective than the use of Euclidean distance based access measurements but has hardware and software limitations (Arafat et al 2008). This research recommended the use of a combine opportunity distance metric for accessibility estimation. The use of a network based accessibility estimation needs very in tensive computation that is time consuming and was only used within the scope of th is research for testing purpose Applying the network based accessibility models in the affordable housing suitability model will be
238 dependent on whether the feasibility of their use can be increased by the use of more powerful computers or by development of software packages that can accelerate the process. Another outcome of incorporating a multi modal transportation system into the allocation process was achieved by usi ng the travel cost and transit accessibility in the opportunity surface. The transit accessibility surface was generated using the methodology explained in Chapter 3 while Chapter 5 compared the methods of generating the travel cost surface. The ARDT surf ace explained in Chapter 7 included transit accessibility and travel cost as preference surfaces in the final affordable housing opportunity, based on very low income population thresholds. The preference was set to discourage driving and increase the use of transit as a mode of transportation. This does not mean that people under the VLI income limit would not use a car but the model promotes affordable housing opportunities in areas that have high transit access and lower travel cost. The traveler may fin ally decide that the use of cars is more convenient. This selection of travel mode may depend on the surrounding households and the characteristics of the traveler. This research did not focus on self selection, which is an important factor to be investiga ted in future research. The models do, however, incorporate travel cost and transit accessibility and as such they promote the use of transit and the reduction of driving travel miles. The use of multi modal transportation continued at the allocation level by the use of a walking biking suitability surface in Orange County. The scenario for compact development in the three counties promoted a greater use of walking, biking and transit and lower driving miles. The scenario tables may also include more refine d conditions
239 by using the transit accessibility and travel cost suitability surfaces in addition to the preference surfaces included in the opportunity grid. This allows the use of a 1 9 suitability range instead of a 1 3 preference range. This also allows more detailed allocation in the scenario. The suitability structure is a deterministic procedure. The miles of travel or the trip length are generated by robust statistics. Some deterministic steps are used to generate the travel cost out of the trip mil es and to transfer the travel cost into a preference surface. Using a stochastic location choice model is outside the scope of this research. However, this research recommends the use of robust statistics in generating the suitability surface as explained in Chapter 5 Introducing Parcel Level Analysis for Affordable Housing Sites The level of analysis was emphasized in Chapter 4 which also utilized the choice of an areal unit. The literature review explained that the level of analysis is dependent on the variable that is analyzed. For accessibility parcel level analysis is recommended ( J ohnston, 2004) ; Wegner, 2005) However, many of the variables used in this research are collective measurements that are recommended to be performed on a neighborhood scale For these variables, Chapter 4 introduced the floating neighborhood, or the surrounding area characteristics, to reduce the impact of the modifiable areal unit on the metrics used in the analysis. The analysis results in Chapter 4 also recommends the Man hattan distance of 1.25 miles as the optimal surrounding distance for collective measurements such as density, connectivity and land use mix metrics. This distance generated a diamond shaped neighborhood that is used to estimate the aggregated value of the metric. However, the 1.25 mile Manhattan distance is suggested for studying the impact of urban form on trip miles and mode
240 choice, and does not change the TOD distance of 0.25 mile for buses and 0.50 mile for rail. This research recommended the use of ne twork or Manhattan sheds for such an analysis. The suitability model is bas ed on raster analysis using a 31 meter x 31 meter cell size, which is a quarter of an acre in size, and can be assumed as being equivalent to a parcel level in terms of spatial loca tion. The output of the AHS model is in raster cells however and not in parcels. Because parcel characteristics are important in the allocation process, the output allocation grids were transformed to a 10 meter x 10 meter cell size. The values of these ce lls were also summarized for each parcel and then joined back to the parcel attribute table so that additional analysis queries could be performed using the vector parcel data. This allows the planner to select suitable parcels for affordable housing and i nvestigates their specific land use characteristics (Figure 8 1). It should be noted, however, that a shift may occur in the raster analysis which is estimated at half a cell, or in this research, is shift in location of 15.5 meters. This shift may h ave a significant effect on the allocation into specific parcels. This research focused on the raster analysis and did not focus on the accuracy of the parcel level vector analysis. Testing the accuracy of such parcel allocation is beyond the scope of this research and will be left for future research.
241 Figure 8 1. Example of parcel level allocation for affordable h ousing
242 APPENDIX A THE A4 SUITABILITY T OOL SOURCE CODE import sys, string, os, arcgisscripting, shutil # Create the Geoprocessor object gp = arcgisscripting.create() gp.SetProduct("ArcView") gp.CheckOutExtension("spatial") # Work space gp.workspace = sys.argv gp.OverwriteOutput = 1 # Load required toolboxes... gp.AddToolbox("C:/Program Files/ArcGIS/ArcToolbox/Toolboxes/Analysis Tools.tbx") gp.AddToolbox("C:/Program Files/ArcGIS/ArcToolbox/Toolboxes/Data Management Tools.tbx") gp.AddToolbox("C:/Program Files/ArcGIS/ArcToolbox/Toolboxes/Spatial Analyst Tools.tbx") classTable = sys.argv infeature = sys.argv # Set the Geoproce ssing environment... inraster = sys.argv inClass = sys.argv reclass = sys.argv gp.compression = "LZ77" gp.rasterStatistics = "STATISTICS 1 1" gp.cartographicCoordinateSystem = "" gp.tileSize = "128 128" gp.pyramid = "PYRAMIDS 1 NEAREST" gp.cellSi ze = "MAXOF" distsupfund = inraster if classTable == '#': cur = gp.SearchCursor(infeature) row = cur.Next() StdSum = 0. MeanSum = 0. MinValue = row.MIN MaxValue = row.MAX i = 0 x = 1 while row:
243 StdValue = row.STD MeanValue = row.MEAN StdSum = StdSum + StdValue MeanSum = MeanSum + MeanValue if (row.MIN < MinValue): MinValue = row.MIN if (row.MAX > MaxValue): Ma xValue = row.MAX i = i + 1 row =cur.Next() StdQuart = StdSum / (i 4) MeanAvarage = MeanSum / i WorkRange1 = MinValue WorkRange2= MeanAvarage WorkRangeD1= MeanAvarage del row, cur if (inClass == "Decreasing Suitability"): # Local variables... URP = gp.workspace table2 = URP + \ \ tablef #Process: Create Table... gp.CreateTable_management(URP, "tablef1", "", "") Range1 = "FROM1" # Process: Add Field... gp.AddField_management(table2, Range1, "DOUBLE", "", "2", "8", "FROM1", "NULLABLE", "NON_REQUIRED", "") gp.AddField_management(table2, "TO", "DOUBLE", "", "2", "8", "", "NULLABLE", "NON_REQUIRED", "") gp.AddField_management(table2, OUT", "LONG", "9", "", "9", "", "NULLABLE", "NON_REQUIRED", "") gp.AddField_management(table2, "MAPPING", "TEXT", "", "", "15", "", "NULLABLE", "NON_REQUIRED", "") gp.deletefield_management(table2, "field1") gp.deletefield_management(table2, "OBJECTID") # check the number of intervals RangeValue = MaxValue MeanAvarage intNum = RangeValue / StdQuart
244 if (intNum < 8): StdQuart = RangeValue / 8 #check f irst Value RangeMinOne = StdQuart 7 checkCell = RangeValue RangeMinOne if (checkCell < 30): StdQuart = RangeValue / 8 rows = gp.InsertCursor(table2) i = 0 while i < 9: Rank = 9 i if (i == 0): FromValue = 0. ToValue = WorkRange2 else: FromValue = WorkRange1 ToValue = WorkRange2 if (i == 8): ToValue = 10000000 row = rows.NewRow() row.FROM1 = FromValue row.TO = ToValue row.OUT = Rank row.MAPPING = "ValueToValue" rows.InsertRow(row) WorkRange1 = WorkRange2 WorkRange2 = WorkRange1 + StdQuart i = i + 1 del rows, row # Local variables... tablef = table2 # Process: Reclass by Table... gp.ReclassByTable_sa(distsupfund, tablef, "FROM1", "TO", "OUT", r eclass, "NODATA") elif (inClass == "Increasing Suitability"): # Local variables... URP = gp.workspace table3 = URP + \ \ tableb1"
245 #Process: Create Table... gp.CreateTable_management(URP, "tableb1", "", "") Range1 = "FROM1" # Process: Add Field... gp.AddField_management(table3, Range1, "DOUBLE", "", "2", "8", "FROM1", "NULLABLE", "NON_REQUIRED", "") gp.AddField_management(table3, "TO", "DOUBLE", "", "2", "8", "", "NULLABLE", "NON_REQU IRED", "") gp.AddField_management(table3, "OUT", "LONG", "9", "", "9", "", "NULLABLE", "NON_REQUIRED", "") gp.AddField_management(table3, "MAPPING", "TEXT", "", "", "15", "", "NULLABLE", "NON_REQUIRED", "") gp.deletefield_management (table3, "field1") gp.deletefield_management(table3, "OBJECTID") # check the number of intervals RangeValue = MeanAvarage MinValue intNum = RangeValue / StdQuart if (intNum < 8): StdQuart = RangeValue / 8 #check first Value RangeMinOne = StdQuart 7 checkCell = RangeValue RangeMinOne if (checkCell < 30): StdQuart = RangeValue / 8 i = 1 while i < 9: WorkRangeD2 = WorkRangeD1 StdQuart FromValue = MinValue if (WorkRangeD1 < MinValue): ToValue = WorkRangeD2 j = i + 1 if (i == 8): ToValue = WorkRangeD2 j = 9 WorkRangeD1 = WorkRangeD2 i = i + 1 rows = gp.InsertCursor(table3)
246 WorkRangeD2 = ToValue + StdQuart WorkRangeD1 = FromValue i = 0 while i < j: Rank = i + 1 if (i == 0): FromValue = 0. ToValue = WorkRangeD2 else: FromValue = WorkRangeD1 ToValue = WorkRangeD2 if (i == 8): ToValue = 10000000 row = rows.NewRow() row.FROM1 = FromValue row.TO = ToValue row.OUT = Rank row.MAPPING = "ValueToValue" rows.InsertRow(row) WorkRangeD2 = WorkRangeD2 + StdQuart WorkRangeD1 = WorkRangeD2 StdQuart i = i + 1 del rows, row # Local variables... tableb = table3 # Process: Reclass by Table... gp.ReclassByTable_sa(distsupfund, tableb, "FROM1", "TO", "OUT", reclass, "N ODATA") else: gp.ReclassByTable_sa(distsupfund, classTable, "FROM1", "TO", "OUT", reclass, "NODATA")
247 APPENDIX B : WEIGHTING TOOLS The Community Value Calculator Source Code Option Explicit Dim GetIndex As Integer Dim intCounter1 As Integer Dim intCounter2 As Integer Dim RankArr(25, 25) As Integer Dim AhpArr(25, 25) As Double Dim columnSumArr(25) As Double Dim normArr(25, 25) As Double Dim rowSumArr(25) As Double Dim wtArr(25) Dim counter As Integer Private Sub cboChooseLayers_Change() End Sub Pr ivate Sub cboChooseLayers_Click() populateLstBox End Sub Private Sub CommandButton1_Click() End Sub Private Sub cmdAdd_Click() Dim pGxDialog As IGxDialog Set pGxDialog = New GxDialog pGxDialog.Title = "Add Layers" pGxDialog.ButtonCaption = "Add" pGxDialog.AllowMultiSelect = False pGxDialog.StartingLocation = "c: \ Dim pGxFilter As IGxObjectFilter Set pGxFilter = New GxFilterRasterDatasets Set pGxDialog.ObjectFilter = pGxFilter Dim pLayerFiles As IEnumGxObject pGxDialog.DoModalOpen 0, pLayerFiles Dim pLayerFile As IGxObject Set pLayerFile = pLayerFiles.Next
248 MsgBox pLayerFile.Name If pLayerFile Is Nothing Then Exit Sub End If MsgBox pLayerFile.Name Dim pGxLayer As IGxObject Set pGxLayer = pLayerFile MsgBox pGxLayer.FullName cboChooseLayers.AddItem (pLayerFile.Name) LstLayerBox.AddItem (pLayerFile.Name) End Sub Private Sub AddShapeFile(folder As String, ShapeName As String) Dim pWorkspaceFactory As IWorkspaceFactory Dim pFeatureWorkspace As IFeatureWorkspace Dim pFeatureLayer As IFeatureLayer Dim pFeatureClass As IFeatureClass Dim pMxDoc As IMxDocument Dim pMap As IMap Dim pLayer As ILayer Dim pEnumLayer As IEnumLayer Set pMxDoc = Appli cation.Document Set pMap = pMxDoc.FocusMap 'Create a new ShapefileWorkspaceFactory object and open a shapefile folder Set pWorkspaceFactory = New ShapefileWorkspaceFactory Set pFeatureWorkspace = pWorkspaceFactory.OpenFromFile(folder, 0) 'Create a new FeatureLayer and assign a shapefile to it Set pFeatureLayer = New FeatureLayer Set pFeatureLayer.FeatureClass = pFeatureWorkspace.OpenFeatureClass(ShapeName) pFeatureLayer.Name = pFeatureLayer.FeatureClass.AliasName pMap.A ddLayer pFeatureLayer pMxDoc.ActiveView.Refresh pMxDoc.UpdateContents End Sub Private Sub CommandButton2_Click() tableBuild End Sub Private Sub CommandButton3_Click()
249 Dim pGxDialog As IGxDialog Set pGxDialog = New GxDialog pGxDialog.Title = "Load Table" pGxDialog.ButtonCaption = "Add" pGxDialog.AllowMultiSelect = False pGxDialog.StartingLocation = txtWorkSpace.Text Dim pGxFilter As IGxObjectFilter Set pGxFilter = New GxFilterTables Set pGxDialog.ObjectFilter = pGxF ilter Dim pLayerFiles As IEnumGxObject pGxDialog.DoModalOpen 0, pLayerFiles Dim pTableFile As IGxObject Set pTableFile = pLayerFiles.Next MsgBox pTableFile.Name If pTableFile Is Nothing Then Exit Sub End If MsgBox pTableFile.Name txtFileName.Text = pTableFile.Name tableRet End Sub Private Sub CommandButton5_Click() Dim icounter As Integer Dim pOIDField Dim pWorkspaceFactory As IWorkspaceFactory Dim pFeatureWorkspace As IFeatureWorkspace Dim workSpace As String workSpace = txtWorkSpace.Text Dim tableName As String If Trim(TextBox5.Text) = "" Then tableName = txtFileName.Text Else tableName = TextBox5.Text End If Set pWorkspaceFactory = New Shapefi leWorkspaceFactory Set pFeatureWorkspace = pWorkspaceFactory.OpenFromFile(workSpace, 0)
250 Dim pTable As ITable Set pTable = pFeatureWorkspace.OpenTable(tableName) Dim pQueryFilter As IQueryFilter Set pQueryFilter = New QueryFilter Dim pCursor As ICursor Set pCursor = pTable.Update(pQueryFilter, False) 'we need to get the featureClass of the layer Dim iCount As Integer Dim pRow As IRow Set pRow = pCursor.NextRow iCount = 0 Do While Not pRow Is Nothing pRow.Value(2) = wtArr(iCount) pRow.Store iCount = iCount + 1 Set pRow = pCursor.NextRow Loop End Sub Private Sub Frame2_Click() End Sub Private Sub Frame3_Click() End Sub Private Sub Frame4_Click() End Sub Private Sub Label1_Click() End Su b Private Sub Label4_Click() End Sub Private Sub ListBox2_Click() End Sub Private Sub ListBox2_DblClick(ByVal Cancel As MSForms.ReturnBoolean) Dim newIndex As Integer newIndex = ListBox2.ListIndex Dim I As Integer Dim Num As Integer Dim sum As Double
251 Num = ListBox2.ListCount 1 For I = 0 To Num wtArr(I) = ListBox2.List(I) Next I wtArr(newIndex) = InputBox("Enter New Weight", "New Weight") ListBox2.Clear sum = 0 For I = 0 To Num ListBox2.AddItem wtArr(I) sum = sum + wtArr(I) Next I TextBox2.Text = sum End S ub Private Sub LstBoxRank_Change() End Sub Private Sub LstBoxRank_Click() End Sub Private Sub LstBoxRank_DblClick(ByVal Cancel As MSForms.ReturnBoolean) Dim strLabel3 As String Dim strLabel2 As String Dim PeopleLim As Integer counter = coun ter + 1 RankArr(intCounter1, intCounter2) = Val(LstBoxRank.Text) If LstBoxRank.ListIndex < 9 Then AhpArr(intCounter1, intCounter2) = AhpArr(intCounter1, intCounter2) + RankArr(intCounter1, intCounter2) Else AhpArr(intCounter1, intCounter2) = AhpArr(intCounter1, intCounter2) + 1 / RankArr(intCounter1, intCounter2) End If MsgBox "Your total Community Value = & AhpArr(intCounter1, intCounter2) & person count = & counter PeopleLim = Tex tBox6.Value If counter < PeopleLim Then Exit Sub Else AhpArr(intCounter1, intCounter2) = AhpArr(intCounter1, intCounter2) / counter
252 MsgBox "You Final avarage Community Value = & AhpArr(intCounter1, intCounter2) & for & counter & persons counter = 0 End If If intCounter2 = LstLayerBox.ListCount 1 Then Exit Sub End If If intCounter2 < LstLayerBox.ListCount 1 Then intCounter2 = intCounter2 + 1 strLabel2 = LstLayerBox.List(intCounter2) Label2 .Caption = strLabel2 End If End Sub Private Sub LstFrame_Click() End Sub Public Sub LstLayerBox_Change() Dim pMxDocument As IMxDocument Set pMxDocument = ThisDocument Dim pMap As IMap Set pMap = pMxDocument.FocusMap intCounter1 = LstLayerBox.ListIndex intCounter2 = intCounter1 + 1 Dim strLabel3 As String Dim strLabel2 As String Dim counter As Integer If intCounter1 = LstLayerBox.ListCount 1 Then Exit Sub strLabel3 = LstLayerBox.List(i ntCounter1) strLabel2 = LstLayerBox.List(intCounter2) Label3.Caption = strLabel3 Label2.Caption = strLabel2 'initializeRankBox End Sub Private Sub LstLayerBox_Click() 'initializeRankBox End Sub Private Sub lstTableBox_Click() End Sub
253 P rivate Sub TextBox1_Change() tableRet End Sub Private Sub TextBox2_Change() End Sub Private Sub TextBox6_Change() End Sub Private Sub UserForm_Click() 'initializeRankBox End Sub Sub initializeRankBox() Dim itemValue As Integer Dim intCounter As Integer LstBoxRank.Clear For intCounter = 1 To 9 itemValue = 10 intCounter LstBoxRank.AddItem (itemValue) Next intCounter For intCounter = 2 To 9 itemValue = intCounter LstBoxRank.AddItem (itemValue) Next intCounter End Sub Private Sub UserForm_ Initialize() intializeLayerMenu initializeRankBox TextBox6.Text = 1 End Sub Private Sub intializeLayerMenu() initialize the list box Dim pMxDocument As IMxDocument Set pMxDocument = ThisDocument Dim pMap As IMap Set pMap = pMxDocument.FocusMap Dim pLayer As ILayer If pMap.LayerCount > 0 Then Dim intIndex As Integer
254 For intIndex = 0 To (pMap.LayerCount 1) Set pLayer = pMap.Layer(intIndex) cboChooseLayers.AddItem pLayer.Name Next intIndex End If select the first as default End Sub Private Sub populateLstBox() GetIndex = cboChooseLayers.ListIndex initialize the list box Dim pMxDocument As IMxDocument Set pMxDocument = ThisDocument define layers Dim pMap As IMap Set pMap = pMxDocument.FocusMap Dim pLayer As ILayer Set pLayer = pMap.Layer(GetIndex) LstLayerBox.AddItem pLayer.Name End Sub Private Sub calculateAhp() Dim icounter As Integer Dim jCounter As Int eger For icounter = 0 To LstLayerBox.ListCount 2 For jCounter = icounter + 1 To LstLayerBox.ListCount 1 AhpArr(jCounter, icounter) = 1 / AhpArr(icounter, jCounter) Next jCounter Next icounter Dim n As Integ er n = LstLayerBox.ListCount 1 For icounter = 0 To n AhpArr(icounter, icounter) = 1 Next icounter For jCounter = 0 To n columnSumArr(jCounter) = 0 For icounter = 0 To n columnSumArr(jCounter) = columnSumArr(jCounter) + AhpArr(icounter, jCounter) Next icounter
255 'normalize For icounter = 0 To n normArr(icounter, jCounter) = AhpArr(icounter, jCounter) / columnSumArr(jCounter) Next icounter Next jCounter find weights For icounter = 0 To n rowSumArr(icounter) = 0 For jCounter = 0 To n rowSumArr(icounter) = rowSumArr(icounter) + normArr(icounter, jCounter) Next jCounter wtArr(icounter) = rowSumArr(icounter) / (n + 1) Next icounter Dim sum As Double sum = 0 For icounter = 0 To n sum = sum + wtArr(icounter) Next icounter TextBox2.Text = sum End Sub Private Sub tableBuild() calculateAhp Dim icounter As Integer Dim pOIDField Dim pWorkspaceFactory As IWorkspaceFactory Dim pFeatureWorkspace As IFeatureWorkspace Dim workSpace As String workSpace = txtWorkSpace.Text Dim tableName As String tableName = TextBox5. Text Set pWorkspaceFactory = New ShapefileWorkspaceFactory Set pFeatureWorkspace = pWorkspaceFactory.OpenFromFile(workSpace, 0) Dim pFieldEdit_nameRas As IFieldEdit Set pFieldEdit_nameRas = New Field Dim pFieldEdit_wtRas As IFieldEdit Set pFieldEdit_wtRas = New Field Dim nameRas
256 Dim wtRas With pFieldEdit_nameRas .Type = esriFieldTypeString .Name = "Raster_Name" .Length = 30 End With With pFieldEdit_wtRas .Type = esriFieldTypeDoub le .Name = "Raster_weight" .Length = 15 End With Dim pFieldsEdit As IFieldsEdit Set pFieldsEdit = New Fields pFieldsEdit.AddField pFieldEdit_nameRas pFieldsEdit.AddField pFieldEdit_wtRas Dim pFields As IFields Set pFields = pFieldsEdit Dim pTable As ITable Set pTable = pFeatureWorkspace.CreateTable(tableName, pFields, Nothing, Nothing, "") Dim pRow As IRow Dim n As Integer n = LstLayerBox.ListCount 1 ListBox2.Clear For icounter = 0 To n Set pRow = pTable.CreateRow pRow.Value(1) = LstLayerBox.List(icounter) pRow.Value(2) = wtArr(icounter) ListBox2.AddItem wtArr(icounter) pRow.Store Next icounter Dim pStandaloneTable As IStandaloneTable Set pStandaloneTable = New Standalone Table 'assign the table from disk to the standalone table object Set pStandaloneTable.Table = pTable Dim pMxDocument As IMxDocument Set pMxDocument = ThisDocument Dim pMap As IMap
257 Set pMap = pMxDocument.FocusMap 'initialize the tables collection to point to the map Dim pStandaloneTableCollection As IStandaloneTableCollection Set pStandaloneTableCollection = pMap 'add the standalone table to the collection of tables pStandaloneTableCollection.AddStandaloneTable pStandaloneTable 'Refresh the TOC for the table to show up pMxDocument.UpdateContents End Sub Private Sub cmdWorkSpace_Click() Dim pGxDialog As IGxDialog Set pGxDialog = New GxDialog pGxDialog.Title = "Work Space" pGxDialog.ButtonCaption = "Add" pGxDialog.AllowMultiS elect = False pGxDialog.StartingLocation = "c: \ Dim pGxFilter As IGxObjectFilter Set pGxFilter = New GxFilterContainers Set pGxDialog.ObjectFilter = pGxFilter Dim pFolders As IEnumGxObject pGxDialog.DoModalOpen 0, pFolders Dim pLayerFolder As IGxObject Set pLayerFolder = pFolders.Next If pLayerFolder Is Nothing Then Exit Sub End If Dim pGxFolder As IGxObject Set pGxFolder = pLayerFolder txtWorkSpace.Text = pGxFolder.FullName End Sub Private Sub tableRet() Dim icounter As Integer Dim pOIDField Dim pWorkspaceFactory As IWorkspaceFactory Dim pFeatureWorkspace As IFeatureWorkspace Dim workSpace As String
258 workSpace = txtWorkSpace.Text Dim tableName As String tableName = txtFileName.Text Set pWorkspaceFactory = New ShapefileWorkspaceFactory Set pFeatureWorkspace = pWorkspaceFactory.OpenFromFile(workSpace, 0) Dim pTable As ITable Set pTable = pFeatureWorkspace.OpenTable(tableName) Dim pQue ryFilter As IQueryFilter Set pQueryFilter = New QueryFilter Dim pCursor As ICursor Set pCursor = pTable.Search(pQueryFilter, False) 'we need to get the featureClass of the layer Dim iCount As Integer Dim pRow As IRow Set pRow = pC ursor.NextRow iCount = 0 Do While Not pRow Is Nothing 'Set pRow = pTable.GetRow(iCount) 'create a cursor to point to the selection LstLayerBox.AddItem pRow.Value(1) Dim xvalue1 As String Dim xvalue2 As Double xvalue1 = pRow.Value(1) xvalue2 = pRow.Value(2) 'iCount = iCount + 1 'counter1 = 0 ListBox1.AddItem xvalue1 ListBox2.AddItem xvalue2 Set pRow = pCursor.NextRow Loop now find the min, and max values and mean End Sub Sub initi alizeTempBoxes() Dim pMxDoc As IMxDocument Dim pMap As IMap Dim pFeatureLayer As IFeatureLayer
259 Set pMxDoc = ThisDocument 'Application.Document Set pMap = pMxDoc.FocusMap 'let's use the States layer that is the second layer in the map 'the tes t below makes sure that the layer is a feature layer (vs image or grid) 'If Not TypeOf pMap.Layer(3) Is IFeatureLayer Then Exit Sub Set pFeatureLayer = pMap.Layer(0) 'Create the query filter Dim pQueryFilter As IQueryFilter Set pQueryFilter = New QueryFilter 'pQueryFilter.WhereClause = "STATE_ABBR = 'FL'" OR STATE_ABBR = 'GA'" 'pQueryFilter.WhereClause = "POP1990 > 5000000" 'we need to get the featureClass of the layer Dim pFeatureClass As IFeatureClass Set pFeatureClass = pFeatureLayer.FeatureClass 'create a cursor to point to the selection Dim pCursor As ICursor Set pCursor = ptable .Search(pQueryFilter, False) Dim lngCounter As Long lngCounter = 0 Dim pFeature As IFeature Set pFeature = pFeatureCursor.NextFeature Do While Not pFeature Is Nothing lngCounter = lngCounter + 1 cboMinTemp.AddItem pFeature.Value(lngFldIndex) cboMaxTemp.AddItem pFeature.Value(lngFldIndex) Set pFeature = pFeatureCursor.NextFeature Loop cboMinTemp.ListIndex = 0 cboMaxTemp.ListIndex = 0 End Sub The A4 W eighting Tool Source Code import sys, os import string, arcgisscripting # Create the Geoprocessor object gp = arcgisscripting.create() gp.SetProduct("ArcView") gp.CheckOutExtension("spatial ")
260 # Work space gp.workspace = sys.argv #gp.workspace = "X: \ \ CentralFlorida \ \ Intermediate" gp.scratchWorkspace = sys.argv gp.overwriteoutput = 1 # Load required toolboxes... gp.AddToolbox("C:/Program Files/ArcGIS/ArcToolbox/Toolboxes/Analysis Tools.tbx") gp.AddToolbox("C:/Program Files/ArcGIS/ArcToolbox/Toolboxes/Data Management Tools.tbx") gp.AddToolbox("C:/Program Files/ArcGIS/ArcToolbox/Toolboxes/Spatial Analyst Tools.tbx") ### Argument 1 is the list of tables to be converted inRasters = sys .argv print inRasters gp.compression = "LZ77" gp.rasterStatistics = "STATISTICS 1 1" gp.cartographicCoordinateSystem = "" gp.tileSize = "128 128" gp.pyramid = "PYRAMIDS 1 NEAREST" gp.cellSize = "MAXOF" ### get the parameter table wtTable = sys.argv ### The list is split by semcolons ";" inRasters = string.split(inRasters, ";") print inRasters ### The output workspace where the shapefiles are created oRas = sys.argv print oRas count = 0 for inRaster in inRasters: print inRaster count = count + 1 i = 0 paraM = "" for inRaster in inRasters: print inRaster i = i + 1 print i # To start, make sure the input exists
261 cur = gp.SearchCursor(wtTable) row = cur.Next() while not row.RASTER_NAM in inRaster: row =cur.Next() RasterName = row.RASTER_NAM WtValue = row.RASTER_WEI print RasterName, WtValue para1 = RasterName + VALUE + str(WtValue) + ";" if i == count: para1 = RasterName + VALUE + str(WtValue) row =cur.Ne xt() print para1 del row, cur paraM = paraM + para1 print paraM gp.WeightedSum_sa(paraM, oRas)
2 62 APPENDIX C ALLOCATION TOOLS Trend Allocation Tool Source Code # Import system modules import sys, string, os, arcgisscripting # Create the Geoprocessor object gp = arcgisscripting.create() # Set the necessary product code gp.SetProduct("ArcInfo") # Check out any necessary licenses gp.CheckOutExtension("spatial") gp.CheckOutExtension("3D") # Work space WorkSpace = sys.argv gp.workspace = WorkSpace # Load required toolboxes... gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Spatial Analyst Tools.tbx") gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Conversion Tools.tbx") gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Data Management Tools.tbx") gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/3D Analyst Tools.tbx") gp. overwriteoutput = 1 Table = sys.argv CalcFeild2 = sys.argv yearP = sys.argv Feild0 = sys.arg v RegionNum = sys.argv CalcFeild1 = sys.argv Mask1 = sys.argv Mask2 = sys.argv Mask3 = sys.argv Mask4 = sys.argv Mask5 = sys.argv Feild1 = sys.argv inCodes1 = sys.argv Feild2 = sys.argv
263 inCodes2 = sys.argv outTable = sys.argv inCode0 = RegionNum Table_View = Table + "_View" Express1 = \ "" + CalcFeild2 + \ "" + = 0" + AND + \ "" + CalcFeild1 + \ "" + > 0 Express2 = AND + \ "" + str(Feild0) + \ "" + " Express3 = "= + inCode0 + " Expres s = "(" + Express1 + Express2 + Express3 + ")" print Express exp = "" expR = "" if Mask1 == '#': expR = "" else: expR = expR + \ "" + Mask1 + \ "" + = 1 if Mask2 == '#': exp = exp + "" else: exp = exp + \ "" + Mask2 + \ "" + = 1 i f Mask3 == '#': exp = exp + "" else: exp = exp + "OR + \ "" + Mask3 + \ "" + = 1 if Mask4 == '#': exp = exp + "" else: exp = exp + "OR + \ "" + Mask4 + \ "" + = 1 if Mask5 == '#': exp = exp + "" else: exp = exp + "OR + \ "" + Mask5 + \ "" + = 1 exp = expR + AND + "(" + exp + ")" print exp inCodes1 = string.split(inCodes1, ";") inCodes2 = string.split(inCodes2, ";") Count = 0 try:
264 for inCode1 in inCodes1: print inCode1 exp2 = \ "" + Feild1 + \ "" + = + inCode1 for inCode2 in inCodes2: print inCode2 Count = Count + 1 print Feild2 exp3 = \ "" + Feild2 + \ "" + >= + inCode2 totExp = Express + AND + exp + AND + e xp2 + AND + exp3 print exp3 print totExp Table_View2 = Table + "_View" + str(Count) print Table_View2 gp.MakeTableView_management(Table, Table_View2, totExp, "") print progress gp.CalculateField_management(Table_View2, CalcFeild2, yearP, "VB") outValue = 1 gp.CreateConstantRaster_sa(outTable, outValue, "INTEGER", "1", "0 0 250 250") except Exception, ErrorDesc: msgStr = gp.GetMessages(2) sys.exit(1) Allocation b y Table Source Code import sys, string, os, arcgisscripting # Create the Geoprocessor object gp = arcgisscripting.create() # Set the necessary product code gp.SetProduct("ArcInfo") # Check out any necessary licenses gp.CheckOutExtension("spatial") gp.CheckOutExtension("3D") # Work space WorkSpace = sys.argv gp.workspace = WorkSpace # Load required toolboxes... gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Spatial Analyst Tools.tbx") gp.AddToolbo x("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Conversion Tools.tbx")
265 gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Data Management Tools.tbx") gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/3D Analyst Tools.tb x") gp. overwriteoutput = 1 Table = sys.argv CalcFeild2 = sys.argv yearP = sys.argv Feild0 = sys.argv CalcFeild1 = sys.argv MaskExp1 = sys.argv MaskExp2 = sys.argv MaskExp3 = sys.argv MaskExp4 = sys.argv MaskExp5 = sys.argv Feild1 = sys.argv inCodes1 = sys.argv Feild2 = sys.argv inCodes2 = sys.argv outTable = sys.argv IterNum = sys.argv PeopleLim = sys.argv PrevAlloc = sys.argv PlanTable = sys.argv print PrevAlloc print CalcFeild2 Table_Vie w = Table + "_View" Express1 = \ "" + CalcFeild2 + \ "" + = 0" + AND + \ "" + CalcFeild1 + \ "" + > 0 Express2 = AND + Feild0 Express = "(" + Express1 + Express2 + ")" print Express exp = "" expR = "" if MaskExp1 == '#': expR = expR + "" else: expR = expR + MaskExp1 if MaskExp2 == '#':
266 exp = exp + "" else: exp = exp + MaskExp2 if MaskExp3 == '#': exp = exp + "" else: exp = exp + OR + MaskExp3 if MaskExp4 == '#': exp = exp + "" else: exp = exp + OR + M askExp4 if MaskExp5 == '#': exp = exp + "" else: exp = exp + OR + MaskExp5 exp = expR + AND + "(" + exp + ")" print exp inCodes1 = string.split(inCodes1, ";") inCodes2 = string.split(inCodes2, ";") Count = 0 try: for inCode1 in inCode s1: print inCode1 exp2 = \ "" + Feild1 + \ "" + = + inCode1 for inCode2 in inCodes2: print inCode2 Count = Count + 1 print Feild2 exp3 = \ "" + Feild2 + \ "" + >= + inCode2 totExp = Express + AND + exp + AND + exp2 + AND + exp3 print exp3 print totExp Table_View2 = Table + "_View" + str(Count) print Table_View2 Phase = 'Iteration' PhaseV = int(IterNum) print PhaseV
267 gp.MakeTableView_management(Table, Table_View2, totExp, "") print "progress gp.CalculateField_management(Table_View2, Phase, PhaseV, "VB") print "progress gp.CalculateField_management(Ta ble_View2, CalcFeild2, yearP, "VB") Express10 = \ "" + CalcFeild2 + \ "" + = + yearP + AND + \ "" + "Iteration" + \ = + IterNum print Express10 sump = 0 cur = gp.SearchCursor(Table, Express10) row = cur.Next() while row: y = row.GetValue (CalcFeild1) sump = sump + y #print sump row = cur.Next() del row, cur print PrevAlloc print sump allocP = int(PrevAlloc) + int(sump) print allocP climit = int(PeopleLim) + 100 print climit AllNeed = int(climit) int(PrevAlloc) print AllNeed if climit < allocP: yearP1 = 0 Table_View2 = Table + "_ViewX" print Table_View2 gp.MakeTableView_management(Table, Table_View2, Express10, "") print "table view com pleted" gp.CalculateField_management(Table_View2, CalcFeild2, yearP1, "VB") print "pop field updated" expr = \ "" + "Iteration" + \ = + IterNum print expr cur = gp.UpdateCursor(Table, expr) print "progress row = cur.Next()
268 sump = 0 yearP = sys.argv while row: if AllNeed >= sump: row.SetValue (CalcFeild2, yearP) cur.updateRow (row) sump = sump + y #print sump row = cur.Next() del row, cur allocP = int(PrevAlloc) + int(sump) print allocP outValue = allocP print outValue cur = gp.UpdateCursor(PlanTable) print "Updating Table" row = cur.Next() FieldIter = "Iter" Popu = "Pop" Prev = "PrevPop" Iter1 = int(IterNum) + 1 print IterNum, Iter1 while row: con = row.GetValue (FieldIter) print "condition printed" print con print IterNum if con == int(IterNum): print "allocate d pop condition granted" row.SetValue (Popu, outValue) cur.updateRow (row) row = cur.Next() elif con == int(Iter1): print "previous pop condition granted" row.SetValue (Prev, outValue) cur.updateRow (row) row = cur.Next()
269 else: print "exception condition granted" row = cur.Next() del row, cur yearP1 = 0 print yearP1 Ex = \ "" + CalcFeild2 + \ "" + = 0" + AND + \ "" + "Iterati on" + \ = + IterNum #Ex = \ "" + CalcFeild2 + \ "" + = + str(yearP1) print "progress print Ex Table_View3 = Table + "_ViewZ" print Table_View3 gp.MakeTableView_management(Table, Table_View3, Ex, "") print "table view completed gp.CalculateField_management(Table_View3, "Iteration", yearP1, "VB") print "Iter field updated" print "Generating Raster" print outValue gp.CreateConstantRaster_sa(outTable, outValue, "INTEGER", "1", "0 0 250 250") except Exception, ErrorD esc: msgStr = gp.GetMessages(2) sys.exit(1) Detailed Allocation Tool Source Code # Import system modules import sys, string, os, arcgisscripting # Create the Geoprocessor object gp = arcgisscripting.create() # Set the necessary product code gp.SetProduct("ArcInfo") # Check out any necessary licenses gp.CheckOutExtension("spatial") gp.CheckOutExtension("3D") # Work space WorkSpace = sys.argv gp.workspace = WorkSpace # Necessary tools # Load required toolboxes...
270 gp.AddToolbox("C:/Prog ram Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Spatial Analyst Tools.tbx") gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Conversion Tools.tbx") gp.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/Data Management Tools.tbx") g p.AddToolbox("C:/Program Files (x86)/ArcGIS/ArcToolbox/Toolboxes/3D Analyst Tools.tbx") # overwrite output gp. overwriteoutput = 1 Table = sys.argv CalcFeild2 = sys.argv yearP = sys.argv Feild0 = sys.argv RegionNum = sys.argv CalcFeild1 = sy s.argv DenFeild = sys.argv PeopleLim = sys.argv Mask1 = sys.argv Mask2 = sys.argv Mask3 = sys.argv Mask4 = sys.argv Mask5 = sys.argv Feild1 = sys.argv inCodes1 = sys.argv Feild2 = sys.argv inCodes2 = sys.argv Feild3 = sys.argv inCodes3 = sys.argv Feild4 = sys.argv inCodes4 = sys.argv Feild5 = sys.argv inCodes5 = sys.argv Prop = sys.argv outRaster = sys.argv inCode0 = RegionNum Table_View = Table + "_View" # gapredn = "gapredn" Ex press1 = \ "" + CalcFeild2 + \ "" + = 0" Express2 = AND + \ "" + str(Feild0) + \ "" + "
271 Express3 = "= + inCode0 + " if DenFeild == '#': Express4 = "" else: Express4 = "AND + \ "" + DenFeild + \ "" + >= 1" Express = "(" + Express1 + Ex press2 + Express3 + Express4 + ")" print Express # Generating expression exp = "" expR = "" if Mask1 == '#': expR = "" else: expR = expR + \ "" + Mask1 + \ "" + = 1 if Mask2 == '#': exp = exp + "" else: exp = exp + \ "" + Mask2 + \ "" + = 1 if Mask3 == '#': exp = exp + "" else: exp = exp + "OR + \ "" + Mask3 + \ "" + = 1 if Mask4 == '#': exp = exp + "" else: exp = exp + "OR + \ "" + Mask4 + \ "" + = 1 if Mask5 == '#': exp = exp + "" else: exp = exp + "OR + \ "" + Mask5 + \ "" + = 1 exp = expR + AND + "(" + exp + ")" print exp Table_View1 = Table + "_View1" # Process: Make Table View... # gp.MakeTableView_management(Table, Table_View, Express, "") #searchCursor1 = gp.searchCursor(Table, Express) if inCodes5 == '#' :
272 inCodes5 = "100000" if inCodes4 == '#' : inCodes4 = "100000" if inCodes3 == '#' : inCodes3 = "100000" ##inCodes1 = "3333;2121;2131;3131" inCodes1 = string.split(inCodes1, ";") ##inCodes2 = "100;99;98;97" inCodes2 = string.split(inCodes2, ";") ##inCodes3 = "2;9;1;3" inCodes3 = string.split(inCodes3, ";") ##inCodes4 = 1" inCodes4 = string.split(inCodes4, ";") ##print inCodes4 ##inCodes5 = 1" inCodes5 = string.split(inCodes5, ;") sump = 0 Count = 1 try: for inCode1 in inCodes1: print inCode1 Count = Count + 1 exp2 = \ "" + Feild1 + \ "" + = + inCode1 totExp = Express + AND + exp + AND + exp2 print exp2 print totExp Table_View2 = Table + "_View" + str(Count) print Table_View2 #gp.MakeTableView_management(Table_View1, Table_View2, exp2, "") for inCode2 in inCodes2: print inCode2 for inCode3 in inCodes3: print inCode 3 for inCode4 in inCodes4: print inCode4 for inCode5 in inCodes5: print inCode5
273 #combine_10 = Table_View2 #combine_10 = Search1 #cur = gp.upd ateCursor(combine_10) #x = 10 #y = 2015 i = 1 #print y #row = cur.Next() print "Done 2 cur = gp.updateCursor(Table, totExp) row = cur.Next() print "done while row: Var3 = row.GetValue (Feild2) print Var3 if Var3 >= int(inCode2): prin t "Second Condition Satisfied" if not Feild3 == '#': Var4 = row.GetValue (Feild3) print Var4 if Feild3 == '#': Var4 = 100000 if Var4 >= int(inCode3): print "third condition Satisfied" if not Feild4 == '#': Var5 = row. GetValue (Feild4) print Var5 if Feild4 == '#': Var5 = 100000 if Var5 >= int(inCode4): print "fou rth condition satisfied" if not Feild5 == '#': Var6 = row.GetValue (Feild5) print Var6 if Feild5 == '#': Var6 = 100000 if Var6 >= int(inCode5):
274 print "Fifth cond" #print "Allocation" if DenFei ld == '#': peop = row.GetValue (CalcFeild1) print peop else: cnt = row.CELLCOUNT den = row.GetValue (DenFeild) peop = cnt den if den == 0: year1 = 0 else: year1 = yearP if Prop == '#': Prop1 = 1 else: Prop1 = row.GetValue (Prop) row.GetValue (CalcFeild2) year1 = yearP print year1 row.SetValue (CalcFeild2, year1) y1 = Prop1 peop y = int(y1) print y row.GetValue (CalcFeild1) row.SetValue (CalcFeild1, y) cur.updaterow (row) sump = sump + y print sump,PeopleLim if sump >= int(PeopleLim) : Express10 = \ "" + CalcFeild2 + \ "" + = + yearP gp.ExtractByAttributes_sa(Table, Express10, outRaster) sys.exit(0) row = cur.Next() else: row = cur.Next() else:
275 row = cur.Next() else: row = cur.Next() else: row = cur.Next() del row, cur Expres s10 = \ "" + CalcFeild2 + \ "" + = + yearP gp.ExtractByAttributes_sa(Table, Express10, outRaster) except Exception, ErrorDesc: msgStr = gp.GetMessages(2) sys.exit(1)
276 LIST OF REFERENCES American Automobile Association (20 11). Your driving cost 2010 edition. Retrieved January 5, 2011, from http://www.aaaexchange.com/Assets/Files/201048935480.Driving%20Costs%20 2010.pdf Arafat A., Srinivasan, S., & Steiner R.L. (2010 ). The impacts of land use and urban form on travel behavior: A parcel level destination choice model applied on south east Florida. Proceeding of the 12th World Conference on Transport Research Conference. Retrieved No vember 1, 2010, from http://www.civil.ist.utl.pt/wctr12_lisboa/proceedings.htm Arafat, A., Steiner, R.L., & Bejleri, I. (2008, July). A method for measuring network distance using network shortest distance and spatial interpolation Paper presented at ACSP AESOP 4th Joint Congress, Chicago, IL. Bejleri I., Steiner R.L.,Wheelock, J., Perez, B., & Fischman A. (2008, July ). Measuring walkability around elementary schools: A comparison of urban form indices in four Florida counties Paper presented at ACSP AESO P 4th Joint Congress, Chicago, IL. Ben Akiva, M. & Bowman, J.L. (1998). Activity based travel demand model systems. In P. Marcotte & S. Nguyen (Eds.), Equilibrium and Advanced Transportation Modeling (pp. 27 46). Boston, MA: Kluwer Academic Publishers. Bha t, C.R., & Guo, J.Y. (2004). A mixed spatially correlated logit model: Formulation and application to residential choice modeling. Transportation Research Part B: Methodological, 38 (2), 147 168. Bhat C.R., Handy, S., Kockelman, K., Mahmassani, H., Gopal, A., Srour, I., & Weston, L. (2002). Development of an urban accessibility index. formulations, aggregation and application (Report No. FHWA/TX 02 4938 4). Austin, TX: University of Texas at Austin Bhat, C. R., & Koppelman, F. S. (1999). A retrospective and prospective survey of time use research Transportation, 26 (2) 119 139 Bhat, C. R., & Koppelman, F. S. (1993). An endogenous switching simultaneous equation system of employment, income, and c ar ownership. Transportation Research Part A: Policy and Practice, 27 (6), 447 459. Bhat, C.R., & Lawton, T.K. (1999). Passenger travel demand forecasting Transportation Research Board Committee on Passenger Travel Demand Forecasting. Retrieved December 15 2009, from http://onlinepubs.trb.org/onlinepubs/millennium/00083.pdf
277 Carr, M. (2008). Key components of environmental planning Gainesville, FL: University of Florida. Carr M., & Zwick P. (2005). Using GIS suitability analysis to identifying future land use conflicts in north central Florida. Journal of Conservation Planning, 1 (1). Retrieved March 15, 2009, from http:// www.journalconsplanning.org/2005/index.html Carr M., & Zwick. P. (2007). Smart land use analysis: The LUCIS model Redlands, CA: ESRI press. Center for Neigborhood Technology CNT (2007). Unco vering the hidden assets of established communities, Housing + Transportation Affordability Report Retrieved January 20, 2010, from http://www .oak forest.org/UserFiles/File/Planning_&_Zoning/Research_&_Studies/H T Oak Forest.pdf Center for Neigborhood Technology CNT (2010). True affordability and location efficiency Retrieved January 30, 2011, from http: //htaindex.cnt.org/ Cervero, R., & Kockelman K. (1997). Travel demand and the 3Ds: density, diversity, and NT design. Transportation Research D 2 199 219. Cervero, R., & Radisch C. (1996). Travel choices in pedestrian versus automobile oriented neighborh oods. Transport Policy 3 127 141. Collins M., Steiner F, & Rushman M. (2001). Land use suitability analysis in the United States: historical development and promising technological achievements. Environmental Management, 28 (5) 611 621. Dill J. (2004). M easuring network connectivity for bicycling and walking Paper presented at the 83th Annual Meeting of the Transportation Research Board (CD ROM), Washington, DC. Driessen, P. M., & Konijn, N. T. (1992) Land use systems analysis Wageningen Agricultural Un iversity: Department of Soil Science and Geology. El Geneidy, A., & Levinson D. (2007, January). Network circuity and the journey to work Paper presented at the University Transport Study Group Conference, Harrowgate, England. Retrieved February 15, 2008, from http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.145.1196 Environmental Systems Research Institute (2011). ArcGIS 9.3 desktop help Retrieved February 18,2011,from http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=welcome
278 Ewing, R., Bartholomew, K., Winkelman, S., Walters, J. & Chen, D. (2008). Growing cooler: The evidence on urban development and climate change Washington, D.C.: Urban Land Insti tute.Ewing, R., & Cervero, R. (2001). Travel and the built environment: A synthesis. Transportation Research Record: Journal of the Transportation Research Board, 1780 87 114. Florida Affordable Housing Model (2009). Affordable housing model methodology University of Florida: Department of Urban and Regional Planning. Frank, L., & Pivo G. (1994). Impacts of mixed use and density on utilization of three modes of travel: Single occupant vehicle, transit, and walking. Transportation Research Record 1466 44 52. Frank L., Sallis J., Conway T., Chapman J., Saelens B., & Bachman W. (2006). Many pathways from land use to health: Associations between neighborhood walkability and active transportation, body mass index, and air quality. Journal of the American Pl anning Association, 72 75 87. Franklin, J.P., Waddell, P., & Britting, J. (2002). Sensitivity analysis approach for an integrated land development & travel demand modeling system. Retrieved January 5, 2010, from http://www.urbansim.org/pub/Research/ResearchPapers/ACSP02.doc Galster, G., Hanson, R. Ratcliffe, M.R. Wolman, H. Coleman, S., & Freihage, J. (2001). Wrestling sprawl to the ground: Defining and measuring an elusiv e concept. Housing Policy Debate, 12 (4), 68 717. Giuliano, G. (2004). Land use impacts of transportation investments: Highway and transit. In S. S. Hanson and G. Giuliano (Eds.), The Geography of Urban Transportation (3 rd ed., pp. 237 273). New York: The Guildford Press. Grengs, J. (2009). Job accessibility and modal mismatch in Detroit. Journal of Transport Geography 18(1), 42 54. doi:10.1016/j.jtrangeo.2009.0.01.012 Guo, J.Y., & Bhat, C.R. (2004). Modifiable areal units: A problem or matter of perceptio n in the context of residential location choice modeling? Transportation Research Record 1898 138 147. Handy, S. (1996). Urban form and pedestrian choices: Study of Austin neighborhoods. Transportation Research Board: Journal of the Transportation Resear ch Board, 1552 135 144. doi:10.3141/1552 19 Handy, S. (2004). Planning for accessibility: In theory and practice. Proceeding from the Access to Destination Conference. Minneapolis, MN: University of Minnesota.
279 Handy, S., Cao, X., Buehler, T. J., & Mokhtar ian, P. L. (2005). The link between the built environment and travel behavior: Correlation or causality? Paper presented at the 84th Annual Meeting of the Transportation Research Board (CD ROM), Washington, D.C. Hanson, S. (2004). The context of urban travel: Concepts and recent trends. In S. Hanson G. Giuliano(Eds.), The Geography of Urban Transportation (3rd ed., pp. 3 29). New York: The Guildford Press. Hess, P., Moudon, A., & Logsdon, M.(2003). Measuring land use patterns for transportation research. Transportation Research Record, 1780 17 24. Iacono, M., & Levinson, D. (2008). Predicting land use change: How much does transportation matter? University of Minnesota: Nexus Research Group. Iacono, M., Levinson, D., & El Geneidy, A. (2008). Models of transportation and land use change. Journal of Planning Literature 22 323 340 Janelle, D. (2004). Impact of information technologies. In S. Hansonand G. Giuliano (Eds.), The Geography of Urban Transportation (3 rd ed., pp. 86 112). New York: The Guildford Press. Jelinski, D., & Wu, J. (1996). The modifiable areal unit problem and implications for landscape ecology Landscape Ecology, 11 (3), 129 140. Johnston, R. A. (2004). The urban transportation planning process In S. Hanson and G. Giuliano (Eds.), The Geography of Urban Transportation (3 rd ed., pp. 115 140). New York: The Guildford Press. Kutz, M. (2003). Handbook of transportation engineering New York: McGraw Hill. Kwan, M., & Weber, J. (2008). Scale and acce ssibility: Implications for the analysis of land use travel interaction. Applied Geography 28 (2), 110 123. Lee, B. (2004). Parcel level measure of public transit accessibility to destinations Proceeding from the Access to Destination Conference. Minneapo lis, MN: University of Minnesota. Lee, R.W., & Cervero, R. (2007). The effect of housing near transit stations on vehicle trip rates and transit trip generation: A summary review of available evidence. Retrieved April 15, 2009, from http://www.reconnecting america.org/assets/Uploads/hcd_tod_resource_paper_2 008_09_20.pdf Leslie, E, Coffee, N., Frank, L., Owen, N., Bauman, A., & Hugo, G. (2007). Walkability of local communities: Using geographic information systems to objectively assess relevant environmental attributes. Health Place, 13 111 122.
280 Levinson, D., & Chen, W. (2004). Paving new ground: A Markov chain model of the change in transportation networks and land use. Proceeding from the Access to Destination Conference. Minneapolis, MN: University of Min nesota. Levinson, D., & Krizek, K.(2008). Planning for place and plexus: Metropolitan land use and transport New York: Routledge. Levinson, D., & Kumar, A. (1994). Integrating feedback into the transportation planning model. Transportation Research Recor d: Journal of the Transportation Research Board, 1413 70 77. Lipman, B. (2006). A heavy load: The combined housing and transportation burdens of working families Washington, DC: Center for Housing Policy. Litman T. (2008). Land use impact on transport: H ow land use affect travel behavior Victoria Transport Policy Institute. Retrieved February 3, 2009, from www.vtpi.org Malczewski, J. (1999). GIS and multicriteria decision analysis New York: John Wiley and Sons. Malczewski, J. (2004). GIS based land use suitability analysis: a critical overview. Progress in Planning, 62 3 65 McHarg, I. L. (1969). Design with nature Garden City, N.Y.: Natural History Press Ndubisi, F. (2002). Ecological planning: a historical an d comparative synthesis Baltimore: Johns Hopkins University Press. Nyerges, T. L., & Jankowski P. (2010). Regional and urban GIS: a decision support approach. New York: Guilford Press. g geographic information systems. Transportation Research Record ,1364 131 138. Ottensmann, J. (2004). Accessibility in the LUCI2 urban simulation model and the importance of accessibility for urban development. Proceeding from the Access to Destination C onference. Minneapolis, MN: University of Minnesota. Pozsgay, M. A., & Bhat, C. R. (2001). Destination choice modeling for home based recreational trips. Transportation Research Record, 1777 47 54. Primerano, F., & Taylor, M. (2004). An accessibility fram ework for evaluating transport policies Proceeding from the Access to Destination Conference. Minneapolis, MN: University of Minnesota. Randolph, J. (2004). Environmental land use planning and management Washington: Island Press.
281 Zuppa, P., & White, D. (2009). The state of University of Florida: Shimberg Center for Affordable Housing. Regional Economic Model Inc REMI (2008). Explore the model Retrieved Octob er 10, 2008, from http://www.remi.com/ Reynolds, H. (1998). The modifiable area unit problem: Empirical analysis by statistical simulation University of Toronto: Graduate Department of Geography. Rodriguez, D., Young, H., & Shneider, R. (2006). An easy me thod to compute index for identifying built environment that supports walking Paper presented at the 85th Annual Meeting of the Transportation Research Board (CD ROM), Washington, DC. Ross, C.L., & Dunning, A.E. (1997). Land use transportation interaction : An examination of the 1995 NPTS data. Atlanta U.S. Department of Transportation: Federal Highway Administration. Saaty, T. L. (1980). The analytic hierarchy process: Planning, priority setting, resource allocation New York: McGraw Hill. Salomon, I., & Mokhtarian, P. L. (1998). What happens when mobility inclined market segments face accessibility enhancing policies? Transportation Research Part D: Transport and Environment, 3 (3), 129 140. Srour, I., Kockelman, K., & Dunn, T. (2002, Janury). Accessibilit y indices: A connection to residential land prices and location choices. Paper presented at the 81st Annual Meeting of the Transportation Research Board (CD ROM), Washington, DC. Steiner, R.L. (1996 ). Traditional neighborhood shopping districts: patterns o f use and modes of access (Ph.D. dissertation). Berkeley: University of California. Steiner, R.L. (1994). Residential density and travel patterns: Review of the literature. Transportation Research Record, 1466 37 43. Steiner, R.L., & Srinivasan, S. (2009) Vehicles miles of travel based traffic impact assessment empirical analysis ( Unpublished manuscript). Gainesville, FL: University of Florida. Steiner R.L., Srinivasan, S., Arafat, A., Provost, R., Anderson, N., & Delarco, L. (2010). VMT based traffic impact assessment: Development of a trip length model (Report No. 2008 007). University of Florida: Center for Multimodal Solutions for Congestion Mitigation. Tomoki, N. (1999). An information statistical approach to the modifiable areal unit problem in in cidence rate maps Kita ku, Kyoto, Japan: Ritsumeikan University
282 Transit Cooperative Research Program (1998). Transit Markets of the Future: The Challenge of Change (TCRP Report 28). Washington, D.C.: National Academy Press. Transit Cooperative Research Pr ogram (2003). Traveler response to transportation system changes chapter 15 land use and site design (TCRP Report 95). Transportation Research Board 82nd Annual Meeting (CD ROM), Washington, D.C. United States Department of Housing and Urban Development (2 011 a ). Affordable housing Community Development Program. Retrieved January 30, 2011, from http://www.hud.gov/offices/cpd/affordablehousing/ United States Department of Housing and Urban Dev elopment (2011 b ). FY income limits 2009 Retrieved February 15, 2011, from http://www.huduser.org/portal/datasets/il/il09/index.html Waddell P. (1998, May) The Oregon prototype metr opolitan land use model. Proceeding of the ACSE Conference on Transportation, Land Use, and Air Quality. Retrieved November 5, 2010, from http://www.urbansim.org/pub/Research/ResearchPapers/ASCE_Model.pdf Waddell, P. (2002). UrbanSim: Modeling urban develo pment for land use, transportation and environmental planning. Journal of the American Planning Association, 68 (3) 297 314. Waddell, P., Bhat, C.R., Eluru, N., Wang, L., & Pendyala, R.M. (2003) Modeling the interdependene in household residence and workplace choices. Transportation Research Record, 2003, 84 92. Waddell, P., Bhat, C.R., Ruiter, E., Bekhor, S., Outwater, M., & Schroer, E. (2001). Land use and travel demand fo recasting models: Review of the literature and operational models Seattle, WA: Puget Sound Regional Council. Waddell, P., Borning, A., Noth, M., Freier, N., Becke, M., & Ulfarsson, G. (2003). UrbanSim: A simulation system for land use and transportation. Networks and Spatial Economics, 3 43 67. Wegener, M. (2005). Urban land use transportation models. In D.J. Macquire, M. Batty, & M.F. Goodchild (Eds.), GIS, Spatial Analysis, and Modeling (1st ed., pp. 203 220). Redlands, CA: ESRI Press. Yamada, H. (197 2). On the theory of residential location: Accessibility, space, leisure, and environmental quality. Papers in Regional Science, 29 125 135. doi: 10.1111/j.1435 5597.1972.tb01537.x
283 Zhang S. (2005). Feasibility study on transit oriented development, using urban form and non urban form variables Proceedings of the Twenty Fifth Annual ESRI User Conference Retrieved November 13, 2009, from http://proceedings.esri.com/libra ry/userconf/proc05/papers/pap1150.pdf Zhao F., Chow, L., Gan, A., & Ubaka, I. (2003). Forecasting transit walk accessibility: regression model alternative to buffer method. Transportation Research Record, 1835 (8), 34 41.
284 BIOGRAPHICAL SKETCH Mr. Arafat received his Ph.D. in Design, Construction and Planning from the University of Florida in the summer of 2011. He has an educational and research background in civil engineering, land use and transportation planning. research interest is in th e coordination of land use and transportation. His research generally focuses on understanding the relationship between urban form and multimodal transportation systems to reduce the vehicle miles of travel and to improve air quality. His research uses dis aggregated and fine spatial resolution approaches using customized GIS automation tools. He has research work in both land use modeling and multimodal transportation systems and has presented his work at many local and international conferences. Mr. Arafa t taught Geographic Information Systems (GIS) in Birzeit University in Palestine before receiving a scholarship from the Palestinian Faculty Development Program to continue his study towards a Ph.D. Upon his completion of his Ph.D. he is interested in cont inuing research work with the University of Florida and teaching in Birzeit University in his country.